diff --git "a/openai_whisper-large-v2_turbo_955MB/AudioEncoder.mlmodelc/model.mil" "b/openai_whisper-large-v2_turbo_955MB/AudioEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/openai_whisper-large-v2_turbo_955MB/AudioEncoder.mlmodelc/model.mil" @@ -0,0 +1,42287 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] +{ + func main(tensor melspectrogram_features) { + tensor var_90 = const()[name = tensor("op_90"), val = tensor([1, 1])]; + tensor var_96 = const()[name = tensor("op_96"), val = tensor([1, 1])]; + tensor var_101 = const()[name = tensor("op_101"), val = tensor(1)]; + tensor var_106_pad_type_0 = const()[name = tensor("op_106_pad_type_0"), val = tensor("custom")]; + tensor var_106_pad_0 = const()[name = tensor("op_106_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_81_to_fp16 = const()[name = tensor("op_81_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor var_87_to_fp16 = const()[name = tensor("op_87_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614528)))]; + tensor var_106_cast_fp16 = conv(bias = var_87_to_fp16, dilations = var_96, groups = var_101, pad = var_106_pad_0, pad_type = var_106_pad_type_0, strides = var_90, weight = var_81_to_fp16, x = melspectrogram_features)[name = tensor("op_106_cast_fp16")]; + tensor hidden_states_1_mode_0 = const()[name = tensor("hidden_states_1_mode_0"), val = tensor("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_106_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor var_130 = const()[name = tensor("op_130"), val = tensor([2, 2])]; + tensor var_136 = const()[name = tensor("op_136"), val = tensor([1, 1])]; + tensor var_141 = const()[name = tensor("op_141"), val = tensor(1)]; + tensor var_146_pad_type_0 = const()[name = tensor("op_146_pad_type_0"), val = tensor("custom")]; + tensor var_146_pad_0 = const()[name = tensor("op_146_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_121_to_fp16 = const()[name = tensor("op_121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617152)))]; + tensor var_127_to_fp16 = const()[name = tensor("op_127_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10447616)))]; + tensor var_146_cast_fp16 = conv(bias = var_127_to_fp16, dilations = var_136, groups = var_141, pad = var_146_pad_0, pad_type = var_146_pad_type_0, strides = var_130, weight = var_121_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor("op_146_cast_fp16")]; + tensor hidden_states_3_mode_0 = const()[name = tensor("hidden_states_3_mode_0"), val = tensor("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_146_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; + tensor var_164_to_fp16 = const()[name = tensor("op_164_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10450240)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_164_to_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor var_174 = const()[name = tensor("op_174"), val = tensor(3)]; + tensor var_199 = const()[name = tensor("op_199"), val = tensor(1)]; + tensor var_200 = const()[name = tensor("op_200"), val = tensor(true)]; + tensor var_210 = const()[name = tensor("op_210"), val = tensor([1])]; + tensor channels_mean_1_cast_fp16 = reduce_mean(axes = var_210, keep_dims = var_200, x = inputs_1_cast_fp16)[name = tensor("channels_mean_1_cast_fp16")]; + tensor zero_mean_1_cast_fp16 = sub(x = inputs_1_cast_fp16, y = channels_mean_1_cast_fp16)[name = tensor("zero_mean_1_cast_fp16")]; + tensor zero_mean_sq_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = zero_mean_1_cast_fp16)[name = tensor("zero_mean_sq_1_cast_fp16")]; + tensor var_214 = const()[name = tensor("op_214"), val = tensor([1])]; + tensor var_215_cast_fp16 = reduce_mean(axes = var_214, keep_dims = var_200, x = zero_mean_sq_1_cast_fp16)[name = tensor("op_215_cast_fp16")]; + tensor var_216_to_fp16 = const()[name = tensor("op_216_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_217_cast_fp16 = add(x = var_215_cast_fp16, y = var_216_to_fp16)[name = tensor("op_217_cast_fp16")]; + tensor denom_1_epsilon_0 = const()[name = tensor("denom_1_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_1_cast_fp16 = rsqrt(epsilon = denom_1_epsilon_0, x = var_217_cast_fp16)[name = tensor("denom_1_cast_fp16")]; + tensor out_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = denom_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = tensor("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14290304)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = tensor("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14292928)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = tensor("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14295552)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = tensor("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14298176)))]; + tensor obj_1_epsilon_0_to_fp16 = const()[name = tensor("obj_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor("obj_1_cast_fp16")]; + tensor var_235 = const()[name = tensor("op_235"), val = tensor([1, 1])]; + tensor var_237 = const()[name = tensor("op_237"), val = tensor([1, 1])]; + tensor pretrained_out_1_pad_type_0 = const()[name = tensor("pretrained_out_1_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_1_pad_0 = const()[name = tensor("pretrained_out_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14300800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15120064))), name = tensor("layers_0_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_0_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15120192)))]; + tensor pretrained_out_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_237, groups = var_199, pad = pretrained_out_1_pad_0, pad_type = pretrained_out_1_pad_type_0, strides = var_235, weight = layers_0_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("pretrained_out_1_cast_fp16")]; + tensor var_241 = const()[name = tensor("op_241"), val = tensor([1, 1])]; + tensor var_243 = const()[name = tensor("op_243"), val = tensor([1, 1])]; + tensor input_1_pad_type_0 = const()[name = tensor("input_1_pad_type_0"), val = tensor("custom")]; + tensor input_1_pad_0 = const()[name = tensor("input_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15122816)))]; + tensor input_1_cast_fp16 = conv(dilations = var_243, groups = var_199, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = var_241, weight = layers_0_self_attn_q_proj_loraA_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor var_247 = const()[name = tensor("op_247"), val = tensor([1, 1])]; + tensor var_249 = const()[name = tensor("op_249"), val = tensor([1, 1])]; + tensor lora_out_1_pad_type_0 = const()[name = tensor("lora_out_1_pad_type_0"), val = tensor("custom")]; + tensor lora_out_1_pad_0 = const()[name = tensor("lora_out_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_3_weight_0_to_fp16 = const()[name = tensor("lora_out_3_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15163840)))]; + tensor lora_out_3_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_249, groups = var_199, pad = lora_out_1_pad_0, pad_type = lora_out_1_pad_type_0, strides = var_247, weight = lora_out_3_weight_0_to_fp16, x = input_1_cast_fp16)[name = tensor("lora_out_3_cast_fp16")]; + tensor query_1_cast_fp16 = add(x = pretrained_out_1_cast_fp16, y = lora_out_3_cast_fp16)[name = tensor("query_1_cast_fp16")]; + tensor var_259 = const()[name = tensor("op_259"), val = tensor([1, 1])]; + tensor var_261 = const()[name = tensor("op_261"), val = tensor([1, 1])]; + tensor pretrained_out_3_pad_type_0 = const()[name = tensor("pretrained_out_3_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_3_pad_0 = const()[name = tensor("pretrained_out_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15204864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16024128))), name = tensor("layers_0_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_3_cast_fp16 = conv(dilations = var_261, groups = var_199, pad = pretrained_out_3_pad_0, pad_type = pretrained_out_3_pad_type_0, strides = var_259, weight = layers_0_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("pretrained_out_3_cast_fp16")]; + tensor var_265 = const()[name = tensor("op_265"), val = tensor([1, 1])]; + tensor var_267 = const()[name = tensor("op_267"), val = tensor([1, 1])]; + tensor input_3_pad_type_0 = const()[name = tensor("input_3_pad_type_0"), val = tensor("custom")]; + tensor input_3_pad_0 = const()[name = tensor("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16024256)))]; + tensor input_3_cast_fp16 = conv(dilations = var_267, groups = var_199, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_265, weight = layers_0_self_attn_k_proj_loraA_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor var_271 = const()[name = tensor("op_271"), val = tensor([1, 1])]; + tensor var_273 = const()[name = tensor("op_273"), val = tensor([1, 1])]; + tensor lora_out_5_pad_type_0 = const()[name = tensor("lora_out_5_pad_type_0"), val = tensor("custom")]; + tensor lora_out_5_pad_0 = const()[name = tensor("lora_out_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_7_weight_0_to_fp16 = const()[name = tensor("lora_out_7_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16065280)))]; + tensor lora_out_7_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_273, groups = var_199, pad = lora_out_5_pad_0, pad_type = lora_out_5_pad_type_0, strides = var_271, weight = lora_out_7_weight_0_to_fp16, x = input_3_cast_fp16)[name = tensor("lora_out_7_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = pretrained_out_3_cast_fp16, y = lora_out_7_cast_fp16)[name = tensor("key_1_cast_fp16")]; + tensor var_284 = const()[name = tensor("op_284"), val = tensor([1, 1])]; + tensor var_286 = const()[name = tensor("op_286"), val = tensor([1, 1])]; + tensor pretrained_out_5_pad_type_0 = const()[name = tensor("pretrained_out_5_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_5_pad_0 = const()[name = tensor("pretrained_out_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16106304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16925568))), name = tensor("layers_0_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_0_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16925696)))]; + tensor pretrained_out_5_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_286, groups = var_199, pad = pretrained_out_5_pad_0, pad_type = pretrained_out_5_pad_type_0, strides = var_284, weight = layers_0_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("pretrained_out_5_cast_fp16")]; + tensor var_290 = const()[name = tensor("op_290"), val = tensor([1, 1])]; + tensor var_292 = const()[name = tensor("op_292"), val = tensor([1, 1])]; + tensor input_5_pad_type_0 = const()[name = tensor("input_5_pad_type_0"), val = tensor("custom")]; + tensor input_5_pad_0 = const()[name = tensor("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16928320)))]; + tensor input_5_cast_fp16 = conv(dilations = var_292, groups = var_199, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = var_290, weight = layers_0_self_attn_v_proj_loraA_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor var_296 = const()[name = tensor("op_296"), val = tensor([1, 1])]; + tensor var_298 = const()[name = tensor("op_298"), val = tensor([1, 1])]; + tensor lora_out_9_pad_type_0 = const()[name = tensor("lora_out_9_pad_type_0"), val = tensor("custom")]; + tensor lora_out_9_pad_0 = const()[name = tensor("lora_out_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_11_weight_0_to_fp16 = const()[name = tensor("lora_out_11_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16969344)))]; + tensor lora_out_11_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_298, groups = var_199, pad = lora_out_9_pad_0, pad_type = lora_out_9_pad_type_0, strides = var_296, weight = lora_out_11_weight_0_to_fp16, x = input_5_cast_fp16)[name = tensor("lora_out_11_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = pretrained_out_5_cast_fp16, y = lora_out_11_cast_fp16)[name = tensor("value_1_cast_fp16")]; + tensor var_308_begin_0 = const()[name = tensor("op_308_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_308_end_0 = const()[name = tensor("op_308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_308_end_mask_0 = const()[name = tensor("op_308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_308_cast_fp16 = slice_by_index(begin = var_308_begin_0, end = var_308_end_0, end_mask = var_308_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_308_cast_fp16")]; + tensor var_312_begin_0 = const()[name = tensor("op_312_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_312_end_0 = const()[name = tensor("op_312_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_312_end_mask_0 = const()[name = tensor("op_312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_312_cast_fp16")]; + tensor var_316_begin_0 = const()[name = tensor("op_316_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_316_end_0 = const()[name = tensor("op_316_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_316_end_mask_0 = const()[name = tensor("op_316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_316_cast_fp16")]; + tensor var_320_begin_0 = const()[name = tensor("op_320_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_320_end_0 = const()[name = tensor("op_320_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_320_end_mask_0 = const()[name = tensor("op_320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_320_cast_fp16 = slice_by_index(begin = var_320_begin_0, end = var_320_end_0, end_mask = var_320_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_320_cast_fp16")]; + tensor var_324_begin_0 = const()[name = tensor("op_324_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_324_end_0 = const()[name = tensor("op_324_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_324_end_mask_0 = const()[name = tensor("op_324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_324_cast_fp16 = slice_by_index(begin = var_324_begin_0, end = var_324_end_0, end_mask = var_324_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_324_cast_fp16")]; + tensor var_328_begin_0 = const()[name = tensor("op_328_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_328_end_0 = const()[name = tensor("op_328_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_328_end_mask_0 = const()[name = tensor("op_328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_328_cast_fp16 = slice_by_index(begin = var_328_begin_0, end = var_328_end_0, end_mask = var_328_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_328_cast_fp16")]; + tensor var_332_begin_0 = const()[name = tensor("op_332_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_332_end_0 = const()[name = tensor("op_332_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_332_end_mask_0 = const()[name = tensor("op_332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_332_cast_fp16")]; + tensor var_336_begin_0 = const()[name = tensor("op_336_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_336_end_0 = const()[name = tensor("op_336_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_336_end_mask_0 = const()[name = tensor("op_336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_336_cast_fp16 = slice_by_index(begin = var_336_begin_0, end = var_336_end_0, end_mask = var_336_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_336_cast_fp16")]; + tensor var_340_begin_0 = const()[name = tensor("op_340_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_340_end_0 = const()[name = tensor("op_340_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_340_end_mask_0 = const()[name = tensor("op_340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_340_cast_fp16 = slice_by_index(begin = var_340_begin_0, end = var_340_end_0, end_mask = var_340_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_340_cast_fp16")]; + tensor var_344_begin_0 = const()[name = tensor("op_344_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_344_end_0 = const()[name = tensor("op_344_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_344_end_mask_0 = const()[name = tensor("op_344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_344_cast_fp16 = slice_by_index(begin = var_344_begin_0, end = var_344_end_0, end_mask = var_344_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_344_cast_fp16")]; + tensor var_348_begin_0 = const()[name = tensor("op_348_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_348_end_0 = const()[name = tensor("op_348_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_348_end_mask_0 = const()[name = tensor("op_348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_348_cast_fp16 = slice_by_index(begin = var_348_begin_0, end = var_348_end_0, end_mask = var_348_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_348_cast_fp16")]; + tensor var_352_begin_0 = const()[name = tensor("op_352_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_352_end_0 = const()[name = tensor("op_352_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_352_end_mask_0 = const()[name = tensor("op_352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_352_cast_fp16 = slice_by_index(begin = var_352_begin_0, end = var_352_end_0, end_mask = var_352_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_352_cast_fp16")]; + tensor var_356_begin_0 = const()[name = tensor("op_356_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_356_end_0 = const()[name = tensor("op_356_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_356_end_mask_0 = const()[name = tensor("op_356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_356_cast_fp16 = slice_by_index(begin = var_356_begin_0, end = var_356_end_0, end_mask = var_356_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_356_cast_fp16")]; + tensor var_360_begin_0 = const()[name = tensor("op_360_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_360_end_0 = const()[name = tensor("op_360_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_360_end_mask_0 = const()[name = tensor("op_360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_360_cast_fp16 = slice_by_index(begin = var_360_begin_0, end = var_360_end_0, end_mask = var_360_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_360_cast_fp16")]; + tensor var_364_begin_0 = const()[name = tensor("op_364_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_364_end_0 = const()[name = tensor("op_364_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_364_end_mask_0 = const()[name = tensor("op_364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_364_cast_fp16 = slice_by_index(begin = var_364_begin_0, end = var_364_end_0, end_mask = var_364_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_364_cast_fp16")]; + tensor var_368_begin_0 = const()[name = tensor("op_368_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_368_end_0 = const()[name = tensor("op_368_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_368_end_mask_0 = const()[name = tensor("op_368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_368_cast_fp16 = slice_by_index(begin = var_368_begin_0, end = var_368_end_0, end_mask = var_368_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_368_cast_fp16")]; + tensor var_372_begin_0 = const()[name = tensor("op_372_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_372_end_0 = const()[name = tensor("op_372_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_372_end_mask_0 = const()[name = tensor("op_372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_372_cast_fp16")]; + tensor var_376_begin_0 = const()[name = tensor("op_376_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_376_end_0 = const()[name = tensor("op_376_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_376_end_mask_0 = const()[name = tensor("op_376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_376_cast_fp16 = slice_by_index(begin = var_376_begin_0, end = var_376_end_0, end_mask = var_376_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_376_cast_fp16")]; + tensor var_380_begin_0 = const()[name = tensor("op_380_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_380_end_0 = const()[name = tensor("op_380_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_380_end_mask_0 = const()[name = tensor("op_380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_380_cast_fp16 = slice_by_index(begin = var_380_begin_0, end = var_380_end_0, end_mask = var_380_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_380_cast_fp16")]; + tensor var_384_begin_0 = const()[name = tensor("op_384_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_384_end_0 = const()[name = tensor("op_384_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_384_end_mask_0 = const()[name = tensor("op_384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_384_cast_fp16 = slice_by_index(begin = var_384_begin_0, end = var_384_end_0, end_mask = var_384_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_384_cast_fp16")]; + tensor var_393_begin_0 = const()[name = tensor("op_393_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_393_end_0 = const()[name = tensor("op_393_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_393_end_mask_0 = const()[name = tensor("op_393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_393_cast_fp16 = slice_by_index(begin = var_393_begin_0, end = var_393_end_0, end_mask = var_393_end_mask_0, x = var_308_cast_fp16)[name = tensor("op_393_cast_fp16")]; + tensor var_400_begin_0 = const()[name = tensor("op_400_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_400_end_0 = const()[name = tensor("op_400_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_400_end_mask_0 = const()[name = tensor("op_400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_400_cast_fp16 = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = var_308_cast_fp16)[name = tensor("op_400_cast_fp16")]; + tensor var_407_begin_0 = const()[name = tensor("op_407_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_407_end_0 = const()[name = tensor("op_407_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_407_end_mask_0 = const()[name = tensor("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = var_308_cast_fp16)[name = tensor("op_407_cast_fp16")]; + tensor var_414_begin_0 = const()[name = tensor("op_414_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_414_end_0 = const()[name = tensor("op_414_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_414_end_mask_0 = const()[name = tensor("op_414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_414_cast_fp16 = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, x = var_308_cast_fp16)[name = tensor("op_414_cast_fp16")]; + tensor var_421_begin_0 = const()[name = tensor("op_421_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_421_end_0 = const()[name = tensor("op_421_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_421_end_mask_0 = const()[name = tensor("op_421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = var_312_cast_fp16)[name = tensor("op_421_cast_fp16")]; + tensor var_428_begin_0 = const()[name = tensor("op_428_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_428_end_0 = const()[name = tensor("op_428_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_428_end_mask_0 = const()[name = tensor("op_428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_428_cast_fp16 = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = var_312_cast_fp16)[name = tensor("op_428_cast_fp16")]; + tensor var_435_begin_0 = const()[name = tensor("op_435_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_435_end_0 = const()[name = tensor("op_435_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_435_end_mask_0 = const()[name = tensor("op_435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_435_cast_fp16 = slice_by_index(begin = var_435_begin_0, end = var_435_end_0, end_mask = var_435_end_mask_0, x = var_312_cast_fp16)[name = tensor("op_435_cast_fp16")]; + tensor var_442_begin_0 = const()[name = tensor("op_442_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_442_end_0 = const()[name = tensor("op_442_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_442_end_mask_0 = const()[name = tensor("op_442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = var_312_cast_fp16)[name = tensor("op_442_cast_fp16")]; + tensor var_449_begin_0 = const()[name = tensor("op_449_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_449_end_0 = const()[name = tensor("op_449_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_449_end_mask_0 = const()[name = tensor("op_449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_449_cast_fp16 = slice_by_index(begin = var_449_begin_0, end = var_449_end_0, end_mask = var_449_end_mask_0, x = var_316_cast_fp16)[name = tensor("op_449_cast_fp16")]; + tensor var_456_begin_0 = const()[name = tensor("op_456_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_456_end_0 = const()[name = tensor("op_456_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_456_end_mask_0 = const()[name = tensor("op_456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_456_cast_fp16 = slice_by_index(begin = var_456_begin_0, end = var_456_end_0, end_mask = var_456_end_mask_0, x = var_316_cast_fp16)[name = tensor("op_456_cast_fp16")]; + tensor var_463_begin_0 = const()[name = tensor("op_463_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_463_end_0 = const()[name = tensor("op_463_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_463_end_mask_0 = const()[name = tensor("op_463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_463_cast_fp16 = slice_by_index(begin = var_463_begin_0, end = var_463_end_0, end_mask = var_463_end_mask_0, x = var_316_cast_fp16)[name = tensor("op_463_cast_fp16")]; + tensor var_470_begin_0 = const()[name = tensor("op_470_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_470_end_0 = const()[name = tensor("op_470_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_470_end_mask_0 = const()[name = tensor("op_470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_470_cast_fp16 = slice_by_index(begin = var_470_begin_0, end = var_470_end_0, end_mask = var_470_end_mask_0, x = var_316_cast_fp16)[name = tensor("op_470_cast_fp16")]; + tensor var_477_begin_0 = const()[name = tensor("op_477_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_477_end_0 = const()[name = tensor("op_477_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_477_end_mask_0 = const()[name = tensor("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = var_320_cast_fp16)[name = tensor("op_477_cast_fp16")]; + tensor var_484_begin_0 = const()[name = tensor("op_484_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_484_end_0 = const()[name = tensor("op_484_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_484_end_mask_0 = const()[name = tensor("op_484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_484_cast_fp16 = slice_by_index(begin = var_484_begin_0, end = var_484_end_0, end_mask = var_484_end_mask_0, x = var_320_cast_fp16)[name = tensor("op_484_cast_fp16")]; + tensor var_491_begin_0 = const()[name = tensor("op_491_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_491_end_0 = const()[name = tensor("op_491_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_491_end_mask_0 = const()[name = tensor("op_491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_491_cast_fp16 = slice_by_index(begin = var_491_begin_0, end = var_491_end_0, end_mask = var_491_end_mask_0, x = var_320_cast_fp16)[name = tensor("op_491_cast_fp16")]; + tensor var_498_begin_0 = const()[name = tensor("op_498_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_498_end_0 = const()[name = tensor("op_498_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_498_end_mask_0 = const()[name = tensor("op_498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_498_cast_fp16 = slice_by_index(begin = var_498_begin_0, end = var_498_end_0, end_mask = var_498_end_mask_0, x = var_320_cast_fp16)[name = tensor("op_498_cast_fp16")]; + tensor var_505_begin_0 = const()[name = tensor("op_505_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_505_end_0 = const()[name = tensor("op_505_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_505_end_mask_0 = const()[name = tensor("op_505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_505_cast_fp16 = slice_by_index(begin = var_505_begin_0, end = var_505_end_0, end_mask = var_505_end_mask_0, x = var_324_cast_fp16)[name = tensor("op_505_cast_fp16")]; + tensor var_512_begin_0 = const()[name = tensor("op_512_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_512_end_0 = const()[name = tensor("op_512_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_512_end_mask_0 = const()[name = tensor("op_512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_512_cast_fp16 = slice_by_index(begin = var_512_begin_0, end = var_512_end_0, end_mask = var_512_end_mask_0, x = var_324_cast_fp16)[name = tensor("op_512_cast_fp16")]; + tensor var_519_begin_0 = const()[name = tensor("op_519_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_519_end_0 = const()[name = tensor("op_519_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_519_end_mask_0 = const()[name = tensor("op_519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = var_324_cast_fp16)[name = tensor("op_519_cast_fp16")]; + tensor var_526_begin_0 = const()[name = tensor("op_526_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_526_end_0 = const()[name = tensor("op_526_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_526_end_mask_0 = const()[name = tensor("op_526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_526_cast_fp16 = slice_by_index(begin = var_526_begin_0, end = var_526_end_0, end_mask = var_526_end_mask_0, x = var_324_cast_fp16)[name = tensor("op_526_cast_fp16")]; + tensor var_533_begin_0 = const()[name = tensor("op_533_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_533_end_0 = const()[name = tensor("op_533_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_533_end_mask_0 = const()[name = tensor("op_533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_533_cast_fp16 = slice_by_index(begin = var_533_begin_0, end = var_533_end_0, end_mask = var_533_end_mask_0, x = var_328_cast_fp16)[name = tensor("op_533_cast_fp16")]; + tensor var_540_begin_0 = const()[name = tensor("op_540_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_540_end_0 = const()[name = tensor("op_540_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_540_end_mask_0 = const()[name = tensor("op_540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_540_cast_fp16 = slice_by_index(begin = var_540_begin_0, end = var_540_end_0, end_mask = var_540_end_mask_0, x = var_328_cast_fp16)[name = tensor("op_540_cast_fp16")]; + tensor var_547_begin_0 = const()[name = tensor("op_547_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_547_end_0 = const()[name = tensor("op_547_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_547_end_mask_0 = const()[name = tensor("op_547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_547_cast_fp16 = slice_by_index(begin = var_547_begin_0, end = var_547_end_0, end_mask = var_547_end_mask_0, x = var_328_cast_fp16)[name = tensor("op_547_cast_fp16")]; + tensor var_554_begin_0 = const()[name = tensor("op_554_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_554_end_0 = const()[name = tensor("op_554_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_554_end_mask_0 = const()[name = tensor("op_554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_554_cast_fp16 = slice_by_index(begin = var_554_begin_0, end = var_554_end_0, end_mask = var_554_end_mask_0, x = var_328_cast_fp16)[name = tensor("op_554_cast_fp16")]; + tensor var_561_begin_0 = const()[name = tensor("op_561_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_561_end_0 = const()[name = tensor("op_561_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_561_end_mask_0 = const()[name = tensor("op_561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_561_cast_fp16 = slice_by_index(begin = var_561_begin_0, end = var_561_end_0, end_mask = var_561_end_mask_0, x = var_332_cast_fp16)[name = tensor("op_561_cast_fp16")]; + tensor var_568_begin_0 = const()[name = tensor("op_568_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_568_end_0 = const()[name = tensor("op_568_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_568_end_mask_0 = const()[name = tensor("op_568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_568_cast_fp16 = slice_by_index(begin = var_568_begin_0, end = var_568_end_0, end_mask = var_568_end_mask_0, x = var_332_cast_fp16)[name = tensor("op_568_cast_fp16")]; + tensor var_575_begin_0 = const()[name = tensor("op_575_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_575_end_0 = const()[name = tensor("op_575_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_575_end_mask_0 = const()[name = tensor("op_575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_575_cast_fp16 = slice_by_index(begin = var_575_begin_0, end = var_575_end_0, end_mask = var_575_end_mask_0, x = var_332_cast_fp16)[name = tensor("op_575_cast_fp16")]; + tensor var_582_begin_0 = const()[name = tensor("op_582_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_582_end_0 = const()[name = tensor("op_582_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_582_end_mask_0 = const()[name = tensor("op_582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_582_cast_fp16 = slice_by_index(begin = var_582_begin_0, end = var_582_end_0, end_mask = var_582_end_mask_0, x = var_332_cast_fp16)[name = tensor("op_582_cast_fp16")]; + tensor var_589_begin_0 = const()[name = tensor("op_589_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_589_end_0 = const()[name = tensor("op_589_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_589_end_mask_0 = const()[name = tensor("op_589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_589_cast_fp16 = slice_by_index(begin = var_589_begin_0, end = var_589_end_0, end_mask = var_589_end_mask_0, x = var_336_cast_fp16)[name = tensor("op_589_cast_fp16")]; + tensor var_596_begin_0 = const()[name = tensor("op_596_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_596_end_0 = const()[name = tensor("op_596_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_596_end_mask_0 = const()[name = tensor("op_596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_596_cast_fp16 = slice_by_index(begin = var_596_begin_0, end = var_596_end_0, end_mask = var_596_end_mask_0, x = var_336_cast_fp16)[name = tensor("op_596_cast_fp16")]; + tensor var_603_begin_0 = const()[name = tensor("op_603_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_603_end_0 = const()[name = tensor("op_603_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_603_end_mask_0 = const()[name = tensor("op_603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_603_cast_fp16 = slice_by_index(begin = var_603_begin_0, end = var_603_end_0, end_mask = var_603_end_mask_0, x = var_336_cast_fp16)[name = tensor("op_603_cast_fp16")]; + tensor var_610_begin_0 = const()[name = tensor("op_610_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_610_end_0 = const()[name = tensor("op_610_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_610_end_mask_0 = const()[name = tensor("op_610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_610_cast_fp16 = slice_by_index(begin = var_610_begin_0, end = var_610_end_0, end_mask = var_610_end_mask_0, x = var_336_cast_fp16)[name = tensor("op_610_cast_fp16")]; + tensor var_617_begin_0 = const()[name = tensor("op_617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_617_end_0 = const()[name = tensor("op_617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_617_end_mask_0 = const()[name = tensor("op_617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_617_cast_fp16 = slice_by_index(begin = var_617_begin_0, end = var_617_end_0, end_mask = var_617_end_mask_0, x = var_340_cast_fp16)[name = tensor("op_617_cast_fp16")]; + tensor var_624_begin_0 = const()[name = tensor("op_624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_624_end_0 = const()[name = tensor("op_624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_624_end_mask_0 = const()[name = tensor("op_624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_624_cast_fp16 = slice_by_index(begin = var_624_begin_0, end = var_624_end_0, end_mask = var_624_end_mask_0, x = var_340_cast_fp16)[name = tensor("op_624_cast_fp16")]; + tensor var_631_begin_0 = const()[name = tensor("op_631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_631_end_0 = const()[name = tensor("op_631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_631_end_mask_0 = const()[name = tensor("op_631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_631_cast_fp16 = slice_by_index(begin = var_631_begin_0, end = var_631_end_0, end_mask = var_631_end_mask_0, x = var_340_cast_fp16)[name = tensor("op_631_cast_fp16")]; + tensor var_638_begin_0 = const()[name = tensor("op_638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_638_end_0 = const()[name = tensor("op_638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_638_end_mask_0 = const()[name = tensor("op_638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_638_cast_fp16 = slice_by_index(begin = var_638_begin_0, end = var_638_end_0, end_mask = var_638_end_mask_0, x = var_340_cast_fp16)[name = tensor("op_638_cast_fp16")]; + tensor var_645_begin_0 = const()[name = tensor("op_645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_645_end_0 = const()[name = tensor("op_645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_645_end_mask_0 = const()[name = tensor("op_645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_645_cast_fp16 = slice_by_index(begin = var_645_begin_0, end = var_645_end_0, end_mask = var_645_end_mask_0, x = var_344_cast_fp16)[name = tensor("op_645_cast_fp16")]; + tensor var_652_begin_0 = const()[name = tensor("op_652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_652_end_0 = const()[name = tensor("op_652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_652_end_mask_0 = const()[name = tensor("op_652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_652_cast_fp16 = slice_by_index(begin = var_652_begin_0, end = var_652_end_0, end_mask = var_652_end_mask_0, x = var_344_cast_fp16)[name = tensor("op_652_cast_fp16")]; + tensor var_659_begin_0 = const()[name = tensor("op_659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_659_end_0 = const()[name = tensor("op_659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_659_end_mask_0 = const()[name = tensor("op_659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_659_cast_fp16 = slice_by_index(begin = var_659_begin_0, end = var_659_end_0, end_mask = var_659_end_mask_0, x = var_344_cast_fp16)[name = tensor("op_659_cast_fp16")]; + tensor var_666_begin_0 = const()[name = tensor("op_666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_666_end_0 = const()[name = tensor("op_666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_666_end_mask_0 = const()[name = tensor("op_666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_666_cast_fp16 = slice_by_index(begin = var_666_begin_0, end = var_666_end_0, end_mask = var_666_end_mask_0, x = var_344_cast_fp16)[name = tensor("op_666_cast_fp16")]; + tensor var_673_begin_0 = const()[name = tensor("op_673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_673_end_0 = const()[name = tensor("op_673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_673_end_mask_0 = const()[name = tensor("op_673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_673_cast_fp16 = slice_by_index(begin = var_673_begin_0, end = var_673_end_0, end_mask = var_673_end_mask_0, x = var_348_cast_fp16)[name = tensor("op_673_cast_fp16")]; + tensor var_680_begin_0 = const()[name = tensor("op_680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_680_end_0 = const()[name = tensor("op_680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_680_end_mask_0 = const()[name = tensor("op_680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_680_cast_fp16 = slice_by_index(begin = var_680_begin_0, end = var_680_end_0, end_mask = var_680_end_mask_0, x = var_348_cast_fp16)[name = tensor("op_680_cast_fp16")]; + tensor var_687_begin_0 = const()[name = tensor("op_687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_687_end_0 = const()[name = tensor("op_687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_687_end_mask_0 = const()[name = tensor("op_687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_687_cast_fp16 = slice_by_index(begin = var_687_begin_0, end = var_687_end_0, end_mask = var_687_end_mask_0, x = var_348_cast_fp16)[name = tensor("op_687_cast_fp16")]; + tensor var_694_begin_0 = const()[name = tensor("op_694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_694_end_0 = const()[name = tensor("op_694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_694_end_mask_0 = const()[name = tensor("op_694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_694_cast_fp16 = slice_by_index(begin = var_694_begin_0, end = var_694_end_0, end_mask = var_694_end_mask_0, x = var_348_cast_fp16)[name = tensor("op_694_cast_fp16")]; + tensor var_701_begin_0 = const()[name = tensor("op_701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_701_end_0 = const()[name = tensor("op_701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_701_end_mask_0 = const()[name = tensor("op_701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_701_cast_fp16 = slice_by_index(begin = var_701_begin_0, end = var_701_end_0, end_mask = var_701_end_mask_0, x = var_352_cast_fp16)[name = tensor("op_701_cast_fp16")]; + tensor var_708_begin_0 = const()[name = tensor("op_708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_708_end_0 = const()[name = tensor("op_708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_708_end_mask_0 = const()[name = tensor("op_708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_708_cast_fp16 = slice_by_index(begin = var_708_begin_0, end = var_708_end_0, end_mask = var_708_end_mask_0, x = var_352_cast_fp16)[name = tensor("op_708_cast_fp16")]; + tensor var_715_begin_0 = const()[name = tensor("op_715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_715_end_0 = const()[name = tensor("op_715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_715_end_mask_0 = const()[name = tensor("op_715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_715_cast_fp16 = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = var_352_cast_fp16)[name = tensor("op_715_cast_fp16")]; + tensor var_722_begin_0 = const()[name = tensor("op_722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_722_end_0 = const()[name = tensor("op_722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_722_end_mask_0 = const()[name = tensor("op_722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_722_cast_fp16 = slice_by_index(begin = var_722_begin_0, end = var_722_end_0, end_mask = var_722_end_mask_0, x = var_352_cast_fp16)[name = tensor("op_722_cast_fp16")]; + tensor var_729_begin_0 = const()[name = tensor("op_729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_729_end_0 = const()[name = tensor("op_729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_729_end_mask_0 = const()[name = tensor("op_729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_729_cast_fp16 = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = var_356_cast_fp16)[name = tensor("op_729_cast_fp16")]; + tensor var_736_begin_0 = const()[name = tensor("op_736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_736_end_0 = const()[name = tensor("op_736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_736_end_mask_0 = const()[name = tensor("op_736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_736_cast_fp16 = slice_by_index(begin = var_736_begin_0, end = var_736_end_0, end_mask = var_736_end_mask_0, x = var_356_cast_fp16)[name = tensor("op_736_cast_fp16")]; + tensor var_743_begin_0 = const()[name = tensor("op_743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_743_end_0 = const()[name = tensor("op_743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_743_end_mask_0 = const()[name = tensor("op_743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_743_cast_fp16 = slice_by_index(begin = var_743_begin_0, end = var_743_end_0, end_mask = var_743_end_mask_0, x = var_356_cast_fp16)[name = tensor("op_743_cast_fp16")]; + tensor var_750_begin_0 = const()[name = tensor("op_750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_750_end_0 = const()[name = tensor("op_750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_750_end_mask_0 = const()[name = tensor("op_750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_750_cast_fp16 = slice_by_index(begin = var_750_begin_0, end = var_750_end_0, end_mask = var_750_end_mask_0, x = var_356_cast_fp16)[name = tensor("op_750_cast_fp16")]; + tensor var_757_begin_0 = const()[name = tensor("op_757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_757_end_0 = const()[name = tensor("op_757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_757_end_mask_0 = const()[name = tensor("op_757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_757_cast_fp16 = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = var_360_cast_fp16)[name = tensor("op_757_cast_fp16")]; + tensor var_764_begin_0 = const()[name = tensor("op_764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_764_end_0 = const()[name = tensor("op_764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_764_end_mask_0 = const()[name = tensor("op_764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_764_cast_fp16 = slice_by_index(begin = var_764_begin_0, end = var_764_end_0, end_mask = var_764_end_mask_0, x = var_360_cast_fp16)[name = tensor("op_764_cast_fp16")]; + tensor var_771_begin_0 = const()[name = tensor("op_771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_771_end_0 = const()[name = tensor("op_771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_771_end_mask_0 = const()[name = tensor("op_771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_771_cast_fp16 = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = var_360_cast_fp16)[name = tensor("op_771_cast_fp16")]; + tensor var_778_begin_0 = const()[name = tensor("op_778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_778_end_0 = const()[name = tensor("op_778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_778_end_mask_0 = const()[name = tensor("op_778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_778_cast_fp16 = slice_by_index(begin = var_778_begin_0, end = var_778_end_0, end_mask = var_778_end_mask_0, x = var_360_cast_fp16)[name = tensor("op_778_cast_fp16")]; + tensor var_785_begin_0 = const()[name = tensor("op_785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_785_end_0 = const()[name = tensor("op_785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_785_end_mask_0 = const()[name = tensor("op_785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_785_cast_fp16 = slice_by_index(begin = var_785_begin_0, end = var_785_end_0, end_mask = var_785_end_mask_0, x = var_364_cast_fp16)[name = tensor("op_785_cast_fp16")]; + tensor var_792_begin_0 = const()[name = tensor("op_792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_792_end_0 = const()[name = tensor("op_792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_792_end_mask_0 = const()[name = tensor("op_792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_792_cast_fp16 = slice_by_index(begin = var_792_begin_0, end = var_792_end_0, end_mask = var_792_end_mask_0, x = var_364_cast_fp16)[name = tensor("op_792_cast_fp16")]; + tensor var_799_begin_0 = const()[name = tensor("op_799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_799_end_0 = const()[name = tensor("op_799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_799_end_mask_0 = const()[name = tensor("op_799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_799_cast_fp16 = slice_by_index(begin = var_799_begin_0, end = var_799_end_0, end_mask = var_799_end_mask_0, x = var_364_cast_fp16)[name = tensor("op_799_cast_fp16")]; + tensor var_806_begin_0 = const()[name = tensor("op_806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_806_end_0 = const()[name = tensor("op_806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_806_end_mask_0 = const()[name = tensor("op_806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_806_cast_fp16 = slice_by_index(begin = var_806_begin_0, end = var_806_end_0, end_mask = var_806_end_mask_0, x = var_364_cast_fp16)[name = tensor("op_806_cast_fp16")]; + tensor var_813_begin_0 = const()[name = tensor("op_813_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_813_end_0 = const()[name = tensor("op_813_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_813_end_mask_0 = const()[name = tensor("op_813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_813_cast_fp16 = slice_by_index(begin = var_813_begin_0, end = var_813_end_0, end_mask = var_813_end_mask_0, x = var_368_cast_fp16)[name = tensor("op_813_cast_fp16")]; + tensor var_820_begin_0 = const()[name = tensor("op_820_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_820_end_0 = const()[name = tensor("op_820_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_820_end_mask_0 = const()[name = tensor("op_820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_820_cast_fp16 = slice_by_index(begin = var_820_begin_0, end = var_820_end_0, end_mask = var_820_end_mask_0, x = var_368_cast_fp16)[name = tensor("op_820_cast_fp16")]; + tensor var_827_begin_0 = const()[name = tensor("op_827_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_827_end_0 = const()[name = tensor("op_827_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_827_end_mask_0 = const()[name = tensor("op_827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_827_cast_fp16 = slice_by_index(begin = var_827_begin_0, end = var_827_end_0, end_mask = var_827_end_mask_0, x = var_368_cast_fp16)[name = tensor("op_827_cast_fp16")]; + tensor var_834_begin_0 = const()[name = tensor("op_834_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_834_end_0 = const()[name = tensor("op_834_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_834_end_mask_0 = const()[name = tensor("op_834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_834_cast_fp16 = slice_by_index(begin = var_834_begin_0, end = var_834_end_0, end_mask = var_834_end_mask_0, x = var_368_cast_fp16)[name = tensor("op_834_cast_fp16")]; + tensor var_841_begin_0 = const()[name = tensor("op_841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_841_end_0 = const()[name = tensor("op_841_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_841_end_mask_0 = const()[name = tensor("op_841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_841_cast_fp16 = slice_by_index(begin = var_841_begin_0, end = var_841_end_0, end_mask = var_841_end_mask_0, x = var_372_cast_fp16)[name = tensor("op_841_cast_fp16")]; + tensor var_848_begin_0 = const()[name = tensor("op_848_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_848_end_0 = const()[name = tensor("op_848_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_848_end_mask_0 = const()[name = tensor("op_848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_848_cast_fp16 = slice_by_index(begin = var_848_begin_0, end = var_848_end_0, end_mask = var_848_end_mask_0, x = var_372_cast_fp16)[name = tensor("op_848_cast_fp16")]; + tensor var_855_begin_0 = const()[name = tensor("op_855_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_855_end_0 = const()[name = tensor("op_855_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_855_end_mask_0 = const()[name = tensor("op_855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_855_cast_fp16 = slice_by_index(begin = var_855_begin_0, end = var_855_end_0, end_mask = var_855_end_mask_0, x = var_372_cast_fp16)[name = tensor("op_855_cast_fp16")]; + tensor var_862_begin_0 = const()[name = tensor("op_862_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_862_end_0 = const()[name = tensor("op_862_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_862_end_mask_0 = const()[name = tensor("op_862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_862_cast_fp16 = slice_by_index(begin = var_862_begin_0, end = var_862_end_0, end_mask = var_862_end_mask_0, x = var_372_cast_fp16)[name = tensor("op_862_cast_fp16")]; + tensor var_869_begin_0 = const()[name = tensor("op_869_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_869_end_0 = const()[name = tensor("op_869_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_869_end_mask_0 = const()[name = tensor("op_869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_869_cast_fp16 = slice_by_index(begin = var_869_begin_0, end = var_869_end_0, end_mask = var_869_end_mask_0, x = var_376_cast_fp16)[name = tensor("op_869_cast_fp16")]; + tensor var_876_begin_0 = const()[name = tensor("op_876_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_876_end_0 = const()[name = tensor("op_876_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_876_end_mask_0 = const()[name = tensor("op_876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_876_cast_fp16 = slice_by_index(begin = var_876_begin_0, end = var_876_end_0, end_mask = var_876_end_mask_0, x = var_376_cast_fp16)[name = tensor("op_876_cast_fp16")]; + tensor var_883_begin_0 = const()[name = tensor("op_883_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_883_end_0 = const()[name = tensor("op_883_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_883_end_mask_0 = const()[name = tensor("op_883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_883_cast_fp16 = slice_by_index(begin = var_883_begin_0, end = var_883_end_0, end_mask = var_883_end_mask_0, x = var_376_cast_fp16)[name = tensor("op_883_cast_fp16")]; + tensor var_890_begin_0 = const()[name = tensor("op_890_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_890_end_0 = const()[name = tensor("op_890_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_890_end_mask_0 = const()[name = tensor("op_890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_890_cast_fp16 = slice_by_index(begin = var_890_begin_0, end = var_890_end_0, end_mask = var_890_end_mask_0, x = var_376_cast_fp16)[name = tensor("op_890_cast_fp16")]; + tensor var_897_begin_0 = const()[name = tensor("op_897_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_897_end_0 = const()[name = tensor("op_897_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_897_end_mask_0 = const()[name = tensor("op_897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_897_cast_fp16 = slice_by_index(begin = var_897_begin_0, end = var_897_end_0, end_mask = var_897_end_mask_0, x = var_380_cast_fp16)[name = tensor("op_897_cast_fp16")]; + tensor var_904_begin_0 = const()[name = tensor("op_904_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_904_end_0 = const()[name = tensor("op_904_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_904_end_mask_0 = const()[name = tensor("op_904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_904_cast_fp16 = slice_by_index(begin = var_904_begin_0, end = var_904_end_0, end_mask = var_904_end_mask_0, x = var_380_cast_fp16)[name = tensor("op_904_cast_fp16")]; + tensor var_911_begin_0 = const()[name = tensor("op_911_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_911_end_0 = const()[name = tensor("op_911_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_911_end_mask_0 = const()[name = tensor("op_911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_911_cast_fp16 = slice_by_index(begin = var_911_begin_0, end = var_911_end_0, end_mask = var_911_end_mask_0, x = var_380_cast_fp16)[name = tensor("op_911_cast_fp16")]; + tensor var_918_begin_0 = const()[name = tensor("op_918_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_918_end_0 = const()[name = tensor("op_918_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_918_end_mask_0 = const()[name = tensor("op_918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_918_cast_fp16 = slice_by_index(begin = var_918_begin_0, end = var_918_end_0, end_mask = var_918_end_mask_0, x = var_380_cast_fp16)[name = tensor("op_918_cast_fp16")]; + tensor var_925_begin_0 = const()[name = tensor("op_925_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_925_end_0 = const()[name = tensor("op_925_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_925_end_mask_0 = const()[name = tensor("op_925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_925_cast_fp16 = slice_by_index(begin = var_925_begin_0, end = var_925_end_0, end_mask = var_925_end_mask_0, x = var_384_cast_fp16)[name = tensor("op_925_cast_fp16")]; + tensor var_932_begin_0 = const()[name = tensor("op_932_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_932_end_0 = const()[name = tensor("op_932_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_932_end_mask_0 = const()[name = tensor("op_932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = var_384_cast_fp16)[name = tensor("op_932_cast_fp16")]; + tensor var_939_begin_0 = const()[name = tensor("op_939_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_939_end_0 = const()[name = tensor("op_939_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_939_end_mask_0 = const()[name = tensor("op_939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_939_cast_fp16 = slice_by_index(begin = var_939_begin_0, end = var_939_end_0, end_mask = var_939_end_mask_0, x = var_384_cast_fp16)[name = tensor("op_939_cast_fp16")]; + tensor var_946_begin_0 = const()[name = tensor("op_946_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_946_end_0 = const()[name = tensor("op_946_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_946_end_mask_0 = const()[name = tensor("op_946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_946_cast_fp16 = slice_by_index(begin = var_946_begin_0, end = var_946_end_0, end_mask = var_946_end_mask_0, x = var_384_cast_fp16)[name = tensor("op_946_cast_fp16")]; + tensor k_1_perm_0 = const()[name = tensor("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_951_begin_0 = const()[name = tensor("op_951_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_951_end_0 = const()[name = tensor("op_951_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_951_end_mask_0 = const()[name = tensor("op_951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_31 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = tensor("transpose_31")]; + tensor var_951_cast_fp16 = slice_by_index(begin = var_951_begin_0, end = var_951_end_0, end_mask = var_951_end_mask_0, x = transpose_31)[name = tensor("op_951_cast_fp16")]; + tensor var_955_begin_0 = const()[name = tensor("op_955_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_955_end_0 = const()[name = tensor("op_955_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_955_end_mask_0 = const()[name = tensor("op_955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_955_cast_fp16 = slice_by_index(begin = var_955_begin_0, end = var_955_end_0, end_mask = var_955_end_mask_0, x = transpose_31)[name = tensor("op_955_cast_fp16")]; + tensor var_959_begin_0 = const()[name = tensor("op_959_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_959_end_0 = const()[name = tensor("op_959_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_959_end_mask_0 = const()[name = tensor("op_959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_959_cast_fp16 = slice_by_index(begin = var_959_begin_0, end = var_959_end_0, end_mask = var_959_end_mask_0, x = transpose_31)[name = tensor("op_959_cast_fp16")]; + tensor var_963_begin_0 = const()[name = tensor("op_963_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_963_end_0 = const()[name = tensor("op_963_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_963_end_mask_0 = const()[name = tensor("op_963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = transpose_31)[name = tensor("op_963_cast_fp16")]; + tensor var_967_begin_0 = const()[name = tensor("op_967_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_967_end_0 = const()[name = tensor("op_967_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_967_end_mask_0 = const()[name = tensor("op_967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_967_cast_fp16 = slice_by_index(begin = var_967_begin_0, end = var_967_end_0, end_mask = var_967_end_mask_0, x = transpose_31)[name = tensor("op_967_cast_fp16")]; + tensor var_971_begin_0 = const()[name = tensor("op_971_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_971_end_0 = const()[name = tensor("op_971_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_971_end_mask_0 = const()[name = tensor("op_971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_971_cast_fp16 = slice_by_index(begin = var_971_begin_0, end = var_971_end_0, end_mask = var_971_end_mask_0, x = transpose_31)[name = tensor("op_971_cast_fp16")]; + tensor var_975_begin_0 = const()[name = tensor("op_975_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_975_end_0 = const()[name = tensor("op_975_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_975_end_mask_0 = const()[name = tensor("op_975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_975_cast_fp16 = slice_by_index(begin = var_975_begin_0, end = var_975_end_0, end_mask = var_975_end_mask_0, x = transpose_31)[name = tensor("op_975_cast_fp16")]; + tensor var_979_begin_0 = const()[name = tensor("op_979_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_979_end_0 = const()[name = tensor("op_979_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_979_end_mask_0 = const()[name = tensor("op_979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_979_cast_fp16 = slice_by_index(begin = var_979_begin_0, end = var_979_end_0, end_mask = var_979_end_mask_0, x = transpose_31)[name = tensor("op_979_cast_fp16")]; + tensor var_983_begin_0 = const()[name = tensor("op_983_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_983_end_0 = const()[name = tensor("op_983_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_983_end_mask_0 = const()[name = tensor("op_983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_983_cast_fp16 = slice_by_index(begin = var_983_begin_0, end = var_983_end_0, end_mask = var_983_end_mask_0, x = transpose_31)[name = tensor("op_983_cast_fp16")]; + tensor var_987_begin_0 = const()[name = tensor("op_987_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_987_end_0 = const()[name = tensor("op_987_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_987_end_mask_0 = const()[name = tensor("op_987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_987_cast_fp16 = slice_by_index(begin = var_987_begin_0, end = var_987_end_0, end_mask = var_987_end_mask_0, x = transpose_31)[name = tensor("op_987_cast_fp16")]; + tensor var_991_begin_0 = const()[name = tensor("op_991_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_991_end_0 = const()[name = tensor("op_991_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_991_end_mask_0 = const()[name = tensor("op_991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = transpose_31)[name = tensor("op_991_cast_fp16")]; + tensor var_995_begin_0 = const()[name = tensor("op_995_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_995_end_0 = const()[name = tensor("op_995_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_995_end_mask_0 = const()[name = tensor("op_995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_995_cast_fp16 = slice_by_index(begin = var_995_begin_0, end = var_995_end_0, end_mask = var_995_end_mask_0, x = transpose_31)[name = tensor("op_995_cast_fp16")]; + tensor var_999_begin_0 = const()[name = tensor("op_999_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_999_end_0 = const()[name = tensor("op_999_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_999_end_mask_0 = const()[name = tensor("op_999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_999_cast_fp16 = slice_by_index(begin = var_999_begin_0, end = var_999_end_0, end_mask = var_999_end_mask_0, x = transpose_31)[name = tensor("op_999_cast_fp16")]; + tensor var_1003_begin_0 = const()[name = tensor("op_1003_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_1003_end_0 = const()[name = tensor("op_1003_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_1003_end_mask_0 = const()[name = tensor("op_1003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1003_cast_fp16 = slice_by_index(begin = var_1003_begin_0, end = var_1003_end_0, end_mask = var_1003_end_mask_0, x = transpose_31)[name = tensor("op_1003_cast_fp16")]; + tensor var_1007_begin_0 = const()[name = tensor("op_1007_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_1007_end_0 = const()[name = tensor("op_1007_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_1007_end_mask_0 = const()[name = tensor("op_1007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1007_cast_fp16 = slice_by_index(begin = var_1007_begin_0, end = var_1007_end_0, end_mask = var_1007_end_mask_0, x = transpose_31)[name = tensor("op_1007_cast_fp16")]; + tensor var_1011_begin_0 = const()[name = tensor("op_1011_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_1011_end_0 = const()[name = tensor("op_1011_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_1011_end_mask_0 = const()[name = tensor("op_1011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1011_cast_fp16 = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = transpose_31)[name = tensor("op_1011_cast_fp16")]; + tensor var_1015_begin_0 = const()[name = tensor("op_1015_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_1015_end_0 = const()[name = tensor("op_1015_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_1015_end_mask_0 = const()[name = tensor("op_1015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1015_cast_fp16 = slice_by_index(begin = var_1015_begin_0, end = var_1015_end_0, end_mask = var_1015_end_mask_0, x = transpose_31)[name = tensor("op_1015_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = tensor("op_1019_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_1019_end_0 = const()[name = tensor("op_1019_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_1019_end_mask_0 = const()[name = tensor("op_1019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = transpose_31)[name = tensor("op_1019_cast_fp16")]; + tensor var_1023_begin_0 = const()[name = tensor("op_1023_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_1023_end_0 = const()[name = tensor("op_1023_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_1023_end_mask_0 = const()[name = tensor("op_1023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1023_cast_fp16 = slice_by_index(begin = var_1023_begin_0, end = var_1023_end_0, end_mask = var_1023_end_mask_0, x = transpose_31)[name = tensor("op_1023_cast_fp16")]; + tensor var_1027_begin_0 = const()[name = tensor("op_1027_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_1027_end_0 = const()[name = tensor("op_1027_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_1027_end_mask_0 = const()[name = tensor("op_1027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1027_cast_fp16 = slice_by_index(begin = var_1027_begin_0, end = var_1027_end_0, end_mask = var_1027_end_mask_0, x = transpose_31)[name = tensor("op_1027_cast_fp16")]; + tensor var_1029_begin_0 = const()[name = tensor("op_1029_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1029_end_0 = const()[name = tensor("op_1029_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1029_end_mask_0 = const()[name = tensor("op_1029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1029_cast_fp16 = slice_by_index(begin = var_1029_begin_0, end = var_1029_end_0, end_mask = var_1029_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1029_cast_fp16")]; + tensor var_1033_begin_0 = const()[name = tensor("op_1033_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1033_end_0 = const()[name = tensor("op_1033_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1033_end_mask_0 = const()[name = tensor("op_1033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1033_cast_fp16 = slice_by_index(begin = var_1033_begin_0, end = var_1033_end_0, end_mask = var_1033_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1033_cast_fp16")]; + tensor var_1037_begin_0 = const()[name = tensor("op_1037_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1037_end_0 = const()[name = tensor("op_1037_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1037_end_mask_0 = const()[name = tensor("op_1037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1037_cast_fp16 = slice_by_index(begin = var_1037_begin_0, end = var_1037_end_0, end_mask = var_1037_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1037_cast_fp16")]; + tensor var_1041_begin_0 = const()[name = tensor("op_1041_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1041_end_0 = const()[name = tensor("op_1041_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1041_end_mask_0 = const()[name = tensor("op_1041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1041_cast_fp16 = slice_by_index(begin = var_1041_begin_0, end = var_1041_end_0, end_mask = var_1041_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1041_cast_fp16")]; + tensor var_1045_begin_0 = const()[name = tensor("op_1045_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1045_end_0 = const()[name = tensor("op_1045_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1045_end_mask_0 = const()[name = tensor("op_1045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1045_cast_fp16 = slice_by_index(begin = var_1045_begin_0, end = var_1045_end_0, end_mask = var_1045_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1045_cast_fp16")]; + tensor var_1049_begin_0 = const()[name = tensor("op_1049_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1049_end_0 = const()[name = tensor("op_1049_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1049_end_mask_0 = const()[name = tensor("op_1049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1049_cast_fp16 = slice_by_index(begin = var_1049_begin_0, end = var_1049_end_0, end_mask = var_1049_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1049_cast_fp16")]; + tensor var_1053_begin_0 = const()[name = tensor("op_1053_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1053_end_0 = const()[name = tensor("op_1053_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1053_end_mask_0 = const()[name = tensor("op_1053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1053_cast_fp16 = slice_by_index(begin = var_1053_begin_0, end = var_1053_end_0, end_mask = var_1053_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1053_cast_fp16")]; + tensor var_1057_begin_0 = const()[name = tensor("op_1057_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1057_end_0 = const()[name = tensor("op_1057_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1057_end_mask_0 = const()[name = tensor("op_1057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1057_cast_fp16 = slice_by_index(begin = var_1057_begin_0, end = var_1057_end_0, end_mask = var_1057_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1057_cast_fp16")]; + tensor var_1061_begin_0 = const()[name = tensor("op_1061_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1061_end_0 = const()[name = tensor("op_1061_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1061_end_mask_0 = const()[name = tensor("op_1061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1061_cast_fp16 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1061_cast_fp16")]; + tensor var_1065_begin_0 = const()[name = tensor("op_1065_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1065_end_0 = const()[name = tensor("op_1065_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1065_end_mask_0 = const()[name = tensor("op_1065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1065_cast_fp16 = slice_by_index(begin = var_1065_begin_0, end = var_1065_end_0, end_mask = var_1065_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1065_cast_fp16")]; + tensor var_1069_begin_0 = const()[name = tensor("op_1069_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1069_end_0 = const()[name = tensor("op_1069_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1069_end_mask_0 = const()[name = tensor("op_1069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1069_cast_fp16 = slice_by_index(begin = var_1069_begin_0, end = var_1069_end_0, end_mask = var_1069_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1069_cast_fp16")]; + tensor var_1073_begin_0 = const()[name = tensor("op_1073_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1073_end_0 = const()[name = tensor("op_1073_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1073_end_mask_0 = const()[name = tensor("op_1073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1073_cast_fp16 = slice_by_index(begin = var_1073_begin_0, end = var_1073_end_0, end_mask = var_1073_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1073_cast_fp16")]; + tensor var_1077_begin_0 = const()[name = tensor("op_1077_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1077_end_0 = const()[name = tensor("op_1077_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1077_end_mask_0 = const()[name = tensor("op_1077_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1077_cast_fp16 = slice_by_index(begin = var_1077_begin_0, end = var_1077_end_0, end_mask = var_1077_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1077_cast_fp16")]; + tensor var_1081_begin_0 = const()[name = tensor("op_1081_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1081_end_0 = const()[name = tensor("op_1081_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1081_end_mask_0 = const()[name = tensor("op_1081_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1081_cast_fp16 = slice_by_index(begin = var_1081_begin_0, end = var_1081_end_0, end_mask = var_1081_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1081_cast_fp16")]; + tensor var_1085_begin_0 = const()[name = tensor("op_1085_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1085_end_0 = const()[name = tensor("op_1085_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1085_end_mask_0 = const()[name = tensor("op_1085_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1085_cast_fp16 = slice_by_index(begin = var_1085_begin_0, end = var_1085_end_0, end_mask = var_1085_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1085_cast_fp16")]; + tensor var_1089_begin_0 = const()[name = tensor("op_1089_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_1089_end_0 = const()[name = tensor("op_1089_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_1089_end_mask_0 = const()[name = tensor("op_1089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1089_cast_fp16")]; + tensor var_1093_begin_0 = const()[name = tensor("op_1093_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_1093_end_0 = const()[name = tensor("op_1093_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_1093_end_mask_0 = const()[name = tensor("op_1093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1093_cast_fp16 = slice_by_index(begin = var_1093_begin_0, end = var_1093_end_0, end_mask = var_1093_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1093_cast_fp16")]; + tensor var_1097_begin_0 = const()[name = tensor("op_1097_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_1097_end_0 = const()[name = tensor("op_1097_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_1097_end_mask_0 = const()[name = tensor("op_1097_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1097_cast_fp16 = slice_by_index(begin = var_1097_begin_0, end = var_1097_end_0, end_mask = var_1097_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1097_cast_fp16")]; + tensor var_1101_begin_0 = const()[name = tensor("op_1101_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_1101_end_0 = const()[name = tensor("op_1101_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_1101_end_mask_0 = const()[name = tensor("op_1101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1101_cast_fp16 = slice_by_index(begin = var_1101_begin_0, end = var_1101_end_0, end_mask = var_1101_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1101_cast_fp16")]; + tensor var_1105_begin_0 = const()[name = tensor("op_1105_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_1105_end_0 = const()[name = tensor("op_1105_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_1105_end_mask_0 = const()[name = tensor("op_1105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1105_cast_fp16 = slice_by_index(begin = var_1105_begin_0, end = var_1105_end_0, end_mask = var_1105_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_1105_cast_fp16")]; + tensor var_1109_equation_0 = const()[name = tensor("op_1109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1109_cast_fp16 = einsum(equation = var_1109_equation_0, values = (var_951_cast_fp16, var_393_cast_fp16))[name = tensor("op_1109_cast_fp16")]; + tensor var_1110_to_fp16 = const()[name = tensor("op_1110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = var_1109_cast_fp16, y = var_1110_to_fp16)[name = tensor("aw_chunk_1_cast_fp16")]; + tensor var_1113_equation_0 = const()[name = tensor("op_1113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1113_cast_fp16 = einsum(equation = var_1113_equation_0, values = (var_951_cast_fp16, var_400_cast_fp16))[name = tensor("op_1113_cast_fp16")]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = var_1113_cast_fp16, y = var_1114_to_fp16)[name = tensor("aw_chunk_3_cast_fp16")]; + tensor var_1117_equation_0 = const()[name = tensor("op_1117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1117_cast_fp16 = einsum(equation = var_1117_equation_0, values = (var_951_cast_fp16, var_407_cast_fp16))[name = tensor("op_1117_cast_fp16")]; + tensor var_1118_to_fp16 = const()[name = tensor("op_1118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = var_1117_cast_fp16, y = var_1118_to_fp16)[name = tensor("aw_chunk_5_cast_fp16")]; + tensor var_1121_equation_0 = const()[name = tensor("op_1121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1121_cast_fp16 = einsum(equation = var_1121_equation_0, values = (var_951_cast_fp16, var_414_cast_fp16))[name = tensor("op_1121_cast_fp16")]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = var_1121_cast_fp16, y = var_1122_to_fp16)[name = tensor("aw_chunk_7_cast_fp16")]; + tensor var_1125_equation_0 = const()[name = tensor("op_1125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1125_cast_fp16 = einsum(equation = var_1125_equation_0, values = (var_955_cast_fp16, var_421_cast_fp16))[name = tensor("op_1125_cast_fp16")]; + tensor var_1126_to_fp16 = const()[name = tensor("op_1126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = var_1125_cast_fp16, y = var_1126_to_fp16)[name = tensor("aw_chunk_9_cast_fp16")]; + tensor var_1129_equation_0 = const()[name = tensor("op_1129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1129_cast_fp16 = einsum(equation = var_1129_equation_0, values = (var_955_cast_fp16, var_428_cast_fp16))[name = tensor("op_1129_cast_fp16")]; + tensor var_1130_to_fp16 = const()[name = tensor("op_1130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = var_1129_cast_fp16, y = var_1130_to_fp16)[name = tensor("aw_chunk_11_cast_fp16")]; + tensor var_1133_equation_0 = const()[name = tensor("op_1133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1133_cast_fp16 = einsum(equation = var_1133_equation_0, values = (var_955_cast_fp16, var_435_cast_fp16))[name = tensor("op_1133_cast_fp16")]; + tensor var_1134_to_fp16 = const()[name = tensor("op_1134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = var_1133_cast_fp16, y = var_1134_to_fp16)[name = tensor("aw_chunk_13_cast_fp16")]; + tensor var_1137_equation_0 = const()[name = tensor("op_1137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1137_cast_fp16 = einsum(equation = var_1137_equation_0, values = (var_955_cast_fp16, var_442_cast_fp16))[name = tensor("op_1137_cast_fp16")]; + tensor var_1138_to_fp16 = const()[name = tensor("op_1138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = var_1137_cast_fp16, y = var_1138_to_fp16)[name = tensor("aw_chunk_15_cast_fp16")]; + tensor var_1141_equation_0 = const()[name = tensor("op_1141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1141_cast_fp16 = einsum(equation = var_1141_equation_0, values = (var_959_cast_fp16, var_449_cast_fp16))[name = tensor("op_1141_cast_fp16")]; + tensor var_1142_to_fp16 = const()[name = tensor("op_1142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = var_1141_cast_fp16, y = var_1142_to_fp16)[name = tensor("aw_chunk_17_cast_fp16")]; + tensor var_1145_equation_0 = const()[name = tensor("op_1145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1145_cast_fp16 = einsum(equation = var_1145_equation_0, values = (var_959_cast_fp16, var_456_cast_fp16))[name = tensor("op_1145_cast_fp16")]; + tensor var_1146_to_fp16 = const()[name = tensor("op_1146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = var_1145_cast_fp16, y = var_1146_to_fp16)[name = tensor("aw_chunk_19_cast_fp16")]; + tensor var_1149_equation_0 = const()[name = tensor("op_1149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1149_cast_fp16 = einsum(equation = var_1149_equation_0, values = (var_959_cast_fp16, var_463_cast_fp16))[name = tensor("op_1149_cast_fp16")]; + tensor var_1150_to_fp16 = const()[name = tensor("op_1150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = var_1149_cast_fp16, y = var_1150_to_fp16)[name = tensor("aw_chunk_21_cast_fp16")]; + tensor var_1153_equation_0 = const()[name = tensor("op_1153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1153_cast_fp16 = einsum(equation = var_1153_equation_0, values = (var_959_cast_fp16, var_470_cast_fp16))[name = tensor("op_1153_cast_fp16")]; + tensor var_1154_to_fp16 = const()[name = tensor("op_1154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = var_1153_cast_fp16, y = var_1154_to_fp16)[name = tensor("aw_chunk_23_cast_fp16")]; + tensor var_1157_equation_0 = const()[name = tensor("op_1157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1157_cast_fp16 = einsum(equation = var_1157_equation_0, values = (var_963_cast_fp16, var_477_cast_fp16))[name = tensor("op_1157_cast_fp16")]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = var_1157_cast_fp16, y = var_1158_to_fp16)[name = tensor("aw_chunk_25_cast_fp16")]; + tensor var_1161_equation_0 = const()[name = tensor("op_1161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1161_cast_fp16 = einsum(equation = var_1161_equation_0, values = (var_963_cast_fp16, var_484_cast_fp16))[name = tensor("op_1161_cast_fp16")]; + tensor var_1162_to_fp16 = const()[name = tensor("op_1162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = var_1161_cast_fp16, y = var_1162_to_fp16)[name = tensor("aw_chunk_27_cast_fp16")]; + tensor var_1165_equation_0 = const()[name = tensor("op_1165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1165_cast_fp16 = einsum(equation = var_1165_equation_0, values = (var_963_cast_fp16, var_491_cast_fp16))[name = tensor("op_1165_cast_fp16")]; + tensor var_1166_to_fp16 = const()[name = tensor("op_1166_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = var_1165_cast_fp16, y = var_1166_to_fp16)[name = tensor("aw_chunk_29_cast_fp16")]; + tensor var_1169_equation_0 = const()[name = tensor("op_1169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1169_cast_fp16 = einsum(equation = var_1169_equation_0, values = (var_963_cast_fp16, var_498_cast_fp16))[name = tensor("op_1169_cast_fp16")]; + tensor var_1170_to_fp16 = const()[name = tensor("op_1170_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = var_1169_cast_fp16, y = var_1170_to_fp16)[name = tensor("aw_chunk_31_cast_fp16")]; + tensor var_1173_equation_0 = const()[name = tensor("op_1173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1173_cast_fp16 = einsum(equation = var_1173_equation_0, values = (var_967_cast_fp16, var_505_cast_fp16))[name = tensor("op_1173_cast_fp16")]; + tensor var_1174_to_fp16 = const()[name = tensor("op_1174_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = var_1173_cast_fp16, y = var_1174_to_fp16)[name = tensor("aw_chunk_33_cast_fp16")]; + tensor var_1177_equation_0 = const()[name = tensor("op_1177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1177_cast_fp16 = einsum(equation = var_1177_equation_0, values = (var_967_cast_fp16, var_512_cast_fp16))[name = tensor("op_1177_cast_fp16")]; + tensor var_1178_to_fp16 = const()[name = tensor("op_1178_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = var_1177_cast_fp16, y = var_1178_to_fp16)[name = tensor("aw_chunk_35_cast_fp16")]; + tensor var_1181_equation_0 = const()[name = tensor("op_1181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1181_cast_fp16 = einsum(equation = var_1181_equation_0, values = (var_967_cast_fp16, var_519_cast_fp16))[name = tensor("op_1181_cast_fp16")]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = var_1181_cast_fp16, y = var_1182_to_fp16)[name = tensor("aw_chunk_37_cast_fp16")]; + tensor var_1185_equation_0 = const()[name = tensor("op_1185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1185_cast_fp16 = einsum(equation = var_1185_equation_0, values = (var_967_cast_fp16, var_526_cast_fp16))[name = tensor("op_1185_cast_fp16")]; + tensor var_1186_to_fp16 = const()[name = tensor("op_1186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = var_1185_cast_fp16, y = var_1186_to_fp16)[name = tensor("aw_chunk_39_cast_fp16")]; + tensor var_1189_equation_0 = const()[name = tensor("op_1189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1189_cast_fp16 = einsum(equation = var_1189_equation_0, values = (var_971_cast_fp16, var_533_cast_fp16))[name = tensor("op_1189_cast_fp16")]; + tensor var_1190_to_fp16 = const()[name = tensor("op_1190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = var_1189_cast_fp16, y = var_1190_to_fp16)[name = tensor("aw_chunk_41_cast_fp16")]; + tensor var_1193_equation_0 = const()[name = tensor("op_1193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1193_cast_fp16 = einsum(equation = var_1193_equation_0, values = (var_971_cast_fp16, var_540_cast_fp16))[name = tensor("op_1193_cast_fp16")]; + tensor var_1194_to_fp16 = const()[name = tensor("op_1194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = var_1193_cast_fp16, y = var_1194_to_fp16)[name = tensor("aw_chunk_43_cast_fp16")]; + tensor var_1197_equation_0 = const()[name = tensor("op_1197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1197_cast_fp16 = einsum(equation = var_1197_equation_0, values = (var_971_cast_fp16, var_547_cast_fp16))[name = tensor("op_1197_cast_fp16")]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = var_1197_cast_fp16, y = var_1198_to_fp16)[name = tensor("aw_chunk_45_cast_fp16")]; + tensor var_1201_equation_0 = const()[name = tensor("op_1201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1201_cast_fp16 = einsum(equation = var_1201_equation_0, values = (var_971_cast_fp16, var_554_cast_fp16))[name = tensor("op_1201_cast_fp16")]; + tensor var_1202_to_fp16 = const()[name = tensor("op_1202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = var_1201_cast_fp16, y = var_1202_to_fp16)[name = tensor("aw_chunk_47_cast_fp16")]; + tensor var_1205_equation_0 = const()[name = tensor("op_1205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1205_cast_fp16 = einsum(equation = var_1205_equation_0, values = (var_975_cast_fp16, var_561_cast_fp16))[name = tensor("op_1205_cast_fp16")]; + tensor var_1206_to_fp16 = const()[name = tensor("op_1206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = var_1205_cast_fp16, y = var_1206_to_fp16)[name = tensor("aw_chunk_49_cast_fp16")]; + tensor var_1209_equation_0 = const()[name = tensor("op_1209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1209_cast_fp16 = einsum(equation = var_1209_equation_0, values = (var_975_cast_fp16, var_568_cast_fp16))[name = tensor("op_1209_cast_fp16")]; + tensor var_1210_to_fp16 = const()[name = tensor("op_1210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = var_1209_cast_fp16, y = var_1210_to_fp16)[name = tensor("aw_chunk_51_cast_fp16")]; + tensor var_1213_equation_0 = const()[name = tensor("op_1213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1213_cast_fp16 = einsum(equation = var_1213_equation_0, values = (var_975_cast_fp16, var_575_cast_fp16))[name = tensor("op_1213_cast_fp16")]; + tensor var_1214_to_fp16 = const()[name = tensor("op_1214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = var_1213_cast_fp16, y = var_1214_to_fp16)[name = tensor("aw_chunk_53_cast_fp16")]; + tensor var_1217_equation_0 = const()[name = tensor("op_1217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1217_cast_fp16 = einsum(equation = var_1217_equation_0, values = (var_975_cast_fp16, var_582_cast_fp16))[name = tensor("op_1217_cast_fp16")]; + tensor var_1218_to_fp16 = const()[name = tensor("op_1218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = var_1217_cast_fp16, y = var_1218_to_fp16)[name = tensor("aw_chunk_55_cast_fp16")]; + tensor var_1221_equation_0 = const()[name = tensor("op_1221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1221_cast_fp16 = einsum(equation = var_1221_equation_0, values = (var_979_cast_fp16, var_589_cast_fp16))[name = tensor("op_1221_cast_fp16")]; + tensor var_1222_to_fp16 = const()[name = tensor("op_1222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = var_1221_cast_fp16, y = var_1222_to_fp16)[name = tensor("aw_chunk_57_cast_fp16")]; + tensor var_1225_equation_0 = const()[name = tensor("op_1225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1225_cast_fp16 = einsum(equation = var_1225_equation_0, values = (var_979_cast_fp16, var_596_cast_fp16))[name = tensor("op_1225_cast_fp16")]; + tensor var_1226_to_fp16 = const()[name = tensor("op_1226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = var_1225_cast_fp16, y = var_1226_to_fp16)[name = tensor("aw_chunk_59_cast_fp16")]; + tensor var_1229_equation_0 = const()[name = tensor("op_1229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1229_cast_fp16 = einsum(equation = var_1229_equation_0, values = (var_979_cast_fp16, var_603_cast_fp16))[name = tensor("op_1229_cast_fp16")]; + tensor var_1230_to_fp16 = const()[name = tensor("op_1230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = var_1229_cast_fp16, y = var_1230_to_fp16)[name = tensor("aw_chunk_61_cast_fp16")]; + tensor var_1233_equation_0 = const()[name = tensor("op_1233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1233_cast_fp16 = einsum(equation = var_1233_equation_0, values = (var_979_cast_fp16, var_610_cast_fp16))[name = tensor("op_1233_cast_fp16")]; + tensor var_1234_to_fp16 = const()[name = tensor("op_1234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = var_1233_cast_fp16, y = var_1234_to_fp16)[name = tensor("aw_chunk_63_cast_fp16")]; + tensor var_1237_equation_0 = const()[name = tensor("op_1237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1237_cast_fp16 = einsum(equation = var_1237_equation_0, values = (var_983_cast_fp16, var_617_cast_fp16))[name = tensor("op_1237_cast_fp16")]; + tensor var_1238_to_fp16 = const()[name = tensor("op_1238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = var_1237_cast_fp16, y = var_1238_to_fp16)[name = tensor("aw_chunk_65_cast_fp16")]; + tensor var_1241_equation_0 = const()[name = tensor("op_1241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1241_cast_fp16 = einsum(equation = var_1241_equation_0, values = (var_983_cast_fp16, var_624_cast_fp16))[name = tensor("op_1241_cast_fp16")]; + tensor var_1242_to_fp16 = const()[name = tensor("op_1242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = var_1241_cast_fp16, y = var_1242_to_fp16)[name = tensor("aw_chunk_67_cast_fp16")]; + tensor var_1245_equation_0 = const()[name = tensor("op_1245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1245_cast_fp16 = einsum(equation = var_1245_equation_0, values = (var_983_cast_fp16, var_631_cast_fp16))[name = tensor("op_1245_cast_fp16")]; + tensor var_1246_to_fp16 = const()[name = tensor("op_1246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = var_1245_cast_fp16, y = var_1246_to_fp16)[name = tensor("aw_chunk_69_cast_fp16")]; + tensor var_1249_equation_0 = const()[name = tensor("op_1249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1249_cast_fp16 = einsum(equation = var_1249_equation_0, values = (var_983_cast_fp16, var_638_cast_fp16))[name = tensor("op_1249_cast_fp16")]; + tensor var_1250_to_fp16 = const()[name = tensor("op_1250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = var_1249_cast_fp16, y = var_1250_to_fp16)[name = tensor("aw_chunk_71_cast_fp16")]; + tensor var_1253_equation_0 = const()[name = tensor("op_1253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1253_cast_fp16 = einsum(equation = var_1253_equation_0, values = (var_987_cast_fp16, var_645_cast_fp16))[name = tensor("op_1253_cast_fp16")]; + tensor var_1254_to_fp16 = const()[name = tensor("op_1254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = var_1253_cast_fp16, y = var_1254_to_fp16)[name = tensor("aw_chunk_73_cast_fp16")]; + tensor var_1257_equation_0 = const()[name = tensor("op_1257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1257_cast_fp16 = einsum(equation = var_1257_equation_0, values = (var_987_cast_fp16, var_652_cast_fp16))[name = tensor("op_1257_cast_fp16")]; + tensor var_1258_to_fp16 = const()[name = tensor("op_1258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = var_1257_cast_fp16, y = var_1258_to_fp16)[name = tensor("aw_chunk_75_cast_fp16")]; + tensor var_1261_equation_0 = const()[name = tensor("op_1261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1261_cast_fp16 = einsum(equation = var_1261_equation_0, values = (var_987_cast_fp16, var_659_cast_fp16))[name = tensor("op_1261_cast_fp16")]; + tensor var_1262_to_fp16 = const()[name = tensor("op_1262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = var_1261_cast_fp16, y = var_1262_to_fp16)[name = tensor("aw_chunk_77_cast_fp16")]; + tensor var_1265_equation_0 = const()[name = tensor("op_1265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1265_cast_fp16 = einsum(equation = var_1265_equation_0, values = (var_987_cast_fp16, var_666_cast_fp16))[name = tensor("op_1265_cast_fp16")]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = var_1265_cast_fp16, y = var_1266_to_fp16)[name = tensor("aw_chunk_79_cast_fp16")]; + tensor var_1269_equation_0 = const()[name = tensor("op_1269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1269_cast_fp16 = einsum(equation = var_1269_equation_0, values = (var_991_cast_fp16, var_673_cast_fp16))[name = tensor("op_1269_cast_fp16")]; + tensor var_1270_to_fp16 = const()[name = tensor("op_1270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = var_1269_cast_fp16, y = var_1270_to_fp16)[name = tensor("aw_chunk_81_cast_fp16")]; + tensor var_1273_equation_0 = const()[name = tensor("op_1273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1273_cast_fp16 = einsum(equation = var_1273_equation_0, values = (var_991_cast_fp16, var_680_cast_fp16))[name = tensor("op_1273_cast_fp16")]; + tensor var_1274_to_fp16 = const()[name = tensor("op_1274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = var_1273_cast_fp16, y = var_1274_to_fp16)[name = tensor("aw_chunk_83_cast_fp16")]; + tensor var_1277_equation_0 = const()[name = tensor("op_1277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1277_cast_fp16 = einsum(equation = var_1277_equation_0, values = (var_991_cast_fp16, var_687_cast_fp16))[name = tensor("op_1277_cast_fp16")]; + tensor var_1278_to_fp16 = const()[name = tensor("op_1278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = var_1277_cast_fp16, y = var_1278_to_fp16)[name = tensor("aw_chunk_85_cast_fp16")]; + tensor var_1281_equation_0 = const()[name = tensor("op_1281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1281_cast_fp16 = einsum(equation = var_1281_equation_0, values = (var_991_cast_fp16, var_694_cast_fp16))[name = tensor("op_1281_cast_fp16")]; + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = var_1281_cast_fp16, y = var_1282_to_fp16)[name = tensor("aw_chunk_87_cast_fp16")]; + tensor var_1285_equation_0 = const()[name = tensor("op_1285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1285_cast_fp16 = einsum(equation = var_1285_equation_0, values = (var_995_cast_fp16, var_701_cast_fp16))[name = tensor("op_1285_cast_fp16")]; + tensor var_1286_to_fp16 = const()[name = tensor("op_1286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = var_1285_cast_fp16, y = var_1286_to_fp16)[name = tensor("aw_chunk_89_cast_fp16")]; + tensor var_1289_equation_0 = const()[name = tensor("op_1289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1289_cast_fp16 = einsum(equation = var_1289_equation_0, values = (var_995_cast_fp16, var_708_cast_fp16))[name = tensor("op_1289_cast_fp16")]; + tensor var_1290_to_fp16 = const()[name = tensor("op_1290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = var_1289_cast_fp16, y = var_1290_to_fp16)[name = tensor("aw_chunk_91_cast_fp16")]; + tensor var_1293_equation_0 = const()[name = tensor("op_1293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1293_cast_fp16 = einsum(equation = var_1293_equation_0, values = (var_995_cast_fp16, var_715_cast_fp16))[name = tensor("op_1293_cast_fp16")]; + tensor var_1294_to_fp16 = const()[name = tensor("op_1294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = var_1293_cast_fp16, y = var_1294_to_fp16)[name = tensor("aw_chunk_93_cast_fp16")]; + tensor var_1297_equation_0 = const()[name = tensor("op_1297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1297_cast_fp16 = einsum(equation = var_1297_equation_0, values = (var_995_cast_fp16, var_722_cast_fp16))[name = tensor("op_1297_cast_fp16")]; + tensor var_1298_to_fp16 = const()[name = tensor("op_1298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = var_1297_cast_fp16, y = var_1298_to_fp16)[name = tensor("aw_chunk_95_cast_fp16")]; + tensor var_1301_equation_0 = const()[name = tensor("op_1301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1301_cast_fp16 = einsum(equation = var_1301_equation_0, values = (var_999_cast_fp16, var_729_cast_fp16))[name = tensor("op_1301_cast_fp16")]; + tensor var_1302_to_fp16 = const()[name = tensor("op_1302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = var_1301_cast_fp16, y = var_1302_to_fp16)[name = tensor("aw_chunk_97_cast_fp16")]; + tensor var_1305_equation_0 = const()[name = tensor("op_1305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1305_cast_fp16 = einsum(equation = var_1305_equation_0, values = (var_999_cast_fp16, var_736_cast_fp16))[name = tensor("op_1305_cast_fp16")]; + tensor var_1306_to_fp16 = const()[name = tensor("op_1306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = var_1305_cast_fp16, y = var_1306_to_fp16)[name = tensor("aw_chunk_99_cast_fp16")]; + tensor var_1309_equation_0 = const()[name = tensor("op_1309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1309_cast_fp16 = einsum(equation = var_1309_equation_0, values = (var_999_cast_fp16, var_743_cast_fp16))[name = tensor("op_1309_cast_fp16")]; + tensor var_1310_to_fp16 = const()[name = tensor("op_1310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = var_1309_cast_fp16, y = var_1310_to_fp16)[name = tensor("aw_chunk_101_cast_fp16")]; + tensor var_1313_equation_0 = const()[name = tensor("op_1313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1313_cast_fp16 = einsum(equation = var_1313_equation_0, values = (var_999_cast_fp16, var_750_cast_fp16))[name = tensor("op_1313_cast_fp16")]; + tensor var_1314_to_fp16 = const()[name = tensor("op_1314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = var_1313_cast_fp16, y = var_1314_to_fp16)[name = tensor("aw_chunk_103_cast_fp16")]; + tensor var_1317_equation_0 = const()[name = tensor("op_1317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1317_cast_fp16 = einsum(equation = var_1317_equation_0, values = (var_1003_cast_fp16, var_757_cast_fp16))[name = tensor("op_1317_cast_fp16")]; + tensor var_1318_to_fp16 = const()[name = tensor("op_1318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = var_1317_cast_fp16, y = var_1318_to_fp16)[name = tensor("aw_chunk_105_cast_fp16")]; + tensor var_1321_equation_0 = const()[name = tensor("op_1321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1321_cast_fp16 = einsum(equation = var_1321_equation_0, values = (var_1003_cast_fp16, var_764_cast_fp16))[name = tensor("op_1321_cast_fp16")]; + tensor var_1322_to_fp16 = const()[name = tensor("op_1322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = var_1321_cast_fp16, y = var_1322_to_fp16)[name = tensor("aw_chunk_107_cast_fp16")]; + tensor var_1325_equation_0 = const()[name = tensor("op_1325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1325_cast_fp16 = einsum(equation = var_1325_equation_0, values = (var_1003_cast_fp16, var_771_cast_fp16))[name = tensor("op_1325_cast_fp16")]; + tensor var_1326_to_fp16 = const()[name = tensor("op_1326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = var_1325_cast_fp16, y = var_1326_to_fp16)[name = tensor("aw_chunk_109_cast_fp16")]; + tensor var_1329_equation_0 = const()[name = tensor("op_1329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1329_cast_fp16 = einsum(equation = var_1329_equation_0, values = (var_1003_cast_fp16, var_778_cast_fp16))[name = tensor("op_1329_cast_fp16")]; + tensor var_1330_to_fp16 = const()[name = tensor("op_1330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = var_1329_cast_fp16, y = var_1330_to_fp16)[name = tensor("aw_chunk_111_cast_fp16")]; + tensor var_1333_equation_0 = const()[name = tensor("op_1333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1333_cast_fp16 = einsum(equation = var_1333_equation_0, values = (var_1007_cast_fp16, var_785_cast_fp16))[name = tensor("op_1333_cast_fp16")]; + tensor var_1334_to_fp16 = const()[name = tensor("op_1334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = var_1333_cast_fp16, y = var_1334_to_fp16)[name = tensor("aw_chunk_113_cast_fp16")]; + tensor var_1337_equation_0 = const()[name = tensor("op_1337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1337_cast_fp16 = einsum(equation = var_1337_equation_0, values = (var_1007_cast_fp16, var_792_cast_fp16))[name = tensor("op_1337_cast_fp16")]; + tensor var_1338_to_fp16 = const()[name = tensor("op_1338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = var_1337_cast_fp16, y = var_1338_to_fp16)[name = tensor("aw_chunk_115_cast_fp16")]; + tensor var_1341_equation_0 = const()[name = tensor("op_1341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1341_cast_fp16 = einsum(equation = var_1341_equation_0, values = (var_1007_cast_fp16, var_799_cast_fp16))[name = tensor("op_1341_cast_fp16")]; + tensor var_1342_to_fp16 = const()[name = tensor("op_1342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = var_1341_cast_fp16, y = var_1342_to_fp16)[name = tensor("aw_chunk_117_cast_fp16")]; + tensor var_1345_equation_0 = const()[name = tensor("op_1345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1345_cast_fp16 = einsum(equation = var_1345_equation_0, values = (var_1007_cast_fp16, var_806_cast_fp16))[name = tensor("op_1345_cast_fp16")]; + tensor var_1346_to_fp16 = const()[name = tensor("op_1346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = var_1345_cast_fp16, y = var_1346_to_fp16)[name = tensor("aw_chunk_119_cast_fp16")]; + tensor var_1349_equation_0 = const()[name = tensor("op_1349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1349_cast_fp16 = einsum(equation = var_1349_equation_0, values = (var_1011_cast_fp16, var_813_cast_fp16))[name = tensor("op_1349_cast_fp16")]; + tensor var_1350_to_fp16 = const()[name = tensor("op_1350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = var_1349_cast_fp16, y = var_1350_to_fp16)[name = tensor("aw_chunk_121_cast_fp16")]; + tensor var_1353_equation_0 = const()[name = tensor("op_1353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1353_cast_fp16 = einsum(equation = var_1353_equation_0, values = (var_1011_cast_fp16, var_820_cast_fp16))[name = tensor("op_1353_cast_fp16")]; + tensor var_1354_to_fp16 = const()[name = tensor("op_1354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = var_1353_cast_fp16, y = var_1354_to_fp16)[name = tensor("aw_chunk_123_cast_fp16")]; + tensor var_1357_equation_0 = const()[name = tensor("op_1357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1357_cast_fp16 = einsum(equation = var_1357_equation_0, values = (var_1011_cast_fp16, var_827_cast_fp16))[name = tensor("op_1357_cast_fp16")]; + tensor var_1358_to_fp16 = const()[name = tensor("op_1358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = var_1357_cast_fp16, y = var_1358_to_fp16)[name = tensor("aw_chunk_125_cast_fp16")]; + tensor var_1361_equation_0 = const()[name = tensor("op_1361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1361_cast_fp16 = einsum(equation = var_1361_equation_0, values = (var_1011_cast_fp16, var_834_cast_fp16))[name = tensor("op_1361_cast_fp16")]; + tensor var_1362_to_fp16 = const()[name = tensor("op_1362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = var_1361_cast_fp16, y = var_1362_to_fp16)[name = tensor("aw_chunk_127_cast_fp16")]; + tensor var_1365_equation_0 = const()[name = tensor("op_1365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1365_cast_fp16 = einsum(equation = var_1365_equation_0, values = (var_1015_cast_fp16, var_841_cast_fp16))[name = tensor("op_1365_cast_fp16")]; + tensor var_1366_to_fp16 = const()[name = tensor("op_1366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = var_1365_cast_fp16, y = var_1366_to_fp16)[name = tensor("aw_chunk_129_cast_fp16")]; + tensor var_1369_equation_0 = const()[name = tensor("op_1369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1369_cast_fp16 = einsum(equation = var_1369_equation_0, values = (var_1015_cast_fp16, var_848_cast_fp16))[name = tensor("op_1369_cast_fp16")]; + tensor var_1370_to_fp16 = const()[name = tensor("op_1370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = var_1369_cast_fp16, y = var_1370_to_fp16)[name = tensor("aw_chunk_131_cast_fp16")]; + tensor var_1373_equation_0 = const()[name = tensor("op_1373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1373_cast_fp16 = einsum(equation = var_1373_equation_0, values = (var_1015_cast_fp16, var_855_cast_fp16))[name = tensor("op_1373_cast_fp16")]; + tensor var_1374_to_fp16 = const()[name = tensor("op_1374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = var_1373_cast_fp16, y = var_1374_to_fp16)[name = tensor("aw_chunk_133_cast_fp16")]; + tensor var_1377_equation_0 = const()[name = tensor("op_1377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1377_cast_fp16 = einsum(equation = var_1377_equation_0, values = (var_1015_cast_fp16, var_862_cast_fp16))[name = tensor("op_1377_cast_fp16")]; + tensor var_1378_to_fp16 = const()[name = tensor("op_1378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = var_1377_cast_fp16, y = var_1378_to_fp16)[name = tensor("aw_chunk_135_cast_fp16")]; + tensor var_1381_equation_0 = const()[name = tensor("op_1381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1381_cast_fp16 = einsum(equation = var_1381_equation_0, values = (var_1019_cast_fp16, var_869_cast_fp16))[name = tensor("op_1381_cast_fp16")]; + tensor var_1382_to_fp16 = const()[name = tensor("op_1382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = var_1381_cast_fp16, y = var_1382_to_fp16)[name = tensor("aw_chunk_137_cast_fp16")]; + tensor var_1385_equation_0 = const()[name = tensor("op_1385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1385_cast_fp16 = einsum(equation = var_1385_equation_0, values = (var_1019_cast_fp16, var_876_cast_fp16))[name = tensor("op_1385_cast_fp16")]; + tensor var_1386_to_fp16 = const()[name = tensor("op_1386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = var_1385_cast_fp16, y = var_1386_to_fp16)[name = tensor("aw_chunk_139_cast_fp16")]; + tensor var_1389_equation_0 = const()[name = tensor("op_1389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1389_cast_fp16 = einsum(equation = var_1389_equation_0, values = (var_1019_cast_fp16, var_883_cast_fp16))[name = tensor("op_1389_cast_fp16")]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = var_1389_cast_fp16, y = var_1390_to_fp16)[name = tensor("aw_chunk_141_cast_fp16")]; + tensor var_1393_equation_0 = const()[name = tensor("op_1393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1393_cast_fp16 = einsum(equation = var_1393_equation_0, values = (var_1019_cast_fp16, var_890_cast_fp16))[name = tensor("op_1393_cast_fp16")]; + tensor var_1394_to_fp16 = const()[name = tensor("op_1394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = var_1393_cast_fp16, y = var_1394_to_fp16)[name = tensor("aw_chunk_143_cast_fp16")]; + tensor var_1397_equation_0 = const()[name = tensor("op_1397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1397_cast_fp16 = einsum(equation = var_1397_equation_0, values = (var_1023_cast_fp16, var_897_cast_fp16))[name = tensor("op_1397_cast_fp16")]; + tensor var_1398_to_fp16 = const()[name = tensor("op_1398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = var_1397_cast_fp16, y = var_1398_to_fp16)[name = tensor("aw_chunk_145_cast_fp16")]; + tensor var_1401_equation_0 = const()[name = tensor("op_1401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1401_cast_fp16 = einsum(equation = var_1401_equation_0, values = (var_1023_cast_fp16, var_904_cast_fp16))[name = tensor("op_1401_cast_fp16")]; + tensor var_1402_to_fp16 = const()[name = tensor("op_1402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = var_1401_cast_fp16, y = var_1402_to_fp16)[name = tensor("aw_chunk_147_cast_fp16")]; + tensor var_1405_equation_0 = const()[name = tensor("op_1405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1405_cast_fp16 = einsum(equation = var_1405_equation_0, values = (var_1023_cast_fp16, var_911_cast_fp16))[name = tensor("op_1405_cast_fp16")]; + tensor var_1406_to_fp16 = const()[name = tensor("op_1406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = var_1405_cast_fp16, y = var_1406_to_fp16)[name = tensor("aw_chunk_149_cast_fp16")]; + tensor var_1409_equation_0 = const()[name = tensor("op_1409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1409_cast_fp16 = einsum(equation = var_1409_equation_0, values = (var_1023_cast_fp16, var_918_cast_fp16))[name = tensor("op_1409_cast_fp16")]; + tensor var_1410_to_fp16 = const()[name = tensor("op_1410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = var_1409_cast_fp16, y = var_1410_to_fp16)[name = tensor("aw_chunk_151_cast_fp16")]; + tensor var_1413_equation_0 = const()[name = tensor("op_1413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1413_cast_fp16 = einsum(equation = var_1413_equation_0, values = (var_1027_cast_fp16, var_925_cast_fp16))[name = tensor("op_1413_cast_fp16")]; + tensor var_1414_to_fp16 = const()[name = tensor("op_1414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = var_1413_cast_fp16, y = var_1414_to_fp16)[name = tensor("aw_chunk_153_cast_fp16")]; + tensor var_1417_equation_0 = const()[name = tensor("op_1417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1417_cast_fp16 = einsum(equation = var_1417_equation_0, values = (var_1027_cast_fp16, var_932_cast_fp16))[name = tensor("op_1417_cast_fp16")]; + tensor var_1418_to_fp16 = const()[name = tensor("op_1418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = var_1417_cast_fp16, y = var_1418_to_fp16)[name = tensor("aw_chunk_155_cast_fp16")]; + tensor var_1421_equation_0 = const()[name = tensor("op_1421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1421_cast_fp16 = einsum(equation = var_1421_equation_0, values = (var_1027_cast_fp16, var_939_cast_fp16))[name = tensor("op_1421_cast_fp16")]; + tensor var_1422_to_fp16 = const()[name = tensor("op_1422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = var_1421_cast_fp16, y = var_1422_to_fp16)[name = tensor("aw_chunk_157_cast_fp16")]; + tensor var_1425_equation_0 = const()[name = tensor("op_1425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_1425_cast_fp16 = einsum(equation = var_1425_equation_0, values = (var_1027_cast_fp16, var_946_cast_fp16))[name = tensor("op_1425_cast_fp16")]; + tensor var_1426_to_fp16 = const()[name = tensor("op_1426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = var_1425_cast_fp16, y = var_1426_to_fp16)[name = tensor("aw_chunk_159_cast_fp16")]; + tensor var_1428_cast_fp16 = softmax(axis = var_199, x = aw_chunk_1_cast_fp16)[name = tensor("op_1428_cast_fp16")]; + tensor var_1429_cast_fp16 = softmax(axis = var_199, x = aw_chunk_3_cast_fp16)[name = tensor("op_1429_cast_fp16")]; + tensor var_1430_cast_fp16 = softmax(axis = var_199, x = aw_chunk_5_cast_fp16)[name = tensor("op_1430_cast_fp16")]; + tensor var_1431_cast_fp16 = softmax(axis = var_199, x = aw_chunk_7_cast_fp16)[name = tensor("op_1431_cast_fp16")]; + tensor var_1432_cast_fp16 = softmax(axis = var_199, x = aw_chunk_9_cast_fp16)[name = tensor("op_1432_cast_fp16")]; + tensor var_1433_cast_fp16 = softmax(axis = var_199, x = aw_chunk_11_cast_fp16)[name = tensor("op_1433_cast_fp16")]; + tensor var_1434_cast_fp16 = softmax(axis = var_199, x = aw_chunk_13_cast_fp16)[name = tensor("op_1434_cast_fp16")]; + tensor var_1435_cast_fp16 = softmax(axis = var_199, x = aw_chunk_15_cast_fp16)[name = tensor("op_1435_cast_fp16")]; + tensor var_1436_cast_fp16 = softmax(axis = var_199, x = aw_chunk_17_cast_fp16)[name = tensor("op_1436_cast_fp16")]; + tensor var_1437_cast_fp16 = softmax(axis = var_199, x = aw_chunk_19_cast_fp16)[name = tensor("op_1437_cast_fp16")]; + tensor var_1438_cast_fp16 = softmax(axis = var_199, x = aw_chunk_21_cast_fp16)[name = tensor("op_1438_cast_fp16")]; + tensor var_1439_cast_fp16 = softmax(axis = var_199, x = aw_chunk_23_cast_fp16)[name = tensor("op_1439_cast_fp16")]; + tensor var_1440_cast_fp16 = softmax(axis = var_199, x = aw_chunk_25_cast_fp16)[name = tensor("op_1440_cast_fp16")]; + tensor var_1441_cast_fp16 = softmax(axis = var_199, x = aw_chunk_27_cast_fp16)[name = tensor("op_1441_cast_fp16")]; + tensor var_1442_cast_fp16 = softmax(axis = var_199, x = aw_chunk_29_cast_fp16)[name = tensor("op_1442_cast_fp16")]; + tensor var_1443_cast_fp16 = softmax(axis = var_199, x = aw_chunk_31_cast_fp16)[name = tensor("op_1443_cast_fp16")]; + tensor var_1444_cast_fp16 = softmax(axis = var_199, x = aw_chunk_33_cast_fp16)[name = tensor("op_1444_cast_fp16")]; + tensor var_1445_cast_fp16 = softmax(axis = var_199, x = aw_chunk_35_cast_fp16)[name = tensor("op_1445_cast_fp16")]; + tensor var_1446_cast_fp16 = softmax(axis = var_199, x = aw_chunk_37_cast_fp16)[name = tensor("op_1446_cast_fp16")]; + tensor var_1447_cast_fp16 = softmax(axis = var_199, x = aw_chunk_39_cast_fp16)[name = tensor("op_1447_cast_fp16")]; + tensor var_1448_cast_fp16 = softmax(axis = var_199, x = aw_chunk_41_cast_fp16)[name = tensor("op_1448_cast_fp16")]; + tensor var_1449_cast_fp16 = softmax(axis = var_199, x = aw_chunk_43_cast_fp16)[name = tensor("op_1449_cast_fp16")]; + tensor var_1450_cast_fp16 = softmax(axis = var_199, x = aw_chunk_45_cast_fp16)[name = tensor("op_1450_cast_fp16")]; + tensor var_1451_cast_fp16 = softmax(axis = var_199, x = aw_chunk_47_cast_fp16)[name = tensor("op_1451_cast_fp16")]; + tensor var_1452_cast_fp16 = softmax(axis = var_199, x = aw_chunk_49_cast_fp16)[name = tensor("op_1452_cast_fp16")]; + tensor var_1453_cast_fp16 = softmax(axis = var_199, x = aw_chunk_51_cast_fp16)[name = tensor("op_1453_cast_fp16")]; + tensor var_1454_cast_fp16 = softmax(axis = var_199, x = aw_chunk_53_cast_fp16)[name = tensor("op_1454_cast_fp16")]; + tensor var_1455_cast_fp16 = softmax(axis = var_199, x = aw_chunk_55_cast_fp16)[name = tensor("op_1455_cast_fp16")]; + tensor var_1456_cast_fp16 = softmax(axis = var_199, x = aw_chunk_57_cast_fp16)[name = tensor("op_1456_cast_fp16")]; + tensor var_1457_cast_fp16 = softmax(axis = var_199, x = aw_chunk_59_cast_fp16)[name = tensor("op_1457_cast_fp16")]; + tensor var_1458_cast_fp16 = softmax(axis = var_199, x = aw_chunk_61_cast_fp16)[name = tensor("op_1458_cast_fp16")]; + tensor var_1459_cast_fp16 = softmax(axis = var_199, x = aw_chunk_63_cast_fp16)[name = tensor("op_1459_cast_fp16")]; + tensor var_1460_cast_fp16 = softmax(axis = var_199, x = aw_chunk_65_cast_fp16)[name = tensor("op_1460_cast_fp16")]; + tensor var_1461_cast_fp16 = softmax(axis = var_199, x = aw_chunk_67_cast_fp16)[name = tensor("op_1461_cast_fp16")]; + tensor var_1462_cast_fp16 = softmax(axis = var_199, x = aw_chunk_69_cast_fp16)[name = tensor("op_1462_cast_fp16")]; + tensor var_1463_cast_fp16 = softmax(axis = var_199, x = aw_chunk_71_cast_fp16)[name = tensor("op_1463_cast_fp16")]; + tensor var_1464_cast_fp16 = softmax(axis = var_199, x = aw_chunk_73_cast_fp16)[name = tensor("op_1464_cast_fp16")]; + tensor var_1465_cast_fp16 = softmax(axis = var_199, x = aw_chunk_75_cast_fp16)[name = tensor("op_1465_cast_fp16")]; + tensor var_1466_cast_fp16 = softmax(axis = var_199, x = aw_chunk_77_cast_fp16)[name = tensor("op_1466_cast_fp16")]; + tensor var_1467_cast_fp16 = softmax(axis = var_199, x = aw_chunk_79_cast_fp16)[name = tensor("op_1467_cast_fp16")]; + tensor var_1468_cast_fp16 = softmax(axis = var_199, x = aw_chunk_81_cast_fp16)[name = tensor("op_1468_cast_fp16")]; + tensor var_1469_cast_fp16 = softmax(axis = var_199, x = aw_chunk_83_cast_fp16)[name = tensor("op_1469_cast_fp16")]; + tensor var_1470_cast_fp16 = softmax(axis = var_199, x = aw_chunk_85_cast_fp16)[name = tensor("op_1470_cast_fp16")]; + tensor var_1471_cast_fp16 = softmax(axis = var_199, x = aw_chunk_87_cast_fp16)[name = tensor("op_1471_cast_fp16")]; + tensor var_1472_cast_fp16 = softmax(axis = var_199, x = aw_chunk_89_cast_fp16)[name = tensor("op_1472_cast_fp16")]; + tensor var_1473_cast_fp16 = softmax(axis = var_199, x = aw_chunk_91_cast_fp16)[name = tensor("op_1473_cast_fp16")]; + tensor var_1474_cast_fp16 = softmax(axis = var_199, x = aw_chunk_93_cast_fp16)[name = tensor("op_1474_cast_fp16")]; + tensor var_1475_cast_fp16 = softmax(axis = var_199, x = aw_chunk_95_cast_fp16)[name = tensor("op_1475_cast_fp16")]; + tensor var_1476_cast_fp16 = softmax(axis = var_199, x = aw_chunk_97_cast_fp16)[name = tensor("op_1476_cast_fp16")]; + tensor var_1477_cast_fp16 = softmax(axis = var_199, x = aw_chunk_99_cast_fp16)[name = tensor("op_1477_cast_fp16")]; + tensor var_1478_cast_fp16 = softmax(axis = var_199, x = aw_chunk_101_cast_fp16)[name = tensor("op_1478_cast_fp16")]; + tensor var_1479_cast_fp16 = softmax(axis = var_199, x = aw_chunk_103_cast_fp16)[name = tensor("op_1479_cast_fp16")]; + tensor var_1480_cast_fp16 = softmax(axis = var_199, x = aw_chunk_105_cast_fp16)[name = tensor("op_1480_cast_fp16")]; + tensor var_1481_cast_fp16 = softmax(axis = var_199, x = aw_chunk_107_cast_fp16)[name = tensor("op_1481_cast_fp16")]; + tensor var_1482_cast_fp16 = softmax(axis = var_199, x = aw_chunk_109_cast_fp16)[name = tensor("op_1482_cast_fp16")]; + tensor var_1483_cast_fp16 = softmax(axis = var_199, x = aw_chunk_111_cast_fp16)[name = tensor("op_1483_cast_fp16")]; + tensor var_1484_cast_fp16 = softmax(axis = var_199, x = aw_chunk_113_cast_fp16)[name = tensor("op_1484_cast_fp16")]; + tensor var_1485_cast_fp16 = softmax(axis = var_199, x = aw_chunk_115_cast_fp16)[name = tensor("op_1485_cast_fp16")]; + tensor var_1486_cast_fp16 = softmax(axis = var_199, x = aw_chunk_117_cast_fp16)[name = tensor("op_1486_cast_fp16")]; + tensor var_1487_cast_fp16 = softmax(axis = var_199, x = aw_chunk_119_cast_fp16)[name = tensor("op_1487_cast_fp16")]; + tensor var_1488_cast_fp16 = softmax(axis = var_199, x = aw_chunk_121_cast_fp16)[name = tensor("op_1488_cast_fp16")]; + tensor var_1489_cast_fp16 = softmax(axis = var_199, x = aw_chunk_123_cast_fp16)[name = tensor("op_1489_cast_fp16")]; + tensor var_1490_cast_fp16 = softmax(axis = var_199, x = aw_chunk_125_cast_fp16)[name = tensor("op_1490_cast_fp16")]; + tensor var_1491_cast_fp16 = softmax(axis = var_199, x = aw_chunk_127_cast_fp16)[name = tensor("op_1491_cast_fp16")]; + tensor var_1492_cast_fp16 = softmax(axis = var_199, x = aw_chunk_129_cast_fp16)[name = tensor("op_1492_cast_fp16")]; + tensor var_1493_cast_fp16 = softmax(axis = var_199, x = aw_chunk_131_cast_fp16)[name = tensor("op_1493_cast_fp16")]; + tensor var_1494_cast_fp16 = softmax(axis = var_199, x = aw_chunk_133_cast_fp16)[name = tensor("op_1494_cast_fp16")]; + tensor var_1495_cast_fp16 = softmax(axis = var_199, x = aw_chunk_135_cast_fp16)[name = tensor("op_1495_cast_fp16")]; + tensor var_1496_cast_fp16 = softmax(axis = var_199, x = aw_chunk_137_cast_fp16)[name = tensor("op_1496_cast_fp16")]; + tensor var_1497_cast_fp16 = softmax(axis = var_199, x = aw_chunk_139_cast_fp16)[name = tensor("op_1497_cast_fp16")]; + tensor var_1498_cast_fp16 = softmax(axis = var_199, x = aw_chunk_141_cast_fp16)[name = tensor("op_1498_cast_fp16")]; + tensor var_1499_cast_fp16 = softmax(axis = var_199, x = aw_chunk_143_cast_fp16)[name = tensor("op_1499_cast_fp16")]; + tensor var_1500_cast_fp16 = softmax(axis = var_199, x = aw_chunk_145_cast_fp16)[name = tensor("op_1500_cast_fp16")]; + tensor var_1501_cast_fp16 = softmax(axis = var_199, x = aw_chunk_147_cast_fp16)[name = tensor("op_1501_cast_fp16")]; + tensor var_1502_cast_fp16 = softmax(axis = var_199, x = aw_chunk_149_cast_fp16)[name = tensor("op_1502_cast_fp16")]; + tensor var_1503_cast_fp16 = softmax(axis = var_199, x = aw_chunk_151_cast_fp16)[name = tensor("op_1503_cast_fp16")]; + tensor var_1504_cast_fp16 = softmax(axis = var_199, x = aw_chunk_153_cast_fp16)[name = tensor("op_1504_cast_fp16")]; + tensor var_1505_cast_fp16 = softmax(axis = var_199, x = aw_chunk_155_cast_fp16)[name = tensor("op_1505_cast_fp16")]; + tensor var_1506_cast_fp16 = softmax(axis = var_199, x = aw_chunk_157_cast_fp16)[name = tensor("op_1506_cast_fp16")]; + tensor var_1507_cast_fp16 = softmax(axis = var_199, x = aw_chunk_159_cast_fp16)[name = tensor("op_1507_cast_fp16")]; + tensor var_1509_equation_0 = const()[name = tensor("op_1509_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1509_cast_fp16 = einsum(equation = var_1509_equation_0, values = (var_1029_cast_fp16, var_1428_cast_fp16))[name = tensor("op_1509_cast_fp16")]; + tensor var_1511_equation_0 = const()[name = tensor("op_1511_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1511_cast_fp16 = einsum(equation = var_1511_equation_0, values = (var_1029_cast_fp16, var_1429_cast_fp16))[name = tensor("op_1511_cast_fp16")]; + tensor var_1513_equation_0 = const()[name = tensor("op_1513_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1513_cast_fp16 = einsum(equation = var_1513_equation_0, values = (var_1029_cast_fp16, var_1430_cast_fp16))[name = tensor("op_1513_cast_fp16")]; + tensor var_1515_equation_0 = const()[name = tensor("op_1515_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1515_cast_fp16 = einsum(equation = var_1515_equation_0, values = (var_1029_cast_fp16, var_1431_cast_fp16))[name = tensor("op_1515_cast_fp16")]; + tensor var_1517_equation_0 = const()[name = tensor("op_1517_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1517_cast_fp16 = einsum(equation = var_1517_equation_0, values = (var_1033_cast_fp16, var_1432_cast_fp16))[name = tensor("op_1517_cast_fp16")]; + tensor var_1519_equation_0 = const()[name = tensor("op_1519_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1519_cast_fp16 = einsum(equation = var_1519_equation_0, values = (var_1033_cast_fp16, var_1433_cast_fp16))[name = tensor("op_1519_cast_fp16")]; + tensor var_1521_equation_0 = const()[name = tensor("op_1521_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1521_cast_fp16 = einsum(equation = var_1521_equation_0, values = (var_1033_cast_fp16, var_1434_cast_fp16))[name = tensor("op_1521_cast_fp16")]; + tensor var_1523_equation_0 = const()[name = tensor("op_1523_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1523_cast_fp16 = einsum(equation = var_1523_equation_0, values = (var_1033_cast_fp16, var_1435_cast_fp16))[name = tensor("op_1523_cast_fp16")]; + tensor var_1525_equation_0 = const()[name = tensor("op_1525_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1525_cast_fp16 = einsum(equation = var_1525_equation_0, values = (var_1037_cast_fp16, var_1436_cast_fp16))[name = tensor("op_1525_cast_fp16")]; + tensor var_1527_equation_0 = const()[name = tensor("op_1527_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1527_cast_fp16 = einsum(equation = var_1527_equation_0, values = (var_1037_cast_fp16, var_1437_cast_fp16))[name = tensor("op_1527_cast_fp16")]; + tensor var_1529_equation_0 = const()[name = tensor("op_1529_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1529_cast_fp16 = einsum(equation = var_1529_equation_0, values = (var_1037_cast_fp16, var_1438_cast_fp16))[name = tensor("op_1529_cast_fp16")]; + tensor var_1531_equation_0 = const()[name = tensor("op_1531_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1531_cast_fp16 = einsum(equation = var_1531_equation_0, values = (var_1037_cast_fp16, var_1439_cast_fp16))[name = tensor("op_1531_cast_fp16")]; + tensor var_1533_equation_0 = const()[name = tensor("op_1533_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1533_cast_fp16 = einsum(equation = var_1533_equation_0, values = (var_1041_cast_fp16, var_1440_cast_fp16))[name = tensor("op_1533_cast_fp16")]; + tensor var_1535_equation_0 = const()[name = tensor("op_1535_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1535_cast_fp16 = einsum(equation = var_1535_equation_0, values = (var_1041_cast_fp16, var_1441_cast_fp16))[name = tensor("op_1535_cast_fp16")]; + tensor var_1537_equation_0 = const()[name = tensor("op_1537_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1537_cast_fp16 = einsum(equation = var_1537_equation_0, values = (var_1041_cast_fp16, var_1442_cast_fp16))[name = tensor("op_1537_cast_fp16")]; + tensor var_1539_equation_0 = const()[name = tensor("op_1539_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1539_cast_fp16 = einsum(equation = var_1539_equation_0, values = (var_1041_cast_fp16, var_1443_cast_fp16))[name = tensor("op_1539_cast_fp16")]; + tensor var_1541_equation_0 = const()[name = tensor("op_1541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1541_cast_fp16 = einsum(equation = var_1541_equation_0, values = (var_1045_cast_fp16, var_1444_cast_fp16))[name = tensor("op_1541_cast_fp16")]; + tensor var_1543_equation_0 = const()[name = tensor("op_1543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1543_cast_fp16 = einsum(equation = var_1543_equation_0, values = (var_1045_cast_fp16, var_1445_cast_fp16))[name = tensor("op_1543_cast_fp16")]; + tensor var_1545_equation_0 = const()[name = tensor("op_1545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1545_cast_fp16 = einsum(equation = var_1545_equation_0, values = (var_1045_cast_fp16, var_1446_cast_fp16))[name = tensor("op_1545_cast_fp16")]; + tensor var_1547_equation_0 = const()[name = tensor("op_1547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1547_cast_fp16 = einsum(equation = var_1547_equation_0, values = (var_1045_cast_fp16, var_1447_cast_fp16))[name = tensor("op_1547_cast_fp16")]; + tensor var_1549_equation_0 = const()[name = tensor("op_1549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1549_cast_fp16 = einsum(equation = var_1549_equation_0, values = (var_1049_cast_fp16, var_1448_cast_fp16))[name = tensor("op_1549_cast_fp16")]; + tensor var_1551_equation_0 = const()[name = tensor("op_1551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1551_cast_fp16 = einsum(equation = var_1551_equation_0, values = (var_1049_cast_fp16, var_1449_cast_fp16))[name = tensor("op_1551_cast_fp16")]; + tensor var_1553_equation_0 = const()[name = tensor("op_1553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1553_cast_fp16 = einsum(equation = var_1553_equation_0, values = (var_1049_cast_fp16, var_1450_cast_fp16))[name = tensor("op_1553_cast_fp16")]; + tensor var_1555_equation_0 = const()[name = tensor("op_1555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1555_cast_fp16 = einsum(equation = var_1555_equation_0, values = (var_1049_cast_fp16, var_1451_cast_fp16))[name = tensor("op_1555_cast_fp16")]; + tensor var_1557_equation_0 = const()[name = tensor("op_1557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1557_cast_fp16 = einsum(equation = var_1557_equation_0, values = (var_1053_cast_fp16, var_1452_cast_fp16))[name = tensor("op_1557_cast_fp16")]; + tensor var_1559_equation_0 = const()[name = tensor("op_1559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1559_cast_fp16 = einsum(equation = var_1559_equation_0, values = (var_1053_cast_fp16, var_1453_cast_fp16))[name = tensor("op_1559_cast_fp16")]; + tensor var_1561_equation_0 = const()[name = tensor("op_1561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1561_cast_fp16 = einsum(equation = var_1561_equation_0, values = (var_1053_cast_fp16, var_1454_cast_fp16))[name = tensor("op_1561_cast_fp16")]; + tensor var_1563_equation_0 = const()[name = tensor("op_1563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1563_cast_fp16 = einsum(equation = var_1563_equation_0, values = (var_1053_cast_fp16, var_1455_cast_fp16))[name = tensor("op_1563_cast_fp16")]; + tensor var_1565_equation_0 = const()[name = tensor("op_1565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1565_cast_fp16 = einsum(equation = var_1565_equation_0, values = (var_1057_cast_fp16, var_1456_cast_fp16))[name = tensor("op_1565_cast_fp16")]; + tensor var_1567_equation_0 = const()[name = tensor("op_1567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1567_cast_fp16 = einsum(equation = var_1567_equation_0, values = (var_1057_cast_fp16, var_1457_cast_fp16))[name = tensor("op_1567_cast_fp16")]; + tensor var_1569_equation_0 = const()[name = tensor("op_1569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1569_cast_fp16 = einsum(equation = var_1569_equation_0, values = (var_1057_cast_fp16, var_1458_cast_fp16))[name = tensor("op_1569_cast_fp16")]; + tensor var_1571_equation_0 = const()[name = tensor("op_1571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1571_cast_fp16 = einsum(equation = var_1571_equation_0, values = (var_1057_cast_fp16, var_1459_cast_fp16))[name = tensor("op_1571_cast_fp16")]; + tensor var_1573_equation_0 = const()[name = tensor("op_1573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1573_cast_fp16 = einsum(equation = var_1573_equation_0, values = (var_1061_cast_fp16, var_1460_cast_fp16))[name = tensor("op_1573_cast_fp16")]; + tensor var_1575_equation_0 = const()[name = tensor("op_1575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1575_cast_fp16 = einsum(equation = var_1575_equation_0, values = (var_1061_cast_fp16, var_1461_cast_fp16))[name = tensor("op_1575_cast_fp16")]; + tensor var_1577_equation_0 = const()[name = tensor("op_1577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1577_cast_fp16 = einsum(equation = var_1577_equation_0, values = (var_1061_cast_fp16, var_1462_cast_fp16))[name = tensor("op_1577_cast_fp16")]; + tensor var_1579_equation_0 = const()[name = tensor("op_1579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1579_cast_fp16 = einsum(equation = var_1579_equation_0, values = (var_1061_cast_fp16, var_1463_cast_fp16))[name = tensor("op_1579_cast_fp16")]; + tensor var_1581_equation_0 = const()[name = tensor("op_1581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1581_cast_fp16 = einsum(equation = var_1581_equation_0, values = (var_1065_cast_fp16, var_1464_cast_fp16))[name = tensor("op_1581_cast_fp16")]; + tensor var_1583_equation_0 = const()[name = tensor("op_1583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1583_cast_fp16 = einsum(equation = var_1583_equation_0, values = (var_1065_cast_fp16, var_1465_cast_fp16))[name = tensor("op_1583_cast_fp16")]; + tensor var_1585_equation_0 = const()[name = tensor("op_1585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1585_cast_fp16 = einsum(equation = var_1585_equation_0, values = (var_1065_cast_fp16, var_1466_cast_fp16))[name = tensor("op_1585_cast_fp16")]; + tensor var_1587_equation_0 = const()[name = tensor("op_1587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1587_cast_fp16 = einsum(equation = var_1587_equation_0, values = (var_1065_cast_fp16, var_1467_cast_fp16))[name = tensor("op_1587_cast_fp16")]; + tensor var_1589_equation_0 = const()[name = tensor("op_1589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1589_cast_fp16 = einsum(equation = var_1589_equation_0, values = (var_1069_cast_fp16, var_1468_cast_fp16))[name = tensor("op_1589_cast_fp16")]; + tensor var_1591_equation_0 = const()[name = tensor("op_1591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1591_cast_fp16 = einsum(equation = var_1591_equation_0, values = (var_1069_cast_fp16, var_1469_cast_fp16))[name = tensor("op_1591_cast_fp16")]; + tensor var_1593_equation_0 = const()[name = tensor("op_1593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1593_cast_fp16 = einsum(equation = var_1593_equation_0, values = (var_1069_cast_fp16, var_1470_cast_fp16))[name = tensor("op_1593_cast_fp16")]; + tensor var_1595_equation_0 = const()[name = tensor("op_1595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1595_cast_fp16 = einsum(equation = var_1595_equation_0, values = (var_1069_cast_fp16, var_1471_cast_fp16))[name = tensor("op_1595_cast_fp16")]; + tensor var_1597_equation_0 = const()[name = tensor("op_1597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1597_cast_fp16 = einsum(equation = var_1597_equation_0, values = (var_1073_cast_fp16, var_1472_cast_fp16))[name = tensor("op_1597_cast_fp16")]; + tensor var_1599_equation_0 = const()[name = tensor("op_1599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1599_cast_fp16 = einsum(equation = var_1599_equation_0, values = (var_1073_cast_fp16, var_1473_cast_fp16))[name = tensor("op_1599_cast_fp16")]; + tensor var_1601_equation_0 = const()[name = tensor("op_1601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1601_cast_fp16 = einsum(equation = var_1601_equation_0, values = (var_1073_cast_fp16, var_1474_cast_fp16))[name = tensor("op_1601_cast_fp16")]; + tensor var_1603_equation_0 = const()[name = tensor("op_1603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1603_cast_fp16 = einsum(equation = var_1603_equation_0, values = (var_1073_cast_fp16, var_1475_cast_fp16))[name = tensor("op_1603_cast_fp16")]; + tensor var_1605_equation_0 = const()[name = tensor("op_1605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1605_cast_fp16 = einsum(equation = var_1605_equation_0, values = (var_1077_cast_fp16, var_1476_cast_fp16))[name = tensor("op_1605_cast_fp16")]; + tensor var_1607_equation_0 = const()[name = tensor("op_1607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1607_cast_fp16 = einsum(equation = var_1607_equation_0, values = (var_1077_cast_fp16, var_1477_cast_fp16))[name = tensor("op_1607_cast_fp16")]; + tensor var_1609_equation_0 = const()[name = tensor("op_1609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1609_cast_fp16 = einsum(equation = var_1609_equation_0, values = (var_1077_cast_fp16, var_1478_cast_fp16))[name = tensor("op_1609_cast_fp16")]; + tensor var_1611_equation_0 = const()[name = tensor("op_1611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1611_cast_fp16 = einsum(equation = var_1611_equation_0, values = (var_1077_cast_fp16, var_1479_cast_fp16))[name = tensor("op_1611_cast_fp16")]; + tensor var_1613_equation_0 = const()[name = tensor("op_1613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1613_cast_fp16 = einsum(equation = var_1613_equation_0, values = (var_1081_cast_fp16, var_1480_cast_fp16))[name = tensor("op_1613_cast_fp16")]; + tensor var_1615_equation_0 = const()[name = tensor("op_1615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1615_cast_fp16 = einsum(equation = var_1615_equation_0, values = (var_1081_cast_fp16, var_1481_cast_fp16))[name = tensor("op_1615_cast_fp16")]; + tensor var_1617_equation_0 = const()[name = tensor("op_1617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1617_cast_fp16 = einsum(equation = var_1617_equation_0, values = (var_1081_cast_fp16, var_1482_cast_fp16))[name = tensor("op_1617_cast_fp16")]; + tensor var_1619_equation_0 = const()[name = tensor("op_1619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1619_cast_fp16 = einsum(equation = var_1619_equation_0, values = (var_1081_cast_fp16, var_1483_cast_fp16))[name = tensor("op_1619_cast_fp16")]; + tensor var_1621_equation_0 = const()[name = tensor("op_1621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1621_cast_fp16 = einsum(equation = var_1621_equation_0, values = (var_1085_cast_fp16, var_1484_cast_fp16))[name = tensor("op_1621_cast_fp16")]; + tensor var_1623_equation_0 = const()[name = tensor("op_1623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1623_cast_fp16 = einsum(equation = var_1623_equation_0, values = (var_1085_cast_fp16, var_1485_cast_fp16))[name = tensor("op_1623_cast_fp16")]; + tensor var_1625_equation_0 = const()[name = tensor("op_1625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1625_cast_fp16 = einsum(equation = var_1625_equation_0, values = (var_1085_cast_fp16, var_1486_cast_fp16))[name = tensor("op_1625_cast_fp16")]; + tensor var_1627_equation_0 = const()[name = tensor("op_1627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1627_cast_fp16 = einsum(equation = var_1627_equation_0, values = (var_1085_cast_fp16, var_1487_cast_fp16))[name = tensor("op_1627_cast_fp16")]; + tensor var_1629_equation_0 = const()[name = tensor("op_1629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1629_cast_fp16 = einsum(equation = var_1629_equation_0, values = (var_1089_cast_fp16, var_1488_cast_fp16))[name = tensor("op_1629_cast_fp16")]; + tensor var_1631_equation_0 = const()[name = tensor("op_1631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1631_cast_fp16 = einsum(equation = var_1631_equation_0, values = (var_1089_cast_fp16, var_1489_cast_fp16))[name = tensor("op_1631_cast_fp16")]; + tensor var_1633_equation_0 = const()[name = tensor("op_1633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1633_cast_fp16 = einsum(equation = var_1633_equation_0, values = (var_1089_cast_fp16, var_1490_cast_fp16))[name = tensor("op_1633_cast_fp16")]; + tensor var_1635_equation_0 = const()[name = tensor("op_1635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1635_cast_fp16 = einsum(equation = var_1635_equation_0, values = (var_1089_cast_fp16, var_1491_cast_fp16))[name = tensor("op_1635_cast_fp16")]; + tensor var_1637_equation_0 = const()[name = tensor("op_1637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1637_cast_fp16 = einsum(equation = var_1637_equation_0, values = (var_1093_cast_fp16, var_1492_cast_fp16))[name = tensor("op_1637_cast_fp16")]; + tensor var_1639_equation_0 = const()[name = tensor("op_1639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1639_cast_fp16 = einsum(equation = var_1639_equation_0, values = (var_1093_cast_fp16, var_1493_cast_fp16))[name = tensor("op_1639_cast_fp16")]; + tensor var_1641_equation_0 = const()[name = tensor("op_1641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1641_cast_fp16 = einsum(equation = var_1641_equation_0, values = (var_1093_cast_fp16, var_1494_cast_fp16))[name = tensor("op_1641_cast_fp16")]; + tensor var_1643_equation_0 = const()[name = tensor("op_1643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1643_cast_fp16 = einsum(equation = var_1643_equation_0, values = (var_1093_cast_fp16, var_1495_cast_fp16))[name = tensor("op_1643_cast_fp16")]; + tensor var_1645_equation_0 = const()[name = tensor("op_1645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1645_cast_fp16 = einsum(equation = var_1645_equation_0, values = (var_1097_cast_fp16, var_1496_cast_fp16))[name = tensor("op_1645_cast_fp16")]; + tensor var_1647_equation_0 = const()[name = tensor("op_1647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1647_cast_fp16 = einsum(equation = var_1647_equation_0, values = (var_1097_cast_fp16, var_1497_cast_fp16))[name = tensor("op_1647_cast_fp16")]; + tensor var_1649_equation_0 = const()[name = tensor("op_1649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1649_cast_fp16 = einsum(equation = var_1649_equation_0, values = (var_1097_cast_fp16, var_1498_cast_fp16))[name = tensor("op_1649_cast_fp16")]; + tensor var_1651_equation_0 = const()[name = tensor("op_1651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1651_cast_fp16 = einsum(equation = var_1651_equation_0, values = (var_1097_cast_fp16, var_1499_cast_fp16))[name = tensor("op_1651_cast_fp16")]; + tensor var_1653_equation_0 = const()[name = tensor("op_1653_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1653_cast_fp16 = einsum(equation = var_1653_equation_0, values = (var_1101_cast_fp16, var_1500_cast_fp16))[name = tensor("op_1653_cast_fp16")]; + tensor var_1655_equation_0 = const()[name = tensor("op_1655_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1655_cast_fp16 = einsum(equation = var_1655_equation_0, values = (var_1101_cast_fp16, var_1501_cast_fp16))[name = tensor("op_1655_cast_fp16")]; + tensor var_1657_equation_0 = const()[name = tensor("op_1657_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1657_cast_fp16 = einsum(equation = var_1657_equation_0, values = (var_1101_cast_fp16, var_1502_cast_fp16))[name = tensor("op_1657_cast_fp16")]; + tensor var_1659_equation_0 = const()[name = tensor("op_1659_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1659_cast_fp16 = einsum(equation = var_1659_equation_0, values = (var_1101_cast_fp16, var_1503_cast_fp16))[name = tensor("op_1659_cast_fp16")]; + tensor var_1661_equation_0 = const()[name = tensor("op_1661_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1661_cast_fp16 = einsum(equation = var_1661_equation_0, values = (var_1105_cast_fp16, var_1504_cast_fp16))[name = tensor("op_1661_cast_fp16")]; + tensor var_1663_equation_0 = const()[name = tensor("op_1663_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1663_cast_fp16 = einsum(equation = var_1663_equation_0, values = (var_1105_cast_fp16, var_1505_cast_fp16))[name = tensor("op_1663_cast_fp16")]; + tensor var_1665_equation_0 = const()[name = tensor("op_1665_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1665_cast_fp16 = einsum(equation = var_1665_equation_0, values = (var_1105_cast_fp16, var_1506_cast_fp16))[name = tensor("op_1665_cast_fp16")]; + tensor var_1667_equation_0 = const()[name = tensor("op_1667_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_1667_cast_fp16 = einsum(equation = var_1667_equation_0, values = (var_1105_cast_fp16, var_1507_cast_fp16))[name = tensor("op_1667_cast_fp16")]; + tensor var_1669_interleave_0 = const()[name = tensor("op_1669_interleave_0"), val = tensor(false)]; + tensor var_1669_cast_fp16 = concat(axis = var_174, interleave = var_1669_interleave_0, values = (var_1509_cast_fp16, var_1511_cast_fp16, var_1513_cast_fp16, var_1515_cast_fp16))[name = tensor("op_1669_cast_fp16")]; + tensor var_1671_interleave_0 = const()[name = tensor("op_1671_interleave_0"), val = tensor(false)]; + tensor var_1671_cast_fp16 = concat(axis = var_174, interleave = var_1671_interleave_0, values = (var_1517_cast_fp16, var_1519_cast_fp16, var_1521_cast_fp16, var_1523_cast_fp16))[name = tensor("op_1671_cast_fp16")]; + tensor var_1673_interleave_0 = const()[name = tensor("op_1673_interleave_0"), val = tensor(false)]; + tensor var_1673_cast_fp16 = concat(axis = var_174, interleave = var_1673_interleave_0, values = (var_1525_cast_fp16, var_1527_cast_fp16, var_1529_cast_fp16, var_1531_cast_fp16))[name = tensor("op_1673_cast_fp16")]; + tensor var_1675_interleave_0 = const()[name = tensor("op_1675_interleave_0"), val = tensor(false)]; + tensor var_1675_cast_fp16 = concat(axis = var_174, interleave = var_1675_interleave_0, values = (var_1533_cast_fp16, var_1535_cast_fp16, var_1537_cast_fp16, var_1539_cast_fp16))[name = tensor("op_1675_cast_fp16")]; + tensor var_1677_interleave_0 = const()[name = tensor("op_1677_interleave_0"), val = tensor(false)]; + tensor var_1677_cast_fp16 = concat(axis = var_174, interleave = var_1677_interleave_0, values = (var_1541_cast_fp16, var_1543_cast_fp16, var_1545_cast_fp16, var_1547_cast_fp16))[name = tensor("op_1677_cast_fp16")]; + tensor var_1679_interleave_0 = const()[name = tensor("op_1679_interleave_0"), val = tensor(false)]; + tensor var_1679_cast_fp16 = concat(axis = var_174, interleave = var_1679_interleave_0, values = (var_1549_cast_fp16, var_1551_cast_fp16, var_1553_cast_fp16, var_1555_cast_fp16))[name = tensor("op_1679_cast_fp16")]; + tensor var_1681_interleave_0 = const()[name = tensor("op_1681_interleave_0"), val = tensor(false)]; + tensor var_1681_cast_fp16 = concat(axis = var_174, interleave = var_1681_interleave_0, values = (var_1557_cast_fp16, var_1559_cast_fp16, var_1561_cast_fp16, var_1563_cast_fp16))[name = tensor("op_1681_cast_fp16")]; + tensor var_1683_interleave_0 = const()[name = tensor("op_1683_interleave_0"), val = tensor(false)]; + tensor var_1683_cast_fp16 = concat(axis = var_174, interleave = var_1683_interleave_0, values = (var_1565_cast_fp16, var_1567_cast_fp16, var_1569_cast_fp16, var_1571_cast_fp16))[name = tensor("op_1683_cast_fp16")]; + tensor var_1685_interleave_0 = const()[name = tensor("op_1685_interleave_0"), val = tensor(false)]; + tensor var_1685_cast_fp16 = concat(axis = var_174, interleave = var_1685_interleave_0, values = (var_1573_cast_fp16, var_1575_cast_fp16, var_1577_cast_fp16, var_1579_cast_fp16))[name = tensor("op_1685_cast_fp16")]; + tensor var_1687_interleave_0 = const()[name = tensor("op_1687_interleave_0"), val = tensor(false)]; + tensor var_1687_cast_fp16 = concat(axis = var_174, interleave = var_1687_interleave_0, values = (var_1581_cast_fp16, var_1583_cast_fp16, var_1585_cast_fp16, var_1587_cast_fp16))[name = tensor("op_1687_cast_fp16")]; + tensor var_1689_interleave_0 = const()[name = tensor("op_1689_interleave_0"), val = tensor(false)]; + tensor var_1689_cast_fp16 = concat(axis = var_174, interleave = var_1689_interleave_0, values = (var_1589_cast_fp16, var_1591_cast_fp16, var_1593_cast_fp16, var_1595_cast_fp16))[name = tensor("op_1689_cast_fp16")]; + tensor var_1691_interleave_0 = const()[name = tensor("op_1691_interleave_0"), val = tensor(false)]; + tensor var_1691_cast_fp16 = concat(axis = var_174, interleave = var_1691_interleave_0, values = (var_1597_cast_fp16, var_1599_cast_fp16, var_1601_cast_fp16, var_1603_cast_fp16))[name = tensor("op_1691_cast_fp16")]; + tensor var_1693_interleave_0 = const()[name = tensor("op_1693_interleave_0"), val = tensor(false)]; + tensor var_1693_cast_fp16 = concat(axis = var_174, interleave = var_1693_interleave_0, values = (var_1605_cast_fp16, var_1607_cast_fp16, var_1609_cast_fp16, var_1611_cast_fp16))[name = tensor("op_1693_cast_fp16")]; + tensor var_1695_interleave_0 = const()[name = tensor("op_1695_interleave_0"), val = tensor(false)]; + tensor var_1695_cast_fp16 = concat(axis = var_174, interleave = var_1695_interleave_0, values = (var_1613_cast_fp16, var_1615_cast_fp16, var_1617_cast_fp16, var_1619_cast_fp16))[name = tensor("op_1695_cast_fp16")]; + tensor var_1697_interleave_0 = const()[name = tensor("op_1697_interleave_0"), val = tensor(false)]; + tensor var_1697_cast_fp16 = concat(axis = var_174, interleave = var_1697_interleave_0, values = (var_1621_cast_fp16, var_1623_cast_fp16, var_1625_cast_fp16, var_1627_cast_fp16))[name = tensor("op_1697_cast_fp16")]; + tensor var_1699_interleave_0 = const()[name = tensor("op_1699_interleave_0"), val = tensor(false)]; + tensor var_1699_cast_fp16 = concat(axis = var_174, interleave = var_1699_interleave_0, values = (var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16, var_1635_cast_fp16))[name = tensor("op_1699_cast_fp16")]; + tensor var_1701_interleave_0 = const()[name = tensor("op_1701_interleave_0"), val = tensor(false)]; + tensor var_1701_cast_fp16 = concat(axis = var_174, interleave = var_1701_interleave_0, values = (var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16, var_1643_cast_fp16))[name = tensor("op_1701_cast_fp16")]; + tensor var_1703_interleave_0 = const()[name = tensor("op_1703_interleave_0"), val = tensor(false)]; + tensor var_1703_cast_fp16 = concat(axis = var_174, interleave = var_1703_interleave_0, values = (var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16, var_1651_cast_fp16))[name = tensor("op_1703_cast_fp16")]; + tensor var_1705_interleave_0 = const()[name = tensor("op_1705_interleave_0"), val = tensor(false)]; + tensor var_1705_cast_fp16 = concat(axis = var_174, interleave = var_1705_interleave_0, values = (var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16, var_1659_cast_fp16))[name = tensor("op_1705_cast_fp16")]; + tensor var_1707_interleave_0 = const()[name = tensor("op_1707_interleave_0"), val = tensor(false)]; + tensor var_1707_cast_fp16 = concat(axis = var_174, interleave = var_1707_interleave_0, values = (var_1661_cast_fp16, var_1663_cast_fp16, var_1665_cast_fp16, var_1667_cast_fp16))[name = tensor("op_1707_cast_fp16")]; + tensor input_7_interleave_0 = const()[name = tensor("input_7_interleave_0"), val = tensor(false)]; + tensor input_7_cast_fp16 = concat(axis = var_199, interleave = input_7_interleave_0, values = (var_1669_cast_fp16, var_1671_cast_fp16, var_1673_cast_fp16, var_1675_cast_fp16, var_1677_cast_fp16, var_1679_cast_fp16, var_1681_cast_fp16, var_1683_cast_fp16, var_1685_cast_fp16, var_1687_cast_fp16, var_1689_cast_fp16, var_1691_cast_fp16, var_1693_cast_fp16, var_1695_cast_fp16, var_1697_cast_fp16, var_1699_cast_fp16, var_1701_cast_fp16, var_1703_cast_fp16, var_1705_cast_fp16, var_1707_cast_fp16))[name = tensor("input_7_cast_fp16")]; + tensor var_1715 = const()[name = tensor("op_1715"), val = tensor([1, 1])]; + tensor var_1717 = const()[name = tensor("op_1717"), val = tensor([1, 1])]; + tensor pretrained_out_7_pad_type_0 = const()[name = tensor("pretrained_out_7_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_7_pad_0 = const()[name = tensor("pretrained_out_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17010368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17829632))), name = tensor("layers_0_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_0_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17829760)))]; + tensor pretrained_out_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_1717, groups = var_199, pad = pretrained_out_7_pad_0, pad_type = pretrained_out_7_pad_type_0, strides = var_1715, weight = layers_0_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor("pretrained_out_7_cast_fp16")]; + tensor var_1721 = const()[name = tensor("op_1721"), val = tensor([1, 1])]; + tensor var_1723 = const()[name = tensor("op_1723"), val = tensor([1, 1])]; + tensor input_9_pad_type_0 = const()[name = tensor("input_9_pad_type_0"), val = tensor("custom")]; + tensor input_9_pad_0 = const()[name = tensor("input_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17832384)))]; + tensor input_9_cast_fp16 = conv(dilations = var_1723, groups = var_199, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = var_1721, weight = layers_0_self_attn_o_proj_loraA_weight_to_fp16, x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor var_1727 = const()[name = tensor("op_1727"), val = tensor([1, 1])]; + tensor var_1729 = const()[name = tensor("op_1729"), val = tensor([1, 1])]; + tensor lora_out_13_pad_type_0 = const()[name = tensor("lora_out_13_pad_type_0"), val = tensor("custom")]; + tensor lora_out_13_pad_0 = const()[name = tensor("lora_out_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_15_weight_0_to_fp16 = const()[name = tensor("lora_out_15_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17873408)))]; + tensor lora_out_15_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_1729, groups = var_199, pad = lora_out_13_pad_0, pad_type = lora_out_13_pad_type_0, strides = var_1727, weight = lora_out_15_weight_0_to_fp16, x = input_9_cast_fp16)[name = tensor("lora_out_15_cast_fp16")]; + tensor obj_3_cast_fp16 = add(x = pretrained_out_7_cast_fp16, y = lora_out_15_cast_fp16)[name = tensor("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor var_1738 = const()[name = tensor("op_1738"), val = tensor([1])]; + tensor channels_mean_3_cast_fp16 = reduce_mean(axes = var_1738, keep_dims = var_200, x = inputs_3_cast_fp16)[name = tensor("channels_mean_3_cast_fp16")]; + tensor zero_mean_3_cast_fp16 = sub(x = inputs_3_cast_fp16, y = channels_mean_3_cast_fp16)[name = tensor("zero_mean_3_cast_fp16")]; + tensor zero_mean_sq_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = zero_mean_3_cast_fp16)[name = tensor("zero_mean_sq_3_cast_fp16")]; + tensor var_1742 = const()[name = tensor("op_1742"), val = tensor([1])]; + tensor var_1743_cast_fp16 = reduce_mean(axes = var_1742, keep_dims = var_200, x = zero_mean_sq_3_cast_fp16)[name = tensor("op_1743_cast_fp16")]; + tensor var_1744_to_fp16 = const()[name = tensor("op_1744_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1745_cast_fp16 = add(x = var_1743_cast_fp16, y = var_1744_to_fp16)[name = tensor("op_1745_cast_fp16")]; + tensor denom_3_epsilon_0 = const()[name = tensor("denom_3_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_3_cast_fp16 = rsqrt(epsilon = denom_3_epsilon_0, x = var_1745_cast_fp16)[name = tensor("denom_3_cast_fp16")]; + tensor out_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = denom_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = tensor("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17914432)))]; + tensor input_11_beta_0_to_fp16 = const()[name = tensor("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17917056)))]; + tensor input_11_epsilon_0_to_fp16 = const()[name = tensor("input_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor var_1759 = const()[name = tensor("op_1759"), val = tensor([1, 1])]; + tensor var_1761 = const()[name = tensor("op_1761"), val = tensor([1, 1])]; + tensor pretrained_out_9_pad_type_0 = const()[name = tensor("pretrained_out_9_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_9_pad_0 = const()[name = tensor("pretrained_out_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17919680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21196544))), name = tensor("layers_0_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_0_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_0_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21196672)))]; + tensor pretrained_out_9_cast_fp16 = conv(bias = layers_0_fc1_pretrained_bias_to_fp16, dilations = var_1761, groups = var_199, pad = pretrained_out_9_pad_0, pad_type = pretrained_out_9_pad_type_0, strides = var_1759, weight = layers_0_fc1_pretrained_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = tensor("pretrained_out_9_cast_fp16")]; + tensor var_1765 = const()[name = tensor("op_1765"), val = tensor([1, 1])]; + tensor var_1767 = const()[name = tensor("op_1767"), val = tensor([1, 1])]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("custom")]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_0_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21206976)))]; + tensor input_13_cast_fp16 = conv(dilations = var_1767, groups = var_199, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_1765, weight = layers_0_fc1_loraA_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor var_1771 = const()[name = tensor("op_1771"), val = tensor([1, 1])]; + tensor var_1773 = const()[name = tensor("op_1773"), val = tensor([1, 1])]; + tensor lora_out_17_pad_type_0 = const()[name = tensor("lora_out_17_pad_type_0"), val = tensor("custom")]; + tensor lora_out_17_pad_0 = const()[name = tensor("lora_out_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_19_weight_0_to_fp16 = const()[name = tensor("lora_out_19_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21248000)))]; + tensor lora_out_19_bias_0_to_fp16 = const()[name = tensor("lora_out_19_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21411904)))]; + tensor lora_out_19_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_1773, groups = var_199, pad = lora_out_17_pad_0, pad_type = lora_out_17_pad_type_0, strides = var_1771, weight = lora_out_19_weight_0_to_fp16, x = input_13_cast_fp16)[name = tensor("lora_out_19_cast_fp16")]; + tensor input_15_cast_fp16 = add(x = pretrained_out_9_cast_fp16, y = lora_out_19_cast_fp16)[name = tensor("input_15_cast_fp16")]; + tensor input_17_mode_0 = const()[name = tensor("input_17_mode_0"), val = tensor("EXACT")]; + tensor input_17_cast_fp16 = gelu(mode = input_17_mode_0, x = input_15_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor var_1785 = const()[name = tensor("op_1785"), val = tensor([1, 1])]; + tensor var_1787 = const()[name = tensor("op_1787"), val = tensor([1, 1])]; + tensor pretrained_out_11_pad_type_0 = const()[name = tensor("pretrained_out_11_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_11_pad_0 = const()[name = tensor("pretrained_out_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21422208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24699072))), name = tensor("layers_0_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_0_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_0_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24699200)))]; + tensor pretrained_out_11_cast_fp16 = conv(bias = layers_0_fc2_pretrained_bias_to_fp16, dilations = var_1787, groups = var_199, pad = pretrained_out_11_pad_0, pad_type = pretrained_out_11_pad_type_0, strides = var_1785, weight = layers_0_fc2_pretrained_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("pretrained_out_11_cast_fp16")]; + tensor var_1791 = const()[name = tensor("op_1791"), val = tensor([1, 1])]; + tensor var_1793 = const()[name = tensor("op_1793"), val = tensor([1, 1])]; + tensor input_19_pad_type_0 = const()[name = tensor("input_19_pad_type_0"), val = tensor("custom")]; + tensor input_19_pad_0 = const()[name = tensor("input_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_0_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_0_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24701824)))]; + tensor input_19_cast_fp16 = conv(dilations = var_1793, groups = var_199, pad = input_19_pad_0, pad_type = input_19_pad_type_0, strides = var_1791, weight = layers_0_fc2_loraA_weight_to_fp16, x = input_17_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor var_1797 = const()[name = tensor("op_1797"), val = tensor([1, 1])]; + tensor var_1799 = const()[name = tensor("op_1799"), val = tensor([1, 1])]; + tensor lora_out_21_pad_type_0 = const()[name = tensor("lora_out_21_pad_type_0"), val = tensor("custom")]; + tensor lora_out_21_pad_0 = const()[name = tensor("lora_out_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_23_weight_0_to_fp16 = const()[name = tensor("lora_out_23_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24865728)))]; + tensor lora_out_23_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_1799, groups = var_199, pad = lora_out_21_pad_0, pad_type = lora_out_21_pad_type_0, strides = var_1797, weight = lora_out_23_weight_0_to_fp16, x = input_19_cast_fp16)[name = tensor("lora_out_23_cast_fp16")]; + tensor hidden_states_5_cast_fp16 = add(x = pretrained_out_11_cast_fp16, y = lora_out_23_cast_fp16)[name = tensor("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor var_1809 = const()[name = tensor("op_1809"), val = tensor(3)]; + tensor var_1834 = const()[name = tensor("op_1834"), val = tensor(1)]; + tensor var_1835 = const()[name = tensor("op_1835"), val = tensor(true)]; + tensor var_1845 = const()[name = tensor("op_1845"), val = tensor([1])]; + tensor channels_mean_5_cast_fp16 = reduce_mean(axes = var_1845, keep_dims = var_1835, x = inputs_5_cast_fp16)[name = tensor("channels_mean_5_cast_fp16")]; + tensor zero_mean_5_cast_fp16 = sub(x = inputs_5_cast_fp16, y = channels_mean_5_cast_fp16)[name = tensor("zero_mean_5_cast_fp16")]; + tensor zero_mean_sq_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = zero_mean_5_cast_fp16)[name = tensor("zero_mean_sq_5_cast_fp16")]; + tensor var_1849 = const()[name = tensor("op_1849"), val = tensor([1])]; + tensor var_1850_cast_fp16 = reduce_mean(axes = var_1849, keep_dims = var_1835, x = zero_mean_sq_5_cast_fp16)[name = tensor("op_1850_cast_fp16")]; + tensor var_1851_to_fp16 = const()[name = tensor("op_1851_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1852_cast_fp16 = add(x = var_1850_cast_fp16, y = var_1851_to_fp16)[name = tensor("op_1852_cast_fp16")]; + tensor denom_5_epsilon_0 = const()[name = tensor("denom_5_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_5_cast_fp16 = rsqrt(epsilon = denom_5_epsilon_0, x = var_1852_cast_fp16)[name = tensor("denom_5_cast_fp16")]; + tensor out_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = denom_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = tensor("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24906752)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = tensor("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24909376)))]; + tensor obj_5_epsilon_0_to_fp16 = const()[name = tensor("obj_5_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor("obj_5_cast_fp16")]; + tensor var_1870 = const()[name = tensor("op_1870"), val = tensor([1, 1])]; + tensor var_1872 = const()[name = tensor("op_1872"), val = tensor([1, 1])]; + tensor pretrained_out_13_pad_type_0 = const()[name = tensor("pretrained_out_13_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_13_pad_0 = const()[name = tensor("pretrained_out_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24912000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25731264))), name = tensor("layers_1_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_1_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25731392)))]; + tensor pretrained_out_13_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_1872, groups = var_1834, pad = pretrained_out_13_pad_0, pad_type = pretrained_out_13_pad_type_0, strides = var_1870, weight = layers_1_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor("pretrained_out_13_cast_fp16")]; + tensor var_1876 = const()[name = tensor("op_1876"), val = tensor([1, 1])]; + tensor var_1878 = const()[name = tensor("op_1878"), val = tensor([1, 1])]; + tensor input_21_pad_type_0 = const()[name = tensor("input_21_pad_type_0"), val = tensor("custom")]; + tensor input_21_pad_0 = const()[name = tensor("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25734016)))]; + tensor input_21_cast_fp16 = conv(dilations = var_1878, groups = var_1834, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = var_1876, weight = layers_1_self_attn_q_proj_loraA_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor var_1882 = const()[name = tensor("op_1882"), val = tensor([1, 1])]; + tensor var_1884 = const()[name = tensor("op_1884"), val = tensor([1, 1])]; + tensor lora_out_25_pad_type_0 = const()[name = tensor("lora_out_25_pad_type_0"), val = tensor("custom")]; + tensor lora_out_25_pad_0 = const()[name = tensor("lora_out_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_27_weight_0_to_fp16 = const()[name = tensor("lora_out_27_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25775040)))]; + tensor lora_out_27_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_1884, groups = var_1834, pad = lora_out_25_pad_0, pad_type = lora_out_25_pad_type_0, strides = var_1882, weight = lora_out_27_weight_0_to_fp16, x = input_21_cast_fp16)[name = tensor("lora_out_27_cast_fp16")]; + tensor query_3_cast_fp16 = add(x = pretrained_out_13_cast_fp16, y = lora_out_27_cast_fp16)[name = tensor("query_3_cast_fp16")]; + tensor var_1894 = const()[name = tensor("op_1894"), val = tensor([1, 1])]; + tensor var_1896 = const()[name = tensor("op_1896"), val = tensor([1, 1])]; + tensor pretrained_out_15_pad_type_0 = const()[name = tensor("pretrained_out_15_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_15_pad_0 = const()[name = tensor("pretrained_out_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25816064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26635328))), name = tensor("layers_1_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_15_cast_fp16 = conv(dilations = var_1896, groups = var_1834, pad = pretrained_out_15_pad_0, pad_type = pretrained_out_15_pad_type_0, strides = var_1894, weight = layers_1_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor("pretrained_out_15_cast_fp16")]; + tensor var_1900 = const()[name = tensor("op_1900"), val = tensor([1, 1])]; + tensor var_1902 = const()[name = tensor("op_1902"), val = tensor([1, 1])]; + tensor input_23_pad_type_0 = const()[name = tensor("input_23_pad_type_0"), val = tensor("custom")]; + tensor input_23_pad_0 = const()[name = tensor("input_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26635456)))]; + tensor input_23_cast_fp16 = conv(dilations = var_1902, groups = var_1834, pad = input_23_pad_0, pad_type = input_23_pad_type_0, strides = var_1900, weight = layers_1_self_attn_k_proj_loraA_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor var_1906 = const()[name = tensor("op_1906"), val = tensor([1, 1])]; + tensor var_1908 = const()[name = tensor("op_1908"), val = tensor([1, 1])]; + tensor lora_out_29_pad_type_0 = const()[name = tensor("lora_out_29_pad_type_0"), val = tensor("custom")]; + tensor lora_out_29_pad_0 = const()[name = tensor("lora_out_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_31_weight_0_to_fp16 = const()[name = tensor("lora_out_31_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26676480)))]; + tensor lora_out_31_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_1908, groups = var_1834, pad = lora_out_29_pad_0, pad_type = lora_out_29_pad_type_0, strides = var_1906, weight = lora_out_31_weight_0_to_fp16, x = input_23_cast_fp16)[name = tensor("lora_out_31_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = pretrained_out_15_cast_fp16, y = lora_out_31_cast_fp16)[name = tensor("key_3_cast_fp16")]; + tensor var_1919 = const()[name = tensor("op_1919"), val = tensor([1, 1])]; + tensor var_1921 = const()[name = tensor("op_1921"), val = tensor([1, 1])]; + tensor pretrained_out_17_pad_type_0 = const()[name = tensor("pretrained_out_17_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_17_pad_0 = const()[name = tensor("pretrained_out_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26717504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27536768))), name = tensor("layers_1_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_1_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27536896)))]; + tensor pretrained_out_17_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_1921, groups = var_1834, pad = pretrained_out_17_pad_0, pad_type = pretrained_out_17_pad_type_0, strides = var_1919, weight = layers_1_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor("pretrained_out_17_cast_fp16")]; + tensor var_1925 = const()[name = tensor("op_1925"), val = tensor([1, 1])]; + tensor var_1927 = const()[name = tensor("op_1927"), val = tensor([1, 1])]; + tensor input_25_pad_type_0 = const()[name = tensor("input_25_pad_type_0"), val = tensor("custom")]; + tensor input_25_pad_0 = const()[name = tensor("input_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27539520)))]; + tensor input_25_cast_fp16 = conv(dilations = var_1927, groups = var_1834, pad = input_25_pad_0, pad_type = input_25_pad_type_0, strides = var_1925, weight = layers_1_self_attn_v_proj_loraA_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor var_1931 = const()[name = tensor("op_1931"), val = tensor([1, 1])]; + tensor var_1933 = const()[name = tensor("op_1933"), val = tensor([1, 1])]; + tensor lora_out_33_pad_type_0 = const()[name = tensor("lora_out_33_pad_type_0"), val = tensor("custom")]; + tensor lora_out_33_pad_0 = const()[name = tensor("lora_out_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_35_weight_0_to_fp16 = const()[name = tensor("lora_out_35_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27580544)))]; + tensor lora_out_35_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_1933, groups = var_1834, pad = lora_out_33_pad_0, pad_type = lora_out_33_pad_type_0, strides = var_1931, weight = lora_out_35_weight_0_to_fp16, x = input_25_cast_fp16)[name = tensor("lora_out_35_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = pretrained_out_17_cast_fp16, y = lora_out_35_cast_fp16)[name = tensor("value_3_cast_fp16")]; + tensor var_1943_begin_0 = const()[name = tensor("op_1943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1943_end_0 = const()[name = tensor("op_1943_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1943_end_mask_0 = const()[name = tensor("op_1943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1943_cast_fp16 = slice_by_index(begin = var_1943_begin_0, end = var_1943_end_0, end_mask = var_1943_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1943_cast_fp16")]; + tensor var_1947_begin_0 = const()[name = tensor("op_1947_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1947_end_0 = const()[name = tensor("op_1947_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1947_end_mask_0 = const()[name = tensor("op_1947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1947_cast_fp16 = slice_by_index(begin = var_1947_begin_0, end = var_1947_end_0, end_mask = var_1947_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1947_cast_fp16")]; + tensor var_1951_begin_0 = const()[name = tensor("op_1951_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1951_end_0 = const()[name = tensor("op_1951_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1951_end_mask_0 = const()[name = tensor("op_1951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1951_cast_fp16 = slice_by_index(begin = var_1951_begin_0, end = var_1951_end_0, end_mask = var_1951_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1951_cast_fp16")]; + tensor var_1955_begin_0 = const()[name = tensor("op_1955_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1955_end_0 = const()[name = tensor("op_1955_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1955_end_mask_0 = const()[name = tensor("op_1955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1955_cast_fp16 = slice_by_index(begin = var_1955_begin_0, end = var_1955_end_0, end_mask = var_1955_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1955_cast_fp16")]; + tensor var_1959_begin_0 = const()[name = tensor("op_1959_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1959_end_0 = const()[name = tensor("op_1959_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1959_end_mask_0 = const()[name = tensor("op_1959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1959_cast_fp16 = slice_by_index(begin = var_1959_begin_0, end = var_1959_end_0, end_mask = var_1959_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1959_cast_fp16")]; + tensor var_1963_begin_0 = const()[name = tensor("op_1963_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1963_end_0 = const()[name = tensor("op_1963_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1963_end_mask_0 = const()[name = tensor("op_1963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1963_cast_fp16 = slice_by_index(begin = var_1963_begin_0, end = var_1963_end_0, end_mask = var_1963_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1963_cast_fp16")]; + tensor var_1967_begin_0 = const()[name = tensor("op_1967_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1967_end_0 = const()[name = tensor("op_1967_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1967_end_mask_0 = const()[name = tensor("op_1967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1967_cast_fp16 = slice_by_index(begin = var_1967_begin_0, end = var_1967_end_0, end_mask = var_1967_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1967_cast_fp16")]; + tensor var_1971_begin_0 = const()[name = tensor("op_1971_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1971_end_0 = const()[name = tensor("op_1971_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1971_end_mask_0 = const()[name = tensor("op_1971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1971_cast_fp16 = slice_by_index(begin = var_1971_begin_0, end = var_1971_end_0, end_mask = var_1971_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1971_cast_fp16")]; + tensor var_1975_begin_0 = const()[name = tensor("op_1975_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_1975_end_0 = const()[name = tensor("op_1975_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_1975_end_mask_0 = const()[name = tensor("op_1975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1975_cast_fp16 = slice_by_index(begin = var_1975_begin_0, end = var_1975_end_0, end_mask = var_1975_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1975_cast_fp16")]; + tensor var_1979_begin_0 = const()[name = tensor("op_1979_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_1979_end_0 = const()[name = tensor("op_1979_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_1979_end_mask_0 = const()[name = tensor("op_1979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1979_cast_fp16 = slice_by_index(begin = var_1979_begin_0, end = var_1979_end_0, end_mask = var_1979_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1979_cast_fp16")]; + tensor var_1983_begin_0 = const()[name = tensor("op_1983_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_1983_end_0 = const()[name = tensor("op_1983_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_1983_end_mask_0 = const()[name = tensor("op_1983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1983_cast_fp16 = slice_by_index(begin = var_1983_begin_0, end = var_1983_end_0, end_mask = var_1983_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1983_cast_fp16")]; + tensor var_1987_begin_0 = const()[name = tensor("op_1987_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_1987_end_0 = const()[name = tensor("op_1987_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_1987_end_mask_0 = const()[name = tensor("op_1987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1987_cast_fp16 = slice_by_index(begin = var_1987_begin_0, end = var_1987_end_0, end_mask = var_1987_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1987_cast_fp16")]; + tensor var_1991_begin_0 = const()[name = tensor("op_1991_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_1991_end_0 = const()[name = tensor("op_1991_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_1991_end_mask_0 = const()[name = tensor("op_1991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1991_cast_fp16 = slice_by_index(begin = var_1991_begin_0, end = var_1991_end_0, end_mask = var_1991_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1991_cast_fp16")]; + tensor var_1995_begin_0 = const()[name = tensor("op_1995_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_1995_end_0 = const()[name = tensor("op_1995_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_1995_end_mask_0 = const()[name = tensor("op_1995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1995_cast_fp16 = slice_by_index(begin = var_1995_begin_0, end = var_1995_end_0, end_mask = var_1995_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1995_cast_fp16")]; + tensor var_1999_begin_0 = const()[name = tensor("op_1999_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_1999_end_0 = const()[name = tensor("op_1999_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_1999_end_mask_0 = const()[name = tensor("op_1999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1999_cast_fp16 = slice_by_index(begin = var_1999_begin_0, end = var_1999_end_0, end_mask = var_1999_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1999_cast_fp16")]; + tensor var_2003_begin_0 = const()[name = tensor("op_2003_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2003_end_0 = const()[name = tensor("op_2003_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_2003_end_mask_0 = const()[name = tensor("op_2003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2003_cast_fp16 = slice_by_index(begin = var_2003_begin_0, end = var_2003_end_0, end_mask = var_2003_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_2003_cast_fp16")]; + tensor var_2007_begin_0 = const()[name = tensor("op_2007_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2007_end_0 = const()[name = tensor("op_2007_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_2007_end_mask_0 = const()[name = tensor("op_2007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2007_cast_fp16 = slice_by_index(begin = var_2007_begin_0, end = var_2007_end_0, end_mask = var_2007_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_2007_cast_fp16")]; + tensor var_2011_begin_0 = const()[name = tensor("op_2011_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2011_end_0 = const()[name = tensor("op_2011_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_2011_end_mask_0 = const()[name = tensor("op_2011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2011_cast_fp16 = slice_by_index(begin = var_2011_begin_0, end = var_2011_end_0, end_mask = var_2011_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_2011_cast_fp16")]; + tensor var_2015_begin_0 = const()[name = tensor("op_2015_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2015_end_0 = const()[name = tensor("op_2015_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_2015_end_mask_0 = const()[name = tensor("op_2015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2015_cast_fp16 = slice_by_index(begin = var_2015_begin_0, end = var_2015_end_0, end_mask = var_2015_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_2015_cast_fp16")]; + tensor var_2019_begin_0 = const()[name = tensor("op_2019_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2019_end_0 = const()[name = tensor("op_2019_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_2019_end_mask_0 = const()[name = tensor("op_2019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2019_cast_fp16 = slice_by_index(begin = var_2019_begin_0, end = var_2019_end_0, end_mask = var_2019_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_2019_cast_fp16")]; + tensor var_2028_begin_0 = const()[name = tensor("op_2028_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2028_end_0 = const()[name = tensor("op_2028_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2028_end_mask_0 = const()[name = tensor("op_2028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = var_1943_cast_fp16)[name = tensor("op_2028_cast_fp16")]; + tensor var_2035_begin_0 = const()[name = tensor("op_2035_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2035_end_0 = const()[name = tensor("op_2035_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2035_end_mask_0 = const()[name = tensor("op_2035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2035_cast_fp16 = slice_by_index(begin = var_2035_begin_0, end = var_2035_end_0, end_mask = var_2035_end_mask_0, x = var_1943_cast_fp16)[name = tensor("op_2035_cast_fp16")]; + tensor var_2042_begin_0 = const()[name = tensor("op_2042_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2042_end_0 = const()[name = tensor("op_2042_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2042_end_mask_0 = const()[name = tensor("op_2042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2042_cast_fp16 = slice_by_index(begin = var_2042_begin_0, end = var_2042_end_0, end_mask = var_2042_end_mask_0, x = var_1943_cast_fp16)[name = tensor("op_2042_cast_fp16")]; + tensor var_2049_begin_0 = const()[name = tensor("op_2049_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2049_end_0 = const()[name = tensor("op_2049_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2049_end_mask_0 = const()[name = tensor("op_2049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2049_cast_fp16 = slice_by_index(begin = var_2049_begin_0, end = var_2049_end_0, end_mask = var_2049_end_mask_0, x = var_1943_cast_fp16)[name = tensor("op_2049_cast_fp16")]; + tensor var_2056_begin_0 = const()[name = tensor("op_2056_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2056_end_0 = const()[name = tensor("op_2056_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2056_end_mask_0 = const()[name = tensor("op_2056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2056_cast_fp16 = slice_by_index(begin = var_2056_begin_0, end = var_2056_end_0, end_mask = var_2056_end_mask_0, x = var_1947_cast_fp16)[name = tensor("op_2056_cast_fp16")]; + tensor var_2063_begin_0 = const()[name = tensor("op_2063_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2063_end_0 = const()[name = tensor("op_2063_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2063_end_mask_0 = const()[name = tensor("op_2063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2063_cast_fp16 = slice_by_index(begin = var_2063_begin_0, end = var_2063_end_0, end_mask = var_2063_end_mask_0, x = var_1947_cast_fp16)[name = tensor("op_2063_cast_fp16")]; + tensor var_2070_begin_0 = const()[name = tensor("op_2070_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2070_end_0 = const()[name = tensor("op_2070_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2070_end_mask_0 = const()[name = tensor("op_2070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2070_cast_fp16 = slice_by_index(begin = var_2070_begin_0, end = var_2070_end_0, end_mask = var_2070_end_mask_0, x = var_1947_cast_fp16)[name = tensor("op_2070_cast_fp16")]; + tensor var_2077_begin_0 = const()[name = tensor("op_2077_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2077_end_0 = const()[name = tensor("op_2077_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2077_end_mask_0 = const()[name = tensor("op_2077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2077_cast_fp16 = slice_by_index(begin = var_2077_begin_0, end = var_2077_end_0, end_mask = var_2077_end_mask_0, x = var_1947_cast_fp16)[name = tensor("op_2077_cast_fp16")]; + tensor var_2084_begin_0 = const()[name = tensor("op_2084_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2084_end_0 = const()[name = tensor("op_2084_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2084_end_mask_0 = const()[name = tensor("op_2084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2084_cast_fp16 = slice_by_index(begin = var_2084_begin_0, end = var_2084_end_0, end_mask = var_2084_end_mask_0, x = var_1951_cast_fp16)[name = tensor("op_2084_cast_fp16")]; + tensor var_2091_begin_0 = const()[name = tensor("op_2091_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2091_end_0 = const()[name = tensor("op_2091_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2091_end_mask_0 = const()[name = tensor("op_2091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2091_cast_fp16 = slice_by_index(begin = var_2091_begin_0, end = var_2091_end_0, end_mask = var_2091_end_mask_0, x = var_1951_cast_fp16)[name = tensor("op_2091_cast_fp16")]; + tensor var_2098_begin_0 = const()[name = tensor("op_2098_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2098_end_0 = const()[name = tensor("op_2098_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2098_end_mask_0 = const()[name = tensor("op_2098_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2098_cast_fp16 = slice_by_index(begin = var_2098_begin_0, end = var_2098_end_0, end_mask = var_2098_end_mask_0, x = var_1951_cast_fp16)[name = tensor("op_2098_cast_fp16")]; + tensor var_2105_begin_0 = const()[name = tensor("op_2105_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2105_end_0 = const()[name = tensor("op_2105_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2105_end_mask_0 = const()[name = tensor("op_2105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, x = var_1951_cast_fp16)[name = tensor("op_2105_cast_fp16")]; + tensor var_2112_begin_0 = const()[name = tensor("op_2112_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2112_end_0 = const()[name = tensor("op_2112_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2112_end_mask_0 = const()[name = tensor("op_2112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2112_cast_fp16 = slice_by_index(begin = var_2112_begin_0, end = var_2112_end_0, end_mask = var_2112_end_mask_0, x = var_1955_cast_fp16)[name = tensor("op_2112_cast_fp16")]; + tensor var_2119_begin_0 = const()[name = tensor("op_2119_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2119_end_0 = const()[name = tensor("op_2119_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2119_end_mask_0 = const()[name = tensor("op_2119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2119_cast_fp16 = slice_by_index(begin = var_2119_begin_0, end = var_2119_end_0, end_mask = var_2119_end_mask_0, x = var_1955_cast_fp16)[name = tensor("op_2119_cast_fp16")]; + tensor var_2126_begin_0 = const()[name = tensor("op_2126_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2126_end_0 = const()[name = tensor("op_2126_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2126_end_mask_0 = const()[name = tensor("op_2126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2126_cast_fp16 = slice_by_index(begin = var_2126_begin_0, end = var_2126_end_0, end_mask = var_2126_end_mask_0, x = var_1955_cast_fp16)[name = tensor("op_2126_cast_fp16")]; + tensor var_2133_begin_0 = const()[name = tensor("op_2133_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2133_end_0 = const()[name = tensor("op_2133_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2133_end_mask_0 = const()[name = tensor("op_2133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2133_cast_fp16 = slice_by_index(begin = var_2133_begin_0, end = var_2133_end_0, end_mask = var_2133_end_mask_0, x = var_1955_cast_fp16)[name = tensor("op_2133_cast_fp16")]; + tensor var_2140_begin_0 = const()[name = tensor("op_2140_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2140_end_0 = const()[name = tensor("op_2140_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2140_end_mask_0 = const()[name = tensor("op_2140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2140_cast_fp16 = slice_by_index(begin = var_2140_begin_0, end = var_2140_end_0, end_mask = var_2140_end_mask_0, x = var_1959_cast_fp16)[name = tensor("op_2140_cast_fp16")]; + tensor var_2147_begin_0 = const()[name = tensor("op_2147_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2147_end_0 = const()[name = tensor("op_2147_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2147_end_mask_0 = const()[name = tensor("op_2147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2147_cast_fp16 = slice_by_index(begin = var_2147_begin_0, end = var_2147_end_0, end_mask = var_2147_end_mask_0, x = var_1959_cast_fp16)[name = tensor("op_2147_cast_fp16")]; + tensor var_2154_begin_0 = const()[name = tensor("op_2154_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2154_end_0 = const()[name = tensor("op_2154_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2154_end_mask_0 = const()[name = tensor("op_2154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2154_cast_fp16 = slice_by_index(begin = var_2154_begin_0, end = var_2154_end_0, end_mask = var_2154_end_mask_0, x = var_1959_cast_fp16)[name = tensor("op_2154_cast_fp16")]; + tensor var_2161_begin_0 = const()[name = tensor("op_2161_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2161_end_0 = const()[name = tensor("op_2161_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2161_end_mask_0 = const()[name = tensor("op_2161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2161_cast_fp16 = slice_by_index(begin = var_2161_begin_0, end = var_2161_end_0, end_mask = var_2161_end_mask_0, x = var_1959_cast_fp16)[name = tensor("op_2161_cast_fp16")]; + tensor var_2168_begin_0 = const()[name = tensor("op_2168_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2168_end_0 = const()[name = tensor("op_2168_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2168_end_mask_0 = const()[name = tensor("op_2168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, x = var_1963_cast_fp16)[name = tensor("op_2168_cast_fp16")]; + tensor var_2175_begin_0 = const()[name = tensor("op_2175_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2175_end_0 = const()[name = tensor("op_2175_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2175_end_mask_0 = const()[name = tensor("op_2175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2175_cast_fp16 = slice_by_index(begin = var_2175_begin_0, end = var_2175_end_0, end_mask = var_2175_end_mask_0, x = var_1963_cast_fp16)[name = tensor("op_2175_cast_fp16")]; + tensor var_2182_begin_0 = const()[name = tensor("op_2182_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2182_end_0 = const()[name = tensor("op_2182_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2182_end_mask_0 = const()[name = tensor("op_2182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2182_cast_fp16 = slice_by_index(begin = var_2182_begin_0, end = var_2182_end_0, end_mask = var_2182_end_mask_0, x = var_1963_cast_fp16)[name = tensor("op_2182_cast_fp16")]; + tensor var_2189_begin_0 = const()[name = tensor("op_2189_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2189_end_0 = const()[name = tensor("op_2189_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2189_end_mask_0 = const()[name = tensor("op_2189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2189_cast_fp16 = slice_by_index(begin = var_2189_begin_0, end = var_2189_end_0, end_mask = var_2189_end_mask_0, x = var_1963_cast_fp16)[name = tensor("op_2189_cast_fp16")]; + tensor var_2196_begin_0 = const()[name = tensor("op_2196_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2196_end_0 = const()[name = tensor("op_2196_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2196_end_mask_0 = const()[name = tensor("op_2196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2196_cast_fp16 = slice_by_index(begin = var_2196_begin_0, end = var_2196_end_0, end_mask = var_2196_end_mask_0, x = var_1967_cast_fp16)[name = tensor("op_2196_cast_fp16")]; + tensor var_2203_begin_0 = const()[name = tensor("op_2203_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2203_end_0 = const()[name = tensor("op_2203_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2203_end_mask_0 = const()[name = tensor("op_2203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2203_cast_fp16 = slice_by_index(begin = var_2203_begin_0, end = var_2203_end_0, end_mask = var_2203_end_mask_0, x = var_1967_cast_fp16)[name = tensor("op_2203_cast_fp16")]; + tensor var_2210_begin_0 = const()[name = tensor("op_2210_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2210_end_0 = const()[name = tensor("op_2210_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2210_end_mask_0 = const()[name = tensor("op_2210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2210_cast_fp16 = slice_by_index(begin = var_2210_begin_0, end = var_2210_end_0, end_mask = var_2210_end_mask_0, x = var_1967_cast_fp16)[name = tensor("op_2210_cast_fp16")]; + tensor var_2217_begin_0 = const()[name = tensor("op_2217_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2217_end_0 = const()[name = tensor("op_2217_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2217_end_mask_0 = const()[name = tensor("op_2217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2217_cast_fp16 = slice_by_index(begin = var_2217_begin_0, end = var_2217_end_0, end_mask = var_2217_end_mask_0, x = var_1967_cast_fp16)[name = tensor("op_2217_cast_fp16")]; + tensor var_2224_begin_0 = const()[name = tensor("op_2224_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2224_end_0 = const()[name = tensor("op_2224_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2224_end_mask_0 = const()[name = tensor("op_2224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2224_cast_fp16 = slice_by_index(begin = var_2224_begin_0, end = var_2224_end_0, end_mask = var_2224_end_mask_0, x = var_1971_cast_fp16)[name = tensor("op_2224_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = tensor("op_2231_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2231_end_0 = const()[name = tensor("op_2231_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2231_end_mask_0 = const()[name = tensor("op_2231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = var_1971_cast_fp16)[name = tensor("op_2231_cast_fp16")]; + tensor var_2238_begin_0 = const()[name = tensor("op_2238_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2238_end_0 = const()[name = tensor("op_2238_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2238_end_mask_0 = const()[name = tensor("op_2238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2238_cast_fp16 = slice_by_index(begin = var_2238_begin_0, end = var_2238_end_0, end_mask = var_2238_end_mask_0, x = var_1971_cast_fp16)[name = tensor("op_2238_cast_fp16")]; + tensor var_2245_begin_0 = const()[name = tensor("op_2245_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2245_end_0 = const()[name = tensor("op_2245_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2245_end_mask_0 = const()[name = tensor("op_2245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2245_cast_fp16 = slice_by_index(begin = var_2245_begin_0, end = var_2245_end_0, end_mask = var_2245_end_mask_0, x = var_1971_cast_fp16)[name = tensor("op_2245_cast_fp16")]; + tensor var_2252_begin_0 = const()[name = tensor("op_2252_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2252_end_0 = const()[name = tensor("op_2252_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2252_end_mask_0 = const()[name = tensor("op_2252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2252_cast_fp16 = slice_by_index(begin = var_2252_begin_0, end = var_2252_end_0, end_mask = var_2252_end_mask_0, x = var_1975_cast_fp16)[name = tensor("op_2252_cast_fp16")]; + tensor var_2259_begin_0 = const()[name = tensor("op_2259_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2259_end_0 = const()[name = tensor("op_2259_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2259_end_mask_0 = const()[name = tensor("op_2259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2259_cast_fp16 = slice_by_index(begin = var_2259_begin_0, end = var_2259_end_0, end_mask = var_2259_end_mask_0, x = var_1975_cast_fp16)[name = tensor("op_2259_cast_fp16")]; + tensor var_2266_begin_0 = const()[name = tensor("op_2266_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2266_end_0 = const()[name = tensor("op_2266_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2266_end_mask_0 = const()[name = tensor("op_2266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2266_cast_fp16 = slice_by_index(begin = var_2266_begin_0, end = var_2266_end_0, end_mask = var_2266_end_mask_0, x = var_1975_cast_fp16)[name = tensor("op_2266_cast_fp16")]; + tensor var_2273_begin_0 = const()[name = tensor("op_2273_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2273_end_0 = const()[name = tensor("op_2273_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2273_end_mask_0 = const()[name = tensor("op_2273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2273_cast_fp16 = slice_by_index(begin = var_2273_begin_0, end = var_2273_end_0, end_mask = var_2273_end_mask_0, x = var_1975_cast_fp16)[name = tensor("op_2273_cast_fp16")]; + tensor var_2280_begin_0 = const()[name = tensor("op_2280_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2280_end_0 = const()[name = tensor("op_2280_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2280_end_mask_0 = const()[name = tensor("op_2280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2280_cast_fp16 = slice_by_index(begin = var_2280_begin_0, end = var_2280_end_0, end_mask = var_2280_end_mask_0, x = var_1979_cast_fp16)[name = tensor("op_2280_cast_fp16")]; + tensor var_2287_begin_0 = const()[name = tensor("op_2287_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2287_end_0 = const()[name = tensor("op_2287_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2287_end_mask_0 = const()[name = tensor("op_2287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2287_cast_fp16 = slice_by_index(begin = var_2287_begin_0, end = var_2287_end_0, end_mask = var_2287_end_mask_0, x = var_1979_cast_fp16)[name = tensor("op_2287_cast_fp16")]; + tensor var_2294_begin_0 = const()[name = tensor("op_2294_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2294_end_0 = const()[name = tensor("op_2294_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2294_end_mask_0 = const()[name = tensor("op_2294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2294_cast_fp16 = slice_by_index(begin = var_2294_begin_0, end = var_2294_end_0, end_mask = var_2294_end_mask_0, x = var_1979_cast_fp16)[name = tensor("op_2294_cast_fp16")]; + tensor var_2301_begin_0 = const()[name = tensor("op_2301_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2301_end_0 = const()[name = tensor("op_2301_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2301_end_mask_0 = const()[name = tensor("op_2301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2301_cast_fp16 = slice_by_index(begin = var_2301_begin_0, end = var_2301_end_0, end_mask = var_2301_end_mask_0, x = var_1979_cast_fp16)[name = tensor("op_2301_cast_fp16")]; + tensor var_2308_begin_0 = const()[name = tensor("op_2308_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2308_end_0 = const()[name = tensor("op_2308_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2308_end_mask_0 = const()[name = tensor("op_2308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2308_cast_fp16 = slice_by_index(begin = var_2308_begin_0, end = var_2308_end_0, end_mask = var_2308_end_mask_0, x = var_1983_cast_fp16)[name = tensor("op_2308_cast_fp16")]; + tensor var_2315_begin_0 = const()[name = tensor("op_2315_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2315_end_0 = const()[name = tensor("op_2315_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2315_end_mask_0 = const()[name = tensor("op_2315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2315_cast_fp16 = slice_by_index(begin = var_2315_begin_0, end = var_2315_end_0, end_mask = var_2315_end_mask_0, x = var_1983_cast_fp16)[name = tensor("op_2315_cast_fp16")]; + tensor var_2322_begin_0 = const()[name = tensor("op_2322_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2322_end_0 = const()[name = tensor("op_2322_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2322_end_mask_0 = const()[name = tensor("op_2322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2322_cast_fp16 = slice_by_index(begin = var_2322_begin_0, end = var_2322_end_0, end_mask = var_2322_end_mask_0, x = var_1983_cast_fp16)[name = tensor("op_2322_cast_fp16")]; + tensor var_2329_begin_0 = const()[name = tensor("op_2329_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2329_end_0 = const()[name = tensor("op_2329_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2329_end_mask_0 = const()[name = tensor("op_2329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2329_cast_fp16 = slice_by_index(begin = var_2329_begin_0, end = var_2329_end_0, end_mask = var_2329_end_mask_0, x = var_1983_cast_fp16)[name = tensor("op_2329_cast_fp16")]; + tensor var_2336_begin_0 = const()[name = tensor("op_2336_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2336_end_0 = const()[name = tensor("op_2336_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2336_end_mask_0 = const()[name = tensor("op_2336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2336_cast_fp16 = slice_by_index(begin = var_2336_begin_0, end = var_2336_end_0, end_mask = var_2336_end_mask_0, x = var_1987_cast_fp16)[name = tensor("op_2336_cast_fp16")]; + tensor var_2343_begin_0 = const()[name = tensor("op_2343_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2343_end_0 = const()[name = tensor("op_2343_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2343_end_mask_0 = const()[name = tensor("op_2343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2343_cast_fp16 = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = var_1987_cast_fp16)[name = tensor("op_2343_cast_fp16")]; + tensor var_2350_begin_0 = const()[name = tensor("op_2350_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2350_end_0 = const()[name = tensor("op_2350_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2350_end_mask_0 = const()[name = tensor("op_2350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2350_cast_fp16 = slice_by_index(begin = var_2350_begin_0, end = var_2350_end_0, end_mask = var_2350_end_mask_0, x = var_1987_cast_fp16)[name = tensor("op_2350_cast_fp16")]; + tensor var_2357_begin_0 = const()[name = tensor("op_2357_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2357_end_0 = const()[name = tensor("op_2357_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2357_end_mask_0 = const()[name = tensor("op_2357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2357_cast_fp16 = slice_by_index(begin = var_2357_begin_0, end = var_2357_end_0, end_mask = var_2357_end_mask_0, x = var_1987_cast_fp16)[name = tensor("op_2357_cast_fp16")]; + tensor var_2364_begin_0 = const()[name = tensor("op_2364_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2364_end_0 = const()[name = tensor("op_2364_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2364_end_mask_0 = const()[name = tensor("op_2364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2364_cast_fp16 = slice_by_index(begin = var_2364_begin_0, end = var_2364_end_0, end_mask = var_2364_end_mask_0, x = var_1991_cast_fp16)[name = tensor("op_2364_cast_fp16")]; + tensor var_2371_begin_0 = const()[name = tensor("op_2371_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2371_end_0 = const()[name = tensor("op_2371_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2371_end_mask_0 = const()[name = tensor("op_2371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2371_cast_fp16 = slice_by_index(begin = var_2371_begin_0, end = var_2371_end_0, end_mask = var_2371_end_mask_0, x = var_1991_cast_fp16)[name = tensor("op_2371_cast_fp16")]; + tensor var_2378_begin_0 = const()[name = tensor("op_2378_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2378_end_0 = const()[name = tensor("op_2378_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2378_end_mask_0 = const()[name = tensor("op_2378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2378_cast_fp16 = slice_by_index(begin = var_2378_begin_0, end = var_2378_end_0, end_mask = var_2378_end_mask_0, x = var_1991_cast_fp16)[name = tensor("op_2378_cast_fp16")]; + tensor var_2385_begin_0 = const()[name = tensor("op_2385_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2385_end_0 = const()[name = tensor("op_2385_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2385_end_mask_0 = const()[name = tensor("op_2385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2385_cast_fp16 = slice_by_index(begin = var_2385_begin_0, end = var_2385_end_0, end_mask = var_2385_end_mask_0, x = var_1991_cast_fp16)[name = tensor("op_2385_cast_fp16")]; + tensor var_2392_begin_0 = const()[name = tensor("op_2392_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2392_end_0 = const()[name = tensor("op_2392_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2392_end_mask_0 = const()[name = tensor("op_2392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2392_cast_fp16 = slice_by_index(begin = var_2392_begin_0, end = var_2392_end_0, end_mask = var_2392_end_mask_0, x = var_1995_cast_fp16)[name = tensor("op_2392_cast_fp16")]; + tensor var_2399_begin_0 = const()[name = tensor("op_2399_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2399_end_0 = const()[name = tensor("op_2399_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2399_end_mask_0 = const()[name = tensor("op_2399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2399_cast_fp16 = slice_by_index(begin = var_2399_begin_0, end = var_2399_end_0, end_mask = var_2399_end_mask_0, x = var_1995_cast_fp16)[name = tensor("op_2399_cast_fp16")]; + tensor var_2406_begin_0 = const()[name = tensor("op_2406_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2406_end_0 = const()[name = tensor("op_2406_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2406_end_mask_0 = const()[name = tensor("op_2406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2406_cast_fp16 = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = var_1995_cast_fp16)[name = tensor("op_2406_cast_fp16")]; + tensor var_2413_begin_0 = const()[name = tensor("op_2413_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2413_end_0 = const()[name = tensor("op_2413_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2413_end_mask_0 = const()[name = tensor("op_2413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2413_cast_fp16 = slice_by_index(begin = var_2413_begin_0, end = var_2413_end_0, end_mask = var_2413_end_mask_0, x = var_1995_cast_fp16)[name = tensor("op_2413_cast_fp16")]; + tensor var_2420_begin_0 = const()[name = tensor("op_2420_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2420_end_0 = const()[name = tensor("op_2420_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2420_end_mask_0 = const()[name = tensor("op_2420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2420_cast_fp16 = slice_by_index(begin = var_2420_begin_0, end = var_2420_end_0, end_mask = var_2420_end_mask_0, x = var_1999_cast_fp16)[name = tensor("op_2420_cast_fp16")]; + tensor var_2427_begin_0 = const()[name = tensor("op_2427_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2427_end_0 = const()[name = tensor("op_2427_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2427_end_mask_0 = const()[name = tensor("op_2427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2427_cast_fp16 = slice_by_index(begin = var_2427_begin_0, end = var_2427_end_0, end_mask = var_2427_end_mask_0, x = var_1999_cast_fp16)[name = tensor("op_2427_cast_fp16")]; + tensor var_2434_begin_0 = const()[name = tensor("op_2434_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2434_end_0 = const()[name = tensor("op_2434_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2434_end_mask_0 = const()[name = tensor("op_2434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2434_cast_fp16 = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = var_1999_cast_fp16)[name = tensor("op_2434_cast_fp16")]; + tensor var_2441_begin_0 = const()[name = tensor("op_2441_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2441_end_0 = const()[name = tensor("op_2441_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2441_end_mask_0 = const()[name = tensor("op_2441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2441_cast_fp16 = slice_by_index(begin = var_2441_begin_0, end = var_2441_end_0, end_mask = var_2441_end_mask_0, x = var_1999_cast_fp16)[name = tensor("op_2441_cast_fp16")]; + tensor var_2448_begin_0 = const()[name = tensor("op_2448_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2448_end_0 = const()[name = tensor("op_2448_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2448_end_mask_0 = const()[name = tensor("op_2448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2448_cast_fp16 = slice_by_index(begin = var_2448_begin_0, end = var_2448_end_0, end_mask = var_2448_end_mask_0, x = var_2003_cast_fp16)[name = tensor("op_2448_cast_fp16")]; + tensor var_2455_begin_0 = const()[name = tensor("op_2455_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2455_end_0 = const()[name = tensor("op_2455_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2455_end_mask_0 = const()[name = tensor("op_2455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2455_cast_fp16 = slice_by_index(begin = var_2455_begin_0, end = var_2455_end_0, end_mask = var_2455_end_mask_0, x = var_2003_cast_fp16)[name = tensor("op_2455_cast_fp16")]; + tensor var_2462_begin_0 = const()[name = tensor("op_2462_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2462_end_0 = const()[name = tensor("op_2462_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2462_end_mask_0 = const()[name = tensor("op_2462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2462_cast_fp16 = slice_by_index(begin = var_2462_begin_0, end = var_2462_end_0, end_mask = var_2462_end_mask_0, x = var_2003_cast_fp16)[name = tensor("op_2462_cast_fp16")]; + tensor var_2469_begin_0 = const()[name = tensor("op_2469_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2469_end_0 = const()[name = tensor("op_2469_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2469_end_mask_0 = const()[name = tensor("op_2469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2469_cast_fp16 = slice_by_index(begin = var_2469_begin_0, end = var_2469_end_0, end_mask = var_2469_end_mask_0, x = var_2003_cast_fp16)[name = tensor("op_2469_cast_fp16")]; + tensor var_2476_begin_0 = const()[name = tensor("op_2476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2476_end_0 = const()[name = tensor("op_2476_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2476_end_mask_0 = const()[name = tensor("op_2476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2476_cast_fp16 = slice_by_index(begin = var_2476_begin_0, end = var_2476_end_0, end_mask = var_2476_end_mask_0, x = var_2007_cast_fp16)[name = tensor("op_2476_cast_fp16")]; + tensor var_2483_begin_0 = const()[name = tensor("op_2483_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2483_end_0 = const()[name = tensor("op_2483_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2483_end_mask_0 = const()[name = tensor("op_2483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2483_cast_fp16 = slice_by_index(begin = var_2483_begin_0, end = var_2483_end_0, end_mask = var_2483_end_mask_0, x = var_2007_cast_fp16)[name = tensor("op_2483_cast_fp16")]; + tensor var_2490_begin_0 = const()[name = tensor("op_2490_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2490_end_0 = const()[name = tensor("op_2490_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2490_end_mask_0 = const()[name = tensor("op_2490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2490_cast_fp16 = slice_by_index(begin = var_2490_begin_0, end = var_2490_end_0, end_mask = var_2490_end_mask_0, x = var_2007_cast_fp16)[name = tensor("op_2490_cast_fp16")]; + tensor var_2497_begin_0 = const()[name = tensor("op_2497_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2497_end_0 = const()[name = tensor("op_2497_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2497_end_mask_0 = const()[name = tensor("op_2497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2497_cast_fp16 = slice_by_index(begin = var_2497_begin_0, end = var_2497_end_0, end_mask = var_2497_end_mask_0, x = var_2007_cast_fp16)[name = tensor("op_2497_cast_fp16")]; + tensor var_2504_begin_0 = const()[name = tensor("op_2504_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2504_end_0 = const()[name = tensor("op_2504_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2504_end_mask_0 = const()[name = tensor("op_2504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2504_cast_fp16 = slice_by_index(begin = var_2504_begin_0, end = var_2504_end_0, end_mask = var_2504_end_mask_0, x = var_2011_cast_fp16)[name = tensor("op_2504_cast_fp16")]; + tensor var_2511_begin_0 = const()[name = tensor("op_2511_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2511_end_0 = const()[name = tensor("op_2511_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2511_end_mask_0 = const()[name = tensor("op_2511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2511_cast_fp16 = slice_by_index(begin = var_2511_begin_0, end = var_2511_end_0, end_mask = var_2511_end_mask_0, x = var_2011_cast_fp16)[name = tensor("op_2511_cast_fp16")]; + tensor var_2518_begin_0 = const()[name = tensor("op_2518_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2518_end_0 = const()[name = tensor("op_2518_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2518_end_mask_0 = const()[name = tensor("op_2518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2518_cast_fp16 = slice_by_index(begin = var_2518_begin_0, end = var_2518_end_0, end_mask = var_2518_end_mask_0, x = var_2011_cast_fp16)[name = tensor("op_2518_cast_fp16")]; + tensor var_2525_begin_0 = const()[name = tensor("op_2525_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2525_end_0 = const()[name = tensor("op_2525_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2525_end_mask_0 = const()[name = tensor("op_2525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2525_cast_fp16 = slice_by_index(begin = var_2525_begin_0, end = var_2525_end_0, end_mask = var_2525_end_mask_0, x = var_2011_cast_fp16)[name = tensor("op_2525_cast_fp16")]; + tensor var_2532_begin_0 = const()[name = tensor("op_2532_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2532_end_0 = const()[name = tensor("op_2532_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2532_end_mask_0 = const()[name = tensor("op_2532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2532_cast_fp16 = slice_by_index(begin = var_2532_begin_0, end = var_2532_end_0, end_mask = var_2532_end_mask_0, x = var_2015_cast_fp16)[name = tensor("op_2532_cast_fp16")]; + tensor var_2539_begin_0 = const()[name = tensor("op_2539_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2539_end_0 = const()[name = tensor("op_2539_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2539_end_mask_0 = const()[name = tensor("op_2539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2539_cast_fp16 = slice_by_index(begin = var_2539_begin_0, end = var_2539_end_0, end_mask = var_2539_end_mask_0, x = var_2015_cast_fp16)[name = tensor("op_2539_cast_fp16")]; + tensor var_2546_begin_0 = const()[name = tensor("op_2546_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2546_end_0 = const()[name = tensor("op_2546_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2546_end_mask_0 = const()[name = tensor("op_2546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2546_cast_fp16 = slice_by_index(begin = var_2546_begin_0, end = var_2546_end_0, end_mask = var_2546_end_mask_0, x = var_2015_cast_fp16)[name = tensor("op_2546_cast_fp16")]; + tensor var_2553_begin_0 = const()[name = tensor("op_2553_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2553_end_0 = const()[name = tensor("op_2553_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2553_end_mask_0 = const()[name = tensor("op_2553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2553_cast_fp16 = slice_by_index(begin = var_2553_begin_0, end = var_2553_end_0, end_mask = var_2553_end_mask_0, x = var_2015_cast_fp16)[name = tensor("op_2553_cast_fp16")]; + tensor var_2560_begin_0 = const()[name = tensor("op_2560_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2560_end_0 = const()[name = tensor("op_2560_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2560_end_mask_0 = const()[name = tensor("op_2560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2560_cast_fp16 = slice_by_index(begin = var_2560_begin_0, end = var_2560_end_0, end_mask = var_2560_end_mask_0, x = var_2019_cast_fp16)[name = tensor("op_2560_cast_fp16")]; + tensor var_2567_begin_0 = const()[name = tensor("op_2567_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2567_end_0 = const()[name = tensor("op_2567_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2567_end_mask_0 = const()[name = tensor("op_2567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2567_cast_fp16 = slice_by_index(begin = var_2567_begin_0, end = var_2567_end_0, end_mask = var_2567_end_mask_0, x = var_2019_cast_fp16)[name = tensor("op_2567_cast_fp16")]; + tensor var_2574_begin_0 = const()[name = tensor("op_2574_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2574_end_0 = const()[name = tensor("op_2574_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2574_end_mask_0 = const()[name = tensor("op_2574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2574_cast_fp16 = slice_by_index(begin = var_2574_begin_0, end = var_2574_end_0, end_mask = var_2574_end_mask_0, x = var_2019_cast_fp16)[name = tensor("op_2574_cast_fp16")]; + tensor var_2581_begin_0 = const()[name = tensor("op_2581_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2581_end_0 = const()[name = tensor("op_2581_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2581_end_mask_0 = const()[name = tensor("op_2581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2581_cast_fp16 = slice_by_index(begin = var_2581_begin_0, end = var_2581_end_0, end_mask = var_2581_end_mask_0, x = var_2019_cast_fp16)[name = tensor("op_2581_cast_fp16")]; + tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2586_begin_0 = const()[name = tensor("op_2586_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2586_end_0 = const()[name = tensor("op_2586_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2586_end_mask_0 = const()[name = tensor("op_2586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_30 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = tensor("transpose_30")]; + tensor var_2586_cast_fp16 = slice_by_index(begin = var_2586_begin_0, end = var_2586_end_0, end_mask = var_2586_end_mask_0, x = transpose_30)[name = tensor("op_2586_cast_fp16")]; + tensor var_2590_begin_0 = const()[name = tensor("op_2590_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2590_end_0 = const()[name = tensor("op_2590_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2590_end_mask_0 = const()[name = tensor("op_2590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2590_cast_fp16 = slice_by_index(begin = var_2590_begin_0, end = var_2590_end_0, end_mask = var_2590_end_mask_0, x = transpose_30)[name = tensor("op_2590_cast_fp16")]; + tensor var_2594_begin_0 = const()[name = tensor("op_2594_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2594_end_0 = const()[name = tensor("op_2594_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2594_end_mask_0 = const()[name = tensor("op_2594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2594_cast_fp16 = slice_by_index(begin = var_2594_begin_0, end = var_2594_end_0, end_mask = var_2594_end_mask_0, x = transpose_30)[name = tensor("op_2594_cast_fp16")]; + tensor var_2598_begin_0 = const()[name = tensor("op_2598_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2598_end_0 = const()[name = tensor("op_2598_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2598_end_mask_0 = const()[name = tensor("op_2598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2598_cast_fp16 = slice_by_index(begin = var_2598_begin_0, end = var_2598_end_0, end_mask = var_2598_end_mask_0, x = transpose_30)[name = tensor("op_2598_cast_fp16")]; + tensor var_2602_begin_0 = const()[name = tensor("op_2602_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2602_end_0 = const()[name = tensor("op_2602_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2602_end_mask_0 = const()[name = tensor("op_2602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2602_cast_fp16 = slice_by_index(begin = var_2602_begin_0, end = var_2602_end_0, end_mask = var_2602_end_mask_0, x = transpose_30)[name = tensor("op_2602_cast_fp16")]; + tensor var_2606_begin_0 = const()[name = tensor("op_2606_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2606_end_0 = const()[name = tensor("op_2606_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2606_end_mask_0 = const()[name = tensor("op_2606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2606_cast_fp16 = slice_by_index(begin = var_2606_begin_0, end = var_2606_end_0, end_mask = var_2606_end_mask_0, x = transpose_30)[name = tensor("op_2606_cast_fp16")]; + tensor var_2610_begin_0 = const()[name = tensor("op_2610_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2610_end_0 = const()[name = tensor("op_2610_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2610_end_mask_0 = const()[name = tensor("op_2610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2610_cast_fp16 = slice_by_index(begin = var_2610_begin_0, end = var_2610_end_0, end_mask = var_2610_end_mask_0, x = transpose_30)[name = tensor("op_2610_cast_fp16")]; + tensor var_2614_begin_0 = const()[name = tensor("op_2614_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2614_end_0 = const()[name = tensor("op_2614_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2614_end_mask_0 = const()[name = tensor("op_2614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2614_cast_fp16 = slice_by_index(begin = var_2614_begin_0, end = var_2614_end_0, end_mask = var_2614_end_mask_0, x = transpose_30)[name = tensor("op_2614_cast_fp16")]; + tensor var_2618_begin_0 = const()[name = tensor("op_2618_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_2618_end_0 = const()[name = tensor("op_2618_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_2618_end_mask_0 = const()[name = tensor("op_2618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2618_cast_fp16 = slice_by_index(begin = var_2618_begin_0, end = var_2618_end_0, end_mask = var_2618_end_mask_0, x = transpose_30)[name = tensor("op_2618_cast_fp16")]; + tensor var_2622_begin_0 = const()[name = tensor("op_2622_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_2622_end_0 = const()[name = tensor("op_2622_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_2622_end_mask_0 = const()[name = tensor("op_2622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2622_cast_fp16 = slice_by_index(begin = var_2622_begin_0, end = var_2622_end_0, end_mask = var_2622_end_mask_0, x = transpose_30)[name = tensor("op_2622_cast_fp16")]; + tensor var_2626_begin_0 = const()[name = tensor("op_2626_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_2626_end_0 = const()[name = tensor("op_2626_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_2626_end_mask_0 = const()[name = tensor("op_2626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2626_cast_fp16 = slice_by_index(begin = var_2626_begin_0, end = var_2626_end_0, end_mask = var_2626_end_mask_0, x = transpose_30)[name = tensor("op_2626_cast_fp16")]; + tensor var_2630_begin_0 = const()[name = tensor("op_2630_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_2630_end_0 = const()[name = tensor("op_2630_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_2630_end_mask_0 = const()[name = tensor("op_2630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2630_cast_fp16 = slice_by_index(begin = var_2630_begin_0, end = var_2630_end_0, end_mask = var_2630_end_mask_0, x = transpose_30)[name = tensor("op_2630_cast_fp16")]; + tensor var_2634_begin_0 = const()[name = tensor("op_2634_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_2634_end_0 = const()[name = tensor("op_2634_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_2634_end_mask_0 = const()[name = tensor("op_2634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2634_cast_fp16 = slice_by_index(begin = var_2634_begin_0, end = var_2634_end_0, end_mask = var_2634_end_mask_0, x = transpose_30)[name = tensor("op_2634_cast_fp16")]; + tensor var_2638_begin_0 = const()[name = tensor("op_2638_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_2638_end_0 = const()[name = tensor("op_2638_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_2638_end_mask_0 = const()[name = tensor("op_2638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2638_cast_fp16 = slice_by_index(begin = var_2638_begin_0, end = var_2638_end_0, end_mask = var_2638_end_mask_0, x = transpose_30)[name = tensor("op_2638_cast_fp16")]; + tensor var_2642_begin_0 = const()[name = tensor("op_2642_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_2642_end_0 = const()[name = tensor("op_2642_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_2642_end_mask_0 = const()[name = tensor("op_2642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2642_cast_fp16 = slice_by_index(begin = var_2642_begin_0, end = var_2642_end_0, end_mask = var_2642_end_mask_0, x = transpose_30)[name = tensor("op_2642_cast_fp16")]; + tensor var_2646_begin_0 = const()[name = tensor("op_2646_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_2646_end_0 = const()[name = tensor("op_2646_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_2646_end_mask_0 = const()[name = tensor("op_2646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2646_cast_fp16 = slice_by_index(begin = var_2646_begin_0, end = var_2646_end_0, end_mask = var_2646_end_mask_0, x = transpose_30)[name = tensor("op_2646_cast_fp16")]; + tensor var_2650_begin_0 = const()[name = tensor("op_2650_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_2650_end_0 = const()[name = tensor("op_2650_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_2650_end_mask_0 = const()[name = tensor("op_2650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2650_cast_fp16 = slice_by_index(begin = var_2650_begin_0, end = var_2650_end_0, end_mask = var_2650_end_mask_0, x = transpose_30)[name = tensor("op_2650_cast_fp16")]; + tensor var_2654_begin_0 = const()[name = tensor("op_2654_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_2654_end_0 = const()[name = tensor("op_2654_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_2654_end_mask_0 = const()[name = tensor("op_2654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2654_cast_fp16 = slice_by_index(begin = var_2654_begin_0, end = var_2654_end_0, end_mask = var_2654_end_mask_0, x = transpose_30)[name = tensor("op_2654_cast_fp16")]; + tensor var_2658_begin_0 = const()[name = tensor("op_2658_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_2658_end_0 = const()[name = tensor("op_2658_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_2658_end_mask_0 = const()[name = tensor("op_2658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2658_cast_fp16 = slice_by_index(begin = var_2658_begin_0, end = var_2658_end_0, end_mask = var_2658_end_mask_0, x = transpose_30)[name = tensor("op_2658_cast_fp16")]; + tensor var_2662_begin_0 = const()[name = tensor("op_2662_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_2662_end_0 = const()[name = tensor("op_2662_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_2662_end_mask_0 = const()[name = tensor("op_2662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2662_cast_fp16 = slice_by_index(begin = var_2662_begin_0, end = var_2662_end_0, end_mask = var_2662_end_mask_0, x = transpose_30)[name = tensor("op_2662_cast_fp16")]; + tensor var_2664_begin_0 = const()[name = tensor("op_2664_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2664_end_0 = const()[name = tensor("op_2664_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2664_end_mask_0 = const()[name = tensor("op_2664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2664_cast_fp16 = slice_by_index(begin = var_2664_begin_0, end = var_2664_end_0, end_mask = var_2664_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2664_cast_fp16")]; + tensor var_2668_begin_0 = const()[name = tensor("op_2668_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2668_end_0 = const()[name = tensor("op_2668_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2668_end_mask_0 = const()[name = tensor("op_2668_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2668_cast_fp16 = slice_by_index(begin = var_2668_begin_0, end = var_2668_end_0, end_mask = var_2668_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2668_cast_fp16")]; + tensor var_2672_begin_0 = const()[name = tensor("op_2672_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2672_end_0 = const()[name = tensor("op_2672_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2672_end_mask_0 = const()[name = tensor("op_2672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2672_cast_fp16 = slice_by_index(begin = var_2672_begin_0, end = var_2672_end_0, end_mask = var_2672_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2672_cast_fp16")]; + tensor var_2676_begin_0 = const()[name = tensor("op_2676_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2676_end_0 = const()[name = tensor("op_2676_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2676_end_mask_0 = const()[name = tensor("op_2676_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2676_cast_fp16 = slice_by_index(begin = var_2676_begin_0, end = var_2676_end_0, end_mask = var_2676_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2676_cast_fp16")]; + tensor var_2680_begin_0 = const()[name = tensor("op_2680_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2680_end_0 = const()[name = tensor("op_2680_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2680_end_mask_0 = const()[name = tensor("op_2680_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2680_cast_fp16 = slice_by_index(begin = var_2680_begin_0, end = var_2680_end_0, end_mask = var_2680_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2680_cast_fp16")]; + tensor var_2684_begin_0 = const()[name = tensor("op_2684_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2684_end_0 = const()[name = tensor("op_2684_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2684_end_mask_0 = const()[name = tensor("op_2684_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2684_cast_fp16 = slice_by_index(begin = var_2684_begin_0, end = var_2684_end_0, end_mask = var_2684_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2684_cast_fp16")]; + tensor var_2688_begin_0 = const()[name = tensor("op_2688_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2688_end_0 = const()[name = tensor("op_2688_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2688_end_mask_0 = const()[name = tensor("op_2688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2688_cast_fp16 = slice_by_index(begin = var_2688_begin_0, end = var_2688_end_0, end_mask = var_2688_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2688_cast_fp16")]; + tensor var_2692_begin_0 = const()[name = tensor("op_2692_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2692_end_0 = const()[name = tensor("op_2692_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2692_end_mask_0 = const()[name = tensor("op_2692_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2692_cast_fp16 = slice_by_index(begin = var_2692_begin_0, end = var_2692_end_0, end_mask = var_2692_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2692_cast_fp16")]; + tensor var_2696_begin_0 = const()[name = tensor("op_2696_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_2696_end_0 = const()[name = tensor("op_2696_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_2696_end_mask_0 = const()[name = tensor("op_2696_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2696_cast_fp16 = slice_by_index(begin = var_2696_begin_0, end = var_2696_end_0, end_mask = var_2696_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2696_cast_fp16")]; + tensor var_2700_begin_0 = const()[name = tensor("op_2700_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_2700_end_0 = const()[name = tensor("op_2700_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_2700_end_mask_0 = const()[name = tensor("op_2700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2700_cast_fp16 = slice_by_index(begin = var_2700_begin_0, end = var_2700_end_0, end_mask = var_2700_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2700_cast_fp16")]; + tensor var_2704_begin_0 = const()[name = tensor("op_2704_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_2704_end_0 = const()[name = tensor("op_2704_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_2704_end_mask_0 = const()[name = tensor("op_2704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2704_cast_fp16 = slice_by_index(begin = var_2704_begin_0, end = var_2704_end_0, end_mask = var_2704_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2704_cast_fp16")]; + tensor var_2708_begin_0 = const()[name = tensor("op_2708_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_2708_end_0 = const()[name = tensor("op_2708_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_2708_end_mask_0 = const()[name = tensor("op_2708_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2708_cast_fp16 = slice_by_index(begin = var_2708_begin_0, end = var_2708_end_0, end_mask = var_2708_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2708_cast_fp16")]; + tensor var_2712_begin_0 = const()[name = tensor("op_2712_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_2712_end_0 = const()[name = tensor("op_2712_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_2712_end_mask_0 = const()[name = tensor("op_2712_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2712_cast_fp16 = slice_by_index(begin = var_2712_begin_0, end = var_2712_end_0, end_mask = var_2712_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2712_cast_fp16")]; + tensor var_2716_begin_0 = const()[name = tensor("op_2716_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_2716_end_0 = const()[name = tensor("op_2716_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_2716_end_mask_0 = const()[name = tensor("op_2716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2716_cast_fp16 = slice_by_index(begin = var_2716_begin_0, end = var_2716_end_0, end_mask = var_2716_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2716_cast_fp16")]; + tensor var_2720_begin_0 = const()[name = tensor("op_2720_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_2720_end_0 = const()[name = tensor("op_2720_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_2720_end_mask_0 = const()[name = tensor("op_2720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2720_cast_fp16 = slice_by_index(begin = var_2720_begin_0, end = var_2720_end_0, end_mask = var_2720_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2720_cast_fp16")]; + tensor var_2724_begin_0 = const()[name = tensor("op_2724_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_2724_end_0 = const()[name = tensor("op_2724_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_2724_end_mask_0 = const()[name = tensor("op_2724_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2724_cast_fp16 = slice_by_index(begin = var_2724_begin_0, end = var_2724_end_0, end_mask = var_2724_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2724_cast_fp16")]; + tensor var_2728_begin_0 = const()[name = tensor("op_2728_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_2728_end_0 = const()[name = tensor("op_2728_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_2728_end_mask_0 = const()[name = tensor("op_2728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2728_cast_fp16 = slice_by_index(begin = var_2728_begin_0, end = var_2728_end_0, end_mask = var_2728_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2728_cast_fp16")]; + tensor var_2732_begin_0 = const()[name = tensor("op_2732_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_2732_end_0 = const()[name = tensor("op_2732_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_2732_end_mask_0 = const()[name = tensor("op_2732_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2732_cast_fp16 = slice_by_index(begin = var_2732_begin_0, end = var_2732_end_0, end_mask = var_2732_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2732_cast_fp16")]; + tensor var_2736_begin_0 = const()[name = tensor("op_2736_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_2736_end_0 = const()[name = tensor("op_2736_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_2736_end_mask_0 = const()[name = tensor("op_2736_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2736_cast_fp16 = slice_by_index(begin = var_2736_begin_0, end = var_2736_end_0, end_mask = var_2736_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2736_cast_fp16")]; + tensor var_2740_begin_0 = const()[name = tensor("op_2740_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_2740_end_0 = const()[name = tensor("op_2740_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_2740_end_mask_0 = const()[name = tensor("op_2740_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2740_cast_fp16 = slice_by_index(begin = var_2740_begin_0, end = var_2740_end_0, end_mask = var_2740_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_2740_cast_fp16")]; + tensor var_2744_equation_0 = const()[name = tensor("op_2744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2744_cast_fp16 = einsum(equation = var_2744_equation_0, values = (var_2586_cast_fp16, var_2028_cast_fp16))[name = tensor("op_2744_cast_fp16")]; + tensor var_2745_to_fp16 = const()[name = tensor("op_2745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = var_2744_cast_fp16, y = var_2745_to_fp16)[name = tensor("aw_chunk_161_cast_fp16")]; + tensor var_2748_equation_0 = const()[name = tensor("op_2748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2748_cast_fp16 = einsum(equation = var_2748_equation_0, values = (var_2586_cast_fp16, var_2035_cast_fp16))[name = tensor("op_2748_cast_fp16")]; + tensor var_2749_to_fp16 = const()[name = tensor("op_2749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = var_2748_cast_fp16, y = var_2749_to_fp16)[name = tensor("aw_chunk_163_cast_fp16")]; + tensor var_2752_equation_0 = const()[name = tensor("op_2752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2752_cast_fp16 = einsum(equation = var_2752_equation_0, values = (var_2586_cast_fp16, var_2042_cast_fp16))[name = tensor("op_2752_cast_fp16")]; + tensor var_2753_to_fp16 = const()[name = tensor("op_2753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = var_2752_cast_fp16, y = var_2753_to_fp16)[name = tensor("aw_chunk_165_cast_fp16")]; + tensor var_2756_equation_0 = const()[name = tensor("op_2756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2756_cast_fp16 = einsum(equation = var_2756_equation_0, values = (var_2586_cast_fp16, var_2049_cast_fp16))[name = tensor("op_2756_cast_fp16")]; + tensor var_2757_to_fp16 = const()[name = tensor("op_2757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = var_2756_cast_fp16, y = var_2757_to_fp16)[name = tensor("aw_chunk_167_cast_fp16")]; + tensor var_2760_equation_0 = const()[name = tensor("op_2760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2760_cast_fp16 = einsum(equation = var_2760_equation_0, values = (var_2590_cast_fp16, var_2056_cast_fp16))[name = tensor("op_2760_cast_fp16")]; + tensor var_2761_to_fp16 = const()[name = tensor("op_2761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = var_2760_cast_fp16, y = var_2761_to_fp16)[name = tensor("aw_chunk_169_cast_fp16")]; + tensor var_2764_equation_0 = const()[name = tensor("op_2764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2764_cast_fp16 = einsum(equation = var_2764_equation_0, values = (var_2590_cast_fp16, var_2063_cast_fp16))[name = tensor("op_2764_cast_fp16")]; + tensor var_2765_to_fp16 = const()[name = tensor("op_2765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = var_2764_cast_fp16, y = var_2765_to_fp16)[name = tensor("aw_chunk_171_cast_fp16")]; + tensor var_2768_equation_0 = const()[name = tensor("op_2768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2768_cast_fp16 = einsum(equation = var_2768_equation_0, values = (var_2590_cast_fp16, var_2070_cast_fp16))[name = tensor("op_2768_cast_fp16")]; + tensor var_2769_to_fp16 = const()[name = tensor("op_2769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = var_2768_cast_fp16, y = var_2769_to_fp16)[name = tensor("aw_chunk_173_cast_fp16")]; + tensor var_2772_equation_0 = const()[name = tensor("op_2772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2772_cast_fp16 = einsum(equation = var_2772_equation_0, values = (var_2590_cast_fp16, var_2077_cast_fp16))[name = tensor("op_2772_cast_fp16")]; + tensor var_2773_to_fp16 = const()[name = tensor("op_2773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = var_2772_cast_fp16, y = var_2773_to_fp16)[name = tensor("aw_chunk_175_cast_fp16")]; + tensor var_2776_equation_0 = const()[name = tensor("op_2776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2776_cast_fp16 = einsum(equation = var_2776_equation_0, values = (var_2594_cast_fp16, var_2084_cast_fp16))[name = tensor("op_2776_cast_fp16")]; + tensor var_2777_to_fp16 = const()[name = tensor("op_2777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = var_2776_cast_fp16, y = var_2777_to_fp16)[name = tensor("aw_chunk_177_cast_fp16")]; + tensor var_2780_equation_0 = const()[name = tensor("op_2780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2780_cast_fp16 = einsum(equation = var_2780_equation_0, values = (var_2594_cast_fp16, var_2091_cast_fp16))[name = tensor("op_2780_cast_fp16")]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = var_2780_cast_fp16, y = var_2781_to_fp16)[name = tensor("aw_chunk_179_cast_fp16")]; + tensor var_2784_equation_0 = const()[name = tensor("op_2784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2784_cast_fp16 = einsum(equation = var_2784_equation_0, values = (var_2594_cast_fp16, var_2098_cast_fp16))[name = tensor("op_2784_cast_fp16")]; + tensor var_2785_to_fp16 = const()[name = tensor("op_2785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = var_2784_cast_fp16, y = var_2785_to_fp16)[name = tensor("aw_chunk_181_cast_fp16")]; + tensor var_2788_equation_0 = const()[name = tensor("op_2788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2788_cast_fp16 = einsum(equation = var_2788_equation_0, values = (var_2594_cast_fp16, var_2105_cast_fp16))[name = tensor("op_2788_cast_fp16")]; + tensor var_2789_to_fp16 = const()[name = tensor("op_2789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = var_2788_cast_fp16, y = var_2789_to_fp16)[name = tensor("aw_chunk_183_cast_fp16")]; + tensor var_2792_equation_0 = const()[name = tensor("op_2792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2792_cast_fp16 = einsum(equation = var_2792_equation_0, values = (var_2598_cast_fp16, var_2112_cast_fp16))[name = tensor("op_2792_cast_fp16")]; + tensor var_2793_to_fp16 = const()[name = tensor("op_2793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = var_2792_cast_fp16, y = var_2793_to_fp16)[name = tensor("aw_chunk_185_cast_fp16")]; + tensor var_2796_equation_0 = const()[name = tensor("op_2796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2796_cast_fp16 = einsum(equation = var_2796_equation_0, values = (var_2598_cast_fp16, var_2119_cast_fp16))[name = tensor("op_2796_cast_fp16")]; + tensor var_2797_to_fp16 = const()[name = tensor("op_2797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = var_2796_cast_fp16, y = var_2797_to_fp16)[name = tensor("aw_chunk_187_cast_fp16")]; + tensor var_2800_equation_0 = const()[name = tensor("op_2800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2800_cast_fp16 = einsum(equation = var_2800_equation_0, values = (var_2598_cast_fp16, var_2126_cast_fp16))[name = tensor("op_2800_cast_fp16")]; + tensor var_2801_to_fp16 = const()[name = tensor("op_2801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = var_2800_cast_fp16, y = var_2801_to_fp16)[name = tensor("aw_chunk_189_cast_fp16")]; + tensor var_2804_equation_0 = const()[name = tensor("op_2804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2804_cast_fp16 = einsum(equation = var_2804_equation_0, values = (var_2598_cast_fp16, var_2133_cast_fp16))[name = tensor("op_2804_cast_fp16")]; + tensor var_2805_to_fp16 = const()[name = tensor("op_2805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = var_2804_cast_fp16, y = var_2805_to_fp16)[name = tensor("aw_chunk_191_cast_fp16")]; + tensor var_2808_equation_0 = const()[name = tensor("op_2808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2808_cast_fp16 = einsum(equation = var_2808_equation_0, values = (var_2602_cast_fp16, var_2140_cast_fp16))[name = tensor("op_2808_cast_fp16")]; + tensor var_2809_to_fp16 = const()[name = tensor("op_2809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = var_2808_cast_fp16, y = var_2809_to_fp16)[name = tensor("aw_chunk_193_cast_fp16")]; + tensor var_2812_equation_0 = const()[name = tensor("op_2812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2812_cast_fp16 = einsum(equation = var_2812_equation_0, values = (var_2602_cast_fp16, var_2147_cast_fp16))[name = tensor("op_2812_cast_fp16")]; + tensor var_2813_to_fp16 = const()[name = tensor("op_2813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = var_2812_cast_fp16, y = var_2813_to_fp16)[name = tensor("aw_chunk_195_cast_fp16")]; + tensor var_2816_equation_0 = const()[name = tensor("op_2816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2816_cast_fp16 = einsum(equation = var_2816_equation_0, values = (var_2602_cast_fp16, var_2154_cast_fp16))[name = tensor("op_2816_cast_fp16")]; + tensor var_2817_to_fp16 = const()[name = tensor("op_2817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = var_2816_cast_fp16, y = var_2817_to_fp16)[name = tensor("aw_chunk_197_cast_fp16")]; + tensor var_2820_equation_0 = const()[name = tensor("op_2820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2820_cast_fp16 = einsum(equation = var_2820_equation_0, values = (var_2602_cast_fp16, var_2161_cast_fp16))[name = tensor("op_2820_cast_fp16")]; + tensor var_2821_to_fp16 = const()[name = tensor("op_2821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = var_2820_cast_fp16, y = var_2821_to_fp16)[name = tensor("aw_chunk_199_cast_fp16")]; + tensor var_2824_equation_0 = const()[name = tensor("op_2824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2824_cast_fp16 = einsum(equation = var_2824_equation_0, values = (var_2606_cast_fp16, var_2168_cast_fp16))[name = tensor("op_2824_cast_fp16")]; + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = var_2824_cast_fp16, y = var_2825_to_fp16)[name = tensor("aw_chunk_201_cast_fp16")]; + tensor var_2828_equation_0 = const()[name = tensor("op_2828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2828_cast_fp16 = einsum(equation = var_2828_equation_0, values = (var_2606_cast_fp16, var_2175_cast_fp16))[name = tensor("op_2828_cast_fp16")]; + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = var_2828_cast_fp16, y = var_2829_to_fp16)[name = tensor("aw_chunk_203_cast_fp16")]; + tensor var_2832_equation_0 = const()[name = tensor("op_2832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2832_cast_fp16 = einsum(equation = var_2832_equation_0, values = (var_2606_cast_fp16, var_2182_cast_fp16))[name = tensor("op_2832_cast_fp16")]; + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = var_2832_cast_fp16, y = var_2833_to_fp16)[name = tensor("aw_chunk_205_cast_fp16")]; + tensor var_2836_equation_0 = const()[name = tensor("op_2836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2836_cast_fp16 = einsum(equation = var_2836_equation_0, values = (var_2606_cast_fp16, var_2189_cast_fp16))[name = tensor("op_2836_cast_fp16")]; + tensor var_2837_to_fp16 = const()[name = tensor("op_2837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = var_2836_cast_fp16, y = var_2837_to_fp16)[name = tensor("aw_chunk_207_cast_fp16")]; + tensor var_2840_equation_0 = const()[name = tensor("op_2840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2840_cast_fp16 = einsum(equation = var_2840_equation_0, values = (var_2610_cast_fp16, var_2196_cast_fp16))[name = tensor("op_2840_cast_fp16")]; + tensor var_2841_to_fp16 = const()[name = tensor("op_2841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = var_2840_cast_fp16, y = var_2841_to_fp16)[name = tensor("aw_chunk_209_cast_fp16")]; + tensor var_2844_equation_0 = const()[name = tensor("op_2844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2844_cast_fp16 = einsum(equation = var_2844_equation_0, values = (var_2610_cast_fp16, var_2203_cast_fp16))[name = tensor("op_2844_cast_fp16")]; + tensor var_2845_to_fp16 = const()[name = tensor("op_2845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = var_2844_cast_fp16, y = var_2845_to_fp16)[name = tensor("aw_chunk_211_cast_fp16")]; + tensor var_2848_equation_0 = const()[name = tensor("op_2848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2848_cast_fp16 = einsum(equation = var_2848_equation_0, values = (var_2610_cast_fp16, var_2210_cast_fp16))[name = tensor("op_2848_cast_fp16")]; + tensor var_2849_to_fp16 = const()[name = tensor("op_2849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = var_2848_cast_fp16, y = var_2849_to_fp16)[name = tensor("aw_chunk_213_cast_fp16")]; + tensor var_2852_equation_0 = const()[name = tensor("op_2852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2852_cast_fp16 = einsum(equation = var_2852_equation_0, values = (var_2610_cast_fp16, var_2217_cast_fp16))[name = tensor("op_2852_cast_fp16")]; + tensor var_2853_to_fp16 = const()[name = tensor("op_2853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = var_2852_cast_fp16, y = var_2853_to_fp16)[name = tensor("aw_chunk_215_cast_fp16")]; + tensor var_2856_equation_0 = const()[name = tensor("op_2856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2856_cast_fp16 = einsum(equation = var_2856_equation_0, values = (var_2614_cast_fp16, var_2224_cast_fp16))[name = tensor("op_2856_cast_fp16")]; + tensor var_2857_to_fp16 = const()[name = tensor("op_2857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = var_2856_cast_fp16, y = var_2857_to_fp16)[name = tensor("aw_chunk_217_cast_fp16")]; + tensor var_2860_equation_0 = const()[name = tensor("op_2860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2860_cast_fp16 = einsum(equation = var_2860_equation_0, values = (var_2614_cast_fp16, var_2231_cast_fp16))[name = tensor("op_2860_cast_fp16")]; + tensor var_2861_to_fp16 = const()[name = tensor("op_2861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = var_2860_cast_fp16, y = var_2861_to_fp16)[name = tensor("aw_chunk_219_cast_fp16")]; + tensor var_2864_equation_0 = const()[name = tensor("op_2864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2864_cast_fp16 = einsum(equation = var_2864_equation_0, values = (var_2614_cast_fp16, var_2238_cast_fp16))[name = tensor("op_2864_cast_fp16")]; + tensor var_2865_to_fp16 = const()[name = tensor("op_2865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = var_2864_cast_fp16, y = var_2865_to_fp16)[name = tensor("aw_chunk_221_cast_fp16")]; + tensor var_2868_equation_0 = const()[name = tensor("op_2868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2868_cast_fp16 = einsum(equation = var_2868_equation_0, values = (var_2614_cast_fp16, var_2245_cast_fp16))[name = tensor("op_2868_cast_fp16")]; + tensor var_2869_to_fp16 = const()[name = tensor("op_2869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = var_2868_cast_fp16, y = var_2869_to_fp16)[name = tensor("aw_chunk_223_cast_fp16")]; + tensor var_2872_equation_0 = const()[name = tensor("op_2872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2872_cast_fp16 = einsum(equation = var_2872_equation_0, values = (var_2618_cast_fp16, var_2252_cast_fp16))[name = tensor("op_2872_cast_fp16")]; + tensor var_2873_to_fp16 = const()[name = tensor("op_2873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = var_2872_cast_fp16, y = var_2873_to_fp16)[name = tensor("aw_chunk_225_cast_fp16")]; + tensor var_2876_equation_0 = const()[name = tensor("op_2876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2876_cast_fp16 = einsum(equation = var_2876_equation_0, values = (var_2618_cast_fp16, var_2259_cast_fp16))[name = tensor("op_2876_cast_fp16")]; + tensor var_2877_to_fp16 = const()[name = tensor("op_2877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = var_2876_cast_fp16, y = var_2877_to_fp16)[name = tensor("aw_chunk_227_cast_fp16")]; + tensor var_2880_equation_0 = const()[name = tensor("op_2880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2880_cast_fp16 = einsum(equation = var_2880_equation_0, values = (var_2618_cast_fp16, var_2266_cast_fp16))[name = tensor("op_2880_cast_fp16")]; + tensor var_2881_to_fp16 = const()[name = tensor("op_2881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = var_2880_cast_fp16, y = var_2881_to_fp16)[name = tensor("aw_chunk_229_cast_fp16")]; + tensor var_2884_equation_0 = const()[name = tensor("op_2884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2884_cast_fp16 = einsum(equation = var_2884_equation_0, values = (var_2618_cast_fp16, var_2273_cast_fp16))[name = tensor("op_2884_cast_fp16")]; + tensor var_2885_to_fp16 = const()[name = tensor("op_2885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = var_2884_cast_fp16, y = var_2885_to_fp16)[name = tensor("aw_chunk_231_cast_fp16")]; + tensor var_2888_equation_0 = const()[name = tensor("op_2888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2888_cast_fp16 = einsum(equation = var_2888_equation_0, values = (var_2622_cast_fp16, var_2280_cast_fp16))[name = tensor("op_2888_cast_fp16")]; + tensor var_2889_to_fp16 = const()[name = tensor("op_2889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = var_2888_cast_fp16, y = var_2889_to_fp16)[name = tensor("aw_chunk_233_cast_fp16")]; + tensor var_2892_equation_0 = const()[name = tensor("op_2892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2892_cast_fp16 = einsum(equation = var_2892_equation_0, values = (var_2622_cast_fp16, var_2287_cast_fp16))[name = tensor("op_2892_cast_fp16")]; + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = var_2892_cast_fp16, y = var_2893_to_fp16)[name = tensor("aw_chunk_235_cast_fp16")]; + tensor var_2896_equation_0 = const()[name = tensor("op_2896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2896_cast_fp16 = einsum(equation = var_2896_equation_0, values = (var_2622_cast_fp16, var_2294_cast_fp16))[name = tensor("op_2896_cast_fp16")]; + tensor var_2897_to_fp16 = const()[name = tensor("op_2897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = var_2896_cast_fp16, y = var_2897_to_fp16)[name = tensor("aw_chunk_237_cast_fp16")]; + tensor var_2900_equation_0 = const()[name = tensor("op_2900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2900_cast_fp16 = einsum(equation = var_2900_equation_0, values = (var_2622_cast_fp16, var_2301_cast_fp16))[name = tensor("op_2900_cast_fp16")]; + tensor var_2901_to_fp16 = const()[name = tensor("op_2901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = var_2900_cast_fp16, y = var_2901_to_fp16)[name = tensor("aw_chunk_239_cast_fp16")]; + tensor var_2904_equation_0 = const()[name = tensor("op_2904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2904_cast_fp16 = einsum(equation = var_2904_equation_0, values = (var_2626_cast_fp16, var_2308_cast_fp16))[name = tensor("op_2904_cast_fp16")]; + tensor var_2905_to_fp16 = const()[name = tensor("op_2905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = var_2904_cast_fp16, y = var_2905_to_fp16)[name = tensor("aw_chunk_241_cast_fp16")]; + tensor var_2908_equation_0 = const()[name = tensor("op_2908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2908_cast_fp16 = einsum(equation = var_2908_equation_0, values = (var_2626_cast_fp16, var_2315_cast_fp16))[name = tensor("op_2908_cast_fp16")]; + tensor var_2909_to_fp16 = const()[name = tensor("op_2909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = var_2908_cast_fp16, y = var_2909_to_fp16)[name = tensor("aw_chunk_243_cast_fp16")]; + tensor var_2912_equation_0 = const()[name = tensor("op_2912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2912_cast_fp16 = einsum(equation = var_2912_equation_0, values = (var_2626_cast_fp16, var_2322_cast_fp16))[name = tensor("op_2912_cast_fp16")]; + tensor var_2913_to_fp16 = const()[name = tensor("op_2913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = var_2912_cast_fp16, y = var_2913_to_fp16)[name = tensor("aw_chunk_245_cast_fp16")]; + tensor var_2916_equation_0 = const()[name = tensor("op_2916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2916_cast_fp16 = einsum(equation = var_2916_equation_0, values = (var_2626_cast_fp16, var_2329_cast_fp16))[name = tensor("op_2916_cast_fp16")]; + tensor var_2917_to_fp16 = const()[name = tensor("op_2917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = var_2916_cast_fp16, y = var_2917_to_fp16)[name = tensor("aw_chunk_247_cast_fp16")]; + tensor var_2920_equation_0 = const()[name = tensor("op_2920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2920_cast_fp16 = einsum(equation = var_2920_equation_0, values = (var_2630_cast_fp16, var_2336_cast_fp16))[name = tensor("op_2920_cast_fp16")]; + tensor var_2921_to_fp16 = const()[name = tensor("op_2921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = var_2920_cast_fp16, y = var_2921_to_fp16)[name = tensor("aw_chunk_249_cast_fp16")]; + tensor var_2924_equation_0 = const()[name = tensor("op_2924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2924_cast_fp16 = einsum(equation = var_2924_equation_0, values = (var_2630_cast_fp16, var_2343_cast_fp16))[name = tensor("op_2924_cast_fp16")]; + tensor var_2925_to_fp16 = const()[name = tensor("op_2925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = var_2924_cast_fp16, y = var_2925_to_fp16)[name = tensor("aw_chunk_251_cast_fp16")]; + tensor var_2928_equation_0 = const()[name = tensor("op_2928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2928_cast_fp16 = einsum(equation = var_2928_equation_0, values = (var_2630_cast_fp16, var_2350_cast_fp16))[name = tensor("op_2928_cast_fp16")]; + tensor var_2929_to_fp16 = const()[name = tensor("op_2929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = var_2928_cast_fp16, y = var_2929_to_fp16)[name = tensor("aw_chunk_253_cast_fp16")]; + tensor var_2932_equation_0 = const()[name = tensor("op_2932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2932_cast_fp16 = einsum(equation = var_2932_equation_0, values = (var_2630_cast_fp16, var_2357_cast_fp16))[name = tensor("op_2932_cast_fp16")]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = var_2932_cast_fp16, y = var_2933_to_fp16)[name = tensor("aw_chunk_255_cast_fp16")]; + tensor var_2936_equation_0 = const()[name = tensor("op_2936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2936_cast_fp16 = einsum(equation = var_2936_equation_0, values = (var_2634_cast_fp16, var_2364_cast_fp16))[name = tensor("op_2936_cast_fp16")]; + tensor var_2937_to_fp16 = const()[name = tensor("op_2937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = var_2936_cast_fp16, y = var_2937_to_fp16)[name = tensor("aw_chunk_257_cast_fp16")]; + tensor var_2940_equation_0 = const()[name = tensor("op_2940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2940_cast_fp16 = einsum(equation = var_2940_equation_0, values = (var_2634_cast_fp16, var_2371_cast_fp16))[name = tensor("op_2940_cast_fp16")]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = var_2940_cast_fp16, y = var_2941_to_fp16)[name = tensor("aw_chunk_259_cast_fp16")]; + tensor var_2944_equation_0 = const()[name = tensor("op_2944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2944_cast_fp16 = einsum(equation = var_2944_equation_0, values = (var_2634_cast_fp16, var_2378_cast_fp16))[name = tensor("op_2944_cast_fp16")]; + tensor var_2945_to_fp16 = const()[name = tensor("op_2945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = var_2944_cast_fp16, y = var_2945_to_fp16)[name = tensor("aw_chunk_261_cast_fp16")]; + tensor var_2948_equation_0 = const()[name = tensor("op_2948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2948_cast_fp16 = einsum(equation = var_2948_equation_0, values = (var_2634_cast_fp16, var_2385_cast_fp16))[name = tensor("op_2948_cast_fp16")]; + tensor var_2949_to_fp16 = const()[name = tensor("op_2949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = var_2948_cast_fp16, y = var_2949_to_fp16)[name = tensor("aw_chunk_263_cast_fp16")]; + tensor var_2952_equation_0 = const()[name = tensor("op_2952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2952_cast_fp16 = einsum(equation = var_2952_equation_0, values = (var_2638_cast_fp16, var_2392_cast_fp16))[name = tensor("op_2952_cast_fp16")]; + tensor var_2953_to_fp16 = const()[name = tensor("op_2953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = var_2952_cast_fp16, y = var_2953_to_fp16)[name = tensor("aw_chunk_265_cast_fp16")]; + tensor var_2956_equation_0 = const()[name = tensor("op_2956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2956_cast_fp16 = einsum(equation = var_2956_equation_0, values = (var_2638_cast_fp16, var_2399_cast_fp16))[name = tensor("op_2956_cast_fp16")]; + tensor var_2957_to_fp16 = const()[name = tensor("op_2957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = var_2956_cast_fp16, y = var_2957_to_fp16)[name = tensor("aw_chunk_267_cast_fp16")]; + tensor var_2960_equation_0 = const()[name = tensor("op_2960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2960_cast_fp16 = einsum(equation = var_2960_equation_0, values = (var_2638_cast_fp16, var_2406_cast_fp16))[name = tensor("op_2960_cast_fp16")]; + tensor var_2961_to_fp16 = const()[name = tensor("op_2961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = var_2960_cast_fp16, y = var_2961_to_fp16)[name = tensor("aw_chunk_269_cast_fp16")]; + tensor var_2964_equation_0 = const()[name = tensor("op_2964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2964_cast_fp16 = einsum(equation = var_2964_equation_0, values = (var_2638_cast_fp16, var_2413_cast_fp16))[name = tensor("op_2964_cast_fp16")]; + tensor var_2965_to_fp16 = const()[name = tensor("op_2965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = var_2964_cast_fp16, y = var_2965_to_fp16)[name = tensor("aw_chunk_271_cast_fp16")]; + tensor var_2968_equation_0 = const()[name = tensor("op_2968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2968_cast_fp16 = einsum(equation = var_2968_equation_0, values = (var_2642_cast_fp16, var_2420_cast_fp16))[name = tensor("op_2968_cast_fp16")]; + tensor var_2969_to_fp16 = const()[name = tensor("op_2969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = var_2968_cast_fp16, y = var_2969_to_fp16)[name = tensor("aw_chunk_273_cast_fp16")]; + tensor var_2972_equation_0 = const()[name = tensor("op_2972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2972_cast_fp16 = einsum(equation = var_2972_equation_0, values = (var_2642_cast_fp16, var_2427_cast_fp16))[name = tensor("op_2972_cast_fp16")]; + tensor var_2973_to_fp16 = const()[name = tensor("op_2973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = var_2972_cast_fp16, y = var_2973_to_fp16)[name = tensor("aw_chunk_275_cast_fp16")]; + tensor var_2976_equation_0 = const()[name = tensor("op_2976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2976_cast_fp16 = einsum(equation = var_2976_equation_0, values = (var_2642_cast_fp16, var_2434_cast_fp16))[name = tensor("op_2976_cast_fp16")]; + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = var_2976_cast_fp16, y = var_2977_to_fp16)[name = tensor("aw_chunk_277_cast_fp16")]; + tensor var_2980_equation_0 = const()[name = tensor("op_2980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2980_cast_fp16 = einsum(equation = var_2980_equation_0, values = (var_2642_cast_fp16, var_2441_cast_fp16))[name = tensor("op_2980_cast_fp16")]; + tensor var_2981_to_fp16 = const()[name = tensor("op_2981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = var_2980_cast_fp16, y = var_2981_to_fp16)[name = tensor("aw_chunk_279_cast_fp16")]; + tensor var_2984_equation_0 = const()[name = tensor("op_2984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2984_cast_fp16 = einsum(equation = var_2984_equation_0, values = (var_2646_cast_fp16, var_2448_cast_fp16))[name = tensor("op_2984_cast_fp16")]; + tensor var_2985_to_fp16 = const()[name = tensor("op_2985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = var_2984_cast_fp16, y = var_2985_to_fp16)[name = tensor("aw_chunk_281_cast_fp16")]; + tensor var_2988_equation_0 = const()[name = tensor("op_2988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2988_cast_fp16 = einsum(equation = var_2988_equation_0, values = (var_2646_cast_fp16, var_2455_cast_fp16))[name = tensor("op_2988_cast_fp16")]; + tensor var_2989_to_fp16 = const()[name = tensor("op_2989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = var_2988_cast_fp16, y = var_2989_to_fp16)[name = tensor("aw_chunk_283_cast_fp16")]; + tensor var_2992_equation_0 = const()[name = tensor("op_2992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2992_cast_fp16 = einsum(equation = var_2992_equation_0, values = (var_2646_cast_fp16, var_2462_cast_fp16))[name = tensor("op_2992_cast_fp16")]; + tensor var_2993_to_fp16 = const()[name = tensor("op_2993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = var_2992_cast_fp16, y = var_2993_to_fp16)[name = tensor("aw_chunk_285_cast_fp16")]; + tensor var_2996_equation_0 = const()[name = tensor("op_2996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_2996_cast_fp16 = einsum(equation = var_2996_equation_0, values = (var_2646_cast_fp16, var_2469_cast_fp16))[name = tensor("op_2996_cast_fp16")]; + tensor var_2997_to_fp16 = const()[name = tensor("op_2997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = var_2996_cast_fp16, y = var_2997_to_fp16)[name = tensor("aw_chunk_287_cast_fp16")]; + tensor var_3000_equation_0 = const()[name = tensor("op_3000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3000_cast_fp16 = einsum(equation = var_3000_equation_0, values = (var_2650_cast_fp16, var_2476_cast_fp16))[name = tensor("op_3000_cast_fp16")]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = var_3000_cast_fp16, y = var_3001_to_fp16)[name = tensor("aw_chunk_289_cast_fp16")]; + tensor var_3004_equation_0 = const()[name = tensor("op_3004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3004_cast_fp16 = einsum(equation = var_3004_equation_0, values = (var_2650_cast_fp16, var_2483_cast_fp16))[name = tensor("op_3004_cast_fp16")]; + tensor var_3005_to_fp16 = const()[name = tensor("op_3005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = var_3004_cast_fp16, y = var_3005_to_fp16)[name = tensor("aw_chunk_291_cast_fp16")]; + tensor var_3008_equation_0 = const()[name = tensor("op_3008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3008_cast_fp16 = einsum(equation = var_3008_equation_0, values = (var_2650_cast_fp16, var_2490_cast_fp16))[name = tensor("op_3008_cast_fp16")]; + tensor var_3009_to_fp16 = const()[name = tensor("op_3009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = var_3008_cast_fp16, y = var_3009_to_fp16)[name = tensor("aw_chunk_293_cast_fp16")]; + tensor var_3012_equation_0 = const()[name = tensor("op_3012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3012_cast_fp16 = einsum(equation = var_3012_equation_0, values = (var_2650_cast_fp16, var_2497_cast_fp16))[name = tensor("op_3012_cast_fp16")]; + tensor var_3013_to_fp16 = const()[name = tensor("op_3013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = var_3012_cast_fp16, y = var_3013_to_fp16)[name = tensor("aw_chunk_295_cast_fp16")]; + tensor var_3016_equation_0 = const()[name = tensor("op_3016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3016_cast_fp16 = einsum(equation = var_3016_equation_0, values = (var_2654_cast_fp16, var_2504_cast_fp16))[name = tensor("op_3016_cast_fp16")]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = var_3016_cast_fp16, y = var_3017_to_fp16)[name = tensor("aw_chunk_297_cast_fp16")]; + tensor var_3020_equation_0 = const()[name = tensor("op_3020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3020_cast_fp16 = einsum(equation = var_3020_equation_0, values = (var_2654_cast_fp16, var_2511_cast_fp16))[name = tensor("op_3020_cast_fp16")]; + tensor var_3021_to_fp16 = const()[name = tensor("op_3021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = var_3020_cast_fp16, y = var_3021_to_fp16)[name = tensor("aw_chunk_299_cast_fp16")]; + tensor var_3024_equation_0 = const()[name = tensor("op_3024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3024_cast_fp16 = einsum(equation = var_3024_equation_0, values = (var_2654_cast_fp16, var_2518_cast_fp16))[name = tensor("op_3024_cast_fp16")]; + tensor var_3025_to_fp16 = const()[name = tensor("op_3025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = var_3024_cast_fp16, y = var_3025_to_fp16)[name = tensor("aw_chunk_301_cast_fp16")]; + tensor var_3028_equation_0 = const()[name = tensor("op_3028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3028_cast_fp16 = einsum(equation = var_3028_equation_0, values = (var_2654_cast_fp16, var_2525_cast_fp16))[name = tensor("op_3028_cast_fp16")]; + tensor var_3029_to_fp16 = const()[name = tensor("op_3029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = var_3028_cast_fp16, y = var_3029_to_fp16)[name = tensor("aw_chunk_303_cast_fp16")]; + tensor var_3032_equation_0 = const()[name = tensor("op_3032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3032_cast_fp16 = einsum(equation = var_3032_equation_0, values = (var_2658_cast_fp16, var_2532_cast_fp16))[name = tensor("op_3032_cast_fp16")]; + tensor var_3033_to_fp16 = const()[name = tensor("op_3033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = var_3032_cast_fp16, y = var_3033_to_fp16)[name = tensor("aw_chunk_305_cast_fp16")]; + tensor var_3036_equation_0 = const()[name = tensor("op_3036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3036_cast_fp16 = einsum(equation = var_3036_equation_0, values = (var_2658_cast_fp16, var_2539_cast_fp16))[name = tensor("op_3036_cast_fp16")]; + tensor var_3037_to_fp16 = const()[name = tensor("op_3037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = var_3036_cast_fp16, y = var_3037_to_fp16)[name = tensor("aw_chunk_307_cast_fp16")]; + tensor var_3040_equation_0 = const()[name = tensor("op_3040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3040_cast_fp16 = einsum(equation = var_3040_equation_0, values = (var_2658_cast_fp16, var_2546_cast_fp16))[name = tensor("op_3040_cast_fp16")]; + tensor var_3041_to_fp16 = const()[name = tensor("op_3041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = var_3040_cast_fp16, y = var_3041_to_fp16)[name = tensor("aw_chunk_309_cast_fp16")]; + tensor var_3044_equation_0 = const()[name = tensor("op_3044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3044_cast_fp16 = einsum(equation = var_3044_equation_0, values = (var_2658_cast_fp16, var_2553_cast_fp16))[name = tensor("op_3044_cast_fp16")]; + tensor var_3045_to_fp16 = const()[name = tensor("op_3045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = var_3044_cast_fp16, y = var_3045_to_fp16)[name = tensor("aw_chunk_311_cast_fp16")]; + tensor var_3048_equation_0 = const()[name = tensor("op_3048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3048_cast_fp16 = einsum(equation = var_3048_equation_0, values = (var_2662_cast_fp16, var_2560_cast_fp16))[name = tensor("op_3048_cast_fp16")]; + tensor var_3049_to_fp16 = const()[name = tensor("op_3049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = var_3048_cast_fp16, y = var_3049_to_fp16)[name = tensor("aw_chunk_313_cast_fp16")]; + tensor var_3052_equation_0 = const()[name = tensor("op_3052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3052_cast_fp16 = einsum(equation = var_3052_equation_0, values = (var_2662_cast_fp16, var_2567_cast_fp16))[name = tensor("op_3052_cast_fp16")]; + tensor var_3053_to_fp16 = const()[name = tensor("op_3053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = var_3052_cast_fp16, y = var_3053_to_fp16)[name = tensor("aw_chunk_315_cast_fp16")]; + tensor var_3056_equation_0 = const()[name = tensor("op_3056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3056_cast_fp16 = einsum(equation = var_3056_equation_0, values = (var_2662_cast_fp16, var_2574_cast_fp16))[name = tensor("op_3056_cast_fp16")]; + tensor var_3057_to_fp16 = const()[name = tensor("op_3057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = var_3056_cast_fp16, y = var_3057_to_fp16)[name = tensor("aw_chunk_317_cast_fp16")]; + tensor var_3060_equation_0 = const()[name = tensor("op_3060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_3060_cast_fp16 = einsum(equation = var_3060_equation_0, values = (var_2662_cast_fp16, var_2581_cast_fp16))[name = tensor("op_3060_cast_fp16")]; + tensor var_3061_to_fp16 = const()[name = tensor("op_3061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = var_3060_cast_fp16, y = var_3061_to_fp16)[name = tensor("aw_chunk_319_cast_fp16")]; + tensor var_3063_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_161_cast_fp16)[name = tensor("op_3063_cast_fp16")]; + tensor var_3064_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_163_cast_fp16)[name = tensor("op_3064_cast_fp16")]; + tensor var_3065_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_165_cast_fp16)[name = tensor("op_3065_cast_fp16")]; + tensor var_3066_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_167_cast_fp16)[name = tensor("op_3066_cast_fp16")]; + tensor var_3067_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_169_cast_fp16)[name = tensor("op_3067_cast_fp16")]; + tensor var_3068_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_171_cast_fp16)[name = tensor("op_3068_cast_fp16")]; + tensor var_3069_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_173_cast_fp16)[name = tensor("op_3069_cast_fp16")]; + tensor var_3070_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_175_cast_fp16)[name = tensor("op_3070_cast_fp16")]; + tensor var_3071_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_177_cast_fp16)[name = tensor("op_3071_cast_fp16")]; + tensor var_3072_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_179_cast_fp16)[name = tensor("op_3072_cast_fp16")]; + tensor var_3073_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_181_cast_fp16)[name = tensor("op_3073_cast_fp16")]; + tensor var_3074_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_183_cast_fp16)[name = tensor("op_3074_cast_fp16")]; + tensor var_3075_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_185_cast_fp16)[name = tensor("op_3075_cast_fp16")]; + tensor var_3076_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_187_cast_fp16)[name = tensor("op_3076_cast_fp16")]; + tensor var_3077_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_189_cast_fp16)[name = tensor("op_3077_cast_fp16")]; + tensor var_3078_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_191_cast_fp16)[name = tensor("op_3078_cast_fp16")]; + tensor var_3079_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_193_cast_fp16)[name = tensor("op_3079_cast_fp16")]; + tensor var_3080_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_195_cast_fp16)[name = tensor("op_3080_cast_fp16")]; + tensor var_3081_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_197_cast_fp16)[name = tensor("op_3081_cast_fp16")]; + tensor var_3082_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_199_cast_fp16)[name = tensor("op_3082_cast_fp16")]; + tensor var_3083_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_201_cast_fp16)[name = tensor("op_3083_cast_fp16")]; + tensor var_3084_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_203_cast_fp16)[name = tensor("op_3084_cast_fp16")]; + tensor var_3085_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_205_cast_fp16)[name = tensor("op_3085_cast_fp16")]; + tensor var_3086_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_207_cast_fp16)[name = tensor("op_3086_cast_fp16")]; + tensor var_3087_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_209_cast_fp16)[name = tensor("op_3087_cast_fp16")]; + tensor var_3088_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_211_cast_fp16)[name = tensor("op_3088_cast_fp16")]; + tensor var_3089_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_213_cast_fp16)[name = tensor("op_3089_cast_fp16")]; + tensor var_3090_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_215_cast_fp16)[name = tensor("op_3090_cast_fp16")]; + tensor var_3091_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_217_cast_fp16)[name = tensor("op_3091_cast_fp16")]; + tensor var_3092_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_219_cast_fp16)[name = tensor("op_3092_cast_fp16")]; + tensor var_3093_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_221_cast_fp16)[name = tensor("op_3093_cast_fp16")]; + tensor var_3094_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_223_cast_fp16)[name = tensor("op_3094_cast_fp16")]; + tensor var_3095_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_225_cast_fp16)[name = tensor("op_3095_cast_fp16")]; + tensor var_3096_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_227_cast_fp16)[name = tensor("op_3096_cast_fp16")]; + tensor var_3097_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_229_cast_fp16)[name = tensor("op_3097_cast_fp16")]; + tensor var_3098_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_231_cast_fp16)[name = tensor("op_3098_cast_fp16")]; + tensor var_3099_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_233_cast_fp16)[name = tensor("op_3099_cast_fp16")]; + tensor var_3100_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_235_cast_fp16)[name = tensor("op_3100_cast_fp16")]; + tensor var_3101_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_237_cast_fp16)[name = tensor("op_3101_cast_fp16")]; + tensor var_3102_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_239_cast_fp16)[name = tensor("op_3102_cast_fp16")]; + tensor var_3103_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_241_cast_fp16)[name = tensor("op_3103_cast_fp16")]; + tensor var_3104_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_243_cast_fp16)[name = tensor("op_3104_cast_fp16")]; + tensor var_3105_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_245_cast_fp16)[name = tensor("op_3105_cast_fp16")]; + tensor var_3106_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_247_cast_fp16)[name = tensor("op_3106_cast_fp16")]; + tensor var_3107_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_249_cast_fp16)[name = tensor("op_3107_cast_fp16")]; + tensor var_3108_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_251_cast_fp16)[name = tensor("op_3108_cast_fp16")]; + tensor var_3109_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_253_cast_fp16)[name = tensor("op_3109_cast_fp16")]; + tensor var_3110_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_255_cast_fp16)[name = tensor("op_3110_cast_fp16")]; + tensor var_3111_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_257_cast_fp16)[name = tensor("op_3111_cast_fp16")]; + tensor var_3112_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_259_cast_fp16)[name = tensor("op_3112_cast_fp16")]; + tensor var_3113_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_261_cast_fp16)[name = tensor("op_3113_cast_fp16")]; + tensor var_3114_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_263_cast_fp16)[name = tensor("op_3114_cast_fp16")]; + tensor var_3115_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_265_cast_fp16)[name = tensor("op_3115_cast_fp16")]; + tensor var_3116_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_267_cast_fp16)[name = tensor("op_3116_cast_fp16")]; + tensor var_3117_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_269_cast_fp16)[name = tensor("op_3117_cast_fp16")]; + tensor var_3118_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_271_cast_fp16)[name = tensor("op_3118_cast_fp16")]; + tensor var_3119_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_273_cast_fp16)[name = tensor("op_3119_cast_fp16")]; + tensor var_3120_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_275_cast_fp16)[name = tensor("op_3120_cast_fp16")]; + tensor var_3121_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_277_cast_fp16)[name = tensor("op_3121_cast_fp16")]; + tensor var_3122_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_279_cast_fp16)[name = tensor("op_3122_cast_fp16")]; + tensor var_3123_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_281_cast_fp16)[name = tensor("op_3123_cast_fp16")]; + tensor var_3124_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_283_cast_fp16)[name = tensor("op_3124_cast_fp16")]; + tensor var_3125_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_285_cast_fp16)[name = tensor("op_3125_cast_fp16")]; + tensor var_3126_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_287_cast_fp16)[name = tensor("op_3126_cast_fp16")]; + tensor var_3127_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_289_cast_fp16)[name = tensor("op_3127_cast_fp16")]; + tensor var_3128_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_291_cast_fp16)[name = tensor("op_3128_cast_fp16")]; + tensor var_3129_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_293_cast_fp16)[name = tensor("op_3129_cast_fp16")]; + tensor var_3130_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_295_cast_fp16)[name = tensor("op_3130_cast_fp16")]; + tensor var_3131_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_297_cast_fp16)[name = tensor("op_3131_cast_fp16")]; + tensor var_3132_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_299_cast_fp16)[name = tensor("op_3132_cast_fp16")]; + tensor var_3133_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_301_cast_fp16)[name = tensor("op_3133_cast_fp16")]; + tensor var_3134_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_303_cast_fp16)[name = tensor("op_3134_cast_fp16")]; + tensor var_3135_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_305_cast_fp16)[name = tensor("op_3135_cast_fp16")]; + tensor var_3136_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_307_cast_fp16)[name = tensor("op_3136_cast_fp16")]; + tensor var_3137_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_309_cast_fp16)[name = tensor("op_3137_cast_fp16")]; + tensor var_3138_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_311_cast_fp16)[name = tensor("op_3138_cast_fp16")]; + tensor var_3139_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_313_cast_fp16)[name = tensor("op_3139_cast_fp16")]; + tensor var_3140_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_315_cast_fp16)[name = tensor("op_3140_cast_fp16")]; + tensor var_3141_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_317_cast_fp16)[name = tensor("op_3141_cast_fp16")]; + tensor var_3142_cast_fp16 = softmax(axis = var_1834, x = aw_chunk_319_cast_fp16)[name = tensor("op_3142_cast_fp16")]; + tensor var_3144_equation_0 = const()[name = tensor("op_3144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3144_cast_fp16 = einsum(equation = var_3144_equation_0, values = (var_2664_cast_fp16, var_3063_cast_fp16))[name = tensor("op_3144_cast_fp16")]; + tensor var_3146_equation_0 = const()[name = tensor("op_3146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3146_cast_fp16 = einsum(equation = var_3146_equation_0, values = (var_2664_cast_fp16, var_3064_cast_fp16))[name = tensor("op_3146_cast_fp16")]; + tensor var_3148_equation_0 = const()[name = tensor("op_3148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3148_cast_fp16 = einsum(equation = var_3148_equation_0, values = (var_2664_cast_fp16, var_3065_cast_fp16))[name = tensor("op_3148_cast_fp16")]; + tensor var_3150_equation_0 = const()[name = tensor("op_3150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3150_cast_fp16 = einsum(equation = var_3150_equation_0, values = (var_2664_cast_fp16, var_3066_cast_fp16))[name = tensor("op_3150_cast_fp16")]; + tensor var_3152_equation_0 = const()[name = tensor("op_3152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3152_cast_fp16 = einsum(equation = var_3152_equation_0, values = (var_2668_cast_fp16, var_3067_cast_fp16))[name = tensor("op_3152_cast_fp16")]; + tensor var_3154_equation_0 = const()[name = tensor("op_3154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3154_cast_fp16 = einsum(equation = var_3154_equation_0, values = (var_2668_cast_fp16, var_3068_cast_fp16))[name = tensor("op_3154_cast_fp16")]; + tensor var_3156_equation_0 = const()[name = tensor("op_3156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3156_cast_fp16 = einsum(equation = var_3156_equation_0, values = (var_2668_cast_fp16, var_3069_cast_fp16))[name = tensor("op_3156_cast_fp16")]; + tensor var_3158_equation_0 = const()[name = tensor("op_3158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3158_cast_fp16 = einsum(equation = var_3158_equation_0, values = (var_2668_cast_fp16, var_3070_cast_fp16))[name = tensor("op_3158_cast_fp16")]; + tensor var_3160_equation_0 = const()[name = tensor("op_3160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3160_cast_fp16 = einsum(equation = var_3160_equation_0, values = (var_2672_cast_fp16, var_3071_cast_fp16))[name = tensor("op_3160_cast_fp16")]; + tensor var_3162_equation_0 = const()[name = tensor("op_3162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3162_cast_fp16 = einsum(equation = var_3162_equation_0, values = (var_2672_cast_fp16, var_3072_cast_fp16))[name = tensor("op_3162_cast_fp16")]; + tensor var_3164_equation_0 = const()[name = tensor("op_3164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3164_cast_fp16 = einsum(equation = var_3164_equation_0, values = (var_2672_cast_fp16, var_3073_cast_fp16))[name = tensor("op_3164_cast_fp16")]; + tensor var_3166_equation_0 = const()[name = tensor("op_3166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3166_cast_fp16 = einsum(equation = var_3166_equation_0, values = (var_2672_cast_fp16, var_3074_cast_fp16))[name = tensor("op_3166_cast_fp16")]; + tensor var_3168_equation_0 = const()[name = tensor("op_3168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3168_cast_fp16 = einsum(equation = var_3168_equation_0, values = (var_2676_cast_fp16, var_3075_cast_fp16))[name = tensor("op_3168_cast_fp16")]; + tensor var_3170_equation_0 = const()[name = tensor("op_3170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3170_cast_fp16 = einsum(equation = var_3170_equation_0, values = (var_2676_cast_fp16, var_3076_cast_fp16))[name = tensor("op_3170_cast_fp16")]; + tensor var_3172_equation_0 = const()[name = tensor("op_3172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3172_cast_fp16 = einsum(equation = var_3172_equation_0, values = (var_2676_cast_fp16, var_3077_cast_fp16))[name = tensor("op_3172_cast_fp16")]; + tensor var_3174_equation_0 = const()[name = tensor("op_3174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3174_cast_fp16 = einsum(equation = var_3174_equation_0, values = (var_2676_cast_fp16, var_3078_cast_fp16))[name = tensor("op_3174_cast_fp16")]; + tensor var_3176_equation_0 = const()[name = tensor("op_3176_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3176_cast_fp16 = einsum(equation = var_3176_equation_0, values = (var_2680_cast_fp16, var_3079_cast_fp16))[name = tensor("op_3176_cast_fp16")]; + tensor var_3178_equation_0 = const()[name = tensor("op_3178_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3178_cast_fp16 = einsum(equation = var_3178_equation_0, values = (var_2680_cast_fp16, var_3080_cast_fp16))[name = tensor("op_3178_cast_fp16")]; + tensor var_3180_equation_0 = const()[name = tensor("op_3180_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3180_cast_fp16 = einsum(equation = var_3180_equation_0, values = (var_2680_cast_fp16, var_3081_cast_fp16))[name = tensor("op_3180_cast_fp16")]; + tensor var_3182_equation_0 = const()[name = tensor("op_3182_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3182_cast_fp16 = einsum(equation = var_3182_equation_0, values = (var_2680_cast_fp16, var_3082_cast_fp16))[name = tensor("op_3182_cast_fp16")]; + tensor var_3184_equation_0 = const()[name = tensor("op_3184_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3184_cast_fp16 = einsum(equation = var_3184_equation_0, values = (var_2684_cast_fp16, var_3083_cast_fp16))[name = tensor("op_3184_cast_fp16")]; + tensor var_3186_equation_0 = const()[name = tensor("op_3186_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3186_cast_fp16 = einsum(equation = var_3186_equation_0, values = (var_2684_cast_fp16, var_3084_cast_fp16))[name = tensor("op_3186_cast_fp16")]; + tensor var_3188_equation_0 = const()[name = tensor("op_3188_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3188_cast_fp16 = einsum(equation = var_3188_equation_0, values = (var_2684_cast_fp16, var_3085_cast_fp16))[name = tensor("op_3188_cast_fp16")]; + tensor var_3190_equation_0 = const()[name = tensor("op_3190_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3190_cast_fp16 = einsum(equation = var_3190_equation_0, values = (var_2684_cast_fp16, var_3086_cast_fp16))[name = tensor("op_3190_cast_fp16")]; + tensor var_3192_equation_0 = const()[name = tensor("op_3192_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3192_cast_fp16 = einsum(equation = var_3192_equation_0, values = (var_2688_cast_fp16, var_3087_cast_fp16))[name = tensor("op_3192_cast_fp16")]; + tensor var_3194_equation_0 = const()[name = tensor("op_3194_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3194_cast_fp16 = einsum(equation = var_3194_equation_0, values = (var_2688_cast_fp16, var_3088_cast_fp16))[name = tensor("op_3194_cast_fp16")]; + tensor var_3196_equation_0 = const()[name = tensor("op_3196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3196_cast_fp16 = einsum(equation = var_3196_equation_0, values = (var_2688_cast_fp16, var_3089_cast_fp16))[name = tensor("op_3196_cast_fp16")]; + tensor var_3198_equation_0 = const()[name = tensor("op_3198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3198_cast_fp16 = einsum(equation = var_3198_equation_0, values = (var_2688_cast_fp16, var_3090_cast_fp16))[name = tensor("op_3198_cast_fp16")]; + tensor var_3200_equation_0 = const()[name = tensor("op_3200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3200_cast_fp16 = einsum(equation = var_3200_equation_0, values = (var_2692_cast_fp16, var_3091_cast_fp16))[name = tensor("op_3200_cast_fp16")]; + tensor var_3202_equation_0 = const()[name = tensor("op_3202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3202_cast_fp16 = einsum(equation = var_3202_equation_0, values = (var_2692_cast_fp16, var_3092_cast_fp16))[name = tensor("op_3202_cast_fp16")]; + tensor var_3204_equation_0 = const()[name = tensor("op_3204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3204_cast_fp16 = einsum(equation = var_3204_equation_0, values = (var_2692_cast_fp16, var_3093_cast_fp16))[name = tensor("op_3204_cast_fp16")]; + tensor var_3206_equation_0 = const()[name = tensor("op_3206_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3206_cast_fp16 = einsum(equation = var_3206_equation_0, values = (var_2692_cast_fp16, var_3094_cast_fp16))[name = tensor("op_3206_cast_fp16")]; + tensor var_3208_equation_0 = const()[name = tensor("op_3208_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3208_cast_fp16 = einsum(equation = var_3208_equation_0, values = (var_2696_cast_fp16, var_3095_cast_fp16))[name = tensor("op_3208_cast_fp16")]; + tensor var_3210_equation_0 = const()[name = tensor("op_3210_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3210_cast_fp16 = einsum(equation = var_3210_equation_0, values = (var_2696_cast_fp16, var_3096_cast_fp16))[name = tensor("op_3210_cast_fp16")]; + tensor var_3212_equation_0 = const()[name = tensor("op_3212_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3212_cast_fp16 = einsum(equation = var_3212_equation_0, values = (var_2696_cast_fp16, var_3097_cast_fp16))[name = tensor("op_3212_cast_fp16")]; + tensor var_3214_equation_0 = const()[name = tensor("op_3214_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3214_cast_fp16 = einsum(equation = var_3214_equation_0, values = (var_2696_cast_fp16, var_3098_cast_fp16))[name = tensor("op_3214_cast_fp16")]; + tensor var_3216_equation_0 = const()[name = tensor("op_3216_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3216_cast_fp16 = einsum(equation = var_3216_equation_0, values = (var_2700_cast_fp16, var_3099_cast_fp16))[name = tensor("op_3216_cast_fp16")]; + tensor var_3218_equation_0 = const()[name = tensor("op_3218_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3218_cast_fp16 = einsum(equation = var_3218_equation_0, values = (var_2700_cast_fp16, var_3100_cast_fp16))[name = tensor("op_3218_cast_fp16")]; + tensor var_3220_equation_0 = const()[name = tensor("op_3220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3220_cast_fp16 = einsum(equation = var_3220_equation_0, values = (var_2700_cast_fp16, var_3101_cast_fp16))[name = tensor("op_3220_cast_fp16")]; + tensor var_3222_equation_0 = const()[name = tensor("op_3222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3222_cast_fp16 = einsum(equation = var_3222_equation_0, values = (var_2700_cast_fp16, var_3102_cast_fp16))[name = tensor("op_3222_cast_fp16")]; + tensor var_3224_equation_0 = const()[name = tensor("op_3224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3224_cast_fp16 = einsum(equation = var_3224_equation_0, values = (var_2704_cast_fp16, var_3103_cast_fp16))[name = tensor("op_3224_cast_fp16")]; + tensor var_3226_equation_0 = const()[name = tensor("op_3226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3226_cast_fp16 = einsum(equation = var_3226_equation_0, values = (var_2704_cast_fp16, var_3104_cast_fp16))[name = tensor("op_3226_cast_fp16")]; + tensor var_3228_equation_0 = const()[name = tensor("op_3228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3228_cast_fp16 = einsum(equation = var_3228_equation_0, values = (var_2704_cast_fp16, var_3105_cast_fp16))[name = tensor("op_3228_cast_fp16")]; + tensor var_3230_equation_0 = const()[name = tensor("op_3230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3230_cast_fp16 = einsum(equation = var_3230_equation_0, values = (var_2704_cast_fp16, var_3106_cast_fp16))[name = tensor("op_3230_cast_fp16")]; + tensor var_3232_equation_0 = const()[name = tensor("op_3232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3232_cast_fp16 = einsum(equation = var_3232_equation_0, values = (var_2708_cast_fp16, var_3107_cast_fp16))[name = tensor("op_3232_cast_fp16")]; + tensor var_3234_equation_0 = const()[name = tensor("op_3234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3234_cast_fp16 = einsum(equation = var_3234_equation_0, values = (var_2708_cast_fp16, var_3108_cast_fp16))[name = tensor("op_3234_cast_fp16")]; + tensor var_3236_equation_0 = const()[name = tensor("op_3236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3236_cast_fp16 = einsum(equation = var_3236_equation_0, values = (var_2708_cast_fp16, var_3109_cast_fp16))[name = tensor("op_3236_cast_fp16")]; + tensor var_3238_equation_0 = const()[name = tensor("op_3238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3238_cast_fp16 = einsum(equation = var_3238_equation_0, values = (var_2708_cast_fp16, var_3110_cast_fp16))[name = tensor("op_3238_cast_fp16")]; + tensor var_3240_equation_0 = const()[name = tensor("op_3240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3240_cast_fp16 = einsum(equation = var_3240_equation_0, values = (var_2712_cast_fp16, var_3111_cast_fp16))[name = tensor("op_3240_cast_fp16")]; + tensor var_3242_equation_0 = const()[name = tensor("op_3242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3242_cast_fp16 = einsum(equation = var_3242_equation_0, values = (var_2712_cast_fp16, var_3112_cast_fp16))[name = tensor("op_3242_cast_fp16")]; + tensor var_3244_equation_0 = const()[name = tensor("op_3244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3244_cast_fp16 = einsum(equation = var_3244_equation_0, values = (var_2712_cast_fp16, var_3113_cast_fp16))[name = tensor("op_3244_cast_fp16")]; + tensor var_3246_equation_0 = const()[name = tensor("op_3246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3246_cast_fp16 = einsum(equation = var_3246_equation_0, values = (var_2712_cast_fp16, var_3114_cast_fp16))[name = tensor("op_3246_cast_fp16")]; + tensor var_3248_equation_0 = const()[name = tensor("op_3248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3248_cast_fp16 = einsum(equation = var_3248_equation_0, values = (var_2716_cast_fp16, var_3115_cast_fp16))[name = tensor("op_3248_cast_fp16")]; + tensor var_3250_equation_0 = const()[name = tensor("op_3250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3250_cast_fp16 = einsum(equation = var_3250_equation_0, values = (var_2716_cast_fp16, var_3116_cast_fp16))[name = tensor("op_3250_cast_fp16")]; + tensor var_3252_equation_0 = const()[name = tensor("op_3252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3252_cast_fp16 = einsum(equation = var_3252_equation_0, values = (var_2716_cast_fp16, var_3117_cast_fp16))[name = tensor("op_3252_cast_fp16")]; + tensor var_3254_equation_0 = const()[name = tensor("op_3254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3254_cast_fp16 = einsum(equation = var_3254_equation_0, values = (var_2716_cast_fp16, var_3118_cast_fp16))[name = tensor("op_3254_cast_fp16")]; + tensor var_3256_equation_0 = const()[name = tensor("op_3256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3256_cast_fp16 = einsum(equation = var_3256_equation_0, values = (var_2720_cast_fp16, var_3119_cast_fp16))[name = tensor("op_3256_cast_fp16")]; + tensor var_3258_equation_0 = const()[name = tensor("op_3258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3258_cast_fp16 = einsum(equation = var_3258_equation_0, values = (var_2720_cast_fp16, var_3120_cast_fp16))[name = tensor("op_3258_cast_fp16")]; + tensor var_3260_equation_0 = const()[name = tensor("op_3260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3260_cast_fp16 = einsum(equation = var_3260_equation_0, values = (var_2720_cast_fp16, var_3121_cast_fp16))[name = tensor("op_3260_cast_fp16")]; + tensor var_3262_equation_0 = const()[name = tensor("op_3262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3262_cast_fp16 = einsum(equation = var_3262_equation_0, values = (var_2720_cast_fp16, var_3122_cast_fp16))[name = tensor("op_3262_cast_fp16")]; + tensor var_3264_equation_0 = const()[name = tensor("op_3264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3264_cast_fp16 = einsum(equation = var_3264_equation_0, values = (var_2724_cast_fp16, var_3123_cast_fp16))[name = tensor("op_3264_cast_fp16")]; + tensor var_3266_equation_0 = const()[name = tensor("op_3266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3266_cast_fp16 = einsum(equation = var_3266_equation_0, values = (var_2724_cast_fp16, var_3124_cast_fp16))[name = tensor("op_3266_cast_fp16")]; + tensor var_3268_equation_0 = const()[name = tensor("op_3268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3268_cast_fp16 = einsum(equation = var_3268_equation_0, values = (var_2724_cast_fp16, var_3125_cast_fp16))[name = tensor("op_3268_cast_fp16")]; + tensor var_3270_equation_0 = const()[name = tensor("op_3270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3270_cast_fp16 = einsum(equation = var_3270_equation_0, values = (var_2724_cast_fp16, var_3126_cast_fp16))[name = tensor("op_3270_cast_fp16")]; + tensor var_3272_equation_0 = const()[name = tensor("op_3272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3272_cast_fp16 = einsum(equation = var_3272_equation_0, values = (var_2728_cast_fp16, var_3127_cast_fp16))[name = tensor("op_3272_cast_fp16")]; + tensor var_3274_equation_0 = const()[name = tensor("op_3274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3274_cast_fp16 = einsum(equation = var_3274_equation_0, values = (var_2728_cast_fp16, var_3128_cast_fp16))[name = tensor("op_3274_cast_fp16")]; + tensor var_3276_equation_0 = const()[name = tensor("op_3276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3276_cast_fp16 = einsum(equation = var_3276_equation_0, values = (var_2728_cast_fp16, var_3129_cast_fp16))[name = tensor("op_3276_cast_fp16")]; + tensor var_3278_equation_0 = const()[name = tensor("op_3278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3278_cast_fp16 = einsum(equation = var_3278_equation_0, values = (var_2728_cast_fp16, var_3130_cast_fp16))[name = tensor("op_3278_cast_fp16")]; + tensor var_3280_equation_0 = const()[name = tensor("op_3280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3280_cast_fp16 = einsum(equation = var_3280_equation_0, values = (var_2732_cast_fp16, var_3131_cast_fp16))[name = tensor("op_3280_cast_fp16")]; + tensor var_3282_equation_0 = const()[name = tensor("op_3282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3282_cast_fp16 = einsum(equation = var_3282_equation_0, values = (var_2732_cast_fp16, var_3132_cast_fp16))[name = tensor("op_3282_cast_fp16")]; + tensor var_3284_equation_0 = const()[name = tensor("op_3284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3284_cast_fp16 = einsum(equation = var_3284_equation_0, values = (var_2732_cast_fp16, var_3133_cast_fp16))[name = tensor("op_3284_cast_fp16")]; + tensor var_3286_equation_0 = const()[name = tensor("op_3286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3286_cast_fp16 = einsum(equation = var_3286_equation_0, values = (var_2732_cast_fp16, var_3134_cast_fp16))[name = tensor("op_3286_cast_fp16")]; + tensor var_3288_equation_0 = const()[name = tensor("op_3288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3288_cast_fp16 = einsum(equation = var_3288_equation_0, values = (var_2736_cast_fp16, var_3135_cast_fp16))[name = tensor("op_3288_cast_fp16")]; + tensor var_3290_equation_0 = const()[name = tensor("op_3290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3290_cast_fp16 = einsum(equation = var_3290_equation_0, values = (var_2736_cast_fp16, var_3136_cast_fp16))[name = tensor("op_3290_cast_fp16")]; + tensor var_3292_equation_0 = const()[name = tensor("op_3292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3292_cast_fp16 = einsum(equation = var_3292_equation_0, values = (var_2736_cast_fp16, var_3137_cast_fp16))[name = tensor("op_3292_cast_fp16")]; + tensor var_3294_equation_0 = const()[name = tensor("op_3294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3294_cast_fp16 = einsum(equation = var_3294_equation_0, values = (var_2736_cast_fp16, var_3138_cast_fp16))[name = tensor("op_3294_cast_fp16")]; + tensor var_3296_equation_0 = const()[name = tensor("op_3296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3296_cast_fp16 = einsum(equation = var_3296_equation_0, values = (var_2740_cast_fp16, var_3139_cast_fp16))[name = tensor("op_3296_cast_fp16")]; + tensor var_3298_equation_0 = const()[name = tensor("op_3298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3298_cast_fp16 = einsum(equation = var_3298_equation_0, values = (var_2740_cast_fp16, var_3140_cast_fp16))[name = tensor("op_3298_cast_fp16")]; + tensor var_3300_equation_0 = const()[name = tensor("op_3300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3300_cast_fp16 = einsum(equation = var_3300_equation_0, values = (var_2740_cast_fp16, var_3141_cast_fp16))[name = tensor("op_3300_cast_fp16")]; + tensor var_3302_equation_0 = const()[name = tensor("op_3302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_3302_cast_fp16 = einsum(equation = var_3302_equation_0, values = (var_2740_cast_fp16, var_3142_cast_fp16))[name = tensor("op_3302_cast_fp16")]; + tensor var_3304_interleave_0 = const()[name = tensor("op_3304_interleave_0"), val = tensor(false)]; + tensor var_3304_cast_fp16 = concat(axis = var_1809, interleave = var_3304_interleave_0, values = (var_3144_cast_fp16, var_3146_cast_fp16, var_3148_cast_fp16, var_3150_cast_fp16))[name = tensor("op_3304_cast_fp16")]; + tensor var_3306_interleave_0 = const()[name = tensor("op_3306_interleave_0"), val = tensor(false)]; + tensor var_3306_cast_fp16 = concat(axis = var_1809, interleave = var_3306_interleave_0, values = (var_3152_cast_fp16, var_3154_cast_fp16, var_3156_cast_fp16, var_3158_cast_fp16))[name = tensor("op_3306_cast_fp16")]; + tensor var_3308_interleave_0 = const()[name = tensor("op_3308_interleave_0"), val = tensor(false)]; + tensor var_3308_cast_fp16 = concat(axis = var_1809, interleave = var_3308_interleave_0, values = (var_3160_cast_fp16, var_3162_cast_fp16, var_3164_cast_fp16, var_3166_cast_fp16))[name = tensor("op_3308_cast_fp16")]; + tensor var_3310_interleave_0 = const()[name = tensor("op_3310_interleave_0"), val = tensor(false)]; + tensor var_3310_cast_fp16 = concat(axis = var_1809, interleave = var_3310_interleave_0, values = (var_3168_cast_fp16, var_3170_cast_fp16, var_3172_cast_fp16, var_3174_cast_fp16))[name = tensor("op_3310_cast_fp16")]; + tensor var_3312_interleave_0 = const()[name = tensor("op_3312_interleave_0"), val = tensor(false)]; + tensor var_3312_cast_fp16 = concat(axis = var_1809, interleave = var_3312_interleave_0, values = (var_3176_cast_fp16, var_3178_cast_fp16, var_3180_cast_fp16, var_3182_cast_fp16))[name = tensor("op_3312_cast_fp16")]; + tensor var_3314_interleave_0 = const()[name = tensor("op_3314_interleave_0"), val = tensor(false)]; + tensor var_3314_cast_fp16 = concat(axis = var_1809, interleave = var_3314_interleave_0, values = (var_3184_cast_fp16, var_3186_cast_fp16, var_3188_cast_fp16, var_3190_cast_fp16))[name = tensor("op_3314_cast_fp16")]; + tensor var_3316_interleave_0 = const()[name = tensor("op_3316_interleave_0"), val = tensor(false)]; + tensor var_3316_cast_fp16 = concat(axis = var_1809, interleave = var_3316_interleave_0, values = (var_3192_cast_fp16, var_3194_cast_fp16, var_3196_cast_fp16, var_3198_cast_fp16))[name = tensor("op_3316_cast_fp16")]; + tensor var_3318_interleave_0 = const()[name = tensor("op_3318_interleave_0"), val = tensor(false)]; + tensor var_3318_cast_fp16 = concat(axis = var_1809, interleave = var_3318_interleave_0, values = (var_3200_cast_fp16, var_3202_cast_fp16, var_3204_cast_fp16, var_3206_cast_fp16))[name = tensor("op_3318_cast_fp16")]; + tensor var_3320_interleave_0 = const()[name = tensor("op_3320_interleave_0"), val = tensor(false)]; + tensor var_3320_cast_fp16 = concat(axis = var_1809, interleave = var_3320_interleave_0, values = (var_3208_cast_fp16, var_3210_cast_fp16, var_3212_cast_fp16, var_3214_cast_fp16))[name = tensor("op_3320_cast_fp16")]; + tensor var_3322_interleave_0 = const()[name = tensor("op_3322_interleave_0"), val = tensor(false)]; + tensor var_3322_cast_fp16 = concat(axis = var_1809, interleave = var_3322_interleave_0, values = (var_3216_cast_fp16, var_3218_cast_fp16, var_3220_cast_fp16, var_3222_cast_fp16))[name = tensor("op_3322_cast_fp16")]; + tensor var_3324_interleave_0 = const()[name = tensor("op_3324_interleave_0"), val = tensor(false)]; + tensor var_3324_cast_fp16 = concat(axis = var_1809, interleave = var_3324_interleave_0, values = (var_3224_cast_fp16, var_3226_cast_fp16, var_3228_cast_fp16, var_3230_cast_fp16))[name = tensor("op_3324_cast_fp16")]; + tensor var_3326_interleave_0 = const()[name = tensor("op_3326_interleave_0"), val = tensor(false)]; + tensor var_3326_cast_fp16 = concat(axis = var_1809, interleave = var_3326_interleave_0, values = (var_3232_cast_fp16, var_3234_cast_fp16, var_3236_cast_fp16, var_3238_cast_fp16))[name = tensor("op_3326_cast_fp16")]; + tensor var_3328_interleave_0 = const()[name = tensor("op_3328_interleave_0"), val = tensor(false)]; + tensor var_3328_cast_fp16 = concat(axis = var_1809, interleave = var_3328_interleave_0, values = (var_3240_cast_fp16, var_3242_cast_fp16, var_3244_cast_fp16, var_3246_cast_fp16))[name = tensor("op_3328_cast_fp16")]; + tensor var_3330_interleave_0 = const()[name = tensor("op_3330_interleave_0"), val = tensor(false)]; + tensor var_3330_cast_fp16 = concat(axis = var_1809, interleave = var_3330_interleave_0, values = (var_3248_cast_fp16, var_3250_cast_fp16, var_3252_cast_fp16, var_3254_cast_fp16))[name = tensor("op_3330_cast_fp16")]; + tensor var_3332_interleave_0 = const()[name = tensor("op_3332_interleave_0"), val = tensor(false)]; + tensor var_3332_cast_fp16 = concat(axis = var_1809, interleave = var_3332_interleave_0, values = (var_3256_cast_fp16, var_3258_cast_fp16, var_3260_cast_fp16, var_3262_cast_fp16))[name = tensor("op_3332_cast_fp16")]; + tensor var_3334_interleave_0 = const()[name = tensor("op_3334_interleave_0"), val = tensor(false)]; + tensor var_3334_cast_fp16 = concat(axis = var_1809, interleave = var_3334_interleave_0, values = (var_3264_cast_fp16, var_3266_cast_fp16, var_3268_cast_fp16, var_3270_cast_fp16))[name = tensor("op_3334_cast_fp16")]; + tensor var_3336_interleave_0 = const()[name = tensor("op_3336_interleave_0"), val = tensor(false)]; + tensor var_3336_cast_fp16 = concat(axis = var_1809, interleave = var_3336_interleave_0, values = (var_3272_cast_fp16, var_3274_cast_fp16, var_3276_cast_fp16, var_3278_cast_fp16))[name = tensor("op_3336_cast_fp16")]; + tensor var_3338_interleave_0 = const()[name = tensor("op_3338_interleave_0"), val = tensor(false)]; + tensor var_3338_cast_fp16 = concat(axis = var_1809, interleave = var_3338_interleave_0, values = (var_3280_cast_fp16, var_3282_cast_fp16, var_3284_cast_fp16, var_3286_cast_fp16))[name = tensor("op_3338_cast_fp16")]; + tensor var_3340_interleave_0 = const()[name = tensor("op_3340_interleave_0"), val = tensor(false)]; + tensor var_3340_cast_fp16 = concat(axis = var_1809, interleave = var_3340_interleave_0, values = (var_3288_cast_fp16, var_3290_cast_fp16, var_3292_cast_fp16, var_3294_cast_fp16))[name = tensor("op_3340_cast_fp16")]; + tensor var_3342_interleave_0 = const()[name = tensor("op_3342_interleave_0"), val = tensor(false)]; + tensor var_3342_cast_fp16 = concat(axis = var_1809, interleave = var_3342_interleave_0, values = (var_3296_cast_fp16, var_3298_cast_fp16, var_3300_cast_fp16, var_3302_cast_fp16))[name = tensor("op_3342_cast_fp16")]; + tensor input_27_interleave_0 = const()[name = tensor("input_27_interleave_0"), val = tensor(false)]; + tensor input_27_cast_fp16 = concat(axis = var_1834, interleave = input_27_interleave_0, values = (var_3304_cast_fp16, var_3306_cast_fp16, var_3308_cast_fp16, var_3310_cast_fp16, var_3312_cast_fp16, var_3314_cast_fp16, var_3316_cast_fp16, var_3318_cast_fp16, var_3320_cast_fp16, var_3322_cast_fp16, var_3324_cast_fp16, var_3326_cast_fp16, var_3328_cast_fp16, var_3330_cast_fp16, var_3332_cast_fp16, var_3334_cast_fp16, var_3336_cast_fp16, var_3338_cast_fp16, var_3340_cast_fp16, var_3342_cast_fp16))[name = tensor("input_27_cast_fp16")]; + tensor var_3350 = const()[name = tensor("op_3350"), val = tensor([1, 1])]; + tensor var_3352 = const()[name = tensor("op_3352"), val = tensor([1, 1])]; + tensor pretrained_out_19_pad_type_0 = const()[name = tensor("pretrained_out_19_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_19_pad_0 = const()[name = tensor("pretrained_out_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27621568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28440832))), name = tensor("layers_1_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_1_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28440960)))]; + tensor pretrained_out_19_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_3352, groups = var_1834, pad = pretrained_out_19_pad_0, pad_type = pretrained_out_19_pad_type_0, strides = var_3350, weight = layers_1_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor("pretrained_out_19_cast_fp16")]; + tensor var_3356 = const()[name = tensor("op_3356"), val = tensor([1, 1])]; + tensor var_3358 = const()[name = tensor("op_3358"), val = tensor([1, 1])]; + tensor input_29_pad_type_0 = const()[name = tensor("input_29_pad_type_0"), val = tensor("custom")]; + tensor input_29_pad_0 = const()[name = tensor("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28443584)))]; + tensor input_29_cast_fp16 = conv(dilations = var_3358, groups = var_1834, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = var_3356, weight = layers_1_self_attn_o_proj_loraA_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor var_3362 = const()[name = tensor("op_3362"), val = tensor([1, 1])]; + tensor var_3364 = const()[name = tensor("op_3364"), val = tensor([1, 1])]; + tensor lora_out_37_pad_type_0 = const()[name = tensor("lora_out_37_pad_type_0"), val = tensor("custom")]; + tensor lora_out_37_pad_0 = const()[name = tensor("lora_out_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_39_weight_0_to_fp16 = const()[name = tensor("lora_out_39_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28484608)))]; + tensor lora_out_39_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_3364, groups = var_1834, pad = lora_out_37_pad_0, pad_type = lora_out_37_pad_type_0, strides = var_3362, weight = lora_out_39_weight_0_to_fp16, x = input_29_cast_fp16)[name = tensor("lora_out_39_cast_fp16")]; + tensor obj_7_cast_fp16 = add(x = pretrained_out_19_cast_fp16, y = lora_out_39_cast_fp16)[name = tensor("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor var_3373 = const()[name = tensor("op_3373"), val = tensor([1])]; + tensor channels_mean_7_cast_fp16 = reduce_mean(axes = var_3373, keep_dims = var_1835, x = inputs_7_cast_fp16)[name = tensor("channels_mean_7_cast_fp16")]; + tensor zero_mean_7_cast_fp16 = sub(x = inputs_7_cast_fp16, y = channels_mean_7_cast_fp16)[name = tensor("zero_mean_7_cast_fp16")]; + tensor zero_mean_sq_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = zero_mean_7_cast_fp16)[name = tensor("zero_mean_sq_7_cast_fp16")]; + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1])]; + tensor var_3378_cast_fp16 = reduce_mean(axes = var_3377, keep_dims = var_1835, x = zero_mean_sq_7_cast_fp16)[name = tensor("op_3378_cast_fp16")]; + tensor var_3379_to_fp16 = const()[name = tensor("op_3379_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3380_cast_fp16 = add(x = var_3378_cast_fp16, y = var_3379_to_fp16)[name = tensor("op_3380_cast_fp16")]; + tensor denom_7_epsilon_0 = const()[name = tensor("denom_7_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_7_cast_fp16 = rsqrt(epsilon = denom_7_epsilon_0, x = var_3380_cast_fp16)[name = tensor("denom_7_cast_fp16")]; + tensor out_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = denom_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; + tensor input_31_gamma_0_to_fp16 = const()[name = tensor("input_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28525632)))]; + tensor input_31_beta_0_to_fp16 = const()[name = tensor("input_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28528256)))]; + tensor input_31_epsilon_0_to_fp16 = const()[name = tensor("input_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_31_cast_fp16 = batch_norm(beta = input_31_beta_0_to_fp16, epsilon = input_31_epsilon_0_to_fp16, gamma = input_31_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor("input_31_cast_fp16")]; + tensor var_3394 = const()[name = tensor("op_3394"), val = tensor([1, 1])]; + tensor var_3396 = const()[name = tensor("op_3396"), val = tensor([1, 1])]; + tensor pretrained_out_21_pad_type_0 = const()[name = tensor("pretrained_out_21_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_21_pad_0 = const()[name = tensor("pretrained_out_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28530880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31807744))), name = tensor("layers_1_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_1_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_1_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31807872)))]; + tensor pretrained_out_21_cast_fp16 = conv(bias = layers_1_fc1_pretrained_bias_to_fp16, dilations = var_3396, groups = var_1834, pad = pretrained_out_21_pad_0, pad_type = pretrained_out_21_pad_type_0, strides = var_3394, weight = layers_1_fc1_pretrained_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor("pretrained_out_21_cast_fp16")]; + tensor var_3400 = const()[name = tensor("op_3400"), val = tensor([1, 1])]; + tensor var_3402 = const()[name = tensor("op_3402"), val = tensor([1, 1])]; + tensor input_33_pad_type_0 = const()[name = tensor("input_33_pad_type_0"), val = tensor("custom")]; + tensor input_33_pad_0 = const()[name = tensor("input_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_1_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31818176)))]; + tensor input_33_cast_fp16 = conv(dilations = var_3402, groups = var_1834, pad = input_33_pad_0, pad_type = input_33_pad_type_0, strides = var_3400, weight = layers_1_fc1_loraA_weight_to_fp16, x = input_31_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor var_3406 = const()[name = tensor("op_3406"), val = tensor([1, 1])]; + tensor var_3408 = const()[name = tensor("op_3408"), val = tensor([1, 1])]; + tensor lora_out_41_pad_type_0 = const()[name = tensor("lora_out_41_pad_type_0"), val = tensor("custom")]; + tensor lora_out_41_pad_0 = const()[name = tensor("lora_out_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_43_weight_0_to_fp16 = const()[name = tensor("lora_out_43_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31859200)))]; + tensor lora_out_43_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_3408, groups = var_1834, pad = lora_out_41_pad_0, pad_type = lora_out_41_pad_type_0, strides = var_3406, weight = lora_out_43_weight_0_to_fp16, x = input_33_cast_fp16)[name = tensor("lora_out_43_cast_fp16")]; + tensor input_35_cast_fp16 = add(x = pretrained_out_21_cast_fp16, y = lora_out_43_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor input_37_mode_0 = const()[name = tensor("input_37_mode_0"), val = tensor("EXACT")]; + tensor input_37_cast_fp16 = gelu(mode = input_37_mode_0, x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor var_3420 = const()[name = tensor("op_3420"), val = tensor([1, 1])]; + tensor var_3422 = const()[name = tensor("op_3422"), val = tensor([1, 1])]; + tensor pretrained_out_23_pad_type_0 = const()[name = tensor("pretrained_out_23_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_23_pad_0 = const()[name = tensor("pretrained_out_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32023104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35299968))), name = tensor("layers_1_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_1_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_1_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35300096)))]; + tensor pretrained_out_23_cast_fp16 = conv(bias = layers_1_fc2_pretrained_bias_to_fp16, dilations = var_3422, groups = var_1834, pad = pretrained_out_23_pad_0, pad_type = pretrained_out_23_pad_type_0, strides = var_3420, weight = layers_1_fc2_pretrained_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("pretrained_out_23_cast_fp16")]; + tensor var_3426 = const()[name = tensor("op_3426"), val = tensor([1, 1])]; + tensor var_3428 = const()[name = tensor("op_3428"), val = tensor([1, 1])]; + tensor input_39_pad_type_0 = const()[name = tensor("input_39_pad_type_0"), val = tensor("custom")]; + tensor input_39_pad_0 = const()[name = tensor("input_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_1_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_1_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35302720)))]; + tensor input_39_cast_fp16 = conv(dilations = var_3428, groups = var_1834, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = var_3426, weight = layers_1_fc2_loraA_weight_to_fp16, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor var_3432 = const()[name = tensor("op_3432"), val = tensor([1, 1])]; + tensor var_3434 = const()[name = tensor("op_3434"), val = tensor([1, 1])]; + tensor lora_out_45_pad_type_0 = const()[name = tensor("lora_out_45_pad_type_0"), val = tensor("custom")]; + tensor lora_out_45_pad_0 = const()[name = tensor("lora_out_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_47_weight_0_to_fp16 = const()[name = tensor("lora_out_47_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35466624)))]; + tensor lora_out_47_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_3434, groups = var_1834, pad = lora_out_45_pad_0, pad_type = lora_out_45_pad_type_0, strides = var_3432, weight = lora_out_47_weight_0_to_fp16, x = input_39_cast_fp16)[name = tensor("lora_out_47_cast_fp16")]; + tensor hidden_states_7_cast_fp16 = add(x = pretrained_out_23_cast_fp16, y = lora_out_47_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor var_3444 = const()[name = tensor("op_3444"), val = tensor(3)]; + tensor var_3469 = const()[name = tensor("op_3469"), val = tensor(1)]; + tensor var_3470 = const()[name = tensor("op_3470"), val = tensor(true)]; + tensor var_3480 = const()[name = tensor("op_3480"), val = tensor([1])]; + tensor channels_mean_9_cast_fp16 = reduce_mean(axes = var_3480, keep_dims = var_3470, x = inputs_9_cast_fp16)[name = tensor("channels_mean_9_cast_fp16")]; + tensor zero_mean_9_cast_fp16 = sub(x = inputs_9_cast_fp16, y = channels_mean_9_cast_fp16)[name = tensor("zero_mean_9_cast_fp16")]; + tensor zero_mean_sq_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = zero_mean_9_cast_fp16)[name = tensor("zero_mean_sq_9_cast_fp16")]; + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1])]; + tensor var_3485_cast_fp16 = reduce_mean(axes = var_3484, keep_dims = var_3470, x = zero_mean_sq_9_cast_fp16)[name = tensor("op_3485_cast_fp16")]; + tensor var_3486_to_fp16 = const()[name = tensor("op_3486_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3487_cast_fp16 = add(x = var_3485_cast_fp16, y = var_3486_to_fp16)[name = tensor("op_3487_cast_fp16")]; + tensor denom_9_epsilon_0 = const()[name = tensor("denom_9_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_9_cast_fp16 = rsqrt(epsilon = denom_9_epsilon_0, x = var_3487_cast_fp16)[name = tensor("denom_9_cast_fp16")]; + tensor out_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = denom_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = tensor("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35507648)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = tensor("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35510272)))]; + tensor obj_9_epsilon_0_to_fp16 = const()[name = tensor("obj_9_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor("obj_9_cast_fp16")]; + tensor var_3505 = const()[name = tensor("op_3505"), val = tensor([1, 1])]; + tensor var_3507 = const()[name = tensor("op_3507"), val = tensor([1, 1])]; + tensor pretrained_out_25_pad_type_0 = const()[name = tensor("pretrained_out_25_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_25_pad_0 = const()[name = tensor("pretrained_out_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35512896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36332160))), name = tensor("layers_2_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_2_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36332288)))]; + tensor pretrained_out_25_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_3507, groups = var_3469, pad = pretrained_out_25_pad_0, pad_type = pretrained_out_25_pad_type_0, strides = var_3505, weight = layers_2_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor("pretrained_out_25_cast_fp16")]; + tensor var_3511 = const()[name = tensor("op_3511"), val = tensor([1, 1])]; + tensor var_3513 = const()[name = tensor("op_3513"), val = tensor([1, 1])]; + tensor input_41_pad_type_0 = const()[name = tensor("input_41_pad_type_0"), val = tensor("custom")]; + tensor input_41_pad_0 = const()[name = tensor("input_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36334912)))]; + tensor input_41_cast_fp16 = conv(dilations = var_3513, groups = var_3469, pad = input_41_pad_0, pad_type = input_41_pad_type_0, strides = var_3511, weight = layers_2_self_attn_q_proj_loraA_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor var_3517 = const()[name = tensor("op_3517"), val = tensor([1, 1])]; + tensor var_3519 = const()[name = tensor("op_3519"), val = tensor([1, 1])]; + tensor lora_out_49_pad_type_0 = const()[name = tensor("lora_out_49_pad_type_0"), val = tensor("custom")]; + tensor lora_out_49_pad_0 = const()[name = tensor("lora_out_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_51_weight_0_to_fp16 = const()[name = tensor("lora_out_51_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36375936)))]; + tensor lora_out_51_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_3519, groups = var_3469, pad = lora_out_49_pad_0, pad_type = lora_out_49_pad_type_0, strides = var_3517, weight = lora_out_51_weight_0_to_fp16, x = input_41_cast_fp16)[name = tensor("lora_out_51_cast_fp16")]; + tensor query_5_cast_fp16 = add(x = pretrained_out_25_cast_fp16, y = lora_out_51_cast_fp16)[name = tensor("query_5_cast_fp16")]; + tensor var_3529 = const()[name = tensor("op_3529"), val = tensor([1, 1])]; + tensor var_3531 = const()[name = tensor("op_3531"), val = tensor([1, 1])]; + tensor pretrained_out_27_pad_type_0 = const()[name = tensor("pretrained_out_27_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_27_pad_0 = const()[name = tensor("pretrained_out_27_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36416960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37236224))), name = tensor("layers_2_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_27_cast_fp16 = conv(dilations = var_3531, groups = var_3469, pad = pretrained_out_27_pad_0, pad_type = pretrained_out_27_pad_type_0, strides = var_3529, weight = layers_2_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor("pretrained_out_27_cast_fp16")]; + tensor var_3535 = const()[name = tensor("op_3535"), val = tensor([1, 1])]; + tensor var_3537 = const()[name = tensor("op_3537"), val = tensor([1, 1])]; + tensor input_43_pad_type_0 = const()[name = tensor("input_43_pad_type_0"), val = tensor("custom")]; + tensor input_43_pad_0 = const()[name = tensor("input_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37236352)))]; + tensor input_43_cast_fp16 = conv(dilations = var_3537, groups = var_3469, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = var_3535, weight = layers_2_self_attn_k_proj_loraA_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor var_3541 = const()[name = tensor("op_3541"), val = tensor([1, 1])]; + tensor var_3543 = const()[name = tensor("op_3543"), val = tensor([1, 1])]; + tensor lora_out_53_pad_type_0 = const()[name = tensor("lora_out_53_pad_type_0"), val = tensor("custom")]; + tensor lora_out_53_pad_0 = const()[name = tensor("lora_out_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_55_weight_0_to_fp16 = const()[name = tensor("lora_out_55_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37277376)))]; + tensor lora_out_55_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_3543, groups = var_3469, pad = lora_out_53_pad_0, pad_type = lora_out_53_pad_type_0, strides = var_3541, weight = lora_out_55_weight_0_to_fp16, x = input_43_cast_fp16)[name = tensor("lora_out_55_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = pretrained_out_27_cast_fp16, y = lora_out_55_cast_fp16)[name = tensor("key_5_cast_fp16")]; + tensor var_3554 = const()[name = tensor("op_3554"), val = tensor([1, 1])]; + tensor var_3556 = const()[name = tensor("op_3556"), val = tensor([1, 1])]; + tensor pretrained_out_29_pad_type_0 = const()[name = tensor("pretrained_out_29_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_29_pad_0 = const()[name = tensor("pretrained_out_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37318400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38137664))), name = tensor("layers_2_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_2_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38137792)))]; + tensor pretrained_out_29_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_3556, groups = var_3469, pad = pretrained_out_29_pad_0, pad_type = pretrained_out_29_pad_type_0, strides = var_3554, weight = layers_2_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor("pretrained_out_29_cast_fp16")]; + tensor var_3560 = const()[name = tensor("op_3560"), val = tensor([1, 1])]; + tensor var_3562 = const()[name = tensor("op_3562"), val = tensor([1, 1])]; + tensor input_45_pad_type_0 = const()[name = tensor("input_45_pad_type_0"), val = tensor("custom")]; + tensor input_45_pad_0 = const()[name = tensor("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38140416)))]; + tensor input_45_cast_fp16 = conv(dilations = var_3562, groups = var_3469, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = var_3560, weight = layers_2_self_attn_v_proj_loraA_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor var_3566 = const()[name = tensor("op_3566"), val = tensor([1, 1])]; + tensor var_3568 = const()[name = tensor("op_3568"), val = tensor([1, 1])]; + tensor lora_out_57_pad_type_0 = const()[name = tensor("lora_out_57_pad_type_0"), val = tensor("custom")]; + tensor lora_out_57_pad_0 = const()[name = tensor("lora_out_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_59_weight_0_to_fp16 = const()[name = tensor("lora_out_59_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38181440)))]; + tensor lora_out_59_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_3568, groups = var_3469, pad = lora_out_57_pad_0, pad_type = lora_out_57_pad_type_0, strides = var_3566, weight = lora_out_59_weight_0_to_fp16, x = input_45_cast_fp16)[name = tensor("lora_out_59_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = pretrained_out_29_cast_fp16, y = lora_out_59_cast_fp16)[name = tensor("value_5_cast_fp16")]; + tensor var_3578_begin_0 = const()[name = tensor("op_3578_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3578_end_0 = const()[name = tensor("op_3578_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3578_end_mask_0 = const()[name = tensor("op_3578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3578_cast_fp16 = slice_by_index(begin = var_3578_begin_0, end = var_3578_end_0, end_mask = var_3578_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3578_cast_fp16")]; + tensor var_3582_begin_0 = const()[name = tensor("op_3582_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3582_end_0 = const()[name = tensor("op_3582_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3582_end_mask_0 = const()[name = tensor("op_3582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3582_cast_fp16 = slice_by_index(begin = var_3582_begin_0, end = var_3582_end_0, end_mask = var_3582_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3582_cast_fp16")]; + tensor var_3586_begin_0 = const()[name = tensor("op_3586_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3586_end_0 = const()[name = tensor("op_3586_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3586_end_mask_0 = const()[name = tensor("op_3586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3586_cast_fp16 = slice_by_index(begin = var_3586_begin_0, end = var_3586_end_0, end_mask = var_3586_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3586_cast_fp16")]; + tensor var_3590_begin_0 = const()[name = tensor("op_3590_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3590_end_0 = const()[name = tensor("op_3590_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3590_end_mask_0 = const()[name = tensor("op_3590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3590_cast_fp16 = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3590_cast_fp16")]; + tensor var_3594_begin_0 = const()[name = tensor("op_3594_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3594_end_0 = const()[name = tensor("op_3594_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3594_end_mask_0 = const()[name = tensor("op_3594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3594_cast_fp16 = slice_by_index(begin = var_3594_begin_0, end = var_3594_end_0, end_mask = var_3594_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3594_cast_fp16")]; + tensor var_3598_begin_0 = const()[name = tensor("op_3598_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3598_end_0 = const()[name = tensor("op_3598_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3598_end_mask_0 = const()[name = tensor("op_3598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3598_cast_fp16 = slice_by_index(begin = var_3598_begin_0, end = var_3598_end_0, end_mask = var_3598_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3598_cast_fp16")]; + tensor var_3602_begin_0 = const()[name = tensor("op_3602_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3602_end_0 = const()[name = tensor("op_3602_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3602_end_mask_0 = const()[name = tensor("op_3602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3602_cast_fp16 = slice_by_index(begin = var_3602_begin_0, end = var_3602_end_0, end_mask = var_3602_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3602_cast_fp16")]; + tensor var_3606_begin_0 = const()[name = tensor("op_3606_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3606_end_0 = const()[name = tensor("op_3606_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3606_end_mask_0 = const()[name = tensor("op_3606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3606_cast_fp16 = slice_by_index(begin = var_3606_begin_0, end = var_3606_end_0, end_mask = var_3606_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3606_cast_fp16")]; + tensor var_3610_begin_0 = const()[name = tensor("op_3610_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_3610_end_0 = const()[name = tensor("op_3610_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_3610_end_mask_0 = const()[name = tensor("op_3610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3610_cast_fp16 = slice_by_index(begin = var_3610_begin_0, end = var_3610_end_0, end_mask = var_3610_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3610_cast_fp16")]; + tensor var_3614_begin_0 = const()[name = tensor("op_3614_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_3614_end_0 = const()[name = tensor("op_3614_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_3614_end_mask_0 = const()[name = tensor("op_3614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3614_cast_fp16 = slice_by_index(begin = var_3614_begin_0, end = var_3614_end_0, end_mask = var_3614_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3614_cast_fp16")]; + tensor var_3618_begin_0 = const()[name = tensor("op_3618_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_3618_end_0 = const()[name = tensor("op_3618_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_3618_end_mask_0 = const()[name = tensor("op_3618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3618_cast_fp16 = slice_by_index(begin = var_3618_begin_0, end = var_3618_end_0, end_mask = var_3618_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3618_cast_fp16")]; + tensor var_3622_begin_0 = const()[name = tensor("op_3622_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_3622_end_0 = const()[name = tensor("op_3622_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_3622_end_mask_0 = const()[name = tensor("op_3622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3622_cast_fp16 = slice_by_index(begin = var_3622_begin_0, end = var_3622_end_0, end_mask = var_3622_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3622_cast_fp16")]; + tensor var_3626_begin_0 = const()[name = tensor("op_3626_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_3626_end_0 = const()[name = tensor("op_3626_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_3626_end_mask_0 = const()[name = tensor("op_3626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3626_cast_fp16 = slice_by_index(begin = var_3626_begin_0, end = var_3626_end_0, end_mask = var_3626_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3626_cast_fp16")]; + tensor var_3630_begin_0 = const()[name = tensor("op_3630_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_3630_end_0 = const()[name = tensor("op_3630_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_3630_end_mask_0 = const()[name = tensor("op_3630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3630_cast_fp16 = slice_by_index(begin = var_3630_begin_0, end = var_3630_end_0, end_mask = var_3630_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3630_cast_fp16")]; + tensor var_3634_begin_0 = const()[name = tensor("op_3634_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_3634_end_0 = const()[name = tensor("op_3634_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_3634_end_mask_0 = const()[name = tensor("op_3634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3634_cast_fp16 = slice_by_index(begin = var_3634_begin_0, end = var_3634_end_0, end_mask = var_3634_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3634_cast_fp16")]; + tensor var_3638_begin_0 = const()[name = tensor("op_3638_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_3638_end_0 = const()[name = tensor("op_3638_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_3638_end_mask_0 = const()[name = tensor("op_3638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3638_cast_fp16 = slice_by_index(begin = var_3638_begin_0, end = var_3638_end_0, end_mask = var_3638_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3638_cast_fp16")]; + tensor var_3642_begin_0 = const()[name = tensor("op_3642_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_3642_end_0 = const()[name = tensor("op_3642_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_3642_end_mask_0 = const()[name = tensor("op_3642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3642_cast_fp16 = slice_by_index(begin = var_3642_begin_0, end = var_3642_end_0, end_mask = var_3642_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3642_cast_fp16")]; + tensor var_3646_begin_0 = const()[name = tensor("op_3646_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_3646_end_0 = const()[name = tensor("op_3646_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_3646_end_mask_0 = const()[name = tensor("op_3646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3646_cast_fp16 = slice_by_index(begin = var_3646_begin_0, end = var_3646_end_0, end_mask = var_3646_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3646_cast_fp16")]; + tensor var_3650_begin_0 = const()[name = tensor("op_3650_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_3650_end_0 = const()[name = tensor("op_3650_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_3650_end_mask_0 = const()[name = tensor("op_3650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3650_cast_fp16 = slice_by_index(begin = var_3650_begin_0, end = var_3650_end_0, end_mask = var_3650_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3650_cast_fp16")]; + tensor var_3654_begin_0 = const()[name = tensor("op_3654_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_3654_end_0 = const()[name = tensor("op_3654_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_3654_end_mask_0 = const()[name = tensor("op_3654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3654_cast_fp16 = slice_by_index(begin = var_3654_begin_0, end = var_3654_end_0, end_mask = var_3654_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_3654_cast_fp16")]; + tensor var_3663_begin_0 = const()[name = tensor("op_3663_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3663_end_0 = const()[name = tensor("op_3663_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3663_end_mask_0 = const()[name = tensor("op_3663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3663_cast_fp16 = slice_by_index(begin = var_3663_begin_0, end = var_3663_end_0, end_mask = var_3663_end_mask_0, x = var_3578_cast_fp16)[name = tensor("op_3663_cast_fp16")]; + tensor var_3670_begin_0 = const()[name = tensor("op_3670_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3670_end_0 = const()[name = tensor("op_3670_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3670_end_mask_0 = const()[name = tensor("op_3670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3670_cast_fp16 = slice_by_index(begin = var_3670_begin_0, end = var_3670_end_0, end_mask = var_3670_end_mask_0, x = var_3578_cast_fp16)[name = tensor("op_3670_cast_fp16")]; + tensor var_3677_begin_0 = const()[name = tensor("op_3677_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3677_end_0 = const()[name = tensor("op_3677_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3677_end_mask_0 = const()[name = tensor("op_3677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3677_cast_fp16 = slice_by_index(begin = var_3677_begin_0, end = var_3677_end_0, end_mask = var_3677_end_mask_0, x = var_3578_cast_fp16)[name = tensor("op_3677_cast_fp16")]; + tensor var_3684_begin_0 = const()[name = tensor("op_3684_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3684_end_0 = const()[name = tensor("op_3684_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3684_end_mask_0 = const()[name = tensor("op_3684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3684_cast_fp16 = slice_by_index(begin = var_3684_begin_0, end = var_3684_end_0, end_mask = var_3684_end_mask_0, x = var_3578_cast_fp16)[name = tensor("op_3684_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = tensor("op_3691_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3691_end_0 = const()[name = tensor("op_3691_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3691_end_mask_0 = const()[name = tensor("op_3691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = var_3582_cast_fp16)[name = tensor("op_3691_cast_fp16")]; + tensor var_3698_begin_0 = const()[name = tensor("op_3698_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3698_end_0 = const()[name = tensor("op_3698_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3698_end_mask_0 = const()[name = tensor("op_3698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = var_3582_cast_fp16)[name = tensor("op_3698_cast_fp16")]; + tensor var_3705_begin_0 = const()[name = tensor("op_3705_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3705_end_0 = const()[name = tensor("op_3705_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3705_end_mask_0 = const()[name = tensor("op_3705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3705_cast_fp16 = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = var_3582_cast_fp16)[name = tensor("op_3705_cast_fp16")]; + tensor var_3712_begin_0 = const()[name = tensor("op_3712_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3712_end_0 = const()[name = tensor("op_3712_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3712_end_mask_0 = const()[name = tensor("op_3712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3712_cast_fp16 = slice_by_index(begin = var_3712_begin_0, end = var_3712_end_0, end_mask = var_3712_end_mask_0, x = var_3582_cast_fp16)[name = tensor("op_3712_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = tensor("op_3719_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3719_end_0 = const()[name = tensor("op_3719_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3719_end_mask_0 = const()[name = tensor("op_3719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = var_3586_cast_fp16)[name = tensor("op_3719_cast_fp16")]; + tensor var_3726_begin_0 = const()[name = tensor("op_3726_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3726_end_0 = const()[name = tensor("op_3726_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3726_end_mask_0 = const()[name = tensor("op_3726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3726_cast_fp16 = slice_by_index(begin = var_3726_begin_0, end = var_3726_end_0, end_mask = var_3726_end_mask_0, x = var_3586_cast_fp16)[name = tensor("op_3726_cast_fp16")]; + tensor var_3733_begin_0 = const()[name = tensor("op_3733_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3733_end_0 = const()[name = tensor("op_3733_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3733_end_mask_0 = const()[name = tensor("op_3733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3733_cast_fp16 = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = var_3586_cast_fp16)[name = tensor("op_3733_cast_fp16")]; + tensor var_3740_begin_0 = const()[name = tensor("op_3740_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3740_end_0 = const()[name = tensor("op_3740_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3740_end_mask_0 = const()[name = tensor("op_3740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3740_cast_fp16 = slice_by_index(begin = var_3740_begin_0, end = var_3740_end_0, end_mask = var_3740_end_mask_0, x = var_3586_cast_fp16)[name = tensor("op_3740_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = tensor("op_3747_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3747_end_0 = const()[name = tensor("op_3747_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3747_end_mask_0 = const()[name = tensor("op_3747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = var_3590_cast_fp16)[name = tensor("op_3747_cast_fp16")]; + tensor var_3754_begin_0 = const()[name = tensor("op_3754_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3754_end_0 = const()[name = tensor("op_3754_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3754_end_mask_0 = const()[name = tensor("op_3754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3754_cast_fp16 = slice_by_index(begin = var_3754_begin_0, end = var_3754_end_0, end_mask = var_3754_end_mask_0, x = var_3590_cast_fp16)[name = tensor("op_3754_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = tensor("op_3761_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3761_end_0 = const()[name = tensor("op_3761_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3761_end_mask_0 = const()[name = tensor("op_3761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = var_3590_cast_fp16)[name = tensor("op_3761_cast_fp16")]; + tensor var_3768_begin_0 = const()[name = tensor("op_3768_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3768_end_0 = const()[name = tensor("op_3768_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3768_end_mask_0 = const()[name = tensor("op_3768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3768_cast_fp16 = slice_by_index(begin = var_3768_begin_0, end = var_3768_end_0, end_mask = var_3768_end_mask_0, x = var_3590_cast_fp16)[name = tensor("op_3768_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = tensor("op_3775_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3775_end_0 = const()[name = tensor("op_3775_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3775_end_mask_0 = const()[name = tensor("op_3775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = var_3594_cast_fp16)[name = tensor("op_3775_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = tensor("op_3782_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3782_end_0 = const()[name = tensor("op_3782_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3782_end_mask_0 = const()[name = tensor("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = var_3594_cast_fp16)[name = tensor("op_3782_cast_fp16")]; + tensor var_3789_begin_0 = const()[name = tensor("op_3789_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3789_end_0 = const()[name = tensor("op_3789_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3789_end_mask_0 = const()[name = tensor("op_3789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3789_cast_fp16 = slice_by_index(begin = var_3789_begin_0, end = var_3789_end_0, end_mask = var_3789_end_mask_0, x = var_3594_cast_fp16)[name = tensor("op_3789_cast_fp16")]; + tensor var_3796_begin_0 = const()[name = tensor("op_3796_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3796_end_0 = const()[name = tensor("op_3796_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3796_end_mask_0 = const()[name = tensor("op_3796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3796_cast_fp16 = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = var_3594_cast_fp16)[name = tensor("op_3796_cast_fp16")]; + tensor var_3803_begin_0 = const()[name = tensor("op_3803_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3803_end_0 = const()[name = tensor("op_3803_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3803_end_mask_0 = const()[name = tensor("op_3803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3803_cast_fp16 = slice_by_index(begin = var_3803_begin_0, end = var_3803_end_0, end_mask = var_3803_end_mask_0, x = var_3598_cast_fp16)[name = tensor("op_3803_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = tensor("op_3810_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3810_end_0 = const()[name = tensor("op_3810_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3810_end_mask_0 = const()[name = tensor("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = var_3598_cast_fp16)[name = tensor("op_3810_cast_fp16")]; + tensor var_3817_begin_0 = const()[name = tensor("op_3817_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3817_end_0 = const()[name = tensor("op_3817_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3817_end_mask_0 = const()[name = tensor("op_3817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = var_3598_cast_fp16)[name = tensor("op_3817_cast_fp16")]; + tensor var_3824_begin_0 = const()[name = tensor("op_3824_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3824_end_0 = const()[name = tensor("op_3824_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3824_end_mask_0 = const()[name = tensor("op_3824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3824_cast_fp16 = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = var_3598_cast_fp16)[name = tensor("op_3824_cast_fp16")]; + tensor var_3831_begin_0 = const()[name = tensor("op_3831_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3831_end_0 = const()[name = tensor("op_3831_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3831_end_mask_0 = const()[name = tensor("op_3831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3831_cast_fp16 = slice_by_index(begin = var_3831_begin_0, end = var_3831_end_0, end_mask = var_3831_end_mask_0, x = var_3602_cast_fp16)[name = tensor("op_3831_cast_fp16")]; + tensor var_3838_begin_0 = const()[name = tensor("op_3838_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3838_end_0 = const()[name = tensor("op_3838_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3838_end_mask_0 = const()[name = tensor("op_3838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3838_cast_fp16 = slice_by_index(begin = var_3838_begin_0, end = var_3838_end_0, end_mask = var_3838_end_mask_0, x = var_3602_cast_fp16)[name = tensor("op_3838_cast_fp16")]; + tensor var_3845_begin_0 = const()[name = tensor("op_3845_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3845_end_0 = const()[name = tensor("op_3845_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3845_end_mask_0 = const()[name = tensor("op_3845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3845_cast_fp16 = slice_by_index(begin = var_3845_begin_0, end = var_3845_end_0, end_mask = var_3845_end_mask_0, x = var_3602_cast_fp16)[name = tensor("op_3845_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = tensor("op_3852_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3852_end_0 = const()[name = tensor("op_3852_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3852_end_mask_0 = const()[name = tensor("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = var_3602_cast_fp16)[name = tensor("op_3852_cast_fp16")]; + tensor var_3859_begin_0 = const()[name = tensor("op_3859_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3859_end_0 = const()[name = tensor("op_3859_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3859_end_mask_0 = const()[name = tensor("op_3859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3859_cast_fp16 = slice_by_index(begin = var_3859_begin_0, end = var_3859_end_0, end_mask = var_3859_end_mask_0, x = var_3606_cast_fp16)[name = tensor("op_3859_cast_fp16")]; + tensor var_3866_begin_0 = const()[name = tensor("op_3866_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3866_end_0 = const()[name = tensor("op_3866_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3866_end_mask_0 = const()[name = tensor("op_3866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = var_3606_cast_fp16)[name = tensor("op_3866_cast_fp16")]; + tensor var_3873_begin_0 = const()[name = tensor("op_3873_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3873_end_0 = const()[name = tensor("op_3873_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3873_end_mask_0 = const()[name = tensor("op_3873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3873_cast_fp16 = slice_by_index(begin = var_3873_begin_0, end = var_3873_end_0, end_mask = var_3873_end_mask_0, x = var_3606_cast_fp16)[name = tensor("op_3873_cast_fp16")]; + tensor var_3880_begin_0 = const()[name = tensor("op_3880_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3880_end_0 = const()[name = tensor("op_3880_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3880_end_mask_0 = const()[name = tensor("op_3880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3880_cast_fp16 = slice_by_index(begin = var_3880_begin_0, end = var_3880_end_0, end_mask = var_3880_end_mask_0, x = var_3606_cast_fp16)[name = tensor("op_3880_cast_fp16")]; + tensor var_3887_begin_0 = const()[name = tensor("op_3887_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3887_end_0 = const()[name = tensor("op_3887_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3887_end_mask_0 = const()[name = tensor("op_3887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3887_cast_fp16 = slice_by_index(begin = var_3887_begin_0, end = var_3887_end_0, end_mask = var_3887_end_mask_0, x = var_3610_cast_fp16)[name = tensor("op_3887_cast_fp16")]; + tensor var_3894_begin_0 = const()[name = tensor("op_3894_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3894_end_0 = const()[name = tensor("op_3894_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3894_end_mask_0 = const()[name = tensor("op_3894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = var_3610_cast_fp16)[name = tensor("op_3894_cast_fp16")]; + tensor var_3901_begin_0 = const()[name = tensor("op_3901_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3901_end_0 = const()[name = tensor("op_3901_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3901_end_mask_0 = const()[name = tensor("op_3901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3901_cast_fp16 = slice_by_index(begin = var_3901_begin_0, end = var_3901_end_0, end_mask = var_3901_end_mask_0, x = var_3610_cast_fp16)[name = tensor("op_3901_cast_fp16")]; + tensor var_3908_begin_0 = const()[name = tensor("op_3908_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3908_end_0 = const()[name = tensor("op_3908_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3908_end_mask_0 = const()[name = tensor("op_3908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3908_cast_fp16 = slice_by_index(begin = var_3908_begin_0, end = var_3908_end_0, end_mask = var_3908_end_mask_0, x = var_3610_cast_fp16)[name = tensor("op_3908_cast_fp16")]; + tensor var_3915_begin_0 = const()[name = tensor("op_3915_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3915_end_0 = const()[name = tensor("op_3915_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3915_end_mask_0 = const()[name = tensor("op_3915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3915_cast_fp16 = slice_by_index(begin = var_3915_begin_0, end = var_3915_end_0, end_mask = var_3915_end_mask_0, x = var_3614_cast_fp16)[name = tensor("op_3915_cast_fp16")]; + tensor var_3922_begin_0 = const()[name = tensor("op_3922_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3922_end_0 = const()[name = tensor("op_3922_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3922_end_mask_0 = const()[name = tensor("op_3922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3922_cast_fp16 = slice_by_index(begin = var_3922_begin_0, end = var_3922_end_0, end_mask = var_3922_end_mask_0, x = var_3614_cast_fp16)[name = tensor("op_3922_cast_fp16")]; + tensor var_3929_begin_0 = const()[name = tensor("op_3929_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3929_end_0 = const()[name = tensor("op_3929_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3929_end_mask_0 = const()[name = tensor("op_3929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3929_cast_fp16 = slice_by_index(begin = var_3929_begin_0, end = var_3929_end_0, end_mask = var_3929_end_mask_0, x = var_3614_cast_fp16)[name = tensor("op_3929_cast_fp16")]; + tensor var_3936_begin_0 = const()[name = tensor("op_3936_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3936_end_0 = const()[name = tensor("op_3936_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3936_end_mask_0 = const()[name = tensor("op_3936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3936_cast_fp16 = slice_by_index(begin = var_3936_begin_0, end = var_3936_end_0, end_mask = var_3936_end_mask_0, x = var_3614_cast_fp16)[name = tensor("op_3936_cast_fp16")]; + tensor var_3943_begin_0 = const()[name = tensor("op_3943_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3943_end_0 = const()[name = tensor("op_3943_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3943_end_mask_0 = const()[name = tensor("op_3943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3943_cast_fp16 = slice_by_index(begin = var_3943_begin_0, end = var_3943_end_0, end_mask = var_3943_end_mask_0, x = var_3618_cast_fp16)[name = tensor("op_3943_cast_fp16")]; + tensor var_3950_begin_0 = const()[name = tensor("op_3950_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3950_end_0 = const()[name = tensor("op_3950_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3950_end_mask_0 = const()[name = tensor("op_3950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3950_cast_fp16 = slice_by_index(begin = var_3950_begin_0, end = var_3950_end_0, end_mask = var_3950_end_mask_0, x = var_3618_cast_fp16)[name = tensor("op_3950_cast_fp16")]; + tensor var_3957_begin_0 = const()[name = tensor("op_3957_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3957_end_0 = const()[name = tensor("op_3957_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3957_end_mask_0 = const()[name = tensor("op_3957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3957_cast_fp16 = slice_by_index(begin = var_3957_begin_0, end = var_3957_end_0, end_mask = var_3957_end_mask_0, x = var_3618_cast_fp16)[name = tensor("op_3957_cast_fp16")]; + tensor var_3964_begin_0 = const()[name = tensor("op_3964_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3964_end_0 = const()[name = tensor("op_3964_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3964_end_mask_0 = const()[name = tensor("op_3964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3964_cast_fp16 = slice_by_index(begin = var_3964_begin_0, end = var_3964_end_0, end_mask = var_3964_end_mask_0, x = var_3618_cast_fp16)[name = tensor("op_3964_cast_fp16")]; + tensor var_3971_begin_0 = const()[name = tensor("op_3971_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3971_end_0 = const()[name = tensor("op_3971_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3971_end_mask_0 = const()[name = tensor("op_3971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3971_cast_fp16 = slice_by_index(begin = var_3971_begin_0, end = var_3971_end_0, end_mask = var_3971_end_mask_0, x = var_3622_cast_fp16)[name = tensor("op_3971_cast_fp16")]; + tensor var_3978_begin_0 = const()[name = tensor("op_3978_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3978_end_0 = const()[name = tensor("op_3978_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3978_end_mask_0 = const()[name = tensor("op_3978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3978_cast_fp16 = slice_by_index(begin = var_3978_begin_0, end = var_3978_end_0, end_mask = var_3978_end_mask_0, x = var_3622_cast_fp16)[name = tensor("op_3978_cast_fp16")]; + tensor var_3985_begin_0 = const()[name = tensor("op_3985_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3985_end_0 = const()[name = tensor("op_3985_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3985_end_mask_0 = const()[name = tensor("op_3985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3985_cast_fp16 = slice_by_index(begin = var_3985_begin_0, end = var_3985_end_0, end_mask = var_3985_end_mask_0, x = var_3622_cast_fp16)[name = tensor("op_3985_cast_fp16")]; + tensor var_3992_begin_0 = const()[name = tensor("op_3992_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3992_end_0 = const()[name = tensor("op_3992_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3992_end_mask_0 = const()[name = tensor("op_3992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3992_cast_fp16 = slice_by_index(begin = var_3992_begin_0, end = var_3992_end_0, end_mask = var_3992_end_mask_0, x = var_3622_cast_fp16)[name = tensor("op_3992_cast_fp16")]; + tensor var_3999_begin_0 = const()[name = tensor("op_3999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3999_end_0 = const()[name = tensor("op_3999_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3999_end_mask_0 = const()[name = tensor("op_3999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3999_cast_fp16 = slice_by_index(begin = var_3999_begin_0, end = var_3999_end_0, end_mask = var_3999_end_mask_0, x = var_3626_cast_fp16)[name = tensor("op_3999_cast_fp16")]; + tensor var_4006_begin_0 = const()[name = tensor("op_4006_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4006_end_0 = const()[name = tensor("op_4006_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4006_end_mask_0 = const()[name = tensor("op_4006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4006_cast_fp16 = slice_by_index(begin = var_4006_begin_0, end = var_4006_end_0, end_mask = var_4006_end_mask_0, x = var_3626_cast_fp16)[name = tensor("op_4006_cast_fp16")]; + tensor var_4013_begin_0 = const()[name = tensor("op_4013_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4013_end_0 = const()[name = tensor("op_4013_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4013_end_mask_0 = const()[name = tensor("op_4013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4013_cast_fp16 = slice_by_index(begin = var_4013_begin_0, end = var_4013_end_0, end_mask = var_4013_end_mask_0, x = var_3626_cast_fp16)[name = tensor("op_4013_cast_fp16")]; + tensor var_4020_begin_0 = const()[name = tensor("op_4020_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4020_end_0 = const()[name = tensor("op_4020_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4020_end_mask_0 = const()[name = tensor("op_4020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4020_cast_fp16 = slice_by_index(begin = var_4020_begin_0, end = var_4020_end_0, end_mask = var_4020_end_mask_0, x = var_3626_cast_fp16)[name = tensor("op_4020_cast_fp16")]; + tensor var_4027_begin_0 = const()[name = tensor("op_4027_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4027_end_0 = const()[name = tensor("op_4027_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4027_end_mask_0 = const()[name = tensor("op_4027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4027_cast_fp16 = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = var_3630_cast_fp16)[name = tensor("op_4027_cast_fp16")]; + tensor var_4034_begin_0 = const()[name = tensor("op_4034_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4034_end_0 = const()[name = tensor("op_4034_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4034_end_mask_0 = const()[name = tensor("op_4034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4034_cast_fp16 = slice_by_index(begin = var_4034_begin_0, end = var_4034_end_0, end_mask = var_4034_end_mask_0, x = var_3630_cast_fp16)[name = tensor("op_4034_cast_fp16")]; + tensor var_4041_begin_0 = const()[name = tensor("op_4041_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4041_end_0 = const()[name = tensor("op_4041_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4041_end_mask_0 = const()[name = tensor("op_4041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4041_cast_fp16 = slice_by_index(begin = var_4041_begin_0, end = var_4041_end_0, end_mask = var_4041_end_mask_0, x = var_3630_cast_fp16)[name = tensor("op_4041_cast_fp16")]; + tensor var_4048_begin_0 = const()[name = tensor("op_4048_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4048_end_0 = const()[name = tensor("op_4048_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4048_end_mask_0 = const()[name = tensor("op_4048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4048_cast_fp16 = slice_by_index(begin = var_4048_begin_0, end = var_4048_end_0, end_mask = var_4048_end_mask_0, x = var_3630_cast_fp16)[name = tensor("op_4048_cast_fp16")]; + tensor var_4055_begin_0 = const()[name = tensor("op_4055_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4055_end_0 = const()[name = tensor("op_4055_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4055_end_mask_0 = const()[name = tensor("op_4055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4055_cast_fp16 = slice_by_index(begin = var_4055_begin_0, end = var_4055_end_0, end_mask = var_4055_end_mask_0, x = var_3634_cast_fp16)[name = tensor("op_4055_cast_fp16")]; + tensor var_4062_begin_0 = const()[name = tensor("op_4062_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4062_end_0 = const()[name = tensor("op_4062_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4062_end_mask_0 = const()[name = tensor("op_4062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4062_cast_fp16 = slice_by_index(begin = var_4062_begin_0, end = var_4062_end_0, end_mask = var_4062_end_mask_0, x = var_3634_cast_fp16)[name = tensor("op_4062_cast_fp16")]; + tensor var_4069_begin_0 = const()[name = tensor("op_4069_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4069_end_0 = const()[name = tensor("op_4069_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4069_end_mask_0 = const()[name = tensor("op_4069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4069_cast_fp16 = slice_by_index(begin = var_4069_begin_0, end = var_4069_end_0, end_mask = var_4069_end_mask_0, x = var_3634_cast_fp16)[name = tensor("op_4069_cast_fp16")]; + tensor var_4076_begin_0 = const()[name = tensor("op_4076_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4076_end_0 = const()[name = tensor("op_4076_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4076_end_mask_0 = const()[name = tensor("op_4076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4076_cast_fp16 = slice_by_index(begin = var_4076_begin_0, end = var_4076_end_0, end_mask = var_4076_end_mask_0, x = var_3634_cast_fp16)[name = tensor("op_4076_cast_fp16")]; + tensor var_4083_begin_0 = const()[name = tensor("op_4083_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4083_end_0 = const()[name = tensor("op_4083_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4083_end_mask_0 = const()[name = tensor("op_4083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4083_cast_fp16 = slice_by_index(begin = var_4083_begin_0, end = var_4083_end_0, end_mask = var_4083_end_mask_0, x = var_3638_cast_fp16)[name = tensor("op_4083_cast_fp16")]; + tensor var_4090_begin_0 = const()[name = tensor("op_4090_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4090_end_0 = const()[name = tensor("op_4090_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4090_end_mask_0 = const()[name = tensor("op_4090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4090_cast_fp16 = slice_by_index(begin = var_4090_begin_0, end = var_4090_end_0, end_mask = var_4090_end_mask_0, x = var_3638_cast_fp16)[name = tensor("op_4090_cast_fp16")]; + tensor var_4097_begin_0 = const()[name = tensor("op_4097_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4097_end_0 = const()[name = tensor("op_4097_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4097_end_mask_0 = const()[name = tensor("op_4097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4097_cast_fp16 = slice_by_index(begin = var_4097_begin_0, end = var_4097_end_0, end_mask = var_4097_end_mask_0, x = var_3638_cast_fp16)[name = tensor("op_4097_cast_fp16")]; + tensor var_4104_begin_0 = const()[name = tensor("op_4104_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4104_end_0 = const()[name = tensor("op_4104_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4104_end_mask_0 = const()[name = tensor("op_4104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4104_cast_fp16 = slice_by_index(begin = var_4104_begin_0, end = var_4104_end_0, end_mask = var_4104_end_mask_0, x = var_3638_cast_fp16)[name = tensor("op_4104_cast_fp16")]; + tensor var_4111_begin_0 = const()[name = tensor("op_4111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4111_end_0 = const()[name = tensor("op_4111_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4111_end_mask_0 = const()[name = tensor("op_4111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4111_cast_fp16 = slice_by_index(begin = var_4111_begin_0, end = var_4111_end_0, end_mask = var_4111_end_mask_0, x = var_3642_cast_fp16)[name = tensor("op_4111_cast_fp16")]; + tensor var_4118_begin_0 = const()[name = tensor("op_4118_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4118_end_0 = const()[name = tensor("op_4118_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4118_end_mask_0 = const()[name = tensor("op_4118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4118_cast_fp16 = slice_by_index(begin = var_4118_begin_0, end = var_4118_end_0, end_mask = var_4118_end_mask_0, x = var_3642_cast_fp16)[name = tensor("op_4118_cast_fp16")]; + tensor var_4125_begin_0 = const()[name = tensor("op_4125_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4125_end_0 = const()[name = tensor("op_4125_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4125_end_mask_0 = const()[name = tensor("op_4125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4125_cast_fp16 = slice_by_index(begin = var_4125_begin_0, end = var_4125_end_0, end_mask = var_4125_end_mask_0, x = var_3642_cast_fp16)[name = tensor("op_4125_cast_fp16")]; + tensor var_4132_begin_0 = const()[name = tensor("op_4132_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4132_end_0 = const()[name = tensor("op_4132_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4132_end_mask_0 = const()[name = tensor("op_4132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4132_cast_fp16 = slice_by_index(begin = var_4132_begin_0, end = var_4132_end_0, end_mask = var_4132_end_mask_0, x = var_3642_cast_fp16)[name = tensor("op_4132_cast_fp16")]; + tensor var_4139_begin_0 = const()[name = tensor("op_4139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4139_end_0 = const()[name = tensor("op_4139_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4139_end_mask_0 = const()[name = tensor("op_4139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4139_cast_fp16 = slice_by_index(begin = var_4139_begin_0, end = var_4139_end_0, end_mask = var_4139_end_mask_0, x = var_3646_cast_fp16)[name = tensor("op_4139_cast_fp16")]; + tensor var_4146_begin_0 = const()[name = tensor("op_4146_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4146_end_0 = const()[name = tensor("op_4146_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4146_end_mask_0 = const()[name = tensor("op_4146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4146_cast_fp16 = slice_by_index(begin = var_4146_begin_0, end = var_4146_end_0, end_mask = var_4146_end_mask_0, x = var_3646_cast_fp16)[name = tensor("op_4146_cast_fp16")]; + tensor var_4153_begin_0 = const()[name = tensor("op_4153_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4153_end_0 = const()[name = tensor("op_4153_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4153_end_mask_0 = const()[name = tensor("op_4153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4153_cast_fp16 = slice_by_index(begin = var_4153_begin_0, end = var_4153_end_0, end_mask = var_4153_end_mask_0, x = var_3646_cast_fp16)[name = tensor("op_4153_cast_fp16")]; + tensor var_4160_begin_0 = const()[name = tensor("op_4160_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4160_end_0 = const()[name = tensor("op_4160_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4160_end_mask_0 = const()[name = tensor("op_4160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4160_cast_fp16 = slice_by_index(begin = var_4160_begin_0, end = var_4160_end_0, end_mask = var_4160_end_mask_0, x = var_3646_cast_fp16)[name = tensor("op_4160_cast_fp16")]; + tensor var_4167_begin_0 = const()[name = tensor("op_4167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4167_end_0 = const()[name = tensor("op_4167_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4167_end_mask_0 = const()[name = tensor("op_4167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4167_cast_fp16 = slice_by_index(begin = var_4167_begin_0, end = var_4167_end_0, end_mask = var_4167_end_mask_0, x = var_3650_cast_fp16)[name = tensor("op_4167_cast_fp16")]; + tensor var_4174_begin_0 = const()[name = tensor("op_4174_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4174_end_0 = const()[name = tensor("op_4174_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4174_end_mask_0 = const()[name = tensor("op_4174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4174_cast_fp16 = slice_by_index(begin = var_4174_begin_0, end = var_4174_end_0, end_mask = var_4174_end_mask_0, x = var_3650_cast_fp16)[name = tensor("op_4174_cast_fp16")]; + tensor var_4181_begin_0 = const()[name = tensor("op_4181_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4181_end_0 = const()[name = tensor("op_4181_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4181_end_mask_0 = const()[name = tensor("op_4181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4181_cast_fp16 = slice_by_index(begin = var_4181_begin_0, end = var_4181_end_0, end_mask = var_4181_end_mask_0, x = var_3650_cast_fp16)[name = tensor("op_4181_cast_fp16")]; + tensor var_4188_begin_0 = const()[name = tensor("op_4188_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4188_end_0 = const()[name = tensor("op_4188_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4188_end_mask_0 = const()[name = tensor("op_4188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4188_cast_fp16 = slice_by_index(begin = var_4188_begin_0, end = var_4188_end_0, end_mask = var_4188_end_mask_0, x = var_3650_cast_fp16)[name = tensor("op_4188_cast_fp16")]; + tensor var_4195_begin_0 = const()[name = tensor("op_4195_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4195_end_0 = const()[name = tensor("op_4195_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_4195_end_mask_0 = const()[name = tensor("op_4195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4195_cast_fp16 = slice_by_index(begin = var_4195_begin_0, end = var_4195_end_0, end_mask = var_4195_end_mask_0, x = var_3654_cast_fp16)[name = tensor("op_4195_cast_fp16")]; + tensor var_4202_begin_0 = const()[name = tensor("op_4202_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_4202_end_0 = const()[name = tensor("op_4202_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_4202_end_mask_0 = const()[name = tensor("op_4202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4202_cast_fp16 = slice_by_index(begin = var_4202_begin_0, end = var_4202_end_0, end_mask = var_4202_end_mask_0, x = var_3654_cast_fp16)[name = tensor("op_4202_cast_fp16")]; + tensor var_4209_begin_0 = const()[name = tensor("op_4209_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_4209_end_0 = const()[name = tensor("op_4209_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_4209_end_mask_0 = const()[name = tensor("op_4209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4209_cast_fp16 = slice_by_index(begin = var_4209_begin_0, end = var_4209_end_0, end_mask = var_4209_end_mask_0, x = var_3654_cast_fp16)[name = tensor("op_4209_cast_fp16")]; + tensor var_4216_begin_0 = const()[name = tensor("op_4216_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_4216_end_0 = const()[name = tensor("op_4216_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4216_end_mask_0 = const()[name = tensor("op_4216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4216_cast_fp16 = slice_by_index(begin = var_4216_begin_0, end = var_4216_end_0, end_mask = var_4216_end_mask_0, x = var_3654_cast_fp16)[name = tensor("op_4216_cast_fp16")]; + tensor k_5_perm_0 = const()[name = tensor("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_4221_begin_0 = const()[name = tensor("op_4221_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4221_end_0 = const()[name = tensor("op_4221_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_4221_end_mask_0 = const()[name = tensor("op_4221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_29 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = tensor("transpose_29")]; + tensor var_4221_cast_fp16 = slice_by_index(begin = var_4221_begin_0, end = var_4221_end_0, end_mask = var_4221_end_mask_0, x = transpose_29)[name = tensor("op_4221_cast_fp16")]; + tensor var_4225_begin_0 = const()[name = tensor("op_4225_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4225_end_0 = const()[name = tensor("op_4225_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_4225_end_mask_0 = const()[name = tensor("op_4225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4225_cast_fp16 = slice_by_index(begin = var_4225_begin_0, end = var_4225_end_0, end_mask = var_4225_end_mask_0, x = transpose_29)[name = tensor("op_4225_cast_fp16")]; + tensor var_4229_begin_0 = const()[name = tensor("op_4229_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_4229_end_0 = const()[name = tensor("op_4229_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_4229_end_mask_0 = const()[name = tensor("op_4229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4229_cast_fp16 = slice_by_index(begin = var_4229_begin_0, end = var_4229_end_0, end_mask = var_4229_end_mask_0, x = transpose_29)[name = tensor("op_4229_cast_fp16")]; + tensor var_4233_begin_0 = const()[name = tensor("op_4233_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_4233_end_0 = const()[name = tensor("op_4233_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_4233_end_mask_0 = const()[name = tensor("op_4233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4233_cast_fp16 = slice_by_index(begin = var_4233_begin_0, end = var_4233_end_0, end_mask = var_4233_end_mask_0, x = transpose_29)[name = tensor("op_4233_cast_fp16")]; + tensor var_4237_begin_0 = const()[name = tensor("op_4237_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_4237_end_0 = const()[name = tensor("op_4237_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_4237_end_mask_0 = const()[name = tensor("op_4237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4237_cast_fp16 = slice_by_index(begin = var_4237_begin_0, end = var_4237_end_0, end_mask = var_4237_end_mask_0, x = transpose_29)[name = tensor("op_4237_cast_fp16")]; + tensor var_4241_begin_0 = const()[name = tensor("op_4241_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_4241_end_0 = const()[name = tensor("op_4241_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_4241_end_mask_0 = const()[name = tensor("op_4241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4241_cast_fp16 = slice_by_index(begin = var_4241_begin_0, end = var_4241_end_0, end_mask = var_4241_end_mask_0, x = transpose_29)[name = tensor("op_4241_cast_fp16")]; + tensor var_4245_begin_0 = const()[name = tensor("op_4245_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_4245_end_0 = const()[name = tensor("op_4245_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_4245_end_mask_0 = const()[name = tensor("op_4245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4245_cast_fp16 = slice_by_index(begin = var_4245_begin_0, end = var_4245_end_0, end_mask = var_4245_end_mask_0, x = transpose_29)[name = tensor("op_4245_cast_fp16")]; + tensor var_4249_begin_0 = const()[name = tensor("op_4249_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_4249_end_0 = const()[name = tensor("op_4249_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_4249_end_mask_0 = const()[name = tensor("op_4249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4249_cast_fp16 = slice_by_index(begin = var_4249_begin_0, end = var_4249_end_0, end_mask = var_4249_end_mask_0, x = transpose_29)[name = tensor("op_4249_cast_fp16")]; + tensor var_4253_begin_0 = const()[name = tensor("op_4253_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_4253_end_0 = const()[name = tensor("op_4253_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_4253_end_mask_0 = const()[name = tensor("op_4253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4253_cast_fp16 = slice_by_index(begin = var_4253_begin_0, end = var_4253_end_0, end_mask = var_4253_end_mask_0, x = transpose_29)[name = tensor("op_4253_cast_fp16")]; + tensor var_4257_begin_0 = const()[name = tensor("op_4257_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_4257_end_0 = const()[name = tensor("op_4257_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_4257_end_mask_0 = const()[name = tensor("op_4257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4257_cast_fp16 = slice_by_index(begin = var_4257_begin_0, end = var_4257_end_0, end_mask = var_4257_end_mask_0, x = transpose_29)[name = tensor("op_4257_cast_fp16")]; + tensor var_4261_begin_0 = const()[name = tensor("op_4261_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_4261_end_0 = const()[name = tensor("op_4261_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_4261_end_mask_0 = const()[name = tensor("op_4261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4261_cast_fp16 = slice_by_index(begin = var_4261_begin_0, end = var_4261_end_0, end_mask = var_4261_end_mask_0, x = transpose_29)[name = tensor("op_4261_cast_fp16")]; + tensor var_4265_begin_0 = const()[name = tensor("op_4265_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_4265_end_0 = const()[name = tensor("op_4265_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_4265_end_mask_0 = const()[name = tensor("op_4265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4265_cast_fp16 = slice_by_index(begin = var_4265_begin_0, end = var_4265_end_0, end_mask = var_4265_end_mask_0, x = transpose_29)[name = tensor("op_4265_cast_fp16")]; + tensor var_4269_begin_0 = const()[name = tensor("op_4269_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_4269_end_0 = const()[name = tensor("op_4269_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_4269_end_mask_0 = const()[name = tensor("op_4269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4269_cast_fp16 = slice_by_index(begin = var_4269_begin_0, end = var_4269_end_0, end_mask = var_4269_end_mask_0, x = transpose_29)[name = tensor("op_4269_cast_fp16")]; + tensor var_4273_begin_0 = const()[name = tensor("op_4273_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_4273_end_0 = const()[name = tensor("op_4273_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_4273_end_mask_0 = const()[name = tensor("op_4273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = transpose_29)[name = tensor("op_4273_cast_fp16")]; + tensor var_4277_begin_0 = const()[name = tensor("op_4277_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_4277_end_0 = const()[name = tensor("op_4277_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_4277_end_mask_0 = const()[name = tensor("op_4277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4277_cast_fp16 = slice_by_index(begin = var_4277_begin_0, end = var_4277_end_0, end_mask = var_4277_end_mask_0, x = transpose_29)[name = tensor("op_4277_cast_fp16")]; + tensor var_4281_begin_0 = const()[name = tensor("op_4281_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_4281_end_0 = const()[name = tensor("op_4281_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_4281_end_mask_0 = const()[name = tensor("op_4281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4281_cast_fp16 = slice_by_index(begin = var_4281_begin_0, end = var_4281_end_0, end_mask = var_4281_end_mask_0, x = transpose_29)[name = tensor("op_4281_cast_fp16")]; + tensor var_4285_begin_0 = const()[name = tensor("op_4285_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_4285_end_0 = const()[name = tensor("op_4285_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_4285_end_mask_0 = const()[name = tensor("op_4285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4285_cast_fp16 = slice_by_index(begin = var_4285_begin_0, end = var_4285_end_0, end_mask = var_4285_end_mask_0, x = transpose_29)[name = tensor("op_4285_cast_fp16")]; + tensor var_4289_begin_0 = const()[name = tensor("op_4289_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_4289_end_0 = const()[name = tensor("op_4289_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_4289_end_mask_0 = const()[name = tensor("op_4289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4289_cast_fp16 = slice_by_index(begin = var_4289_begin_0, end = var_4289_end_0, end_mask = var_4289_end_mask_0, x = transpose_29)[name = tensor("op_4289_cast_fp16")]; + tensor var_4293_begin_0 = const()[name = tensor("op_4293_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_4293_end_0 = const()[name = tensor("op_4293_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_4293_end_mask_0 = const()[name = tensor("op_4293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4293_cast_fp16 = slice_by_index(begin = var_4293_begin_0, end = var_4293_end_0, end_mask = var_4293_end_mask_0, x = transpose_29)[name = tensor("op_4293_cast_fp16")]; + tensor var_4297_begin_0 = const()[name = tensor("op_4297_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_4297_end_0 = const()[name = tensor("op_4297_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_4297_end_mask_0 = const()[name = tensor("op_4297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4297_cast_fp16 = slice_by_index(begin = var_4297_begin_0, end = var_4297_end_0, end_mask = var_4297_end_mask_0, x = transpose_29)[name = tensor("op_4297_cast_fp16")]; + tensor var_4299_begin_0 = const()[name = tensor("op_4299_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4299_end_0 = const()[name = tensor("op_4299_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_4299_end_mask_0 = const()[name = tensor("op_4299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4299_cast_fp16 = slice_by_index(begin = var_4299_begin_0, end = var_4299_end_0, end_mask = var_4299_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4299_cast_fp16")]; + tensor var_4303_begin_0 = const()[name = tensor("op_4303_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_4303_end_0 = const()[name = tensor("op_4303_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_4303_end_mask_0 = const()[name = tensor("op_4303_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4303_cast_fp16 = slice_by_index(begin = var_4303_begin_0, end = var_4303_end_0, end_mask = var_4303_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4303_cast_fp16")]; + tensor var_4307_begin_0 = const()[name = tensor("op_4307_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_4307_end_0 = const()[name = tensor("op_4307_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_4307_end_mask_0 = const()[name = tensor("op_4307_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4307_cast_fp16 = slice_by_index(begin = var_4307_begin_0, end = var_4307_end_0, end_mask = var_4307_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4307_cast_fp16")]; + tensor var_4311_begin_0 = const()[name = tensor("op_4311_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_4311_end_0 = const()[name = tensor("op_4311_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_4311_end_mask_0 = const()[name = tensor("op_4311_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4311_cast_fp16 = slice_by_index(begin = var_4311_begin_0, end = var_4311_end_0, end_mask = var_4311_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4311_cast_fp16")]; + tensor var_4315_begin_0 = const()[name = tensor("op_4315_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_4315_end_0 = const()[name = tensor("op_4315_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_4315_end_mask_0 = const()[name = tensor("op_4315_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4315_cast_fp16 = slice_by_index(begin = var_4315_begin_0, end = var_4315_end_0, end_mask = var_4315_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4315_cast_fp16")]; + tensor var_4319_begin_0 = const()[name = tensor("op_4319_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_4319_end_0 = const()[name = tensor("op_4319_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_4319_end_mask_0 = const()[name = tensor("op_4319_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4319_cast_fp16 = slice_by_index(begin = var_4319_begin_0, end = var_4319_end_0, end_mask = var_4319_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4319_cast_fp16")]; + tensor var_4323_begin_0 = const()[name = tensor("op_4323_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_4323_end_0 = const()[name = tensor("op_4323_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_4323_end_mask_0 = const()[name = tensor("op_4323_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4323_cast_fp16 = slice_by_index(begin = var_4323_begin_0, end = var_4323_end_0, end_mask = var_4323_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4323_cast_fp16")]; + tensor var_4327_begin_0 = const()[name = tensor("op_4327_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_4327_end_0 = const()[name = tensor("op_4327_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_4327_end_mask_0 = const()[name = tensor("op_4327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4327_cast_fp16 = slice_by_index(begin = var_4327_begin_0, end = var_4327_end_0, end_mask = var_4327_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4327_cast_fp16")]; + tensor var_4331_begin_0 = const()[name = tensor("op_4331_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_4331_end_0 = const()[name = tensor("op_4331_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_4331_end_mask_0 = const()[name = tensor("op_4331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4331_cast_fp16 = slice_by_index(begin = var_4331_begin_0, end = var_4331_end_0, end_mask = var_4331_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4331_cast_fp16")]; + tensor var_4335_begin_0 = const()[name = tensor("op_4335_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_4335_end_0 = const()[name = tensor("op_4335_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_4335_end_mask_0 = const()[name = tensor("op_4335_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4335_cast_fp16 = slice_by_index(begin = var_4335_begin_0, end = var_4335_end_0, end_mask = var_4335_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4335_cast_fp16")]; + tensor var_4339_begin_0 = const()[name = tensor("op_4339_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_4339_end_0 = const()[name = tensor("op_4339_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_4339_end_mask_0 = const()[name = tensor("op_4339_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4339_cast_fp16 = slice_by_index(begin = var_4339_begin_0, end = var_4339_end_0, end_mask = var_4339_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4339_cast_fp16")]; + tensor var_4343_begin_0 = const()[name = tensor("op_4343_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_4343_end_0 = const()[name = tensor("op_4343_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_4343_end_mask_0 = const()[name = tensor("op_4343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4343_cast_fp16")]; + tensor var_4347_begin_0 = const()[name = tensor("op_4347_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_4347_end_0 = const()[name = tensor("op_4347_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_4347_end_mask_0 = const()[name = tensor("op_4347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4347_cast_fp16 = slice_by_index(begin = var_4347_begin_0, end = var_4347_end_0, end_mask = var_4347_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4347_cast_fp16")]; + tensor var_4351_begin_0 = const()[name = tensor("op_4351_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_4351_end_0 = const()[name = tensor("op_4351_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_4351_end_mask_0 = const()[name = tensor("op_4351_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4351_cast_fp16 = slice_by_index(begin = var_4351_begin_0, end = var_4351_end_0, end_mask = var_4351_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4351_cast_fp16")]; + tensor var_4355_begin_0 = const()[name = tensor("op_4355_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_4355_end_0 = const()[name = tensor("op_4355_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_4355_end_mask_0 = const()[name = tensor("op_4355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4355_cast_fp16 = slice_by_index(begin = var_4355_begin_0, end = var_4355_end_0, end_mask = var_4355_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4355_cast_fp16")]; + tensor var_4359_begin_0 = const()[name = tensor("op_4359_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_4359_end_0 = const()[name = tensor("op_4359_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_4359_end_mask_0 = const()[name = tensor("op_4359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4359_cast_fp16 = slice_by_index(begin = var_4359_begin_0, end = var_4359_end_0, end_mask = var_4359_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4359_cast_fp16")]; + tensor var_4363_begin_0 = const()[name = tensor("op_4363_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_4363_end_0 = const()[name = tensor("op_4363_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_4363_end_mask_0 = const()[name = tensor("op_4363_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4363_cast_fp16 = slice_by_index(begin = var_4363_begin_0, end = var_4363_end_0, end_mask = var_4363_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4363_cast_fp16")]; + tensor var_4367_begin_0 = const()[name = tensor("op_4367_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_4367_end_0 = const()[name = tensor("op_4367_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_4367_end_mask_0 = const()[name = tensor("op_4367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4367_cast_fp16 = slice_by_index(begin = var_4367_begin_0, end = var_4367_end_0, end_mask = var_4367_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4367_cast_fp16")]; + tensor var_4371_begin_0 = const()[name = tensor("op_4371_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_4371_end_0 = const()[name = tensor("op_4371_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_4371_end_mask_0 = const()[name = tensor("op_4371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4371_cast_fp16 = slice_by_index(begin = var_4371_begin_0, end = var_4371_end_0, end_mask = var_4371_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4371_cast_fp16")]; + tensor var_4375_begin_0 = const()[name = tensor("op_4375_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_4375_end_0 = const()[name = tensor("op_4375_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_4375_end_mask_0 = const()[name = tensor("op_4375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4375_cast_fp16 = slice_by_index(begin = var_4375_begin_0, end = var_4375_end_0, end_mask = var_4375_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_4375_cast_fp16")]; + tensor var_4379_equation_0 = const()[name = tensor("op_4379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4379_cast_fp16 = einsum(equation = var_4379_equation_0, values = (var_4221_cast_fp16, var_3663_cast_fp16))[name = tensor("op_4379_cast_fp16")]; + tensor var_4380_to_fp16 = const()[name = tensor("op_4380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = var_4379_cast_fp16, y = var_4380_to_fp16)[name = tensor("aw_chunk_321_cast_fp16")]; + tensor var_4383_equation_0 = const()[name = tensor("op_4383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4383_cast_fp16 = einsum(equation = var_4383_equation_0, values = (var_4221_cast_fp16, var_3670_cast_fp16))[name = tensor("op_4383_cast_fp16")]; + tensor var_4384_to_fp16 = const()[name = tensor("op_4384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = var_4383_cast_fp16, y = var_4384_to_fp16)[name = tensor("aw_chunk_323_cast_fp16")]; + tensor var_4387_equation_0 = const()[name = tensor("op_4387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4387_cast_fp16 = einsum(equation = var_4387_equation_0, values = (var_4221_cast_fp16, var_3677_cast_fp16))[name = tensor("op_4387_cast_fp16")]; + tensor var_4388_to_fp16 = const()[name = tensor("op_4388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = var_4387_cast_fp16, y = var_4388_to_fp16)[name = tensor("aw_chunk_325_cast_fp16")]; + tensor var_4391_equation_0 = const()[name = tensor("op_4391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4391_cast_fp16 = einsum(equation = var_4391_equation_0, values = (var_4221_cast_fp16, var_3684_cast_fp16))[name = tensor("op_4391_cast_fp16")]; + tensor var_4392_to_fp16 = const()[name = tensor("op_4392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = var_4391_cast_fp16, y = var_4392_to_fp16)[name = tensor("aw_chunk_327_cast_fp16")]; + tensor var_4395_equation_0 = const()[name = tensor("op_4395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4395_cast_fp16 = einsum(equation = var_4395_equation_0, values = (var_4225_cast_fp16, var_3691_cast_fp16))[name = tensor("op_4395_cast_fp16")]; + tensor var_4396_to_fp16 = const()[name = tensor("op_4396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = var_4395_cast_fp16, y = var_4396_to_fp16)[name = tensor("aw_chunk_329_cast_fp16")]; + tensor var_4399_equation_0 = const()[name = tensor("op_4399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4399_cast_fp16 = einsum(equation = var_4399_equation_0, values = (var_4225_cast_fp16, var_3698_cast_fp16))[name = tensor("op_4399_cast_fp16")]; + tensor var_4400_to_fp16 = const()[name = tensor("op_4400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = var_4399_cast_fp16, y = var_4400_to_fp16)[name = tensor("aw_chunk_331_cast_fp16")]; + tensor var_4403_equation_0 = const()[name = tensor("op_4403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4403_cast_fp16 = einsum(equation = var_4403_equation_0, values = (var_4225_cast_fp16, var_3705_cast_fp16))[name = tensor("op_4403_cast_fp16")]; + tensor var_4404_to_fp16 = const()[name = tensor("op_4404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = var_4403_cast_fp16, y = var_4404_to_fp16)[name = tensor("aw_chunk_333_cast_fp16")]; + tensor var_4407_equation_0 = const()[name = tensor("op_4407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4407_cast_fp16 = einsum(equation = var_4407_equation_0, values = (var_4225_cast_fp16, var_3712_cast_fp16))[name = tensor("op_4407_cast_fp16")]; + tensor var_4408_to_fp16 = const()[name = tensor("op_4408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = var_4407_cast_fp16, y = var_4408_to_fp16)[name = tensor("aw_chunk_335_cast_fp16")]; + tensor var_4411_equation_0 = const()[name = tensor("op_4411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4411_cast_fp16 = einsum(equation = var_4411_equation_0, values = (var_4229_cast_fp16, var_3719_cast_fp16))[name = tensor("op_4411_cast_fp16")]; + tensor var_4412_to_fp16 = const()[name = tensor("op_4412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = var_4411_cast_fp16, y = var_4412_to_fp16)[name = tensor("aw_chunk_337_cast_fp16")]; + tensor var_4415_equation_0 = const()[name = tensor("op_4415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4415_cast_fp16 = einsum(equation = var_4415_equation_0, values = (var_4229_cast_fp16, var_3726_cast_fp16))[name = tensor("op_4415_cast_fp16")]; + tensor var_4416_to_fp16 = const()[name = tensor("op_4416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = var_4415_cast_fp16, y = var_4416_to_fp16)[name = tensor("aw_chunk_339_cast_fp16")]; + tensor var_4419_equation_0 = const()[name = tensor("op_4419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4419_cast_fp16 = einsum(equation = var_4419_equation_0, values = (var_4229_cast_fp16, var_3733_cast_fp16))[name = tensor("op_4419_cast_fp16")]; + tensor var_4420_to_fp16 = const()[name = tensor("op_4420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = var_4419_cast_fp16, y = var_4420_to_fp16)[name = tensor("aw_chunk_341_cast_fp16")]; + tensor var_4423_equation_0 = const()[name = tensor("op_4423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4423_cast_fp16 = einsum(equation = var_4423_equation_0, values = (var_4229_cast_fp16, var_3740_cast_fp16))[name = tensor("op_4423_cast_fp16")]; + tensor var_4424_to_fp16 = const()[name = tensor("op_4424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = var_4423_cast_fp16, y = var_4424_to_fp16)[name = tensor("aw_chunk_343_cast_fp16")]; + tensor var_4427_equation_0 = const()[name = tensor("op_4427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4427_cast_fp16 = einsum(equation = var_4427_equation_0, values = (var_4233_cast_fp16, var_3747_cast_fp16))[name = tensor("op_4427_cast_fp16")]; + tensor var_4428_to_fp16 = const()[name = tensor("op_4428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = var_4427_cast_fp16, y = var_4428_to_fp16)[name = tensor("aw_chunk_345_cast_fp16")]; + tensor var_4431_equation_0 = const()[name = tensor("op_4431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4431_cast_fp16 = einsum(equation = var_4431_equation_0, values = (var_4233_cast_fp16, var_3754_cast_fp16))[name = tensor("op_4431_cast_fp16")]; + tensor var_4432_to_fp16 = const()[name = tensor("op_4432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = var_4431_cast_fp16, y = var_4432_to_fp16)[name = tensor("aw_chunk_347_cast_fp16")]; + tensor var_4435_equation_0 = const()[name = tensor("op_4435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4435_cast_fp16 = einsum(equation = var_4435_equation_0, values = (var_4233_cast_fp16, var_3761_cast_fp16))[name = tensor("op_4435_cast_fp16")]; + tensor var_4436_to_fp16 = const()[name = tensor("op_4436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = var_4435_cast_fp16, y = var_4436_to_fp16)[name = tensor("aw_chunk_349_cast_fp16")]; + tensor var_4439_equation_0 = const()[name = tensor("op_4439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4439_cast_fp16 = einsum(equation = var_4439_equation_0, values = (var_4233_cast_fp16, var_3768_cast_fp16))[name = tensor("op_4439_cast_fp16")]; + tensor var_4440_to_fp16 = const()[name = tensor("op_4440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = var_4439_cast_fp16, y = var_4440_to_fp16)[name = tensor("aw_chunk_351_cast_fp16")]; + tensor var_4443_equation_0 = const()[name = tensor("op_4443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4443_cast_fp16 = einsum(equation = var_4443_equation_0, values = (var_4237_cast_fp16, var_3775_cast_fp16))[name = tensor("op_4443_cast_fp16")]; + tensor var_4444_to_fp16 = const()[name = tensor("op_4444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = var_4443_cast_fp16, y = var_4444_to_fp16)[name = tensor("aw_chunk_353_cast_fp16")]; + tensor var_4447_equation_0 = const()[name = tensor("op_4447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4447_cast_fp16 = einsum(equation = var_4447_equation_0, values = (var_4237_cast_fp16, var_3782_cast_fp16))[name = tensor("op_4447_cast_fp16")]; + tensor var_4448_to_fp16 = const()[name = tensor("op_4448_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = var_4447_cast_fp16, y = var_4448_to_fp16)[name = tensor("aw_chunk_355_cast_fp16")]; + tensor var_4451_equation_0 = const()[name = tensor("op_4451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4451_cast_fp16 = einsum(equation = var_4451_equation_0, values = (var_4237_cast_fp16, var_3789_cast_fp16))[name = tensor("op_4451_cast_fp16")]; + tensor var_4452_to_fp16 = const()[name = tensor("op_4452_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = var_4451_cast_fp16, y = var_4452_to_fp16)[name = tensor("aw_chunk_357_cast_fp16")]; + tensor var_4455_equation_0 = const()[name = tensor("op_4455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4455_cast_fp16 = einsum(equation = var_4455_equation_0, values = (var_4237_cast_fp16, var_3796_cast_fp16))[name = tensor("op_4455_cast_fp16")]; + tensor var_4456_to_fp16 = const()[name = tensor("op_4456_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = var_4455_cast_fp16, y = var_4456_to_fp16)[name = tensor("aw_chunk_359_cast_fp16")]; + tensor var_4459_equation_0 = const()[name = tensor("op_4459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4459_cast_fp16 = einsum(equation = var_4459_equation_0, values = (var_4241_cast_fp16, var_3803_cast_fp16))[name = tensor("op_4459_cast_fp16")]; + tensor var_4460_to_fp16 = const()[name = tensor("op_4460_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = var_4459_cast_fp16, y = var_4460_to_fp16)[name = tensor("aw_chunk_361_cast_fp16")]; + tensor var_4463_equation_0 = const()[name = tensor("op_4463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4463_cast_fp16 = einsum(equation = var_4463_equation_0, values = (var_4241_cast_fp16, var_3810_cast_fp16))[name = tensor("op_4463_cast_fp16")]; + tensor var_4464_to_fp16 = const()[name = tensor("op_4464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = var_4463_cast_fp16, y = var_4464_to_fp16)[name = tensor("aw_chunk_363_cast_fp16")]; + tensor var_4467_equation_0 = const()[name = tensor("op_4467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4467_cast_fp16 = einsum(equation = var_4467_equation_0, values = (var_4241_cast_fp16, var_3817_cast_fp16))[name = tensor("op_4467_cast_fp16")]; + tensor var_4468_to_fp16 = const()[name = tensor("op_4468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = var_4467_cast_fp16, y = var_4468_to_fp16)[name = tensor("aw_chunk_365_cast_fp16")]; + tensor var_4471_equation_0 = const()[name = tensor("op_4471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4471_cast_fp16 = einsum(equation = var_4471_equation_0, values = (var_4241_cast_fp16, var_3824_cast_fp16))[name = tensor("op_4471_cast_fp16")]; + tensor var_4472_to_fp16 = const()[name = tensor("op_4472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = var_4471_cast_fp16, y = var_4472_to_fp16)[name = tensor("aw_chunk_367_cast_fp16")]; + tensor var_4475_equation_0 = const()[name = tensor("op_4475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4475_cast_fp16 = einsum(equation = var_4475_equation_0, values = (var_4245_cast_fp16, var_3831_cast_fp16))[name = tensor("op_4475_cast_fp16")]; + tensor var_4476_to_fp16 = const()[name = tensor("op_4476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = var_4475_cast_fp16, y = var_4476_to_fp16)[name = tensor("aw_chunk_369_cast_fp16")]; + tensor var_4479_equation_0 = const()[name = tensor("op_4479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4479_cast_fp16 = einsum(equation = var_4479_equation_0, values = (var_4245_cast_fp16, var_3838_cast_fp16))[name = tensor("op_4479_cast_fp16")]; + tensor var_4480_to_fp16 = const()[name = tensor("op_4480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = var_4479_cast_fp16, y = var_4480_to_fp16)[name = tensor("aw_chunk_371_cast_fp16")]; + tensor var_4483_equation_0 = const()[name = tensor("op_4483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4483_cast_fp16 = einsum(equation = var_4483_equation_0, values = (var_4245_cast_fp16, var_3845_cast_fp16))[name = tensor("op_4483_cast_fp16")]; + tensor var_4484_to_fp16 = const()[name = tensor("op_4484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = var_4483_cast_fp16, y = var_4484_to_fp16)[name = tensor("aw_chunk_373_cast_fp16")]; + tensor var_4487_equation_0 = const()[name = tensor("op_4487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4487_cast_fp16 = einsum(equation = var_4487_equation_0, values = (var_4245_cast_fp16, var_3852_cast_fp16))[name = tensor("op_4487_cast_fp16")]; + tensor var_4488_to_fp16 = const()[name = tensor("op_4488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = var_4487_cast_fp16, y = var_4488_to_fp16)[name = tensor("aw_chunk_375_cast_fp16")]; + tensor var_4491_equation_0 = const()[name = tensor("op_4491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4491_cast_fp16 = einsum(equation = var_4491_equation_0, values = (var_4249_cast_fp16, var_3859_cast_fp16))[name = tensor("op_4491_cast_fp16")]; + tensor var_4492_to_fp16 = const()[name = tensor("op_4492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = var_4491_cast_fp16, y = var_4492_to_fp16)[name = tensor("aw_chunk_377_cast_fp16")]; + tensor var_4495_equation_0 = const()[name = tensor("op_4495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4495_cast_fp16 = einsum(equation = var_4495_equation_0, values = (var_4249_cast_fp16, var_3866_cast_fp16))[name = tensor("op_4495_cast_fp16")]; + tensor var_4496_to_fp16 = const()[name = tensor("op_4496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = var_4495_cast_fp16, y = var_4496_to_fp16)[name = tensor("aw_chunk_379_cast_fp16")]; + tensor var_4499_equation_0 = const()[name = tensor("op_4499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4499_cast_fp16 = einsum(equation = var_4499_equation_0, values = (var_4249_cast_fp16, var_3873_cast_fp16))[name = tensor("op_4499_cast_fp16")]; + tensor var_4500_to_fp16 = const()[name = tensor("op_4500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = var_4499_cast_fp16, y = var_4500_to_fp16)[name = tensor("aw_chunk_381_cast_fp16")]; + tensor var_4503_equation_0 = const()[name = tensor("op_4503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4503_cast_fp16 = einsum(equation = var_4503_equation_0, values = (var_4249_cast_fp16, var_3880_cast_fp16))[name = tensor("op_4503_cast_fp16")]; + tensor var_4504_to_fp16 = const()[name = tensor("op_4504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_383_cast_fp16 = mul(x = var_4503_cast_fp16, y = var_4504_to_fp16)[name = tensor("aw_chunk_383_cast_fp16")]; + tensor var_4507_equation_0 = const()[name = tensor("op_4507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4507_cast_fp16 = einsum(equation = var_4507_equation_0, values = (var_4253_cast_fp16, var_3887_cast_fp16))[name = tensor("op_4507_cast_fp16")]; + tensor var_4508_to_fp16 = const()[name = tensor("op_4508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_385_cast_fp16 = mul(x = var_4507_cast_fp16, y = var_4508_to_fp16)[name = tensor("aw_chunk_385_cast_fp16")]; + tensor var_4511_equation_0 = const()[name = tensor("op_4511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4511_cast_fp16 = einsum(equation = var_4511_equation_0, values = (var_4253_cast_fp16, var_3894_cast_fp16))[name = tensor("op_4511_cast_fp16")]; + tensor var_4512_to_fp16 = const()[name = tensor("op_4512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_387_cast_fp16 = mul(x = var_4511_cast_fp16, y = var_4512_to_fp16)[name = tensor("aw_chunk_387_cast_fp16")]; + tensor var_4515_equation_0 = const()[name = tensor("op_4515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4515_cast_fp16 = einsum(equation = var_4515_equation_0, values = (var_4253_cast_fp16, var_3901_cast_fp16))[name = tensor("op_4515_cast_fp16")]; + tensor var_4516_to_fp16 = const()[name = tensor("op_4516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_389_cast_fp16 = mul(x = var_4515_cast_fp16, y = var_4516_to_fp16)[name = tensor("aw_chunk_389_cast_fp16")]; + tensor var_4519_equation_0 = const()[name = tensor("op_4519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4519_cast_fp16 = einsum(equation = var_4519_equation_0, values = (var_4253_cast_fp16, var_3908_cast_fp16))[name = tensor("op_4519_cast_fp16")]; + tensor var_4520_to_fp16 = const()[name = tensor("op_4520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_391_cast_fp16 = mul(x = var_4519_cast_fp16, y = var_4520_to_fp16)[name = tensor("aw_chunk_391_cast_fp16")]; + tensor var_4523_equation_0 = const()[name = tensor("op_4523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4523_cast_fp16 = einsum(equation = var_4523_equation_0, values = (var_4257_cast_fp16, var_3915_cast_fp16))[name = tensor("op_4523_cast_fp16")]; + tensor var_4524_to_fp16 = const()[name = tensor("op_4524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_393_cast_fp16 = mul(x = var_4523_cast_fp16, y = var_4524_to_fp16)[name = tensor("aw_chunk_393_cast_fp16")]; + tensor var_4527_equation_0 = const()[name = tensor("op_4527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4527_cast_fp16 = einsum(equation = var_4527_equation_0, values = (var_4257_cast_fp16, var_3922_cast_fp16))[name = tensor("op_4527_cast_fp16")]; + tensor var_4528_to_fp16 = const()[name = tensor("op_4528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_395_cast_fp16 = mul(x = var_4527_cast_fp16, y = var_4528_to_fp16)[name = tensor("aw_chunk_395_cast_fp16")]; + tensor var_4531_equation_0 = const()[name = tensor("op_4531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4531_cast_fp16 = einsum(equation = var_4531_equation_0, values = (var_4257_cast_fp16, var_3929_cast_fp16))[name = tensor("op_4531_cast_fp16")]; + tensor var_4532_to_fp16 = const()[name = tensor("op_4532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_397_cast_fp16 = mul(x = var_4531_cast_fp16, y = var_4532_to_fp16)[name = tensor("aw_chunk_397_cast_fp16")]; + tensor var_4535_equation_0 = const()[name = tensor("op_4535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4535_cast_fp16 = einsum(equation = var_4535_equation_0, values = (var_4257_cast_fp16, var_3936_cast_fp16))[name = tensor("op_4535_cast_fp16")]; + tensor var_4536_to_fp16 = const()[name = tensor("op_4536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_399_cast_fp16 = mul(x = var_4535_cast_fp16, y = var_4536_to_fp16)[name = tensor("aw_chunk_399_cast_fp16")]; + tensor var_4539_equation_0 = const()[name = tensor("op_4539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4539_cast_fp16 = einsum(equation = var_4539_equation_0, values = (var_4261_cast_fp16, var_3943_cast_fp16))[name = tensor("op_4539_cast_fp16")]; + tensor var_4540_to_fp16 = const()[name = tensor("op_4540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_401_cast_fp16 = mul(x = var_4539_cast_fp16, y = var_4540_to_fp16)[name = tensor("aw_chunk_401_cast_fp16")]; + tensor var_4543_equation_0 = const()[name = tensor("op_4543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4543_cast_fp16 = einsum(equation = var_4543_equation_0, values = (var_4261_cast_fp16, var_3950_cast_fp16))[name = tensor("op_4543_cast_fp16")]; + tensor var_4544_to_fp16 = const()[name = tensor("op_4544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_403_cast_fp16 = mul(x = var_4543_cast_fp16, y = var_4544_to_fp16)[name = tensor("aw_chunk_403_cast_fp16")]; + tensor var_4547_equation_0 = const()[name = tensor("op_4547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4547_cast_fp16 = einsum(equation = var_4547_equation_0, values = (var_4261_cast_fp16, var_3957_cast_fp16))[name = tensor("op_4547_cast_fp16")]; + tensor var_4548_to_fp16 = const()[name = tensor("op_4548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_405_cast_fp16 = mul(x = var_4547_cast_fp16, y = var_4548_to_fp16)[name = tensor("aw_chunk_405_cast_fp16")]; + tensor var_4551_equation_0 = const()[name = tensor("op_4551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4551_cast_fp16 = einsum(equation = var_4551_equation_0, values = (var_4261_cast_fp16, var_3964_cast_fp16))[name = tensor("op_4551_cast_fp16")]; + tensor var_4552_to_fp16 = const()[name = tensor("op_4552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_407_cast_fp16 = mul(x = var_4551_cast_fp16, y = var_4552_to_fp16)[name = tensor("aw_chunk_407_cast_fp16")]; + tensor var_4555_equation_0 = const()[name = tensor("op_4555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4555_cast_fp16 = einsum(equation = var_4555_equation_0, values = (var_4265_cast_fp16, var_3971_cast_fp16))[name = tensor("op_4555_cast_fp16")]; + tensor var_4556_to_fp16 = const()[name = tensor("op_4556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_409_cast_fp16 = mul(x = var_4555_cast_fp16, y = var_4556_to_fp16)[name = tensor("aw_chunk_409_cast_fp16")]; + tensor var_4559_equation_0 = const()[name = tensor("op_4559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4559_cast_fp16 = einsum(equation = var_4559_equation_0, values = (var_4265_cast_fp16, var_3978_cast_fp16))[name = tensor("op_4559_cast_fp16")]; + tensor var_4560_to_fp16 = const()[name = tensor("op_4560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_411_cast_fp16 = mul(x = var_4559_cast_fp16, y = var_4560_to_fp16)[name = tensor("aw_chunk_411_cast_fp16")]; + tensor var_4563_equation_0 = const()[name = tensor("op_4563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4563_cast_fp16 = einsum(equation = var_4563_equation_0, values = (var_4265_cast_fp16, var_3985_cast_fp16))[name = tensor("op_4563_cast_fp16")]; + tensor var_4564_to_fp16 = const()[name = tensor("op_4564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_413_cast_fp16 = mul(x = var_4563_cast_fp16, y = var_4564_to_fp16)[name = tensor("aw_chunk_413_cast_fp16")]; + tensor var_4567_equation_0 = const()[name = tensor("op_4567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4567_cast_fp16 = einsum(equation = var_4567_equation_0, values = (var_4265_cast_fp16, var_3992_cast_fp16))[name = tensor("op_4567_cast_fp16")]; + tensor var_4568_to_fp16 = const()[name = tensor("op_4568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_415_cast_fp16 = mul(x = var_4567_cast_fp16, y = var_4568_to_fp16)[name = tensor("aw_chunk_415_cast_fp16")]; + tensor var_4571_equation_0 = const()[name = tensor("op_4571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4571_cast_fp16 = einsum(equation = var_4571_equation_0, values = (var_4269_cast_fp16, var_3999_cast_fp16))[name = tensor("op_4571_cast_fp16")]; + tensor var_4572_to_fp16 = const()[name = tensor("op_4572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_417_cast_fp16 = mul(x = var_4571_cast_fp16, y = var_4572_to_fp16)[name = tensor("aw_chunk_417_cast_fp16")]; + tensor var_4575_equation_0 = const()[name = tensor("op_4575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4575_cast_fp16 = einsum(equation = var_4575_equation_0, values = (var_4269_cast_fp16, var_4006_cast_fp16))[name = tensor("op_4575_cast_fp16")]; + tensor var_4576_to_fp16 = const()[name = tensor("op_4576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_419_cast_fp16 = mul(x = var_4575_cast_fp16, y = var_4576_to_fp16)[name = tensor("aw_chunk_419_cast_fp16")]; + tensor var_4579_equation_0 = const()[name = tensor("op_4579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4579_cast_fp16 = einsum(equation = var_4579_equation_0, values = (var_4269_cast_fp16, var_4013_cast_fp16))[name = tensor("op_4579_cast_fp16")]; + tensor var_4580_to_fp16 = const()[name = tensor("op_4580_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_421_cast_fp16 = mul(x = var_4579_cast_fp16, y = var_4580_to_fp16)[name = tensor("aw_chunk_421_cast_fp16")]; + tensor var_4583_equation_0 = const()[name = tensor("op_4583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4583_cast_fp16 = einsum(equation = var_4583_equation_0, values = (var_4269_cast_fp16, var_4020_cast_fp16))[name = tensor("op_4583_cast_fp16")]; + tensor var_4584_to_fp16 = const()[name = tensor("op_4584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_423_cast_fp16 = mul(x = var_4583_cast_fp16, y = var_4584_to_fp16)[name = tensor("aw_chunk_423_cast_fp16")]; + tensor var_4587_equation_0 = const()[name = tensor("op_4587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4587_cast_fp16 = einsum(equation = var_4587_equation_0, values = (var_4273_cast_fp16, var_4027_cast_fp16))[name = tensor("op_4587_cast_fp16")]; + tensor var_4588_to_fp16 = const()[name = tensor("op_4588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_425_cast_fp16 = mul(x = var_4587_cast_fp16, y = var_4588_to_fp16)[name = tensor("aw_chunk_425_cast_fp16")]; + tensor var_4591_equation_0 = const()[name = tensor("op_4591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4591_cast_fp16 = einsum(equation = var_4591_equation_0, values = (var_4273_cast_fp16, var_4034_cast_fp16))[name = tensor("op_4591_cast_fp16")]; + tensor var_4592_to_fp16 = const()[name = tensor("op_4592_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_427_cast_fp16 = mul(x = var_4591_cast_fp16, y = var_4592_to_fp16)[name = tensor("aw_chunk_427_cast_fp16")]; + tensor var_4595_equation_0 = const()[name = tensor("op_4595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4595_cast_fp16 = einsum(equation = var_4595_equation_0, values = (var_4273_cast_fp16, var_4041_cast_fp16))[name = tensor("op_4595_cast_fp16")]; + tensor var_4596_to_fp16 = const()[name = tensor("op_4596_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_429_cast_fp16 = mul(x = var_4595_cast_fp16, y = var_4596_to_fp16)[name = tensor("aw_chunk_429_cast_fp16")]; + tensor var_4599_equation_0 = const()[name = tensor("op_4599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4599_cast_fp16 = einsum(equation = var_4599_equation_0, values = (var_4273_cast_fp16, var_4048_cast_fp16))[name = tensor("op_4599_cast_fp16")]; + tensor var_4600_to_fp16 = const()[name = tensor("op_4600_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_431_cast_fp16 = mul(x = var_4599_cast_fp16, y = var_4600_to_fp16)[name = tensor("aw_chunk_431_cast_fp16")]; + tensor var_4603_equation_0 = const()[name = tensor("op_4603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4603_cast_fp16 = einsum(equation = var_4603_equation_0, values = (var_4277_cast_fp16, var_4055_cast_fp16))[name = tensor("op_4603_cast_fp16")]; + tensor var_4604_to_fp16 = const()[name = tensor("op_4604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_433_cast_fp16 = mul(x = var_4603_cast_fp16, y = var_4604_to_fp16)[name = tensor("aw_chunk_433_cast_fp16")]; + tensor var_4607_equation_0 = const()[name = tensor("op_4607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4607_cast_fp16 = einsum(equation = var_4607_equation_0, values = (var_4277_cast_fp16, var_4062_cast_fp16))[name = tensor("op_4607_cast_fp16")]; + tensor var_4608_to_fp16 = const()[name = tensor("op_4608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_435_cast_fp16 = mul(x = var_4607_cast_fp16, y = var_4608_to_fp16)[name = tensor("aw_chunk_435_cast_fp16")]; + tensor var_4611_equation_0 = const()[name = tensor("op_4611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4611_cast_fp16 = einsum(equation = var_4611_equation_0, values = (var_4277_cast_fp16, var_4069_cast_fp16))[name = tensor("op_4611_cast_fp16")]; + tensor var_4612_to_fp16 = const()[name = tensor("op_4612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_437_cast_fp16 = mul(x = var_4611_cast_fp16, y = var_4612_to_fp16)[name = tensor("aw_chunk_437_cast_fp16")]; + tensor var_4615_equation_0 = const()[name = tensor("op_4615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4615_cast_fp16 = einsum(equation = var_4615_equation_0, values = (var_4277_cast_fp16, var_4076_cast_fp16))[name = tensor("op_4615_cast_fp16")]; + tensor var_4616_to_fp16 = const()[name = tensor("op_4616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_439_cast_fp16 = mul(x = var_4615_cast_fp16, y = var_4616_to_fp16)[name = tensor("aw_chunk_439_cast_fp16")]; + tensor var_4619_equation_0 = const()[name = tensor("op_4619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4619_cast_fp16 = einsum(equation = var_4619_equation_0, values = (var_4281_cast_fp16, var_4083_cast_fp16))[name = tensor("op_4619_cast_fp16")]; + tensor var_4620_to_fp16 = const()[name = tensor("op_4620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_441_cast_fp16 = mul(x = var_4619_cast_fp16, y = var_4620_to_fp16)[name = tensor("aw_chunk_441_cast_fp16")]; + tensor var_4623_equation_0 = const()[name = tensor("op_4623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4623_cast_fp16 = einsum(equation = var_4623_equation_0, values = (var_4281_cast_fp16, var_4090_cast_fp16))[name = tensor("op_4623_cast_fp16")]; + tensor var_4624_to_fp16 = const()[name = tensor("op_4624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_443_cast_fp16 = mul(x = var_4623_cast_fp16, y = var_4624_to_fp16)[name = tensor("aw_chunk_443_cast_fp16")]; + tensor var_4627_equation_0 = const()[name = tensor("op_4627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4627_cast_fp16 = einsum(equation = var_4627_equation_0, values = (var_4281_cast_fp16, var_4097_cast_fp16))[name = tensor("op_4627_cast_fp16")]; + tensor var_4628_to_fp16 = const()[name = tensor("op_4628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_445_cast_fp16 = mul(x = var_4627_cast_fp16, y = var_4628_to_fp16)[name = tensor("aw_chunk_445_cast_fp16")]; + tensor var_4631_equation_0 = const()[name = tensor("op_4631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4631_cast_fp16 = einsum(equation = var_4631_equation_0, values = (var_4281_cast_fp16, var_4104_cast_fp16))[name = tensor("op_4631_cast_fp16")]; + tensor var_4632_to_fp16 = const()[name = tensor("op_4632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_447_cast_fp16 = mul(x = var_4631_cast_fp16, y = var_4632_to_fp16)[name = tensor("aw_chunk_447_cast_fp16")]; + tensor var_4635_equation_0 = const()[name = tensor("op_4635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4635_cast_fp16 = einsum(equation = var_4635_equation_0, values = (var_4285_cast_fp16, var_4111_cast_fp16))[name = tensor("op_4635_cast_fp16")]; + tensor var_4636_to_fp16 = const()[name = tensor("op_4636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_449_cast_fp16 = mul(x = var_4635_cast_fp16, y = var_4636_to_fp16)[name = tensor("aw_chunk_449_cast_fp16")]; + tensor var_4639_equation_0 = const()[name = tensor("op_4639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4639_cast_fp16 = einsum(equation = var_4639_equation_0, values = (var_4285_cast_fp16, var_4118_cast_fp16))[name = tensor("op_4639_cast_fp16")]; + tensor var_4640_to_fp16 = const()[name = tensor("op_4640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_451_cast_fp16 = mul(x = var_4639_cast_fp16, y = var_4640_to_fp16)[name = tensor("aw_chunk_451_cast_fp16")]; + tensor var_4643_equation_0 = const()[name = tensor("op_4643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4643_cast_fp16 = einsum(equation = var_4643_equation_0, values = (var_4285_cast_fp16, var_4125_cast_fp16))[name = tensor("op_4643_cast_fp16")]; + tensor var_4644_to_fp16 = const()[name = tensor("op_4644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_453_cast_fp16 = mul(x = var_4643_cast_fp16, y = var_4644_to_fp16)[name = tensor("aw_chunk_453_cast_fp16")]; + tensor var_4647_equation_0 = const()[name = tensor("op_4647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4647_cast_fp16 = einsum(equation = var_4647_equation_0, values = (var_4285_cast_fp16, var_4132_cast_fp16))[name = tensor("op_4647_cast_fp16")]; + tensor var_4648_to_fp16 = const()[name = tensor("op_4648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_455_cast_fp16 = mul(x = var_4647_cast_fp16, y = var_4648_to_fp16)[name = tensor("aw_chunk_455_cast_fp16")]; + tensor var_4651_equation_0 = const()[name = tensor("op_4651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4651_cast_fp16 = einsum(equation = var_4651_equation_0, values = (var_4289_cast_fp16, var_4139_cast_fp16))[name = tensor("op_4651_cast_fp16")]; + tensor var_4652_to_fp16 = const()[name = tensor("op_4652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_457_cast_fp16 = mul(x = var_4651_cast_fp16, y = var_4652_to_fp16)[name = tensor("aw_chunk_457_cast_fp16")]; + tensor var_4655_equation_0 = const()[name = tensor("op_4655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4655_cast_fp16 = einsum(equation = var_4655_equation_0, values = (var_4289_cast_fp16, var_4146_cast_fp16))[name = tensor("op_4655_cast_fp16")]; + tensor var_4656_to_fp16 = const()[name = tensor("op_4656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_459_cast_fp16 = mul(x = var_4655_cast_fp16, y = var_4656_to_fp16)[name = tensor("aw_chunk_459_cast_fp16")]; + tensor var_4659_equation_0 = const()[name = tensor("op_4659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4659_cast_fp16 = einsum(equation = var_4659_equation_0, values = (var_4289_cast_fp16, var_4153_cast_fp16))[name = tensor("op_4659_cast_fp16")]; + tensor var_4660_to_fp16 = const()[name = tensor("op_4660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_461_cast_fp16 = mul(x = var_4659_cast_fp16, y = var_4660_to_fp16)[name = tensor("aw_chunk_461_cast_fp16")]; + tensor var_4663_equation_0 = const()[name = tensor("op_4663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4663_cast_fp16 = einsum(equation = var_4663_equation_0, values = (var_4289_cast_fp16, var_4160_cast_fp16))[name = tensor("op_4663_cast_fp16")]; + tensor var_4664_to_fp16 = const()[name = tensor("op_4664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_463_cast_fp16 = mul(x = var_4663_cast_fp16, y = var_4664_to_fp16)[name = tensor("aw_chunk_463_cast_fp16")]; + tensor var_4667_equation_0 = const()[name = tensor("op_4667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4667_cast_fp16 = einsum(equation = var_4667_equation_0, values = (var_4293_cast_fp16, var_4167_cast_fp16))[name = tensor("op_4667_cast_fp16")]; + tensor var_4668_to_fp16 = const()[name = tensor("op_4668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_465_cast_fp16 = mul(x = var_4667_cast_fp16, y = var_4668_to_fp16)[name = tensor("aw_chunk_465_cast_fp16")]; + tensor var_4671_equation_0 = const()[name = tensor("op_4671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4671_cast_fp16 = einsum(equation = var_4671_equation_0, values = (var_4293_cast_fp16, var_4174_cast_fp16))[name = tensor("op_4671_cast_fp16")]; + tensor var_4672_to_fp16 = const()[name = tensor("op_4672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_467_cast_fp16 = mul(x = var_4671_cast_fp16, y = var_4672_to_fp16)[name = tensor("aw_chunk_467_cast_fp16")]; + tensor var_4675_equation_0 = const()[name = tensor("op_4675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4675_cast_fp16 = einsum(equation = var_4675_equation_0, values = (var_4293_cast_fp16, var_4181_cast_fp16))[name = tensor("op_4675_cast_fp16")]; + tensor var_4676_to_fp16 = const()[name = tensor("op_4676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_469_cast_fp16 = mul(x = var_4675_cast_fp16, y = var_4676_to_fp16)[name = tensor("aw_chunk_469_cast_fp16")]; + tensor var_4679_equation_0 = const()[name = tensor("op_4679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4679_cast_fp16 = einsum(equation = var_4679_equation_0, values = (var_4293_cast_fp16, var_4188_cast_fp16))[name = tensor("op_4679_cast_fp16")]; + tensor var_4680_to_fp16 = const()[name = tensor("op_4680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_471_cast_fp16 = mul(x = var_4679_cast_fp16, y = var_4680_to_fp16)[name = tensor("aw_chunk_471_cast_fp16")]; + tensor var_4683_equation_0 = const()[name = tensor("op_4683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4683_cast_fp16 = einsum(equation = var_4683_equation_0, values = (var_4297_cast_fp16, var_4195_cast_fp16))[name = tensor("op_4683_cast_fp16")]; + tensor var_4684_to_fp16 = const()[name = tensor("op_4684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_473_cast_fp16 = mul(x = var_4683_cast_fp16, y = var_4684_to_fp16)[name = tensor("aw_chunk_473_cast_fp16")]; + tensor var_4687_equation_0 = const()[name = tensor("op_4687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4687_cast_fp16 = einsum(equation = var_4687_equation_0, values = (var_4297_cast_fp16, var_4202_cast_fp16))[name = tensor("op_4687_cast_fp16")]; + tensor var_4688_to_fp16 = const()[name = tensor("op_4688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_475_cast_fp16 = mul(x = var_4687_cast_fp16, y = var_4688_to_fp16)[name = tensor("aw_chunk_475_cast_fp16")]; + tensor var_4691_equation_0 = const()[name = tensor("op_4691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4691_cast_fp16 = einsum(equation = var_4691_equation_0, values = (var_4297_cast_fp16, var_4209_cast_fp16))[name = tensor("op_4691_cast_fp16")]; + tensor var_4692_to_fp16 = const()[name = tensor("op_4692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_477_cast_fp16 = mul(x = var_4691_cast_fp16, y = var_4692_to_fp16)[name = tensor("aw_chunk_477_cast_fp16")]; + tensor var_4695_equation_0 = const()[name = tensor("op_4695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_4695_cast_fp16 = einsum(equation = var_4695_equation_0, values = (var_4297_cast_fp16, var_4216_cast_fp16))[name = tensor("op_4695_cast_fp16")]; + tensor var_4696_to_fp16 = const()[name = tensor("op_4696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_479_cast_fp16 = mul(x = var_4695_cast_fp16, y = var_4696_to_fp16)[name = tensor("aw_chunk_479_cast_fp16")]; + tensor var_4698_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_321_cast_fp16)[name = tensor("op_4698_cast_fp16")]; + tensor var_4699_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_323_cast_fp16)[name = tensor("op_4699_cast_fp16")]; + tensor var_4700_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_325_cast_fp16)[name = tensor("op_4700_cast_fp16")]; + tensor var_4701_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_327_cast_fp16)[name = tensor("op_4701_cast_fp16")]; + tensor var_4702_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_329_cast_fp16)[name = tensor("op_4702_cast_fp16")]; + tensor var_4703_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_331_cast_fp16)[name = tensor("op_4703_cast_fp16")]; + tensor var_4704_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_333_cast_fp16)[name = tensor("op_4704_cast_fp16")]; + tensor var_4705_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_335_cast_fp16)[name = tensor("op_4705_cast_fp16")]; + tensor var_4706_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_337_cast_fp16)[name = tensor("op_4706_cast_fp16")]; + tensor var_4707_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_339_cast_fp16)[name = tensor("op_4707_cast_fp16")]; + tensor var_4708_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_341_cast_fp16)[name = tensor("op_4708_cast_fp16")]; + tensor var_4709_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_343_cast_fp16)[name = tensor("op_4709_cast_fp16")]; + tensor var_4710_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_345_cast_fp16)[name = tensor("op_4710_cast_fp16")]; + tensor var_4711_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_347_cast_fp16)[name = tensor("op_4711_cast_fp16")]; + tensor var_4712_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_349_cast_fp16)[name = tensor("op_4712_cast_fp16")]; + tensor var_4713_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_351_cast_fp16)[name = tensor("op_4713_cast_fp16")]; + tensor var_4714_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_353_cast_fp16)[name = tensor("op_4714_cast_fp16")]; + tensor var_4715_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_355_cast_fp16)[name = tensor("op_4715_cast_fp16")]; + tensor var_4716_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_357_cast_fp16)[name = tensor("op_4716_cast_fp16")]; + tensor var_4717_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_359_cast_fp16)[name = tensor("op_4717_cast_fp16")]; + tensor var_4718_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_361_cast_fp16)[name = tensor("op_4718_cast_fp16")]; + tensor var_4719_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_363_cast_fp16)[name = tensor("op_4719_cast_fp16")]; + tensor var_4720_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_365_cast_fp16)[name = tensor("op_4720_cast_fp16")]; + tensor var_4721_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_367_cast_fp16)[name = tensor("op_4721_cast_fp16")]; + tensor var_4722_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_369_cast_fp16)[name = tensor("op_4722_cast_fp16")]; + tensor var_4723_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_371_cast_fp16)[name = tensor("op_4723_cast_fp16")]; + tensor var_4724_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_373_cast_fp16)[name = tensor("op_4724_cast_fp16")]; + tensor var_4725_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_375_cast_fp16)[name = tensor("op_4725_cast_fp16")]; + tensor var_4726_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_377_cast_fp16)[name = tensor("op_4726_cast_fp16")]; + tensor var_4727_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_379_cast_fp16)[name = tensor("op_4727_cast_fp16")]; + tensor var_4728_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_381_cast_fp16)[name = tensor("op_4728_cast_fp16")]; + tensor var_4729_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_383_cast_fp16)[name = tensor("op_4729_cast_fp16")]; + tensor var_4730_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_385_cast_fp16)[name = tensor("op_4730_cast_fp16")]; + tensor var_4731_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_387_cast_fp16)[name = tensor("op_4731_cast_fp16")]; + tensor var_4732_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_389_cast_fp16)[name = tensor("op_4732_cast_fp16")]; + tensor var_4733_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_391_cast_fp16)[name = tensor("op_4733_cast_fp16")]; + tensor var_4734_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_393_cast_fp16)[name = tensor("op_4734_cast_fp16")]; + tensor var_4735_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_395_cast_fp16)[name = tensor("op_4735_cast_fp16")]; + tensor var_4736_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_397_cast_fp16)[name = tensor("op_4736_cast_fp16")]; + tensor var_4737_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_399_cast_fp16)[name = tensor("op_4737_cast_fp16")]; + tensor var_4738_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_401_cast_fp16)[name = tensor("op_4738_cast_fp16")]; + tensor var_4739_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_403_cast_fp16)[name = tensor("op_4739_cast_fp16")]; + tensor var_4740_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_405_cast_fp16)[name = tensor("op_4740_cast_fp16")]; + tensor var_4741_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_407_cast_fp16)[name = tensor("op_4741_cast_fp16")]; + tensor var_4742_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_409_cast_fp16)[name = tensor("op_4742_cast_fp16")]; + tensor var_4743_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_411_cast_fp16)[name = tensor("op_4743_cast_fp16")]; + tensor var_4744_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_413_cast_fp16)[name = tensor("op_4744_cast_fp16")]; + tensor var_4745_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_415_cast_fp16)[name = tensor("op_4745_cast_fp16")]; + tensor var_4746_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_417_cast_fp16)[name = tensor("op_4746_cast_fp16")]; + tensor var_4747_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_419_cast_fp16)[name = tensor("op_4747_cast_fp16")]; + tensor var_4748_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_421_cast_fp16)[name = tensor("op_4748_cast_fp16")]; + tensor var_4749_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_423_cast_fp16)[name = tensor("op_4749_cast_fp16")]; + tensor var_4750_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_425_cast_fp16)[name = tensor("op_4750_cast_fp16")]; + tensor var_4751_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_427_cast_fp16)[name = tensor("op_4751_cast_fp16")]; + tensor var_4752_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_429_cast_fp16)[name = tensor("op_4752_cast_fp16")]; + tensor var_4753_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_431_cast_fp16)[name = tensor("op_4753_cast_fp16")]; + tensor var_4754_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_433_cast_fp16)[name = tensor("op_4754_cast_fp16")]; + tensor var_4755_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_435_cast_fp16)[name = tensor("op_4755_cast_fp16")]; + tensor var_4756_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_437_cast_fp16)[name = tensor("op_4756_cast_fp16")]; + tensor var_4757_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_439_cast_fp16)[name = tensor("op_4757_cast_fp16")]; + tensor var_4758_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_441_cast_fp16)[name = tensor("op_4758_cast_fp16")]; + tensor var_4759_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_443_cast_fp16)[name = tensor("op_4759_cast_fp16")]; + tensor var_4760_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_445_cast_fp16)[name = tensor("op_4760_cast_fp16")]; + tensor var_4761_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_447_cast_fp16)[name = tensor("op_4761_cast_fp16")]; + tensor var_4762_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_449_cast_fp16)[name = tensor("op_4762_cast_fp16")]; + tensor var_4763_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_451_cast_fp16)[name = tensor("op_4763_cast_fp16")]; + tensor var_4764_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_453_cast_fp16)[name = tensor("op_4764_cast_fp16")]; + tensor var_4765_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_455_cast_fp16)[name = tensor("op_4765_cast_fp16")]; + tensor var_4766_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_457_cast_fp16)[name = tensor("op_4766_cast_fp16")]; + tensor var_4767_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_459_cast_fp16)[name = tensor("op_4767_cast_fp16")]; + tensor var_4768_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_461_cast_fp16)[name = tensor("op_4768_cast_fp16")]; + tensor var_4769_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_463_cast_fp16)[name = tensor("op_4769_cast_fp16")]; + tensor var_4770_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_465_cast_fp16)[name = tensor("op_4770_cast_fp16")]; + tensor var_4771_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_467_cast_fp16)[name = tensor("op_4771_cast_fp16")]; + tensor var_4772_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_469_cast_fp16)[name = tensor("op_4772_cast_fp16")]; + tensor var_4773_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_471_cast_fp16)[name = tensor("op_4773_cast_fp16")]; + tensor var_4774_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_473_cast_fp16)[name = tensor("op_4774_cast_fp16")]; + tensor var_4775_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_475_cast_fp16)[name = tensor("op_4775_cast_fp16")]; + tensor var_4776_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_477_cast_fp16)[name = tensor("op_4776_cast_fp16")]; + tensor var_4777_cast_fp16 = softmax(axis = var_3469, x = aw_chunk_479_cast_fp16)[name = tensor("op_4777_cast_fp16")]; + tensor var_4779_equation_0 = const()[name = tensor("op_4779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4779_cast_fp16 = einsum(equation = var_4779_equation_0, values = (var_4299_cast_fp16, var_4698_cast_fp16))[name = tensor("op_4779_cast_fp16")]; + tensor var_4781_equation_0 = const()[name = tensor("op_4781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4781_cast_fp16 = einsum(equation = var_4781_equation_0, values = (var_4299_cast_fp16, var_4699_cast_fp16))[name = tensor("op_4781_cast_fp16")]; + tensor var_4783_equation_0 = const()[name = tensor("op_4783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4783_cast_fp16 = einsum(equation = var_4783_equation_0, values = (var_4299_cast_fp16, var_4700_cast_fp16))[name = tensor("op_4783_cast_fp16")]; + tensor var_4785_equation_0 = const()[name = tensor("op_4785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4785_cast_fp16 = einsum(equation = var_4785_equation_0, values = (var_4299_cast_fp16, var_4701_cast_fp16))[name = tensor("op_4785_cast_fp16")]; + tensor var_4787_equation_0 = const()[name = tensor("op_4787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4787_cast_fp16 = einsum(equation = var_4787_equation_0, values = (var_4303_cast_fp16, var_4702_cast_fp16))[name = tensor("op_4787_cast_fp16")]; + tensor var_4789_equation_0 = const()[name = tensor("op_4789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4789_cast_fp16 = einsum(equation = var_4789_equation_0, values = (var_4303_cast_fp16, var_4703_cast_fp16))[name = tensor("op_4789_cast_fp16")]; + tensor var_4791_equation_0 = const()[name = tensor("op_4791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4791_cast_fp16 = einsum(equation = var_4791_equation_0, values = (var_4303_cast_fp16, var_4704_cast_fp16))[name = tensor("op_4791_cast_fp16")]; + tensor var_4793_equation_0 = const()[name = tensor("op_4793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4793_cast_fp16 = einsum(equation = var_4793_equation_0, values = (var_4303_cast_fp16, var_4705_cast_fp16))[name = tensor("op_4793_cast_fp16")]; + tensor var_4795_equation_0 = const()[name = tensor("op_4795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4795_cast_fp16 = einsum(equation = var_4795_equation_0, values = (var_4307_cast_fp16, var_4706_cast_fp16))[name = tensor("op_4795_cast_fp16")]; + tensor var_4797_equation_0 = const()[name = tensor("op_4797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4797_cast_fp16 = einsum(equation = var_4797_equation_0, values = (var_4307_cast_fp16, var_4707_cast_fp16))[name = tensor("op_4797_cast_fp16")]; + tensor var_4799_equation_0 = const()[name = tensor("op_4799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4799_cast_fp16 = einsum(equation = var_4799_equation_0, values = (var_4307_cast_fp16, var_4708_cast_fp16))[name = tensor("op_4799_cast_fp16")]; + tensor var_4801_equation_0 = const()[name = tensor("op_4801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4801_cast_fp16 = einsum(equation = var_4801_equation_0, values = (var_4307_cast_fp16, var_4709_cast_fp16))[name = tensor("op_4801_cast_fp16")]; + tensor var_4803_equation_0 = const()[name = tensor("op_4803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4803_cast_fp16 = einsum(equation = var_4803_equation_0, values = (var_4311_cast_fp16, var_4710_cast_fp16))[name = tensor("op_4803_cast_fp16")]; + tensor var_4805_equation_0 = const()[name = tensor("op_4805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4805_cast_fp16 = einsum(equation = var_4805_equation_0, values = (var_4311_cast_fp16, var_4711_cast_fp16))[name = tensor("op_4805_cast_fp16")]; + tensor var_4807_equation_0 = const()[name = tensor("op_4807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4807_cast_fp16 = einsum(equation = var_4807_equation_0, values = (var_4311_cast_fp16, var_4712_cast_fp16))[name = tensor("op_4807_cast_fp16")]; + tensor var_4809_equation_0 = const()[name = tensor("op_4809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4809_cast_fp16 = einsum(equation = var_4809_equation_0, values = (var_4311_cast_fp16, var_4713_cast_fp16))[name = tensor("op_4809_cast_fp16")]; + tensor var_4811_equation_0 = const()[name = tensor("op_4811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4811_cast_fp16 = einsum(equation = var_4811_equation_0, values = (var_4315_cast_fp16, var_4714_cast_fp16))[name = tensor("op_4811_cast_fp16")]; + tensor var_4813_equation_0 = const()[name = tensor("op_4813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4813_cast_fp16 = einsum(equation = var_4813_equation_0, values = (var_4315_cast_fp16, var_4715_cast_fp16))[name = tensor("op_4813_cast_fp16")]; + tensor var_4815_equation_0 = const()[name = tensor("op_4815_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4815_cast_fp16 = einsum(equation = var_4815_equation_0, values = (var_4315_cast_fp16, var_4716_cast_fp16))[name = tensor("op_4815_cast_fp16")]; + tensor var_4817_equation_0 = const()[name = tensor("op_4817_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4817_cast_fp16 = einsum(equation = var_4817_equation_0, values = (var_4315_cast_fp16, var_4717_cast_fp16))[name = tensor("op_4817_cast_fp16")]; + tensor var_4819_equation_0 = const()[name = tensor("op_4819_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4819_cast_fp16 = einsum(equation = var_4819_equation_0, values = (var_4319_cast_fp16, var_4718_cast_fp16))[name = tensor("op_4819_cast_fp16")]; + tensor var_4821_equation_0 = const()[name = tensor("op_4821_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4821_cast_fp16 = einsum(equation = var_4821_equation_0, values = (var_4319_cast_fp16, var_4719_cast_fp16))[name = tensor("op_4821_cast_fp16")]; + tensor var_4823_equation_0 = const()[name = tensor("op_4823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4823_cast_fp16 = einsum(equation = var_4823_equation_0, values = (var_4319_cast_fp16, var_4720_cast_fp16))[name = tensor("op_4823_cast_fp16")]; + tensor var_4825_equation_0 = const()[name = tensor("op_4825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4825_cast_fp16 = einsum(equation = var_4825_equation_0, values = (var_4319_cast_fp16, var_4721_cast_fp16))[name = tensor("op_4825_cast_fp16")]; + tensor var_4827_equation_0 = const()[name = tensor("op_4827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4827_cast_fp16 = einsum(equation = var_4827_equation_0, values = (var_4323_cast_fp16, var_4722_cast_fp16))[name = tensor("op_4827_cast_fp16")]; + tensor var_4829_equation_0 = const()[name = tensor("op_4829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4829_cast_fp16 = einsum(equation = var_4829_equation_0, values = (var_4323_cast_fp16, var_4723_cast_fp16))[name = tensor("op_4829_cast_fp16")]; + tensor var_4831_equation_0 = const()[name = tensor("op_4831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4831_cast_fp16 = einsum(equation = var_4831_equation_0, values = (var_4323_cast_fp16, var_4724_cast_fp16))[name = tensor("op_4831_cast_fp16")]; + tensor var_4833_equation_0 = const()[name = tensor("op_4833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4833_cast_fp16 = einsum(equation = var_4833_equation_0, values = (var_4323_cast_fp16, var_4725_cast_fp16))[name = tensor("op_4833_cast_fp16")]; + tensor var_4835_equation_0 = const()[name = tensor("op_4835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4835_cast_fp16 = einsum(equation = var_4835_equation_0, values = (var_4327_cast_fp16, var_4726_cast_fp16))[name = tensor("op_4835_cast_fp16")]; + tensor var_4837_equation_0 = const()[name = tensor("op_4837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4837_cast_fp16 = einsum(equation = var_4837_equation_0, values = (var_4327_cast_fp16, var_4727_cast_fp16))[name = tensor("op_4837_cast_fp16")]; + tensor var_4839_equation_0 = const()[name = tensor("op_4839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4839_cast_fp16 = einsum(equation = var_4839_equation_0, values = (var_4327_cast_fp16, var_4728_cast_fp16))[name = tensor("op_4839_cast_fp16")]; + tensor var_4841_equation_0 = const()[name = tensor("op_4841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4841_cast_fp16 = einsum(equation = var_4841_equation_0, values = (var_4327_cast_fp16, var_4729_cast_fp16))[name = tensor("op_4841_cast_fp16")]; + tensor var_4843_equation_0 = const()[name = tensor("op_4843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4843_cast_fp16 = einsum(equation = var_4843_equation_0, values = (var_4331_cast_fp16, var_4730_cast_fp16))[name = tensor("op_4843_cast_fp16")]; + tensor var_4845_equation_0 = const()[name = tensor("op_4845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4845_cast_fp16 = einsum(equation = var_4845_equation_0, values = (var_4331_cast_fp16, var_4731_cast_fp16))[name = tensor("op_4845_cast_fp16")]; + tensor var_4847_equation_0 = const()[name = tensor("op_4847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4847_cast_fp16 = einsum(equation = var_4847_equation_0, values = (var_4331_cast_fp16, var_4732_cast_fp16))[name = tensor("op_4847_cast_fp16")]; + tensor var_4849_equation_0 = const()[name = tensor("op_4849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4849_cast_fp16 = einsum(equation = var_4849_equation_0, values = (var_4331_cast_fp16, var_4733_cast_fp16))[name = tensor("op_4849_cast_fp16")]; + tensor var_4851_equation_0 = const()[name = tensor("op_4851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4851_cast_fp16 = einsum(equation = var_4851_equation_0, values = (var_4335_cast_fp16, var_4734_cast_fp16))[name = tensor("op_4851_cast_fp16")]; + tensor var_4853_equation_0 = const()[name = tensor("op_4853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4853_cast_fp16 = einsum(equation = var_4853_equation_0, values = (var_4335_cast_fp16, var_4735_cast_fp16))[name = tensor("op_4853_cast_fp16")]; + tensor var_4855_equation_0 = const()[name = tensor("op_4855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4855_cast_fp16 = einsum(equation = var_4855_equation_0, values = (var_4335_cast_fp16, var_4736_cast_fp16))[name = tensor("op_4855_cast_fp16")]; + tensor var_4857_equation_0 = const()[name = tensor("op_4857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4857_cast_fp16 = einsum(equation = var_4857_equation_0, values = (var_4335_cast_fp16, var_4737_cast_fp16))[name = tensor("op_4857_cast_fp16")]; + tensor var_4859_equation_0 = const()[name = tensor("op_4859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4859_cast_fp16 = einsum(equation = var_4859_equation_0, values = (var_4339_cast_fp16, var_4738_cast_fp16))[name = tensor("op_4859_cast_fp16")]; + tensor var_4861_equation_0 = const()[name = tensor("op_4861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4861_cast_fp16 = einsum(equation = var_4861_equation_0, values = (var_4339_cast_fp16, var_4739_cast_fp16))[name = tensor("op_4861_cast_fp16")]; + tensor var_4863_equation_0 = const()[name = tensor("op_4863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4863_cast_fp16 = einsum(equation = var_4863_equation_0, values = (var_4339_cast_fp16, var_4740_cast_fp16))[name = tensor("op_4863_cast_fp16")]; + tensor var_4865_equation_0 = const()[name = tensor("op_4865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4865_cast_fp16 = einsum(equation = var_4865_equation_0, values = (var_4339_cast_fp16, var_4741_cast_fp16))[name = tensor("op_4865_cast_fp16")]; + tensor var_4867_equation_0 = const()[name = tensor("op_4867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4867_cast_fp16 = einsum(equation = var_4867_equation_0, values = (var_4343_cast_fp16, var_4742_cast_fp16))[name = tensor("op_4867_cast_fp16")]; + tensor var_4869_equation_0 = const()[name = tensor("op_4869_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4869_cast_fp16 = einsum(equation = var_4869_equation_0, values = (var_4343_cast_fp16, var_4743_cast_fp16))[name = tensor("op_4869_cast_fp16")]; + tensor var_4871_equation_0 = const()[name = tensor("op_4871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4871_cast_fp16 = einsum(equation = var_4871_equation_0, values = (var_4343_cast_fp16, var_4744_cast_fp16))[name = tensor("op_4871_cast_fp16")]; + tensor var_4873_equation_0 = const()[name = tensor("op_4873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4873_cast_fp16 = einsum(equation = var_4873_equation_0, values = (var_4343_cast_fp16, var_4745_cast_fp16))[name = tensor("op_4873_cast_fp16")]; + tensor var_4875_equation_0 = const()[name = tensor("op_4875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4875_cast_fp16 = einsum(equation = var_4875_equation_0, values = (var_4347_cast_fp16, var_4746_cast_fp16))[name = tensor("op_4875_cast_fp16")]; + tensor var_4877_equation_0 = const()[name = tensor("op_4877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4877_cast_fp16 = einsum(equation = var_4877_equation_0, values = (var_4347_cast_fp16, var_4747_cast_fp16))[name = tensor("op_4877_cast_fp16")]; + tensor var_4879_equation_0 = const()[name = tensor("op_4879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4879_cast_fp16 = einsum(equation = var_4879_equation_0, values = (var_4347_cast_fp16, var_4748_cast_fp16))[name = tensor("op_4879_cast_fp16")]; + tensor var_4881_equation_0 = const()[name = tensor("op_4881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4881_cast_fp16 = einsum(equation = var_4881_equation_0, values = (var_4347_cast_fp16, var_4749_cast_fp16))[name = tensor("op_4881_cast_fp16")]; + tensor var_4883_equation_0 = const()[name = tensor("op_4883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4883_cast_fp16 = einsum(equation = var_4883_equation_0, values = (var_4351_cast_fp16, var_4750_cast_fp16))[name = tensor("op_4883_cast_fp16")]; + tensor var_4885_equation_0 = const()[name = tensor("op_4885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4885_cast_fp16 = einsum(equation = var_4885_equation_0, values = (var_4351_cast_fp16, var_4751_cast_fp16))[name = tensor("op_4885_cast_fp16")]; + tensor var_4887_equation_0 = const()[name = tensor("op_4887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4887_cast_fp16 = einsum(equation = var_4887_equation_0, values = (var_4351_cast_fp16, var_4752_cast_fp16))[name = tensor("op_4887_cast_fp16")]; + tensor var_4889_equation_0 = const()[name = tensor("op_4889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4889_cast_fp16 = einsum(equation = var_4889_equation_0, values = (var_4351_cast_fp16, var_4753_cast_fp16))[name = tensor("op_4889_cast_fp16")]; + tensor var_4891_equation_0 = const()[name = tensor("op_4891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4891_cast_fp16 = einsum(equation = var_4891_equation_0, values = (var_4355_cast_fp16, var_4754_cast_fp16))[name = tensor("op_4891_cast_fp16")]; + tensor var_4893_equation_0 = const()[name = tensor("op_4893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4893_cast_fp16 = einsum(equation = var_4893_equation_0, values = (var_4355_cast_fp16, var_4755_cast_fp16))[name = tensor("op_4893_cast_fp16")]; + tensor var_4895_equation_0 = const()[name = tensor("op_4895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4895_cast_fp16 = einsum(equation = var_4895_equation_0, values = (var_4355_cast_fp16, var_4756_cast_fp16))[name = tensor("op_4895_cast_fp16")]; + tensor var_4897_equation_0 = const()[name = tensor("op_4897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4897_cast_fp16 = einsum(equation = var_4897_equation_0, values = (var_4355_cast_fp16, var_4757_cast_fp16))[name = tensor("op_4897_cast_fp16")]; + tensor var_4899_equation_0 = const()[name = tensor("op_4899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4899_cast_fp16 = einsum(equation = var_4899_equation_0, values = (var_4359_cast_fp16, var_4758_cast_fp16))[name = tensor("op_4899_cast_fp16")]; + tensor var_4901_equation_0 = const()[name = tensor("op_4901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4901_cast_fp16 = einsum(equation = var_4901_equation_0, values = (var_4359_cast_fp16, var_4759_cast_fp16))[name = tensor("op_4901_cast_fp16")]; + tensor var_4903_equation_0 = const()[name = tensor("op_4903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4903_cast_fp16 = einsum(equation = var_4903_equation_0, values = (var_4359_cast_fp16, var_4760_cast_fp16))[name = tensor("op_4903_cast_fp16")]; + tensor var_4905_equation_0 = const()[name = tensor("op_4905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4905_cast_fp16 = einsum(equation = var_4905_equation_0, values = (var_4359_cast_fp16, var_4761_cast_fp16))[name = tensor("op_4905_cast_fp16")]; + tensor var_4907_equation_0 = const()[name = tensor("op_4907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4907_cast_fp16 = einsum(equation = var_4907_equation_0, values = (var_4363_cast_fp16, var_4762_cast_fp16))[name = tensor("op_4907_cast_fp16")]; + tensor var_4909_equation_0 = const()[name = tensor("op_4909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4909_cast_fp16 = einsum(equation = var_4909_equation_0, values = (var_4363_cast_fp16, var_4763_cast_fp16))[name = tensor("op_4909_cast_fp16")]; + tensor var_4911_equation_0 = const()[name = tensor("op_4911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4911_cast_fp16 = einsum(equation = var_4911_equation_0, values = (var_4363_cast_fp16, var_4764_cast_fp16))[name = tensor("op_4911_cast_fp16")]; + tensor var_4913_equation_0 = const()[name = tensor("op_4913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4913_cast_fp16 = einsum(equation = var_4913_equation_0, values = (var_4363_cast_fp16, var_4765_cast_fp16))[name = tensor("op_4913_cast_fp16")]; + tensor var_4915_equation_0 = const()[name = tensor("op_4915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4915_cast_fp16 = einsum(equation = var_4915_equation_0, values = (var_4367_cast_fp16, var_4766_cast_fp16))[name = tensor("op_4915_cast_fp16")]; + tensor var_4917_equation_0 = const()[name = tensor("op_4917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4917_cast_fp16 = einsum(equation = var_4917_equation_0, values = (var_4367_cast_fp16, var_4767_cast_fp16))[name = tensor("op_4917_cast_fp16")]; + tensor var_4919_equation_0 = const()[name = tensor("op_4919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4919_cast_fp16 = einsum(equation = var_4919_equation_0, values = (var_4367_cast_fp16, var_4768_cast_fp16))[name = tensor("op_4919_cast_fp16")]; + tensor var_4921_equation_0 = const()[name = tensor("op_4921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4921_cast_fp16 = einsum(equation = var_4921_equation_0, values = (var_4367_cast_fp16, var_4769_cast_fp16))[name = tensor("op_4921_cast_fp16")]; + tensor var_4923_equation_0 = const()[name = tensor("op_4923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4923_cast_fp16 = einsum(equation = var_4923_equation_0, values = (var_4371_cast_fp16, var_4770_cast_fp16))[name = tensor("op_4923_cast_fp16")]; + tensor var_4925_equation_0 = const()[name = tensor("op_4925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4925_cast_fp16 = einsum(equation = var_4925_equation_0, values = (var_4371_cast_fp16, var_4771_cast_fp16))[name = tensor("op_4925_cast_fp16")]; + tensor var_4927_equation_0 = const()[name = tensor("op_4927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4927_cast_fp16 = einsum(equation = var_4927_equation_0, values = (var_4371_cast_fp16, var_4772_cast_fp16))[name = tensor("op_4927_cast_fp16")]; + tensor var_4929_equation_0 = const()[name = tensor("op_4929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4929_cast_fp16 = einsum(equation = var_4929_equation_0, values = (var_4371_cast_fp16, var_4773_cast_fp16))[name = tensor("op_4929_cast_fp16")]; + tensor var_4931_equation_0 = const()[name = tensor("op_4931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4931_cast_fp16 = einsum(equation = var_4931_equation_0, values = (var_4375_cast_fp16, var_4774_cast_fp16))[name = tensor("op_4931_cast_fp16")]; + tensor var_4933_equation_0 = const()[name = tensor("op_4933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4933_cast_fp16 = einsum(equation = var_4933_equation_0, values = (var_4375_cast_fp16, var_4775_cast_fp16))[name = tensor("op_4933_cast_fp16")]; + tensor var_4935_equation_0 = const()[name = tensor("op_4935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4935_cast_fp16 = einsum(equation = var_4935_equation_0, values = (var_4375_cast_fp16, var_4776_cast_fp16))[name = tensor("op_4935_cast_fp16")]; + tensor var_4937_equation_0 = const()[name = tensor("op_4937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_4937_cast_fp16 = einsum(equation = var_4937_equation_0, values = (var_4375_cast_fp16, var_4777_cast_fp16))[name = tensor("op_4937_cast_fp16")]; + tensor var_4939_interleave_0 = const()[name = tensor("op_4939_interleave_0"), val = tensor(false)]; + tensor var_4939_cast_fp16 = concat(axis = var_3444, interleave = var_4939_interleave_0, values = (var_4779_cast_fp16, var_4781_cast_fp16, var_4783_cast_fp16, var_4785_cast_fp16))[name = tensor("op_4939_cast_fp16")]; + tensor var_4941_interleave_0 = const()[name = tensor("op_4941_interleave_0"), val = tensor(false)]; + tensor var_4941_cast_fp16 = concat(axis = var_3444, interleave = var_4941_interleave_0, values = (var_4787_cast_fp16, var_4789_cast_fp16, var_4791_cast_fp16, var_4793_cast_fp16))[name = tensor("op_4941_cast_fp16")]; + tensor var_4943_interleave_0 = const()[name = tensor("op_4943_interleave_0"), val = tensor(false)]; + tensor var_4943_cast_fp16 = concat(axis = var_3444, interleave = var_4943_interleave_0, values = (var_4795_cast_fp16, var_4797_cast_fp16, var_4799_cast_fp16, var_4801_cast_fp16))[name = tensor("op_4943_cast_fp16")]; + tensor var_4945_interleave_0 = const()[name = tensor("op_4945_interleave_0"), val = tensor(false)]; + tensor var_4945_cast_fp16 = concat(axis = var_3444, interleave = var_4945_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16))[name = tensor("op_4945_cast_fp16")]; + tensor var_4947_interleave_0 = const()[name = tensor("op_4947_interleave_0"), val = tensor(false)]; + tensor var_4947_cast_fp16 = concat(axis = var_3444, interleave = var_4947_interleave_0, values = (var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16))[name = tensor("op_4947_cast_fp16")]; + tensor var_4949_interleave_0 = const()[name = tensor("op_4949_interleave_0"), val = tensor(false)]; + tensor var_4949_cast_fp16 = concat(axis = var_3444, interleave = var_4949_interleave_0, values = (var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16))[name = tensor("op_4949_cast_fp16")]; + tensor var_4951_interleave_0 = const()[name = tensor("op_4951_interleave_0"), val = tensor(false)]; + tensor var_4951_cast_fp16 = concat(axis = var_3444, interleave = var_4951_interleave_0, values = (var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = tensor("op_4951_cast_fp16")]; + tensor var_4953_interleave_0 = const()[name = tensor("op_4953_interleave_0"), val = tensor(false)]; + tensor var_4953_cast_fp16 = concat(axis = var_3444, interleave = var_4953_interleave_0, values = (var_4835_cast_fp16, var_4837_cast_fp16, var_4839_cast_fp16, var_4841_cast_fp16))[name = tensor("op_4953_cast_fp16")]; + tensor var_4955_interleave_0 = const()[name = tensor("op_4955_interleave_0"), val = tensor(false)]; + tensor var_4955_cast_fp16 = concat(axis = var_3444, interleave = var_4955_interleave_0, values = (var_4843_cast_fp16, var_4845_cast_fp16, var_4847_cast_fp16, var_4849_cast_fp16))[name = tensor("op_4955_cast_fp16")]; + tensor var_4957_interleave_0 = const()[name = tensor("op_4957_interleave_0"), val = tensor(false)]; + tensor var_4957_cast_fp16 = concat(axis = var_3444, interleave = var_4957_interleave_0, values = (var_4851_cast_fp16, var_4853_cast_fp16, var_4855_cast_fp16, var_4857_cast_fp16))[name = tensor("op_4957_cast_fp16")]; + tensor var_4959_interleave_0 = const()[name = tensor("op_4959_interleave_0"), val = tensor(false)]; + tensor var_4959_cast_fp16 = concat(axis = var_3444, interleave = var_4959_interleave_0, values = (var_4859_cast_fp16, var_4861_cast_fp16, var_4863_cast_fp16, var_4865_cast_fp16))[name = tensor("op_4959_cast_fp16")]; + tensor var_4961_interleave_0 = const()[name = tensor("op_4961_interleave_0"), val = tensor(false)]; + tensor var_4961_cast_fp16 = concat(axis = var_3444, interleave = var_4961_interleave_0, values = (var_4867_cast_fp16, var_4869_cast_fp16, var_4871_cast_fp16, var_4873_cast_fp16))[name = tensor("op_4961_cast_fp16")]; + tensor var_4963_interleave_0 = const()[name = tensor("op_4963_interleave_0"), val = tensor(false)]; + tensor var_4963_cast_fp16 = concat(axis = var_3444, interleave = var_4963_interleave_0, values = (var_4875_cast_fp16, var_4877_cast_fp16, var_4879_cast_fp16, var_4881_cast_fp16))[name = tensor("op_4963_cast_fp16")]; + tensor var_4965_interleave_0 = const()[name = tensor("op_4965_interleave_0"), val = tensor(false)]; + tensor var_4965_cast_fp16 = concat(axis = var_3444, interleave = var_4965_interleave_0, values = (var_4883_cast_fp16, var_4885_cast_fp16, var_4887_cast_fp16, var_4889_cast_fp16))[name = tensor("op_4965_cast_fp16")]; + tensor var_4967_interleave_0 = const()[name = tensor("op_4967_interleave_0"), val = tensor(false)]; + tensor var_4967_cast_fp16 = concat(axis = var_3444, interleave = var_4967_interleave_0, values = (var_4891_cast_fp16, var_4893_cast_fp16, var_4895_cast_fp16, var_4897_cast_fp16))[name = tensor("op_4967_cast_fp16")]; + tensor var_4969_interleave_0 = const()[name = tensor("op_4969_interleave_0"), val = tensor(false)]; + tensor var_4969_cast_fp16 = concat(axis = var_3444, interleave = var_4969_interleave_0, values = (var_4899_cast_fp16, var_4901_cast_fp16, var_4903_cast_fp16, var_4905_cast_fp16))[name = tensor("op_4969_cast_fp16")]; + tensor var_4971_interleave_0 = const()[name = tensor("op_4971_interleave_0"), val = tensor(false)]; + tensor var_4971_cast_fp16 = concat(axis = var_3444, interleave = var_4971_interleave_0, values = (var_4907_cast_fp16, var_4909_cast_fp16, var_4911_cast_fp16, var_4913_cast_fp16))[name = tensor("op_4971_cast_fp16")]; + tensor var_4973_interleave_0 = const()[name = tensor("op_4973_interleave_0"), val = tensor(false)]; + tensor var_4973_cast_fp16 = concat(axis = var_3444, interleave = var_4973_interleave_0, values = (var_4915_cast_fp16, var_4917_cast_fp16, var_4919_cast_fp16, var_4921_cast_fp16))[name = tensor("op_4973_cast_fp16")]; + tensor var_4975_interleave_0 = const()[name = tensor("op_4975_interleave_0"), val = tensor(false)]; + tensor var_4975_cast_fp16 = concat(axis = var_3444, interleave = var_4975_interleave_0, values = (var_4923_cast_fp16, var_4925_cast_fp16, var_4927_cast_fp16, var_4929_cast_fp16))[name = tensor("op_4975_cast_fp16")]; + tensor var_4977_interleave_0 = const()[name = tensor("op_4977_interleave_0"), val = tensor(false)]; + tensor var_4977_cast_fp16 = concat(axis = var_3444, interleave = var_4977_interleave_0, values = (var_4931_cast_fp16, var_4933_cast_fp16, var_4935_cast_fp16, var_4937_cast_fp16))[name = tensor("op_4977_cast_fp16")]; + tensor input_47_interleave_0 = const()[name = tensor("input_47_interleave_0"), val = tensor(false)]; + tensor input_47_cast_fp16 = concat(axis = var_3469, interleave = input_47_interleave_0, values = (var_4939_cast_fp16, var_4941_cast_fp16, var_4943_cast_fp16, var_4945_cast_fp16, var_4947_cast_fp16, var_4949_cast_fp16, var_4951_cast_fp16, var_4953_cast_fp16, var_4955_cast_fp16, var_4957_cast_fp16, var_4959_cast_fp16, var_4961_cast_fp16, var_4963_cast_fp16, var_4965_cast_fp16, var_4967_cast_fp16, var_4969_cast_fp16, var_4971_cast_fp16, var_4973_cast_fp16, var_4975_cast_fp16, var_4977_cast_fp16))[name = tensor("input_47_cast_fp16")]; + tensor var_4985 = const()[name = tensor("op_4985"), val = tensor([1, 1])]; + tensor var_4987 = const()[name = tensor("op_4987"), val = tensor([1, 1])]; + tensor pretrained_out_31_pad_type_0 = const()[name = tensor("pretrained_out_31_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_31_pad_0 = const()[name = tensor("pretrained_out_31_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38222464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39041728))), name = tensor("layers_2_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_2_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39041856)))]; + tensor pretrained_out_31_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_4987, groups = var_3469, pad = pretrained_out_31_pad_0, pad_type = pretrained_out_31_pad_type_0, strides = var_4985, weight = layers_2_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = tensor("pretrained_out_31_cast_fp16")]; + tensor var_4991 = const()[name = tensor("op_4991"), val = tensor([1, 1])]; + tensor var_4993 = const()[name = tensor("op_4993"), val = tensor([1, 1])]; + tensor input_49_pad_type_0 = const()[name = tensor("input_49_pad_type_0"), val = tensor("custom")]; + tensor input_49_pad_0 = const()[name = tensor("input_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39044480)))]; + tensor input_49_cast_fp16 = conv(dilations = var_4993, groups = var_3469, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = var_4991, weight = layers_2_self_attn_o_proj_loraA_weight_to_fp16, x = input_47_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor var_4997 = const()[name = tensor("op_4997"), val = tensor([1, 1])]; + tensor var_4999 = const()[name = tensor("op_4999"), val = tensor([1, 1])]; + tensor lora_out_61_pad_type_0 = const()[name = tensor("lora_out_61_pad_type_0"), val = tensor("custom")]; + tensor lora_out_61_pad_0 = const()[name = tensor("lora_out_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_63_weight_0_to_fp16 = const()[name = tensor("lora_out_63_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39085504)))]; + tensor lora_out_63_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_4999, groups = var_3469, pad = lora_out_61_pad_0, pad_type = lora_out_61_pad_type_0, strides = var_4997, weight = lora_out_63_weight_0_to_fp16, x = input_49_cast_fp16)[name = tensor("lora_out_63_cast_fp16")]; + tensor obj_11_cast_fp16 = add(x = pretrained_out_31_cast_fp16, y = lora_out_63_cast_fp16)[name = tensor("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor var_5008 = const()[name = tensor("op_5008"), val = tensor([1])]; + tensor channels_mean_11_cast_fp16 = reduce_mean(axes = var_5008, keep_dims = var_3470, x = inputs_11_cast_fp16)[name = tensor("channels_mean_11_cast_fp16")]; + tensor zero_mean_11_cast_fp16 = sub(x = inputs_11_cast_fp16, y = channels_mean_11_cast_fp16)[name = tensor("zero_mean_11_cast_fp16")]; + tensor zero_mean_sq_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = zero_mean_11_cast_fp16)[name = tensor("zero_mean_sq_11_cast_fp16")]; + tensor var_5012 = const()[name = tensor("op_5012"), val = tensor([1])]; + tensor var_5013_cast_fp16 = reduce_mean(axes = var_5012, keep_dims = var_3470, x = zero_mean_sq_11_cast_fp16)[name = tensor("op_5013_cast_fp16")]; + tensor var_5014_to_fp16 = const()[name = tensor("op_5014_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_5015_cast_fp16 = add(x = var_5013_cast_fp16, y = var_5014_to_fp16)[name = tensor("op_5015_cast_fp16")]; + tensor denom_11_epsilon_0 = const()[name = tensor("denom_11_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_11_cast_fp16 = rsqrt(epsilon = denom_11_epsilon_0, x = var_5015_cast_fp16)[name = tensor("denom_11_cast_fp16")]; + tensor out_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = denom_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = tensor("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39126528)))]; + tensor input_51_beta_0_to_fp16 = const()[name = tensor("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39129152)))]; + tensor input_51_epsilon_0_to_fp16 = const()[name = tensor("input_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor("input_51_cast_fp16")]; + tensor var_5029 = const()[name = tensor("op_5029"), val = tensor([1, 1])]; + tensor var_5031 = const()[name = tensor("op_5031"), val = tensor([1, 1])]; + tensor pretrained_out_33_pad_type_0 = const()[name = tensor("pretrained_out_33_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_33_pad_0 = const()[name = tensor("pretrained_out_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39131776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42408640))), name = tensor("layers_2_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_2_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_2_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42408768)))]; + tensor pretrained_out_33_cast_fp16 = conv(bias = layers_2_fc1_pretrained_bias_to_fp16, dilations = var_5031, groups = var_3469, pad = pretrained_out_33_pad_0, pad_type = pretrained_out_33_pad_type_0, strides = var_5029, weight = layers_2_fc1_pretrained_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = tensor("pretrained_out_33_cast_fp16")]; + tensor var_5035 = const()[name = tensor("op_5035"), val = tensor([1, 1])]; + tensor var_5037 = const()[name = tensor("op_5037"), val = tensor([1, 1])]; + tensor input_53_pad_type_0 = const()[name = tensor("input_53_pad_type_0"), val = tensor("custom")]; + tensor input_53_pad_0 = const()[name = tensor("input_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_2_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42419072)))]; + tensor input_53_cast_fp16 = conv(dilations = var_5037, groups = var_3469, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = var_5035, weight = layers_2_fc1_loraA_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor var_5041 = const()[name = tensor("op_5041"), val = tensor([1, 1])]; + tensor var_5043 = const()[name = tensor("op_5043"), val = tensor([1, 1])]; + tensor lora_out_65_pad_type_0 = const()[name = tensor("lora_out_65_pad_type_0"), val = tensor("custom")]; + tensor lora_out_65_pad_0 = const()[name = tensor("lora_out_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_67_weight_0_to_fp16 = const()[name = tensor("lora_out_67_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42460096)))]; + tensor lora_out_67_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_5043, groups = var_3469, pad = lora_out_65_pad_0, pad_type = lora_out_65_pad_type_0, strides = var_5041, weight = lora_out_67_weight_0_to_fp16, x = input_53_cast_fp16)[name = tensor("lora_out_67_cast_fp16")]; + tensor input_55_cast_fp16 = add(x = pretrained_out_33_cast_fp16, y = lora_out_67_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor input_57_mode_0 = const()[name = tensor("input_57_mode_0"), val = tensor("EXACT")]; + tensor input_57_cast_fp16 = gelu(mode = input_57_mode_0, x = input_55_cast_fp16)[name = tensor("input_57_cast_fp16")]; + tensor var_5055 = const()[name = tensor("op_5055"), val = tensor([1, 1])]; + tensor var_5057 = const()[name = tensor("op_5057"), val = tensor([1, 1])]; + tensor pretrained_out_35_pad_type_0 = const()[name = tensor("pretrained_out_35_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_35_pad_0 = const()[name = tensor("pretrained_out_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42624000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45900864))), name = tensor("layers_2_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_2_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_2_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45900992)))]; + tensor pretrained_out_35_cast_fp16 = conv(bias = layers_2_fc2_pretrained_bias_to_fp16, dilations = var_5057, groups = var_3469, pad = pretrained_out_35_pad_0, pad_type = pretrained_out_35_pad_type_0, strides = var_5055, weight = layers_2_fc2_pretrained_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("pretrained_out_35_cast_fp16")]; + tensor var_5061 = const()[name = tensor("op_5061"), val = tensor([1, 1])]; + tensor var_5063 = const()[name = tensor("op_5063"), val = tensor([1, 1])]; + tensor input_59_pad_type_0 = const()[name = tensor("input_59_pad_type_0"), val = tensor("custom")]; + tensor input_59_pad_0 = const()[name = tensor("input_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_2_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_2_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45903616)))]; + tensor input_59_cast_fp16 = conv(dilations = var_5063, groups = var_3469, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = var_5061, weight = layers_2_fc2_loraA_weight_to_fp16, x = input_57_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor var_5067 = const()[name = tensor("op_5067"), val = tensor([1, 1])]; + tensor var_5069 = const()[name = tensor("op_5069"), val = tensor([1, 1])]; + tensor lora_out_69_pad_type_0 = const()[name = tensor("lora_out_69_pad_type_0"), val = tensor("custom")]; + tensor lora_out_69_pad_0 = const()[name = tensor("lora_out_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_71_weight_0_to_fp16 = const()[name = tensor("lora_out_71_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46067520)))]; + tensor lora_out_71_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_5069, groups = var_3469, pad = lora_out_69_pad_0, pad_type = lora_out_69_pad_type_0, strides = var_5067, weight = lora_out_71_weight_0_to_fp16, x = input_59_cast_fp16)[name = tensor("lora_out_71_cast_fp16")]; + tensor hidden_states_9_cast_fp16 = add(x = pretrained_out_35_cast_fp16, y = lora_out_71_cast_fp16)[name = tensor("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor var_5079 = const()[name = tensor("op_5079"), val = tensor(3)]; + tensor var_5104 = const()[name = tensor("op_5104"), val = tensor(1)]; + tensor var_5105 = const()[name = tensor("op_5105"), val = tensor(true)]; + tensor var_5115 = const()[name = tensor("op_5115"), val = tensor([1])]; + tensor channels_mean_13_cast_fp16 = reduce_mean(axes = var_5115, keep_dims = var_5105, x = inputs_13_cast_fp16)[name = tensor("channels_mean_13_cast_fp16")]; + tensor zero_mean_13_cast_fp16 = sub(x = inputs_13_cast_fp16, y = channels_mean_13_cast_fp16)[name = tensor("zero_mean_13_cast_fp16")]; + tensor zero_mean_sq_13_cast_fp16 = mul(x = zero_mean_13_cast_fp16, y = zero_mean_13_cast_fp16)[name = tensor("zero_mean_sq_13_cast_fp16")]; + tensor var_5119 = const()[name = tensor("op_5119"), val = tensor([1])]; + tensor var_5120_cast_fp16 = reduce_mean(axes = var_5119, keep_dims = var_5105, x = zero_mean_sq_13_cast_fp16)[name = tensor("op_5120_cast_fp16")]; + tensor var_5121_to_fp16 = const()[name = tensor("op_5121_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_5122_cast_fp16 = add(x = var_5120_cast_fp16, y = var_5121_to_fp16)[name = tensor("op_5122_cast_fp16")]; + tensor denom_13_epsilon_0 = const()[name = tensor("denom_13_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_13_cast_fp16 = rsqrt(epsilon = denom_13_epsilon_0, x = var_5122_cast_fp16)[name = tensor("denom_13_cast_fp16")]; + tensor out_13_cast_fp16 = mul(x = zero_mean_13_cast_fp16, y = denom_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = tensor("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46108544)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = tensor("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46111168)))]; + tensor obj_13_epsilon_0_to_fp16 = const()[name = tensor("obj_13_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor("obj_13_cast_fp16")]; + tensor var_5140 = const()[name = tensor("op_5140"), val = tensor([1, 1])]; + tensor var_5142 = const()[name = tensor("op_5142"), val = tensor([1, 1])]; + tensor pretrained_out_37_pad_type_0 = const()[name = tensor("pretrained_out_37_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_37_pad_0 = const()[name = tensor("pretrained_out_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46113792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46933056))), name = tensor("layers_3_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_3_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46933184)))]; + tensor pretrained_out_37_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_5142, groups = var_5104, pad = pretrained_out_37_pad_0, pad_type = pretrained_out_37_pad_type_0, strides = var_5140, weight = layers_3_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor("pretrained_out_37_cast_fp16")]; + tensor var_5146 = const()[name = tensor("op_5146"), val = tensor([1, 1])]; + tensor var_5148 = const()[name = tensor("op_5148"), val = tensor([1, 1])]; + tensor input_61_pad_type_0 = const()[name = tensor("input_61_pad_type_0"), val = tensor("custom")]; + tensor input_61_pad_0 = const()[name = tensor("input_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46935808)))]; + tensor input_61_cast_fp16 = conv(dilations = var_5148, groups = var_5104, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = var_5146, weight = layers_3_self_attn_q_proj_loraA_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor var_5152 = const()[name = tensor("op_5152"), val = tensor([1, 1])]; + tensor var_5154 = const()[name = tensor("op_5154"), val = tensor([1, 1])]; + tensor lora_out_73_pad_type_0 = const()[name = tensor("lora_out_73_pad_type_0"), val = tensor("custom")]; + tensor lora_out_73_pad_0 = const()[name = tensor("lora_out_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_75_weight_0_to_fp16 = const()[name = tensor("lora_out_75_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46976832)))]; + tensor lora_out_75_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_5154, groups = var_5104, pad = lora_out_73_pad_0, pad_type = lora_out_73_pad_type_0, strides = var_5152, weight = lora_out_75_weight_0_to_fp16, x = input_61_cast_fp16)[name = tensor("lora_out_75_cast_fp16")]; + tensor query_7_cast_fp16 = add(x = pretrained_out_37_cast_fp16, y = lora_out_75_cast_fp16)[name = tensor("query_7_cast_fp16")]; + tensor var_5164 = const()[name = tensor("op_5164"), val = tensor([1, 1])]; + tensor var_5166 = const()[name = tensor("op_5166"), val = tensor([1, 1])]; + tensor pretrained_out_39_pad_type_0 = const()[name = tensor("pretrained_out_39_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_39_pad_0 = const()[name = tensor("pretrained_out_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47017856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47837120))), name = tensor("layers_3_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_39_cast_fp16 = conv(dilations = var_5166, groups = var_5104, pad = pretrained_out_39_pad_0, pad_type = pretrained_out_39_pad_type_0, strides = var_5164, weight = layers_3_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor("pretrained_out_39_cast_fp16")]; + tensor var_5170 = const()[name = tensor("op_5170"), val = tensor([1, 1])]; + tensor var_5172 = const()[name = tensor("op_5172"), val = tensor([1, 1])]; + tensor input_63_pad_type_0 = const()[name = tensor("input_63_pad_type_0"), val = tensor("custom")]; + tensor input_63_pad_0 = const()[name = tensor("input_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47837248)))]; + tensor input_63_cast_fp16 = conv(dilations = var_5172, groups = var_5104, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_5170, weight = layers_3_self_attn_k_proj_loraA_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("input_63_cast_fp16")]; + tensor var_5176 = const()[name = tensor("op_5176"), val = tensor([1, 1])]; + tensor var_5178 = const()[name = tensor("op_5178"), val = tensor([1, 1])]; + tensor lora_out_77_pad_type_0 = const()[name = tensor("lora_out_77_pad_type_0"), val = tensor("custom")]; + tensor lora_out_77_pad_0 = const()[name = tensor("lora_out_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_79_weight_0_to_fp16 = const()[name = tensor("lora_out_79_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47878272)))]; + tensor lora_out_79_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_5178, groups = var_5104, pad = lora_out_77_pad_0, pad_type = lora_out_77_pad_type_0, strides = var_5176, weight = lora_out_79_weight_0_to_fp16, x = input_63_cast_fp16)[name = tensor("lora_out_79_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = pretrained_out_39_cast_fp16, y = lora_out_79_cast_fp16)[name = tensor("key_7_cast_fp16")]; + tensor var_5189 = const()[name = tensor("op_5189"), val = tensor([1, 1])]; + tensor var_5191 = const()[name = tensor("op_5191"), val = tensor([1, 1])]; + tensor pretrained_out_41_pad_type_0 = const()[name = tensor("pretrained_out_41_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_41_pad_0 = const()[name = tensor("pretrained_out_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47919296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48738560))), name = tensor("layers_3_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_3_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48738688)))]; + tensor pretrained_out_41_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_5191, groups = var_5104, pad = pretrained_out_41_pad_0, pad_type = pretrained_out_41_pad_type_0, strides = var_5189, weight = layers_3_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor("pretrained_out_41_cast_fp16")]; + tensor var_5195 = const()[name = tensor("op_5195"), val = tensor([1, 1])]; + tensor var_5197 = const()[name = tensor("op_5197"), val = tensor([1, 1])]; + tensor input_65_pad_type_0 = const()[name = tensor("input_65_pad_type_0"), val = tensor("custom")]; + tensor input_65_pad_0 = const()[name = tensor("input_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48741312)))]; + tensor input_65_cast_fp16 = conv(dilations = var_5197, groups = var_5104, pad = input_65_pad_0, pad_type = input_65_pad_type_0, strides = var_5195, weight = layers_3_self_attn_v_proj_loraA_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor var_5201 = const()[name = tensor("op_5201"), val = tensor([1, 1])]; + tensor var_5203 = const()[name = tensor("op_5203"), val = tensor([1, 1])]; + tensor lora_out_81_pad_type_0 = const()[name = tensor("lora_out_81_pad_type_0"), val = tensor("custom")]; + tensor lora_out_81_pad_0 = const()[name = tensor("lora_out_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_83_weight_0_to_fp16 = const()[name = tensor("lora_out_83_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48782336)))]; + tensor lora_out_83_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_5203, groups = var_5104, pad = lora_out_81_pad_0, pad_type = lora_out_81_pad_type_0, strides = var_5201, weight = lora_out_83_weight_0_to_fp16, x = input_65_cast_fp16)[name = tensor("lora_out_83_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = pretrained_out_41_cast_fp16, y = lora_out_83_cast_fp16)[name = tensor("value_7_cast_fp16")]; + tensor var_5213_begin_0 = const()[name = tensor("op_5213_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5213_end_0 = const()[name = tensor("op_5213_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5213_end_mask_0 = const()[name = tensor("op_5213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5213_cast_fp16 = slice_by_index(begin = var_5213_begin_0, end = var_5213_end_0, end_mask = var_5213_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5213_cast_fp16")]; + tensor var_5217_begin_0 = const()[name = tensor("op_5217_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5217_end_0 = const()[name = tensor("op_5217_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5217_end_mask_0 = const()[name = tensor("op_5217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5217_cast_fp16 = slice_by_index(begin = var_5217_begin_0, end = var_5217_end_0, end_mask = var_5217_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5217_cast_fp16")]; + tensor var_5221_begin_0 = const()[name = tensor("op_5221_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5221_end_0 = const()[name = tensor("op_5221_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5221_end_mask_0 = const()[name = tensor("op_5221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5221_cast_fp16 = slice_by_index(begin = var_5221_begin_0, end = var_5221_end_0, end_mask = var_5221_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5221_cast_fp16")]; + tensor var_5225_begin_0 = const()[name = tensor("op_5225_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5225_end_0 = const()[name = tensor("op_5225_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5225_end_mask_0 = const()[name = tensor("op_5225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5225_cast_fp16 = slice_by_index(begin = var_5225_begin_0, end = var_5225_end_0, end_mask = var_5225_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5225_cast_fp16")]; + tensor var_5229_begin_0 = const()[name = tensor("op_5229_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5229_end_0 = const()[name = tensor("op_5229_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5229_end_mask_0 = const()[name = tensor("op_5229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5229_cast_fp16 = slice_by_index(begin = var_5229_begin_0, end = var_5229_end_0, end_mask = var_5229_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5229_cast_fp16")]; + tensor var_5233_begin_0 = const()[name = tensor("op_5233_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5233_end_0 = const()[name = tensor("op_5233_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5233_end_mask_0 = const()[name = tensor("op_5233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5233_cast_fp16 = slice_by_index(begin = var_5233_begin_0, end = var_5233_end_0, end_mask = var_5233_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5233_cast_fp16")]; + tensor var_5237_begin_0 = const()[name = tensor("op_5237_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5237_end_0 = const()[name = tensor("op_5237_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5237_end_mask_0 = const()[name = tensor("op_5237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5237_cast_fp16 = slice_by_index(begin = var_5237_begin_0, end = var_5237_end_0, end_mask = var_5237_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5237_cast_fp16")]; + tensor var_5241_begin_0 = const()[name = tensor("op_5241_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5241_end_0 = const()[name = tensor("op_5241_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5241_end_mask_0 = const()[name = tensor("op_5241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5241_cast_fp16 = slice_by_index(begin = var_5241_begin_0, end = var_5241_end_0, end_mask = var_5241_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5241_cast_fp16")]; + tensor var_5245_begin_0 = const()[name = tensor("op_5245_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5245_end_0 = const()[name = tensor("op_5245_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5245_end_mask_0 = const()[name = tensor("op_5245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5245_cast_fp16 = slice_by_index(begin = var_5245_begin_0, end = var_5245_end_0, end_mask = var_5245_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5245_cast_fp16")]; + tensor var_5249_begin_0 = const()[name = tensor("op_5249_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5249_end_0 = const()[name = tensor("op_5249_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5249_end_mask_0 = const()[name = tensor("op_5249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5249_cast_fp16 = slice_by_index(begin = var_5249_begin_0, end = var_5249_end_0, end_mask = var_5249_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5249_cast_fp16")]; + tensor var_5253_begin_0 = const()[name = tensor("op_5253_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5253_end_0 = const()[name = tensor("op_5253_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5253_end_mask_0 = const()[name = tensor("op_5253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5253_cast_fp16 = slice_by_index(begin = var_5253_begin_0, end = var_5253_end_0, end_mask = var_5253_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5253_cast_fp16")]; + tensor var_5257_begin_0 = const()[name = tensor("op_5257_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5257_end_0 = const()[name = tensor("op_5257_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5257_end_mask_0 = const()[name = tensor("op_5257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5257_cast_fp16 = slice_by_index(begin = var_5257_begin_0, end = var_5257_end_0, end_mask = var_5257_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5257_cast_fp16")]; + tensor var_5261_begin_0 = const()[name = tensor("op_5261_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5261_end_0 = const()[name = tensor("op_5261_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_5261_end_mask_0 = const()[name = tensor("op_5261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5261_cast_fp16 = slice_by_index(begin = var_5261_begin_0, end = var_5261_end_0, end_mask = var_5261_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5261_cast_fp16")]; + tensor var_5265_begin_0 = const()[name = tensor("op_5265_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5265_end_0 = const()[name = tensor("op_5265_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_5265_end_mask_0 = const()[name = tensor("op_5265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5265_cast_fp16 = slice_by_index(begin = var_5265_begin_0, end = var_5265_end_0, end_mask = var_5265_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5265_cast_fp16")]; + tensor var_5269_begin_0 = const()[name = tensor("op_5269_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5269_end_0 = const()[name = tensor("op_5269_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_5269_end_mask_0 = const()[name = tensor("op_5269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5269_cast_fp16 = slice_by_index(begin = var_5269_begin_0, end = var_5269_end_0, end_mask = var_5269_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5269_cast_fp16")]; + tensor var_5273_begin_0 = const()[name = tensor("op_5273_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5273_end_0 = const()[name = tensor("op_5273_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_5273_end_mask_0 = const()[name = tensor("op_5273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5273_cast_fp16 = slice_by_index(begin = var_5273_begin_0, end = var_5273_end_0, end_mask = var_5273_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5273_cast_fp16")]; + tensor var_5277_begin_0 = const()[name = tensor("op_5277_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5277_end_0 = const()[name = tensor("op_5277_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_5277_end_mask_0 = const()[name = tensor("op_5277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5277_cast_fp16 = slice_by_index(begin = var_5277_begin_0, end = var_5277_end_0, end_mask = var_5277_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5277_cast_fp16")]; + tensor var_5281_begin_0 = const()[name = tensor("op_5281_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_5281_end_0 = const()[name = tensor("op_5281_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_5281_end_mask_0 = const()[name = tensor("op_5281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5281_cast_fp16 = slice_by_index(begin = var_5281_begin_0, end = var_5281_end_0, end_mask = var_5281_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5281_cast_fp16")]; + tensor var_5285_begin_0 = const()[name = tensor("op_5285_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_5285_end_0 = const()[name = tensor("op_5285_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_5285_end_mask_0 = const()[name = tensor("op_5285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5285_cast_fp16 = slice_by_index(begin = var_5285_begin_0, end = var_5285_end_0, end_mask = var_5285_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5285_cast_fp16")]; + tensor var_5289_begin_0 = const()[name = tensor("op_5289_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_5289_end_0 = const()[name = tensor("op_5289_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_5289_end_mask_0 = const()[name = tensor("op_5289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5289_cast_fp16 = slice_by_index(begin = var_5289_begin_0, end = var_5289_end_0, end_mask = var_5289_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_5289_cast_fp16")]; + tensor var_5298_begin_0 = const()[name = tensor("op_5298_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5298_end_0 = const()[name = tensor("op_5298_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5298_end_mask_0 = const()[name = tensor("op_5298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5298_cast_fp16 = slice_by_index(begin = var_5298_begin_0, end = var_5298_end_0, end_mask = var_5298_end_mask_0, x = var_5213_cast_fp16)[name = tensor("op_5298_cast_fp16")]; + tensor var_5305_begin_0 = const()[name = tensor("op_5305_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5305_end_0 = const()[name = tensor("op_5305_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5305_end_mask_0 = const()[name = tensor("op_5305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5305_cast_fp16 = slice_by_index(begin = var_5305_begin_0, end = var_5305_end_0, end_mask = var_5305_end_mask_0, x = var_5213_cast_fp16)[name = tensor("op_5305_cast_fp16")]; + tensor var_5312_begin_0 = const()[name = tensor("op_5312_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5312_end_0 = const()[name = tensor("op_5312_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5312_end_mask_0 = const()[name = tensor("op_5312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5312_cast_fp16 = slice_by_index(begin = var_5312_begin_0, end = var_5312_end_0, end_mask = var_5312_end_mask_0, x = var_5213_cast_fp16)[name = tensor("op_5312_cast_fp16")]; + tensor var_5319_begin_0 = const()[name = tensor("op_5319_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5319_end_0 = const()[name = tensor("op_5319_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5319_end_mask_0 = const()[name = tensor("op_5319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5319_cast_fp16 = slice_by_index(begin = var_5319_begin_0, end = var_5319_end_0, end_mask = var_5319_end_mask_0, x = var_5213_cast_fp16)[name = tensor("op_5319_cast_fp16")]; + tensor var_5326_begin_0 = const()[name = tensor("op_5326_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5326_end_0 = const()[name = tensor("op_5326_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5326_end_mask_0 = const()[name = tensor("op_5326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5326_cast_fp16 = slice_by_index(begin = var_5326_begin_0, end = var_5326_end_0, end_mask = var_5326_end_mask_0, x = var_5217_cast_fp16)[name = tensor("op_5326_cast_fp16")]; + tensor var_5333_begin_0 = const()[name = tensor("op_5333_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5333_end_0 = const()[name = tensor("op_5333_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5333_end_mask_0 = const()[name = tensor("op_5333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5333_cast_fp16 = slice_by_index(begin = var_5333_begin_0, end = var_5333_end_0, end_mask = var_5333_end_mask_0, x = var_5217_cast_fp16)[name = tensor("op_5333_cast_fp16")]; + tensor var_5340_begin_0 = const()[name = tensor("op_5340_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5340_end_0 = const()[name = tensor("op_5340_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5340_end_mask_0 = const()[name = tensor("op_5340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5340_cast_fp16 = slice_by_index(begin = var_5340_begin_0, end = var_5340_end_0, end_mask = var_5340_end_mask_0, x = var_5217_cast_fp16)[name = tensor("op_5340_cast_fp16")]; + tensor var_5347_begin_0 = const()[name = tensor("op_5347_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5347_end_0 = const()[name = tensor("op_5347_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5347_end_mask_0 = const()[name = tensor("op_5347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5347_cast_fp16 = slice_by_index(begin = var_5347_begin_0, end = var_5347_end_0, end_mask = var_5347_end_mask_0, x = var_5217_cast_fp16)[name = tensor("op_5347_cast_fp16")]; + tensor var_5354_begin_0 = const()[name = tensor("op_5354_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5354_end_0 = const()[name = tensor("op_5354_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5354_end_mask_0 = const()[name = tensor("op_5354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5354_cast_fp16 = slice_by_index(begin = var_5354_begin_0, end = var_5354_end_0, end_mask = var_5354_end_mask_0, x = var_5221_cast_fp16)[name = tensor("op_5354_cast_fp16")]; + tensor var_5361_begin_0 = const()[name = tensor("op_5361_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5361_end_0 = const()[name = tensor("op_5361_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5361_end_mask_0 = const()[name = tensor("op_5361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5361_cast_fp16 = slice_by_index(begin = var_5361_begin_0, end = var_5361_end_0, end_mask = var_5361_end_mask_0, x = var_5221_cast_fp16)[name = tensor("op_5361_cast_fp16")]; + tensor var_5368_begin_0 = const()[name = tensor("op_5368_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5368_end_0 = const()[name = tensor("op_5368_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5368_end_mask_0 = const()[name = tensor("op_5368_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5368_cast_fp16 = slice_by_index(begin = var_5368_begin_0, end = var_5368_end_0, end_mask = var_5368_end_mask_0, x = var_5221_cast_fp16)[name = tensor("op_5368_cast_fp16")]; + tensor var_5375_begin_0 = const()[name = tensor("op_5375_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5375_end_0 = const()[name = tensor("op_5375_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5375_end_mask_0 = const()[name = tensor("op_5375_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5375_cast_fp16 = slice_by_index(begin = var_5375_begin_0, end = var_5375_end_0, end_mask = var_5375_end_mask_0, x = var_5221_cast_fp16)[name = tensor("op_5375_cast_fp16")]; + tensor var_5382_begin_0 = const()[name = tensor("op_5382_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5382_end_0 = const()[name = tensor("op_5382_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5382_end_mask_0 = const()[name = tensor("op_5382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5382_cast_fp16 = slice_by_index(begin = var_5382_begin_0, end = var_5382_end_0, end_mask = var_5382_end_mask_0, x = var_5225_cast_fp16)[name = tensor("op_5382_cast_fp16")]; + tensor var_5389_begin_0 = const()[name = tensor("op_5389_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5389_end_0 = const()[name = tensor("op_5389_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5389_end_mask_0 = const()[name = tensor("op_5389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5389_cast_fp16 = slice_by_index(begin = var_5389_begin_0, end = var_5389_end_0, end_mask = var_5389_end_mask_0, x = var_5225_cast_fp16)[name = tensor("op_5389_cast_fp16")]; + tensor var_5396_begin_0 = const()[name = tensor("op_5396_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5396_end_0 = const()[name = tensor("op_5396_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5396_end_mask_0 = const()[name = tensor("op_5396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5396_cast_fp16 = slice_by_index(begin = var_5396_begin_0, end = var_5396_end_0, end_mask = var_5396_end_mask_0, x = var_5225_cast_fp16)[name = tensor("op_5396_cast_fp16")]; + tensor var_5403_begin_0 = const()[name = tensor("op_5403_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5403_end_0 = const()[name = tensor("op_5403_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5403_end_mask_0 = const()[name = tensor("op_5403_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5403_cast_fp16 = slice_by_index(begin = var_5403_begin_0, end = var_5403_end_0, end_mask = var_5403_end_mask_0, x = var_5225_cast_fp16)[name = tensor("op_5403_cast_fp16")]; + tensor var_5410_begin_0 = const()[name = tensor("op_5410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5410_end_0 = const()[name = tensor("op_5410_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5410_end_mask_0 = const()[name = tensor("op_5410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5410_cast_fp16 = slice_by_index(begin = var_5410_begin_0, end = var_5410_end_0, end_mask = var_5410_end_mask_0, x = var_5229_cast_fp16)[name = tensor("op_5410_cast_fp16")]; + tensor var_5417_begin_0 = const()[name = tensor("op_5417_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5417_end_0 = const()[name = tensor("op_5417_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5417_end_mask_0 = const()[name = tensor("op_5417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5417_cast_fp16 = slice_by_index(begin = var_5417_begin_0, end = var_5417_end_0, end_mask = var_5417_end_mask_0, x = var_5229_cast_fp16)[name = tensor("op_5417_cast_fp16")]; + tensor var_5424_begin_0 = const()[name = tensor("op_5424_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5424_end_0 = const()[name = tensor("op_5424_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5424_end_mask_0 = const()[name = tensor("op_5424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5424_cast_fp16 = slice_by_index(begin = var_5424_begin_0, end = var_5424_end_0, end_mask = var_5424_end_mask_0, x = var_5229_cast_fp16)[name = tensor("op_5424_cast_fp16")]; + tensor var_5431_begin_0 = const()[name = tensor("op_5431_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5431_end_0 = const()[name = tensor("op_5431_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5431_end_mask_0 = const()[name = tensor("op_5431_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5431_cast_fp16 = slice_by_index(begin = var_5431_begin_0, end = var_5431_end_0, end_mask = var_5431_end_mask_0, x = var_5229_cast_fp16)[name = tensor("op_5431_cast_fp16")]; + tensor var_5438_begin_0 = const()[name = tensor("op_5438_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5438_end_0 = const()[name = tensor("op_5438_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5438_end_mask_0 = const()[name = tensor("op_5438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5438_cast_fp16 = slice_by_index(begin = var_5438_begin_0, end = var_5438_end_0, end_mask = var_5438_end_mask_0, x = var_5233_cast_fp16)[name = tensor("op_5438_cast_fp16")]; + tensor var_5445_begin_0 = const()[name = tensor("op_5445_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5445_end_0 = const()[name = tensor("op_5445_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5445_end_mask_0 = const()[name = tensor("op_5445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5445_cast_fp16 = slice_by_index(begin = var_5445_begin_0, end = var_5445_end_0, end_mask = var_5445_end_mask_0, x = var_5233_cast_fp16)[name = tensor("op_5445_cast_fp16")]; + tensor var_5452_begin_0 = const()[name = tensor("op_5452_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5452_end_0 = const()[name = tensor("op_5452_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5452_end_mask_0 = const()[name = tensor("op_5452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5452_cast_fp16 = slice_by_index(begin = var_5452_begin_0, end = var_5452_end_0, end_mask = var_5452_end_mask_0, x = var_5233_cast_fp16)[name = tensor("op_5452_cast_fp16")]; + tensor var_5459_begin_0 = const()[name = tensor("op_5459_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5459_end_0 = const()[name = tensor("op_5459_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5459_end_mask_0 = const()[name = tensor("op_5459_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5459_cast_fp16 = slice_by_index(begin = var_5459_begin_0, end = var_5459_end_0, end_mask = var_5459_end_mask_0, x = var_5233_cast_fp16)[name = tensor("op_5459_cast_fp16")]; + tensor var_5466_begin_0 = const()[name = tensor("op_5466_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5466_end_0 = const()[name = tensor("op_5466_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5466_end_mask_0 = const()[name = tensor("op_5466_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5466_cast_fp16 = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = var_5237_cast_fp16)[name = tensor("op_5466_cast_fp16")]; + tensor var_5473_begin_0 = const()[name = tensor("op_5473_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5473_end_0 = const()[name = tensor("op_5473_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5473_end_mask_0 = const()[name = tensor("op_5473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5473_cast_fp16 = slice_by_index(begin = var_5473_begin_0, end = var_5473_end_0, end_mask = var_5473_end_mask_0, x = var_5237_cast_fp16)[name = tensor("op_5473_cast_fp16")]; + tensor var_5480_begin_0 = const()[name = tensor("op_5480_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5480_end_0 = const()[name = tensor("op_5480_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5480_end_mask_0 = const()[name = tensor("op_5480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5480_cast_fp16 = slice_by_index(begin = var_5480_begin_0, end = var_5480_end_0, end_mask = var_5480_end_mask_0, x = var_5237_cast_fp16)[name = tensor("op_5480_cast_fp16")]; + tensor var_5487_begin_0 = const()[name = tensor("op_5487_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5487_end_0 = const()[name = tensor("op_5487_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5487_end_mask_0 = const()[name = tensor("op_5487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5487_cast_fp16 = slice_by_index(begin = var_5487_begin_0, end = var_5487_end_0, end_mask = var_5487_end_mask_0, x = var_5237_cast_fp16)[name = tensor("op_5487_cast_fp16")]; + tensor var_5494_begin_0 = const()[name = tensor("op_5494_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5494_end_0 = const()[name = tensor("op_5494_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5494_end_mask_0 = const()[name = tensor("op_5494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5494_cast_fp16 = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = var_5241_cast_fp16)[name = tensor("op_5494_cast_fp16")]; + tensor var_5501_begin_0 = const()[name = tensor("op_5501_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5501_end_0 = const()[name = tensor("op_5501_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5501_end_mask_0 = const()[name = tensor("op_5501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5501_cast_fp16 = slice_by_index(begin = var_5501_begin_0, end = var_5501_end_0, end_mask = var_5501_end_mask_0, x = var_5241_cast_fp16)[name = tensor("op_5501_cast_fp16")]; + tensor var_5508_begin_0 = const()[name = tensor("op_5508_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5508_end_0 = const()[name = tensor("op_5508_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5508_end_mask_0 = const()[name = tensor("op_5508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5508_cast_fp16 = slice_by_index(begin = var_5508_begin_0, end = var_5508_end_0, end_mask = var_5508_end_mask_0, x = var_5241_cast_fp16)[name = tensor("op_5508_cast_fp16")]; + tensor var_5515_begin_0 = const()[name = tensor("op_5515_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5515_end_0 = const()[name = tensor("op_5515_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5515_end_mask_0 = const()[name = tensor("op_5515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5515_cast_fp16 = slice_by_index(begin = var_5515_begin_0, end = var_5515_end_0, end_mask = var_5515_end_mask_0, x = var_5241_cast_fp16)[name = tensor("op_5515_cast_fp16")]; + tensor var_5522_begin_0 = const()[name = tensor("op_5522_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5522_end_0 = const()[name = tensor("op_5522_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5522_end_mask_0 = const()[name = tensor("op_5522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5522_cast_fp16 = slice_by_index(begin = var_5522_begin_0, end = var_5522_end_0, end_mask = var_5522_end_mask_0, x = var_5245_cast_fp16)[name = tensor("op_5522_cast_fp16")]; + tensor var_5529_begin_0 = const()[name = tensor("op_5529_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5529_end_0 = const()[name = tensor("op_5529_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5529_end_mask_0 = const()[name = tensor("op_5529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5529_cast_fp16 = slice_by_index(begin = var_5529_begin_0, end = var_5529_end_0, end_mask = var_5529_end_mask_0, x = var_5245_cast_fp16)[name = tensor("op_5529_cast_fp16")]; + tensor var_5536_begin_0 = const()[name = tensor("op_5536_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5536_end_0 = const()[name = tensor("op_5536_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5536_end_mask_0 = const()[name = tensor("op_5536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5536_cast_fp16 = slice_by_index(begin = var_5536_begin_0, end = var_5536_end_0, end_mask = var_5536_end_mask_0, x = var_5245_cast_fp16)[name = tensor("op_5536_cast_fp16")]; + tensor var_5543_begin_0 = const()[name = tensor("op_5543_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5543_end_0 = const()[name = tensor("op_5543_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5543_end_mask_0 = const()[name = tensor("op_5543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5543_cast_fp16 = slice_by_index(begin = var_5543_begin_0, end = var_5543_end_0, end_mask = var_5543_end_mask_0, x = var_5245_cast_fp16)[name = tensor("op_5543_cast_fp16")]; + tensor var_5550_begin_0 = const()[name = tensor("op_5550_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5550_end_0 = const()[name = tensor("op_5550_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5550_end_mask_0 = const()[name = tensor("op_5550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5550_cast_fp16 = slice_by_index(begin = var_5550_begin_0, end = var_5550_end_0, end_mask = var_5550_end_mask_0, x = var_5249_cast_fp16)[name = tensor("op_5550_cast_fp16")]; + tensor var_5557_begin_0 = const()[name = tensor("op_5557_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5557_end_0 = const()[name = tensor("op_5557_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5557_end_mask_0 = const()[name = tensor("op_5557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5557_cast_fp16 = slice_by_index(begin = var_5557_begin_0, end = var_5557_end_0, end_mask = var_5557_end_mask_0, x = var_5249_cast_fp16)[name = tensor("op_5557_cast_fp16")]; + tensor var_5564_begin_0 = const()[name = tensor("op_5564_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5564_end_0 = const()[name = tensor("op_5564_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5564_end_mask_0 = const()[name = tensor("op_5564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5564_cast_fp16 = slice_by_index(begin = var_5564_begin_0, end = var_5564_end_0, end_mask = var_5564_end_mask_0, x = var_5249_cast_fp16)[name = tensor("op_5564_cast_fp16")]; + tensor var_5571_begin_0 = const()[name = tensor("op_5571_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5571_end_0 = const()[name = tensor("op_5571_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5571_end_mask_0 = const()[name = tensor("op_5571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5571_cast_fp16 = slice_by_index(begin = var_5571_begin_0, end = var_5571_end_0, end_mask = var_5571_end_mask_0, x = var_5249_cast_fp16)[name = tensor("op_5571_cast_fp16")]; + tensor var_5578_begin_0 = const()[name = tensor("op_5578_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5578_end_0 = const()[name = tensor("op_5578_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5578_end_mask_0 = const()[name = tensor("op_5578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5578_cast_fp16 = slice_by_index(begin = var_5578_begin_0, end = var_5578_end_0, end_mask = var_5578_end_mask_0, x = var_5253_cast_fp16)[name = tensor("op_5578_cast_fp16")]; + tensor var_5585_begin_0 = const()[name = tensor("op_5585_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5585_end_0 = const()[name = tensor("op_5585_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5585_end_mask_0 = const()[name = tensor("op_5585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5585_cast_fp16 = slice_by_index(begin = var_5585_begin_0, end = var_5585_end_0, end_mask = var_5585_end_mask_0, x = var_5253_cast_fp16)[name = tensor("op_5585_cast_fp16")]; + tensor var_5592_begin_0 = const()[name = tensor("op_5592_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5592_end_0 = const()[name = tensor("op_5592_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5592_end_mask_0 = const()[name = tensor("op_5592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5592_cast_fp16 = slice_by_index(begin = var_5592_begin_0, end = var_5592_end_0, end_mask = var_5592_end_mask_0, x = var_5253_cast_fp16)[name = tensor("op_5592_cast_fp16")]; + tensor var_5599_begin_0 = const()[name = tensor("op_5599_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5599_end_0 = const()[name = tensor("op_5599_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5599_end_mask_0 = const()[name = tensor("op_5599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5599_cast_fp16 = slice_by_index(begin = var_5599_begin_0, end = var_5599_end_0, end_mask = var_5599_end_mask_0, x = var_5253_cast_fp16)[name = tensor("op_5599_cast_fp16")]; + tensor var_5606_begin_0 = const()[name = tensor("op_5606_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5606_end_0 = const()[name = tensor("op_5606_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5606_end_mask_0 = const()[name = tensor("op_5606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5606_cast_fp16 = slice_by_index(begin = var_5606_begin_0, end = var_5606_end_0, end_mask = var_5606_end_mask_0, x = var_5257_cast_fp16)[name = tensor("op_5606_cast_fp16")]; + tensor var_5613_begin_0 = const()[name = tensor("op_5613_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5613_end_0 = const()[name = tensor("op_5613_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5613_end_mask_0 = const()[name = tensor("op_5613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5613_cast_fp16 = slice_by_index(begin = var_5613_begin_0, end = var_5613_end_0, end_mask = var_5613_end_mask_0, x = var_5257_cast_fp16)[name = tensor("op_5613_cast_fp16")]; + tensor var_5620_begin_0 = const()[name = tensor("op_5620_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5620_end_0 = const()[name = tensor("op_5620_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5620_end_mask_0 = const()[name = tensor("op_5620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5620_cast_fp16 = slice_by_index(begin = var_5620_begin_0, end = var_5620_end_0, end_mask = var_5620_end_mask_0, x = var_5257_cast_fp16)[name = tensor("op_5620_cast_fp16")]; + tensor var_5627_begin_0 = const()[name = tensor("op_5627_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5627_end_0 = const()[name = tensor("op_5627_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5627_end_mask_0 = const()[name = tensor("op_5627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5627_cast_fp16 = slice_by_index(begin = var_5627_begin_0, end = var_5627_end_0, end_mask = var_5627_end_mask_0, x = var_5257_cast_fp16)[name = tensor("op_5627_cast_fp16")]; + tensor var_5634_begin_0 = const()[name = tensor("op_5634_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5634_end_0 = const()[name = tensor("op_5634_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5634_end_mask_0 = const()[name = tensor("op_5634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5634_cast_fp16 = slice_by_index(begin = var_5634_begin_0, end = var_5634_end_0, end_mask = var_5634_end_mask_0, x = var_5261_cast_fp16)[name = tensor("op_5634_cast_fp16")]; + tensor var_5641_begin_0 = const()[name = tensor("op_5641_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5641_end_0 = const()[name = tensor("op_5641_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5641_end_mask_0 = const()[name = tensor("op_5641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5641_cast_fp16 = slice_by_index(begin = var_5641_begin_0, end = var_5641_end_0, end_mask = var_5641_end_mask_0, x = var_5261_cast_fp16)[name = tensor("op_5641_cast_fp16")]; + tensor var_5648_begin_0 = const()[name = tensor("op_5648_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5648_end_0 = const()[name = tensor("op_5648_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5648_end_mask_0 = const()[name = tensor("op_5648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5648_cast_fp16 = slice_by_index(begin = var_5648_begin_0, end = var_5648_end_0, end_mask = var_5648_end_mask_0, x = var_5261_cast_fp16)[name = tensor("op_5648_cast_fp16")]; + tensor var_5655_begin_0 = const()[name = tensor("op_5655_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5655_end_0 = const()[name = tensor("op_5655_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5655_end_mask_0 = const()[name = tensor("op_5655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5655_cast_fp16 = slice_by_index(begin = var_5655_begin_0, end = var_5655_end_0, end_mask = var_5655_end_mask_0, x = var_5261_cast_fp16)[name = tensor("op_5655_cast_fp16")]; + tensor var_5662_begin_0 = const()[name = tensor("op_5662_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5662_end_0 = const()[name = tensor("op_5662_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5662_end_mask_0 = const()[name = tensor("op_5662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5662_cast_fp16 = slice_by_index(begin = var_5662_begin_0, end = var_5662_end_0, end_mask = var_5662_end_mask_0, x = var_5265_cast_fp16)[name = tensor("op_5662_cast_fp16")]; + tensor var_5669_begin_0 = const()[name = tensor("op_5669_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5669_end_0 = const()[name = tensor("op_5669_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5669_end_mask_0 = const()[name = tensor("op_5669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5669_cast_fp16 = slice_by_index(begin = var_5669_begin_0, end = var_5669_end_0, end_mask = var_5669_end_mask_0, x = var_5265_cast_fp16)[name = tensor("op_5669_cast_fp16")]; + tensor var_5676_begin_0 = const()[name = tensor("op_5676_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5676_end_0 = const()[name = tensor("op_5676_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5676_end_mask_0 = const()[name = tensor("op_5676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5676_cast_fp16 = slice_by_index(begin = var_5676_begin_0, end = var_5676_end_0, end_mask = var_5676_end_mask_0, x = var_5265_cast_fp16)[name = tensor("op_5676_cast_fp16")]; + tensor var_5683_begin_0 = const()[name = tensor("op_5683_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5683_end_0 = const()[name = tensor("op_5683_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5683_end_mask_0 = const()[name = tensor("op_5683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5683_cast_fp16 = slice_by_index(begin = var_5683_begin_0, end = var_5683_end_0, end_mask = var_5683_end_mask_0, x = var_5265_cast_fp16)[name = tensor("op_5683_cast_fp16")]; + tensor var_5690_begin_0 = const()[name = tensor("op_5690_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5690_end_0 = const()[name = tensor("op_5690_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5690_end_mask_0 = const()[name = tensor("op_5690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5690_cast_fp16 = slice_by_index(begin = var_5690_begin_0, end = var_5690_end_0, end_mask = var_5690_end_mask_0, x = var_5269_cast_fp16)[name = tensor("op_5690_cast_fp16")]; + tensor var_5697_begin_0 = const()[name = tensor("op_5697_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5697_end_0 = const()[name = tensor("op_5697_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5697_end_mask_0 = const()[name = tensor("op_5697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5697_cast_fp16 = slice_by_index(begin = var_5697_begin_0, end = var_5697_end_0, end_mask = var_5697_end_mask_0, x = var_5269_cast_fp16)[name = tensor("op_5697_cast_fp16")]; + tensor var_5704_begin_0 = const()[name = tensor("op_5704_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5704_end_0 = const()[name = tensor("op_5704_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5704_end_mask_0 = const()[name = tensor("op_5704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5704_cast_fp16 = slice_by_index(begin = var_5704_begin_0, end = var_5704_end_0, end_mask = var_5704_end_mask_0, x = var_5269_cast_fp16)[name = tensor("op_5704_cast_fp16")]; + tensor var_5711_begin_0 = const()[name = tensor("op_5711_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5711_end_0 = const()[name = tensor("op_5711_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5711_end_mask_0 = const()[name = tensor("op_5711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5711_cast_fp16 = slice_by_index(begin = var_5711_begin_0, end = var_5711_end_0, end_mask = var_5711_end_mask_0, x = var_5269_cast_fp16)[name = tensor("op_5711_cast_fp16")]; + tensor var_5718_begin_0 = const()[name = tensor("op_5718_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5718_end_0 = const()[name = tensor("op_5718_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5718_end_mask_0 = const()[name = tensor("op_5718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5718_cast_fp16 = slice_by_index(begin = var_5718_begin_0, end = var_5718_end_0, end_mask = var_5718_end_mask_0, x = var_5273_cast_fp16)[name = tensor("op_5718_cast_fp16")]; + tensor var_5725_begin_0 = const()[name = tensor("op_5725_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5725_end_0 = const()[name = tensor("op_5725_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5725_end_mask_0 = const()[name = tensor("op_5725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5725_cast_fp16 = slice_by_index(begin = var_5725_begin_0, end = var_5725_end_0, end_mask = var_5725_end_mask_0, x = var_5273_cast_fp16)[name = tensor("op_5725_cast_fp16")]; + tensor var_5732_begin_0 = const()[name = tensor("op_5732_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5732_end_0 = const()[name = tensor("op_5732_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5732_end_mask_0 = const()[name = tensor("op_5732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5732_cast_fp16 = slice_by_index(begin = var_5732_begin_0, end = var_5732_end_0, end_mask = var_5732_end_mask_0, x = var_5273_cast_fp16)[name = tensor("op_5732_cast_fp16")]; + tensor var_5739_begin_0 = const()[name = tensor("op_5739_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5739_end_0 = const()[name = tensor("op_5739_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5739_end_mask_0 = const()[name = tensor("op_5739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5739_cast_fp16 = slice_by_index(begin = var_5739_begin_0, end = var_5739_end_0, end_mask = var_5739_end_mask_0, x = var_5273_cast_fp16)[name = tensor("op_5739_cast_fp16")]; + tensor var_5746_begin_0 = const()[name = tensor("op_5746_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5746_end_0 = const()[name = tensor("op_5746_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5746_end_mask_0 = const()[name = tensor("op_5746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5746_cast_fp16 = slice_by_index(begin = var_5746_begin_0, end = var_5746_end_0, end_mask = var_5746_end_mask_0, x = var_5277_cast_fp16)[name = tensor("op_5746_cast_fp16")]; + tensor var_5753_begin_0 = const()[name = tensor("op_5753_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5753_end_0 = const()[name = tensor("op_5753_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5753_end_mask_0 = const()[name = tensor("op_5753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5753_cast_fp16 = slice_by_index(begin = var_5753_begin_0, end = var_5753_end_0, end_mask = var_5753_end_mask_0, x = var_5277_cast_fp16)[name = tensor("op_5753_cast_fp16")]; + tensor var_5760_begin_0 = const()[name = tensor("op_5760_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5760_end_0 = const()[name = tensor("op_5760_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5760_end_mask_0 = const()[name = tensor("op_5760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5760_cast_fp16 = slice_by_index(begin = var_5760_begin_0, end = var_5760_end_0, end_mask = var_5760_end_mask_0, x = var_5277_cast_fp16)[name = tensor("op_5760_cast_fp16")]; + tensor var_5767_begin_0 = const()[name = tensor("op_5767_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5767_end_0 = const()[name = tensor("op_5767_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5767_end_mask_0 = const()[name = tensor("op_5767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5767_cast_fp16 = slice_by_index(begin = var_5767_begin_0, end = var_5767_end_0, end_mask = var_5767_end_mask_0, x = var_5277_cast_fp16)[name = tensor("op_5767_cast_fp16")]; + tensor var_5774_begin_0 = const()[name = tensor("op_5774_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5774_end_0 = const()[name = tensor("op_5774_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5774_end_mask_0 = const()[name = tensor("op_5774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5774_cast_fp16 = slice_by_index(begin = var_5774_begin_0, end = var_5774_end_0, end_mask = var_5774_end_mask_0, x = var_5281_cast_fp16)[name = tensor("op_5774_cast_fp16")]; + tensor var_5781_begin_0 = const()[name = tensor("op_5781_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5781_end_0 = const()[name = tensor("op_5781_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5781_end_mask_0 = const()[name = tensor("op_5781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5781_cast_fp16 = slice_by_index(begin = var_5781_begin_0, end = var_5781_end_0, end_mask = var_5781_end_mask_0, x = var_5281_cast_fp16)[name = tensor("op_5781_cast_fp16")]; + tensor var_5788_begin_0 = const()[name = tensor("op_5788_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5788_end_0 = const()[name = tensor("op_5788_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5788_end_mask_0 = const()[name = tensor("op_5788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5788_cast_fp16 = slice_by_index(begin = var_5788_begin_0, end = var_5788_end_0, end_mask = var_5788_end_mask_0, x = var_5281_cast_fp16)[name = tensor("op_5788_cast_fp16")]; + tensor var_5795_begin_0 = const()[name = tensor("op_5795_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5795_end_0 = const()[name = tensor("op_5795_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5795_end_mask_0 = const()[name = tensor("op_5795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5795_cast_fp16 = slice_by_index(begin = var_5795_begin_0, end = var_5795_end_0, end_mask = var_5795_end_mask_0, x = var_5281_cast_fp16)[name = tensor("op_5795_cast_fp16")]; + tensor var_5802_begin_0 = const()[name = tensor("op_5802_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5802_end_0 = const()[name = tensor("op_5802_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5802_end_mask_0 = const()[name = tensor("op_5802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5802_cast_fp16 = slice_by_index(begin = var_5802_begin_0, end = var_5802_end_0, end_mask = var_5802_end_mask_0, x = var_5285_cast_fp16)[name = tensor("op_5802_cast_fp16")]; + tensor var_5809_begin_0 = const()[name = tensor("op_5809_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5809_end_0 = const()[name = tensor("op_5809_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5809_end_mask_0 = const()[name = tensor("op_5809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5809_cast_fp16 = slice_by_index(begin = var_5809_begin_0, end = var_5809_end_0, end_mask = var_5809_end_mask_0, x = var_5285_cast_fp16)[name = tensor("op_5809_cast_fp16")]; + tensor var_5816_begin_0 = const()[name = tensor("op_5816_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5816_end_0 = const()[name = tensor("op_5816_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5816_end_mask_0 = const()[name = tensor("op_5816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5816_cast_fp16 = slice_by_index(begin = var_5816_begin_0, end = var_5816_end_0, end_mask = var_5816_end_mask_0, x = var_5285_cast_fp16)[name = tensor("op_5816_cast_fp16")]; + tensor var_5823_begin_0 = const()[name = tensor("op_5823_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5823_end_0 = const()[name = tensor("op_5823_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5823_end_mask_0 = const()[name = tensor("op_5823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5823_cast_fp16 = slice_by_index(begin = var_5823_begin_0, end = var_5823_end_0, end_mask = var_5823_end_mask_0, x = var_5285_cast_fp16)[name = tensor("op_5823_cast_fp16")]; + tensor var_5830_begin_0 = const()[name = tensor("op_5830_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5830_end_0 = const()[name = tensor("op_5830_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_5830_end_mask_0 = const()[name = tensor("op_5830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5830_cast_fp16 = slice_by_index(begin = var_5830_begin_0, end = var_5830_end_0, end_mask = var_5830_end_mask_0, x = var_5289_cast_fp16)[name = tensor("op_5830_cast_fp16")]; + tensor var_5837_begin_0 = const()[name = tensor("op_5837_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_5837_end_0 = const()[name = tensor("op_5837_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_5837_end_mask_0 = const()[name = tensor("op_5837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5837_cast_fp16 = slice_by_index(begin = var_5837_begin_0, end = var_5837_end_0, end_mask = var_5837_end_mask_0, x = var_5289_cast_fp16)[name = tensor("op_5837_cast_fp16")]; + tensor var_5844_begin_0 = const()[name = tensor("op_5844_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_5844_end_0 = const()[name = tensor("op_5844_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_5844_end_mask_0 = const()[name = tensor("op_5844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5844_cast_fp16 = slice_by_index(begin = var_5844_begin_0, end = var_5844_end_0, end_mask = var_5844_end_mask_0, x = var_5289_cast_fp16)[name = tensor("op_5844_cast_fp16")]; + tensor var_5851_begin_0 = const()[name = tensor("op_5851_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_5851_end_0 = const()[name = tensor("op_5851_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5851_end_mask_0 = const()[name = tensor("op_5851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5851_cast_fp16 = slice_by_index(begin = var_5851_begin_0, end = var_5851_end_0, end_mask = var_5851_end_mask_0, x = var_5289_cast_fp16)[name = tensor("op_5851_cast_fp16")]; + tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_5856_begin_0 = const()[name = tensor("op_5856_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5856_end_0 = const()[name = tensor("op_5856_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_5856_end_mask_0 = const()[name = tensor("op_5856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_28 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = tensor("transpose_28")]; + tensor var_5856_cast_fp16 = slice_by_index(begin = var_5856_begin_0, end = var_5856_end_0, end_mask = var_5856_end_mask_0, x = transpose_28)[name = tensor("op_5856_cast_fp16")]; + tensor var_5860_begin_0 = const()[name = tensor("op_5860_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_5860_end_0 = const()[name = tensor("op_5860_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_5860_end_mask_0 = const()[name = tensor("op_5860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5860_cast_fp16 = slice_by_index(begin = var_5860_begin_0, end = var_5860_end_0, end_mask = var_5860_end_mask_0, x = transpose_28)[name = tensor("op_5860_cast_fp16")]; + tensor var_5864_begin_0 = const()[name = tensor("op_5864_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_5864_end_0 = const()[name = tensor("op_5864_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_5864_end_mask_0 = const()[name = tensor("op_5864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5864_cast_fp16 = slice_by_index(begin = var_5864_begin_0, end = var_5864_end_0, end_mask = var_5864_end_mask_0, x = transpose_28)[name = tensor("op_5864_cast_fp16")]; + tensor var_5868_begin_0 = const()[name = tensor("op_5868_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_5868_end_0 = const()[name = tensor("op_5868_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_5868_end_mask_0 = const()[name = tensor("op_5868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5868_cast_fp16 = slice_by_index(begin = var_5868_begin_0, end = var_5868_end_0, end_mask = var_5868_end_mask_0, x = transpose_28)[name = tensor("op_5868_cast_fp16")]; + tensor var_5872_begin_0 = const()[name = tensor("op_5872_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_5872_end_0 = const()[name = tensor("op_5872_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_5872_end_mask_0 = const()[name = tensor("op_5872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5872_cast_fp16 = slice_by_index(begin = var_5872_begin_0, end = var_5872_end_0, end_mask = var_5872_end_mask_0, x = transpose_28)[name = tensor("op_5872_cast_fp16")]; + tensor var_5876_begin_0 = const()[name = tensor("op_5876_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_5876_end_0 = const()[name = tensor("op_5876_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_5876_end_mask_0 = const()[name = tensor("op_5876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5876_cast_fp16 = slice_by_index(begin = var_5876_begin_0, end = var_5876_end_0, end_mask = var_5876_end_mask_0, x = transpose_28)[name = tensor("op_5876_cast_fp16")]; + tensor var_5880_begin_0 = const()[name = tensor("op_5880_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_5880_end_0 = const()[name = tensor("op_5880_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_5880_end_mask_0 = const()[name = tensor("op_5880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5880_cast_fp16 = slice_by_index(begin = var_5880_begin_0, end = var_5880_end_0, end_mask = var_5880_end_mask_0, x = transpose_28)[name = tensor("op_5880_cast_fp16")]; + tensor var_5884_begin_0 = const()[name = tensor("op_5884_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_5884_end_0 = const()[name = tensor("op_5884_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_5884_end_mask_0 = const()[name = tensor("op_5884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5884_cast_fp16 = slice_by_index(begin = var_5884_begin_0, end = var_5884_end_0, end_mask = var_5884_end_mask_0, x = transpose_28)[name = tensor("op_5884_cast_fp16")]; + tensor var_5888_begin_0 = const()[name = tensor("op_5888_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_5888_end_0 = const()[name = tensor("op_5888_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_5888_end_mask_0 = const()[name = tensor("op_5888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5888_cast_fp16 = slice_by_index(begin = var_5888_begin_0, end = var_5888_end_0, end_mask = var_5888_end_mask_0, x = transpose_28)[name = tensor("op_5888_cast_fp16")]; + tensor var_5892_begin_0 = const()[name = tensor("op_5892_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_5892_end_0 = const()[name = tensor("op_5892_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_5892_end_mask_0 = const()[name = tensor("op_5892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5892_cast_fp16 = slice_by_index(begin = var_5892_begin_0, end = var_5892_end_0, end_mask = var_5892_end_mask_0, x = transpose_28)[name = tensor("op_5892_cast_fp16")]; + tensor var_5896_begin_0 = const()[name = tensor("op_5896_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_5896_end_0 = const()[name = tensor("op_5896_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_5896_end_mask_0 = const()[name = tensor("op_5896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5896_cast_fp16 = slice_by_index(begin = var_5896_begin_0, end = var_5896_end_0, end_mask = var_5896_end_mask_0, x = transpose_28)[name = tensor("op_5896_cast_fp16")]; + tensor var_5900_begin_0 = const()[name = tensor("op_5900_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_5900_end_0 = const()[name = tensor("op_5900_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_5900_end_mask_0 = const()[name = tensor("op_5900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5900_cast_fp16 = slice_by_index(begin = var_5900_begin_0, end = var_5900_end_0, end_mask = var_5900_end_mask_0, x = transpose_28)[name = tensor("op_5900_cast_fp16")]; + tensor var_5904_begin_0 = const()[name = tensor("op_5904_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_5904_end_0 = const()[name = tensor("op_5904_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_5904_end_mask_0 = const()[name = tensor("op_5904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5904_cast_fp16 = slice_by_index(begin = var_5904_begin_0, end = var_5904_end_0, end_mask = var_5904_end_mask_0, x = transpose_28)[name = tensor("op_5904_cast_fp16")]; + tensor var_5908_begin_0 = const()[name = tensor("op_5908_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_5908_end_0 = const()[name = tensor("op_5908_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_5908_end_mask_0 = const()[name = tensor("op_5908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5908_cast_fp16 = slice_by_index(begin = var_5908_begin_0, end = var_5908_end_0, end_mask = var_5908_end_mask_0, x = transpose_28)[name = tensor("op_5908_cast_fp16")]; + tensor var_5912_begin_0 = const()[name = tensor("op_5912_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_5912_end_0 = const()[name = tensor("op_5912_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_5912_end_mask_0 = const()[name = tensor("op_5912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5912_cast_fp16 = slice_by_index(begin = var_5912_begin_0, end = var_5912_end_0, end_mask = var_5912_end_mask_0, x = transpose_28)[name = tensor("op_5912_cast_fp16")]; + tensor var_5916_begin_0 = const()[name = tensor("op_5916_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_5916_end_0 = const()[name = tensor("op_5916_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_5916_end_mask_0 = const()[name = tensor("op_5916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5916_cast_fp16 = slice_by_index(begin = var_5916_begin_0, end = var_5916_end_0, end_mask = var_5916_end_mask_0, x = transpose_28)[name = tensor("op_5916_cast_fp16")]; + tensor var_5920_begin_0 = const()[name = tensor("op_5920_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_5920_end_0 = const()[name = tensor("op_5920_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_5920_end_mask_0 = const()[name = tensor("op_5920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5920_cast_fp16 = slice_by_index(begin = var_5920_begin_0, end = var_5920_end_0, end_mask = var_5920_end_mask_0, x = transpose_28)[name = tensor("op_5920_cast_fp16")]; + tensor var_5924_begin_0 = const()[name = tensor("op_5924_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_5924_end_0 = const()[name = tensor("op_5924_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_5924_end_mask_0 = const()[name = tensor("op_5924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5924_cast_fp16 = slice_by_index(begin = var_5924_begin_0, end = var_5924_end_0, end_mask = var_5924_end_mask_0, x = transpose_28)[name = tensor("op_5924_cast_fp16")]; + tensor var_5928_begin_0 = const()[name = tensor("op_5928_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_5928_end_0 = const()[name = tensor("op_5928_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_5928_end_mask_0 = const()[name = tensor("op_5928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5928_cast_fp16 = slice_by_index(begin = var_5928_begin_0, end = var_5928_end_0, end_mask = var_5928_end_mask_0, x = transpose_28)[name = tensor("op_5928_cast_fp16")]; + tensor var_5932_begin_0 = const()[name = tensor("op_5932_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_5932_end_0 = const()[name = tensor("op_5932_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_5932_end_mask_0 = const()[name = tensor("op_5932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5932_cast_fp16 = slice_by_index(begin = var_5932_begin_0, end = var_5932_end_0, end_mask = var_5932_end_mask_0, x = transpose_28)[name = tensor("op_5932_cast_fp16")]; + tensor var_5934_begin_0 = const()[name = tensor("op_5934_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5934_end_0 = const()[name = tensor("op_5934_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_5934_end_mask_0 = const()[name = tensor("op_5934_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5934_cast_fp16 = slice_by_index(begin = var_5934_begin_0, end = var_5934_end_0, end_mask = var_5934_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5934_cast_fp16")]; + tensor var_5938_begin_0 = const()[name = tensor("op_5938_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_5938_end_0 = const()[name = tensor("op_5938_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_5938_end_mask_0 = const()[name = tensor("op_5938_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5938_cast_fp16 = slice_by_index(begin = var_5938_begin_0, end = var_5938_end_0, end_mask = var_5938_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5938_cast_fp16")]; + tensor var_5942_begin_0 = const()[name = tensor("op_5942_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_5942_end_0 = const()[name = tensor("op_5942_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_5942_end_mask_0 = const()[name = tensor("op_5942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5942_cast_fp16 = slice_by_index(begin = var_5942_begin_0, end = var_5942_end_0, end_mask = var_5942_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5942_cast_fp16")]; + tensor var_5946_begin_0 = const()[name = tensor("op_5946_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_5946_end_0 = const()[name = tensor("op_5946_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_5946_end_mask_0 = const()[name = tensor("op_5946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5946_cast_fp16 = slice_by_index(begin = var_5946_begin_0, end = var_5946_end_0, end_mask = var_5946_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5946_cast_fp16")]; + tensor var_5950_begin_0 = const()[name = tensor("op_5950_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_5950_end_0 = const()[name = tensor("op_5950_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_5950_end_mask_0 = const()[name = tensor("op_5950_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5950_cast_fp16 = slice_by_index(begin = var_5950_begin_0, end = var_5950_end_0, end_mask = var_5950_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5950_cast_fp16")]; + tensor var_5954_begin_0 = const()[name = tensor("op_5954_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_5954_end_0 = const()[name = tensor("op_5954_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_5954_end_mask_0 = const()[name = tensor("op_5954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5954_cast_fp16 = slice_by_index(begin = var_5954_begin_0, end = var_5954_end_0, end_mask = var_5954_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5954_cast_fp16")]; + tensor var_5958_begin_0 = const()[name = tensor("op_5958_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_5958_end_0 = const()[name = tensor("op_5958_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_5958_end_mask_0 = const()[name = tensor("op_5958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5958_cast_fp16 = slice_by_index(begin = var_5958_begin_0, end = var_5958_end_0, end_mask = var_5958_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5958_cast_fp16")]; + tensor var_5962_begin_0 = const()[name = tensor("op_5962_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_5962_end_0 = const()[name = tensor("op_5962_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_5962_end_mask_0 = const()[name = tensor("op_5962_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5962_cast_fp16 = slice_by_index(begin = var_5962_begin_0, end = var_5962_end_0, end_mask = var_5962_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5962_cast_fp16")]; + tensor var_5966_begin_0 = const()[name = tensor("op_5966_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_5966_end_0 = const()[name = tensor("op_5966_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_5966_end_mask_0 = const()[name = tensor("op_5966_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5966_cast_fp16 = slice_by_index(begin = var_5966_begin_0, end = var_5966_end_0, end_mask = var_5966_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5966_cast_fp16")]; + tensor var_5970_begin_0 = const()[name = tensor("op_5970_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_5970_end_0 = const()[name = tensor("op_5970_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_5970_end_mask_0 = const()[name = tensor("op_5970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5970_cast_fp16 = slice_by_index(begin = var_5970_begin_0, end = var_5970_end_0, end_mask = var_5970_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5970_cast_fp16")]; + tensor var_5974_begin_0 = const()[name = tensor("op_5974_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_5974_end_0 = const()[name = tensor("op_5974_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_5974_end_mask_0 = const()[name = tensor("op_5974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5974_cast_fp16 = slice_by_index(begin = var_5974_begin_0, end = var_5974_end_0, end_mask = var_5974_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5974_cast_fp16")]; + tensor var_5978_begin_0 = const()[name = tensor("op_5978_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_5978_end_0 = const()[name = tensor("op_5978_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_5978_end_mask_0 = const()[name = tensor("op_5978_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5978_cast_fp16 = slice_by_index(begin = var_5978_begin_0, end = var_5978_end_0, end_mask = var_5978_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5978_cast_fp16")]; + tensor var_5982_begin_0 = const()[name = tensor("op_5982_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_5982_end_0 = const()[name = tensor("op_5982_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_5982_end_mask_0 = const()[name = tensor("op_5982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5982_cast_fp16 = slice_by_index(begin = var_5982_begin_0, end = var_5982_end_0, end_mask = var_5982_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5982_cast_fp16")]; + tensor var_5986_begin_0 = const()[name = tensor("op_5986_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_5986_end_0 = const()[name = tensor("op_5986_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_5986_end_mask_0 = const()[name = tensor("op_5986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5986_cast_fp16 = slice_by_index(begin = var_5986_begin_0, end = var_5986_end_0, end_mask = var_5986_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5986_cast_fp16")]; + tensor var_5990_begin_0 = const()[name = tensor("op_5990_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_5990_end_0 = const()[name = tensor("op_5990_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_5990_end_mask_0 = const()[name = tensor("op_5990_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5990_cast_fp16 = slice_by_index(begin = var_5990_begin_0, end = var_5990_end_0, end_mask = var_5990_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5990_cast_fp16")]; + tensor var_5994_begin_0 = const()[name = tensor("op_5994_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_5994_end_0 = const()[name = tensor("op_5994_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_5994_end_mask_0 = const()[name = tensor("op_5994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5994_cast_fp16 = slice_by_index(begin = var_5994_begin_0, end = var_5994_end_0, end_mask = var_5994_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5994_cast_fp16")]; + tensor var_5998_begin_0 = const()[name = tensor("op_5998_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_5998_end_0 = const()[name = tensor("op_5998_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_5998_end_mask_0 = const()[name = tensor("op_5998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_5998_cast_fp16 = slice_by_index(begin = var_5998_begin_0, end = var_5998_end_0, end_mask = var_5998_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_5998_cast_fp16")]; + tensor var_6002_begin_0 = const()[name = tensor("op_6002_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6002_end_0 = const()[name = tensor("op_6002_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_6002_end_mask_0 = const()[name = tensor("op_6002_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6002_cast_fp16 = slice_by_index(begin = var_6002_begin_0, end = var_6002_end_0, end_mask = var_6002_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_6002_cast_fp16")]; + tensor var_6006_begin_0 = const()[name = tensor("op_6006_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6006_end_0 = const()[name = tensor("op_6006_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_6006_end_mask_0 = const()[name = tensor("op_6006_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6006_cast_fp16 = slice_by_index(begin = var_6006_begin_0, end = var_6006_end_0, end_mask = var_6006_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_6006_cast_fp16")]; + tensor var_6010_begin_0 = const()[name = tensor("op_6010_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6010_end_0 = const()[name = tensor("op_6010_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_6010_end_mask_0 = const()[name = tensor("op_6010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6010_cast_fp16 = slice_by_index(begin = var_6010_begin_0, end = var_6010_end_0, end_mask = var_6010_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_6010_cast_fp16")]; + tensor var_6014_equation_0 = const()[name = tensor("op_6014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6014_cast_fp16 = einsum(equation = var_6014_equation_0, values = (var_5856_cast_fp16, var_5298_cast_fp16))[name = tensor("op_6014_cast_fp16")]; + tensor var_6015_to_fp16 = const()[name = tensor("op_6015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_481_cast_fp16 = mul(x = var_6014_cast_fp16, y = var_6015_to_fp16)[name = tensor("aw_chunk_481_cast_fp16")]; + tensor var_6018_equation_0 = const()[name = tensor("op_6018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6018_cast_fp16 = einsum(equation = var_6018_equation_0, values = (var_5856_cast_fp16, var_5305_cast_fp16))[name = tensor("op_6018_cast_fp16")]; + tensor var_6019_to_fp16 = const()[name = tensor("op_6019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_483_cast_fp16 = mul(x = var_6018_cast_fp16, y = var_6019_to_fp16)[name = tensor("aw_chunk_483_cast_fp16")]; + tensor var_6022_equation_0 = const()[name = tensor("op_6022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6022_cast_fp16 = einsum(equation = var_6022_equation_0, values = (var_5856_cast_fp16, var_5312_cast_fp16))[name = tensor("op_6022_cast_fp16")]; + tensor var_6023_to_fp16 = const()[name = tensor("op_6023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_485_cast_fp16 = mul(x = var_6022_cast_fp16, y = var_6023_to_fp16)[name = tensor("aw_chunk_485_cast_fp16")]; + tensor var_6026_equation_0 = const()[name = tensor("op_6026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6026_cast_fp16 = einsum(equation = var_6026_equation_0, values = (var_5856_cast_fp16, var_5319_cast_fp16))[name = tensor("op_6026_cast_fp16")]; + tensor var_6027_to_fp16 = const()[name = tensor("op_6027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_487_cast_fp16 = mul(x = var_6026_cast_fp16, y = var_6027_to_fp16)[name = tensor("aw_chunk_487_cast_fp16")]; + tensor var_6030_equation_0 = const()[name = tensor("op_6030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6030_cast_fp16 = einsum(equation = var_6030_equation_0, values = (var_5860_cast_fp16, var_5326_cast_fp16))[name = tensor("op_6030_cast_fp16")]; + tensor var_6031_to_fp16 = const()[name = tensor("op_6031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_489_cast_fp16 = mul(x = var_6030_cast_fp16, y = var_6031_to_fp16)[name = tensor("aw_chunk_489_cast_fp16")]; + tensor var_6034_equation_0 = const()[name = tensor("op_6034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6034_cast_fp16 = einsum(equation = var_6034_equation_0, values = (var_5860_cast_fp16, var_5333_cast_fp16))[name = tensor("op_6034_cast_fp16")]; + tensor var_6035_to_fp16 = const()[name = tensor("op_6035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_491_cast_fp16 = mul(x = var_6034_cast_fp16, y = var_6035_to_fp16)[name = tensor("aw_chunk_491_cast_fp16")]; + tensor var_6038_equation_0 = const()[name = tensor("op_6038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6038_cast_fp16 = einsum(equation = var_6038_equation_0, values = (var_5860_cast_fp16, var_5340_cast_fp16))[name = tensor("op_6038_cast_fp16")]; + tensor var_6039_to_fp16 = const()[name = tensor("op_6039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_493_cast_fp16 = mul(x = var_6038_cast_fp16, y = var_6039_to_fp16)[name = tensor("aw_chunk_493_cast_fp16")]; + tensor var_6042_equation_0 = const()[name = tensor("op_6042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6042_cast_fp16 = einsum(equation = var_6042_equation_0, values = (var_5860_cast_fp16, var_5347_cast_fp16))[name = tensor("op_6042_cast_fp16")]; + tensor var_6043_to_fp16 = const()[name = tensor("op_6043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_495_cast_fp16 = mul(x = var_6042_cast_fp16, y = var_6043_to_fp16)[name = tensor("aw_chunk_495_cast_fp16")]; + tensor var_6046_equation_0 = const()[name = tensor("op_6046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6046_cast_fp16 = einsum(equation = var_6046_equation_0, values = (var_5864_cast_fp16, var_5354_cast_fp16))[name = tensor("op_6046_cast_fp16")]; + tensor var_6047_to_fp16 = const()[name = tensor("op_6047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_497_cast_fp16 = mul(x = var_6046_cast_fp16, y = var_6047_to_fp16)[name = tensor("aw_chunk_497_cast_fp16")]; + tensor var_6050_equation_0 = const()[name = tensor("op_6050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6050_cast_fp16 = einsum(equation = var_6050_equation_0, values = (var_5864_cast_fp16, var_5361_cast_fp16))[name = tensor("op_6050_cast_fp16")]; + tensor var_6051_to_fp16 = const()[name = tensor("op_6051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_499_cast_fp16 = mul(x = var_6050_cast_fp16, y = var_6051_to_fp16)[name = tensor("aw_chunk_499_cast_fp16")]; + tensor var_6054_equation_0 = const()[name = tensor("op_6054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6054_cast_fp16 = einsum(equation = var_6054_equation_0, values = (var_5864_cast_fp16, var_5368_cast_fp16))[name = tensor("op_6054_cast_fp16")]; + tensor var_6055_to_fp16 = const()[name = tensor("op_6055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_501_cast_fp16 = mul(x = var_6054_cast_fp16, y = var_6055_to_fp16)[name = tensor("aw_chunk_501_cast_fp16")]; + tensor var_6058_equation_0 = const()[name = tensor("op_6058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6058_cast_fp16 = einsum(equation = var_6058_equation_0, values = (var_5864_cast_fp16, var_5375_cast_fp16))[name = tensor("op_6058_cast_fp16")]; + tensor var_6059_to_fp16 = const()[name = tensor("op_6059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_503_cast_fp16 = mul(x = var_6058_cast_fp16, y = var_6059_to_fp16)[name = tensor("aw_chunk_503_cast_fp16")]; + tensor var_6062_equation_0 = const()[name = tensor("op_6062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6062_cast_fp16 = einsum(equation = var_6062_equation_0, values = (var_5868_cast_fp16, var_5382_cast_fp16))[name = tensor("op_6062_cast_fp16")]; + tensor var_6063_to_fp16 = const()[name = tensor("op_6063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_505_cast_fp16 = mul(x = var_6062_cast_fp16, y = var_6063_to_fp16)[name = tensor("aw_chunk_505_cast_fp16")]; + tensor var_6066_equation_0 = const()[name = tensor("op_6066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6066_cast_fp16 = einsum(equation = var_6066_equation_0, values = (var_5868_cast_fp16, var_5389_cast_fp16))[name = tensor("op_6066_cast_fp16")]; + tensor var_6067_to_fp16 = const()[name = tensor("op_6067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_507_cast_fp16 = mul(x = var_6066_cast_fp16, y = var_6067_to_fp16)[name = tensor("aw_chunk_507_cast_fp16")]; + tensor var_6070_equation_0 = const()[name = tensor("op_6070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6070_cast_fp16 = einsum(equation = var_6070_equation_0, values = (var_5868_cast_fp16, var_5396_cast_fp16))[name = tensor("op_6070_cast_fp16")]; + tensor var_6071_to_fp16 = const()[name = tensor("op_6071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_509_cast_fp16 = mul(x = var_6070_cast_fp16, y = var_6071_to_fp16)[name = tensor("aw_chunk_509_cast_fp16")]; + tensor var_6074_equation_0 = const()[name = tensor("op_6074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6074_cast_fp16 = einsum(equation = var_6074_equation_0, values = (var_5868_cast_fp16, var_5403_cast_fp16))[name = tensor("op_6074_cast_fp16")]; + tensor var_6075_to_fp16 = const()[name = tensor("op_6075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_511_cast_fp16 = mul(x = var_6074_cast_fp16, y = var_6075_to_fp16)[name = tensor("aw_chunk_511_cast_fp16")]; + tensor var_6078_equation_0 = const()[name = tensor("op_6078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6078_cast_fp16 = einsum(equation = var_6078_equation_0, values = (var_5872_cast_fp16, var_5410_cast_fp16))[name = tensor("op_6078_cast_fp16")]; + tensor var_6079_to_fp16 = const()[name = tensor("op_6079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_513_cast_fp16 = mul(x = var_6078_cast_fp16, y = var_6079_to_fp16)[name = tensor("aw_chunk_513_cast_fp16")]; + tensor var_6082_equation_0 = const()[name = tensor("op_6082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6082_cast_fp16 = einsum(equation = var_6082_equation_0, values = (var_5872_cast_fp16, var_5417_cast_fp16))[name = tensor("op_6082_cast_fp16")]; + tensor var_6083_to_fp16 = const()[name = tensor("op_6083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_515_cast_fp16 = mul(x = var_6082_cast_fp16, y = var_6083_to_fp16)[name = tensor("aw_chunk_515_cast_fp16")]; + tensor var_6086_equation_0 = const()[name = tensor("op_6086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6086_cast_fp16 = einsum(equation = var_6086_equation_0, values = (var_5872_cast_fp16, var_5424_cast_fp16))[name = tensor("op_6086_cast_fp16")]; + tensor var_6087_to_fp16 = const()[name = tensor("op_6087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_517_cast_fp16 = mul(x = var_6086_cast_fp16, y = var_6087_to_fp16)[name = tensor("aw_chunk_517_cast_fp16")]; + tensor var_6090_equation_0 = const()[name = tensor("op_6090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6090_cast_fp16 = einsum(equation = var_6090_equation_0, values = (var_5872_cast_fp16, var_5431_cast_fp16))[name = tensor("op_6090_cast_fp16")]; + tensor var_6091_to_fp16 = const()[name = tensor("op_6091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_519_cast_fp16 = mul(x = var_6090_cast_fp16, y = var_6091_to_fp16)[name = tensor("aw_chunk_519_cast_fp16")]; + tensor var_6094_equation_0 = const()[name = tensor("op_6094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6094_cast_fp16 = einsum(equation = var_6094_equation_0, values = (var_5876_cast_fp16, var_5438_cast_fp16))[name = tensor("op_6094_cast_fp16")]; + tensor var_6095_to_fp16 = const()[name = tensor("op_6095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_521_cast_fp16 = mul(x = var_6094_cast_fp16, y = var_6095_to_fp16)[name = tensor("aw_chunk_521_cast_fp16")]; + tensor var_6098_equation_0 = const()[name = tensor("op_6098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6098_cast_fp16 = einsum(equation = var_6098_equation_0, values = (var_5876_cast_fp16, var_5445_cast_fp16))[name = tensor("op_6098_cast_fp16")]; + tensor var_6099_to_fp16 = const()[name = tensor("op_6099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_523_cast_fp16 = mul(x = var_6098_cast_fp16, y = var_6099_to_fp16)[name = tensor("aw_chunk_523_cast_fp16")]; + tensor var_6102_equation_0 = const()[name = tensor("op_6102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6102_cast_fp16 = einsum(equation = var_6102_equation_0, values = (var_5876_cast_fp16, var_5452_cast_fp16))[name = tensor("op_6102_cast_fp16")]; + tensor var_6103_to_fp16 = const()[name = tensor("op_6103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_525_cast_fp16 = mul(x = var_6102_cast_fp16, y = var_6103_to_fp16)[name = tensor("aw_chunk_525_cast_fp16")]; + tensor var_6106_equation_0 = const()[name = tensor("op_6106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6106_cast_fp16 = einsum(equation = var_6106_equation_0, values = (var_5876_cast_fp16, var_5459_cast_fp16))[name = tensor("op_6106_cast_fp16")]; + tensor var_6107_to_fp16 = const()[name = tensor("op_6107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_527_cast_fp16 = mul(x = var_6106_cast_fp16, y = var_6107_to_fp16)[name = tensor("aw_chunk_527_cast_fp16")]; + tensor var_6110_equation_0 = const()[name = tensor("op_6110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6110_cast_fp16 = einsum(equation = var_6110_equation_0, values = (var_5880_cast_fp16, var_5466_cast_fp16))[name = tensor("op_6110_cast_fp16")]; + tensor var_6111_to_fp16 = const()[name = tensor("op_6111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_529_cast_fp16 = mul(x = var_6110_cast_fp16, y = var_6111_to_fp16)[name = tensor("aw_chunk_529_cast_fp16")]; + tensor var_6114_equation_0 = const()[name = tensor("op_6114_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6114_cast_fp16 = einsum(equation = var_6114_equation_0, values = (var_5880_cast_fp16, var_5473_cast_fp16))[name = tensor("op_6114_cast_fp16")]; + tensor var_6115_to_fp16 = const()[name = tensor("op_6115_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_531_cast_fp16 = mul(x = var_6114_cast_fp16, y = var_6115_to_fp16)[name = tensor("aw_chunk_531_cast_fp16")]; + tensor var_6118_equation_0 = const()[name = tensor("op_6118_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6118_cast_fp16 = einsum(equation = var_6118_equation_0, values = (var_5880_cast_fp16, var_5480_cast_fp16))[name = tensor("op_6118_cast_fp16")]; + tensor var_6119_to_fp16 = const()[name = tensor("op_6119_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_533_cast_fp16 = mul(x = var_6118_cast_fp16, y = var_6119_to_fp16)[name = tensor("aw_chunk_533_cast_fp16")]; + tensor var_6122_equation_0 = const()[name = tensor("op_6122_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6122_cast_fp16 = einsum(equation = var_6122_equation_0, values = (var_5880_cast_fp16, var_5487_cast_fp16))[name = tensor("op_6122_cast_fp16")]; + tensor var_6123_to_fp16 = const()[name = tensor("op_6123_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_535_cast_fp16 = mul(x = var_6122_cast_fp16, y = var_6123_to_fp16)[name = tensor("aw_chunk_535_cast_fp16")]; + tensor var_6126_equation_0 = const()[name = tensor("op_6126_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6126_cast_fp16 = einsum(equation = var_6126_equation_0, values = (var_5884_cast_fp16, var_5494_cast_fp16))[name = tensor("op_6126_cast_fp16")]; + tensor var_6127_to_fp16 = const()[name = tensor("op_6127_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_537_cast_fp16 = mul(x = var_6126_cast_fp16, y = var_6127_to_fp16)[name = tensor("aw_chunk_537_cast_fp16")]; + tensor var_6130_equation_0 = const()[name = tensor("op_6130_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6130_cast_fp16 = einsum(equation = var_6130_equation_0, values = (var_5884_cast_fp16, var_5501_cast_fp16))[name = tensor("op_6130_cast_fp16")]; + tensor var_6131_to_fp16 = const()[name = tensor("op_6131_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_539_cast_fp16 = mul(x = var_6130_cast_fp16, y = var_6131_to_fp16)[name = tensor("aw_chunk_539_cast_fp16")]; + tensor var_6134_equation_0 = const()[name = tensor("op_6134_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6134_cast_fp16 = einsum(equation = var_6134_equation_0, values = (var_5884_cast_fp16, var_5508_cast_fp16))[name = tensor("op_6134_cast_fp16")]; + tensor var_6135_to_fp16 = const()[name = tensor("op_6135_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_541_cast_fp16 = mul(x = var_6134_cast_fp16, y = var_6135_to_fp16)[name = tensor("aw_chunk_541_cast_fp16")]; + tensor var_6138_equation_0 = const()[name = tensor("op_6138_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6138_cast_fp16 = einsum(equation = var_6138_equation_0, values = (var_5884_cast_fp16, var_5515_cast_fp16))[name = tensor("op_6138_cast_fp16")]; + tensor var_6139_to_fp16 = const()[name = tensor("op_6139_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_543_cast_fp16 = mul(x = var_6138_cast_fp16, y = var_6139_to_fp16)[name = tensor("aw_chunk_543_cast_fp16")]; + tensor var_6142_equation_0 = const()[name = tensor("op_6142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6142_cast_fp16 = einsum(equation = var_6142_equation_0, values = (var_5888_cast_fp16, var_5522_cast_fp16))[name = tensor("op_6142_cast_fp16")]; + tensor var_6143_to_fp16 = const()[name = tensor("op_6143_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_545_cast_fp16 = mul(x = var_6142_cast_fp16, y = var_6143_to_fp16)[name = tensor("aw_chunk_545_cast_fp16")]; + tensor var_6146_equation_0 = const()[name = tensor("op_6146_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6146_cast_fp16 = einsum(equation = var_6146_equation_0, values = (var_5888_cast_fp16, var_5529_cast_fp16))[name = tensor("op_6146_cast_fp16")]; + tensor var_6147_to_fp16 = const()[name = tensor("op_6147_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_547_cast_fp16 = mul(x = var_6146_cast_fp16, y = var_6147_to_fp16)[name = tensor("aw_chunk_547_cast_fp16")]; + tensor var_6150_equation_0 = const()[name = tensor("op_6150_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6150_cast_fp16 = einsum(equation = var_6150_equation_0, values = (var_5888_cast_fp16, var_5536_cast_fp16))[name = tensor("op_6150_cast_fp16")]; + tensor var_6151_to_fp16 = const()[name = tensor("op_6151_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_549_cast_fp16 = mul(x = var_6150_cast_fp16, y = var_6151_to_fp16)[name = tensor("aw_chunk_549_cast_fp16")]; + tensor var_6154_equation_0 = const()[name = tensor("op_6154_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6154_cast_fp16 = einsum(equation = var_6154_equation_0, values = (var_5888_cast_fp16, var_5543_cast_fp16))[name = tensor("op_6154_cast_fp16")]; + tensor var_6155_to_fp16 = const()[name = tensor("op_6155_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_551_cast_fp16 = mul(x = var_6154_cast_fp16, y = var_6155_to_fp16)[name = tensor("aw_chunk_551_cast_fp16")]; + tensor var_6158_equation_0 = const()[name = tensor("op_6158_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6158_cast_fp16 = einsum(equation = var_6158_equation_0, values = (var_5892_cast_fp16, var_5550_cast_fp16))[name = tensor("op_6158_cast_fp16")]; + tensor var_6159_to_fp16 = const()[name = tensor("op_6159_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_553_cast_fp16 = mul(x = var_6158_cast_fp16, y = var_6159_to_fp16)[name = tensor("aw_chunk_553_cast_fp16")]; + tensor var_6162_equation_0 = const()[name = tensor("op_6162_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6162_cast_fp16 = einsum(equation = var_6162_equation_0, values = (var_5892_cast_fp16, var_5557_cast_fp16))[name = tensor("op_6162_cast_fp16")]; + tensor var_6163_to_fp16 = const()[name = tensor("op_6163_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_555_cast_fp16 = mul(x = var_6162_cast_fp16, y = var_6163_to_fp16)[name = tensor("aw_chunk_555_cast_fp16")]; + tensor var_6166_equation_0 = const()[name = tensor("op_6166_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6166_cast_fp16 = einsum(equation = var_6166_equation_0, values = (var_5892_cast_fp16, var_5564_cast_fp16))[name = tensor("op_6166_cast_fp16")]; + tensor var_6167_to_fp16 = const()[name = tensor("op_6167_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_557_cast_fp16 = mul(x = var_6166_cast_fp16, y = var_6167_to_fp16)[name = tensor("aw_chunk_557_cast_fp16")]; + tensor var_6170_equation_0 = const()[name = tensor("op_6170_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6170_cast_fp16 = einsum(equation = var_6170_equation_0, values = (var_5892_cast_fp16, var_5571_cast_fp16))[name = tensor("op_6170_cast_fp16")]; + tensor var_6171_to_fp16 = const()[name = tensor("op_6171_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_559_cast_fp16 = mul(x = var_6170_cast_fp16, y = var_6171_to_fp16)[name = tensor("aw_chunk_559_cast_fp16")]; + tensor var_6174_equation_0 = const()[name = tensor("op_6174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6174_cast_fp16 = einsum(equation = var_6174_equation_0, values = (var_5896_cast_fp16, var_5578_cast_fp16))[name = tensor("op_6174_cast_fp16")]; + tensor var_6175_to_fp16 = const()[name = tensor("op_6175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_561_cast_fp16 = mul(x = var_6174_cast_fp16, y = var_6175_to_fp16)[name = tensor("aw_chunk_561_cast_fp16")]; + tensor var_6178_equation_0 = const()[name = tensor("op_6178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6178_cast_fp16 = einsum(equation = var_6178_equation_0, values = (var_5896_cast_fp16, var_5585_cast_fp16))[name = tensor("op_6178_cast_fp16")]; + tensor var_6179_to_fp16 = const()[name = tensor("op_6179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_563_cast_fp16 = mul(x = var_6178_cast_fp16, y = var_6179_to_fp16)[name = tensor("aw_chunk_563_cast_fp16")]; + tensor var_6182_equation_0 = const()[name = tensor("op_6182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6182_cast_fp16 = einsum(equation = var_6182_equation_0, values = (var_5896_cast_fp16, var_5592_cast_fp16))[name = tensor("op_6182_cast_fp16")]; + tensor var_6183_to_fp16 = const()[name = tensor("op_6183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_565_cast_fp16 = mul(x = var_6182_cast_fp16, y = var_6183_to_fp16)[name = tensor("aw_chunk_565_cast_fp16")]; + tensor var_6186_equation_0 = const()[name = tensor("op_6186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6186_cast_fp16 = einsum(equation = var_6186_equation_0, values = (var_5896_cast_fp16, var_5599_cast_fp16))[name = tensor("op_6186_cast_fp16")]; + tensor var_6187_to_fp16 = const()[name = tensor("op_6187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_567_cast_fp16 = mul(x = var_6186_cast_fp16, y = var_6187_to_fp16)[name = tensor("aw_chunk_567_cast_fp16")]; + tensor var_6190_equation_0 = const()[name = tensor("op_6190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6190_cast_fp16 = einsum(equation = var_6190_equation_0, values = (var_5900_cast_fp16, var_5606_cast_fp16))[name = tensor("op_6190_cast_fp16")]; + tensor var_6191_to_fp16 = const()[name = tensor("op_6191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_569_cast_fp16 = mul(x = var_6190_cast_fp16, y = var_6191_to_fp16)[name = tensor("aw_chunk_569_cast_fp16")]; + tensor var_6194_equation_0 = const()[name = tensor("op_6194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6194_cast_fp16 = einsum(equation = var_6194_equation_0, values = (var_5900_cast_fp16, var_5613_cast_fp16))[name = tensor("op_6194_cast_fp16")]; + tensor var_6195_to_fp16 = const()[name = tensor("op_6195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_571_cast_fp16 = mul(x = var_6194_cast_fp16, y = var_6195_to_fp16)[name = tensor("aw_chunk_571_cast_fp16")]; + tensor var_6198_equation_0 = const()[name = tensor("op_6198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6198_cast_fp16 = einsum(equation = var_6198_equation_0, values = (var_5900_cast_fp16, var_5620_cast_fp16))[name = tensor("op_6198_cast_fp16")]; + tensor var_6199_to_fp16 = const()[name = tensor("op_6199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_573_cast_fp16 = mul(x = var_6198_cast_fp16, y = var_6199_to_fp16)[name = tensor("aw_chunk_573_cast_fp16")]; + tensor var_6202_equation_0 = const()[name = tensor("op_6202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6202_cast_fp16 = einsum(equation = var_6202_equation_0, values = (var_5900_cast_fp16, var_5627_cast_fp16))[name = tensor("op_6202_cast_fp16")]; + tensor var_6203_to_fp16 = const()[name = tensor("op_6203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_575_cast_fp16 = mul(x = var_6202_cast_fp16, y = var_6203_to_fp16)[name = tensor("aw_chunk_575_cast_fp16")]; + tensor var_6206_equation_0 = const()[name = tensor("op_6206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6206_cast_fp16 = einsum(equation = var_6206_equation_0, values = (var_5904_cast_fp16, var_5634_cast_fp16))[name = tensor("op_6206_cast_fp16")]; + tensor var_6207_to_fp16 = const()[name = tensor("op_6207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_577_cast_fp16 = mul(x = var_6206_cast_fp16, y = var_6207_to_fp16)[name = tensor("aw_chunk_577_cast_fp16")]; + tensor var_6210_equation_0 = const()[name = tensor("op_6210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6210_cast_fp16 = einsum(equation = var_6210_equation_0, values = (var_5904_cast_fp16, var_5641_cast_fp16))[name = tensor("op_6210_cast_fp16")]; + tensor var_6211_to_fp16 = const()[name = tensor("op_6211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_579_cast_fp16 = mul(x = var_6210_cast_fp16, y = var_6211_to_fp16)[name = tensor("aw_chunk_579_cast_fp16")]; + tensor var_6214_equation_0 = const()[name = tensor("op_6214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6214_cast_fp16 = einsum(equation = var_6214_equation_0, values = (var_5904_cast_fp16, var_5648_cast_fp16))[name = tensor("op_6214_cast_fp16")]; + tensor var_6215_to_fp16 = const()[name = tensor("op_6215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_581_cast_fp16 = mul(x = var_6214_cast_fp16, y = var_6215_to_fp16)[name = tensor("aw_chunk_581_cast_fp16")]; + tensor var_6218_equation_0 = const()[name = tensor("op_6218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6218_cast_fp16 = einsum(equation = var_6218_equation_0, values = (var_5904_cast_fp16, var_5655_cast_fp16))[name = tensor("op_6218_cast_fp16")]; + tensor var_6219_to_fp16 = const()[name = tensor("op_6219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_583_cast_fp16 = mul(x = var_6218_cast_fp16, y = var_6219_to_fp16)[name = tensor("aw_chunk_583_cast_fp16")]; + tensor var_6222_equation_0 = const()[name = tensor("op_6222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6222_cast_fp16 = einsum(equation = var_6222_equation_0, values = (var_5908_cast_fp16, var_5662_cast_fp16))[name = tensor("op_6222_cast_fp16")]; + tensor var_6223_to_fp16 = const()[name = tensor("op_6223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_585_cast_fp16 = mul(x = var_6222_cast_fp16, y = var_6223_to_fp16)[name = tensor("aw_chunk_585_cast_fp16")]; + tensor var_6226_equation_0 = const()[name = tensor("op_6226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6226_cast_fp16 = einsum(equation = var_6226_equation_0, values = (var_5908_cast_fp16, var_5669_cast_fp16))[name = tensor("op_6226_cast_fp16")]; + tensor var_6227_to_fp16 = const()[name = tensor("op_6227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_587_cast_fp16 = mul(x = var_6226_cast_fp16, y = var_6227_to_fp16)[name = tensor("aw_chunk_587_cast_fp16")]; + tensor var_6230_equation_0 = const()[name = tensor("op_6230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6230_cast_fp16 = einsum(equation = var_6230_equation_0, values = (var_5908_cast_fp16, var_5676_cast_fp16))[name = tensor("op_6230_cast_fp16")]; + tensor var_6231_to_fp16 = const()[name = tensor("op_6231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_589_cast_fp16 = mul(x = var_6230_cast_fp16, y = var_6231_to_fp16)[name = tensor("aw_chunk_589_cast_fp16")]; + tensor var_6234_equation_0 = const()[name = tensor("op_6234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6234_cast_fp16 = einsum(equation = var_6234_equation_0, values = (var_5908_cast_fp16, var_5683_cast_fp16))[name = tensor("op_6234_cast_fp16")]; + tensor var_6235_to_fp16 = const()[name = tensor("op_6235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_591_cast_fp16 = mul(x = var_6234_cast_fp16, y = var_6235_to_fp16)[name = tensor("aw_chunk_591_cast_fp16")]; + tensor var_6238_equation_0 = const()[name = tensor("op_6238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6238_cast_fp16 = einsum(equation = var_6238_equation_0, values = (var_5912_cast_fp16, var_5690_cast_fp16))[name = tensor("op_6238_cast_fp16")]; + tensor var_6239_to_fp16 = const()[name = tensor("op_6239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_593_cast_fp16 = mul(x = var_6238_cast_fp16, y = var_6239_to_fp16)[name = tensor("aw_chunk_593_cast_fp16")]; + tensor var_6242_equation_0 = const()[name = tensor("op_6242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6242_cast_fp16 = einsum(equation = var_6242_equation_0, values = (var_5912_cast_fp16, var_5697_cast_fp16))[name = tensor("op_6242_cast_fp16")]; + tensor var_6243_to_fp16 = const()[name = tensor("op_6243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_595_cast_fp16 = mul(x = var_6242_cast_fp16, y = var_6243_to_fp16)[name = tensor("aw_chunk_595_cast_fp16")]; + tensor var_6246_equation_0 = const()[name = tensor("op_6246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6246_cast_fp16 = einsum(equation = var_6246_equation_0, values = (var_5912_cast_fp16, var_5704_cast_fp16))[name = tensor("op_6246_cast_fp16")]; + tensor var_6247_to_fp16 = const()[name = tensor("op_6247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_597_cast_fp16 = mul(x = var_6246_cast_fp16, y = var_6247_to_fp16)[name = tensor("aw_chunk_597_cast_fp16")]; + tensor var_6250_equation_0 = const()[name = tensor("op_6250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6250_cast_fp16 = einsum(equation = var_6250_equation_0, values = (var_5912_cast_fp16, var_5711_cast_fp16))[name = tensor("op_6250_cast_fp16")]; + tensor var_6251_to_fp16 = const()[name = tensor("op_6251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_599_cast_fp16 = mul(x = var_6250_cast_fp16, y = var_6251_to_fp16)[name = tensor("aw_chunk_599_cast_fp16")]; + tensor var_6254_equation_0 = const()[name = tensor("op_6254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6254_cast_fp16 = einsum(equation = var_6254_equation_0, values = (var_5916_cast_fp16, var_5718_cast_fp16))[name = tensor("op_6254_cast_fp16")]; + tensor var_6255_to_fp16 = const()[name = tensor("op_6255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_601_cast_fp16 = mul(x = var_6254_cast_fp16, y = var_6255_to_fp16)[name = tensor("aw_chunk_601_cast_fp16")]; + tensor var_6258_equation_0 = const()[name = tensor("op_6258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6258_cast_fp16 = einsum(equation = var_6258_equation_0, values = (var_5916_cast_fp16, var_5725_cast_fp16))[name = tensor("op_6258_cast_fp16")]; + tensor var_6259_to_fp16 = const()[name = tensor("op_6259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_603_cast_fp16 = mul(x = var_6258_cast_fp16, y = var_6259_to_fp16)[name = tensor("aw_chunk_603_cast_fp16")]; + tensor var_6262_equation_0 = const()[name = tensor("op_6262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6262_cast_fp16 = einsum(equation = var_6262_equation_0, values = (var_5916_cast_fp16, var_5732_cast_fp16))[name = tensor("op_6262_cast_fp16")]; + tensor var_6263_to_fp16 = const()[name = tensor("op_6263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_605_cast_fp16 = mul(x = var_6262_cast_fp16, y = var_6263_to_fp16)[name = tensor("aw_chunk_605_cast_fp16")]; + tensor var_6266_equation_0 = const()[name = tensor("op_6266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6266_cast_fp16 = einsum(equation = var_6266_equation_0, values = (var_5916_cast_fp16, var_5739_cast_fp16))[name = tensor("op_6266_cast_fp16")]; + tensor var_6267_to_fp16 = const()[name = tensor("op_6267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_607_cast_fp16 = mul(x = var_6266_cast_fp16, y = var_6267_to_fp16)[name = tensor("aw_chunk_607_cast_fp16")]; + tensor var_6270_equation_0 = const()[name = tensor("op_6270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6270_cast_fp16 = einsum(equation = var_6270_equation_0, values = (var_5920_cast_fp16, var_5746_cast_fp16))[name = tensor("op_6270_cast_fp16")]; + tensor var_6271_to_fp16 = const()[name = tensor("op_6271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_609_cast_fp16 = mul(x = var_6270_cast_fp16, y = var_6271_to_fp16)[name = tensor("aw_chunk_609_cast_fp16")]; + tensor var_6274_equation_0 = const()[name = tensor("op_6274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6274_cast_fp16 = einsum(equation = var_6274_equation_0, values = (var_5920_cast_fp16, var_5753_cast_fp16))[name = tensor("op_6274_cast_fp16")]; + tensor var_6275_to_fp16 = const()[name = tensor("op_6275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_611_cast_fp16 = mul(x = var_6274_cast_fp16, y = var_6275_to_fp16)[name = tensor("aw_chunk_611_cast_fp16")]; + tensor var_6278_equation_0 = const()[name = tensor("op_6278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6278_cast_fp16 = einsum(equation = var_6278_equation_0, values = (var_5920_cast_fp16, var_5760_cast_fp16))[name = tensor("op_6278_cast_fp16")]; + tensor var_6279_to_fp16 = const()[name = tensor("op_6279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_613_cast_fp16 = mul(x = var_6278_cast_fp16, y = var_6279_to_fp16)[name = tensor("aw_chunk_613_cast_fp16")]; + tensor var_6282_equation_0 = const()[name = tensor("op_6282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6282_cast_fp16 = einsum(equation = var_6282_equation_0, values = (var_5920_cast_fp16, var_5767_cast_fp16))[name = tensor("op_6282_cast_fp16")]; + tensor var_6283_to_fp16 = const()[name = tensor("op_6283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_615_cast_fp16 = mul(x = var_6282_cast_fp16, y = var_6283_to_fp16)[name = tensor("aw_chunk_615_cast_fp16")]; + tensor var_6286_equation_0 = const()[name = tensor("op_6286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6286_cast_fp16 = einsum(equation = var_6286_equation_0, values = (var_5924_cast_fp16, var_5774_cast_fp16))[name = tensor("op_6286_cast_fp16")]; + tensor var_6287_to_fp16 = const()[name = tensor("op_6287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_617_cast_fp16 = mul(x = var_6286_cast_fp16, y = var_6287_to_fp16)[name = tensor("aw_chunk_617_cast_fp16")]; + tensor var_6290_equation_0 = const()[name = tensor("op_6290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6290_cast_fp16 = einsum(equation = var_6290_equation_0, values = (var_5924_cast_fp16, var_5781_cast_fp16))[name = tensor("op_6290_cast_fp16")]; + tensor var_6291_to_fp16 = const()[name = tensor("op_6291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_619_cast_fp16 = mul(x = var_6290_cast_fp16, y = var_6291_to_fp16)[name = tensor("aw_chunk_619_cast_fp16")]; + tensor var_6294_equation_0 = const()[name = tensor("op_6294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6294_cast_fp16 = einsum(equation = var_6294_equation_0, values = (var_5924_cast_fp16, var_5788_cast_fp16))[name = tensor("op_6294_cast_fp16")]; + tensor var_6295_to_fp16 = const()[name = tensor("op_6295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_621_cast_fp16 = mul(x = var_6294_cast_fp16, y = var_6295_to_fp16)[name = tensor("aw_chunk_621_cast_fp16")]; + tensor var_6298_equation_0 = const()[name = tensor("op_6298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6298_cast_fp16 = einsum(equation = var_6298_equation_0, values = (var_5924_cast_fp16, var_5795_cast_fp16))[name = tensor("op_6298_cast_fp16")]; + tensor var_6299_to_fp16 = const()[name = tensor("op_6299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_623_cast_fp16 = mul(x = var_6298_cast_fp16, y = var_6299_to_fp16)[name = tensor("aw_chunk_623_cast_fp16")]; + tensor var_6302_equation_0 = const()[name = tensor("op_6302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6302_cast_fp16 = einsum(equation = var_6302_equation_0, values = (var_5928_cast_fp16, var_5802_cast_fp16))[name = tensor("op_6302_cast_fp16")]; + tensor var_6303_to_fp16 = const()[name = tensor("op_6303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_625_cast_fp16 = mul(x = var_6302_cast_fp16, y = var_6303_to_fp16)[name = tensor("aw_chunk_625_cast_fp16")]; + tensor var_6306_equation_0 = const()[name = tensor("op_6306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6306_cast_fp16 = einsum(equation = var_6306_equation_0, values = (var_5928_cast_fp16, var_5809_cast_fp16))[name = tensor("op_6306_cast_fp16")]; + tensor var_6307_to_fp16 = const()[name = tensor("op_6307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_627_cast_fp16 = mul(x = var_6306_cast_fp16, y = var_6307_to_fp16)[name = tensor("aw_chunk_627_cast_fp16")]; + tensor var_6310_equation_0 = const()[name = tensor("op_6310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6310_cast_fp16 = einsum(equation = var_6310_equation_0, values = (var_5928_cast_fp16, var_5816_cast_fp16))[name = tensor("op_6310_cast_fp16")]; + tensor var_6311_to_fp16 = const()[name = tensor("op_6311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_629_cast_fp16 = mul(x = var_6310_cast_fp16, y = var_6311_to_fp16)[name = tensor("aw_chunk_629_cast_fp16")]; + tensor var_6314_equation_0 = const()[name = tensor("op_6314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6314_cast_fp16 = einsum(equation = var_6314_equation_0, values = (var_5928_cast_fp16, var_5823_cast_fp16))[name = tensor("op_6314_cast_fp16")]; + tensor var_6315_to_fp16 = const()[name = tensor("op_6315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_631_cast_fp16 = mul(x = var_6314_cast_fp16, y = var_6315_to_fp16)[name = tensor("aw_chunk_631_cast_fp16")]; + tensor var_6318_equation_0 = const()[name = tensor("op_6318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6318_cast_fp16 = einsum(equation = var_6318_equation_0, values = (var_5932_cast_fp16, var_5830_cast_fp16))[name = tensor("op_6318_cast_fp16")]; + tensor var_6319_to_fp16 = const()[name = tensor("op_6319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_633_cast_fp16 = mul(x = var_6318_cast_fp16, y = var_6319_to_fp16)[name = tensor("aw_chunk_633_cast_fp16")]; + tensor var_6322_equation_0 = const()[name = tensor("op_6322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6322_cast_fp16 = einsum(equation = var_6322_equation_0, values = (var_5932_cast_fp16, var_5837_cast_fp16))[name = tensor("op_6322_cast_fp16")]; + tensor var_6323_to_fp16 = const()[name = tensor("op_6323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_635_cast_fp16 = mul(x = var_6322_cast_fp16, y = var_6323_to_fp16)[name = tensor("aw_chunk_635_cast_fp16")]; + tensor var_6326_equation_0 = const()[name = tensor("op_6326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6326_cast_fp16 = einsum(equation = var_6326_equation_0, values = (var_5932_cast_fp16, var_5844_cast_fp16))[name = tensor("op_6326_cast_fp16")]; + tensor var_6327_to_fp16 = const()[name = tensor("op_6327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_637_cast_fp16 = mul(x = var_6326_cast_fp16, y = var_6327_to_fp16)[name = tensor("aw_chunk_637_cast_fp16")]; + tensor var_6330_equation_0 = const()[name = tensor("op_6330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_6330_cast_fp16 = einsum(equation = var_6330_equation_0, values = (var_5932_cast_fp16, var_5851_cast_fp16))[name = tensor("op_6330_cast_fp16")]; + tensor var_6331_to_fp16 = const()[name = tensor("op_6331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_639_cast_fp16 = mul(x = var_6330_cast_fp16, y = var_6331_to_fp16)[name = tensor("aw_chunk_639_cast_fp16")]; + tensor var_6333_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_481_cast_fp16)[name = tensor("op_6333_cast_fp16")]; + tensor var_6334_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_483_cast_fp16)[name = tensor("op_6334_cast_fp16")]; + tensor var_6335_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_485_cast_fp16)[name = tensor("op_6335_cast_fp16")]; + tensor var_6336_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_487_cast_fp16)[name = tensor("op_6336_cast_fp16")]; + tensor var_6337_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_489_cast_fp16)[name = tensor("op_6337_cast_fp16")]; + tensor var_6338_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_491_cast_fp16)[name = tensor("op_6338_cast_fp16")]; + tensor var_6339_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_493_cast_fp16)[name = tensor("op_6339_cast_fp16")]; + tensor var_6340_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_495_cast_fp16)[name = tensor("op_6340_cast_fp16")]; + tensor var_6341_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_497_cast_fp16)[name = tensor("op_6341_cast_fp16")]; + tensor var_6342_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_499_cast_fp16)[name = tensor("op_6342_cast_fp16")]; + tensor var_6343_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_501_cast_fp16)[name = tensor("op_6343_cast_fp16")]; + tensor var_6344_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_503_cast_fp16)[name = tensor("op_6344_cast_fp16")]; + tensor var_6345_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_505_cast_fp16)[name = tensor("op_6345_cast_fp16")]; + tensor var_6346_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_507_cast_fp16)[name = tensor("op_6346_cast_fp16")]; + tensor var_6347_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_509_cast_fp16)[name = tensor("op_6347_cast_fp16")]; + tensor var_6348_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_511_cast_fp16)[name = tensor("op_6348_cast_fp16")]; + tensor var_6349_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_513_cast_fp16)[name = tensor("op_6349_cast_fp16")]; + tensor var_6350_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_515_cast_fp16)[name = tensor("op_6350_cast_fp16")]; + tensor var_6351_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_517_cast_fp16)[name = tensor("op_6351_cast_fp16")]; + tensor var_6352_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_519_cast_fp16)[name = tensor("op_6352_cast_fp16")]; + tensor var_6353_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_521_cast_fp16)[name = tensor("op_6353_cast_fp16")]; + tensor var_6354_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_523_cast_fp16)[name = tensor("op_6354_cast_fp16")]; + tensor var_6355_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_525_cast_fp16)[name = tensor("op_6355_cast_fp16")]; + tensor var_6356_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_527_cast_fp16)[name = tensor("op_6356_cast_fp16")]; + tensor var_6357_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_529_cast_fp16)[name = tensor("op_6357_cast_fp16")]; + tensor var_6358_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_531_cast_fp16)[name = tensor("op_6358_cast_fp16")]; + tensor var_6359_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_533_cast_fp16)[name = tensor("op_6359_cast_fp16")]; + tensor var_6360_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_535_cast_fp16)[name = tensor("op_6360_cast_fp16")]; + tensor var_6361_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_537_cast_fp16)[name = tensor("op_6361_cast_fp16")]; + tensor var_6362_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_539_cast_fp16)[name = tensor("op_6362_cast_fp16")]; + tensor var_6363_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_541_cast_fp16)[name = tensor("op_6363_cast_fp16")]; + tensor var_6364_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_543_cast_fp16)[name = tensor("op_6364_cast_fp16")]; + tensor var_6365_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_545_cast_fp16)[name = tensor("op_6365_cast_fp16")]; + tensor var_6366_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_547_cast_fp16)[name = tensor("op_6366_cast_fp16")]; + tensor var_6367_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_549_cast_fp16)[name = tensor("op_6367_cast_fp16")]; + tensor var_6368_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_551_cast_fp16)[name = tensor("op_6368_cast_fp16")]; + tensor var_6369_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_553_cast_fp16)[name = tensor("op_6369_cast_fp16")]; + tensor var_6370_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_555_cast_fp16)[name = tensor("op_6370_cast_fp16")]; + tensor var_6371_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_557_cast_fp16)[name = tensor("op_6371_cast_fp16")]; + tensor var_6372_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_559_cast_fp16)[name = tensor("op_6372_cast_fp16")]; + tensor var_6373_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_561_cast_fp16)[name = tensor("op_6373_cast_fp16")]; + tensor var_6374_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_563_cast_fp16)[name = tensor("op_6374_cast_fp16")]; + tensor var_6375_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_565_cast_fp16)[name = tensor("op_6375_cast_fp16")]; + tensor var_6376_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_567_cast_fp16)[name = tensor("op_6376_cast_fp16")]; + tensor var_6377_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_569_cast_fp16)[name = tensor("op_6377_cast_fp16")]; + tensor var_6378_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_571_cast_fp16)[name = tensor("op_6378_cast_fp16")]; + tensor var_6379_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_573_cast_fp16)[name = tensor("op_6379_cast_fp16")]; + tensor var_6380_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_575_cast_fp16)[name = tensor("op_6380_cast_fp16")]; + tensor var_6381_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_577_cast_fp16)[name = tensor("op_6381_cast_fp16")]; + tensor var_6382_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_579_cast_fp16)[name = tensor("op_6382_cast_fp16")]; + tensor var_6383_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_581_cast_fp16)[name = tensor("op_6383_cast_fp16")]; + tensor var_6384_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_583_cast_fp16)[name = tensor("op_6384_cast_fp16")]; + tensor var_6385_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_585_cast_fp16)[name = tensor("op_6385_cast_fp16")]; + tensor var_6386_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_587_cast_fp16)[name = tensor("op_6386_cast_fp16")]; + tensor var_6387_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_589_cast_fp16)[name = tensor("op_6387_cast_fp16")]; + tensor var_6388_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_591_cast_fp16)[name = tensor("op_6388_cast_fp16")]; + tensor var_6389_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_593_cast_fp16)[name = tensor("op_6389_cast_fp16")]; + tensor var_6390_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_595_cast_fp16)[name = tensor("op_6390_cast_fp16")]; + tensor var_6391_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_597_cast_fp16)[name = tensor("op_6391_cast_fp16")]; + tensor var_6392_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_599_cast_fp16)[name = tensor("op_6392_cast_fp16")]; + tensor var_6393_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_601_cast_fp16)[name = tensor("op_6393_cast_fp16")]; + tensor var_6394_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_603_cast_fp16)[name = tensor("op_6394_cast_fp16")]; + tensor var_6395_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_605_cast_fp16)[name = tensor("op_6395_cast_fp16")]; + tensor var_6396_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_607_cast_fp16)[name = tensor("op_6396_cast_fp16")]; + tensor var_6397_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_609_cast_fp16)[name = tensor("op_6397_cast_fp16")]; + tensor var_6398_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_611_cast_fp16)[name = tensor("op_6398_cast_fp16")]; + tensor var_6399_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_613_cast_fp16)[name = tensor("op_6399_cast_fp16")]; + tensor var_6400_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_615_cast_fp16)[name = tensor("op_6400_cast_fp16")]; + tensor var_6401_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_617_cast_fp16)[name = tensor("op_6401_cast_fp16")]; + tensor var_6402_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_619_cast_fp16)[name = tensor("op_6402_cast_fp16")]; + tensor var_6403_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_621_cast_fp16)[name = tensor("op_6403_cast_fp16")]; + tensor var_6404_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_623_cast_fp16)[name = tensor("op_6404_cast_fp16")]; + tensor var_6405_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_625_cast_fp16)[name = tensor("op_6405_cast_fp16")]; + tensor var_6406_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_627_cast_fp16)[name = tensor("op_6406_cast_fp16")]; + tensor var_6407_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_629_cast_fp16)[name = tensor("op_6407_cast_fp16")]; + tensor var_6408_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_631_cast_fp16)[name = tensor("op_6408_cast_fp16")]; + tensor var_6409_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_633_cast_fp16)[name = tensor("op_6409_cast_fp16")]; + tensor var_6410_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_635_cast_fp16)[name = tensor("op_6410_cast_fp16")]; + tensor var_6411_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_637_cast_fp16)[name = tensor("op_6411_cast_fp16")]; + tensor var_6412_cast_fp16 = softmax(axis = var_5104, x = aw_chunk_639_cast_fp16)[name = tensor("op_6412_cast_fp16")]; + tensor var_6414_equation_0 = const()[name = tensor("op_6414_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6414_cast_fp16 = einsum(equation = var_6414_equation_0, values = (var_5934_cast_fp16, var_6333_cast_fp16))[name = tensor("op_6414_cast_fp16")]; + tensor var_6416_equation_0 = const()[name = tensor("op_6416_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6416_cast_fp16 = einsum(equation = var_6416_equation_0, values = (var_5934_cast_fp16, var_6334_cast_fp16))[name = tensor("op_6416_cast_fp16")]; + tensor var_6418_equation_0 = const()[name = tensor("op_6418_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6418_cast_fp16 = einsum(equation = var_6418_equation_0, values = (var_5934_cast_fp16, var_6335_cast_fp16))[name = tensor("op_6418_cast_fp16")]; + tensor var_6420_equation_0 = const()[name = tensor("op_6420_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6420_cast_fp16 = einsum(equation = var_6420_equation_0, values = (var_5934_cast_fp16, var_6336_cast_fp16))[name = tensor("op_6420_cast_fp16")]; + tensor var_6422_equation_0 = const()[name = tensor("op_6422_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6422_cast_fp16 = einsum(equation = var_6422_equation_0, values = (var_5938_cast_fp16, var_6337_cast_fp16))[name = tensor("op_6422_cast_fp16")]; + tensor var_6424_equation_0 = const()[name = tensor("op_6424_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6424_cast_fp16 = einsum(equation = var_6424_equation_0, values = (var_5938_cast_fp16, var_6338_cast_fp16))[name = tensor("op_6424_cast_fp16")]; + tensor var_6426_equation_0 = const()[name = tensor("op_6426_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6426_cast_fp16 = einsum(equation = var_6426_equation_0, values = (var_5938_cast_fp16, var_6339_cast_fp16))[name = tensor("op_6426_cast_fp16")]; + tensor var_6428_equation_0 = const()[name = tensor("op_6428_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6428_cast_fp16 = einsum(equation = var_6428_equation_0, values = (var_5938_cast_fp16, var_6340_cast_fp16))[name = tensor("op_6428_cast_fp16")]; + tensor var_6430_equation_0 = const()[name = tensor("op_6430_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6430_cast_fp16 = einsum(equation = var_6430_equation_0, values = (var_5942_cast_fp16, var_6341_cast_fp16))[name = tensor("op_6430_cast_fp16")]; + tensor var_6432_equation_0 = const()[name = tensor("op_6432_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6432_cast_fp16 = einsum(equation = var_6432_equation_0, values = (var_5942_cast_fp16, var_6342_cast_fp16))[name = tensor("op_6432_cast_fp16")]; + tensor var_6434_equation_0 = const()[name = tensor("op_6434_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6434_cast_fp16 = einsum(equation = var_6434_equation_0, values = (var_5942_cast_fp16, var_6343_cast_fp16))[name = tensor("op_6434_cast_fp16")]; + tensor var_6436_equation_0 = const()[name = tensor("op_6436_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6436_cast_fp16 = einsum(equation = var_6436_equation_0, values = (var_5942_cast_fp16, var_6344_cast_fp16))[name = tensor("op_6436_cast_fp16")]; + tensor var_6438_equation_0 = const()[name = tensor("op_6438_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6438_cast_fp16 = einsum(equation = var_6438_equation_0, values = (var_5946_cast_fp16, var_6345_cast_fp16))[name = tensor("op_6438_cast_fp16")]; + tensor var_6440_equation_0 = const()[name = tensor("op_6440_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6440_cast_fp16 = einsum(equation = var_6440_equation_0, values = (var_5946_cast_fp16, var_6346_cast_fp16))[name = tensor("op_6440_cast_fp16")]; + tensor var_6442_equation_0 = const()[name = tensor("op_6442_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6442_cast_fp16 = einsum(equation = var_6442_equation_0, values = (var_5946_cast_fp16, var_6347_cast_fp16))[name = tensor("op_6442_cast_fp16")]; + tensor var_6444_equation_0 = const()[name = tensor("op_6444_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6444_cast_fp16 = einsum(equation = var_6444_equation_0, values = (var_5946_cast_fp16, var_6348_cast_fp16))[name = tensor("op_6444_cast_fp16")]; + tensor var_6446_equation_0 = const()[name = tensor("op_6446_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6446_cast_fp16 = einsum(equation = var_6446_equation_0, values = (var_5950_cast_fp16, var_6349_cast_fp16))[name = tensor("op_6446_cast_fp16")]; + tensor var_6448_equation_0 = const()[name = tensor("op_6448_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6448_cast_fp16 = einsum(equation = var_6448_equation_0, values = (var_5950_cast_fp16, var_6350_cast_fp16))[name = tensor("op_6448_cast_fp16")]; + tensor var_6450_equation_0 = const()[name = tensor("op_6450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6450_cast_fp16 = einsum(equation = var_6450_equation_0, values = (var_5950_cast_fp16, var_6351_cast_fp16))[name = tensor("op_6450_cast_fp16")]; + tensor var_6452_equation_0 = const()[name = tensor("op_6452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6452_cast_fp16 = einsum(equation = var_6452_equation_0, values = (var_5950_cast_fp16, var_6352_cast_fp16))[name = tensor("op_6452_cast_fp16")]; + tensor var_6454_equation_0 = const()[name = tensor("op_6454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6454_cast_fp16 = einsum(equation = var_6454_equation_0, values = (var_5954_cast_fp16, var_6353_cast_fp16))[name = tensor("op_6454_cast_fp16")]; + tensor var_6456_equation_0 = const()[name = tensor("op_6456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6456_cast_fp16 = einsum(equation = var_6456_equation_0, values = (var_5954_cast_fp16, var_6354_cast_fp16))[name = tensor("op_6456_cast_fp16")]; + tensor var_6458_equation_0 = const()[name = tensor("op_6458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6458_cast_fp16 = einsum(equation = var_6458_equation_0, values = (var_5954_cast_fp16, var_6355_cast_fp16))[name = tensor("op_6458_cast_fp16")]; + tensor var_6460_equation_0 = const()[name = tensor("op_6460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6460_cast_fp16 = einsum(equation = var_6460_equation_0, values = (var_5954_cast_fp16, var_6356_cast_fp16))[name = tensor("op_6460_cast_fp16")]; + tensor var_6462_equation_0 = const()[name = tensor("op_6462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6462_cast_fp16 = einsum(equation = var_6462_equation_0, values = (var_5958_cast_fp16, var_6357_cast_fp16))[name = tensor("op_6462_cast_fp16")]; + tensor var_6464_equation_0 = const()[name = tensor("op_6464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6464_cast_fp16 = einsum(equation = var_6464_equation_0, values = (var_5958_cast_fp16, var_6358_cast_fp16))[name = tensor("op_6464_cast_fp16")]; + tensor var_6466_equation_0 = const()[name = tensor("op_6466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6466_cast_fp16 = einsum(equation = var_6466_equation_0, values = (var_5958_cast_fp16, var_6359_cast_fp16))[name = tensor("op_6466_cast_fp16")]; + tensor var_6468_equation_0 = const()[name = tensor("op_6468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6468_cast_fp16 = einsum(equation = var_6468_equation_0, values = (var_5958_cast_fp16, var_6360_cast_fp16))[name = tensor("op_6468_cast_fp16")]; + tensor var_6470_equation_0 = const()[name = tensor("op_6470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6470_cast_fp16 = einsum(equation = var_6470_equation_0, values = (var_5962_cast_fp16, var_6361_cast_fp16))[name = tensor("op_6470_cast_fp16")]; + tensor var_6472_equation_0 = const()[name = tensor("op_6472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6472_cast_fp16 = einsum(equation = var_6472_equation_0, values = (var_5962_cast_fp16, var_6362_cast_fp16))[name = tensor("op_6472_cast_fp16")]; + tensor var_6474_equation_0 = const()[name = tensor("op_6474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6474_cast_fp16 = einsum(equation = var_6474_equation_0, values = (var_5962_cast_fp16, var_6363_cast_fp16))[name = tensor("op_6474_cast_fp16")]; + tensor var_6476_equation_0 = const()[name = tensor("op_6476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6476_cast_fp16 = einsum(equation = var_6476_equation_0, values = (var_5962_cast_fp16, var_6364_cast_fp16))[name = tensor("op_6476_cast_fp16")]; + tensor var_6478_equation_0 = const()[name = tensor("op_6478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6478_cast_fp16 = einsum(equation = var_6478_equation_0, values = (var_5966_cast_fp16, var_6365_cast_fp16))[name = tensor("op_6478_cast_fp16")]; + tensor var_6480_equation_0 = const()[name = tensor("op_6480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6480_cast_fp16 = einsum(equation = var_6480_equation_0, values = (var_5966_cast_fp16, var_6366_cast_fp16))[name = tensor("op_6480_cast_fp16")]; + tensor var_6482_equation_0 = const()[name = tensor("op_6482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6482_cast_fp16 = einsum(equation = var_6482_equation_0, values = (var_5966_cast_fp16, var_6367_cast_fp16))[name = tensor("op_6482_cast_fp16")]; + tensor var_6484_equation_0 = const()[name = tensor("op_6484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6484_cast_fp16 = einsum(equation = var_6484_equation_0, values = (var_5966_cast_fp16, var_6368_cast_fp16))[name = tensor("op_6484_cast_fp16")]; + tensor var_6486_equation_0 = const()[name = tensor("op_6486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6486_cast_fp16 = einsum(equation = var_6486_equation_0, values = (var_5970_cast_fp16, var_6369_cast_fp16))[name = tensor("op_6486_cast_fp16")]; + tensor var_6488_equation_0 = const()[name = tensor("op_6488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6488_cast_fp16 = einsum(equation = var_6488_equation_0, values = (var_5970_cast_fp16, var_6370_cast_fp16))[name = tensor("op_6488_cast_fp16")]; + tensor var_6490_equation_0 = const()[name = tensor("op_6490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6490_cast_fp16 = einsum(equation = var_6490_equation_0, values = (var_5970_cast_fp16, var_6371_cast_fp16))[name = tensor("op_6490_cast_fp16")]; + tensor var_6492_equation_0 = const()[name = tensor("op_6492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6492_cast_fp16 = einsum(equation = var_6492_equation_0, values = (var_5970_cast_fp16, var_6372_cast_fp16))[name = tensor("op_6492_cast_fp16")]; + tensor var_6494_equation_0 = const()[name = tensor("op_6494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6494_cast_fp16 = einsum(equation = var_6494_equation_0, values = (var_5974_cast_fp16, var_6373_cast_fp16))[name = tensor("op_6494_cast_fp16")]; + tensor var_6496_equation_0 = const()[name = tensor("op_6496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6496_cast_fp16 = einsum(equation = var_6496_equation_0, values = (var_5974_cast_fp16, var_6374_cast_fp16))[name = tensor("op_6496_cast_fp16")]; + tensor var_6498_equation_0 = const()[name = tensor("op_6498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6498_cast_fp16 = einsum(equation = var_6498_equation_0, values = (var_5974_cast_fp16, var_6375_cast_fp16))[name = tensor("op_6498_cast_fp16")]; + tensor var_6500_equation_0 = const()[name = tensor("op_6500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6500_cast_fp16 = einsum(equation = var_6500_equation_0, values = (var_5974_cast_fp16, var_6376_cast_fp16))[name = tensor("op_6500_cast_fp16")]; + tensor var_6502_equation_0 = const()[name = tensor("op_6502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6502_cast_fp16 = einsum(equation = var_6502_equation_0, values = (var_5978_cast_fp16, var_6377_cast_fp16))[name = tensor("op_6502_cast_fp16")]; + tensor var_6504_equation_0 = const()[name = tensor("op_6504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6504_cast_fp16 = einsum(equation = var_6504_equation_0, values = (var_5978_cast_fp16, var_6378_cast_fp16))[name = tensor("op_6504_cast_fp16")]; + tensor var_6506_equation_0 = const()[name = tensor("op_6506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6506_cast_fp16 = einsum(equation = var_6506_equation_0, values = (var_5978_cast_fp16, var_6379_cast_fp16))[name = tensor("op_6506_cast_fp16")]; + tensor var_6508_equation_0 = const()[name = tensor("op_6508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6508_cast_fp16 = einsum(equation = var_6508_equation_0, values = (var_5978_cast_fp16, var_6380_cast_fp16))[name = tensor("op_6508_cast_fp16")]; + tensor var_6510_equation_0 = const()[name = tensor("op_6510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6510_cast_fp16 = einsum(equation = var_6510_equation_0, values = (var_5982_cast_fp16, var_6381_cast_fp16))[name = tensor("op_6510_cast_fp16")]; + tensor var_6512_equation_0 = const()[name = tensor("op_6512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6512_cast_fp16 = einsum(equation = var_6512_equation_0, values = (var_5982_cast_fp16, var_6382_cast_fp16))[name = tensor("op_6512_cast_fp16")]; + tensor var_6514_equation_0 = const()[name = tensor("op_6514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6514_cast_fp16 = einsum(equation = var_6514_equation_0, values = (var_5982_cast_fp16, var_6383_cast_fp16))[name = tensor("op_6514_cast_fp16")]; + tensor var_6516_equation_0 = const()[name = tensor("op_6516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6516_cast_fp16 = einsum(equation = var_6516_equation_0, values = (var_5982_cast_fp16, var_6384_cast_fp16))[name = tensor("op_6516_cast_fp16")]; + tensor var_6518_equation_0 = const()[name = tensor("op_6518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6518_cast_fp16 = einsum(equation = var_6518_equation_0, values = (var_5986_cast_fp16, var_6385_cast_fp16))[name = tensor("op_6518_cast_fp16")]; + tensor var_6520_equation_0 = const()[name = tensor("op_6520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6520_cast_fp16 = einsum(equation = var_6520_equation_0, values = (var_5986_cast_fp16, var_6386_cast_fp16))[name = tensor("op_6520_cast_fp16")]; + tensor var_6522_equation_0 = const()[name = tensor("op_6522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6522_cast_fp16 = einsum(equation = var_6522_equation_0, values = (var_5986_cast_fp16, var_6387_cast_fp16))[name = tensor("op_6522_cast_fp16")]; + tensor var_6524_equation_0 = const()[name = tensor("op_6524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6524_cast_fp16 = einsum(equation = var_6524_equation_0, values = (var_5986_cast_fp16, var_6388_cast_fp16))[name = tensor("op_6524_cast_fp16")]; + tensor var_6526_equation_0 = const()[name = tensor("op_6526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6526_cast_fp16 = einsum(equation = var_6526_equation_0, values = (var_5990_cast_fp16, var_6389_cast_fp16))[name = tensor("op_6526_cast_fp16")]; + tensor var_6528_equation_0 = const()[name = tensor("op_6528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6528_cast_fp16 = einsum(equation = var_6528_equation_0, values = (var_5990_cast_fp16, var_6390_cast_fp16))[name = tensor("op_6528_cast_fp16")]; + tensor var_6530_equation_0 = const()[name = tensor("op_6530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6530_cast_fp16 = einsum(equation = var_6530_equation_0, values = (var_5990_cast_fp16, var_6391_cast_fp16))[name = tensor("op_6530_cast_fp16")]; + tensor var_6532_equation_0 = const()[name = tensor("op_6532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6532_cast_fp16 = einsum(equation = var_6532_equation_0, values = (var_5990_cast_fp16, var_6392_cast_fp16))[name = tensor("op_6532_cast_fp16")]; + tensor var_6534_equation_0 = const()[name = tensor("op_6534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6534_cast_fp16 = einsum(equation = var_6534_equation_0, values = (var_5994_cast_fp16, var_6393_cast_fp16))[name = tensor("op_6534_cast_fp16")]; + tensor var_6536_equation_0 = const()[name = tensor("op_6536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6536_cast_fp16 = einsum(equation = var_6536_equation_0, values = (var_5994_cast_fp16, var_6394_cast_fp16))[name = tensor("op_6536_cast_fp16")]; + tensor var_6538_equation_0 = const()[name = tensor("op_6538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6538_cast_fp16 = einsum(equation = var_6538_equation_0, values = (var_5994_cast_fp16, var_6395_cast_fp16))[name = tensor("op_6538_cast_fp16")]; + tensor var_6540_equation_0 = const()[name = tensor("op_6540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6540_cast_fp16 = einsum(equation = var_6540_equation_0, values = (var_5994_cast_fp16, var_6396_cast_fp16))[name = tensor("op_6540_cast_fp16")]; + tensor var_6542_equation_0 = const()[name = tensor("op_6542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6542_cast_fp16 = einsum(equation = var_6542_equation_0, values = (var_5998_cast_fp16, var_6397_cast_fp16))[name = tensor("op_6542_cast_fp16")]; + tensor var_6544_equation_0 = const()[name = tensor("op_6544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6544_cast_fp16 = einsum(equation = var_6544_equation_0, values = (var_5998_cast_fp16, var_6398_cast_fp16))[name = tensor("op_6544_cast_fp16")]; + tensor var_6546_equation_0 = const()[name = tensor("op_6546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6546_cast_fp16 = einsum(equation = var_6546_equation_0, values = (var_5998_cast_fp16, var_6399_cast_fp16))[name = tensor("op_6546_cast_fp16")]; + tensor var_6548_equation_0 = const()[name = tensor("op_6548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6548_cast_fp16 = einsum(equation = var_6548_equation_0, values = (var_5998_cast_fp16, var_6400_cast_fp16))[name = tensor("op_6548_cast_fp16")]; + tensor var_6550_equation_0 = const()[name = tensor("op_6550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6550_cast_fp16 = einsum(equation = var_6550_equation_0, values = (var_6002_cast_fp16, var_6401_cast_fp16))[name = tensor("op_6550_cast_fp16")]; + tensor var_6552_equation_0 = const()[name = tensor("op_6552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6552_cast_fp16 = einsum(equation = var_6552_equation_0, values = (var_6002_cast_fp16, var_6402_cast_fp16))[name = tensor("op_6552_cast_fp16")]; + tensor var_6554_equation_0 = const()[name = tensor("op_6554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6554_cast_fp16 = einsum(equation = var_6554_equation_0, values = (var_6002_cast_fp16, var_6403_cast_fp16))[name = tensor("op_6554_cast_fp16")]; + tensor var_6556_equation_0 = const()[name = tensor("op_6556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6556_cast_fp16 = einsum(equation = var_6556_equation_0, values = (var_6002_cast_fp16, var_6404_cast_fp16))[name = tensor("op_6556_cast_fp16")]; + tensor var_6558_equation_0 = const()[name = tensor("op_6558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6558_cast_fp16 = einsum(equation = var_6558_equation_0, values = (var_6006_cast_fp16, var_6405_cast_fp16))[name = tensor("op_6558_cast_fp16")]; + tensor var_6560_equation_0 = const()[name = tensor("op_6560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6560_cast_fp16 = einsum(equation = var_6560_equation_0, values = (var_6006_cast_fp16, var_6406_cast_fp16))[name = tensor("op_6560_cast_fp16")]; + tensor var_6562_equation_0 = const()[name = tensor("op_6562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6562_cast_fp16 = einsum(equation = var_6562_equation_0, values = (var_6006_cast_fp16, var_6407_cast_fp16))[name = tensor("op_6562_cast_fp16")]; + tensor var_6564_equation_0 = const()[name = tensor("op_6564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6564_cast_fp16 = einsum(equation = var_6564_equation_0, values = (var_6006_cast_fp16, var_6408_cast_fp16))[name = tensor("op_6564_cast_fp16")]; + tensor var_6566_equation_0 = const()[name = tensor("op_6566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6566_cast_fp16 = einsum(equation = var_6566_equation_0, values = (var_6010_cast_fp16, var_6409_cast_fp16))[name = tensor("op_6566_cast_fp16")]; + tensor var_6568_equation_0 = const()[name = tensor("op_6568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6568_cast_fp16 = einsum(equation = var_6568_equation_0, values = (var_6010_cast_fp16, var_6410_cast_fp16))[name = tensor("op_6568_cast_fp16")]; + tensor var_6570_equation_0 = const()[name = tensor("op_6570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6570_cast_fp16 = einsum(equation = var_6570_equation_0, values = (var_6010_cast_fp16, var_6411_cast_fp16))[name = tensor("op_6570_cast_fp16")]; + tensor var_6572_equation_0 = const()[name = tensor("op_6572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_6572_cast_fp16 = einsum(equation = var_6572_equation_0, values = (var_6010_cast_fp16, var_6412_cast_fp16))[name = tensor("op_6572_cast_fp16")]; + tensor var_6574_interleave_0 = const()[name = tensor("op_6574_interleave_0"), val = tensor(false)]; + tensor var_6574_cast_fp16 = concat(axis = var_5079, interleave = var_6574_interleave_0, values = (var_6414_cast_fp16, var_6416_cast_fp16, var_6418_cast_fp16, var_6420_cast_fp16))[name = tensor("op_6574_cast_fp16")]; + tensor var_6576_interleave_0 = const()[name = tensor("op_6576_interleave_0"), val = tensor(false)]; + tensor var_6576_cast_fp16 = concat(axis = var_5079, interleave = var_6576_interleave_0, values = (var_6422_cast_fp16, var_6424_cast_fp16, var_6426_cast_fp16, var_6428_cast_fp16))[name = tensor("op_6576_cast_fp16")]; + tensor var_6578_interleave_0 = const()[name = tensor("op_6578_interleave_0"), val = tensor(false)]; + tensor var_6578_cast_fp16 = concat(axis = var_5079, interleave = var_6578_interleave_0, values = (var_6430_cast_fp16, var_6432_cast_fp16, var_6434_cast_fp16, var_6436_cast_fp16))[name = tensor("op_6578_cast_fp16")]; + tensor var_6580_interleave_0 = const()[name = tensor("op_6580_interleave_0"), val = tensor(false)]; + tensor var_6580_cast_fp16 = concat(axis = var_5079, interleave = var_6580_interleave_0, values = (var_6438_cast_fp16, var_6440_cast_fp16, var_6442_cast_fp16, var_6444_cast_fp16))[name = tensor("op_6580_cast_fp16")]; + tensor var_6582_interleave_0 = const()[name = tensor("op_6582_interleave_0"), val = tensor(false)]; + tensor var_6582_cast_fp16 = concat(axis = var_5079, interleave = var_6582_interleave_0, values = (var_6446_cast_fp16, var_6448_cast_fp16, var_6450_cast_fp16, var_6452_cast_fp16))[name = tensor("op_6582_cast_fp16")]; + tensor var_6584_interleave_0 = const()[name = tensor("op_6584_interleave_0"), val = tensor(false)]; + tensor var_6584_cast_fp16 = concat(axis = var_5079, interleave = var_6584_interleave_0, values = (var_6454_cast_fp16, var_6456_cast_fp16, var_6458_cast_fp16, var_6460_cast_fp16))[name = tensor("op_6584_cast_fp16")]; + tensor var_6586_interleave_0 = const()[name = tensor("op_6586_interleave_0"), val = tensor(false)]; + tensor var_6586_cast_fp16 = concat(axis = var_5079, interleave = var_6586_interleave_0, values = (var_6462_cast_fp16, var_6464_cast_fp16, var_6466_cast_fp16, var_6468_cast_fp16))[name = tensor("op_6586_cast_fp16")]; + tensor var_6588_interleave_0 = const()[name = tensor("op_6588_interleave_0"), val = tensor(false)]; + tensor var_6588_cast_fp16 = concat(axis = var_5079, interleave = var_6588_interleave_0, values = (var_6470_cast_fp16, var_6472_cast_fp16, var_6474_cast_fp16, var_6476_cast_fp16))[name = tensor("op_6588_cast_fp16")]; + tensor var_6590_interleave_0 = const()[name = tensor("op_6590_interleave_0"), val = tensor(false)]; + tensor var_6590_cast_fp16 = concat(axis = var_5079, interleave = var_6590_interleave_0, values = (var_6478_cast_fp16, var_6480_cast_fp16, var_6482_cast_fp16, var_6484_cast_fp16))[name = tensor("op_6590_cast_fp16")]; + tensor var_6592_interleave_0 = const()[name = tensor("op_6592_interleave_0"), val = tensor(false)]; + tensor var_6592_cast_fp16 = concat(axis = var_5079, interleave = var_6592_interleave_0, values = (var_6486_cast_fp16, var_6488_cast_fp16, var_6490_cast_fp16, var_6492_cast_fp16))[name = tensor("op_6592_cast_fp16")]; + tensor var_6594_interleave_0 = const()[name = tensor("op_6594_interleave_0"), val = tensor(false)]; + tensor var_6594_cast_fp16 = concat(axis = var_5079, interleave = var_6594_interleave_0, values = (var_6494_cast_fp16, var_6496_cast_fp16, var_6498_cast_fp16, var_6500_cast_fp16))[name = tensor("op_6594_cast_fp16")]; + tensor var_6596_interleave_0 = const()[name = tensor("op_6596_interleave_0"), val = tensor(false)]; + tensor var_6596_cast_fp16 = concat(axis = var_5079, interleave = var_6596_interleave_0, values = (var_6502_cast_fp16, var_6504_cast_fp16, var_6506_cast_fp16, var_6508_cast_fp16))[name = tensor("op_6596_cast_fp16")]; + tensor var_6598_interleave_0 = const()[name = tensor("op_6598_interleave_0"), val = tensor(false)]; + tensor var_6598_cast_fp16 = concat(axis = var_5079, interleave = var_6598_interleave_0, values = (var_6510_cast_fp16, var_6512_cast_fp16, var_6514_cast_fp16, var_6516_cast_fp16))[name = tensor("op_6598_cast_fp16")]; + tensor var_6600_interleave_0 = const()[name = tensor("op_6600_interleave_0"), val = tensor(false)]; + tensor var_6600_cast_fp16 = concat(axis = var_5079, interleave = var_6600_interleave_0, values = (var_6518_cast_fp16, var_6520_cast_fp16, var_6522_cast_fp16, var_6524_cast_fp16))[name = tensor("op_6600_cast_fp16")]; + tensor var_6602_interleave_0 = const()[name = tensor("op_6602_interleave_0"), val = tensor(false)]; + tensor var_6602_cast_fp16 = concat(axis = var_5079, interleave = var_6602_interleave_0, values = (var_6526_cast_fp16, var_6528_cast_fp16, var_6530_cast_fp16, var_6532_cast_fp16))[name = tensor("op_6602_cast_fp16")]; + tensor var_6604_interleave_0 = const()[name = tensor("op_6604_interleave_0"), val = tensor(false)]; + tensor var_6604_cast_fp16 = concat(axis = var_5079, interleave = var_6604_interleave_0, values = (var_6534_cast_fp16, var_6536_cast_fp16, var_6538_cast_fp16, var_6540_cast_fp16))[name = tensor("op_6604_cast_fp16")]; + tensor var_6606_interleave_0 = const()[name = tensor("op_6606_interleave_0"), val = tensor(false)]; + tensor var_6606_cast_fp16 = concat(axis = var_5079, interleave = var_6606_interleave_0, values = (var_6542_cast_fp16, var_6544_cast_fp16, var_6546_cast_fp16, var_6548_cast_fp16))[name = tensor("op_6606_cast_fp16")]; + tensor var_6608_interleave_0 = const()[name = tensor("op_6608_interleave_0"), val = tensor(false)]; + tensor var_6608_cast_fp16 = concat(axis = var_5079, interleave = var_6608_interleave_0, values = (var_6550_cast_fp16, var_6552_cast_fp16, var_6554_cast_fp16, var_6556_cast_fp16))[name = tensor("op_6608_cast_fp16")]; + tensor var_6610_interleave_0 = const()[name = tensor("op_6610_interleave_0"), val = tensor(false)]; + tensor var_6610_cast_fp16 = concat(axis = var_5079, interleave = var_6610_interleave_0, values = (var_6558_cast_fp16, var_6560_cast_fp16, var_6562_cast_fp16, var_6564_cast_fp16))[name = tensor("op_6610_cast_fp16")]; + tensor var_6612_interleave_0 = const()[name = tensor("op_6612_interleave_0"), val = tensor(false)]; + tensor var_6612_cast_fp16 = concat(axis = var_5079, interleave = var_6612_interleave_0, values = (var_6566_cast_fp16, var_6568_cast_fp16, var_6570_cast_fp16, var_6572_cast_fp16))[name = tensor("op_6612_cast_fp16")]; + tensor input_67_interleave_0 = const()[name = tensor("input_67_interleave_0"), val = tensor(false)]; + tensor input_67_cast_fp16 = concat(axis = var_5104, interleave = input_67_interleave_0, values = (var_6574_cast_fp16, var_6576_cast_fp16, var_6578_cast_fp16, var_6580_cast_fp16, var_6582_cast_fp16, var_6584_cast_fp16, var_6586_cast_fp16, var_6588_cast_fp16, var_6590_cast_fp16, var_6592_cast_fp16, var_6594_cast_fp16, var_6596_cast_fp16, var_6598_cast_fp16, var_6600_cast_fp16, var_6602_cast_fp16, var_6604_cast_fp16, var_6606_cast_fp16, var_6608_cast_fp16, var_6610_cast_fp16, var_6612_cast_fp16))[name = tensor("input_67_cast_fp16")]; + tensor var_6620 = const()[name = tensor("op_6620"), val = tensor([1, 1])]; + tensor var_6622 = const()[name = tensor("op_6622"), val = tensor([1, 1])]; + tensor pretrained_out_43_pad_type_0 = const()[name = tensor("pretrained_out_43_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_43_pad_0 = const()[name = tensor("pretrained_out_43_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48823360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49642624))), name = tensor("layers_3_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_3_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49642752)))]; + tensor pretrained_out_43_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_6622, groups = var_5104, pad = pretrained_out_43_pad_0, pad_type = pretrained_out_43_pad_type_0, strides = var_6620, weight = layers_3_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor("pretrained_out_43_cast_fp16")]; + tensor var_6626 = const()[name = tensor("op_6626"), val = tensor([1, 1])]; + tensor var_6628 = const()[name = tensor("op_6628"), val = tensor([1, 1])]; + tensor input_69_pad_type_0 = const()[name = tensor("input_69_pad_type_0"), val = tensor("custom")]; + tensor input_69_pad_0 = const()[name = tensor("input_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49645376)))]; + tensor input_69_cast_fp16 = conv(dilations = var_6628, groups = var_5104, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = var_6626, weight = layers_3_self_attn_o_proj_loraA_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor var_6632 = const()[name = tensor("op_6632"), val = tensor([1, 1])]; + tensor var_6634 = const()[name = tensor("op_6634"), val = tensor([1, 1])]; + tensor lora_out_85_pad_type_0 = const()[name = tensor("lora_out_85_pad_type_0"), val = tensor("custom")]; + tensor lora_out_85_pad_0 = const()[name = tensor("lora_out_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_87_weight_0_to_fp16 = const()[name = tensor("lora_out_87_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49686400)))]; + tensor lora_out_87_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_6634, groups = var_5104, pad = lora_out_85_pad_0, pad_type = lora_out_85_pad_type_0, strides = var_6632, weight = lora_out_87_weight_0_to_fp16, x = input_69_cast_fp16)[name = tensor("lora_out_87_cast_fp16")]; + tensor obj_15_cast_fp16 = add(x = pretrained_out_43_cast_fp16, y = lora_out_87_cast_fp16)[name = tensor("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor var_6643 = const()[name = tensor("op_6643"), val = tensor([1])]; + tensor channels_mean_15_cast_fp16 = reduce_mean(axes = var_6643, keep_dims = var_5105, x = inputs_15_cast_fp16)[name = tensor("channels_mean_15_cast_fp16")]; + tensor zero_mean_15_cast_fp16 = sub(x = inputs_15_cast_fp16, y = channels_mean_15_cast_fp16)[name = tensor("zero_mean_15_cast_fp16")]; + tensor zero_mean_sq_15_cast_fp16 = mul(x = zero_mean_15_cast_fp16, y = zero_mean_15_cast_fp16)[name = tensor("zero_mean_sq_15_cast_fp16")]; + tensor var_6647 = const()[name = tensor("op_6647"), val = tensor([1])]; + tensor var_6648_cast_fp16 = reduce_mean(axes = var_6647, keep_dims = var_5105, x = zero_mean_sq_15_cast_fp16)[name = tensor("op_6648_cast_fp16")]; + tensor var_6649_to_fp16 = const()[name = tensor("op_6649_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_6650_cast_fp16 = add(x = var_6648_cast_fp16, y = var_6649_to_fp16)[name = tensor("op_6650_cast_fp16")]; + tensor denom_15_epsilon_0 = const()[name = tensor("denom_15_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_15_cast_fp16 = rsqrt(epsilon = denom_15_epsilon_0, x = var_6650_cast_fp16)[name = tensor("denom_15_cast_fp16")]; + tensor out_15_cast_fp16 = mul(x = zero_mean_15_cast_fp16, y = denom_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; + tensor input_71_gamma_0_to_fp16 = const()[name = tensor("input_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49727424)))]; + tensor input_71_beta_0_to_fp16 = const()[name = tensor("input_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49730048)))]; + tensor input_71_epsilon_0_to_fp16 = const()[name = tensor("input_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_71_cast_fp16 = batch_norm(beta = input_71_beta_0_to_fp16, epsilon = input_71_epsilon_0_to_fp16, gamma = input_71_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor var_6664 = const()[name = tensor("op_6664"), val = tensor([1, 1])]; + tensor var_6666 = const()[name = tensor("op_6666"), val = tensor([1, 1])]; + tensor pretrained_out_45_pad_type_0 = const()[name = tensor("pretrained_out_45_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_45_pad_0 = const()[name = tensor("pretrained_out_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49732672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53009536))), name = tensor("layers_3_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_3_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_3_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53009664)))]; + tensor pretrained_out_45_cast_fp16 = conv(bias = layers_3_fc1_pretrained_bias_to_fp16, dilations = var_6666, groups = var_5104, pad = pretrained_out_45_pad_0, pad_type = pretrained_out_45_pad_type_0, strides = var_6664, weight = layers_3_fc1_pretrained_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = tensor("pretrained_out_45_cast_fp16")]; + tensor var_6670 = const()[name = tensor("op_6670"), val = tensor([1, 1])]; + tensor var_6672 = const()[name = tensor("op_6672"), val = tensor([1, 1])]; + tensor input_73_pad_type_0 = const()[name = tensor("input_73_pad_type_0"), val = tensor("custom")]; + tensor input_73_pad_0 = const()[name = tensor("input_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_3_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53019968)))]; + tensor input_73_cast_fp16 = conv(dilations = var_6672, groups = var_5104, pad = input_73_pad_0, pad_type = input_73_pad_type_0, strides = var_6670, weight = layers_3_fc1_loraA_weight_to_fp16, x = input_71_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor var_6676 = const()[name = tensor("op_6676"), val = tensor([1, 1])]; + tensor var_6678 = const()[name = tensor("op_6678"), val = tensor([1, 1])]; + tensor lora_out_89_pad_type_0 = const()[name = tensor("lora_out_89_pad_type_0"), val = tensor("custom")]; + tensor lora_out_89_pad_0 = const()[name = tensor("lora_out_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_91_weight_0_to_fp16 = const()[name = tensor("lora_out_91_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53060992)))]; + tensor lora_out_91_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_6678, groups = var_5104, pad = lora_out_89_pad_0, pad_type = lora_out_89_pad_type_0, strides = var_6676, weight = lora_out_91_weight_0_to_fp16, x = input_73_cast_fp16)[name = tensor("lora_out_91_cast_fp16")]; + tensor input_75_cast_fp16 = add(x = pretrained_out_45_cast_fp16, y = lora_out_91_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor input_77_mode_0 = const()[name = tensor("input_77_mode_0"), val = tensor("EXACT")]; + tensor input_77_cast_fp16 = gelu(mode = input_77_mode_0, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor var_6690 = const()[name = tensor("op_6690"), val = tensor([1, 1])]; + tensor var_6692 = const()[name = tensor("op_6692"), val = tensor([1, 1])]; + tensor pretrained_out_47_pad_type_0 = const()[name = tensor("pretrained_out_47_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_47_pad_0 = const()[name = tensor("pretrained_out_47_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53224896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56501760))), name = tensor("layers_3_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_3_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_3_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56501888)))]; + tensor pretrained_out_47_cast_fp16 = conv(bias = layers_3_fc2_pretrained_bias_to_fp16, dilations = var_6692, groups = var_5104, pad = pretrained_out_47_pad_0, pad_type = pretrained_out_47_pad_type_0, strides = var_6690, weight = layers_3_fc2_pretrained_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("pretrained_out_47_cast_fp16")]; + tensor var_6696 = const()[name = tensor("op_6696"), val = tensor([1, 1])]; + tensor var_6698 = const()[name = tensor("op_6698"), val = tensor([1, 1])]; + tensor input_79_pad_type_0 = const()[name = tensor("input_79_pad_type_0"), val = tensor("custom")]; + tensor input_79_pad_0 = const()[name = tensor("input_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_3_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_3_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56504512)))]; + tensor input_79_cast_fp16 = conv(dilations = var_6698, groups = var_5104, pad = input_79_pad_0, pad_type = input_79_pad_type_0, strides = var_6696, weight = layers_3_fc2_loraA_weight_to_fp16, x = input_77_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor var_6702 = const()[name = tensor("op_6702"), val = tensor([1, 1])]; + tensor var_6704 = const()[name = tensor("op_6704"), val = tensor([1, 1])]; + tensor lora_out_93_pad_type_0 = const()[name = tensor("lora_out_93_pad_type_0"), val = tensor("custom")]; + tensor lora_out_93_pad_0 = const()[name = tensor("lora_out_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_95_weight_0_to_fp16 = const()[name = tensor("lora_out_95_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56668416)))]; + tensor lora_out_95_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_6704, groups = var_5104, pad = lora_out_93_pad_0, pad_type = lora_out_93_pad_type_0, strides = var_6702, weight = lora_out_95_weight_0_to_fp16, x = input_79_cast_fp16)[name = tensor("lora_out_95_cast_fp16")]; + tensor hidden_states_11_cast_fp16 = add(x = pretrained_out_47_cast_fp16, y = lora_out_95_cast_fp16)[name = tensor("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor var_6714 = const()[name = tensor("op_6714"), val = tensor(3)]; + tensor var_6739 = const()[name = tensor("op_6739"), val = tensor(1)]; + tensor var_6740 = const()[name = tensor("op_6740"), val = tensor(true)]; + tensor var_6750 = const()[name = tensor("op_6750"), val = tensor([1])]; + tensor channels_mean_17_cast_fp16 = reduce_mean(axes = var_6750, keep_dims = var_6740, x = inputs_17_cast_fp16)[name = tensor("channels_mean_17_cast_fp16")]; + tensor zero_mean_17_cast_fp16 = sub(x = inputs_17_cast_fp16, y = channels_mean_17_cast_fp16)[name = tensor("zero_mean_17_cast_fp16")]; + tensor zero_mean_sq_17_cast_fp16 = mul(x = zero_mean_17_cast_fp16, y = zero_mean_17_cast_fp16)[name = tensor("zero_mean_sq_17_cast_fp16")]; + tensor var_6754 = const()[name = tensor("op_6754"), val = tensor([1])]; + tensor var_6755_cast_fp16 = reduce_mean(axes = var_6754, keep_dims = var_6740, x = zero_mean_sq_17_cast_fp16)[name = tensor("op_6755_cast_fp16")]; + tensor var_6756_to_fp16 = const()[name = tensor("op_6756_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_6757_cast_fp16 = add(x = var_6755_cast_fp16, y = var_6756_to_fp16)[name = tensor("op_6757_cast_fp16")]; + tensor denom_17_epsilon_0 = const()[name = tensor("denom_17_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_17_cast_fp16 = rsqrt(epsilon = denom_17_epsilon_0, x = var_6757_cast_fp16)[name = tensor("denom_17_cast_fp16")]; + tensor out_17_cast_fp16 = mul(x = zero_mean_17_cast_fp16, y = denom_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = tensor("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56709440)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = tensor("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56712064)))]; + tensor obj_17_epsilon_0_to_fp16 = const()[name = tensor("obj_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor("obj_17_cast_fp16")]; + tensor var_6775 = const()[name = tensor("op_6775"), val = tensor([1, 1])]; + tensor var_6777 = const()[name = tensor("op_6777"), val = tensor([1, 1])]; + tensor pretrained_out_49_pad_type_0 = const()[name = tensor("pretrained_out_49_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_49_pad_0 = const()[name = tensor("pretrained_out_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56714688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57533952))), name = tensor("layers_4_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_4_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57534080)))]; + tensor pretrained_out_49_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_6777, groups = var_6739, pad = pretrained_out_49_pad_0, pad_type = pretrained_out_49_pad_type_0, strides = var_6775, weight = layers_4_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor("pretrained_out_49_cast_fp16")]; + tensor var_6781 = const()[name = tensor("op_6781"), val = tensor([1, 1])]; + tensor var_6783 = const()[name = tensor("op_6783"), val = tensor([1, 1])]; + tensor input_81_pad_type_0 = const()[name = tensor("input_81_pad_type_0"), val = tensor("custom")]; + tensor input_81_pad_0 = const()[name = tensor("input_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57536704)))]; + tensor input_81_cast_fp16 = conv(dilations = var_6783, groups = var_6739, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = var_6781, weight = layers_4_self_attn_q_proj_loraA_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor var_6787 = const()[name = tensor("op_6787"), val = tensor([1, 1])]; + tensor var_6789 = const()[name = tensor("op_6789"), val = tensor([1, 1])]; + tensor lora_out_97_pad_type_0 = const()[name = tensor("lora_out_97_pad_type_0"), val = tensor("custom")]; + tensor lora_out_97_pad_0 = const()[name = tensor("lora_out_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_99_weight_0_to_fp16 = const()[name = tensor("lora_out_99_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57577728)))]; + tensor lora_out_99_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_6789, groups = var_6739, pad = lora_out_97_pad_0, pad_type = lora_out_97_pad_type_0, strides = var_6787, weight = lora_out_99_weight_0_to_fp16, x = input_81_cast_fp16)[name = tensor("lora_out_99_cast_fp16")]; + tensor query_9_cast_fp16 = add(x = pretrained_out_49_cast_fp16, y = lora_out_99_cast_fp16)[name = tensor("query_9_cast_fp16")]; + tensor var_6799 = const()[name = tensor("op_6799"), val = tensor([1, 1])]; + tensor var_6801 = const()[name = tensor("op_6801"), val = tensor([1, 1])]; + tensor pretrained_out_51_pad_type_0 = const()[name = tensor("pretrained_out_51_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_51_pad_0 = const()[name = tensor("pretrained_out_51_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57618752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58438016))), name = tensor("layers_4_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_51_cast_fp16 = conv(dilations = var_6801, groups = var_6739, pad = pretrained_out_51_pad_0, pad_type = pretrained_out_51_pad_type_0, strides = var_6799, weight = layers_4_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor("pretrained_out_51_cast_fp16")]; + tensor var_6805 = const()[name = tensor("op_6805"), val = tensor([1, 1])]; + tensor var_6807 = const()[name = tensor("op_6807"), val = tensor([1, 1])]; + tensor input_83_pad_type_0 = const()[name = tensor("input_83_pad_type_0"), val = tensor("custom")]; + tensor input_83_pad_0 = const()[name = tensor("input_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58438144)))]; + tensor input_83_cast_fp16 = conv(dilations = var_6807, groups = var_6739, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = var_6805, weight = layers_4_self_attn_k_proj_loraA_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor var_6811 = const()[name = tensor("op_6811"), val = tensor([1, 1])]; + tensor var_6813 = const()[name = tensor("op_6813"), val = tensor([1, 1])]; + tensor lora_out_101_pad_type_0 = const()[name = tensor("lora_out_101_pad_type_0"), val = tensor("custom")]; + tensor lora_out_101_pad_0 = const()[name = tensor("lora_out_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_103_weight_0_to_fp16 = const()[name = tensor("lora_out_103_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58479168)))]; + tensor lora_out_103_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_6813, groups = var_6739, pad = lora_out_101_pad_0, pad_type = lora_out_101_pad_type_0, strides = var_6811, weight = lora_out_103_weight_0_to_fp16, x = input_83_cast_fp16)[name = tensor("lora_out_103_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = pretrained_out_51_cast_fp16, y = lora_out_103_cast_fp16)[name = tensor("key_9_cast_fp16")]; + tensor var_6824 = const()[name = tensor("op_6824"), val = tensor([1, 1])]; + tensor var_6826 = const()[name = tensor("op_6826"), val = tensor([1, 1])]; + tensor pretrained_out_53_pad_type_0 = const()[name = tensor("pretrained_out_53_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_53_pad_0 = const()[name = tensor("pretrained_out_53_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58520192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59339456))), name = tensor("layers_4_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_4_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59339584)))]; + tensor pretrained_out_53_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_6826, groups = var_6739, pad = pretrained_out_53_pad_0, pad_type = pretrained_out_53_pad_type_0, strides = var_6824, weight = layers_4_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor("pretrained_out_53_cast_fp16")]; + tensor var_6830 = const()[name = tensor("op_6830"), val = tensor([1, 1])]; + tensor var_6832 = const()[name = tensor("op_6832"), val = tensor([1, 1])]; + tensor input_85_pad_type_0 = const()[name = tensor("input_85_pad_type_0"), val = tensor("custom")]; + tensor input_85_pad_0 = const()[name = tensor("input_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59342208)))]; + tensor input_85_cast_fp16 = conv(dilations = var_6832, groups = var_6739, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = var_6830, weight = layers_4_self_attn_v_proj_loraA_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor var_6836 = const()[name = tensor("op_6836"), val = tensor([1, 1])]; + tensor var_6838 = const()[name = tensor("op_6838"), val = tensor([1, 1])]; + tensor lora_out_105_pad_type_0 = const()[name = tensor("lora_out_105_pad_type_0"), val = tensor("custom")]; + tensor lora_out_105_pad_0 = const()[name = tensor("lora_out_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_107_weight_0_to_fp16 = const()[name = tensor("lora_out_107_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59383232)))]; + tensor lora_out_107_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_6838, groups = var_6739, pad = lora_out_105_pad_0, pad_type = lora_out_105_pad_type_0, strides = var_6836, weight = lora_out_107_weight_0_to_fp16, x = input_85_cast_fp16)[name = tensor("lora_out_107_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = pretrained_out_53_cast_fp16, y = lora_out_107_cast_fp16)[name = tensor("value_9_cast_fp16")]; + tensor var_6848_begin_0 = const()[name = tensor("op_6848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6848_end_0 = const()[name = tensor("op_6848_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6848_end_mask_0 = const()[name = tensor("op_6848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6848_cast_fp16 = slice_by_index(begin = var_6848_begin_0, end = var_6848_end_0, end_mask = var_6848_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6848_cast_fp16")]; + tensor var_6852_begin_0 = const()[name = tensor("op_6852_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_6852_end_0 = const()[name = tensor("op_6852_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_6852_end_mask_0 = const()[name = tensor("op_6852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6852_cast_fp16 = slice_by_index(begin = var_6852_begin_0, end = var_6852_end_0, end_mask = var_6852_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6852_cast_fp16")]; + tensor var_6856_begin_0 = const()[name = tensor("op_6856_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_6856_end_0 = const()[name = tensor("op_6856_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_6856_end_mask_0 = const()[name = tensor("op_6856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6856_cast_fp16 = slice_by_index(begin = var_6856_begin_0, end = var_6856_end_0, end_mask = var_6856_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6856_cast_fp16")]; + tensor var_6860_begin_0 = const()[name = tensor("op_6860_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_6860_end_0 = const()[name = tensor("op_6860_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_6860_end_mask_0 = const()[name = tensor("op_6860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6860_cast_fp16 = slice_by_index(begin = var_6860_begin_0, end = var_6860_end_0, end_mask = var_6860_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6860_cast_fp16")]; + tensor var_6864_begin_0 = const()[name = tensor("op_6864_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_6864_end_0 = const()[name = tensor("op_6864_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_6864_end_mask_0 = const()[name = tensor("op_6864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6864_cast_fp16 = slice_by_index(begin = var_6864_begin_0, end = var_6864_end_0, end_mask = var_6864_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6864_cast_fp16")]; + tensor var_6868_begin_0 = const()[name = tensor("op_6868_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_6868_end_0 = const()[name = tensor("op_6868_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_6868_end_mask_0 = const()[name = tensor("op_6868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6868_cast_fp16 = slice_by_index(begin = var_6868_begin_0, end = var_6868_end_0, end_mask = var_6868_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6868_cast_fp16")]; + tensor var_6872_begin_0 = const()[name = tensor("op_6872_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_6872_end_0 = const()[name = tensor("op_6872_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_6872_end_mask_0 = const()[name = tensor("op_6872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6872_cast_fp16 = slice_by_index(begin = var_6872_begin_0, end = var_6872_end_0, end_mask = var_6872_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6872_cast_fp16")]; + tensor var_6876_begin_0 = const()[name = tensor("op_6876_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_6876_end_0 = const()[name = tensor("op_6876_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_6876_end_mask_0 = const()[name = tensor("op_6876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6876_cast_fp16 = slice_by_index(begin = var_6876_begin_0, end = var_6876_end_0, end_mask = var_6876_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6876_cast_fp16")]; + tensor var_6880_begin_0 = const()[name = tensor("op_6880_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_6880_end_0 = const()[name = tensor("op_6880_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_6880_end_mask_0 = const()[name = tensor("op_6880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6880_cast_fp16 = slice_by_index(begin = var_6880_begin_0, end = var_6880_end_0, end_mask = var_6880_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6880_cast_fp16")]; + tensor var_6884_begin_0 = const()[name = tensor("op_6884_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_6884_end_0 = const()[name = tensor("op_6884_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_6884_end_mask_0 = const()[name = tensor("op_6884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6884_cast_fp16 = slice_by_index(begin = var_6884_begin_0, end = var_6884_end_0, end_mask = var_6884_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6884_cast_fp16")]; + tensor var_6888_begin_0 = const()[name = tensor("op_6888_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_6888_end_0 = const()[name = tensor("op_6888_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_6888_end_mask_0 = const()[name = tensor("op_6888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6888_cast_fp16 = slice_by_index(begin = var_6888_begin_0, end = var_6888_end_0, end_mask = var_6888_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6888_cast_fp16")]; + tensor var_6892_begin_0 = const()[name = tensor("op_6892_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_6892_end_0 = const()[name = tensor("op_6892_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_6892_end_mask_0 = const()[name = tensor("op_6892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6892_cast_fp16 = slice_by_index(begin = var_6892_begin_0, end = var_6892_end_0, end_mask = var_6892_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6892_cast_fp16")]; + tensor var_6896_begin_0 = const()[name = tensor("op_6896_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_6896_end_0 = const()[name = tensor("op_6896_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_6896_end_mask_0 = const()[name = tensor("op_6896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6896_cast_fp16 = slice_by_index(begin = var_6896_begin_0, end = var_6896_end_0, end_mask = var_6896_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6896_cast_fp16")]; + tensor var_6900_begin_0 = const()[name = tensor("op_6900_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_6900_end_0 = const()[name = tensor("op_6900_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_6900_end_mask_0 = const()[name = tensor("op_6900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6900_cast_fp16 = slice_by_index(begin = var_6900_begin_0, end = var_6900_end_0, end_mask = var_6900_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6900_cast_fp16")]; + tensor var_6904_begin_0 = const()[name = tensor("op_6904_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_6904_end_0 = const()[name = tensor("op_6904_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_6904_end_mask_0 = const()[name = tensor("op_6904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6904_cast_fp16 = slice_by_index(begin = var_6904_begin_0, end = var_6904_end_0, end_mask = var_6904_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6904_cast_fp16")]; + tensor var_6908_begin_0 = const()[name = tensor("op_6908_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_6908_end_0 = const()[name = tensor("op_6908_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_6908_end_mask_0 = const()[name = tensor("op_6908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6908_cast_fp16 = slice_by_index(begin = var_6908_begin_0, end = var_6908_end_0, end_mask = var_6908_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6908_cast_fp16")]; + tensor var_6912_begin_0 = const()[name = tensor("op_6912_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_6912_end_0 = const()[name = tensor("op_6912_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_6912_end_mask_0 = const()[name = tensor("op_6912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6912_cast_fp16 = slice_by_index(begin = var_6912_begin_0, end = var_6912_end_0, end_mask = var_6912_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6912_cast_fp16")]; + tensor var_6916_begin_0 = const()[name = tensor("op_6916_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_6916_end_0 = const()[name = tensor("op_6916_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_6916_end_mask_0 = const()[name = tensor("op_6916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6916_cast_fp16 = slice_by_index(begin = var_6916_begin_0, end = var_6916_end_0, end_mask = var_6916_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6916_cast_fp16")]; + tensor var_6920_begin_0 = const()[name = tensor("op_6920_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_6920_end_0 = const()[name = tensor("op_6920_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_6920_end_mask_0 = const()[name = tensor("op_6920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6920_cast_fp16 = slice_by_index(begin = var_6920_begin_0, end = var_6920_end_0, end_mask = var_6920_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6920_cast_fp16")]; + tensor var_6924_begin_0 = const()[name = tensor("op_6924_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_6924_end_0 = const()[name = tensor("op_6924_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_6924_end_mask_0 = const()[name = tensor("op_6924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_6924_cast_fp16 = slice_by_index(begin = var_6924_begin_0, end = var_6924_end_0, end_mask = var_6924_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_6924_cast_fp16")]; + tensor var_6933_begin_0 = const()[name = tensor("op_6933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6933_end_0 = const()[name = tensor("op_6933_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6933_end_mask_0 = const()[name = tensor("op_6933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6933_cast_fp16 = slice_by_index(begin = var_6933_begin_0, end = var_6933_end_0, end_mask = var_6933_end_mask_0, x = var_6848_cast_fp16)[name = tensor("op_6933_cast_fp16")]; + tensor var_6940_begin_0 = const()[name = tensor("op_6940_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6940_end_0 = const()[name = tensor("op_6940_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6940_end_mask_0 = const()[name = tensor("op_6940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6940_cast_fp16 = slice_by_index(begin = var_6940_begin_0, end = var_6940_end_0, end_mask = var_6940_end_mask_0, x = var_6848_cast_fp16)[name = tensor("op_6940_cast_fp16")]; + tensor var_6947_begin_0 = const()[name = tensor("op_6947_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6947_end_0 = const()[name = tensor("op_6947_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6947_end_mask_0 = const()[name = tensor("op_6947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6947_cast_fp16 = slice_by_index(begin = var_6947_begin_0, end = var_6947_end_0, end_mask = var_6947_end_mask_0, x = var_6848_cast_fp16)[name = tensor("op_6947_cast_fp16")]; + tensor var_6954_begin_0 = const()[name = tensor("op_6954_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6954_end_0 = const()[name = tensor("op_6954_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6954_end_mask_0 = const()[name = tensor("op_6954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6954_cast_fp16 = slice_by_index(begin = var_6954_begin_0, end = var_6954_end_0, end_mask = var_6954_end_mask_0, x = var_6848_cast_fp16)[name = tensor("op_6954_cast_fp16")]; + tensor var_6961_begin_0 = const()[name = tensor("op_6961_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6961_end_0 = const()[name = tensor("op_6961_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6961_end_mask_0 = const()[name = tensor("op_6961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6961_cast_fp16 = slice_by_index(begin = var_6961_begin_0, end = var_6961_end_0, end_mask = var_6961_end_mask_0, x = var_6852_cast_fp16)[name = tensor("op_6961_cast_fp16")]; + tensor var_6968_begin_0 = const()[name = tensor("op_6968_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6968_end_0 = const()[name = tensor("op_6968_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6968_end_mask_0 = const()[name = tensor("op_6968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6968_cast_fp16 = slice_by_index(begin = var_6968_begin_0, end = var_6968_end_0, end_mask = var_6968_end_mask_0, x = var_6852_cast_fp16)[name = tensor("op_6968_cast_fp16")]; + tensor var_6975_begin_0 = const()[name = tensor("op_6975_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_6975_end_0 = const()[name = tensor("op_6975_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_6975_end_mask_0 = const()[name = tensor("op_6975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6975_cast_fp16 = slice_by_index(begin = var_6975_begin_0, end = var_6975_end_0, end_mask = var_6975_end_mask_0, x = var_6852_cast_fp16)[name = tensor("op_6975_cast_fp16")]; + tensor var_6982_begin_0 = const()[name = tensor("op_6982_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_6982_end_0 = const()[name = tensor("op_6982_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_6982_end_mask_0 = const()[name = tensor("op_6982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6982_cast_fp16 = slice_by_index(begin = var_6982_begin_0, end = var_6982_end_0, end_mask = var_6982_end_mask_0, x = var_6852_cast_fp16)[name = tensor("op_6982_cast_fp16")]; + tensor var_6989_begin_0 = const()[name = tensor("op_6989_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6989_end_0 = const()[name = tensor("op_6989_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_6989_end_mask_0 = const()[name = tensor("op_6989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6989_cast_fp16 = slice_by_index(begin = var_6989_begin_0, end = var_6989_end_0, end_mask = var_6989_end_mask_0, x = var_6856_cast_fp16)[name = tensor("op_6989_cast_fp16")]; + tensor var_6996_begin_0 = const()[name = tensor("op_6996_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_6996_end_0 = const()[name = tensor("op_6996_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_6996_end_mask_0 = const()[name = tensor("op_6996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6996_cast_fp16 = slice_by_index(begin = var_6996_begin_0, end = var_6996_end_0, end_mask = var_6996_end_mask_0, x = var_6856_cast_fp16)[name = tensor("op_6996_cast_fp16")]; + tensor var_7003_begin_0 = const()[name = tensor("op_7003_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7003_end_0 = const()[name = tensor("op_7003_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7003_end_mask_0 = const()[name = tensor("op_7003_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7003_cast_fp16 = slice_by_index(begin = var_7003_begin_0, end = var_7003_end_0, end_mask = var_7003_end_mask_0, x = var_6856_cast_fp16)[name = tensor("op_7003_cast_fp16")]; + tensor var_7010_begin_0 = const()[name = tensor("op_7010_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7010_end_0 = const()[name = tensor("op_7010_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7010_end_mask_0 = const()[name = tensor("op_7010_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7010_cast_fp16 = slice_by_index(begin = var_7010_begin_0, end = var_7010_end_0, end_mask = var_7010_end_mask_0, x = var_6856_cast_fp16)[name = tensor("op_7010_cast_fp16")]; + tensor var_7017_begin_0 = const()[name = tensor("op_7017_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7017_end_0 = const()[name = tensor("op_7017_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7017_end_mask_0 = const()[name = tensor("op_7017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7017_cast_fp16 = slice_by_index(begin = var_7017_begin_0, end = var_7017_end_0, end_mask = var_7017_end_mask_0, x = var_6860_cast_fp16)[name = tensor("op_7017_cast_fp16")]; + tensor var_7024_begin_0 = const()[name = tensor("op_7024_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7024_end_0 = const()[name = tensor("op_7024_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7024_end_mask_0 = const()[name = tensor("op_7024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7024_cast_fp16 = slice_by_index(begin = var_7024_begin_0, end = var_7024_end_0, end_mask = var_7024_end_mask_0, x = var_6860_cast_fp16)[name = tensor("op_7024_cast_fp16")]; + tensor var_7031_begin_0 = const()[name = tensor("op_7031_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7031_end_0 = const()[name = tensor("op_7031_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7031_end_mask_0 = const()[name = tensor("op_7031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7031_cast_fp16 = slice_by_index(begin = var_7031_begin_0, end = var_7031_end_0, end_mask = var_7031_end_mask_0, x = var_6860_cast_fp16)[name = tensor("op_7031_cast_fp16")]; + tensor var_7038_begin_0 = const()[name = tensor("op_7038_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7038_end_0 = const()[name = tensor("op_7038_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7038_end_mask_0 = const()[name = tensor("op_7038_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7038_cast_fp16 = slice_by_index(begin = var_7038_begin_0, end = var_7038_end_0, end_mask = var_7038_end_mask_0, x = var_6860_cast_fp16)[name = tensor("op_7038_cast_fp16")]; + tensor var_7045_begin_0 = const()[name = tensor("op_7045_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7045_end_0 = const()[name = tensor("op_7045_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7045_end_mask_0 = const()[name = tensor("op_7045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7045_cast_fp16 = slice_by_index(begin = var_7045_begin_0, end = var_7045_end_0, end_mask = var_7045_end_mask_0, x = var_6864_cast_fp16)[name = tensor("op_7045_cast_fp16")]; + tensor var_7052_begin_0 = const()[name = tensor("op_7052_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7052_end_0 = const()[name = tensor("op_7052_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7052_end_mask_0 = const()[name = tensor("op_7052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7052_cast_fp16 = slice_by_index(begin = var_7052_begin_0, end = var_7052_end_0, end_mask = var_7052_end_mask_0, x = var_6864_cast_fp16)[name = tensor("op_7052_cast_fp16")]; + tensor var_7059_begin_0 = const()[name = tensor("op_7059_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7059_end_0 = const()[name = tensor("op_7059_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7059_end_mask_0 = const()[name = tensor("op_7059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7059_cast_fp16 = slice_by_index(begin = var_7059_begin_0, end = var_7059_end_0, end_mask = var_7059_end_mask_0, x = var_6864_cast_fp16)[name = tensor("op_7059_cast_fp16")]; + tensor var_7066_begin_0 = const()[name = tensor("op_7066_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7066_end_0 = const()[name = tensor("op_7066_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7066_end_mask_0 = const()[name = tensor("op_7066_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7066_cast_fp16 = slice_by_index(begin = var_7066_begin_0, end = var_7066_end_0, end_mask = var_7066_end_mask_0, x = var_6864_cast_fp16)[name = tensor("op_7066_cast_fp16")]; + tensor var_7073_begin_0 = const()[name = tensor("op_7073_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7073_end_0 = const()[name = tensor("op_7073_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7073_end_mask_0 = const()[name = tensor("op_7073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7073_cast_fp16 = slice_by_index(begin = var_7073_begin_0, end = var_7073_end_0, end_mask = var_7073_end_mask_0, x = var_6868_cast_fp16)[name = tensor("op_7073_cast_fp16")]; + tensor var_7080_begin_0 = const()[name = tensor("op_7080_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7080_end_0 = const()[name = tensor("op_7080_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7080_end_mask_0 = const()[name = tensor("op_7080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7080_cast_fp16 = slice_by_index(begin = var_7080_begin_0, end = var_7080_end_0, end_mask = var_7080_end_mask_0, x = var_6868_cast_fp16)[name = tensor("op_7080_cast_fp16")]; + tensor var_7087_begin_0 = const()[name = tensor("op_7087_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7087_end_0 = const()[name = tensor("op_7087_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7087_end_mask_0 = const()[name = tensor("op_7087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7087_cast_fp16 = slice_by_index(begin = var_7087_begin_0, end = var_7087_end_0, end_mask = var_7087_end_mask_0, x = var_6868_cast_fp16)[name = tensor("op_7087_cast_fp16")]; + tensor var_7094_begin_0 = const()[name = tensor("op_7094_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7094_end_0 = const()[name = tensor("op_7094_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7094_end_mask_0 = const()[name = tensor("op_7094_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7094_cast_fp16 = slice_by_index(begin = var_7094_begin_0, end = var_7094_end_0, end_mask = var_7094_end_mask_0, x = var_6868_cast_fp16)[name = tensor("op_7094_cast_fp16")]; + tensor var_7101_begin_0 = const()[name = tensor("op_7101_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7101_end_0 = const()[name = tensor("op_7101_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7101_end_mask_0 = const()[name = tensor("op_7101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7101_cast_fp16 = slice_by_index(begin = var_7101_begin_0, end = var_7101_end_0, end_mask = var_7101_end_mask_0, x = var_6872_cast_fp16)[name = tensor("op_7101_cast_fp16")]; + tensor var_7108_begin_0 = const()[name = tensor("op_7108_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7108_end_0 = const()[name = tensor("op_7108_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7108_end_mask_0 = const()[name = tensor("op_7108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7108_cast_fp16 = slice_by_index(begin = var_7108_begin_0, end = var_7108_end_0, end_mask = var_7108_end_mask_0, x = var_6872_cast_fp16)[name = tensor("op_7108_cast_fp16")]; + tensor var_7115_begin_0 = const()[name = tensor("op_7115_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7115_end_0 = const()[name = tensor("op_7115_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7115_end_mask_0 = const()[name = tensor("op_7115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7115_cast_fp16 = slice_by_index(begin = var_7115_begin_0, end = var_7115_end_0, end_mask = var_7115_end_mask_0, x = var_6872_cast_fp16)[name = tensor("op_7115_cast_fp16")]; + tensor var_7122_begin_0 = const()[name = tensor("op_7122_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7122_end_0 = const()[name = tensor("op_7122_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7122_end_mask_0 = const()[name = tensor("op_7122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7122_cast_fp16 = slice_by_index(begin = var_7122_begin_0, end = var_7122_end_0, end_mask = var_7122_end_mask_0, x = var_6872_cast_fp16)[name = tensor("op_7122_cast_fp16")]; + tensor var_7129_begin_0 = const()[name = tensor("op_7129_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7129_end_0 = const()[name = tensor("op_7129_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7129_end_mask_0 = const()[name = tensor("op_7129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7129_cast_fp16 = slice_by_index(begin = var_7129_begin_0, end = var_7129_end_0, end_mask = var_7129_end_mask_0, x = var_6876_cast_fp16)[name = tensor("op_7129_cast_fp16")]; + tensor var_7136_begin_0 = const()[name = tensor("op_7136_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7136_end_0 = const()[name = tensor("op_7136_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7136_end_mask_0 = const()[name = tensor("op_7136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7136_cast_fp16 = slice_by_index(begin = var_7136_begin_0, end = var_7136_end_0, end_mask = var_7136_end_mask_0, x = var_6876_cast_fp16)[name = tensor("op_7136_cast_fp16")]; + tensor var_7143_begin_0 = const()[name = tensor("op_7143_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7143_end_0 = const()[name = tensor("op_7143_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7143_end_mask_0 = const()[name = tensor("op_7143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7143_cast_fp16 = slice_by_index(begin = var_7143_begin_0, end = var_7143_end_0, end_mask = var_7143_end_mask_0, x = var_6876_cast_fp16)[name = tensor("op_7143_cast_fp16")]; + tensor var_7150_begin_0 = const()[name = tensor("op_7150_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7150_end_0 = const()[name = tensor("op_7150_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7150_end_mask_0 = const()[name = tensor("op_7150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7150_cast_fp16 = slice_by_index(begin = var_7150_begin_0, end = var_7150_end_0, end_mask = var_7150_end_mask_0, x = var_6876_cast_fp16)[name = tensor("op_7150_cast_fp16")]; + tensor var_7157_begin_0 = const()[name = tensor("op_7157_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7157_end_0 = const()[name = tensor("op_7157_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7157_end_mask_0 = const()[name = tensor("op_7157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7157_cast_fp16 = slice_by_index(begin = var_7157_begin_0, end = var_7157_end_0, end_mask = var_7157_end_mask_0, x = var_6880_cast_fp16)[name = tensor("op_7157_cast_fp16")]; + tensor var_7164_begin_0 = const()[name = tensor("op_7164_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7164_end_0 = const()[name = tensor("op_7164_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7164_end_mask_0 = const()[name = tensor("op_7164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7164_cast_fp16 = slice_by_index(begin = var_7164_begin_0, end = var_7164_end_0, end_mask = var_7164_end_mask_0, x = var_6880_cast_fp16)[name = tensor("op_7164_cast_fp16")]; + tensor var_7171_begin_0 = const()[name = tensor("op_7171_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7171_end_0 = const()[name = tensor("op_7171_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7171_end_mask_0 = const()[name = tensor("op_7171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7171_cast_fp16 = slice_by_index(begin = var_7171_begin_0, end = var_7171_end_0, end_mask = var_7171_end_mask_0, x = var_6880_cast_fp16)[name = tensor("op_7171_cast_fp16")]; + tensor var_7178_begin_0 = const()[name = tensor("op_7178_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7178_end_0 = const()[name = tensor("op_7178_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7178_end_mask_0 = const()[name = tensor("op_7178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7178_cast_fp16 = slice_by_index(begin = var_7178_begin_0, end = var_7178_end_0, end_mask = var_7178_end_mask_0, x = var_6880_cast_fp16)[name = tensor("op_7178_cast_fp16")]; + tensor var_7185_begin_0 = const()[name = tensor("op_7185_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7185_end_0 = const()[name = tensor("op_7185_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7185_end_mask_0 = const()[name = tensor("op_7185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7185_cast_fp16 = slice_by_index(begin = var_7185_begin_0, end = var_7185_end_0, end_mask = var_7185_end_mask_0, x = var_6884_cast_fp16)[name = tensor("op_7185_cast_fp16")]; + tensor var_7192_begin_0 = const()[name = tensor("op_7192_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7192_end_0 = const()[name = tensor("op_7192_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7192_end_mask_0 = const()[name = tensor("op_7192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7192_cast_fp16 = slice_by_index(begin = var_7192_begin_0, end = var_7192_end_0, end_mask = var_7192_end_mask_0, x = var_6884_cast_fp16)[name = tensor("op_7192_cast_fp16")]; + tensor var_7199_begin_0 = const()[name = tensor("op_7199_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7199_end_0 = const()[name = tensor("op_7199_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7199_end_mask_0 = const()[name = tensor("op_7199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7199_cast_fp16 = slice_by_index(begin = var_7199_begin_0, end = var_7199_end_0, end_mask = var_7199_end_mask_0, x = var_6884_cast_fp16)[name = tensor("op_7199_cast_fp16")]; + tensor var_7206_begin_0 = const()[name = tensor("op_7206_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7206_end_0 = const()[name = tensor("op_7206_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7206_end_mask_0 = const()[name = tensor("op_7206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7206_cast_fp16 = slice_by_index(begin = var_7206_begin_0, end = var_7206_end_0, end_mask = var_7206_end_mask_0, x = var_6884_cast_fp16)[name = tensor("op_7206_cast_fp16")]; + tensor var_7213_begin_0 = const()[name = tensor("op_7213_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7213_end_0 = const()[name = tensor("op_7213_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7213_end_mask_0 = const()[name = tensor("op_7213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7213_cast_fp16 = slice_by_index(begin = var_7213_begin_0, end = var_7213_end_0, end_mask = var_7213_end_mask_0, x = var_6888_cast_fp16)[name = tensor("op_7213_cast_fp16")]; + tensor var_7220_begin_0 = const()[name = tensor("op_7220_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7220_end_0 = const()[name = tensor("op_7220_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7220_end_mask_0 = const()[name = tensor("op_7220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7220_cast_fp16 = slice_by_index(begin = var_7220_begin_0, end = var_7220_end_0, end_mask = var_7220_end_mask_0, x = var_6888_cast_fp16)[name = tensor("op_7220_cast_fp16")]; + tensor var_7227_begin_0 = const()[name = tensor("op_7227_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7227_end_0 = const()[name = tensor("op_7227_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7227_end_mask_0 = const()[name = tensor("op_7227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7227_cast_fp16 = slice_by_index(begin = var_7227_begin_0, end = var_7227_end_0, end_mask = var_7227_end_mask_0, x = var_6888_cast_fp16)[name = tensor("op_7227_cast_fp16")]; + tensor var_7234_begin_0 = const()[name = tensor("op_7234_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7234_end_0 = const()[name = tensor("op_7234_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7234_end_mask_0 = const()[name = tensor("op_7234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7234_cast_fp16 = slice_by_index(begin = var_7234_begin_0, end = var_7234_end_0, end_mask = var_7234_end_mask_0, x = var_6888_cast_fp16)[name = tensor("op_7234_cast_fp16")]; + tensor var_7241_begin_0 = const()[name = tensor("op_7241_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7241_end_0 = const()[name = tensor("op_7241_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7241_end_mask_0 = const()[name = tensor("op_7241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7241_cast_fp16 = slice_by_index(begin = var_7241_begin_0, end = var_7241_end_0, end_mask = var_7241_end_mask_0, x = var_6892_cast_fp16)[name = tensor("op_7241_cast_fp16")]; + tensor var_7248_begin_0 = const()[name = tensor("op_7248_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7248_end_0 = const()[name = tensor("op_7248_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7248_end_mask_0 = const()[name = tensor("op_7248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7248_cast_fp16 = slice_by_index(begin = var_7248_begin_0, end = var_7248_end_0, end_mask = var_7248_end_mask_0, x = var_6892_cast_fp16)[name = tensor("op_7248_cast_fp16")]; + tensor var_7255_begin_0 = const()[name = tensor("op_7255_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7255_end_0 = const()[name = tensor("op_7255_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7255_end_mask_0 = const()[name = tensor("op_7255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7255_cast_fp16 = slice_by_index(begin = var_7255_begin_0, end = var_7255_end_0, end_mask = var_7255_end_mask_0, x = var_6892_cast_fp16)[name = tensor("op_7255_cast_fp16")]; + tensor var_7262_begin_0 = const()[name = tensor("op_7262_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7262_end_0 = const()[name = tensor("op_7262_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7262_end_mask_0 = const()[name = tensor("op_7262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7262_cast_fp16 = slice_by_index(begin = var_7262_begin_0, end = var_7262_end_0, end_mask = var_7262_end_mask_0, x = var_6892_cast_fp16)[name = tensor("op_7262_cast_fp16")]; + tensor var_7269_begin_0 = const()[name = tensor("op_7269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7269_end_0 = const()[name = tensor("op_7269_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7269_end_mask_0 = const()[name = tensor("op_7269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7269_cast_fp16 = slice_by_index(begin = var_7269_begin_0, end = var_7269_end_0, end_mask = var_7269_end_mask_0, x = var_6896_cast_fp16)[name = tensor("op_7269_cast_fp16")]; + tensor var_7276_begin_0 = const()[name = tensor("op_7276_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7276_end_0 = const()[name = tensor("op_7276_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7276_end_mask_0 = const()[name = tensor("op_7276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7276_cast_fp16 = slice_by_index(begin = var_7276_begin_0, end = var_7276_end_0, end_mask = var_7276_end_mask_0, x = var_6896_cast_fp16)[name = tensor("op_7276_cast_fp16")]; + tensor var_7283_begin_0 = const()[name = tensor("op_7283_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7283_end_0 = const()[name = tensor("op_7283_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7283_end_mask_0 = const()[name = tensor("op_7283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7283_cast_fp16 = slice_by_index(begin = var_7283_begin_0, end = var_7283_end_0, end_mask = var_7283_end_mask_0, x = var_6896_cast_fp16)[name = tensor("op_7283_cast_fp16")]; + tensor var_7290_begin_0 = const()[name = tensor("op_7290_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7290_end_0 = const()[name = tensor("op_7290_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7290_end_mask_0 = const()[name = tensor("op_7290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7290_cast_fp16 = slice_by_index(begin = var_7290_begin_0, end = var_7290_end_0, end_mask = var_7290_end_mask_0, x = var_6896_cast_fp16)[name = tensor("op_7290_cast_fp16")]; + tensor var_7297_begin_0 = const()[name = tensor("op_7297_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7297_end_0 = const()[name = tensor("op_7297_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7297_end_mask_0 = const()[name = tensor("op_7297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7297_cast_fp16 = slice_by_index(begin = var_7297_begin_0, end = var_7297_end_0, end_mask = var_7297_end_mask_0, x = var_6900_cast_fp16)[name = tensor("op_7297_cast_fp16")]; + tensor var_7304_begin_0 = const()[name = tensor("op_7304_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7304_end_0 = const()[name = tensor("op_7304_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7304_end_mask_0 = const()[name = tensor("op_7304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7304_cast_fp16 = slice_by_index(begin = var_7304_begin_0, end = var_7304_end_0, end_mask = var_7304_end_mask_0, x = var_6900_cast_fp16)[name = tensor("op_7304_cast_fp16")]; + tensor var_7311_begin_0 = const()[name = tensor("op_7311_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7311_end_0 = const()[name = tensor("op_7311_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7311_end_mask_0 = const()[name = tensor("op_7311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7311_cast_fp16 = slice_by_index(begin = var_7311_begin_0, end = var_7311_end_0, end_mask = var_7311_end_mask_0, x = var_6900_cast_fp16)[name = tensor("op_7311_cast_fp16")]; + tensor var_7318_begin_0 = const()[name = tensor("op_7318_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7318_end_0 = const()[name = tensor("op_7318_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7318_end_mask_0 = const()[name = tensor("op_7318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7318_cast_fp16 = slice_by_index(begin = var_7318_begin_0, end = var_7318_end_0, end_mask = var_7318_end_mask_0, x = var_6900_cast_fp16)[name = tensor("op_7318_cast_fp16")]; + tensor var_7325_begin_0 = const()[name = tensor("op_7325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7325_end_0 = const()[name = tensor("op_7325_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7325_end_mask_0 = const()[name = tensor("op_7325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7325_cast_fp16 = slice_by_index(begin = var_7325_begin_0, end = var_7325_end_0, end_mask = var_7325_end_mask_0, x = var_6904_cast_fp16)[name = tensor("op_7325_cast_fp16")]; + tensor var_7332_begin_0 = const()[name = tensor("op_7332_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7332_end_0 = const()[name = tensor("op_7332_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7332_end_mask_0 = const()[name = tensor("op_7332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7332_cast_fp16 = slice_by_index(begin = var_7332_begin_0, end = var_7332_end_0, end_mask = var_7332_end_mask_0, x = var_6904_cast_fp16)[name = tensor("op_7332_cast_fp16")]; + tensor var_7339_begin_0 = const()[name = tensor("op_7339_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7339_end_0 = const()[name = tensor("op_7339_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7339_end_mask_0 = const()[name = tensor("op_7339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7339_cast_fp16 = slice_by_index(begin = var_7339_begin_0, end = var_7339_end_0, end_mask = var_7339_end_mask_0, x = var_6904_cast_fp16)[name = tensor("op_7339_cast_fp16")]; + tensor var_7346_begin_0 = const()[name = tensor("op_7346_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7346_end_0 = const()[name = tensor("op_7346_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7346_end_mask_0 = const()[name = tensor("op_7346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7346_cast_fp16 = slice_by_index(begin = var_7346_begin_0, end = var_7346_end_0, end_mask = var_7346_end_mask_0, x = var_6904_cast_fp16)[name = tensor("op_7346_cast_fp16")]; + tensor var_7353_begin_0 = const()[name = tensor("op_7353_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7353_end_0 = const()[name = tensor("op_7353_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7353_end_mask_0 = const()[name = tensor("op_7353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7353_cast_fp16 = slice_by_index(begin = var_7353_begin_0, end = var_7353_end_0, end_mask = var_7353_end_mask_0, x = var_6908_cast_fp16)[name = tensor("op_7353_cast_fp16")]; + tensor var_7360_begin_0 = const()[name = tensor("op_7360_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7360_end_0 = const()[name = tensor("op_7360_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7360_end_mask_0 = const()[name = tensor("op_7360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7360_cast_fp16 = slice_by_index(begin = var_7360_begin_0, end = var_7360_end_0, end_mask = var_7360_end_mask_0, x = var_6908_cast_fp16)[name = tensor("op_7360_cast_fp16")]; + tensor var_7367_begin_0 = const()[name = tensor("op_7367_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7367_end_0 = const()[name = tensor("op_7367_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7367_end_mask_0 = const()[name = tensor("op_7367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7367_cast_fp16 = slice_by_index(begin = var_7367_begin_0, end = var_7367_end_0, end_mask = var_7367_end_mask_0, x = var_6908_cast_fp16)[name = tensor("op_7367_cast_fp16")]; + tensor var_7374_begin_0 = const()[name = tensor("op_7374_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7374_end_0 = const()[name = tensor("op_7374_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7374_end_mask_0 = const()[name = tensor("op_7374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7374_cast_fp16 = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = var_6908_cast_fp16)[name = tensor("op_7374_cast_fp16")]; + tensor var_7381_begin_0 = const()[name = tensor("op_7381_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7381_end_0 = const()[name = tensor("op_7381_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7381_end_mask_0 = const()[name = tensor("op_7381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7381_cast_fp16 = slice_by_index(begin = var_7381_begin_0, end = var_7381_end_0, end_mask = var_7381_end_mask_0, x = var_6912_cast_fp16)[name = tensor("op_7381_cast_fp16")]; + tensor var_7388_begin_0 = const()[name = tensor("op_7388_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7388_end_0 = const()[name = tensor("op_7388_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7388_end_mask_0 = const()[name = tensor("op_7388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7388_cast_fp16 = slice_by_index(begin = var_7388_begin_0, end = var_7388_end_0, end_mask = var_7388_end_mask_0, x = var_6912_cast_fp16)[name = tensor("op_7388_cast_fp16")]; + tensor var_7395_begin_0 = const()[name = tensor("op_7395_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7395_end_0 = const()[name = tensor("op_7395_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7395_end_mask_0 = const()[name = tensor("op_7395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7395_cast_fp16 = slice_by_index(begin = var_7395_begin_0, end = var_7395_end_0, end_mask = var_7395_end_mask_0, x = var_6912_cast_fp16)[name = tensor("op_7395_cast_fp16")]; + tensor var_7402_begin_0 = const()[name = tensor("op_7402_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7402_end_0 = const()[name = tensor("op_7402_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7402_end_mask_0 = const()[name = tensor("op_7402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7402_cast_fp16 = slice_by_index(begin = var_7402_begin_0, end = var_7402_end_0, end_mask = var_7402_end_mask_0, x = var_6912_cast_fp16)[name = tensor("op_7402_cast_fp16")]; + tensor var_7409_begin_0 = const()[name = tensor("op_7409_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7409_end_0 = const()[name = tensor("op_7409_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7409_end_mask_0 = const()[name = tensor("op_7409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7409_cast_fp16 = slice_by_index(begin = var_7409_begin_0, end = var_7409_end_0, end_mask = var_7409_end_mask_0, x = var_6916_cast_fp16)[name = tensor("op_7409_cast_fp16")]; + tensor var_7416_begin_0 = const()[name = tensor("op_7416_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7416_end_0 = const()[name = tensor("op_7416_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7416_end_mask_0 = const()[name = tensor("op_7416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7416_cast_fp16 = slice_by_index(begin = var_7416_begin_0, end = var_7416_end_0, end_mask = var_7416_end_mask_0, x = var_6916_cast_fp16)[name = tensor("op_7416_cast_fp16")]; + tensor var_7423_begin_0 = const()[name = tensor("op_7423_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7423_end_0 = const()[name = tensor("op_7423_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7423_end_mask_0 = const()[name = tensor("op_7423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7423_cast_fp16 = slice_by_index(begin = var_7423_begin_0, end = var_7423_end_0, end_mask = var_7423_end_mask_0, x = var_6916_cast_fp16)[name = tensor("op_7423_cast_fp16")]; + tensor var_7430_begin_0 = const()[name = tensor("op_7430_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7430_end_0 = const()[name = tensor("op_7430_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7430_end_mask_0 = const()[name = tensor("op_7430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7430_cast_fp16 = slice_by_index(begin = var_7430_begin_0, end = var_7430_end_0, end_mask = var_7430_end_mask_0, x = var_6916_cast_fp16)[name = tensor("op_7430_cast_fp16")]; + tensor var_7437_begin_0 = const()[name = tensor("op_7437_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7437_end_0 = const()[name = tensor("op_7437_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7437_end_mask_0 = const()[name = tensor("op_7437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7437_cast_fp16 = slice_by_index(begin = var_7437_begin_0, end = var_7437_end_0, end_mask = var_7437_end_mask_0, x = var_6920_cast_fp16)[name = tensor("op_7437_cast_fp16")]; + tensor var_7444_begin_0 = const()[name = tensor("op_7444_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7444_end_0 = const()[name = tensor("op_7444_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7444_end_mask_0 = const()[name = tensor("op_7444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7444_cast_fp16 = slice_by_index(begin = var_7444_begin_0, end = var_7444_end_0, end_mask = var_7444_end_mask_0, x = var_6920_cast_fp16)[name = tensor("op_7444_cast_fp16")]; + tensor var_7451_begin_0 = const()[name = tensor("op_7451_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7451_end_0 = const()[name = tensor("op_7451_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7451_end_mask_0 = const()[name = tensor("op_7451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7451_cast_fp16 = slice_by_index(begin = var_7451_begin_0, end = var_7451_end_0, end_mask = var_7451_end_mask_0, x = var_6920_cast_fp16)[name = tensor("op_7451_cast_fp16")]; + tensor var_7458_begin_0 = const()[name = tensor("op_7458_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7458_end_0 = const()[name = tensor("op_7458_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7458_end_mask_0 = const()[name = tensor("op_7458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7458_cast_fp16 = slice_by_index(begin = var_7458_begin_0, end = var_7458_end_0, end_mask = var_7458_end_mask_0, x = var_6920_cast_fp16)[name = tensor("op_7458_cast_fp16")]; + tensor var_7465_begin_0 = const()[name = tensor("op_7465_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7465_end_0 = const()[name = tensor("op_7465_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_7465_end_mask_0 = const()[name = tensor("op_7465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7465_cast_fp16 = slice_by_index(begin = var_7465_begin_0, end = var_7465_end_0, end_mask = var_7465_end_mask_0, x = var_6924_cast_fp16)[name = tensor("op_7465_cast_fp16")]; + tensor var_7472_begin_0 = const()[name = tensor("op_7472_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_7472_end_0 = const()[name = tensor("op_7472_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_7472_end_mask_0 = const()[name = tensor("op_7472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7472_cast_fp16 = slice_by_index(begin = var_7472_begin_0, end = var_7472_end_0, end_mask = var_7472_end_mask_0, x = var_6924_cast_fp16)[name = tensor("op_7472_cast_fp16")]; + tensor var_7479_begin_0 = const()[name = tensor("op_7479_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_7479_end_0 = const()[name = tensor("op_7479_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_7479_end_mask_0 = const()[name = tensor("op_7479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7479_cast_fp16 = slice_by_index(begin = var_7479_begin_0, end = var_7479_end_0, end_mask = var_7479_end_mask_0, x = var_6924_cast_fp16)[name = tensor("op_7479_cast_fp16")]; + tensor var_7486_begin_0 = const()[name = tensor("op_7486_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_7486_end_0 = const()[name = tensor("op_7486_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7486_end_mask_0 = const()[name = tensor("op_7486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7486_cast_fp16 = slice_by_index(begin = var_7486_begin_0, end = var_7486_end_0, end_mask = var_7486_end_mask_0, x = var_6924_cast_fp16)[name = tensor("op_7486_cast_fp16")]; + tensor k_9_perm_0 = const()[name = tensor("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_7491_begin_0 = const()[name = tensor("op_7491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7491_end_0 = const()[name = tensor("op_7491_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_7491_end_mask_0 = const()[name = tensor("op_7491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_27 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = tensor("transpose_27")]; + tensor var_7491_cast_fp16 = slice_by_index(begin = var_7491_begin_0, end = var_7491_end_0, end_mask = var_7491_end_mask_0, x = transpose_27)[name = tensor("op_7491_cast_fp16")]; + tensor var_7495_begin_0 = const()[name = tensor("op_7495_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_7495_end_0 = const()[name = tensor("op_7495_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_7495_end_mask_0 = const()[name = tensor("op_7495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7495_cast_fp16 = slice_by_index(begin = var_7495_begin_0, end = var_7495_end_0, end_mask = var_7495_end_mask_0, x = transpose_27)[name = tensor("op_7495_cast_fp16")]; + tensor var_7499_begin_0 = const()[name = tensor("op_7499_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_7499_end_0 = const()[name = tensor("op_7499_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_7499_end_mask_0 = const()[name = tensor("op_7499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7499_cast_fp16 = slice_by_index(begin = var_7499_begin_0, end = var_7499_end_0, end_mask = var_7499_end_mask_0, x = transpose_27)[name = tensor("op_7499_cast_fp16")]; + tensor var_7503_begin_0 = const()[name = tensor("op_7503_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_7503_end_0 = const()[name = tensor("op_7503_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_7503_end_mask_0 = const()[name = tensor("op_7503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7503_cast_fp16 = slice_by_index(begin = var_7503_begin_0, end = var_7503_end_0, end_mask = var_7503_end_mask_0, x = transpose_27)[name = tensor("op_7503_cast_fp16")]; + tensor var_7507_begin_0 = const()[name = tensor("op_7507_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_7507_end_0 = const()[name = tensor("op_7507_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_7507_end_mask_0 = const()[name = tensor("op_7507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7507_cast_fp16 = slice_by_index(begin = var_7507_begin_0, end = var_7507_end_0, end_mask = var_7507_end_mask_0, x = transpose_27)[name = tensor("op_7507_cast_fp16")]; + tensor var_7511_begin_0 = const()[name = tensor("op_7511_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_7511_end_0 = const()[name = tensor("op_7511_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_7511_end_mask_0 = const()[name = tensor("op_7511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7511_cast_fp16 = slice_by_index(begin = var_7511_begin_0, end = var_7511_end_0, end_mask = var_7511_end_mask_0, x = transpose_27)[name = tensor("op_7511_cast_fp16")]; + tensor var_7515_begin_0 = const()[name = tensor("op_7515_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_7515_end_0 = const()[name = tensor("op_7515_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_7515_end_mask_0 = const()[name = tensor("op_7515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7515_cast_fp16 = slice_by_index(begin = var_7515_begin_0, end = var_7515_end_0, end_mask = var_7515_end_mask_0, x = transpose_27)[name = tensor("op_7515_cast_fp16")]; + tensor var_7519_begin_0 = const()[name = tensor("op_7519_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_7519_end_0 = const()[name = tensor("op_7519_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_7519_end_mask_0 = const()[name = tensor("op_7519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7519_cast_fp16 = slice_by_index(begin = var_7519_begin_0, end = var_7519_end_0, end_mask = var_7519_end_mask_0, x = transpose_27)[name = tensor("op_7519_cast_fp16")]; + tensor var_7523_begin_0 = const()[name = tensor("op_7523_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_7523_end_0 = const()[name = tensor("op_7523_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_7523_end_mask_0 = const()[name = tensor("op_7523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7523_cast_fp16 = slice_by_index(begin = var_7523_begin_0, end = var_7523_end_0, end_mask = var_7523_end_mask_0, x = transpose_27)[name = tensor("op_7523_cast_fp16")]; + tensor var_7527_begin_0 = const()[name = tensor("op_7527_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_7527_end_0 = const()[name = tensor("op_7527_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_7527_end_mask_0 = const()[name = tensor("op_7527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7527_cast_fp16 = slice_by_index(begin = var_7527_begin_0, end = var_7527_end_0, end_mask = var_7527_end_mask_0, x = transpose_27)[name = tensor("op_7527_cast_fp16")]; + tensor var_7531_begin_0 = const()[name = tensor("op_7531_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_7531_end_0 = const()[name = tensor("op_7531_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_7531_end_mask_0 = const()[name = tensor("op_7531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7531_cast_fp16 = slice_by_index(begin = var_7531_begin_0, end = var_7531_end_0, end_mask = var_7531_end_mask_0, x = transpose_27)[name = tensor("op_7531_cast_fp16")]; + tensor var_7535_begin_0 = const()[name = tensor("op_7535_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_7535_end_0 = const()[name = tensor("op_7535_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_7535_end_mask_0 = const()[name = tensor("op_7535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7535_cast_fp16 = slice_by_index(begin = var_7535_begin_0, end = var_7535_end_0, end_mask = var_7535_end_mask_0, x = transpose_27)[name = tensor("op_7535_cast_fp16")]; + tensor var_7539_begin_0 = const()[name = tensor("op_7539_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_7539_end_0 = const()[name = tensor("op_7539_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_7539_end_mask_0 = const()[name = tensor("op_7539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7539_cast_fp16 = slice_by_index(begin = var_7539_begin_0, end = var_7539_end_0, end_mask = var_7539_end_mask_0, x = transpose_27)[name = tensor("op_7539_cast_fp16")]; + tensor var_7543_begin_0 = const()[name = tensor("op_7543_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_7543_end_0 = const()[name = tensor("op_7543_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_7543_end_mask_0 = const()[name = tensor("op_7543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7543_cast_fp16 = slice_by_index(begin = var_7543_begin_0, end = var_7543_end_0, end_mask = var_7543_end_mask_0, x = transpose_27)[name = tensor("op_7543_cast_fp16")]; + tensor var_7547_begin_0 = const()[name = tensor("op_7547_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_7547_end_0 = const()[name = tensor("op_7547_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_7547_end_mask_0 = const()[name = tensor("op_7547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7547_cast_fp16 = slice_by_index(begin = var_7547_begin_0, end = var_7547_end_0, end_mask = var_7547_end_mask_0, x = transpose_27)[name = tensor("op_7547_cast_fp16")]; + tensor var_7551_begin_0 = const()[name = tensor("op_7551_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_7551_end_0 = const()[name = tensor("op_7551_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_7551_end_mask_0 = const()[name = tensor("op_7551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7551_cast_fp16 = slice_by_index(begin = var_7551_begin_0, end = var_7551_end_0, end_mask = var_7551_end_mask_0, x = transpose_27)[name = tensor("op_7551_cast_fp16")]; + tensor var_7555_begin_0 = const()[name = tensor("op_7555_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_7555_end_0 = const()[name = tensor("op_7555_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_7555_end_mask_0 = const()[name = tensor("op_7555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7555_cast_fp16 = slice_by_index(begin = var_7555_begin_0, end = var_7555_end_0, end_mask = var_7555_end_mask_0, x = transpose_27)[name = tensor("op_7555_cast_fp16")]; + tensor var_7559_begin_0 = const()[name = tensor("op_7559_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_7559_end_0 = const()[name = tensor("op_7559_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_7559_end_mask_0 = const()[name = tensor("op_7559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7559_cast_fp16 = slice_by_index(begin = var_7559_begin_0, end = var_7559_end_0, end_mask = var_7559_end_mask_0, x = transpose_27)[name = tensor("op_7559_cast_fp16")]; + tensor var_7563_begin_0 = const()[name = tensor("op_7563_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_7563_end_0 = const()[name = tensor("op_7563_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_7563_end_mask_0 = const()[name = tensor("op_7563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7563_cast_fp16 = slice_by_index(begin = var_7563_begin_0, end = var_7563_end_0, end_mask = var_7563_end_mask_0, x = transpose_27)[name = tensor("op_7563_cast_fp16")]; + tensor var_7567_begin_0 = const()[name = tensor("op_7567_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_7567_end_0 = const()[name = tensor("op_7567_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_7567_end_mask_0 = const()[name = tensor("op_7567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7567_cast_fp16 = slice_by_index(begin = var_7567_begin_0, end = var_7567_end_0, end_mask = var_7567_end_mask_0, x = transpose_27)[name = tensor("op_7567_cast_fp16")]; + tensor var_7569_begin_0 = const()[name = tensor("op_7569_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7569_end_0 = const()[name = tensor("op_7569_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_7569_end_mask_0 = const()[name = tensor("op_7569_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7569_cast_fp16 = slice_by_index(begin = var_7569_begin_0, end = var_7569_end_0, end_mask = var_7569_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7569_cast_fp16")]; + tensor var_7573_begin_0 = const()[name = tensor("op_7573_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_7573_end_0 = const()[name = tensor("op_7573_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_7573_end_mask_0 = const()[name = tensor("op_7573_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7573_cast_fp16 = slice_by_index(begin = var_7573_begin_0, end = var_7573_end_0, end_mask = var_7573_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7573_cast_fp16")]; + tensor var_7577_begin_0 = const()[name = tensor("op_7577_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_7577_end_0 = const()[name = tensor("op_7577_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_7577_end_mask_0 = const()[name = tensor("op_7577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7577_cast_fp16 = slice_by_index(begin = var_7577_begin_0, end = var_7577_end_0, end_mask = var_7577_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7577_cast_fp16")]; + tensor var_7581_begin_0 = const()[name = tensor("op_7581_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_7581_end_0 = const()[name = tensor("op_7581_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_7581_end_mask_0 = const()[name = tensor("op_7581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7581_cast_fp16 = slice_by_index(begin = var_7581_begin_0, end = var_7581_end_0, end_mask = var_7581_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7581_cast_fp16")]; + tensor var_7585_begin_0 = const()[name = tensor("op_7585_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_7585_end_0 = const()[name = tensor("op_7585_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_7585_end_mask_0 = const()[name = tensor("op_7585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7585_cast_fp16 = slice_by_index(begin = var_7585_begin_0, end = var_7585_end_0, end_mask = var_7585_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7585_cast_fp16")]; + tensor var_7589_begin_0 = const()[name = tensor("op_7589_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_7589_end_0 = const()[name = tensor("op_7589_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_7589_end_mask_0 = const()[name = tensor("op_7589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7589_cast_fp16 = slice_by_index(begin = var_7589_begin_0, end = var_7589_end_0, end_mask = var_7589_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7589_cast_fp16")]; + tensor var_7593_begin_0 = const()[name = tensor("op_7593_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_7593_end_0 = const()[name = tensor("op_7593_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_7593_end_mask_0 = const()[name = tensor("op_7593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7593_cast_fp16 = slice_by_index(begin = var_7593_begin_0, end = var_7593_end_0, end_mask = var_7593_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7593_cast_fp16")]; + tensor var_7597_begin_0 = const()[name = tensor("op_7597_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_7597_end_0 = const()[name = tensor("op_7597_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_7597_end_mask_0 = const()[name = tensor("op_7597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7597_cast_fp16 = slice_by_index(begin = var_7597_begin_0, end = var_7597_end_0, end_mask = var_7597_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7597_cast_fp16")]; + tensor var_7601_begin_0 = const()[name = tensor("op_7601_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_7601_end_0 = const()[name = tensor("op_7601_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_7601_end_mask_0 = const()[name = tensor("op_7601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7601_cast_fp16 = slice_by_index(begin = var_7601_begin_0, end = var_7601_end_0, end_mask = var_7601_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7601_cast_fp16")]; + tensor var_7605_begin_0 = const()[name = tensor("op_7605_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_7605_end_0 = const()[name = tensor("op_7605_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_7605_end_mask_0 = const()[name = tensor("op_7605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7605_cast_fp16 = slice_by_index(begin = var_7605_begin_0, end = var_7605_end_0, end_mask = var_7605_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7605_cast_fp16")]; + tensor var_7609_begin_0 = const()[name = tensor("op_7609_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_7609_end_0 = const()[name = tensor("op_7609_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_7609_end_mask_0 = const()[name = tensor("op_7609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7609_cast_fp16 = slice_by_index(begin = var_7609_begin_0, end = var_7609_end_0, end_mask = var_7609_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7609_cast_fp16")]; + tensor var_7613_begin_0 = const()[name = tensor("op_7613_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_7613_end_0 = const()[name = tensor("op_7613_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_7613_end_mask_0 = const()[name = tensor("op_7613_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7613_cast_fp16 = slice_by_index(begin = var_7613_begin_0, end = var_7613_end_0, end_mask = var_7613_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7613_cast_fp16")]; + tensor var_7617_begin_0 = const()[name = tensor("op_7617_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_7617_end_0 = const()[name = tensor("op_7617_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_7617_end_mask_0 = const()[name = tensor("op_7617_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7617_cast_fp16 = slice_by_index(begin = var_7617_begin_0, end = var_7617_end_0, end_mask = var_7617_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7617_cast_fp16")]; + tensor var_7621_begin_0 = const()[name = tensor("op_7621_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_7621_end_0 = const()[name = tensor("op_7621_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_7621_end_mask_0 = const()[name = tensor("op_7621_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7621_cast_fp16 = slice_by_index(begin = var_7621_begin_0, end = var_7621_end_0, end_mask = var_7621_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7621_cast_fp16")]; + tensor var_7625_begin_0 = const()[name = tensor("op_7625_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_7625_end_0 = const()[name = tensor("op_7625_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_7625_end_mask_0 = const()[name = tensor("op_7625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7625_cast_fp16 = slice_by_index(begin = var_7625_begin_0, end = var_7625_end_0, end_mask = var_7625_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7625_cast_fp16")]; + tensor var_7629_begin_0 = const()[name = tensor("op_7629_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_7629_end_0 = const()[name = tensor("op_7629_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_7629_end_mask_0 = const()[name = tensor("op_7629_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7629_cast_fp16 = slice_by_index(begin = var_7629_begin_0, end = var_7629_end_0, end_mask = var_7629_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7629_cast_fp16")]; + tensor var_7633_begin_0 = const()[name = tensor("op_7633_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_7633_end_0 = const()[name = tensor("op_7633_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_7633_end_mask_0 = const()[name = tensor("op_7633_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7633_cast_fp16 = slice_by_index(begin = var_7633_begin_0, end = var_7633_end_0, end_mask = var_7633_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7633_cast_fp16")]; + tensor var_7637_begin_0 = const()[name = tensor("op_7637_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_7637_end_0 = const()[name = tensor("op_7637_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_7637_end_mask_0 = const()[name = tensor("op_7637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7637_cast_fp16 = slice_by_index(begin = var_7637_begin_0, end = var_7637_end_0, end_mask = var_7637_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7637_cast_fp16")]; + tensor var_7641_begin_0 = const()[name = tensor("op_7641_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_7641_end_0 = const()[name = tensor("op_7641_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_7641_end_mask_0 = const()[name = tensor("op_7641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7641_cast_fp16 = slice_by_index(begin = var_7641_begin_0, end = var_7641_end_0, end_mask = var_7641_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7641_cast_fp16")]; + tensor var_7645_begin_0 = const()[name = tensor("op_7645_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_7645_end_0 = const()[name = tensor("op_7645_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_7645_end_mask_0 = const()[name = tensor("op_7645_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_7645_cast_fp16 = slice_by_index(begin = var_7645_begin_0, end = var_7645_end_0, end_mask = var_7645_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_7645_cast_fp16")]; + tensor var_7649_equation_0 = const()[name = tensor("op_7649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7649_cast_fp16 = einsum(equation = var_7649_equation_0, values = (var_7491_cast_fp16, var_6933_cast_fp16))[name = tensor("op_7649_cast_fp16")]; + tensor var_7650_to_fp16 = const()[name = tensor("op_7650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_641_cast_fp16 = mul(x = var_7649_cast_fp16, y = var_7650_to_fp16)[name = tensor("aw_chunk_641_cast_fp16")]; + tensor var_7653_equation_0 = const()[name = tensor("op_7653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7653_cast_fp16 = einsum(equation = var_7653_equation_0, values = (var_7491_cast_fp16, var_6940_cast_fp16))[name = tensor("op_7653_cast_fp16")]; + tensor var_7654_to_fp16 = const()[name = tensor("op_7654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_643_cast_fp16 = mul(x = var_7653_cast_fp16, y = var_7654_to_fp16)[name = tensor("aw_chunk_643_cast_fp16")]; + tensor var_7657_equation_0 = const()[name = tensor("op_7657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7657_cast_fp16 = einsum(equation = var_7657_equation_0, values = (var_7491_cast_fp16, var_6947_cast_fp16))[name = tensor("op_7657_cast_fp16")]; + tensor var_7658_to_fp16 = const()[name = tensor("op_7658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_645_cast_fp16 = mul(x = var_7657_cast_fp16, y = var_7658_to_fp16)[name = tensor("aw_chunk_645_cast_fp16")]; + tensor var_7661_equation_0 = const()[name = tensor("op_7661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7661_cast_fp16 = einsum(equation = var_7661_equation_0, values = (var_7491_cast_fp16, var_6954_cast_fp16))[name = tensor("op_7661_cast_fp16")]; + tensor var_7662_to_fp16 = const()[name = tensor("op_7662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_647_cast_fp16 = mul(x = var_7661_cast_fp16, y = var_7662_to_fp16)[name = tensor("aw_chunk_647_cast_fp16")]; + tensor var_7665_equation_0 = const()[name = tensor("op_7665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7665_cast_fp16 = einsum(equation = var_7665_equation_0, values = (var_7495_cast_fp16, var_6961_cast_fp16))[name = tensor("op_7665_cast_fp16")]; + tensor var_7666_to_fp16 = const()[name = tensor("op_7666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_649_cast_fp16 = mul(x = var_7665_cast_fp16, y = var_7666_to_fp16)[name = tensor("aw_chunk_649_cast_fp16")]; + tensor var_7669_equation_0 = const()[name = tensor("op_7669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7669_cast_fp16 = einsum(equation = var_7669_equation_0, values = (var_7495_cast_fp16, var_6968_cast_fp16))[name = tensor("op_7669_cast_fp16")]; + tensor var_7670_to_fp16 = const()[name = tensor("op_7670_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_651_cast_fp16 = mul(x = var_7669_cast_fp16, y = var_7670_to_fp16)[name = tensor("aw_chunk_651_cast_fp16")]; + tensor var_7673_equation_0 = const()[name = tensor("op_7673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7673_cast_fp16 = einsum(equation = var_7673_equation_0, values = (var_7495_cast_fp16, var_6975_cast_fp16))[name = tensor("op_7673_cast_fp16")]; + tensor var_7674_to_fp16 = const()[name = tensor("op_7674_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_653_cast_fp16 = mul(x = var_7673_cast_fp16, y = var_7674_to_fp16)[name = tensor("aw_chunk_653_cast_fp16")]; + tensor var_7677_equation_0 = const()[name = tensor("op_7677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7677_cast_fp16 = einsum(equation = var_7677_equation_0, values = (var_7495_cast_fp16, var_6982_cast_fp16))[name = tensor("op_7677_cast_fp16")]; + tensor var_7678_to_fp16 = const()[name = tensor("op_7678_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_655_cast_fp16 = mul(x = var_7677_cast_fp16, y = var_7678_to_fp16)[name = tensor("aw_chunk_655_cast_fp16")]; + tensor var_7681_equation_0 = const()[name = tensor("op_7681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7681_cast_fp16 = einsum(equation = var_7681_equation_0, values = (var_7499_cast_fp16, var_6989_cast_fp16))[name = tensor("op_7681_cast_fp16")]; + tensor var_7682_to_fp16 = const()[name = tensor("op_7682_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_657_cast_fp16 = mul(x = var_7681_cast_fp16, y = var_7682_to_fp16)[name = tensor("aw_chunk_657_cast_fp16")]; + tensor var_7685_equation_0 = const()[name = tensor("op_7685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7685_cast_fp16 = einsum(equation = var_7685_equation_0, values = (var_7499_cast_fp16, var_6996_cast_fp16))[name = tensor("op_7685_cast_fp16")]; + tensor var_7686_to_fp16 = const()[name = tensor("op_7686_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_659_cast_fp16 = mul(x = var_7685_cast_fp16, y = var_7686_to_fp16)[name = tensor("aw_chunk_659_cast_fp16")]; + tensor var_7689_equation_0 = const()[name = tensor("op_7689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7689_cast_fp16 = einsum(equation = var_7689_equation_0, values = (var_7499_cast_fp16, var_7003_cast_fp16))[name = tensor("op_7689_cast_fp16")]; + tensor var_7690_to_fp16 = const()[name = tensor("op_7690_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_661_cast_fp16 = mul(x = var_7689_cast_fp16, y = var_7690_to_fp16)[name = tensor("aw_chunk_661_cast_fp16")]; + tensor var_7693_equation_0 = const()[name = tensor("op_7693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7693_cast_fp16 = einsum(equation = var_7693_equation_0, values = (var_7499_cast_fp16, var_7010_cast_fp16))[name = tensor("op_7693_cast_fp16")]; + tensor var_7694_to_fp16 = const()[name = tensor("op_7694_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_663_cast_fp16 = mul(x = var_7693_cast_fp16, y = var_7694_to_fp16)[name = tensor("aw_chunk_663_cast_fp16")]; + tensor var_7697_equation_0 = const()[name = tensor("op_7697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7697_cast_fp16 = einsum(equation = var_7697_equation_0, values = (var_7503_cast_fp16, var_7017_cast_fp16))[name = tensor("op_7697_cast_fp16")]; + tensor var_7698_to_fp16 = const()[name = tensor("op_7698_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_665_cast_fp16 = mul(x = var_7697_cast_fp16, y = var_7698_to_fp16)[name = tensor("aw_chunk_665_cast_fp16")]; + tensor var_7701_equation_0 = const()[name = tensor("op_7701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7701_cast_fp16 = einsum(equation = var_7701_equation_0, values = (var_7503_cast_fp16, var_7024_cast_fp16))[name = tensor("op_7701_cast_fp16")]; + tensor var_7702_to_fp16 = const()[name = tensor("op_7702_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_667_cast_fp16 = mul(x = var_7701_cast_fp16, y = var_7702_to_fp16)[name = tensor("aw_chunk_667_cast_fp16")]; + tensor var_7705_equation_0 = const()[name = tensor("op_7705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7705_cast_fp16 = einsum(equation = var_7705_equation_0, values = (var_7503_cast_fp16, var_7031_cast_fp16))[name = tensor("op_7705_cast_fp16")]; + tensor var_7706_to_fp16 = const()[name = tensor("op_7706_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_669_cast_fp16 = mul(x = var_7705_cast_fp16, y = var_7706_to_fp16)[name = tensor("aw_chunk_669_cast_fp16")]; + tensor var_7709_equation_0 = const()[name = tensor("op_7709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7709_cast_fp16 = einsum(equation = var_7709_equation_0, values = (var_7503_cast_fp16, var_7038_cast_fp16))[name = tensor("op_7709_cast_fp16")]; + tensor var_7710_to_fp16 = const()[name = tensor("op_7710_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_671_cast_fp16 = mul(x = var_7709_cast_fp16, y = var_7710_to_fp16)[name = tensor("aw_chunk_671_cast_fp16")]; + tensor var_7713_equation_0 = const()[name = tensor("op_7713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7713_cast_fp16 = einsum(equation = var_7713_equation_0, values = (var_7507_cast_fp16, var_7045_cast_fp16))[name = tensor("op_7713_cast_fp16")]; + tensor var_7714_to_fp16 = const()[name = tensor("op_7714_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_673_cast_fp16 = mul(x = var_7713_cast_fp16, y = var_7714_to_fp16)[name = tensor("aw_chunk_673_cast_fp16")]; + tensor var_7717_equation_0 = const()[name = tensor("op_7717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7717_cast_fp16 = einsum(equation = var_7717_equation_0, values = (var_7507_cast_fp16, var_7052_cast_fp16))[name = tensor("op_7717_cast_fp16")]; + tensor var_7718_to_fp16 = const()[name = tensor("op_7718_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_675_cast_fp16 = mul(x = var_7717_cast_fp16, y = var_7718_to_fp16)[name = tensor("aw_chunk_675_cast_fp16")]; + tensor var_7721_equation_0 = const()[name = tensor("op_7721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7721_cast_fp16 = einsum(equation = var_7721_equation_0, values = (var_7507_cast_fp16, var_7059_cast_fp16))[name = tensor("op_7721_cast_fp16")]; + tensor var_7722_to_fp16 = const()[name = tensor("op_7722_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_677_cast_fp16 = mul(x = var_7721_cast_fp16, y = var_7722_to_fp16)[name = tensor("aw_chunk_677_cast_fp16")]; + tensor var_7725_equation_0 = const()[name = tensor("op_7725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7725_cast_fp16 = einsum(equation = var_7725_equation_0, values = (var_7507_cast_fp16, var_7066_cast_fp16))[name = tensor("op_7725_cast_fp16")]; + tensor var_7726_to_fp16 = const()[name = tensor("op_7726_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_679_cast_fp16 = mul(x = var_7725_cast_fp16, y = var_7726_to_fp16)[name = tensor("aw_chunk_679_cast_fp16")]; + tensor var_7729_equation_0 = const()[name = tensor("op_7729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7729_cast_fp16 = einsum(equation = var_7729_equation_0, values = (var_7511_cast_fp16, var_7073_cast_fp16))[name = tensor("op_7729_cast_fp16")]; + tensor var_7730_to_fp16 = const()[name = tensor("op_7730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_681_cast_fp16 = mul(x = var_7729_cast_fp16, y = var_7730_to_fp16)[name = tensor("aw_chunk_681_cast_fp16")]; + tensor var_7733_equation_0 = const()[name = tensor("op_7733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7733_cast_fp16 = einsum(equation = var_7733_equation_0, values = (var_7511_cast_fp16, var_7080_cast_fp16))[name = tensor("op_7733_cast_fp16")]; + tensor var_7734_to_fp16 = const()[name = tensor("op_7734_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_683_cast_fp16 = mul(x = var_7733_cast_fp16, y = var_7734_to_fp16)[name = tensor("aw_chunk_683_cast_fp16")]; + tensor var_7737_equation_0 = const()[name = tensor("op_7737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7737_cast_fp16 = einsum(equation = var_7737_equation_0, values = (var_7511_cast_fp16, var_7087_cast_fp16))[name = tensor("op_7737_cast_fp16")]; + tensor var_7738_to_fp16 = const()[name = tensor("op_7738_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_685_cast_fp16 = mul(x = var_7737_cast_fp16, y = var_7738_to_fp16)[name = tensor("aw_chunk_685_cast_fp16")]; + tensor var_7741_equation_0 = const()[name = tensor("op_7741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7741_cast_fp16 = einsum(equation = var_7741_equation_0, values = (var_7511_cast_fp16, var_7094_cast_fp16))[name = tensor("op_7741_cast_fp16")]; + tensor var_7742_to_fp16 = const()[name = tensor("op_7742_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_687_cast_fp16 = mul(x = var_7741_cast_fp16, y = var_7742_to_fp16)[name = tensor("aw_chunk_687_cast_fp16")]; + tensor var_7745_equation_0 = const()[name = tensor("op_7745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7745_cast_fp16 = einsum(equation = var_7745_equation_0, values = (var_7515_cast_fp16, var_7101_cast_fp16))[name = tensor("op_7745_cast_fp16")]; + tensor var_7746_to_fp16 = const()[name = tensor("op_7746_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_689_cast_fp16 = mul(x = var_7745_cast_fp16, y = var_7746_to_fp16)[name = tensor("aw_chunk_689_cast_fp16")]; + tensor var_7749_equation_0 = const()[name = tensor("op_7749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7749_cast_fp16 = einsum(equation = var_7749_equation_0, values = (var_7515_cast_fp16, var_7108_cast_fp16))[name = tensor("op_7749_cast_fp16")]; + tensor var_7750_to_fp16 = const()[name = tensor("op_7750_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_691_cast_fp16 = mul(x = var_7749_cast_fp16, y = var_7750_to_fp16)[name = tensor("aw_chunk_691_cast_fp16")]; + tensor var_7753_equation_0 = const()[name = tensor("op_7753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7753_cast_fp16 = einsum(equation = var_7753_equation_0, values = (var_7515_cast_fp16, var_7115_cast_fp16))[name = tensor("op_7753_cast_fp16")]; + tensor var_7754_to_fp16 = const()[name = tensor("op_7754_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_693_cast_fp16 = mul(x = var_7753_cast_fp16, y = var_7754_to_fp16)[name = tensor("aw_chunk_693_cast_fp16")]; + tensor var_7757_equation_0 = const()[name = tensor("op_7757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7757_cast_fp16 = einsum(equation = var_7757_equation_0, values = (var_7515_cast_fp16, var_7122_cast_fp16))[name = tensor("op_7757_cast_fp16")]; + tensor var_7758_to_fp16 = const()[name = tensor("op_7758_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_695_cast_fp16 = mul(x = var_7757_cast_fp16, y = var_7758_to_fp16)[name = tensor("aw_chunk_695_cast_fp16")]; + tensor var_7761_equation_0 = const()[name = tensor("op_7761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7761_cast_fp16 = einsum(equation = var_7761_equation_0, values = (var_7519_cast_fp16, var_7129_cast_fp16))[name = tensor("op_7761_cast_fp16")]; + tensor var_7762_to_fp16 = const()[name = tensor("op_7762_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_697_cast_fp16 = mul(x = var_7761_cast_fp16, y = var_7762_to_fp16)[name = tensor("aw_chunk_697_cast_fp16")]; + tensor var_7765_equation_0 = const()[name = tensor("op_7765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7765_cast_fp16 = einsum(equation = var_7765_equation_0, values = (var_7519_cast_fp16, var_7136_cast_fp16))[name = tensor("op_7765_cast_fp16")]; + tensor var_7766_to_fp16 = const()[name = tensor("op_7766_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_699_cast_fp16 = mul(x = var_7765_cast_fp16, y = var_7766_to_fp16)[name = tensor("aw_chunk_699_cast_fp16")]; + tensor var_7769_equation_0 = const()[name = tensor("op_7769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7769_cast_fp16 = einsum(equation = var_7769_equation_0, values = (var_7519_cast_fp16, var_7143_cast_fp16))[name = tensor("op_7769_cast_fp16")]; + tensor var_7770_to_fp16 = const()[name = tensor("op_7770_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_701_cast_fp16 = mul(x = var_7769_cast_fp16, y = var_7770_to_fp16)[name = tensor("aw_chunk_701_cast_fp16")]; + tensor var_7773_equation_0 = const()[name = tensor("op_7773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7773_cast_fp16 = einsum(equation = var_7773_equation_0, values = (var_7519_cast_fp16, var_7150_cast_fp16))[name = tensor("op_7773_cast_fp16")]; + tensor var_7774_to_fp16 = const()[name = tensor("op_7774_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_703_cast_fp16 = mul(x = var_7773_cast_fp16, y = var_7774_to_fp16)[name = tensor("aw_chunk_703_cast_fp16")]; + tensor var_7777_equation_0 = const()[name = tensor("op_7777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7777_cast_fp16 = einsum(equation = var_7777_equation_0, values = (var_7523_cast_fp16, var_7157_cast_fp16))[name = tensor("op_7777_cast_fp16")]; + tensor var_7778_to_fp16 = const()[name = tensor("op_7778_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_705_cast_fp16 = mul(x = var_7777_cast_fp16, y = var_7778_to_fp16)[name = tensor("aw_chunk_705_cast_fp16")]; + tensor var_7781_equation_0 = const()[name = tensor("op_7781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7781_cast_fp16 = einsum(equation = var_7781_equation_0, values = (var_7523_cast_fp16, var_7164_cast_fp16))[name = tensor("op_7781_cast_fp16")]; + tensor var_7782_to_fp16 = const()[name = tensor("op_7782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_707_cast_fp16 = mul(x = var_7781_cast_fp16, y = var_7782_to_fp16)[name = tensor("aw_chunk_707_cast_fp16")]; + tensor var_7785_equation_0 = const()[name = tensor("op_7785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7785_cast_fp16 = einsum(equation = var_7785_equation_0, values = (var_7523_cast_fp16, var_7171_cast_fp16))[name = tensor("op_7785_cast_fp16")]; + tensor var_7786_to_fp16 = const()[name = tensor("op_7786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_709_cast_fp16 = mul(x = var_7785_cast_fp16, y = var_7786_to_fp16)[name = tensor("aw_chunk_709_cast_fp16")]; + tensor var_7789_equation_0 = const()[name = tensor("op_7789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7789_cast_fp16 = einsum(equation = var_7789_equation_0, values = (var_7523_cast_fp16, var_7178_cast_fp16))[name = tensor("op_7789_cast_fp16")]; + tensor var_7790_to_fp16 = const()[name = tensor("op_7790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_711_cast_fp16 = mul(x = var_7789_cast_fp16, y = var_7790_to_fp16)[name = tensor("aw_chunk_711_cast_fp16")]; + tensor var_7793_equation_0 = const()[name = tensor("op_7793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7793_cast_fp16 = einsum(equation = var_7793_equation_0, values = (var_7527_cast_fp16, var_7185_cast_fp16))[name = tensor("op_7793_cast_fp16")]; + tensor var_7794_to_fp16 = const()[name = tensor("op_7794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_713_cast_fp16 = mul(x = var_7793_cast_fp16, y = var_7794_to_fp16)[name = tensor("aw_chunk_713_cast_fp16")]; + tensor var_7797_equation_0 = const()[name = tensor("op_7797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7797_cast_fp16 = einsum(equation = var_7797_equation_0, values = (var_7527_cast_fp16, var_7192_cast_fp16))[name = tensor("op_7797_cast_fp16")]; + tensor var_7798_to_fp16 = const()[name = tensor("op_7798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_715_cast_fp16 = mul(x = var_7797_cast_fp16, y = var_7798_to_fp16)[name = tensor("aw_chunk_715_cast_fp16")]; + tensor var_7801_equation_0 = const()[name = tensor("op_7801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7801_cast_fp16 = einsum(equation = var_7801_equation_0, values = (var_7527_cast_fp16, var_7199_cast_fp16))[name = tensor("op_7801_cast_fp16")]; + tensor var_7802_to_fp16 = const()[name = tensor("op_7802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_717_cast_fp16 = mul(x = var_7801_cast_fp16, y = var_7802_to_fp16)[name = tensor("aw_chunk_717_cast_fp16")]; + tensor var_7805_equation_0 = const()[name = tensor("op_7805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7805_cast_fp16 = einsum(equation = var_7805_equation_0, values = (var_7527_cast_fp16, var_7206_cast_fp16))[name = tensor("op_7805_cast_fp16")]; + tensor var_7806_to_fp16 = const()[name = tensor("op_7806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_719_cast_fp16 = mul(x = var_7805_cast_fp16, y = var_7806_to_fp16)[name = tensor("aw_chunk_719_cast_fp16")]; + tensor var_7809_equation_0 = const()[name = tensor("op_7809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7809_cast_fp16 = einsum(equation = var_7809_equation_0, values = (var_7531_cast_fp16, var_7213_cast_fp16))[name = tensor("op_7809_cast_fp16")]; + tensor var_7810_to_fp16 = const()[name = tensor("op_7810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_721_cast_fp16 = mul(x = var_7809_cast_fp16, y = var_7810_to_fp16)[name = tensor("aw_chunk_721_cast_fp16")]; + tensor var_7813_equation_0 = const()[name = tensor("op_7813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7813_cast_fp16 = einsum(equation = var_7813_equation_0, values = (var_7531_cast_fp16, var_7220_cast_fp16))[name = tensor("op_7813_cast_fp16")]; + tensor var_7814_to_fp16 = const()[name = tensor("op_7814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_723_cast_fp16 = mul(x = var_7813_cast_fp16, y = var_7814_to_fp16)[name = tensor("aw_chunk_723_cast_fp16")]; + tensor var_7817_equation_0 = const()[name = tensor("op_7817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7817_cast_fp16 = einsum(equation = var_7817_equation_0, values = (var_7531_cast_fp16, var_7227_cast_fp16))[name = tensor("op_7817_cast_fp16")]; + tensor var_7818_to_fp16 = const()[name = tensor("op_7818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_725_cast_fp16 = mul(x = var_7817_cast_fp16, y = var_7818_to_fp16)[name = tensor("aw_chunk_725_cast_fp16")]; + tensor var_7821_equation_0 = const()[name = tensor("op_7821_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7821_cast_fp16 = einsum(equation = var_7821_equation_0, values = (var_7531_cast_fp16, var_7234_cast_fp16))[name = tensor("op_7821_cast_fp16")]; + tensor var_7822_to_fp16 = const()[name = tensor("op_7822_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_727_cast_fp16 = mul(x = var_7821_cast_fp16, y = var_7822_to_fp16)[name = tensor("aw_chunk_727_cast_fp16")]; + tensor var_7825_equation_0 = const()[name = tensor("op_7825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7825_cast_fp16 = einsum(equation = var_7825_equation_0, values = (var_7535_cast_fp16, var_7241_cast_fp16))[name = tensor("op_7825_cast_fp16")]; + tensor var_7826_to_fp16 = const()[name = tensor("op_7826_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_729_cast_fp16 = mul(x = var_7825_cast_fp16, y = var_7826_to_fp16)[name = tensor("aw_chunk_729_cast_fp16")]; + tensor var_7829_equation_0 = const()[name = tensor("op_7829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7829_cast_fp16 = einsum(equation = var_7829_equation_0, values = (var_7535_cast_fp16, var_7248_cast_fp16))[name = tensor("op_7829_cast_fp16")]; + tensor var_7830_to_fp16 = const()[name = tensor("op_7830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_731_cast_fp16 = mul(x = var_7829_cast_fp16, y = var_7830_to_fp16)[name = tensor("aw_chunk_731_cast_fp16")]; + tensor var_7833_equation_0 = const()[name = tensor("op_7833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7833_cast_fp16 = einsum(equation = var_7833_equation_0, values = (var_7535_cast_fp16, var_7255_cast_fp16))[name = tensor("op_7833_cast_fp16")]; + tensor var_7834_to_fp16 = const()[name = tensor("op_7834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_733_cast_fp16 = mul(x = var_7833_cast_fp16, y = var_7834_to_fp16)[name = tensor("aw_chunk_733_cast_fp16")]; + tensor var_7837_equation_0 = const()[name = tensor("op_7837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7837_cast_fp16 = einsum(equation = var_7837_equation_0, values = (var_7535_cast_fp16, var_7262_cast_fp16))[name = tensor("op_7837_cast_fp16")]; + tensor var_7838_to_fp16 = const()[name = tensor("op_7838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_735_cast_fp16 = mul(x = var_7837_cast_fp16, y = var_7838_to_fp16)[name = tensor("aw_chunk_735_cast_fp16")]; + tensor var_7841_equation_0 = const()[name = tensor("op_7841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7841_cast_fp16 = einsum(equation = var_7841_equation_0, values = (var_7539_cast_fp16, var_7269_cast_fp16))[name = tensor("op_7841_cast_fp16")]; + tensor var_7842_to_fp16 = const()[name = tensor("op_7842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_737_cast_fp16 = mul(x = var_7841_cast_fp16, y = var_7842_to_fp16)[name = tensor("aw_chunk_737_cast_fp16")]; + tensor var_7845_equation_0 = const()[name = tensor("op_7845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7845_cast_fp16 = einsum(equation = var_7845_equation_0, values = (var_7539_cast_fp16, var_7276_cast_fp16))[name = tensor("op_7845_cast_fp16")]; + tensor var_7846_to_fp16 = const()[name = tensor("op_7846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_739_cast_fp16 = mul(x = var_7845_cast_fp16, y = var_7846_to_fp16)[name = tensor("aw_chunk_739_cast_fp16")]; + tensor var_7849_equation_0 = const()[name = tensor("op_7849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7849_cast_fp16 = einsum(equation = var_7849_equation_0, values = (var_7539_cast_fp16, var_7283_cast_fp16))[name = tensor("op_7849_cast_fp16")]; + tensor var_7850_to_fp16 = const()[name = tensor("op_7850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_741_cast_fp16 = mul(x = var_7849_cast_fp16, y = var_7850_to_fp16)[name = tensor("aw_chunk_741_cast_fp16")]; + tensor var_7853_equation_0 = const()[name = tensor("op_7853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7853_cast_fp16 = einsum(equation = var_7853_equation_0, values = (var_7539_cast_fp16, var_7290_cast_fp16))[name = tensor("op_7853_cast_fp16")]; + tensor var_7854_to_fp16 = const()[name = tensor("op_7854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_743_cast_fp16 = mul(x = var_7853_cast_fp16, y = var_7854_to_fp16)[name = tensor("aw_chunk_743_cast_fp16")]; + tensor var_7857_equation_0 = const()[name = tensor("op_7857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7857_cast_fp16 = einsum(equation = var_7857_equation_0, values = (var_7543_cast_fp16, var_7297_cast_fp16))[name = tensor("op_7857_cast_fp16")]; + tensor var_7858_to_fp16 = const()[name = tensor("op_7858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_745_cast_fp16 = mul(x = var_7857_cast_fp16, y = var_7858_to_fp16)[name = tensor("aw_chunk_745_cast_fp16")]; + tensor var_7861_equation_0 = const()[name = tensor("op_7861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7861_cast_fp16 = einsum(equation = var_7861_equation_0, values = (var_7543_cast_fp16, var_7304_cast_fp16))[name = tensor("op_7861_cast_fp16")]; + tensor var_7862_to_fp16 = const()[name = tensor("op_7862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_747_cast_fp16 = mul(x = var_7861_cast_fp16, y = var_7862_to_fp16)[name = tensor("aw_chunk_747_cast_fp16")]; + tensor var_7865_equation_0 = const()[name = tensor("op_7865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7865_cast_fp16 = einsum(equation = var_7865_equation_0, values = (var_7543_cast_fp16, var_7311_cast_fp16))[name = tensor("op_7865_cast_fp16")]; + tensor var_7866_to_fp16 = const()[name = tensor("op_7866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_749_cast_fp16 = mul(x = var_7865_cast_fp16, y = var_7866_to_fp16)[name = tensor("aw_chunk_749_cast_fp16")]; + tensor var_7869_equation_0 = const()[name = tensor("op_7869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7869_cast_fp16 = einsum(equation = var_7869_equation_0, values = (var_7543_cast_fp16, var_7318_cast_fp16))[name = tensor("op_7869_cast_fp16")]; + tensor var_7870_to_fp16 = const()[name = tensor("op_7870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_751_cast_fp16 = mul(x = var_7869_cast_fp16, y = var_7870_to_fp16)[name = tensor("aw_chunk_751_cast_fp16")]; + tensor var_7873_equation_0 = const()[name = tensor("op_7873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7873_cast_fp16 = einsum(equation = var_7873_equation_0, values = (var_7547_cast_fp16, var_7325_cast_fp16))[name = tensor("op_7873_cast_fp16")]; + tensor var_7874_to_fp16 = const()[name = tensor("op_7874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_753_cast_fp16 = mul(x = var_7873_cast_fp16, y = var_7874_to_fp16)[name = tensor("aw_chunk_753_cast_fp16")]; + tensor var_7877_equation_0 = const()[name = tensor("op_7877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7877_cast_fp16 = einsum(equation = var_7877_equation_0, values = (var_7547_cast_fp16, var_7332_cast_fp16))[name = tensor("op_7877_cast_fp16")]; + tensor var_7878_to_fp16 = const()[name = tensor("op_7878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_755_cast_fp16 = mul(x = var_7877_cast_fp16, y = var_7878_to_fp16)[name = tensor("aw_chunk_755_cast_fp16")]; + tensor var_7881_equation_0 = const()[name = tensor("op_7881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7881_cast_fp16 = einsum(equation = var_7881_equation_0, values = (var_7547_cast_fp16, var_7339_cast_fp16))[name = tensor("op_7881_cast_fp16")]; + tensor var_7882_to_fp16 = const()[name = tensor("op_7882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_757_cast_fp16 = mul(x = var_7881_cast_fp16, y = var_7882_to_fp16)[name = tensor("aw_chunk_757_cast_fp16")]; + tensor var_7885_equation_0 = const()[name = tensor("op_7885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7885_cast_fp16 = einsum(equation = var_7885_equation_0, values = (var_7547_cast_fp16, var_7346_cast_fp16))[name = tensor("op_7885_cast_fp16")]; + tensor var_7886_to_fp16 = const()[name = tensor("op_7886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_759_cast_fp16 = mul(x = var_7885_cast_fp16, y = var_7886_to_fp16)[name = tensor("aw_chunk_759_cast_fp16")]; + tensor var_7889_equation_0 = const()[name = tensor("op_7889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7889_cast_fp16 = einsum(equation = var_7889_equation_0, values = (var_7551_cast_fp16, var_7353_cast_fp16))[name = tensor("op_7889_cast_fp16")]; + tensor var_7890_to_fp16 = const()[name = tensor("op_7890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_761_cast_fp16 = mul(x = var_7889_cast_fp16, y = var_7890_to_fp16)[name = tensor("aw_chunk_761_cast_fp16")]; + tensor var_7893_equation_0 = const()[name = tensor("op_7893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7893_cast_fp16 = einsum(equation = var_7893_equation_0, values = (var_7551_cast_fp16, var_7360_cast_fp16))[name = tensor("op_7893_cast_fp16")]; + tensor var_7894_to_fp16 = const()[name = tensor("op_7894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_763_cast_fp16 = mul(x = var_7893_cast_fp16, y = var_7894_to_fp16)[name = tensor("aw_chunk_763_cast_fp16")]; + tensor var_7897_equation_0 = const()[name = tensor("op_7897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7897_cast_fp16 = einsum(equation = var_7897_equation_0, values = (var_7551_cast_fp16, var_7367_cast_fp16))[name = tensor("op_7897_cast_fp16")]; + tensor var_7898_to_fp16 = const()[name = tensor("op_7898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_765_cast_fp16 = mul(x = var_7897_cast_fp16, y = var_7898_to_fp16)[name = tensor("aw_chunk_765_cast_fp16")]; + tensor var_7901_equation_0 = const()[name = tensor("op_7901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7901_cast_fp16 = einsum(equation = var_7901_equation_0, values = (var_7551_cast_fp16, var_7374_cast_fp16))[name = tensor("op_7901_cast_fp16")]; + tensor var_7902_to_fp16 = const()[name = tensor("op_7902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_767_cast_fp16 = mul(x = var_7901_cast_fp16, y = var_7902_to_fp16)[name = tensor("aw_chunk_767_cast_fp16")]; + tensor var_7905_equation_0 = const()[name = tensor("op_7905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7905_cast_fp16 = einsum(equation = var_7905_equation_0, values = (var_7555_cast_fp16, var_7381_cast_fp16))[name = tensor("op_7905_cast_fp16")]; + tensor var_7906_to_fp16 = const()[name = tensor("op_7906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_769_cast_fp16 = mul(x = var_7905_cast_fp16, y = var_7906_to_fp16)[name = tensor("aw_chunk_769_cast_fp16")]; + tensor var_7909_equation_0 = const()[name = tensor("op_7909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7909_cast_fp16 = einsum(equation = var_7909_equation_0, values = (var_7555_cast_fp16, var_7388_cast_fp16))[name = tensor("op_7909_cast_fp16")]; + tensor var_7910_to_fp16 = const()[name = tensor("op_7910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_771_cast_fp16 = mul(x = var_7909_cast_fp16, y = var_7910_to_fp16)[name = tensor("aw_chunk_771_cast_fp16")]; + tensor var_7913_equation_0 = const()[name = tensor("op_7913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7913_cast_fp16 = einsum(equation = var_7913_equation_0, values = (var_7555_cast_fp16, var_7395_cast_fp16))[name = tensor("op_7913_cast_fp16")]; + tensor var_7914_to_fp16 = const()[name = tensor("op_7914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_773_cast_fp16 = mul(x = var_7913_cast_fp16, y = var_7914_to_fp16)[name = tensor("aw_chunk_773_cast_fp16")]; + tensor var_7917_equation_0 = const()[name = tensor("op_7917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7917_cast_fp16 = einsum(equation = var_7917_equation_0, values = (var_7555_cast_fp16, var_7402_cast_fp16))[name = tensor("op_7917_cast_fp16")]; + tensor var_7918_to_fp16 = const()[name = tensor("op_7918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_775_cast_fp16 = mul(x = var_7917_cast_fp16, y = var_7918_to_fp16)[name = tensor("aw_chunk_775_cast_fp16")]; + tensor var_7921_equation_0 = const()[name = tensor("op_7921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7921_cast_fp16 = einsum(equation = var_7921_equation_0, values = (var_7559_cast_fp16, var_7409_cast_fp16))[name = tensor("op_7921_cast_fp16")]; + tensor var_7922_to_fp16 = const()[name = tensor("op_7922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_777_cast_fp16 = mul(x = var_7921_cast_fp16, y = var_7922_to_fp16)[name = tensor("aw_chunk_777_cast_fp16")]; + tensor var_7925_equation_0 = const()[name = tensor("op_7925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7925_cast_fp16 = einsum(equation = var_7925_equation_0, values = (var_7559_cast_fp16, var_7416_cast_fp16))[name = tensor("op_7925_cast_fp16")]; + tensor var_7926_to_fp16 = const()[name = tensor("op_7926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_779_cast_fp16 = mul(x = var_7925_cast_fp16, y = var_7926_to_fp16)[name = tensor("aw_chunk_779_cast_fp16")]; + tensor var_7929_equation_0 = const()[name = tensor("op_7929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7929_cast_fp16 = einsum(equation = var_7929_equation_0, values = (var_7559_cast_fp16, var_7423_cast_fp16))[name = tensor("op_7929_cast_fp16")]; + tensor var_7930_to_fp16 = const()[name = tensor("op_7930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_781_cast_fp16 = mul(x = var_7929_cast_fp16, y = var_7930_to_fp16)[name = tensor("aw_chunk_781_cast_fp16")]; + tensor var_7933_equation_0 = const()[name = tensor("op_7933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7933_cast_fp16 = einsum(equation = var_7933_equation_0, values = (var_7559_cast_fp16, var_7430_cast_fp16))[name = tensor("op_7933_cast_fp16")]; + tensor var_7934_to_fp16 = const()[name = tensor("op_7934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_783_cast_fp16 = mul(x = var_7933_cast_fp16, y = var_7934_to_fp16)[name = tensor("aw_chunk_783_cast_fp16")]; + tensor var_7937_equation_0 = const()[name = tensor("op_7937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7937_cast_fp16 = einsum(equation = var_7937_equation_0, values = (var_7563_cast_fp16, var_7437_cast_fp16))[name = tensor("op_7937_cast_fp16")]; + tensor var_7938_to_fp16 = const()[name = tensor("op_7938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_785_cast_fp16 = mul(x = var_7937_cast_fp16, y = var_7938_to_fp16)[name = tensor("aw_chunk_785_cast_fp16")]; + tensor var_7941_equation_0 = const()[name = tensor("op_7941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7941_cast_fp16 = einsum(equation = var_7941_equation_0, values = (var_7563_cast_fp16, var_7444_cast_fp16))[name = tensor("op_7941_cast_fp16")]; + tensor var_7942_to_fp16 = const()[name = tensor("op_7942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_787_cast_fp16 = mul(x = var_7941_cast_fp16, y = var_7942_to_fp16)[name = tensor("aw_chunk_787_cast_fp16")]; + tensor var_7945_equation_0 = const()[name = tensor("op_7945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7945_cast_fp16 = einsum(equation = var_7945_equation_0, values = (var_7563_cast_fp16, var_7451_cast_fp16))[name = tensor("op_7945_cast_fp16")]; + tensor var_7946_to_fp16 = const()[name = tensor("op_7946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_789_cast_fp16 = mul(x = var_7945_cast_fp16, y = var_7946_to_fp16)[name = tensor("aw_chunk_789_cast_fp16")]; + tensor var_7949_equation_0 = const()[name = tensor("op_7949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7949_cast_fp16 = einsum(equation = var_7949_equation_0, values = (var_7563_cast_fp16, var_7458_cast_fp16))[name = tensor("op_7949_cast_fp16")]; + tensor var_7950_to_fp16 = const()[name = tensor("op_7950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_791_cast_fp16 = mul(x = var_7949_cast_fp16, y = var_7950_to_fp16)[name = tensor("aw_chunk_791_cast_fp16")]; + tensor var_7953_equation_0 = const()[name = tensor("op_7953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7953_cast_fp16 = einsum(equation = var_7953_equation_0, values = (var_7567_cast_fp16, var_7465_cast_fp16))[name = tensor("op_7953_cast_fp16")]; + tensor var_7954_to_fp16 = const()[name = tensor("op_7954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_793_cast_fp16 = mul(x = var_7953_cast_fp16, y = var_7954_to_fp16)[name = tensor("aw_chunk_793_cast_fp16")]; + tensor var_7957_equation_0 = const()[name = tensor("op_7957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7957_cast_fp16 = einsum(equation = var_7957_equation_0, values = (var_7567_cast_fp16, var_7472_cast_fp16))[name = tensor("op_7957_cast_fp16")]; + tensor var_7958_to_fp16 = const()[name = tensor("op_7958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_795_cast_fp16 = mul(x = var_7957_cast_fp16, y = var_7958_to_fp16)[name = tensor("aw_chunk_795_cast_fp16")]; + tensor var_7961_equation_0 = const()[name = tensor("op_7961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7961_cast_fp16 = einsum(equation = var_7961_equation_0, values = (var_7567_cast_fp16, var_7479_cast_fp16))[name = tensor("op_7961_cast_fp16")]; + tensor var_7962_to_fp16 = const()[name = tensor("op_7962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_797_cast_fp16 = mul(x = var_7961_cast_fp16, y = var_7962_to_fp16)[name = tensor("aw_chunk_797_cast_fp16")]; + tensor var_7965_equation_0 = const()[name = tensor("op_7965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_7965_cast_fp16 = einsum(equation = var_7965_equation_0, values = (var_7567_cast_fp16, var_7486_cast_fp16))[name = tensor("op_7965_cast_fp16")]; + tensor var_7966_to_fp16 = const()[name = tensor("op_7966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_799_cast_fp16 = mul(x = var_7965_cast_fp16, y = var_7966_to_fp16)[name = tensor("aw_chunk_799_cast_fp16")]; + tensor var_7968_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_641_cast_fp16)[name = tensor("op_7968_cast_fp16")]; + tensor var_7969_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_643_cast_fp16)[name = tensor("op_7969_cast_fp16")]; + tensor var_7970_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_645_cast_fp16)[name = tensor("op_7970_cast_fp16")]; + tensor var_7971_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_647_cast_fp16)[name = tensor("op_7971_cast_fp16")]; + tensor var_7972_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_649_cast_fp16)[name = tensor("op_7972_cast_fp16")]; + tensor var_7973_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_651_cast_fp16)[name = tensor("op_7973_cast_fp16")]; + tensor var_7974_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_653_cast_fp16)[name = tensor("op_7974_cast_fp16")]; + tensor var_7975_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_655_cast_fp16)[name = tensor("op_7975_cast_fp16")]; + tensor var_7976_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_657_cast_fp16)[name = tensor("op_7976_cast_fp16")]; + tensor var_7977_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_659_cast_fp16)[name = tensor("op_7977_cast_fp16")]; + tensor var_7978_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_661_cast_fp16)[name = tensor("op_7978_cast_fp16")]; + tensor var_7979_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_663_cast_fp16)[name = tensor("op_7979_cast_fp16")]; + tensor var_7980_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_665_cast_fp16)[name = tensor("op_7980_cast_fp16")]; + tensor var_7981_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_667_cast_fp16)[name = tensor("op_7981_cast_fp16")]; + tensor var_7982_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_669_cast_fp16)[name = tensor("op_7982_cast_fp16")]; + tensor var_7983_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_671_cast_fp16)[name = tensor("op_7983_cast_fp16")]; + tensor var_7984_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_673_cast_fp16)[name = tensor("op_7984_cast_fp16")]; + tensor var_7985_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_675_cast_fp16)[name = tensor("op_7985_cast_fp16")]; + tensor var_7986_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_677_cast_fp16)[name = tensor("op_7986_cast_fp16")]; + tensor var_7987_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_679_cast_fp16)[name = tensor("op_7987_cast_fp16")]; + tensor var_7988_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_681_cast_fp16)[name = tensor("op_7988_cast_fp16")]; + tensor var_7989_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_683_cast_fp16)[name = tensor("op_7989_cast_fp16")]; + tensor var_7990_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_685_cast_fp16)[name = tensor("op_7990_cast_fp16")]; + tensor var_7991_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_687_cast_fp16)[name = tensor("op_7991_cast_fp16")]; + tensor var_7992_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_689_cast_fp16)[name = tensor("op_7992_cast_fp16")]; + tensor var_7993_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_691_cast_fp16)[name = tensor("op_7993_cast_fp16")]; + tensor var_7994_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_693_cast_fp16)[name = tensor("op_7994_cast_fp16")]; + tensor var_7995_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_695_cast_fp16)[name = tensor("op_7995_cast_fp16")]; + tensor var_7996_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_697_cast_fp16)[name = tensor("op_7996_cast_fp16")]; + tensor var_7997_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_699_cast_fp16)[name = tensor("op_7997_cast_fp16")]; + tensor var_7998_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_701_cast_fp16)[name = tensor("op_7998_cast_fp16")]; + tensor var_7999_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_703_cast_fp16)[name = tensor("op_7999_cast_fp16")]; + tensor var_8000_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_705_cast_fp16)[name = tensor("op_8000_cast_fp16")]; + tensor var_8001_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_707_cast_fp16)[name = tensor("op_8001_cast_fp16")]; + tensor var_8002_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_709_cast_fp16)[name = tensor("op_8002_cast_fp16")]; + tensor var_8003_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_711_cast_fp16)[name = tensor("op_8003_cast_fp16")]; + tensor var_8004_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_713_cast_fp16)[name = tensor("op_8004_cast_fp16")]; + tensor var_8005_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_715_cast_fp16)[name = tensor("op_8005_cast_fp16")]; + tensor var_8006_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_717_cast_fp16)[name = tensor("op_8006_cast_fp16")]; + tensor var_8007_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_719_cast_fp16)[name = tensor("op_8007_cast_fp16")]; + tensor var_8008_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_721_cast_fp16)[name = tensor("op_8008_cast_fp16")]; + tensor var_8009_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_723_cast_fp16)[name = tensor("op_8009_cast_fp16")]; + tensor var_8010_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_725_cast_fp16)[name = tensor("op_8010_cast_fp16")]; + tensor var_8011_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_727_cast_fp16)[name = tensor("op_8011_cast_fp16")]; + tensor var_8012_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_729_cast_fp16)[name = tensor("op_8012_cast_fp16")]; + tensor var_8013_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_731_cast_fp16)[name = tensor("op_8013_cast_fp16")]; + tensor var_8014_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_733_cast_fp16)[name = tensor("op_8014_cast_fp16")]; + tensor var_8015_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_735_cast_fp16)[name = tensor("op_8015_cast_fp16")]; + tensor var_8016_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_737_cast_fp16)[name = tensor("op_8016_cast_fp16")]; + tensor var_8017_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_739_cast_fp16)[name = tensor("op_8017_cast_fp16")]; + tensor var_8018_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_741_cast_fp16)[name = tensor("op_8018_cast_fp16")]; + tensor var_8019_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_743_cast_fp16)[name = tensor("op_8019_cast_fp16")]; + tensor var_8020_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_745_cast_fp16)[name = tensor("op_8020_cast_fp16")]; + tensor var_8021_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_747_cast_fp16)[name = tensor("op_8021_cast_fp16")]; + tensor var_8022_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_749_cast_fp16)[name = tensor("op_8022_cast_fp16")]; + tensor var_8023_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_751_cast_fp16)[name = tensor("op_8023_cast_fp16")]; + tensor var_8024_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_753_cast_fp16)[name = tensor("op_8024_cast_fp16")]; + tensor var_8025_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_755_cast_fp16)[name = tensor("op_8025_cast_fp16")]; + tensor var_8026_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_757_cast_fp16)[name = tensor("op_8026_cast_fp16")]; + tensor var_8027_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_759_cast_fp16)[name = tensor("op_8027_cast_fp16")]; + tensor var_8028_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_761_cast_fp16)[name = tensor("op_8028_cast_fp16")]; + tensor var_8029_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_763_cast_fp16)[name = tensor("op_8029_cast_fp16")]; + tensor var_8030_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_765_cast_fp16)[name = tensor("op_8030_cast_fp16")]; + tensor var_8031_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_767_cast_fp16)[name = tensor("op_8031_cast_fp16")]; + tensor var_8032_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_769_cast_fp16)[name = tensor("op_8032_cast_fp16")]; + tensor var_8033_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_771_cast_fp16)[name = tensor("op_8033_cast_fp16")]; + tensor var_8034_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_773_cast_fp16)[name = tensor("op_8034_cast_fp16")]; + tensor var_8035_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_775_cast_fp16)[name = tensor("op_8035_cast_fp16")]; + tensor var_8036_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_777_cast_fp16)[name = tensor("op_8036_cast_fp16")]; + tensor var_8037_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_779_cast_fp16)[name = tensor("op_8037_cast_fp16")]; + tensor var_8038_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_781_cast_fp16)[name = tensor("op_8038_cast_fp16")]; + tensor var_8039_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_783_cast_fp16)[name = tensor("op_8039_cast_fp16")]; + tensor var_8040_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_785_cast_fp16)[name = tensor("op_8040_cast_fp16")]; + tensor var_8041_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_787_cast_fp16)[name = tensor("op_8041_cast_fp16")]; + tensor var_8042_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_789_cast_fp16)[name = tensor("op_8042_cast_fp16")]; + tensor var_8043_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_791_cast_fp16)[name = tensor("op_8043_cast_fp16")]; + tensor var_8044_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_793_cast_fp16)[name = tensor("op_8044_cast_fp16")]; + tensor var_8045_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_795_cast_fp16)[name = tensor("op_8045_cast_fp16")]; + tensor var_8046_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_797_cast_fp16)[name = tensor("op_8046_cast_fp16")]; + tensor var_8047_cast_fp16 = softmax(axis = var_6739, x = aw_chunk_799_cast_fp16)[name = tensor("op_8047_cast_fp16")]; + tensor var_8049_equation_0 = const()[name = tensor("op_8049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8049_cast_fp16 = einsum(equation = var_8049_equation_0, values = (var_7569_cast_fp16, var_7968_cast_fp16))[name = tensor("op_8049_cast_fp16")]; + tensor var_8051_equation_0 = const()[name = tensor("op_8051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8051_cast_fp16 = einsum(equation = var_8051_equation_0, values = (var_7569_cast_fp16, var_7969_cast_fp16))[name = tensor("op_8051_cast_fp16")]; + tensor var_8053_equation_0 = const()[name = tensor("op_8053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8053_cast_fp16 = einsum(equation = var_8053_equation_0, values = (var_7569_cast_fp16, var_7970_cast_fp16))[name = tensor("op_8053_cast_fp16")]; + tensor var_8055_equation_0 = const()[name = tensor("op_8055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8055_cast_fp16 = einsum(equation = var_8055_equation_0, values = (var_7569_cast_fp16, var_7971_cast_fp16))[name = tensor("op_8055_cast_fp16")]; + tensor var_8057_equation_0 = const()[name = tensor("op_8057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8057_cast_fp16 = einsum(equation = var_8057_equation_0, values = (var_7573_cast_fp16, var_7972_cast_fp16))[name = tensor("op_8057_cast_fp16")]; + tensor var_8059_equation_0 = const()[name = tensor("op_8059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8059_cast_fp16 = einsum(equation = var_8059_equation_0, values = (var_7573_cast_fp16, var_7973_cast_fp16))[name = tensor("op_8059_cast_fp16")]; + tensor var_8061_equation_0 = const()[name = tensor("op_8061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8061_cast_fp16 = einsum(equation = var_8061_equation_0, values = (var_7573_cast_fp16, var_7974_cast_fp16))[name = tensor("op_8061_cast_fp16")]; + tensor var_8063_equation_0 = const()[name = tensor("op_8063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8063_cast_fp16 = einsum(equation = var_8063_equation_0, values = (var_7573_cast_fp16, var_7975_cast_fp16))[name = tensor("op_8063_cast_fp16")]; + tensor var_8065_equation_0 = const()[name = tensor("op_8065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8065_cast_fp16 = einsum(equation = var_8065_equation_0, values = (var_7577_cast_fp16, var_7976_cast_fp16))[name = tensor("op_8065_cast_fp16")]; + tensor var_8067_equation_0 = const()[name = tensor("op_8067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8067_cast_fp16 = einsum(equation = var_8067_equation_0, values = (var_7577_cast_fp16, var_7977_cast_fp16))[name = tensor("op_8067_cast_fp16")]; + tensor var_8069_equation_0 = const()[name = tensor("op_8069_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8069_cast_fp16 = einsum(equation = var_8069_equation_0, values = (var_7577_cast_fp16, var_7978_cast_fp16))[name = tensor("op_8069_cast_fp16")]; + tensor var_8071_equation_0 = const()[name = tensor("op_8071_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8071_cast_fp16 = einsum(equation = var_8071_equation_0, values = (var_7577_cast_fp16, var_7979_cast_fp16))[name = tensor("op_8071_cast_fp16")]; + tensor var_8073_equation_0 = const()[name = tensor("op_8073_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8073_cast_fp16 = einsum(equation = var_8073_equation_0, values = (var_7581_cast_fp16, var_7980_cast_fp16))[name = tensor("op_8073_cast_fp16")]; + tensor var_8075_equation_0 = const()[name = tensor("op_8075_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8075_cast_fp16 = einsum(equation = var_8075_equation_0, values = (var_7581_cast_fp16, var_7981_cast_fp16))[name = tensor("op_8075_cast_fp16")]; + tensor var_8077_equation_0 = const()[name = tensor("op_8077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8077_cast_fp16 = einsum(equation = var_8077_equation_0, values = (var_7581_cast_fp16, var_7982_cast_fp16))[name = tensor("op_8077_cast_fp16")]; + tensor var_8079_equation_0 = const()[name = tensor("op_8079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8079_cast_fp16 = einsum(equation = var_8079_equation_0, values = (var_7581_cast_fp16, var_7983_cast_fp16))[name = tensor("op_8079_cast_fp16")]; + tensor var_8081_equation_0 = const()[name = tensor("op_8081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8081_cast_fp16 = einsum(equation = var_8081_equation_0, values = (var_7585_cast_fp16, var_7984_cast_fp16))[name = tensor("op_8081_cast_fp16")]; + tensor var_8083_equation_0 = const()[name = tensor("op_8083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8083_cast_fp16 = einsum(equation = var_8083_equation_0, values = (var_7585_cast_fp16, var_7985_cast_fp16))[name = tensor("op_8083_cast_fp16")]; + tensor var_8085_equation_0 = const()[name = tensor("op_8085_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8085_cast_fp16 = einsum(equation = var_8085_equation_0, values = (var_7585_cast_fp16, var_7986_cast_fp16))[name = tensor("op_8085_cast_fp16")]; + tensor var_8087_equation_0 = const()[name = tensor("op_8087_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8087_cast_fp16 = einsum(equation = var_8087_equation_0, values = (var_7585_cast_fp16, var_7987_cast_fp16))[name = tensor("op_8087_cast_fp16")]; + tensor var_8089_equation_0 = const()[name = tensor("op_8089_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8089_cast_fp16 = einsum(equation = var_8089_equation_0, values = (var_7589_cast_fp16, var_7988_cast_fp16))[name = tensor("op_8089_cast_fp16")]; + tensor var_8091_equation_0 = const()[name = tensor("op_8091_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8091_cast_fp16 = einsum(equation = var_8091_equation_0, values = (var_7589_cast_fp16, var_7989_cast_fp16))[name = tensor("op_8091_cast_fp16")]; + tensor var_8093_equation_0 = const()[name = tensor("op_8093_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8093_cast_fp16 = einsum(equation = var_8093_equation_0, values = (var_7589_cast_fp16, var_7990_cast_fp16))[name = tensor("op_8093_cast_fp16")]; + tensor var_8095_equation_0 = const()[name = tensor("op_8095_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8095_cast_fp16 = einsum(equation = var_8095_equation_0, values = (var_7589_cast_fp16, var_7991_cast_fp16))[name = tensor("op_8095_cast_fp16")]; + tensor var_8097_equation_0 = const()[name = tensor("op_8097_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8097_cast_fp16 = einsum(equation = var_8097_equation_0, values = (var_7593_cast_fp16, var_7992_cast_fp16))[name = tensor("op_8097_cast_fp16")]; + tensor var_8099_equation_0 = const()[name = tensor("op_8099_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8099_cast_fp16 = einsum(equation = var_8099_equation_0, values = (var_7593_cast_fp16, var_7993_cast_fp16))[name = tensor("op_8099_cast_fp16")]; + tensor var_8101_equation_0 = const()[name = tensor("op_8101_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8101_cast_fp16 = einsum(equation = var_8101_equation_0, values = (var_7593_cast_fp16, var_7994_cast_fp16))[name = tensor("op_8101_cast_fp16")]; + tensor var_8103_equation_0 = const()[name = tensor("op_8103_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8103_cast_fp16 = einsum(equation = var_8103_equation_0, values = (var_7593_cast_fp16, var_7995_cast_fp16))[name = tensor("op_8103_cast_fp16")]; + tensor var_8105_equation_0 = const()[name = tensor("op_8105_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8105_cast_fp16 = einsum(equation = var_8105_equation_0, values = (var_7597_cast_fp16, var_7996_cast_fp16))[name = tensor("op_8105_cast_fp16")]; + tensor var_8107_equation_0 = const()[name = tensor("op_8107_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8107_cast_fp16 = einsum(equation = var_8107_equation_0, values = (var_7597_cast_fp16, var_7997_cast_fp16))[name = tensor("op_8107_cast_fp16")]; + tensor var_8109_equation_0 = const()[name = tensor("op_8109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8109_cast_fp16 = einsum(equation = var_8109_equation_0, values = (var_7597_cast_fp16, var_7998_cast_fp16))[name = tensor("op_8109_cast_fp16")]; + tensor var_8111_equation_0 = const()[name = tensor("op_8111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8111_cast_fp16 = einsum(equation = var_8111_equation_0, values = (var_7597_cast_fp16, var_7999_cast_fp16))[name = tensor("op_8111_cast_fp16")]; + tensor var_8113_equation_0 = const()[name = tensor("op_8113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8113_cast_fp16 = einsum(equation = var_8113_equation_0, values = (var_7601_cast_fp16, var_8000_cast_fp16))[name = tensor("op_8113_cast_fp16")]; + tensor var_8115_equation_0 = const()[name = tensor("op_8115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8115_cast_fp16 = einsum(equation = var_8115_equation_0, values = (var_7601_cast_fp16, var_8001_cast_fp16))[name = tensor("op_8115_cast_fp16")]; + tensor var_8117_equation_0 = const()[name = tensor("op_8117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8117_cast_fp16 = einsum(equation = var_8117_equation_0, values = (var_7601_cast_fp16, var_8002_cast_fp16))[name = tensor("op_8117_cast_fp16")]; + tensor var_8119_equation_0 = const()[name = tensor("op_8119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8119_cast_fp16 = einsum(equation = var_8119_equation_0, values = (var_7601_cast_fp16, var_8003_cast_fp16))[name = tensor("op_8119_cast_fp16")]; + tensor var_8121_equation_0 = const()[name = tensor("op_8121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8121_cast_fp16 = einsum(equation = var_8121_equation_0, values = (var_7605_cast_fp16, var_8004_cast_fp16))[name = tensor("op_8121_cast_fp16")]; + tensor var_8123_equation_0 = const()[name = tensor("op_8123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8123_cast_fp16 = einsum(equation = var_8123_equation_0, values = (var_7605_cast_fp16, var_8005_cast_fp16))[name = tensor("op_8123_cast_fp16")]; + tensor var_8125_equation_0 = const()[name = tensor("op_8125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8125_cast_fp16 = einsum(equation = var_8125_equation_0, values = (var_7605_cast_fp16, var_8006_cast_fp16))[name = tensor("op_8125_cast_fp16")]; + tensor var_8127_equation_0 = const()[name = tensor("op_8127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8127_cast_fp16 = einsum(equation = var_8127_equation_0, values = (var_7605_cast_fp16, var_8007_cast_fp16))[name = tensor("op_8127_cast_fp16")]; + tensor var_8129_equation_0 = const()[name = tensor("op_8129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8129_cast_fp16 = einsum(equation = var_8129_equation_0, values = (var_7609_cast_fp16, var_8008_cast_fp16))[name = tensor("op_8129_cast_fp16")]; + tensor var_8131_equation_0 = const()[name = tensor("op_8131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8131_cast_fp16 = einsum(equation = var_8131_equation_0, values = (var_7609_cast_fp16, var_8009_cast_fp16))[name = tensor("op_8131_cast_fp16")]; + tensor var_8133_equation_0 = const()[name = tensor("op_8133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8133_cast_fp16 = einsum(equation = var_8133_equation_0, values = (var_7609_cast_fp16, var_8010_cast_fp16))[name = tensor("op_8133_cast_fp16")]; + tensor var_8135_equation_0 = const()[name = tensor("op_8135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8135_cast_fp16 = einsum(equation = var_8135_equation_0, values = (var_7609_cast_fp16, var_8011_cast_fp16))[name = tensor("op_8135_cast_fp16")]; + tensor var_8137_equation_0 = const()[name = tensor("op_8137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8137_cast_fp16 = einsum(equation = var_8137_equation_0, values = (var_7613_cast_fp16, var_8012_cast_fp16))[name = tensor("op_8137_cast_fp16")]; + tensor var_8139_equation_0 = const()[name = tensor("op_8139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8139_cast_fp16 = einsum(equation = var_8139_equation_0, values = (var_7613_cast_fp16, var_8013_cast_fp16))[name = tensor("op_8139_cast_fp16")]; + tensor var_8141_equation_0 = const()[name = tensor("op_8141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8141_cast_fp16 = einsum(equation = var_8141_equation_0, values = (var_7613_cast_fp16, var_8014_cast_fp16))[name = tensor("op_8141_cast_fp16")]; + tensor var_8143_equation_0 = const()[name = tensor("op_8143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8143_cast_fp16 = einsum(equation = var_8143_equation_0, values = (var_7613_cast_fp16, var_8015_cast_fp16))[name = tensor("op_8143_cast_fp16")]; + tensor var_8145_equation_0 = const()[name = tensor("op_8145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8145_cast_fp16 = einsum(equation = var_8145_equation_0, values = (var_7617_cast_fp16, var_8016_cast_fp16))[name = tensor("op_8145_cast_fp16")]; + tensor var_8147_equation_0 = const()[name = tensor("op_8147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8147_cast_fp16 = einsum(equation = var_8147_equation_0, values = (var_7617_cast_fp16, var_8017_cast_fp16))[name = tensor("op_8147_cast_fp16")]; + tensor var_8149_equation_0 = const()[name = tensor("op_8149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8149_cast_fp16 = einsum(equation = var_8149_equation_0, values = (var_7617_cast_fp16, var_8018_cast_fp16))[name = tensor("op_8149_cast_fp16")]; + tensor var_8151_equation_0 = const()[name = tensor("op_8151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8151_cast_fp16 = einsum(equation = var_8151_equation_0, values = (var_7617_cast_fp16, var_8019_cast_fp16))[name = tensor("op_8151_cast_fp16")]; + tensor var_8153_equation_0 = const()[name = tensor("op_8153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8153_cast_fp16 = einsum(equation = var_8153_equation_0, values = (var_7621_cast_fp16, var_8020_cast_fp16))[name = tensor("op_8153_cast_fp16")]; + tensor var_8155_equation_0 = const()[name = tensor("op_8155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8155_cast_fp16 = einsum(equation = var_8155_equation_0, values = (var_7621_cast_fp16, var_8021_cast_fp16))[name = tensor("op_8155_cast_fp16")]; + tensor var_8157_equation_0 = const()[name = tensor("op_8157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8157_cast_fp16 = einsum(equation = var_8157_equation_0, values = (var_7621_cast_fp16, var_8022_cast_fp16))[name = tensor("op_8157_cast_fp16")]; + tensor var_8159_equation_0 = const()[name = tensor("op_8159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8159_cast_fp16 = einsum(equation = var_8159_equation_0, values = (var_7621_cast_fp16, var_8023_cast_fp16))[name = tensor("op_8159_cast_fp16")]; + tensor var_8161_equation_0 = const()[name = tensor("op_8161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8161_cast_fp16 = einsum(equation = var_8161_equation_0, values = (var_7625_cast_fp16, var_8024_cast_fp16))[name = tensor("op_8161_cast_fp16")]; + tensor var_8163_equation_0 = const()[name = tensor("op_8163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8163_cast_fp16 = einsum(equation = var_8163_equation_0, values = (var_7625_cast_fp16, var_8025_cast_fp16))[name = tensor("op_8163_cast_fp16")]; + tensor var_8165_equation_0 = const()[name = tensor("op_8165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8165_cast_fp16 = einsum(equation = var_8165_equation_0, values = (var_7625_cast_fp16, var_8026_cast_fp16))[name = tensor("op_8165_cast_fp16")]; + tensor var_8167_equation_0 = const()[name = tensor("op_8167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8167_cast_fp16 = einsum(equation = var_8167_equation_0, values = (var_7625_cast_fp16, var_8027_cast_fp16))[name = tensor("op_8167_cast_fp16")]; + tensor var_8169_equation_0 = const()[name = tensor("op_8169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8169_cast_fp16 = einsum(equation = var_8169_equation_0, values = (var_7629_cast_fp16, var_8028_cast_fp16))[name = tensor("op_8169_cast_fp16")]; + tensor var_8171_equation_0 = const()[name = tensor("op_8171_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8171_cast_fp16 = einsum(equation = var_8171_equation_0, values = (var_7629_cast_fp16, var_8029_cast_fp16))[name = tensor("op_8171_cast_fp16")]; + tensor var_8173_equation_0 = const()[name = tensor("op_8173_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8173_cast_fp16 = einsum(equation = var_8173_equation_0, values = (var_7629_cast_fp16, var_8030_cast_fp16))[name = tensor("op_8173_cast_fp16")]; + tensor var_8175_equation_0 = const()[name = tensor("op_8175_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8175_cast_fp16 = einsum(equation = var_8175_equation_0, values = (var_7629_cast_fp16, var_8031_cast_fp16))[name = tensor("op_8175_cast_fp16")]; + tensor var_8177_equation_0 = const()[name = tensor("op_8177_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8177_cast_fp16 = einsum(equation = var_8177_equation_0, values = (var_7633_cast_fp16, var_8032_cast_fp16))[name = tensor("op_8177_cast_fp16")]; + tensor var_8179_equation_0 = const()[name = tensor("op_8179_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8179_cast_fp16 = einsum(equation = var_8179_equation_0, values = (var_7633_cast_fp16, var_8033_cast_fp16))[name = tensor("op_8179_cast_fp16")]; + tensor var_8181_equation_0 = const()[name = tensor("op_8181_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8181_cast_fp16 = einsum(equation = var_8181_equation_0, values = (var_7633_cast_fp16, var_8034_cast_fp16))[name = tensor("op_8181_cast_fp16")]; + tensor var_8183_equation_0 = const()[name = tensor("op_8183_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8183_cast_fp16 = einsum(equation = var_8183_equation_0, values = (var_7633_cast_fp16, var_8035_cast_fp16))[name = tensor("op_8183_cast_fp16")]; + tensor var_8185_equation_0 = const()[name = tensor("op_8185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8185_cast_fp16 = einsum(equation = var_8185_equation_0, values = (var_7637_cast_fp16, var_8036_cast_fp16))[name = tensor("op_8185_cast_fp16")]; + tensor var_8187_equation_0 = const()[name = tensor("op_8187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8187_cast_fp16 = einsum(equation = var_8187_equation_0, values = (var_7637_cast_fp16, var_8037_cast_fp16))[name = tensor("op_8187_cast_fp16")]; + tensor var_8189_equation_0 = const()[name = tensor("op_8189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8189_cast_fp16 = einsum(equation = var_8189_equation_0, values = (var_7637_cast_fp16, var_8038_cast_fp16))[name = tensor("op_8189_cast_fp16")]; + tensor var_8191_equation_0 = const()[name = tensor("op_8191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8191_cast_fp16 = einsum(equation = var_8191_equation_0, values = (var_7637_cast_fp16, var_8039_cast_fp16))[name = tensor("op_8191_cast_fp16")]; + tensor var_8193_equation_0 = const()[name = tensor("op_8193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8193_cast_fp16 = einsum(equation = var_8193_equation_0, values = (var_7641_cast_fp16, var_8040_cast_fp16))[name = tensor("op_8193_cast_fp16")]; + tensor var_8195_equation_0 = const()[name = tensor("op_8195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8195_cast_fp16 = einsum(equation = var_8195_equation_0, values = (var_7641_cast_fp16, var_8041_cast_fp16))[name = tensor("op_8195_cast_fp16")]; + tensor var_8197_equation_0 = const()[name = tensor("op_8197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8197_cast_fp16 = einsum(equation = var_8197_equation_0, values = (var_7641_cast_fp16, var_8042_cast_fp16))[name = tensor("op_8197_cast_fp16")]; + tensor var_8199_equation_0 = const()[name = tensor("op_8199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8199_cast_fp16 = einsum(equation = var_8199_equation_0, values = (var_7641_cast_fp16, var_8043_cast_fp16))[name = tensor("op_8199_cast_fp16")]; + tensor var_8201_equation_0 = const()[name = tensor("op_8201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8201_cast_fp16 = einsum(equation = var_8201_equation_0, values = (var_7645_cast_fp16, var_8044_cast_fp16))[name = tensor("op_8201_cast_fp16")]; + tensor var_8203_equation_0 = const()[name = tensor("op_8203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8203_cast_fp16 = einsum(equation = var_8203_equation_0, values = (var_7645_cast_fp16, var_8045_cast_fp16))[name = tensor("op_8203_cast_fp16")]; + tensor var_8205_equation_0 = const()[name = tensor("op_8205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8205_cast_fp16 = einsum(equation = var_8205_equation_0, values = (var_7645_cast_fp16, var_8046_cast_fp16))[name = tensor("op_8205_cast_fp16")]; + tensor var_8207_equation_0 = const()[name = tensor("op_8207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_8207_cast_fp16 = einsum(equation = var_8207_equation_0, values = (var_7645_cast_fp16, var_8047_cast_fp16))[name = tensor("op_8207_cast_fp16")]; + tensor var_8209_interleave_0 = const()[name = tensor("op_8209_interleave_0"), val = tensor(false)]; + tensor var_8209_cast_fp16 = concat(axis = var_6714, interleave = var_8209_interleave_0, values = (var_8049_cast_fp16, var_8051_cast_fp16, var_8053_cast_fp16, var_8055_cast_fp16))[name = tensor("op_8209_cast_fp16")]; + tensor var_8211_interleave_0 = const()[name = tensor("op_8211_interleave_0"), val = tensor(false)]; + tensor var_8211_cast_fp16 = concat(axis = var_6714, interleave = var_8211_interleave_0, values = (var_8057_cast_fp16, var_8059_cast_fp16, var_8061_cast_fp16, var_8063_cast_fp16))[name = tensor("op_8211_cast_fp16")]; + tensor var_8213_interleave_0 = const()[name = tensor("op_8213_interleave_0"), val = tensor(false)]; + tensor var_8213_cast_fp16 = concat(axis = var_6714, interleave = var_8213_interleave_0, values = (var_8065_cast_fp16, var_8067_cast_fp16, var_8069_cast_fp16, var_8071_cast_fp16))[name = tensor("op_8213_cast_fp16")]; + tensor var_8215_interleave_0 = const()[name = tensor("op_8215_interleave_0"), val = tensor(false)]; + tensor var_8215_cast_fp16 = concat(axis = var_6714, interleave = var_8215_interleave_0, values = (var_8073_cast_fp16, var_8075_cast_fp16, var_8077_cast_fp16, var_8079_cast_fp16))[name = tensor("op_8215_cast_fp16")]; + tensor var_8217_interleave_0 = const()[name = tensor("op_8217_interleave_0"), val = tensor(false)]; + tensor var_8217_cast_fp16 = concat(axis = var_6714, interleave = var_8217_interleave_0, values = (var_8081_cast_fp16, var_8083_cast_fp16, var_8085_cast_fp16, var_8087_cast_fp16))[name = tensor("op_8217_cast_fp16")]; + tensor var_8219_interleave_0 = const()[name = tensor("op_8219_interleave_0"), val = tensor(false)]; + tensor var_8219_cast_fp16 = concat(axis = var_6714, interleave = var_8219_interleave_0, values = (var_8089_cast_fp16, var_8091_cast_fp16, var_8093_cast_fp16, var_8095_cast_fp16))[name = tensor("op_8219_cast_fp16")]; + tensor var_8221_interleave_0 = const()[name = tensor("op_8221_interleave_0"), val = tensor(false)]; + tensor var_8221_cast_fp16 = concat(axis = var_6714, interleave = var_8221_interleave_0, values = (var_8097_cast_fp16, var_8099_cast_fp16, var_8101_cast_fp16, var_8103_cast_fp16))[name = tensor("op_8221_cast_fp16")]; + tensor var_8223_interleave_0 = const()[name = tensor("op_8223_interleave_0"), val = tensor(false)]; + tensor var_8223_cast_fp16 = concat(axis = var_6714, interleave = var_8223_interleave_0, values = (var_8105_cast_fp16, var_8107_cast_fp16, var_8109_cast_fp16, var_8111_cast_fp16))[name = tensor("op_8223_cast_fp16")]; + tensor var_8225_interleave_0 = const()[name = tensor("op_8225_interleave_0"), val = tensor(false)]; + tensor var_8225_cast_fp16 = concat(axis = var_6714, interleave = var_8225_interleave_0, values = (var_8113_cast_fp16, var_8115_cast_fp16, var_8117_cast_fp16, var_8119_cast_fp16))[name = tensor("op_8225_cast_fp16")]; + tensor var_8227_interleave_0 = const()[name = tensor("op_8227_interleave_0"), val = tensor(false)]; + tensor var_8227_cast_fp16 = concat(axis = var_6714, interleave = var_8227_interleave_0, values = (var_8121_cast_fp16, var_8123_cast_fp16, var_8125_cast_fp16, var_8127_cast_fp16))[name = tensor("op_8227_cast_fp16")]; + tensor var_8229_interleave_0 = const()[name = tensor("op_8229_interleave_0"), val = tensor(false)]; + tensor var_8229_cast_fp16 = concat(axis = var_6714, interleave = var_8229_interleave_0, values = (var_8129_cast_fp16, var_8131_cast_fp16, var_8133_cast_fp16, var_8135_cast_fp16))[name = tensor("op_8229_cast_fp16")]; + tensor var_8231_interleave_0 = const()[name = tensor("op_8231_interleave_0"), val = tensor(false)]; + tensor var_8231_cast_fp16 = concat(axis = var_6714, interleave = var_8231_interleave_0, values = (var_8137_cast_fp16, var_8139_cast_fp16, var_8141_cast_fp16, var_8143_cast_fp16))[name = tensor("op_8231_cast_fp16")]; + tensor var_8233_interleave_0 = const()[name = tensor("op_8233_interleave_0"), val = tensor(false)]; + tensor var_8233_cast_fp16 = concat(axis = var_6714, interleave = var_8233_interleave_0, values = (var_8145_cast_fp16, var_8147_cast_fp16, var_8149_cast_fp16, var_8151_cast_fp16))[name = tensor("op_8233_cast_fp16")]; + tensor var_8235_interleave_0 = const()[name = tensor("op_8235_interleave_0"), val = tensor(false)]; + tensor var_8235_cast_fp16 = concat(axis = var_6714, interleave = var_8235_interleave_0, values = (var_8153_cast_fp16, var_8155_cast_fp16, var_8157_cast_fp16, var_8159_cast_fp16))[name = tensor("op_8235_cast_fp16")]; + tensor var_8237_interleave_0 = const()[name = tensor("op_8237_interleave_0"), val = tensor(false)]; + tensor var_8237_cast_fp16 = concat(axis = var_6714, interleave = var_8237_interleave_0, values = (var_8161_cast_fp16, var_8163_cast_fp16, var_8165_cast_fp16, var_8167_cast_fp16))[name = tensor("op_8237_cast_fp16")]; + tensor var_8239_interleave_0 = const()[name = tensor("op_8239_interleave_0"), val = tensor(false)]; + tensor var_8239_cast_fp16 = concat(axis = var_6714, interleave = var_8239_interleave_0, values = (var_8169_cast_fp16, var_8171_cast_fp16, var_8173_cast_fp16, var_8175_cast_fp16))[name = tensor("op_8239_cast_fp16")]; + tensor var_8241_interleave_0 = const()[name = tensor("op_8241_interleave_0"), val = tensor(false)]; + tensor var_8241_cast_fp16 = concat(axis = var_6714, interleave = var_8241_interleave_0, values = (var_8177_cast_fp16, var_8179_cast_fp16, var_8181_cast_fp16, var_8183_cast_fp16))[name = tensor("op_8241_cast_fp16")]; + tensor var_8243_interleave_0 = const()[name = tensor("op_8243_interleave_0"), val = tensor(false)]; + tensor var_8243_cast_fp16 = concat(axis = var_6714, interleave = var_8243_interleave_0, values = (var_8185_cast_fp16, var_8187_cast_fp16, var_8189_cast_fp16, var_8191_cast_fp16))[name = tensor("op_8243_cast_fp16")]; + tensor var_8245_interleave_0 = const()[name = tensor("op_8245_interleave_0"), val = tensor(false)]; + tensor var_8245_cast_fp16 = concat(axis = var_6714, interleave = var_8245_interleave_0, values = (var_8193_cast_fp16, var_8195_cast_fp16, var_8197_cast_fp16, var_8199_cast_fp16))[name = tensor("op_8245_cast_fp16")]; + tensor var_8247_interleave_0 = const()[name = tensor("op_8247_interleave_0"), val = tensor(false)]; + tensor var_8247_cast_fp16 = concat(axis = var_6714, interleave = var_8247_interleave_0, values = (var_8201_cast_fp16, var_8203_cast_fp16, var_8205_cast_fp16, var_8207_cast_fp16))[name = tensor("op_8247_cast_fp16")]; + tensor input_87_interleave_0 = const()[name = tensor("input_87_interleave_0"), val = tensor(false)]; + tensor input_87_cast_fp16 = concat(axis = var_6739, interleave = input_87_interleave_0, values = (var_8209_cast_fp16, var_8211_cast_fp16, var_8213_cast_fp16, var_8215_cast_fp16, var_8217_cast_fp16, var_8219_cast_fp16, var_8221_cast_fp16, var_8223_cast_fp16, var_8225_cast_fp16, var_8227_cast_fp16, var_8229_cast_fp16, var_8231_cast_fp16, var_8233_cast_fp16, var_8235_cast_fp16, var_8237_cast_fp16, var_8239_cast_fp16, var_8241_cast_fp16, var_8243_cast_fp16, var_8245_cast_fp16, var_8247_cast_fp16))[name = tensor("input_87_cast_fp16")]; + tensor var_8255 = const()[name = tensor("op_8255"), val = tensor([1, 1])]; + tensor var_8257 = const()[name = tensor("op_8257"), val = tensor([1, 1])]; + tensor pretrained_out_55_pad_type_0 = const()[name = tensor("pretrained_out_55_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_55_pad_0 = const()[name = tensor("pretrained_out_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59424256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60243520))), name = tensor("layers_4_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_4_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60243648)))]; + tensor pretrained_out_55_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_8257, groups = var_6739, pad = pretrained_out_55_pad_0, pad_type = pretrained_out_55_pad_type_0, strides = var_8255, weight = layers_4_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("pretrained_out_55_cast_fp16")]; + tensor var_8261 = const()[name = tensor("op_8261"), val = tensor([1, 1])]; + tensor var_8263 = const()[name = tensor("op_8263"), val = tensor([1, 1])]; + tensor input_89_pad_type_0 = const()[name = tensor("input_89_pad_type_0"), val = tensor("custom")]; + tensor input_89_pad_0 = const()[name = tensor("input_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60246272)))]; + tensor input_89_cast_fp16 = conv(dilations = var_8263, groups = var_6739, pad = input_89_pad_0, pad_type = input_89_pad_type_0, strides = var_8261, weight = layers_4_self_attn_o_proj_loraA_weight_to_fp16, x = input_87_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor var_8267 = const()[name = tensor("op_8267"), val = tensor([1, 1])]; + tensor var_8269 = const()[name = tensor("op_8269"), val = tensor([1, 1])]; + tensor lora_out_109_pad_type_0 = const()[name = tensor("lora_out_109_pad_type_0"), val = tensor("custom")]; + tensor lora_out_109_pad_0 = const()[name = tensor("lora_out_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_111_weight_0_to_fp16 = const()[name = tensor("lora_out_111_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60287296)))]; + tensor lora_out_111_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_8269, groups = var_6739, pad = lora_out_109_pad_0, pad_type = lora_out_109_pad_type_0, strides = var_8267, weight = lora_out_111_weight_0_to_fp16, x = input_89_cast_fp16)[name = tensor("lora_out_111_cast_fp16")]; + tensor obj_19_cast_fp16 = add(x = pretrained_out_55_cast_fp16, y = lora_out_111_cast_fp16)[name = tensor("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor var_8278 = const()[name = tensor("op_8278"), val = tensor([1])]; + tensor channels_mean_19_cast_fp16 = reduce_mean(axes = var_8278, keep_dims = var_6740, x = inputs_19_cast_fp16)[name = tensor("channels_mean_19_cast_fp16")]; + tensor zero_mean_19_cast_fp16 = sub(x = inputs_19_cast_fp16, y = channels_mean_19_cast_fp16)[name = tensor("zero_mean_19_cast_fp16")]; + tensor zero_mean_sq_19_cast_fp16 = mul(x = zero_mean_19_cast_fp16, y = zero_mean_19_cast_fp16)[name = tensor("zero_mean_sq_19_cast_fp16")]; + tensor var_8282 = const()[name = tensor("op_8282"), val = tensor([1])]; + tensor var_8283_cast_fp16 = reduce_mean(axes = var_8282, keep_dims = var_6740, x = zero_mean_sq_19_cast_fp16)[name = tensor("op_8283_cast_fp16")]; + tensor var_8284_to_fp16 = const()[name = tensor("op_8284_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_8285_cast_fp16 = add(x = var_8283_cast_fp16, y = var_8284_to_fp16)[name = tensor("op_8285_cast_fp16")]; + tensor denom_19_epsilon_0 = const()[name = tensor("denom_19_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_19_cast_fp16 = rsqrt(epsilon = denom_19_epsilon_0, x = var_8285_cast_fp16)[name = tensor("denom_19_cast_fp16")]; + tensor out_19_cast_fp16 = mul(x = zero_mean_19_cast_fp16, y = denom_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; + tensor input_91_gamma_0_to_fp16 = const()[name = tensor("input_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60328320)))]; + tensor input_91_beta_0_to_fp16 = const()[name = tensor("input_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60330944)))]; + tensor input_91_epsilon_0_to_fp16 = const()[name = tensor("input_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor var_8299 = const()[name = tensor("op_8299"), val = tensor([1, 1])]; + tensor var_8301 = const()[name = tensor("op_8301"), val = tensor([1, 1])]; + tensor pretrained_out_57_pad_type_0 = const()[name = tensor("pretrained_out_57_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_57_pad_0 = const()[name = tensor("pretrained_out_57_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60333568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63610432))), name = tensor("layers_4_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_4_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_4_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63610560)))]; + tensor pretrained_out_57_cast_fp16 = conv(bias = layers_4_fc1_pretrained_bias_to_fp16, dilations = var_8301, groups = var_6739, pad = pretrained_out_57_pad_0, pad_type = pretrained_out_57_pad_type_0, strides = var_8299, weight = layers_4_fc1_pretrained_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor("pretrained_out_57_cast_fp16")]; + tensor var_8305 = const()[name = tensor("op_8305"), val = tensor([1, 1])]; + tensor var_8307 = const()[name = tensor("op_8307"), val = tensor([1, 1])]; + tensor input_93_pad_type_0 = const()[name = tensor("input_93_pad_type_0"), val = tensor("custom")]; + tensor input_93_pad_0 = const()[name = tensor("input_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_4_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63620864)))]; + tensor input_93_cast_fp16 = conv(dilations = var_8307, groups = var_6739, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = var_8305, weight = layers_4_fc1_loraA_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor var_8311 = const()[name = tensor("op_8311"), val = tensor([1, 1])]; + tensor var_8313 = const()[name = tensor("op_8313"), val = tensor([1, 1])]; + tensor lora_out_113_pad_type_0 = const()[name = tensor("lora_out_113_pad_type_0"), val = tensor("custom")]; + tensor lora_out_113_pad_0 = const()[name = tensor("lora_out_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_115_weight_0_to_fp16 = const()[name = tensor("lora_out_115_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63661888)))]; + tensor lora_out_115_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_8313, groups = var_6739, pad = lora_out_113_pad_0, pad_type = lora_out_113_pad_type_0, strides = var_8311, weight = lora_out_115_weight_0_to_fp16, x = input_93_cast_fp16)[name = tensor("lora_out_115_cast_fp16")]; + tensor input_95_cast_fp16 = add(x = pretrained_out_57_cast_fp16, y = lora_out_115_cast_fp16)[name = tensor("input_95_cast_fp16")]; + tensor input_97_mode_0 = const()[name = tensor("input_97_mode_0"), val = tensor("EXACT")]; + tensor input_97_cast_fp16 = gelu(mode = input_97_mode_0, x = input_95_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor var_8325 = const()[name = tensor("op_8325"), val = tensor([1, 1])]; + tensor var_8327 = const()[name = tensor("op_8327"), val = tensor([1, 1])]; + tensor pretrained_out_59_pad_type_0 = const()[name = tensor("pretrained_out_59_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_59_pad_0 = const()[name = tensor("pretrained_out_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63825792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67102656))), name = tensor("layers_4_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_4_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_4_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67102784)))]; + tensor pretrained_out_59_cast_fp16 = conv(bias = layers_4_fc2_pretrained_bias_to_fp16, dilations = var_8327, groups = var_6739, pad = pretrained_out_59_pad_0, pad_type = pretrained_out_59_pad_type_0, strides = var_8325, weight = layers_4_fc2_pretrained_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor("pretrained_out_59_cast_fp16")]; + tensor var_8331 = const()[name = tensor("op_8331"), val = tensor([1, 1])]; + tensor var_8333 = const()[name = tensor("op_8333"), val = tensor([1, 1])]; + tensor input_99_pad_type_0 = const()[name = tensor("input_99_pad_type_0"), val = tensor("custom")]; + tensor input_99_pad_0 = const()[name = tensor("input_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_4_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_4_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67105408)))]; + tensor input_99_cast_fp16 = conv(dilations = var_8333, groups = var_6739, pad = input_99_pad_0, pad_type = input_99_pad_type_0, strides = var_8331, weight = layers_4_fc2_loraA_weight_to_fp16, x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor var_8337 = const()[name = tensor("op_8337"), val = tensor([1, 1])]; + tensor var_8339 = const()[name = tensor("op_8339"), val = tensor([1, 1])]; + tensor lora_out_117_pad_type_0 = const()[name = tensor("lora_out_117_pad_type_0"), val = tensor("custom")]; + tensor lora_out_117_pad_0 = const()[name = tensor("lora_out_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_119_weight_0_to_fp16 = const()[name = tensor("lora_out_119_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67269312)))]; + tensor lora_out_119_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_8339, groups = var_6739, pad = lora_out_117_pad_0, pad_type = lora_out_117_pad_type_0, strides = var_8337, weight = lora_out_119_weight_0_to_fp16, x = input_99_cast_fp16)[name = tensor("lora_out_119_cast_fp16")]; + tensor hidden_states_13_cast_fp16 = add(x = pretrained_out_59_cast_fp16, y = lora_out_119_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor var_8349 = const()[name = tensor("op_8349"), val = tensor(3)]; + tensor var_8374 = const()[name = tensor("op_8374"), val = tensor(1)]; + tensor var_8375 = const()[name = tensor("op_8375"), val = tensor(true)]; + tensor var_8385 = const()[name = tensor("op_8385"), val = tensor([1])]; + tensor channels_mean_21_cast_fp16 = reduce_mean(axes = var_8385, keep_dims = var_8375, x = inputs_21_cast_fp16)[name = tensor("channels_mean_21_cast_fp16")]; + tensor zero_mean_21_cast_fp16 = sub(x = inputs_21_cast_fp16, y = channels_mean_21_cast_fp16)[name = tensor("zero_mean_21_cast_fp16")]; + tensor zero_mean_sq_21_cast_fp16 = mul(x = zero_mean_21_cast_fp16, y = zero_mean_21_cast_fp16)[name = tensor("zero_mean_sq_21_cast_fp16")]; + tensor var_8389 = const()[name = tensor("op_8389"), val = tensor([1])]; + tensor var_8390_cast_fp16 = reduce_mean(axes = var_8389, keep_dims = var_8375, x = zero_mean_sq_21_cast_fp16)[name = tensor("op_8390_cast_fp16")]; + tensor var_8391_to_fp16 = const()[name = tensor("op_8391_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_8392_cast_fp16 = add(x = var_8390_cast_fp16, y = var_8391_to_fp16)[name = tensor("op_8392_cast_fp16")]; + tensor denom_21_epsilon_0 = const()[name = tensor("denom_21_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_21_cast_fp16 = rsqrt(epsilon = denom_21_epsilon_0, x = var_8392_cast_fp16)[name = tensor("denom_21_cast_fp16")]; + tensor out_21_cast_fp16 = mul(x = zero_mean_21_cast_fp16, y = denom_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = tensor("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67310336)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = tensor("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67312960)))]; + tensor obj_21_epsilon_0_to_fp16 = const()[name = tensor("obj_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor("obj_21_cast_fp16")]; + tensor var_8410 = const()[name = tensor("op_8410"), val = tensor([1, 1])]; + tensor var_8412 = const()[name = tensor("op_8412"), val = tensor([1, 1])]; + tensor pretrained_out_61_pad_type_0 = const()[name = tensor("pretrained_out_61_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_61_pad_0 = const()[name = tensor("pretrained_out_61_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67315584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68134848))), name = tensor("layers_5_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_5_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68134976)))]; + tensor pretrained_out_61_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_8412, groups = var_8374, pad = pretrained_out_61_pad_0, pad_type = pretrained_out_61_pad_type_0, strides = var_8410, weight = layers_5_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor("pretrained_out_61_cast_fp16")]; + tensor var_8416 = const()[name = tensor("op_8416"), val = tensor([1, 1])]; + tensor var_8418 = const()[name = tensor("op_8418"), val = tensor([1, 1])]; + tensor input_101_pad_type_0 = const()[name = tensor("input_101_pad_type_0"), val = tensor("custom")]; + tensor input_101_pad_0 = const()[name = tensor("input_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68137600)))]; + tensor input_101_cast_fp16 = conv(dilations = var_8418, groups = var_8374, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = var_8416, weight = layers_5_self_attn_q_proj_loraA_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor var_8422 = const()[name = tensor("op_8422"), val = tensor([1, 1])]; + tensor var_8424 = const()[name = tensor("op_8424"), val = tensor([1, 1])]; + tensor lora_out_121_pad_type_0 = const()[name = tensor("lora_out_121_pad_type_0"), val = tensor("custom")]; + tensor lora_out_121_pad_0 = const()[name = tensor("lora_out_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_123_weight_0_to_fp16 = const()[name = tensor("lora_out_123_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68178624)))]; + tensor lora_out_123_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_8424, groups = var_8374, pad = lora_out_121_pad_0, pad_type = lora_out_121_pad_type_0, strides = var_8422, weight = lora_out_123_weight_0_to_fp16, x = input_101_cast_fp16)[name = tensor("lora_out_123_cast_fp16")]; + tensor query_11_cast_fp16 = add(x = pretrained_out_61_cast_fp16, y = lora_out_123_cast_fp16)[name = tensor("query_11_cast_fp16")]; + tensor var_8434 = const()[name = tensor("op_8434"), val = tensor([1, 1])]; + tensor var_8436 = const()[name = tensor("op_8436"), val = tensor([1, 1])]; + tensor pretrained_out_63_pad_type_0 = const()[name = tensor("pretrained_out_63_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_63_pad_0 = const()[name = tensor("pretrained_out_63_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68219648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69038912))), name = tensor("layers_5_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_63_cast_fp16 = conv(dilations = var_8436, groups = var_8374, pad = pretrained_out_63_pad_0, pad_type = pretrained_out_63_pad_type_0, strides = var_8434, weight = layers_5_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor("pretrained_out_63_cast_fp16")]; + tensor var_8440 = const()[name = tensor("op_8440"), val = tensor([1, 1])]; + tensor var_8442 = const()[name = tensor("op_8442"), val = tensor([1, 1])]; + tensor input_103_pad_type_0 = const()[name = tensor("input_103_pad_type_0"), val = tensor("custom")]; + tensor input_103_pad_0 = const()[name = tensor("input_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69039040)))]; + tensor input_103_cast_fp16 = conv(dilations = var_8442, groups = var_8374, pad = input_103_pad_0, pad_type = input_103_pad_type_0, strides = var_8440, weight = layers_5_self_attn_k_proj_loraA_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor var_8446 = const()[name = tensor("op_8446"), val = tensor([1, 1])]; + tensor var_8448 = const()[name = tensor("op_8448"), val = tensor([1, 1])]; + tensor lora_out_125_pad_type_0 = const()[name = tensor("lora_out_125_pad_type_0"), val = tensor("custom")]; + tensor lora_out_125_pad_0 = const()[name = tensor("lora_out_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_127_weight_0_to_fp16 = const()[name = tensor("lora_out_127_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69080064)))]; + tensor lora_out_127_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_8448, groups = var_8374, pad = lora_out_125_pad_0, pad_type = lora_out_125_pad_type_0, strides = var_8446, weight = lora_out_127_weight_0_to_fp16, x = input_103_cast_fp16)[name = tensor("lora_out_127_cast_fp16")]; + tensor key_11_cast_fp16 = add(x = pretrained_out_63_cast_fp16, y = lora_out_127_cast_fp16)[name = tensor("key_11_cast_fp16")]; + tensor var_8459 = const()[name = tensor("op_8459"), val = tensor([1, 1])]; + tensor var_8461 = const()[name = tensor("op_8461"), val = tensor([1, 1])]; + tensor pretrained_out_65_pad_type_0 = const()[name = tensor("pretrained_out_65_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_65_pad_0 = const()[name = tensor("pretrained_out_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69121088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69940352))), name = tensor("layers_5_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_5_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69940480)))]; + tensor pretrained_out_65_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_8461, groups = var_8374, pad = pretrained_out_65_pad_0, pad_type = pretrained_out_65_pad_type_0, strides = var_8459, weight = layers_5_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor("pretrained_out_65_cast_fp16")]; + tensor var_8465 = const()[name = tensor("op_8465"), val = tensor([1, 1])]; + tensor var_8467 = const()[name = tensor("op_8467"), val = tensor([1, 1])]; + tensor input_105_pad_type_0 = const()[name = tensor("input_105_pad_type_0"), val = tensor("custom")]; + tensor input_105_pad_0 = const()[name = tensor("input_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69943104)))]; + tensor input_105_cast_fp16 = conv(dilations = var_8467, groups = var_8374, pad = input_105_pad_0, pad_type = input_105_pad_type_0, strides = var_8465, weight = layers_5_self_attn_v_proj_loraA_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor var_8471 = const()[name = tensor("op_8471"), val = tensor([1, 1])]; + tensor var_8473 = const()[name = tensor("op_8473"), val = tensor([1, 1])]; + tensor lora_out_129_pad_type_0 = const()[name = tensor("lora_out_129_pad_type_0"), val = tensor("custom")]; + tensor lora_out_129_pad_0 = const()[name = tensor("lora_out_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_131_weight_0_to_fp16 = const()[name = tensor("lora_out_131_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69984128)))]; + tensor lora_out_131_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_8473, groups = var_8374, pad = lora_out_129_pad_0, pad_type = lora_out_129_pad_type_0, strides = var_8471, weight = lora_out_131_weight_0_to_fp16, x = input_105_cast_fp16)[name = tensor("lora_out_131_cast_fp16")]; + tensor value_11_cast_fp16 = add(x = pretrained_out_65_cast_fp16, y = lora_out_131_cast_fp16)[name = tensor("value_11_cast_fp16")]; + tensor var_8483_begin_0 = const()[name = tensor("op_8483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8483_end_0 = const()[name = tensor("op_8483_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8483_end_mask_0 = const()[name = tensor("op_8483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8483_cast_fp16 = slice_by_index(begin = var_8483_begin_0, end = var_8483_end_0, end_mask = var_8483_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8483_cast_fp16")]; + tensor var_8487_begin_0 = const()[name = tensor("op_8487_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_8487_end_0 = const()[name = tensor("op_8487_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_8487_end_mask_0 = const()[name = tensor("op_8487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8487_cast_fp16 = slice_by_index(begin = var_8487_begin_0, end = var_8487_end_0, end_mask = var_8487_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8487_cast_fp16")]; + tensor var_8491_begin_0 = const()[name = tensor("op_8491_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_8491_end_0 = const()[name = tensor("op_8491_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_8491_end_mask_0 = const()[name = tensor("op_8491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8491_cast_fp16 = slice_by_index(begin = var_8491_begin_0, end = var_8491_end_0, end_mask = var_8491_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8491_cast_fp16")]; + tensor var_8495_begin_0 = const()[name = tensor("op_8495_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_8495_end_0 = const()[name = tensor("op_8495_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_8495_end_mask_0 = const()[name = tensor("op_8495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8495_cast_fp16 = slice_by_index(begin = var_8495_begin_0, end = var_8495_end_0, end_mask = var_8495_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8495_cast_fp16")]; + tensor var_8499_begin_0 = const()[name = tensor("op_8499_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_8499_end_0 = const()[name = tensor("op_8499_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_8499_end_mask_0 = const()[name = tensor("op_8499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8499_cast_fp16 = slice_by_index(begin = var_8499_begin_0, end = var_8499_end_0, end_mask = var_8499_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8499_cast_fp16")]; + tensor var_8503_begin_0 = const()[name = tensor("op_8503_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_8503_end_0 = const()[name = tensor("op_8503_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_8503_end_mask_0 = const()[name = tensor("op_8503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8503_cast_fp16 = slice_by_index(begin = var_8503_begin_0, end = var_8503_end_0, end_mask = var_8503_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8503_cast_fp16")]; + tensor var_8507_begin_0 = const()[name = tensor("op_8507_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_8507_end_0 = const()[name = tensor("op_8507_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_8507_end_mask_0 = const()[name = tensor("op_8507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8507_cast_fp16 = slice_by_index(begin = var_8507_begin_0, end = var_8507_end_0, end_mask = var_8507_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8507_cast_fp16")]; + tensor var_8511_begin_0 = const()[name = tensor("op_8511_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_8511_end_0 = const()[name = tensor("op_8511_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_8511_end_mask_0 = const()[name = tensor("op_8511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8511_cast_fp16 = slice_by_index(begin = var_8511_begin_0, end = var_8511_end_0, end_mask = var_8511_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8511_cast_fp16")]; + tensor var_8515_begin_0 = const()[name = tensor("op_8515_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_8515_end_0 = const()[name = tensor("op_8515_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_8515_end_mask_0 = const()[name = tensor("op_8515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8515_cast_fp16 = slice_by_index(begin = var_8515_begin_0, end = var_8515_end_0, end_mask = var_8515_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8515_cast_fp16")]; + tensor var_8519_begin_0 = const()[name = tensor("op_8519_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_8519_end_0 = const()[name = tensor("op_8519_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_8519_end_mask_0 = const()[name = tensor("op_8519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8519_cast_fp16 = slice_by_index(begin = var_8519_begin_0, end = var_8519_end_0, end_mask = var_8519_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8519_cast_fp16")]; + tensor var_8523_begin_0 = const()[name = tensor("op_8523_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_8523_end_0 = const()[name = tensor("op_8523_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_8523_end_mask_0 = const()[name = tensor("op_8523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8523_cast_fp16 = slice_by_index(begin = var_8523_begin_0, end = var_8523_end_0, end_mask = var_8523_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8523_cast_fp16")]; + tensor var_8527_begin_0 = const()[name = tensor("op_8527_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_8527_end_0 = const()[name = tensor("op_8527_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_8527_end_mask_0 = const()[name = tensor("op_8527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8527_cast_fp16 = slice_by_index(begin = var_8527_begin_0, end = var_8527_end_0, end_mask = var_8527_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8527_cast_fp16")]; + tensor var_8531_begin_0 = const()[name = tensor("op_8531_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_8531_end_0 = const()[name = tensor("op_8531_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_8531_end_mask_0 = const()[name = tensor("op_8531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8531_cast_fp16 = slice_by_index(begin = var_8531_begin_0, end = var_8531_end_0, end_mask = var_8531_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8531_cast_fp16")]; + tensor var_8535_begin_0 = const()[name = tensor("op_8535_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_8535_end_0 = const()[name = tensor("op_8535_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_8535_end_mask_0 = const()[name = tensor("op_8535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8535_cast_fp16 = slice_by_index(begin = var_8535_begin_0, end = var_8535_end_0, end_mask = var_8535_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8535_cast_fp16")]; + tensor var_8539_begin_0 = const()[name = tensor("op_8539_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_8539_end_0 = const()[name = tensor("op_8539_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_8539_end_mask_0 = const()[name = tensor("op_8539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8539_cast_fp16 = slice_by_index(begin = var_8539_begin_0, end = var_8539_end_0, end_mask = var_8539_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8539_cast_fp16")]; + tensor var_8543_begin_0 = const()[name = tensor("op_8543_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_8543_end_0 = const()[name = tensor("op_8543_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_8543_end_mask_0 = const()[name = tensor("op_8543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8543_cast_fp16 = slice_by_index(begin = var_8543_begin_0, end = var_8543_end_0, end_mask = var_8543_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8543_cast_fp16")]; + tensor var_8547_begin_0 = const()[name = tensor("op_8547_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_8547_end_0 = const()[name = tensor("op_8547_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_8547_end_mask_0 = const()[name = tensor("op_8547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8547_cast_fp16 = slice_by_index(begin = var_8547_begin_0, end = var_8547_end_0, end_mask = var_8547_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8547_cast_fp16")]; + tensor var_8551_begin_0 = const()[name = tensor("op_8551_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_8551_end_0 = const()[name = tensor("op_8551_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_8551_end_mask_0 = const()[name = tensor("op_8551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8551_cast_fp16 = slice_by_index(begin = var_8551_begin_0, end = var_8551_end_0, end_mask = var_8551_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8551_cast_fp16")]; + tensor var_8555_begin_0 = const()[name = tensor("op_8555_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_8555_end_0 = const()[name = tensor("op_8555_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_8555_end_mask_0 = const()[name = tensor("op_8555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8555_cast_fp16 = slice_by_index(begin = var_8555_begin_0, end = var_8555_end_0, end_mask = var_8555_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8555_cast_fp16")]; + tensor var_8559_begin_0 = const()[name = tensor("op_8559_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_8559_end_0 = const()[name = tensor("op_8559_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_8559_end_mask_0 = const()[name = tensor("op_8559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_8559_cast_fp16 = slice_by_index(begin = var_8559_begin_0, end = var_8559_end_0, end_mask = var_8559_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_8559_cast_fp16")]; + tensor var_8568_begin_0 = const()[name = tensor("op_8568_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8568_end_0 = const()[name = tensor("op_8568_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8568_end_mask_0 = const()[name = tensor("op_8568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8568_cast_fp16 = slice_by_index(begin = var_8568_begin_0, end = var_8568_end_0, end_mask = var_8568_end_mask_0, x = var_8483_cast_fp16)[name = tensor("op_8568_cast_fp16")]; + tensor var_8575_begin_0 = const()[name = tensor("op_8575_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8575_end_0 = const()[name = tensor("op_8575_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8575_end_mask_0 = const()[name = tensor("op_8575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8575_cast_fp16 = slice_by_index(begin = var_8575_begin_0, end = var_8575_end_0, end_mask = var_8575_end_mask_0, x = var_8483_cast_fp16)[name = tensor("op_8575_cast_fp16")]; + tensor var_8582_begin_0 = const()[name = tensor("op_8582_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8582_end_0 = const()[name = tensor("op_8582_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8582_end_mask_0 = const()[name = tensor("op_8582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8582_cast_fp16 = slice_by_index(begin = var_8582_begin_0, end = var_8582_end_0, end_mask = var_8582_end_mask_0, x = var_8483_cast_fp16)[name = tensor("op_8582_cast_fp16")]; + tensor var_8589_begin_0 = const()[name = tensor("op_8589_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8589_end_0 = const()[name = tensor("op_8589_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8589_end_mask_0 = const()[name = tensor("op_8589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8589_cast_fp16 = slice_by_index(begin = var_8589_begin_0, end = var_8589_end_0, end_mask = var_8589_end_mask_0, x = var_8483_cast_fp16)[name = tensor("op_8589_cast_fp16")]; + tensor var_8596_begin_0 = const()[name = tensor("op_8596_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8596_end_0 = const()[name = tensor("op_8596_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8596_end_mask_0 = const()[name = tensor("op_8596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8596_cast_fp16 = slice_by_index(begin = var_8596_begin_0, end = var_8596_end_0, end_mask = var_8596_end_mask_0, x = var_8487_cast_fp16)[name = tensor("op_8596_cast_fp16")]; + tensor var_8603_begin_0 = const()[name = tensor("op_8603_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8603_end_0 = const()[name = tensor("op_8603_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8603_end_mask_0 = const()[name = tensor("op_8603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8603_cast_fp16 = slice_by_index(begin = var_8603_begin_0, end = var_8603_end_0, end_mask = var_8603_end_mask_0, x = var_8487_cast_fp16)[name = tensor("op_8603_cast_fp16")]; + tensor var_8610_begin_0 = const()[name = tensor("op_8610_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8610_end_0 = const()[name = tensor("op_8610_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8610_end_mask_0 = const()[name = tensor("op_8610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8610_cast_fp16 = slice_by_index(begin = var_8610_begin_0, end = var_8610_end_0, end_mask = var_8610_end_mask_0, x = var_8487_cast_fp16)[name = tensor("op_8610_cast_fp16")]; + tensor var_8617_begin_0 = const()[name = tensor("op_8617_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8617_end_0 = const()[name = tensor("op_8617_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8617_end_mask_0 = const()[name = tensor("op_8617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8617_cast_fp16 = slice_by_index(begin = var_8617_begin_0, end = var_8617_end_0, end_mask = var_8617_end_mask_0, x = var_8487_cast_fp16)[name = tensor("op_8617_cast_fp16")]; + tensor var_8624_begin_0 = const()[name = tensor("op_8624_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8624_end_0 = const()[name = tensor("op_8624_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8624_end_mask_0 = const()[name = tensor("op_8624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8624_cast_fp16 = slice_by_index(begin = var_8624_begin_0, end = var_8624_end_0, end_mask = var_8624_end_mask_0, x = var_8491_cast_fp16)[name = tensor("op_8624_cast_fp16")]; + tensor var_8631_begin_0 = const()[name = tensor("op_8631_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8631_end_0 = const()[name = tensor("op_8631_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8631_end_mask_0 = const()[name = tensor("op_8631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8631_cast_fp16 = slice_by_index(begin = var_8631_begin_0, end = var_8631_end_0, end_mask = var_8631_end_mask_0, x = var_8491_cast_fp16)[name = tensor("op_8631_cast_fp16")]; + tensor var_8638_begin_0 = const()[name = tensor("op_8638_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8638_end_0 = const()[name = tensor("op_8638_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8638_end_mask_0 = const()[name = tensor("op_8638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8638_cast_fp16 = slice_by_index(begin = var_8638_begin_0, end = var_8638_end_0, end_mask = var_8638_end_mask_0, x = var_8491_cast_fp16)[name = tensor("op_8638_cast_fp16")]; + tensor var_8645_begin_0 = const()[name = tensor("op_8645_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8645_end_0 = const()[name = tensor("op_8645_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8645_end_mask_0 = const()[name = tensor("op_8645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8645_cast_fp16 = slice_by_index(begin = var_8645_begin_0, end = var_8645_end_0, end_mask = var_8645_end_mask_0, x = var_8491_cast_fp16)[name = tensor("op_8645_cast_fp16")]; + tensor var_8652_begin_0 = const()[name = tensor("op_8652_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8652_end_0 = const()[name = tensor("op_8652_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8652_end_mask_0 = const()[name = tensor("op_8652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8652_cast_fp16 = slice_by_index(begin = var_8652_begin_0, end = var_8652_end_0, end_mask = var_8652_end_mask_0, x = var_8495_cast_fp16)[name = tensor("op_8652_cast_fp16")]; + tensor var_8659_begin_0 = const()[name = tensor("op_8659_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8659_end_0 = const()[name = tensor("op_8659_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8659_end_mask_0 = const()[name = tensor("op_8659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8659_cast_fp16 = slice_by_index(begin = var_8659_begin_0, end = var_8659_end_0, end_mask = var_8659_end_mask_0, x = var_8495_cast_fp16)[name = tensor("op_8659_cast_fp16")]; + tensor var_8666_begin_0 = const()[name = tensor("op_8666_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8666_end_0 = const()[name = tensor("op_8666_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8666_end_mask_0 = const()[name = tensor("op_8666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8666_cast_fp16 = slice_by_index(begin = var_8666_begin_0, end = var_8666_end_0, end_mask = var_8666_end_mask_0, x = var_8495_cast_fp16)[name = tensor("op_8666_cast_fp16")]; + tensor var_8673_begin_0 = const()[name = tensor("op_8673_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8673_end_0 = const()[name = tensor("op_8673_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8673_end_mask_0 = const()[name = tensor("op_8673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8673_cast_fp16 = slice_by_index(begin = var_8673_begin_0, end = var_8673_end_0, end_mask = var_8673_end_mask_0, x = var_8495_cast_fp16)[name = tensor("op_8673_cast_fp16")]; + tensor var_8680_begin_0 = const()[name = tensor("op_8680_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8680_end_0 = const()[name = tensor("op_8680_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8680_end_mask_0 = const()[name = tensor("op_8680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8680_cast_fp16 = slice_by_index(begin = var_8680_begin_0, end = var_8680_end_0, end_mask = var_8680_end_mask_0, x = var_8499_cast_fp16)[name = tensor("op_8680_cast_fp16")]; + tensor var_8687_begin_0 = const()[name = tensor("op_8687_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8687_end_0 = const()[name = tensor("op_8687_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8687_end_mask_0 = const()[name = tensor("op_8687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8687_cast_fp16 = slice_by_index(begin = var_8687_begin_0, end = var_8687_end_0, end_mask = var_8687_end_mask_0, x = var_8499_cast_fp16)[name = tensor("op_8687_cast_fp16")]; + tensor var_8694_begin_0 = const()[name = tensor("op_8694_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8694_end_0 = const()[name = tensor("op_8694_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8694_end_mask_0 = const()[name = tensor("op_8694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8694_cast_fp16 = slice_by_index(begin = var_8694_begin_0, end = var_8694_end_0, end_mask = var_8694_end_mask_0, x = var_8499_cast_fp16)[name = tensor("op_8694_cast_fp16")]; + tensor var_8701_begin_0 = const()[name = tensor("op_8701_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8701_end_0 = const()[name = tensor("op_8701_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8701_end_mask_0 = const()[name = tensor("op_8701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8701_cast_fp16 = slice_by_index(begin = var_8701_begin_0, end = var_8701_end_0, end_mask = var_8701_end_mask_0, x = var_8499_cast_fp16)[name = tensor("op_8701_cast_fp16")]; + tensor var_8708_begin_0 = const()[name = tensor("op_8708_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8708_end_0 = const()[name = tensor("op_8708_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8708_end_mask_0 = const()[name = tensor("op_8708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8708_cast_fp16 = slice_by_index(begin = var_8708_begin_0, end = var_8708_end_0, end_mask = var_8708_end_mask_0, x = var_8503_cast_fp16)[name = tensor("op_8708_cast_fp16")]; + tensor var_8715_begin_0 = const()[name = tensor("op_8715_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8715_end_0 = const()[name = tensor("op_8715_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8715_end_mask_0 = const()[name = tensor("op_8715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8715_cast_fp16 = slice_by_index(begin = var_8715_begin_0, end = var_8715_end_0, end_mask = var_8715_end_mask_0, x = var_8503_cast_fp16)[name = tensor("op_8715_cast_fp16")]; + tensor var_8722_begin_0 = const()[name = tensor("op_8722_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8722_end_0 = const()[name = tensor("op_8722_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8722_end_mask_0 = const()[name = tensor("op_8722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8722_cast_fp16 = slice_by_index(begin = var_8722_begin_0, end = var_8722_end_0, end_mask = var_8722_end_mask_0, x = var_8503_cast_fp16)[name = tensor("op_8722_cast_fp16")]; + tensor var_8729_begin_0 = const()[name = tensor("op_8729_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8729_end_0 = const()[name = tensor("op_8729_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8729_end_mask_0 = const()[name = tensor("op_8729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8729_cast_fp16 = slice_by_index(begin = var_8729_begin_0, end = var_8729_end_0, end_mask = var_8729_end_mask_0, x = var_8503_cast_fp16)[name = tensor("op_8729_cast_fp16")]; + tensor var_8736_begin_0 = const()[name = tensor("op_8736_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8736_end_0 = const()[name = tensor("op_8736_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8736_end_mask_0 = const()[name = tensor("op_8736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8736_cast_fp16 = slice_by_index(begin = var_8736_begin_0, end = var_8736_end_0, end_mask = var_8736_end_mask_0, x = var_8507_cast_fp16)[name = tensor("op_8736_cast_fp16")]; + tensor var_8743_begin_0 = const()[name = tensor("op_8743_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8743_end_0 = const()[name = tensor("op_8743_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8743_end_mask_0 = const()[name = tensor("op_8743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8743_cast_fp16 = slice_by_index(begin = var_8743_begin_0, end = var_8743_end_0, end_mask = var_8743_end_mask_0, x = var_8507_cast_fp16)[name = tensor("op_8743_cast_fp16")]; + tensor var_8750_begin_0 = const()[name = tensor("op_8750_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8750_end_0 = const()[name = tensor("op_8750_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8750_end_mask_0 = const()[name = tensor("op_8750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8750_cast_fp16 = slice_by_index(begin = var_8750_begin_0, end = var_8750_end_0, end_mask = var_8750_end_mask_0, x = var_8507_cast_fp16)[name = tensor("op_8750_cast_fp16")]; + tensor var_8757_begin_0 = const()[name = tensor("op_8757_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8757_end_0 = const()[name = tensor("op_8757_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8757_end_mask_0 = const()[name = tensor("op_8757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8757_cast_fp16 = slice_by_index(begin = var_8757_begin_0, end = var_8757_end_0, end_mask = var_8757_end_mask_0, x = var_8507_cast_fp16)[name = tensor("op_8757_cast_fp16")]; + tensor var_8764_begin_0 = const()[name = tensor("op_8764_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8764_end_0 = const()[name = tensor("op_8764_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8764_end_mask_0 = const()[name = tensor("op_8764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8764_cast_fp16 = slice_by_index(begin = var_8764_begin_0, end = var_8764_end_0, end_mask = var_8764_end_mask_0, x = var_8511_cast_fp16)[name = tensor("op_8764_cast_fp16")]; + tensor var_8771_begin_0 = const()[name = tensor("op_8771_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8771_end_0 = const()[name = tensor("op_8771_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8771_end_mask_0 = const()[name = tensor("op_8771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8771_cast_fp16 = slice_by_index(begin = var_8771_begin_0, end = var_8771_end_0, end_mask = var_8771_end_mask_0, x = var_8511_cast_fp16)[name = tensor("op_8771_cast_fp16")]; + tensor var_8778_begin_0 = const()[name = tensor("op_8778_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8778_end_0 = const()[name = tensor("op_8778_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8778_end_mask_0 = const()[name = tensor("op_8778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8778_cast_fp16 = slice_by_index(begin = var_8778_begin_0, end = var_8778_end_0, end_mask = var_8778_end_mask_0, x = var_8511_cast_fp16)[name = tensor("op_8778_cast_fp16")]; + tensor var_8785_begin_0 = const()[name = tensor("op_8785_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8785_end_0 = const()[name = tensor("op_8785_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8785_end_mask_0 = const()[name = tensor("op_8785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8785_cast_fp16 = slice_by_index(begin = var_8785_begin_0, end = var_8785_end_0, end_mask = var_8785_end_mask_0, x = var_8511_cast_fp16)[name = tensor("op_8785_cast_fp16")]; + tensor var_8792_begin_0 = const()[name = tensor("op_8792_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8792_end_0 = const()[name = tensor("op_8792_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8792_end_mask_0 = const()[name = tensor("op_8792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8792_cast_fp16 = slice_by_index(begin = var_8792_begin_0, end = var_8792_end_0, end_mask = var_8792_end_mask_0, x = var_8515_cast_fp16)[name = tensor("op_8792_cast_fp16")]; + tensor var_8799_begin_0 = const()[name = tensor("op_8799_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8799_end_0 = const()[name = tensor("op_8799_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8799_end_mask_0 = const()[name = tensor("op_8799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8799_cast_fp16 = slice_by_index(begin = var_8799_begin_0, end = var_8799_end_0, end_mask = var_8799_end_mask_0, x = var_8515_cast_fp16)[name = tensor("op_8799_cast_fp16")]; + tensor var_8806_begin_0 = const()[name = tensor("op_8806_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8806_end_0 = const()[name = tensor("op_8806_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8806_end_mask_0 = const()[name = tensor("op_8806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8806_cast_fp16 = slice_by_index(begin = var_8806_begin_0, end = var_8806_end_0, end_mask = var_8806_end_mask_0, x = var_8515_cast_fp16)[name = tensor("op_8806_cast_fp16")]; + tensor var_8813_begin_0 = const()[name = tensor("op_8813_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8813_end_0 = const()[name = tensor("op_8813_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8813_end_mask_0 = const()[name = tensor("op_8813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8813_cast_fp16 = slice_by_index(begin = var_8813_begin_0, end = var_8813_end_0, end_mask = var_8813_end_mask_0, x = var_8515_cast_fp16)[name = tensor("op_8813_cast_fp16")]; + tensor var_8820_begin_0 = const()[name = tensor("op_8820_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8820_end_0 = const()[name = tensor("op_8820_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8820_end_mask_0 = const()[name = tensor("op_8820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8820_cast_fp16 = slice_by_index(begin = var_8820_begin_0, end = var_8820_end_0, end_mask = var_8820_end_mask_0, x = var_8519_cast_fp16)[name = tensor("op_8820_cast_fp16")]; + tensor var_8827_begin_0 = const()[name = tensor("op_8827_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8827_end_0 = const()[name = tensor("op_8827_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8827_end_mask_0 = const()[name = tensor("op_8827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8827_cast_fp16 = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = var_8519_cast_fp16)[name = tensor("op_8827_cast_fp16")]; + tensor var_8834_begin_0 = const()[name = tensor("op_8834_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8834_end_0 = const()[name = tensor("op_8834_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8834_end_mask_0 = const()[name = tensor("op_8834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8834_cast_fp16 = slice_by_index(begin = var_8834_begin_0, end = var_8834_end_0, end_mask = var_8834_end_mask_0, x = var_8519_cast_fp16)[name = tensor("op_8834_cast_fp16")]; + tensor var_8841_begin_0 = const()[name = tensor("op_8841_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8841_end_0 = const()[name = tensor("op_8841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8841_end_mask_0 = const()[name = tensor("op_8841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8841_cast_fp16 = slice_by_index(begin = var_8841_begin_0, end = var_8841_end_0, end_mask = var_8841_end_mask_0, x = var_8519_cast_fp16)[name = tensor("op_8841_cast_fp16")]; + tensor var_8848_begin_0 = const()[name = tensor("op_8848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8848_end_0 = const()[name = tensor("op_8848_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8848_end_mask_0 = const()[name = tensor("op_8848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8848_cast_fp16 = slice_by_index(begin = var_8848_begin_0, end = var_8848_end_0, end_mask = var_8848_end_mask_0, x = var_8523_cast_fp16)[name = tensor("op_8848_cast_fp16")]; + tensor var_8855_begin_0 = const()[name = tensor("op_8855_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8855_end_0 = const()[name = tensor("op_8855_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8855_end_mask_0 = const()[name = tensor("op_8855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8855_cast_fp16 = slice_by_index(begin = var_8855_begin_0, end = var_8855_end_0, end_mask = var_8855_end_mask_0, x = var_8523_cast_fp16)[name = tensor("op_8855_cast_fp16")]; + tensor var_8862_begin_0 = const()[name = tensor("op_8862_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8862_end_0 = const()[name = tensor("op_8862_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8862_end_mask_0 = const()[name = tensor("op_8862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8862_cast_fp16 = slice_by_index(begin = var_8862_begin_0, end = var_8862_end_0, end_mask = var_8862_end_mask_0, x = var_8523_cast_fp16)[name = tensor("op_8862_cast_fp16")]; + tensor var_8869_begin_0 = const()[name = tensor("op_8869_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8869_end_0 = const()[name = tensor("op_8869_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8869_end_mask_0 = const()[name = tensor("op_8869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8869_cast_fp16 = slice_by_index(begin = var_8869_begin_0, end = var_8869_end_0, end_mask = var_8869_end_mask_0, x = var_8523_cast_fp16)[name = tensor("op_8869_cast_fp16")]; + tensor var_8876_begin_0 = const()[name = tensor("op_8876_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8876_end_0 = const()[name = tensor("op_8876_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8876_end_mask_0 = const()[name = tensor("op_8876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8876_cast_fp16 = slice_by_index(begin = var_8876_begin_0, end = var_8876_end_0, end_mask = var_8876_end_mask_0, x = var_8527_cast_fp16)[name = tensor("op_8876_cast_fp16")]; + tensor var_8883_begin_0 = const()[name = tensor("op_8883_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8883_end_0 = const()[name = tensor("op_8883_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8883_end_mask_0 = const()[name = tensor("op_8883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8883_cast_fp16 = slice_by_index(begin = var_8883_begin_0, end = var_8883_end_0, end_mask = var_8883_end_mask_0, x = var_8527_cast_fp16)[name = tensor("op_8883_cast_fp16")]; + tensor var_8890_begin_0 = const()[name = tensor("op_8890_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8890_end_0 = const()[name = tensor("op_8890_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8890_end_mask_0 = const()[name = tensor("op_8890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8890_cast_fp16 = slice_by_index(begin = var_8890_begin_0, end = var_8890_end_0, end_mask = var_8890_end_mask_0, x = var_8527_cast_fp16)[name = tensor("op_8890_cast_fp16")]; + tensor var_8897_begin_0 = const()[name = tensor("op_8897_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8897_end_0 = const()[name = tensor("op_8897_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8897_end_mask_0 = const()[name = tensor("op_8897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8897_cast_fp16 = slice_by_index(begin = var_8897_begin_0, end = var_8897_end_0, end_mask = var_8897_end_mask_0, x = var_8527_cast_fp16)[name = tensor("op_8897_cast_fp16")]; + tensor var_8904_begin_0 = const()[name = tensor("op_8904_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8904_end_0 = const()[name = tensor("op_8904_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8904_end_mask_0 = const()[name = tensor("op_8904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8904_cast_fp16 = slice_by_index(begin = var_8904_begin_0, end = var_8904_end_0, end_mask = var_8904_end_mask_0, x = var_8531_cast_fp16)[name = tensor("op_8904_cast_fp16")]; + tensor var_8911_begin_0 = const()[name = tensor("op_8911_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8911_end_0 = const()[name = tensor("op_8911_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8911_end_mask_0 = const()[name = tensor("op_8911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8911_cast_fp16 = slice_by_index(begin = var_8911_begin_0, end = var_8911_end_0, end_mask = var_8911_end_mask_0, x = var_8531_cast_fp16)[name = tensor("op_8911_cast_fp16")]; + tensor var_8918_begin_0 = const()[name = tensor("op_8918_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8918_end_0 = const()[name = tensor("op_8918_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8918_end_mask_0 = const()[name = tensor("op_8918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8918_cast_fp16 = slice_by_index(begin = var_8918_begin_0, end = var_8918_end_0, end_mask = var_8918_end_mask_0, x = var_8531_cast_fp16)[name = tensor("op_8918_cast_fp16")]; + tensor var_8925_begin_0 = const()[name = tensor("op_8925_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8925_end_0 = const()[name = tensor("op_8925_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8925_end_mask_0 = const()[name = tensor("op_8925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8925_cast_fp16 = slice_by_index(begin = var_8925_begin_0, end = var_8925_end_0, end_mask = var_8925_end_mask_0, x = var_8531_cast_fp16)[name = tensor("op_8925_cast_fp16")]; + tensor var_8932_begin_0 = const()[name = tensor("op_8932_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8932_end_0 = const()[name = tensor("op_8932_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8932_end_mask_0 = const()[name = tensor("op_8932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8932_cast_fp16 = slice_by_index(begin = var_8932_begin_0, end = var_8932_end_0, end_mask = var_8932_end_mask_0, x = var_8535_cast_fp16)[name = tensor("op_8932_cast_fp16")]; + tensor var_8939_begin_0 = const()[name = tensor("op_8939_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8939_end_0 = const()[name = tensor("op_8939_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8939_end_mask_0 = const()[name = tensor("op_8939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8939_cast_fp16 = slice_by_index(begin = var_8939_begin_0, end = var_8939_end_0, end_mask = var_8939_end_mask_0, x = var_8535_cast_fp16)[name = tensor("op_8939_cast_fp16")]; + tensor var_8946_begin_0 = const()[name = tensor("op_8946_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8946_end_0 = const()[name = tensor("op_8946_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8946_end_mask_0 = const()[name = tensor("op_8946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8946_cast_fp16 = slice_by_index(begin = var_8946_begin_0, end = var_8946_end_0, end_mask = var_8946_end_mask_0, x = var_8535_cast_fp16)[name = tensor("op_8946_cast_fp16")]; + tensor var_8953_begin_0 = const()[name = tensor("op_8953_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8953_end_0 = const()[name = tensor("op_8953_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8953_end_mask_0 = const()[name = tensor("op_8953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8953_cast_fp16 = slice_by_index(begin = var_8953_begin_0, end = var_8953_end_0, end_mask = var_8953_end_mask_0, x = var_8535_cast_fp16)[name = tensor("op_8953_cast_fp16")]; + tensor var_8960_begin_0 = const()[name = tensor("op_8960_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8960_end_0 = const()[name = tensor("op_8960_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8960_end_mask_0 = const()[name = tensor("op_8960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8960_cast_fp16 = slice_by_index(begin = var_8960_begin_0, end = var_8960_end_0, end_mask = var_8960_end_mask_0, x = var_8539_cast_fp16)[name = tensor("op_8960_cast_fp16")]; + tensor var_8967_begin_0 = const()[name = tensor("op_8967_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8967_end_0 = const()[name = tensor("op_8967_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8967_end_mask_0 = const()[name = tensor("op_8967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8967_cast_fp16 = slice_by_index(begin = var_8967_begin_0, end = var_8967_end_0, end_mask = var_8967_end_mask_0, x = var_8539_cast_fp16)[name = tensor("op_8967_cast_fp16")]; + tensor var_8974_begin_0 = const()[name = tensor("op_8974_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_8974_end_0 = const()[name = tensor("op_8974_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_8974_end_mask_0 = const()[name = tensor("op_8974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8974_cast_fp16 = slice_by_index(begin = var_8974_begin_0, end = var_8974_end_0, end_mask = var_8974_end_mask_0, x = var_8539_cast_fp16)[name = tensor("op_8974_cast_fp16")]; + tensor var_8981_begin_0 = const()[name = tensor("op_8981_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_8981_end_0 = const()[name = tensor("op_8981_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_8981_end_mask_0 = const()[name = tensor("op_8981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8981_cast_fp16 = slice_by_index(begin = var_8981_begin_0, end = var_8981_end_0, end_mask = var_8981_end_mask_0, x = var_8539_cast_fp16)[name = tensor("op_8981_cast_fp16")]; + tensor var_8988_begin_0 = const()[name = tensor("op_8988_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8988_end_0 = const()[name = tensor("op_8988_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_8988_end_mask_0 = const()[name = tensor("op_8988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8988_cast_fp16 = slice_by_index(begin = var_8988_begin_0, end = var_8988_end_0, end_mask = var_8988_end_mask_0, x = var_8543_cast_fp16)[name = tensor("op_8988_cast_fp16")]; + tensor var_8995_begin_0 = const()[name = tensor("op_8995_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_8995_end_0 = const()[name = tensor("op_8995_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_8995_end_mask_0 = const()[name = tensor("op_8995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8995_cast_fp16 = slice_by_index(begin = var_8995_begin_0, end = var_8995_end_0, end_mask = var_8995_end_mask_0, x = var_8543_cast_fp16)[name = tensor("op_8995_cast_fp16")]; + tensor var_9002_begin_0 = const()[name = tensor("op_9002_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9002_end_0 = const()[name = tensor("op_9002_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9002_end_mask_0 = const()[name = tensor("op_9002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9002_cast_fp16 = slice_by_index(begin = var_9002_begin_0, end = var_9002_end_0, end_mask = var_9002_end_mask_0, x = var_8543_cast_fp16)[name = tensor("op_9002_cast_fp16")]; + tensor var_9009_begin_0 = const()[name = tensor("op_9009_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9009_end_0 = const()[name = tensor("op_9009_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9009_end_mask_0 = const()[name = tensor("op_9009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9009_cast_fp16 = slice_by_index(begin = var_9009_begin_0, end = var_9009_end_0, end_mask = var_9009_end_mask_0, x = var_8543_cast_fp16)[name = tensor("op_9009_cast_fp16")]; + tensor var_9016_begin_0 = const()[name = tensor("op_9016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9016_end_0 = const()[name = tensor("op_9016_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9016_end_mask_0 = const()[name = tensor("op_9016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9016_cast_fp16 = slice_by_index(begin = var_9016_begin_0, end = var_9016_end_0, end_mask = var_9016_end_mask_0, x = var_8547_cast_fp16)[name = tensor("op_9016_cast_fp16")]; + tensor var_9023_begin_0 = const()[name = tensor("op_9023_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9023_end_0 = const()[name = tensor("op_9023_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9023_end_mask_0 = const()[name = tensor("op_9023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9023_cast_fp16 = slice_by_index(begin = var_9023_begin_0, end = var_9023_end_0, end_mask = var_9023_end_mask_0, x = var_8547_cast_fp16)[name = tensor("op_9023_cast_fp16")]; + tensor var_9030_begin_0 = const()[name = tensor("op_9030_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9030_end_0 = const()[name = tensor("op_9030_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9030_end_mask_0 = const()[name = tensor("op_9030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9030_cast_fp16 = slice_by_index(begin = var_9030_begin_0, end = var_9030_end_0, end_mask = var_9030_end_mask_0, x = var_8547_cast_fp16)[name = tensor("op_9030_cast_fp16")]; + tensor var_9037_begin_0 = const()[name = tensor("op_9037_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9037_end_0 = const()[name = tensor("op_9037_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9037_end_mask_0 = const()[name = tensor("op_9037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9037_cast_fp16 = slice_by_index(begin = var_9037_begin_0, end = var_9037_end_0, end_mask = var_9037_end_mask_0, x = var_8547_cast_fp16)[name = tensor("op_9037_cast_fp16")]; + tensor var_9044_begin_0 = const()[name = tensor("op_9044_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9044_end_0 = const()[name = tensor("op_9044_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9044_end_mask_0 = const()[name = tensor("op_9044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9044_cast_fp16 = slice_by_index(begin = var_9044_begin_0, end = var_9044_end_0, end_mask = var_9044_end_mask_0, x = var_8551_cast_fp16)[name = tensor("op_9044_cast_fp16")]; + tensor var_9051_begin_0 = const()[name = tensor("op_9051_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9051_end_0 = const()[name = tensor("op_9051_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9051_end_mask_0 = const()[name = tensor("op_9051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9051_cast_fp16 = slice_by_index(begin = var_9051_begin_0, end = var_9051_end_0, end_mask = var_9051_end_mask_0, x = var_8551_cast_fp16)[name = tensor("op_9051_cast_fp16")]; + tensor var_9058_begin_0 = const()[name = tensor("op_9058_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9058_end_0 = const()[name = tensor("op_9058_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9058_end_mask_0 = const()[name = tensor("op_9058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9058_cast_fp16 = slice_by_index(begin = var_9058_begin_0, end = var_9058_end_0, end_mask = var_9058_end_mask_0, x = var_8551_cast_fp16)[name = tensor("op_9058_cast_fp16")]; + tensor var_9065_begin_0 = const()[name = tensor("op_9065_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9065_end_0 = const()[name = tensor("op_9065_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9065_end_mask_0 = const()[name = tensor("op_9065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9065_cast_fp16 = slice_by_index(begin = var_9065_begin_0, end = var_9065_end_0, end_mask = var_9065_end_mask_0, x = var_8551_cast_fp16)[name = tensor("op_9065_cast_fp16")]; + tensor var_9072_begin_0 = const()[name = tensor("op_9072_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9072_end_0 = const()[name = tensor("op_9072_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9072_end_mask_0 = const()[name = tensor("op_9072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9072_cast_fp16 = slice_by_index(begin = var_9072_begin_0, end = var_9072_end_0, end_mask = var_9072_end_mask_0, x = var_8555_cast_fp16)[name = tensor("op_9072_cast_fp16")]; + tensor var_9079_begin_0 = const()[name = tensor("op_9079_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9079_end_0 = const()[name = tensor("op_9079_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9079_end_mask_0 = const()[name = tensor("op_9079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9079_cast_fp16 = slice_by_index(begin = var_9079_begin_0, end = var_9079_end_0, end_mask = var_9079_end_mask_0, x = var_8555_cast_fp16)[name = tensor("op_9079_cast_fp16")]; + tensor var_9086_begin_0 = const()[name = tensor("op_9086_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9086_end_0 = const()[name = tensor("op_9086_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9086_end_mask_0 = const()[name = tensor("op_9086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9086_cast_fp16 = slice_by_index(begin = var_9086_begin_0, end = var_9086_end_0, end_mask = var_9086_end_mask_0, x = var_8555_cast_fp16)[name = tensor("op_9086_cast_fp16")]; + tensor var_9093_begin_0 = const()[name = tensor("op_9093_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9093_end_0 = const()[name = tensor("op_9093_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9093_end_mask_0 = const()[name = tensor("op_9093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9093_cast_fp16 = slice_by_index(begin = var_9093_begin_0, end = var_9093_end_0, end_mask = var_9093_end_mask_0, x = var_8555_cast_fp16)[name = tensor("op_9093_cast_fp16")]; + tensor var_9100_begin_0 = const()[name = tensor("op_9100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9100_end_0 = const()[name = tensor("op_9100_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_9100_end_mask_0 = const()[name = tensor("op_9100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9100_cast_fp16 = slice_by_index(begin = var_9100_begin_0, end = var_9100_end_0, end_mask = var_9100_end_mask_0, x = var_8559_cast_fp16)[name = tensor("op_9100_cast_fp16")]; + tensor var_9107_begin_0 = const()[name = tensor("op_9107_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_9107_end_0 = const()[name = tensor("op_9107_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_9107_end_mask_0 = const()[name = tensor("op_9107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9107_cast_fp16 = slice_by_index(begin = var_9107_begin_0, end = var_9107_end_0, end_mask = var_9107_end_mask_0, x = var_8559_cast_fp16)[name = tensor("op_9107_cast_fp16")]; + tensor var_9114_begin_0 = const()[name = tensor("op_9114_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_9114_end_0 = const()[name = tensor("op_9114_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_9114_end_mask_0 = const()[name = tensor("op_9114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9114_cast_fp16 = slice_by_index(begin = var_9114_begin_0, end = var_9114_end_0, end_mask = var_9114_end_mask_0, x = var_8559_cast_fp16)[name = tensor("op_9114_cast_fp16")]; + tensor var_9121_begin_0 = const()[name = tensor("op_9121_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_9121_end_0 = const()[name = tensor("op_9121_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9121_end_mask_0 = const()[name = tensor("op_9121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9121_cast_fp16 = slice_by_index(begin = var_9121_begin_0, end = var_9121_end_0, end_mask = var_9121_end_mask_0, x = var_8559_cast_fp16)[name = tensor("op_9121_cast_fp16")]; + tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_9126_begin_0 = const()[name = tensor("op_9126_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9126_end_0 = const()[name = tensor("op_9126_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_9126_end_mask_0 = const()[name = tensor("op_9126_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_26 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = tensor("transpose_26")]; + tensor var_9126_cast_fp16 = slice_by_index(begin = var_9126_begin_0, end = var_9126_end_0, end_mask = var_9126_end_mask_0, x = transpose_26)[name = tensor("op_9126_cast_fp16")]; + tensor var_9130_begin_0 = const()[name = tensor("op_9130_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_9130_end_0 = const()[name = tensor("op_9130_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_9130_end_mask_0 = const()[name = tensor("op_9130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9130_cast_fp16 = slice_by_index(begin = var_9130_begin_0, end = var_9130_end_0, end_mask = var_9130_end_mask_0, x = transpose_26)[name = tensor("op_9130_cast_fp16")]; + tensor var_9134_begin_0 = const()[name = tensor("op_9134_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_9134_end_0 = const()[name = tensor("op_9134_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_9134_end_mask_0 = const()[name = tensor("op_9134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9134_cast_fp16 = slice_by_index(begin = var_9134_begin_0, end = var_9134_end_0, end_mask = var_9134_end_mask_0, x = transpose_26)[name = tensor("op_9134_cast_fp16")]; + tensor var_9138_begin_0 = const()[name = tensor("op_9138_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_9138_end_0 = const()[name = tensor("op_9138_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_9138_end_mask_0 = const()[name = tensor("op_9138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9138_cast_fp16 = slice_by_index(begin = var_9138_begin_0, end = var_9138_end_0, end_mask = var_9138_end_mask_0, x = transpose_26)[name = tensor("op_9138_cast_fp16")]; + tensor var_9142_begin_0 = const()[name = tensor("op_9142_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_9142_end_0 = const()[name = tensor("op_9142_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_9142_end_mask_0 = const()[name = tensor("op_9142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9142_cast_fp16 = slice_by_index(begin = var_9142_begin_0, end = var_9142_end_0, end_mask = var_9142_end_mask_0, x = transpose_26)[name = tensor("op_9142_cast_fp16")]; + tensor var_9146_begin_0 = const()[name = tensor("op_9146_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_9146_end_0 = const()[name = tensor("op_9146_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_9146_end_mask_0 = const()[name = tensor("op_9146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9146_cast_fp16 = slice_by_index(begin = var_9146_begin_0, end = var_9146_end_0, end_mask = var_9146_end_mask_0, x = transpose_26)[name = tensor("op_9146_cast_fp16")]; + tensor var_9150_begin_0 = const()[name = tensor("op_9150_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_9150_end_0 = const()[name = tensor("op_9150_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_9150_end_mask_0 = const()[name = tensor("op_9150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9150_cast_fp16 = slice_by_index(begin = var_9150_begin_0, end = var_9150_end_0, end_mask = var_9150_end_mask_0, x = transpose_26)[name = tensor("op_9150_cast_fp16")]; + tensor var_9154_begin_0 = const()[name = tensor("op_9154_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_9154_end_0 = const()[name = tensor("op_9154_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_9154_end_mask_0 = const()[name = tensor("op_9154_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9154_cast_fp16 = slice_by_index(begin = var_9154_begin_0, end = var_9154_end_0, end_mask = var_9154_end_mask_0, x = transpose_26)[name = tensor("op_9154_cast_fp16")]; + tensor var_9158_begin_0 = const()[name = tensor("op_9158_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_9158_end_0 = const()[name = tensor("op_9158_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_9158_end_mask_0 = const()[name = tensor("op_9158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9158_cast_fp16 = slice_by_index(begin = var_9158_begin_0, end = var_9158_end_0, end_mask = var_9158_end_mask_0, x = transpose_26)[name = tensor("op_9158_cast_fp16")]; + tensor var_9162_begin_0 = const()[name = tensor("op_9162_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_9162_end_0 = const()[name = tensor("op_9162_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_9162_end_mask_0 = const()[name = tensor("op_9162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9162_cast_fp16 = slice_by_index(begin = var_9162_begin_0, end = var_9162_end_0, end_mask = var_9162_end_mask_0, x = transpose_26)[name = tensor("op_9162_cast_fp16")]; + tensor var_9166_begin_0 = const()[name = tensor("op_9166_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_9166_end_0 = const()[name = tensor("op_9166_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_9166_end_mask_0 = const()[name = tensor("op_9166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9166_cast_fp16 = slice_by_index(begin = var_9166_begin_0, end = var_9166_end_0, end_mask = var_9166_end_mask_0, x = transpose_26)[name = tensor("op_9166_cast_fp16")]; + tensor var_9170_begin_0 = const()[name = tensor("op_9170_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_9170_end_0 = const()[name = tensor("op_9170_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_9170_end_mask_0 = const()[name = tensor("op_9170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9170_cast_fp16 = slice_by_index(begin = var_9170_begin_0, end = var_9170_end_0, end_mask = var_9170_end_mask_0, x = transpose_26)[name = tensor("op_9170_cast_fp16")]; + tensor var_9174_begin_0 = const()[name = tensor("op_9174_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_9174_end_0 = const()[name = tensor("op_9174_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_9174_end_mask_0 = const()[name = tensor("op_9174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9174_cast_fp16 = slice_by_index(begin = var_9174_begin_0, end = var_9174_end_0, end_mask = var_9174_end_mask_0, x = transpose_26)[name = tensor("op_9174_cast_fp16")]; + tensor var_9178_begin_0 = const()[name = tensor("op_9178_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_9178_end_0 = const()[name = tensor("op_9178_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_9178_end_mask_0 = const()[name = tensor("op_9178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9178_cast_fp16 = slice_by_index(begin = var_9178_begin_0, end = var_9178_end_0, end_mask = var_9178_end_mask_0, x = transpose_26)[name = tensor("op_9178_cast_fp16")]; + tensor var_9182_begin_0 = const()[name = tensor("op_9182_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_9182_end_0 = const()[name = tensor("op_9182_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_9182_end_mask_0 = const()[name = tensor("op_9182_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9182_cast_fp16 = slice_by_index(begin = var_9182_begin_0, end = var_9182_end_0, end_mask = var_9182_end_mask_0, x = transpose_26)[name = tensor("op_9182_cast_fp16")]; + tensor var_9186_begin_0 = const()[name = tensor("op_9186_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_9186_end_0 = const()[name = tensor("op_9186_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_9186_end_mask_0 = const()[name = tensor("op_9186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9186_cast_fp16 = slice_by_index(begin = var_9186_begin_0, end = var_9186_end_0, end_mask = var_9186_end_mask_0, x = transpose_26)[name = tensor("op_9186_cast_fp16")]; + tensor var_9190_begin_0 = const()[name = tensor("op_9190_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_9190_end_0 = const()[name = tensor("op_9190_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_9190_end_mask_0 = const()[name = tensor("op_9190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9190_cast_fp16 = slice_by_index(begin = var_9190_begin_0, end = var_9190_end_0, end_mask = var_9190_end_mask_0, x = transpose_26)[name = tensor("op_9190_cast_fp16")]; + tensor var_9194_begin_0 = const()[name = tensor("op_9194_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_9194_end_0 = const()[name = tensor("op_9194_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_9194_end_mask_0 = const()[name = tensor("op_9194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9194_cast_fp16 = slice_by_index(begin = var_9194_begin_0, end = var_9194_end_0, end_mask = var_9194_end_mask_0, x = transpose_26)[name = tensor("op_9194_cast_fp16")]; + tensor var_9198_begin_0 = const()[name = tensor("op_9198_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_9198_end_0 = const()[name = tensor("op_9198_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_9198_end_mask_0 = const()[name = tensor("op_9198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9198_cast_fp16 = slice_by_index(begin = var_9198_begin_0, end = var_9198_end_0, end_mask = var_9198_end_mask_0, x = transpose_26)[name = tensor("op_9198_cast_fp16")]; + tensor var_9202_begin_0 = const()[name = tensor("op_9202_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_9202_end_0 = const()[name = tensor("op_9202_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_9202_end_mask_0 = const()[name = tensor("op_9202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9202_cast_fp16 = slice_by_index(begin = var_9202_begin_0, end = var_9202_end_0, end_mask = var_9202_end_mask_0, x = transpose_26)[name = tensor("op_9202_cast_fp16")]; + tensor var_9204_begin_0 = const()[name = tensor("op_9204_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9204_end_0 = const()[name = tensor("op_9204_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_9204_end_mask_0 = const()[name = tensor("op_9204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9204_cast_fp16 = slice_by_index(begin = var_9204_begin_0, end = var_9204_end_0, end_mask = var_9204_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9204_cast_fp16")]; + tensor var_9208_begin_0 = const()[name = tensor("op_9208_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_9208_end_0 = const()[name = tensor("op_9208_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_9208_end_mask_0 = const()[name = tensor("op_9208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9208_cast_fp16 = slice_by_index(begin = var_9208_begin_0, end = var_9208_end_0, end_mask = var_9208_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9208_cast_fp16")]; + tensor var_9212_begin_0 = const()[name = tensor("op_9212_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_9212_end_0 = const()[name = tensor("op_9212_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_9212_end_mask_0 = const()[name = tensor("op_9212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9212_cast_fp16 = slice_by_index(begin = var_9212_begin_0, end = var_9212_end_0, end_mask = var_9212_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9212_cast_fp16")]; + tensor var_9216_begin_0 = const()[name = tensor("op_9216_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_9216_end_0 = const()[name = tensor("op_9216_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_9216_end_mask_0 = const()[name = tensor("op_9216_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9216_cast_fp16 = slice_by_index(begin = var_9216_begin_0, end = var_9216_end_0, end_mask = var_9216_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9216_cast_fp16")]; + tensor var_9220_begin_0 = const()[name = tensor("op_9220_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_9220_end_0 = const()[name = tensor("op_9220_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_9220_end_mask_0 = const()[name = tensor("op_9220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9220_cast_fp16 = slice_by_index(begin = var_9220_begin_0, end = var_9220_end_0, end_mask = var_9220_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9220_cast_fp16")]; + tensor var_9224_begin_0 = const()[name = tensor("op_9224_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_9224_end_0 = const()[name = tensor("op_9224_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_9224_end_mask_0 = const()[name = tensor("op_9224_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9224_cast_fp16 = slice_by_index(begin = var_9224_begin_0, end = var_9224_end_0, end_mask = var_9224_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9224_cast_fp16")]; + tensor var_9228_begin_0 = const()[name = tensor("op_9228_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_9228_end_0 = const()[name = tensor("op_9228_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_9228_end_mask_0 = const()[name = tensor("op_9228_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9228_cast_fp16 = slice_by_index(begin = var_9228_begin_0, end = var_9228_end_0, end_mask = var_9228_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9228_cast_fp16")]; + tensor var_9232_begin_0 = const()[name = tensor("op_9232_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_9232_end_0 = const()[name = tensor("op_9232_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_9232_end_mask_0 = const()[name = tensor("op_9232_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9232_cast_fp16 = slice_by_index(begin = var_9232_begin_0, end = var_9232_end_0, end_mask = var_9232_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9232_cast_fp16")]; + tensor var_9236_begin_0 = const()[name = tensor("op_9236_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_9236_end_0 = const()[name = tensor("op_9236_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_9236_end_mask_0 = const()[name = tensor("op_9236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9236_cast_fp16 = slice_by_index(begin = var_9236_begin_0, end = var_9236_end_0, end_mask = var_9236_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9236_cast_fp16")]; + tensor var_9240_begin_0 = const()[name = tensor("op_9240_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_9240_end_0 = const()[name = tensor("op_9240_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_9240_end_mask_0 = const()[name = tensor("op_9240_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9240_cast_fp16 = slice_by_index(begin = var_9240_begin_0, end = var_9240_end_0, end_mask = var_9240_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9240_cast_fp16")]; + tensor var_9244_begin_0 = const()[name = tensor("op_9244_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_9244_end_0 = const()[name = tensor("op_9244_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_9244_end_mask_0 = const()[name = tensor("op_9244_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9244_cast_fp16 = slice_by_index(begin = var_9244_begin_0, end = var_9244_end_0, end_mask = var_9244_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9244_cast_fp16")]; + tensor var_9248_begin_0 = const()[name = tensor("op_9248_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_9248_end_0 = const()[name = tensor("op_9248_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_9248_end_mask_0 = const()[name = tensor("op_9248_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9248_cast_fp16 = slice_by_index(begin = var_9248_begin_0, end = var_9248_end_0, end_mask = var_9248_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9248_cast_fp16")]; + tensor var_9252_begin_0 = const()[name = tensor("op_9252_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_9252_end_0 = const()[name = tensor("op_9252_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_9252_end_mask_0 = const()[name = tensor("op_9252_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9252_cast_fp16 = slice_by_index(begin = var_9252_begin_0, end = var_9252_end_0, end_mask = var_9252_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9252_cast_fp16")]; + tensor var_9256_begin_0 = const()[name = tensor("op_9256_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_9256_end_0 = const()[name = tensor("op_9256_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_9256_end_mask_0 = const()[name = tensor("op_9256_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9256_cast_fp16 = slice_by_index(begin = var_9256_begin_0, end = var_9256_end_0, end_mask = var_9256_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9256_cast_fp16")]; + tensor var_9260_begin_0 = const()[name = tensor("op_9260_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_9260_end_0 = const()[name = tensor("op_9260_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_9260_end_mask_0 = const()[name = tensor("op_9260_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9260_cast_fp16 = slice_by_index(begin = var_9260_begin_0, end = var_9260_end_0, end_mask = var_9260_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9260_cast_fp16")]; + tensor var_9264_begin_0 = const()[name = tensor("op_9264_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_9264_end_0 = const()[name = tensor("op_9264_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_9264_end_mask_0 = const()[name = tensor("op_9264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9264_cast_fp16 = slice_by_index(begin = var_9264_begin_0, end = var_9264_end_0, end_mask = var_9264_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9264_cast_fp16")]; + tensor var_9268_begin_0 = const()[name = tensor("op_9268_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_9268_end_0 = const()[name = tensor("op_9268_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_9268_end_mask_0 = const()[name = tensor("op_9268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9268_cast_fp16 = slice_by_index(begin = var_9268_begin_0, end = var_9268_end_0, end_mask = var_9268_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9268_cast_fp16")]; + tensor var_9272_begin_0 = const()[name = tensor("op_9272_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_9272_end_0 = const()[name = tensor("op_9272_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_9272_end_mask_0 = const()[name = tensor("op_9272_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9272_cast_fp16 = slice_by_index(begin = var_9272_begin_0, end = var_9272_end_0, end_mask = var_9272_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9272_cast_fp16")]; + tensor var_9276_begin_0 = const()[name = tensor("op_9276_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_9276_end_0 = const()[name = tensor("op_9276_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_9276_end_mask_0 = const()[name = tensor("op_9276_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9276_cast_fp16 = slice_by_index(begin = var_9276_begin_0, end = var_9276_end_0, end_mask = var_9276_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9276_cast_fp16")]; + tensor var_9280_begin_0 = const()[name = tensor("op_9280_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_9280_end_0 = const()[name = tensor("op_9280_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_9280_end_mask_0 = const()[name = tensor("op_9280_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_9280_cast_fp16 = slice_by_index(begin = var_9280_begin_0, end = var_9280_end_0, end_mask = var_9280_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_9280_cast_fp16")]; + tensor var_9284_equation_0 = const()[name = tensor("op_9284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9284_cast_fp16 = einsum(equation = var_9284_equation_0, values = (var_9126_cast_fp16, var_8568_cast_fp16))[name = tensor("op_9284_cast_fp16")]; + tensor var_9285_to_fp16 = const()[name = tensor("op_9285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_801_cast_fp16 = mul(x = var_9284_cast_fp16, y = var_9285_to_fp16)[name = tensor("aw_chunk_801_cast_fp16")]; + tensor var_9288_equation_0 = const()[name = tensor("op_9288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9288_cast_fp16 = einsum(equation = var_9288_equation_0, values = (var_9126_cast_fp16, var_8575_cast_fp16))[name = tensor("op_9288_cast_fp16")]; + tensor var_9289_to_fp16 = const()[name = tensor("op_9289_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_803_cast_fp16 = mul(x = var_9288_cast_fp16, y = var_9289_to_fp16)[name = tensor("aw_chunk_803_cast_fp16")]; + tensor var_9292_equation_0 = const()[name = tensor("op_9292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9292_cast_fp16 = einsum(equation = var_9292_equation_0, values = (var_9126_cast_fp16, var_8582_cast_fp16))[name = tensor("op_9292_cast_fp16")]; + tensor var_9293_to_fp16 = const()[name = tensor("op_9293_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_805_cast_fp16 = mul(x = var_9292_cast_fp16, y = var_9293_to_fp16)[name = tensor("aw_chunk_805_cast_fp16")]; + tensor var_9296_equation_0 = const()[name = tensor("op_9296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9296_cast_fp16 = einsum(equation = var_9296_equation_0, values = (var_9126_cast_fp16, var_8589_cast_fp16))[name = tensor("op_9296_cast_fp16")]; + tensor var_9297_to_fp16 = const()[name = tensor("op_9297_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_807_cast_fp16 = mul(x = var_9296_cast_fp16, y = var_9297_to_fp16)[name = tensor("aw_chunk_807_cast_fp16")]; + tensor var_9300_equation_0 = const()[name = tensor("op_9300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9300_cast_fp16 = einsum(equation = var_9300_equation_0, values = (var_9130_cast_fp16, var_8596_cast_fp16))[name = tensor("op_9300_cast_fp16")]; + tensor var_9301_to_fp16 = const()[name = tensor("op_9301_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_809_cast_fp16 = mul(x = var_9300_cast_fp16, y = var_9301_to_fp16)[name = tensor("aw_chunk_809_cast_fp16")]; + tensor var_9304_equation_0 = const()[name = tensor("op_9304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9304_cast_fp16 = einsum(equation = var_9304_equation_0, values = (var_9130_cast_fp16, var_8603_cast_fp16))[name = tensor("op_9304_cast_fp16")]; + tensor var_9305_to_fp16 = const()[name = tensor("op_9305_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_811_cast_fp16 = mul(x = var_9304_cast_fp16, y = var_9305_to_fp16)[name = tensor("aw_chunk_811_cast_fp16")]; + tensor var_9308_equation_0 = const()[name = tensor("op_9308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9308_cast_fp16 = einsum(equation = var_9308_equation_0, values = (var_9130_cast_fp16, var_8610_cast_fp16))[name = tensor("op_9308_cast_fp16")]; + tensor var_9309_to_fp16 = const()[name = tensor("op_9309_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_813_cast_fp16 = mul(x = var_9308_cast_fp16, y = var_9309_to_fp16)[name = tensor("aw_chunk_813_cast_fp16")]; + tensor var_9312_equation_0 = const()[name = tensor("op_9312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9312_cast_fp16 = einsum(equation = var_9312_equation_0, values = (var_9130_cast_fp16, var_8617_cast_fp16))[name = tensor("op_9312_cast_fp16")]; + tensor var_9313_to_fp16 = const()[name = tensor("op_9313_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_815_cast_fp16 = mul(x = var_9312_cast_fp16, y = var_9313_to_fp16)[name = tensor("aw_chunk_815_cast_fp16")]; + tensor var_9316_equation_0 = const()[name = tensor("op_9316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9316_cast_fp16 = einsum(equation = var_9316_equation_0, values = (var_9134_cast_fp16, var_8624_cast_fp16))[name = tensor("op_9316_cast_fp16")]; + tensor var_9317_to_fp16 = const()[name = tensor("op_9317_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_817_cast_fp16 = mul(x = var_9316_cast_fp16, y = var_9317_to_fp16)[name = tensor("aw_chunk_817_cast_fp16")]; + tensor var_9320_equation_0 = const()[name = tensor("op_9320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9320_cast_fp16 = einsum(equation = var_9320_equation_0, values = (var_9134_cast_fp16, var_8631_cast_fp16))[name = tensor("op_9320_cast_fp16")]; + tensor var_9321_to_fp16 = const()[name = tensor("op_9321_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_819_cast_fp16 = mul(x = var_9320_cast_fp16, y = var_9321_to_fp16)[name = tensor("aw_chunk_819_cast_fp16")]; + tensor var_9324_equation_0 = const()[name = tensor("op_9324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9324_cast_fp16 = einsum(equation = var_9324_equation_0, values = (var_9134_cast_fp16, var_8638_cast_fp16))[name = tensor("op_9324_cast_fp16")]; + tensor var_9325_to_fp16 = const()[name = tensor("op_9325_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_821_cast_fp16 = mul(x = var_9324_cast_fp16, y = var_9325_to_fp16)[name = tensor("aw_chunk_821_cast_fp16")]; + tensor var_9328_equation_0 = const()[name = tensor("op_9328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9328_cast_fp16 = einsum(equation = var_9328_equation_0, values = (var_9134_cast_fp16, var_8645_cast_fp16))[name = tensor("op_9328_cast_fp16")]; + tensor var_9329_to_fp16 = const()[name = tensor("op_9329_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_823_cast_fp16 = mul(x = var_9328_cast_fp16, y = var_9329_to_fp16)[name = tensor("aw_chunk_823_cast_fp16")]; + tensor var_9332_equation_0 = const()[name = tensor("op_9332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9332_cast_fp16 = einsum(equation = var_9332_equation_0, values = (var_9138_cast_fp16, var_8652_cast_fp16))[name = tensor("op_9332_cast_fp16")]; + tensor var_9333_to_fp16 = const()[name = tensor("op_9333_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_825_cast_fp16 = mul(x = var_9332_cast_fp16, y = var_9333_to_fp16)[name = tensor("aw_chunk_825_cast_fp16")]; + tensor var_9336_equation_0 = const()[name = tensor("op_9336_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9336_cast_fp16 = einsum(equation = var_9336_equation_0, values = (var_9138_cast_fp16, var_8659_cast_fp16))[name = tensor("op_9336_cast_fp16")]; + tensor var_9337_to_fp16 = const()[name = tensor("op_9337_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_827_cast_fp16 = mul(x = var_9336_cast_fp16, y = var_9337_to_fp16)[name = tensor("aw_chunk_827_cast_fp16")]; + tensor var_9340_equation_0 = const()[name = tensor("op_9340_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9340_cast_fp16 = einsum(equation = var_9340_equation_0, values = (var_9138_cast_fp16, var_8666_cast_fp16))[name = tensor("op_9340_cast_fp16")]; + tensor var_9341_to_fp16 = const()[name = tensor("op_9341_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_829_cast_fp16 = mul(x = var_9340_cast_fp16, y = var_9341_to_fp16)[name = tensor("aw_chunk_829_cast_fp16")]; + tensor var_9344_equation_0 = const()[name = tensor("op_9344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9344_cast_fp16 = einsum(equation = var_9344_equation_0, values = (var_9138_cast_fp16, var_8673_cast_fp16))[name = tensor("op_9344_cast_fp16")]; + tensor var_9345_to_fp16 = const()[name = tensor("op_9345_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_831_cast_fp16 = mul(x = var_9344_cast_fp16, y = var_9345_to_fp16)[name = tensor("aw_chunk_831_cast_fp16")]; + tensor var_9348_equation_0 = const()[name = tensor("op_9348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9348_cast_fp16 = einsum(equation = var_9348_equation_0, values = (var_9142_cast_fp16, var_8680_cast_fp16))[name = tensor("op_9348_cast_fp16")]; + tensor var_9349_to_fp16 = const()[name = tensor("op_9349_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_833_cast_fp16 = mul(x = var_9348_cast_fp16, y = var_9349_to_fp16)[name = tensor("aw_chunk_833_cast_fp16")]; + tensor var_9352_equation_0 = const()[name = tensor("op_9352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9352_cast_fp16 = einsum(equation = var_9352_equation_0, values = (var_9142_cast_fp16, var_8687_cast_fp16))[name = tensor("op_9352_cast_fp16")]; + tensor var_9353_to_fp16 = const()[name = tensor("op_9353_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_835_cast_fp16 = mul(x = var_9352_cast_fp16, y = var_9353_to_fp16)[name = tensor("aw_chunk_835_cast_fp16")]; + tensor var_9356_equation_0 = const()[name = tensor("op_9356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9356_cast_fp16 = einsum(equation = var_9356_equation_0, values = (var_9142_cast_fp16, var_8694_cast_fp16))[name = tensor("op_9356_cast_fp16")]; + tensor var_9357_to_fp16 = const()[name = tensor("op_9357_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_837_cast_fp16 = mul(x = var_9356_cast_fp16, y = var_9357_to_fp16)[name = tensor("aw_chunk_837_cast_fp16")]; + tensor var_9360_equation_0 = const()[name = tensor("op_9360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9360_cast_fp16 = einsum(equation = var_9360_equation_0, values = (var_9142_cast_fp16, var_8701_cast_fp16))[name = tensor("op_9360_cast_fp16")]; + tensor var_9361_to_fp16 = const()[name = tensor("op_9361_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_839_cast_fp16 = mul(x = var_9360_cast_fp16, y = var_9361_to_fp16)[name = tensor("aw_chunk_839_cast_fp16")]; + tensor var_9364_equation_0 = const()[name = tensor("op_9364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9364_cast_fp16 = einsum(equation = var_9364_equation_0, values = (var_9146_cast_fp16, var_8708_cast_fp16))[name = tensor("op_9364_cast_fp16")]; + tensor var_9365_to_fp16 = const()[name = tensor("op_9365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_841_cast_fp16 = mul(x = var_9364_cast_fp16, y = var_9365_to_fp16)[name = tensor("aw_chunk_841_cast_fp16")]; + tensor var_9368_equation_0 = const()[name = tensor("op_9368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9368_cast_fp16 = einsum(equation = var_9368_equation_0, values = (var_9146_cast_fp16, var_8715_cast_fp16))[name = tensor("op_9368_cast_fp16")]; + tensor var_9369_to_fp16 = const()[name = tensor("op_9369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_843_cast_fp16 = mul(x = var_9368_cast_fp16, y = var_9369_to_fp16)[name = tensor("aw_chunk_843_cast_fp16")]; + tensor var_9372_equation_0 = const()[name = tensor("op_9372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9372_cast_fp16 = einsum(equation = var_9372_equation_0, values = (var_9146_cast_fp16, var_8722_cast_fp16))[name = tensor("op_9372_cast_fp16")]; + tensor var_9373_to_fp16 = const()[name = tensor("op_9373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_845_cast_fp16 = mul(x = var_9372_cast_fp16, y = var_9373_to_fp16)[name = tensor("aw_chunk_845_cast_fp16")]; + tensor var_9376_equation_0 = const()[name = tensor("op_9376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9376_cast_fp16 = einsum(equation = var_9376_equation_0, values = (var_9146_cast_fp16, var_8729_cast_fp16))[name = tensor("op_9376_cast_fp16")]; + tensor var_9377_to_fp16 = const()[name = tensor("op_9377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_847_cast_fp16 = mul(x = var_9376_cast_fp16, y = var_9377_to_fp16)[name = tensor("aw_chunk_847_cast_fp16")]; + tensor var_9380_equation_0 = const()[name = tensor("op_9380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9380_cast_fp16 = einsum(equation = var_9380_equation_0, values = (var_9150_cast_fp16, var_8736_cast_fp16))[name = tensor("op_9380_cast_fp16")]; + tensor var_9381_to_fp16 = const()[name = tensor("op_9381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_849_cast_fp16 = mul(x = var_9380_cast_fp16, y = var_9381_to_fp16)[name = tensor("aw_chunk_849_cast_fp16")]; + tensor var_9384_equation_0 = const()[name = tensor("op_9384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9384_cast_fp16 = einsum(equation = var_9384_equation_0, values = (var_9150_cast_fp16, var_8743_cast_fp16))[name = tensor("op_9384_cast_fp16")]; + tensor var_9385_to_fp16 = const()[name = tensor("op_9385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_851_cast_fp16 = mul(x = var_9384_cast_fp16, y = var_9385_to_fp16)[name = tensor("aw_chunk_851_cast_fp16")]; + tensor var_9388_equation_0 = const()[name = tensor("op_9388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9388_cast_fp16 = einsum(equation = var_9388_equation_0, values = (var_9150_cast_fp16, var_8750_cast_fp16))[name = tensor("op_9388_cast_fp16")]; + tensor var_9389_to_fp16 = const()[name = tensor("op_9389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_853_cast_fp16 = mul(x = var_9388_cast_fp16, y = var_9389_to_fp16)[name = tensor("aw_chunk_853_cast_fp16")]; + tensor var_9392_equation_0 = const()[name = tensor("op_9392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9392_cast_fp16 = einsum(equation = var_9392_equation_0, values = (var_9150_cast_fp16, var_8757_cast_fp16))[name = tensor("op_9392_cast_fp16")]; + tensor var_9393_to_fp16 = const()[name = tensor("op_9393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_855_cast_fp16 = mul(x = var_9392_cast_fp16, y = var_9393_to_fp16)[name = tensor("aw_chunk_855_cast_fp16")]; + tensor var_9396_equation_0 = const()[name = tensor("op_9396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9396_cast_fp16 = einsum(equation = var_9396_equation_0, values = (var_9154_cast_fp16, var_8764_cast_fp16))[name = tensor("op_9396_cast_fp16")]; + tensor var_9397_to_fp16 = const()[name = tensor("op_9397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_857_cast_fp16 = mul(x = var_9396_cast_fp16, y = var_9397_to_fp16)[name = tensor("aw_chunk_857_cast_fp16")]; + tensor var_9400_equation_0 = const()[name = tensor("op_9400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9400_cast_fp16 = einsum(equation = var_9400_equation_0, values = (var_9154_cast_fp16, var_8771_cast_fp16))[name = tensor("op_9400_cast_fp16")]; + tensor var_9401_to_fp16 = const()[name = tensor("op_9401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_859_cast_fp16 = mul(x = var_9400_cast_fp16, y = var_9401_to_fp16)[name = tensor("aw_chunk_859_cast_fp16")]; + tensor var_9404_equation_0 = const()[name = tensor("op_9404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9404_cast_fp16 = einsum(equation = var_9404_equation_0, values = (var_9154_cast_fp16, var_8778_cast_fp16))[name = tensor("op_9404_cast_fp16")]; + tensor var_9405_to_fp16 = const()[name = tensor("op_9405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_861_cast_fp16 = mul(x = var_9404_cast_fp16, y = var_9405_to_fp16)[name = tensor("aw_chunk_861_cast_fp16")]; + tensor var_9408_equation_0 = const()[name = tensor("op_9408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9408_cast_fp16 = einsum(equation = var_9408_equation_0, values = (var_9154_cast_fp16, var_8785_cast_fp16))[name = tensor("op_9408_cast_fp16")]; + tensor var_9409_to_fp16 = const()[name = tensor("op_9409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_863_cast_fp16 = mul(x = var_9408_cast_fp16, y = var_9409_to_fp16)[name = tensor("aw_chunk_863_cast_fp16")]; + tensor var_9412_equation_0 = const()[name = tensor("op_9412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9412_cast_fp16 = einsum(equation = var_9412_equation_0, values = (var_9158_cast_fp16, var_8792_cast_fp16))[name = tensor("op_9412_cast_fp16")]; + tensor var_9413_to_fp16 = const()[name = tensor("op_9413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_865_cast_fp16 = mul(x = var_9412_cast_fp16, y = var_9413_to_fp16)[name = tensor("aw_chunk_865_cast_fp16")]; + tensor var_9416_equation_0 = const()[name = tensor("op_9416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9416_cast_fp16 = einsum(equation = var_9416_equation_0, values = (var_9158_cast_fp16, var_8799_cast_fp16))[name = tensor("op_9416_cast_fp16")]; + tensor var_9417_to_fp16 = const()[name = tensor("op_9417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_867_cast_fp16 = mul(x = var_9416_cast_fp16, y = var_9417_to_fp16)[name = tensor("aw_chunk_867_cast_fp16")]; + tensor var_9420_equation_0 = const()[name = tensor("op_9420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9420_cast_fp16 = einsum(equation = var_9420_equation_0, values = (var_9158_cast_fp16, var_8806_cast_fp16))[name = tensor("op_9420_cast_fp16")]; + tensor var_9421_to_fp16 = const()[name = tensor("op_9421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_869_cast_fp16 = mul(x = var_9420_cast_fp16, y = var_9421_to_fp16)[name = tensor("aw_chunk_869_cast_fp16")]; + tensor var_9424_equation_0 = const()[name = tensor("op_9424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9424_cast_fp16 = einsum(equation = var_9424_equation_0, values = (var_9158_cast_fp16, var_8813_cast_fp16))[name = tensor("op_9424_cast_fp16")]; + tensor var_9425_to_fp16 = const()[name = tensor("op_9425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_871_cast_fp16 = mul(x = var_9424_cast_fp16, y = var_9425_to_fp16)[name = tensor("aw_chunk_871_cast_fp16")]; + tensor var_9428_equation_0 = const()[name = tensor("op_9428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9428_cast_fp16 = einsum(equation = var_9428_equation_0, values = (var_9162_cast_fp16, var_8820_cast_fp16))[name = tensor("op_9428_cast_fp16")]; + tensor var_9429_to_fp16 = const()[name = tensor("op_9429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_873_cast_fp16 = mul(x = var_9428_cast_fp16, y = var_9429_to_fp16)[name = tensor("aw_chunk_873_cast_fp16")]; + tensor var_9432_equation_0 = const()[name = tensor("op_9432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9432_cast_fp16 = einsum(equation = var_9432_equation_0, values = (var_9162_cast_fp16, var_8827_cast_fp16))[name = tensor("op_9432_cast_fp16")]; + tensor var_9433_to_fp16 = const()[name = tensor("op_9433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_875_cast_fp16 = mul(x = var_9432_cast_fp16, y = var_9433_to_fp16)[name = tensor("aw_chunk_875_cast_fp16")]; + tensor var_9436_equation_0 = const()[name = tensor("op_9436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9436_cast_fp16 = einsum(equation = var_9436_equation_0, values = (var_9162_cast_fp16, var_8834_cast_fp16))[name = tensor("op_9436_cast_fp16")]; + tensor var_9437_to_fp16 = const()[name = tensor("op_9437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_877_cast_fp16 = mul(x = var_9436_cast_fp16, y = var_9437_to_fp16)[name = tensor("aw_chunk_877_cast_fp16")]; + tensor var_9440_equation_0 = const()[name = tensor("op_9440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9440_cast_fp16 = einsum(equation = var_9440_equation_0, values = (var_9162_cast_fp16, var_8841_cast_fp16))[name = tensor("op_9440_cast_fp16")]; + tensor var_9441_to_fp16 = const()[name = tensor("op_9441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_879_cast_fp16 = mul(x = var_9440_cast_fp16, y = var_9441_to_fp16)[name = tensor("aw_chunk_879_cast_fp16")]; + tensor var_9444_equation_0 = const()[name = tensor("op_9444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9444_cast_fp16 = einsum(equation = var_9444_equation_0, values = (var_9166_cast_fp16, var_8848_cast_fp16))[name = tensor("op_9444_cast_fp16")]; + tensor var_9445_to_fp16 = const()[name = tensor("op_9445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_881_cast_fp16 = mul(x = var_9444_cast_fp16, y = var_9445_to_fp16)[name = tensor("aw_chunk_881_cast_fp16")]; + tensor var_9448_equation_0 = const()[name = tensor("op_9448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9448_cast_fp16 = einsum(equation = var_9448_equation_0, values = (var_9166_cast_fp16, var_8855_cast_fp16))[name = tensor("op_9448_cast_fp16")]; + tensor var_9449_to_fp16 = const()[name = tensor("op_9449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_883_cast_fp16 = mul(x = var_9448_cast_fp16, y = var_9449_to_fp16)[name = tensor("aw_chunk_883_cast_fp16")]; + tensor var_9452_equation_0 = const()[name = tensor("op_9452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9452_cast_fp16 = einsum(equation = var_9452_equation_0, values = (var_9166_cast_fp16, var_8862_cast_fp16))[name = tensor("op_9452_cast_fp16")]; + tensor var_9453_to_fp16 = const()[name = tensor("op_9453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_885_cast_fp16 = mul(x = var_9452_cast_fp16, y = var_9453_to_fp16)[name = tensor("aw_chunk_885_cast_fp16")]; + tensor var_9456_equation_0 = const()[name = tensor("op_9456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9456_cast_fp16 = einsum(equation = var_9456_equation_0, values = (var_9166_cast_fp16, var_8869_cast_fp16))[name = tensor("op_9456_cast_fp16")]; + tensor var_9457_to_fp16 = const()[name = tensor("op_9457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_887_cast_fp16 = mul(x = var_9456_cast_fp16, y = var_9457_to_fp16)[name = tensor("aw_chunk_887_cast_fp16")]; + tensor var_9460_equation_0 = const()[name = tensor("op_9460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9460_cast_fp16 = einsum(equation = var_9460_equation_0, values = (var_9170_cast_fp16, var_8876_cast_fp16))[name = tensor("op_9460_cast_fp16")]; + tensor var_9461_to_fp16 = const()[name = tensor("op_9461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_889_cast_fp16 = mul(x = var_9460_cast_fp16, y = var_9461_to_fp16)[name = tensor("aw_chunk_889_cast_fp16")]; + tensor var_9464_equation_0 = const()[name = tensor("op_9464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9464_cast_fp16 = einsum(equation = var_9464_equation_0, values = (var_9170_cast_fp16, var_8883_cast_fp16))[name = tensor("op_9464_cast_fp16")]; + tensor var_9465_to_fp16 = const()[name = tensor("op_9465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_891_cast_fp16 = mul(x = var_9464_cast_fp16, y = var_9465_to_fp16)[name = tensor("aw_chunk_891_cast_fp16")]; + tensor var_9468_equation_0 = const()[name = tensor("op_9468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9468_cast_fp16 = einsum(equation = var_9468_equation_0, values = (var_9170_cast_fp16, var_8890_cast_fp16))[name = tensor("op_9468_cast_fp16")]; + tensor var_9469_to_fp16 = const()[name = tensor("op_9469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_893_cast_fp16 = mul(x = var_9468_cast_fp16, y = var_9469_to_fp16)[name = tensor("aw_chunk_893_cast_fp16")]; + tensor var_9472_equation_0 = const()[name = tensor("op_9472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9472_cast_fp16 = einsum(equation = var_9472_equation_0, values = (var_9170_cast_fp16, var_8897_cast_fp16))[name = tensor("op_9472_cast_fp16")]; + tensor var_9473_to_fp16 = const()[name = tensor("op_9473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_895_cast_fp16 = mul(x = var_9472_cast_fp16, y = var_9473_to_fp16)[name = tensor("aw_chunk_895_cast_fp16")]; + tensor var_9476_equation_0 = const()[name = tensor("op_9476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9476_cast_fp16 = einsum(equation = var_9476_equation_0, values = (var_9174_cast_fp16, var_8904_cast_fp16))[name = tensor("op_9476_cast_fp16")]; + tensor var_9477_to_fp16 = const()[name = tensor("op_9477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_897_cast_fp16 = mul(x = var_9476_cast_fp16, y = var_9477_to_fp16)[name = tensor("aw_chunk_897_cast_fp16")]; + tensor var_9480_equation_0 = const()[name = tensor("op_9480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9480_cast_fp16 = einsum(equation = var_9480_equation_0, values = (var_9174_cast_fp16, var_8911_cast_fp16))[name = tensor("op_9480_cast_fp16")]; + tensor var_9481_to_fp16 = const()[name = tensor("op_9481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_899_cast_fp16 = mul(x = var_9480_cast_fp16, y = var_9481_to_fp16)[name = tensor("aw_chunk_899_cast_fp16")]; + tensor var_9484_equation_0 = const()[name = tensor("op_9484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9484_cast_fp16 = einsum(equation = var_9484_equation_0, values = (var_9174_cast_fp16, var_8918_cast_fp16))[name = tensor("op_9484_cast_fp16")]; + tensor var_9485_to_fp16 = const()[name = tensor("op_9485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_901_cast_fp16 = mul(x = var_9484_cast_fp16, y = var_9485_to_fp16)[name = tensor("aw_chunk_901_cast_fp16")]; + tensor var_9488_equation_0 = const()[name = tensor("op_9488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9488_cast_fp16 = einsum(equation = var_9488_equation_0, values = (var_9174_cast_fp16, var_8925_cast_fp16))[name = tensor("op_9488_cast_fp16")]; + tensor var_9489_to_fp16 = const()[name = tensor("op_9489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_903_cast_fp16 = mul(x = var_9488_cast_fp16, y = var_9489_to_fp16)[name = tensor("aw_chunk_903_cast_fp16")]; + tensor var_9492_equation_0 = const()[name = tensor("op_9492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9492_cast_fp16 = einsum(equation = var_9492_equation_0, values = (var_9178_cast_fp16, var_8932_cast_fp16))[name = tensor("op_9492_cast_fp16")]; + tensor var_9493_to_fp16 = const()[name = tensor("op_9493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_905_cast_fp16 = mul(x = var_9492_cast_fp16, y = var_9493_to_fp16)[name = tensor("aw_chunk_905_cast_fp16")]; + tensor var_9496_equation_0 = const()[name = tensor("op_9496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9496_cast_fp16 = einsum(equation = var_9496_equation_0, values = (var_9178_cast_fp16, var_8939_cast_fp16))[name = tensor("op_9496_cast_fp16")]; + tensor var_9497_to_fp16 = const()[name = tensor("op_9497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_907_cast_fp16 = mul(x = var_9496_cast_fp16, y = var_9497_to_fp16)[name = tensor("aw_chunk_907_cast_fp16")]; + tensor var_9500_equation_0 = const()[name = tensor("op_9500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9500_cast_fp16 = einsum(equation = var_9500_equation_0, values = (var_9178_cast_fp16, var_8946_cast_fp16))[name = tensor("op_9500_cast_fp16")]; + tensor var_9501_to_fp16 = const()[name = tensor("op_9501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_909_cast_fp16 = mul(x = var_9500_cast_fp16, y = var_9501_to_fp16)[name = tensor("aw_chunk_909_cast_fp16")]; + tensor var_9504_equation_0 = const()[name = tensor("op_9504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9504_cast_fp16 = einsum(equation = var_9504_equation_0, values = (var_9178_cast_fp16, var_8953_cast_fp16))[name = tensor("op_9504_cast_fp16")]; + tensor var_9505_to_fp16 = const()[name = tensor("op_9505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_911_cast_fp16 = mul(x = var_9504_cast_fp16, y = var_9505_to_fp16)[name = tensor("aw_chunk_911_cast_fp16")]; + tensor var_9508_equation_0 = const()[name = tensor("op_9508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9508_cast_fp16 = einsum(equation = var_9508_equation_0, values = (var_9182_cast_fp16, var_8960_cast_fp16))[name = tensor("op_9508_cast_fp16")]; + tensor var_9509_to_fp16 = const()[name = tensor("op_9509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_913_cast_fp16 = mul(x = var_9508_cast_fp16, y = var_9509_to_fp16)[name = tensor("aw_chunk_913_cast_fp16")]; + tensor var_9512_equation_0 = const()[name = tensor("op_9512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9512_cast_fp16 = einsum(equation = var_9512_equation_0, values = (var_9182_cast_fp16, var_8967_cast_fp16))[name = tensor("op_9512_cast_fp16")]; + tensor var_9513_to_fp16 = const()[name = tensor("op_9513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_915_cast_fp16 = mul(x = var_9512_cast_fp16, y = var_9513_to_fp16)[name = tensor("aw_chunk_915_cast_fp16")]; + tensor var_9516_equation_0 = const()[name = tensor("op_9516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9516_cast_fp16 = einsum(equation = var_9516_equation_0, values = (var_9182_cast_fp16, var_8974_cast_fp16))[name = tensor("op_9516_cast_fp16")]; + tensor var_9517_to_fp16 = const()[name = tensor("op_9517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_917_cast_fp16 = mul(x = var_9516_cast_fp16, y = var_9517_to_fp16)[name = tensor("aw_chunk_917_cast_fp16")]; + tensor var_9520_equation_0 = const()[name = tensor("op_9520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9520_cast_fp16 = einsum(equation = var_9520_equation_0, values = (var_9182_cast_fp16, var_8981_cast_fp16))[name = tensor("op_9520_cast_fp16")]; + tensor var_9521_to_fp16 = const()[name = tensor("op_9521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_919_cast_fp16 = mul(x = var_9520_cast_fp16, y = var_9521_to_fp16)[name = tensor("aw_chunk_919_cast_fp16")]; + tensor var_9524_equation_0 = const()[name = tensor("op_9524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9524_cast_fp16 = einsum(equation = var_9524_equation_0, values = (var_9186_cast_fp16, var_8988_cast_fp16))[name = tensor("op_9524_cast_fp16")]; + tensor var_9525_to_fp16 = const()[name = tensor("op_9525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_921_cast_fp16 = mul(x = var_9524_cast_fp16, y = var_9525_to_fp16)[name = tensor("aw_chunk_921_cast_fp16")]; + tensor var_9528_equation_0 = const()[name = tensor("op_9528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9528_cast_fp16 = einsum(equation = var_9528_equation_0, values = (var_9186_cast_fp16, var_8995_cast_fp16))[name = tensor("op_9528_cast_fp16")]; + tensor var_9529_to_fp16 = const()[name = tensor("op_9529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_923_cast_fp16 = mul(x = var_9528_cast_fp16, y = var_9529_to_fp16)[name = tensor("aw_chunk_923_cast_fp16")]; + tensor var_9532_equation_0 = const()[name = tensor("op_9532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9532_cast_fp16 = einsum(equation = var_9532_equation_0, values = (var_9186_cast_fp16, var_9002_cast_fp16))[name = tensor("op_9532_cast_fp16")]; + tensor var_9533_to_fp16 = const()[name = tensor("op_9533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_925_cast_fp16 = mul(x = var_9532_cast_fp16, y = var_9533_to_fp16)[name = tensor("aw_chunk_925_cast_fp16")]; + tensor var_9536_equation_0 = const()[name = tensor("op_9536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9536_cast_fp16 = einsum(equation = var_9536_equation_0, values = (var_9186_cast_fp16, var_9009_cast_fp16))[name = tensor("op_9536_cast_fp16")]; + tensor var_9537_to_fp16 = const()[name = tensor("op_9537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_927_cast_fp16 = mul(x = var_9536_cast_fp16, y = var_9537_to_fp16)[name = tensor("aw_chunk_927_cast_fp16")]; + tensor var_9540_equation_0 = const()[name = tensor("op_9540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9540_cast_fp16 = einsum(equation = var_9540_equation_0, values = (var_9190_cast_fp16, var_9016_cast_fp16))[name = tensor("op_9540_cast_fp16")]; + tensor var_9541_to_fp16 = const()[name = tensor("op_9541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_929_cast_fp16 = mul(x = var_9540_cast_fp16, y = var_9541_to_fp16)[name = tensor("aw_chunk_929_cast_fp16")]; + tensor var_9544_equation_0 = const()[name = tensor("op_9544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9544_cast_fp16 = einsum(equation = var_9544_equation_0, values = (var_9190_cast_fp16, var_9023_cast_fp16))[name = tensor("op_9544_cast_fp16")]; + tensor var_9545_to_fp16 = const()[name = tensor("op_9545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_931_cast_fp16 = mul(x = var_9544_cast_fp16, y = var_9545_to_fp16)[name = tensor("aw_chunk_931_cast_fp16")]; + tensor var_9548_equation_0 = const()[name = tensor("op_9548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9548_cast_fp16 = einsum(equation = var_9548_equation_0, values = (var_9190_cast_fp16, var_9030_cast_fp16))[name = tensor("op_9548_cast_fp16")]; + tensor var_9549_to_fp16 = const()[name = tensor("op_9549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_933_cast_fp16 = mul(x = var_9548_cast_fp16, y = var_9549_to_fp16)[name = tensor("aw_chunk_933_cast_fp16")]; + tensor var_9552_equation_0 = const()[name = tensor("op_9552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9552_cast_fp16 = einsum(equation = var_9552_equation_0, values = (var_9190_cast_fp16, var_9037_cast_fp16))[name = tensor("op_9552_cast_fp16")]; + tensor var_9553_to_fp16 = const()[name = tensor("op_9553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_935_cast_fp16 = mul(x = var_9552_cast_fp16, y = var_9553_to_fp16)[name = tensor("aw_chunk_935_cast_fp16")]; + tensor var_9556_equation_0 = const()[name = tensor("op_9556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9556_cast_fp16 = einsum(equation = var_9556_equation_0, values = (var_9194_cast_fp16, var_9044_cast_fp16))[name = tensor("op_9556_cast_fp16")]; + tensor var_9557_to_fp16 = const()[name = tensor("op_9557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_937_cast_fp16 = mul(x = var_9556_cast_fp16, y = var_9557_to_fp16)[name = tensor("aw_chunk_937_cast_fp16")]; + tensor var_9560_equation_0 = const()[name = tensor("op_9560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9560_cast_fp16 = einsum(equation = var_9560_equation_0, values = (var_9194_cast_fp16, var_9051_cast_fp16))[name = tensor("op_9560_cast_fp16")]; + tensor var_9561_to_fp16 = const()[name = tensor("op_9561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_939_cast_fp16 = mul(x = var_9560_cast_fp16, y = var_9561_to_fp16)[name = tensor("aw_chunk_939_cast_fp16")]; + tensor var_9564_equation_0 = const()[name = tensor("op_9564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9564_cast_fp16 = einsum(equation = var_9564_equation_0, values = (var_9194_cast_fp16, var_9058_cast_fp16))[name = tensor("op_9564_cast_fp16")]; + tensor var_9565_to_fp16 = const()[name = tensor("op_9565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_941_cast_fp16 = mul(x = var_9564_cast_fp16, y = var_9565_to_fp16)[name = tensor("aw_chunk_941_cast_fp16")]; + tensor var_9568_equation_0 = const()[name = tensor("op_9568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9568_cast_fp16 = einsum(equation = var_9568_equation_0, values = (var_9194_cast_fp16, var_9065_cast_fp16))[name = tensor("op_9568_cast_fp16")]; + tensor var_9569_to_fp16 = const()[name = tensor("op_9569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_943_cast_fp16 = mul(x = var_9568_cast_fp16, y = var_9569_to_fp16)[name = tensor("aw_chunk_943_cast_fp16")]; + tensor var_9572_equation_0 = const()[name = tensor("op_9572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9572_cast_fp16 = einsum(equation = var_9572_equation_0, values = (var_9198_cast_fp16, var_9072_cast_fp16))[name = tensor("op_9572_cast_fp16")]; + tensor var_9573_to_fp16 = const()[name = tensor("op_9573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_945_cast_fp16 = mul(x = var_9572_cast_fp16, y = var_9573_to_fp16)[name = tensor("aw_chunk_945_cast_fp16")]; + tensor var_9576_equation_0 = const()[name = tensor("op_9576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9576_cast_fp16 = einsum(equation = var_9576_equation_0, values = (var_9198_cast_fp16, var_9079_cast_fp16))[name = tensor("op_9576_cast_fp16")]; + tensor var_9577_to_fp16 = const()[name = tensor("op_9577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_947_cast_fp16 = mul(x = var_9576_cast_fp16, y = var_9577_to_fp16)[name = tensor("aw_chunk_947_cast_fp16")]; + tensor var_9580_equation_0 = const()[name = tensor("op_9580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9580_cast_fp16 = einsum(equation = var_9580_equation_0, values = (var_9198_cast_fp16, var_9086_cast_fp16))[name = tensor("op_9580_cast_fp16")]; + tensor var_9581_to_fp16 = const()[name = tensor("op_9581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_949_cast_fp16 = mul(x = var_9580_cast_fp16, y = var_9581_to_fp16)[name = tensor("aw_chunk_949_cast_fp16")]; + tensor var_9584_equation_0 = const()[name = tensor("op_9584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9584_cast_fp16 = einsum(equation = var_9584_equation_0, values = (var_9198_cast_fp16, var_9093_cast_fp16))[name = tensor("op_9584_cast_fp16")]; + tensor var_9585_to_fp16 = const()[name = tensor("op_9585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_951_cast_fp16 = mul(x = var_9584_cast_fp16, y = var_9585_to_fp16)[name = tensor("aw_chunk_951_cast_fp16")]; + tensor var_9588_equation_0 = const()[name = tensor("op_9588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9588_cast_fp16 = einsum(equation = var_9588_equation_0, values = (var_9202_cast_fp16, var_9100_cast_fp16))[name = tensor("op_9588_cast_fp16")]; + tensor var_9589_to_fp16 = const()[name = tensor("op_9589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_953_cast_fp16 = mul(x = var_9588_cast_fp16, y = var_9589_to_fp16)[name = tensor("aw_chunk_953_cast_fp16")]; + tensor var_9592_equation_0 = const()[name = tensor("op_9592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9592_cast_fp16 = einsum(equation = var_9592_equation_0, values = (var_9202_cast_fp16, var_9107_cast_fp16))[name = tensor("op_9592_cast_fp16")]; + tensor var_9593_to_fp16 = const()[name = tensor("op_9593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_955_cast_fp16 = mul(x = var_9592_cast_fp16, y = var_9593_to_fp16)[name = tensor("aw_chunk_955_cast_fp16")]; + tensor var_9596_equation_0 = const()[name = tensor("op_9596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9596_cast_fp16 = einsum(equation = var_9596_equation_0, values = (var_9202_cast_fp16, var_9114_cast_fp16))[name = tensor("op_9596_cast_fp16")]; + tensor var_9597_to_fp16 = const()[name = tensor("op_9597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_957_cast_fp16 = mul(x = var_9596_cast_fp16, y = var_9597_to_fp16)[name = tensor("aw_chunk_957_cast_fp16")]; + tensor var_9600_equation_0 = const()[name = tensor("op_9600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_9600_cast_fp16 = einsum(equation = var_9600_equation_0, values = (var_9202_cast_fp16, var_9121_cast_fp16))[name = tensor("op_9600_cast_fp16")]; + tensor var_9601_to_fp16 = const()[name = tensor("op_9601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_959_cast_fp16 = mul(x = var_9600_cast_fp16, y = var_9601_to_fp16)[name = tensor("aw_chunk_959_cast_fp16")]; + tensor var_9603_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_801_cast_fp16)[name = tensor("op_9603_cast_fp16")]; + tensor var_9604_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_803_cast_fp16)[name = tensor("op_9604_cast_fp16")]; + tensor var_9605_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_805_cast_fp16)[name = tensor("op_9605_cast_fp16")]; + tensor var_9606_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_807_cast_fp16)[name = tensor("op_9606_cast_fp16")]; + tensor var_9607_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_809_cast_fp16)[name = tensor("op_9607_cast_fp16")]; + tensor var_9608_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_811_cast_fp16)[name = tensor("op_9608_cast_fp16")]; + tensor var_9609_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_813_cast_fp16)[name = tensor("op_9609_cast_fp16")]; + tensor var_9610_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_815_cast_fp16)[name = tensor("op_9610_cast_fp16")]; + tensor var_9611_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_817_cast_fp16)[name = tensor("op_9611_cast_fp16")]; + tensor var_9612_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_819_cast_fp16)[name = tensor("op_9612_cast_fp16")]; + tensor var_9613_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_821_cast_fp16)[name = tensor("op_9613_cast_fp16")]; + tensor var_9614_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_823_cast_fp16)[name = tensor("op_9614_cast_fp16")]; + tensor var_9615_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_825_cast_fp16)[name = tensor("op_9615_cast_fp16")]; + tensor var_9616_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_827_cast_fp16)[name = tensor("op_9616_cast_fp16")]; + tensor var_9617_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_829_cast_fp16)[name = tensor("op_9617_cast_fp16")]; + tensor var_9618_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_831_cast_fp16)[name = tensor("op_9618_cast_fp16")]; + tensor var_9619_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_833_cast_fp16)[name = tensor("op_9619_cast_fp16")]; + tensor var_9620_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_835_cast_fp16)[name = tensor("op_9620_cast_fp16")]; + tensor var_9621_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_837_cast_fp16)[name = tensor("op_9621_cast_fp16")]; + tensor var_9622_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_839_cast_fp16)[name = tensor("op_9622_cast_fp16")]; + tensor var_9623_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_841_cast_fp16)[name = tensor("op_9623_cast_fp16")]; + tensor var_9624_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_843_cast_fp16)[name = tensor("op_9624_cast_fp16")]; + tensor var_9625_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_845_cast_fp16)[name = tensor("op_9625_cast_fp16")]; + tensor var_9626_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_847_cast_fp16)[name = tensor("op_9626_cast_fp16")]; + tensor var_9627_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_849_cast_fp16)[name = tensor("op_9627_cast_fp16")]; + tensor var_9628_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_851_cast_fp16)[name = tensor("op_9628_cast_fp16")]; + tensor var_9629_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_853_cast_fp16)[name = tensor("op_9629_cast_fp16")]; + tensor var_9630_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_855_cast_fp16)[name = tensor("op_9630_cast_fp16")]; + tensor var_9631_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_857_cast_fp16)[name = tensor("op_9631_cast_fp16")]; + tensor var_9632_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_859_cast_fp16)[name = tensor("op_9632_cast_fp16")]; + tensor var_9633_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_861_cast_fp16)[name = tensor("op_9633_cast_fp16")]; + tensor var_9634_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_863_cast_fp16)[name = tensor("op_9634_cast_fp16")]; + tensor var_9635_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_865_cast_fp16)[name = tensor("op_9635_cast_fp16")]; + tensor var_9636_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_867_cast_fp16)[name = tensor("op_9636_cast_fp16")]; + tensor var_9637_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_869_cast_fp16)[name = tensor("op_9637_cast_fp16")]; + tensor var_9638_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_871_cast_fp16)[name = tensor("op_9638_cast_fp16")]; + tensor var_9639_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_873_cast_fp16)[name = tensor("op_9639_cast_fp16")]; + tensor var_9640_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_875_cast_fp16)[name = tensor("op_9640_cast_fp16")]; + tensor var_9641_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_877_cast_fp16)[name = tensor("op_9641_cast_fp16")]; + tensor var_9642_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_879_cast_fp16)[name = tensor("op_9642_cast_fp16")]; + tensor var_9643_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_881_cast_fp16)[name = tensor("op_9643_cast_fp16")]; + tensor var_9644_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_883_cast_fp16)[name = tensor("op_9644_cast_fp16")]; + tensor var_9645_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_885_cast_fp16)[name = tensor("op_9645_cast_fp16")]; + tensor var_9646_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_887_cast_fp16)[name = tensor("op_9646_cast_fp16")]; + tensor var_9647_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_889_cast_fp16)[name = tensor("op_9647_cast_fp16")]; + tensor var_9648_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_891_cast_fp16)[name = tensor("op_9648_cast_fp16")]; + tensor var_9649_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_893_cast_fp16)[name = tensor("op_9649_cast_fp16")]; + tensor var_9650_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_895_cast_fp16)[name = tensor("op_9650_cast_fp16")]; + tensor var_9651_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_897_cast_fp16)[name = tensor("op_9651_cast_fp16")]; + tensor var_9652_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_899_cast_fp16)[name = tensor("op_9652_cast_fp16")]; + tensor var_9653_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_901_cast_fp16)[name = tensor("op_9653_cast_fp16")]; + tensor var_9654_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_903_cast_fp16)[name = tensor("op_9654_cast_fp16")]; + tensor var_9655_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_905_cast_fp16)[name = tensor("op_9655_cast_fp16")]; + tensor var_9656_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_907_cast_fp16)[name = tensor("op_9656_cast_fp16")]; + tensor var_9657_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_909_cast_fp16)[name = tensor("op_9657_cast_fp16")]; + tensor var_9658_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_911_cast_fp16)[name = tensor("op_9658_cast_fp16")]; + tensor var_9659_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_913_cast_fp16)[name = tensor("op_9659_cast_fp16")]; + tensor var_9660_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_915_cast_fp16)[name = tensor("op_9660_cast_fp16")]; + tensor var_9661_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_917_cast_fp16)[name = tensor("op_9661_cast_fp16")]; + tensor var_9662_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_919_cast_fp16)[name = tensor("op_9662_cast_fp16")]; + tensor var_9663_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_921_cast_fp16)[name = tensor("op_9663_cast_fp16")]; + tensor var_9664_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_923_cast_fp16)[name = tensor("op_9664_cast_fp16")]; + tensor var_9665_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_925_cast_fp16)[name = tensor("op_9665_cast_fp16")]; + tensor var_9666_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_927_cast_fp16)[name = tensor("op_9666_cast_fp16")]; + tensor var_9667_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_929_cast_fp16)[name = tensor("op_9667_cast_fp16")]; + tensor var_9668_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_931_cast_fp16)[name = tensor("op_9668_cast_fp16")]; + tensor var_9669_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_933_cast_fp16)[name = tensor("op_9669_cast_fp16")]; + tensor var_9670_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_935_cast_fp16)[name = tensor("op_9670_cast_fp16")]; + tensor var_9671_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_937_cast_fp16)[name = tensor("op_9671_cast_fp16")]; + tensor var_9672_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_939_cast_fp16)[name = tensor("op_9672_cast_fp16")]; + tensor var_9673_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_941_cast_fp16)[name = tensor("op_9673_cast_fp16")]; + tensor var_9674_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_943_cast_fp16)[name = tensor("op_9674_cast_fp16")]; + tensor var_9675_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_945_cast_fp16)[name = tensor("op_9675_cast_fp16")]; + tensor var_9676_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_947_cast_fp16)[name = tensor("op_9676_cast_fp16")]; + tensor var_9677_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_949_cast_fp16)[name = tensor("op_9677_cast_fp16")]; + tensor var_9678_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_951_cast_fp16)[name = tensor("op_9678_cast_fp16")]; + tensor var_9679_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_953_cast_fp16)[name = tensor("op_9679_cast_fp16")]; + tensor var_9680_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_955_cast_fp16)[name = tensor("op_9680_cast_fp16")]; + tensor var_9681_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_957_cast_fp16)[name = tensor("op_9681_cast_fp16")]; + tensor var_9682_cast_fp16 = softmax(axis = var_8374, x = aw_chunk_959_cast_fp16)[name = tensor("op_9682_cast_fp16")]; + tensor var_9684_equation_0 = const()[name = tensor("op_9684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9684_cast_fp16 = einsum(equation = var_9684_equation_0, values = (var_9204_cast_fp16, var_9603_cast_fp16))[name = tensor("op_9684_cast_fp16")]; + tensor var_9686_equation_0 = const()[name = tensor("op_9686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9686_cast_fp16 = einsum(equation = var_9686_equation_0, values = (var_9204_cast_fp16, var_9604_cast_fp16))[name = tensor("op_9686_cast_fp16")]; + tensor var_9688_equation_0 = const()[name = tensor("op_9688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9688_cast_fp16 = einsum(equation = var_9688_equation_0, values = (var_9204_cast_fp16, var_9605_cast_fp16))[name = tensor("op_9688_cast_fp16")]; + tensor var_9690_equation_0 = const()[name = tensor("op_9690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9690_cast_fp16 = einsum(equation = var_9690_equation_0, values = (var_9204_cast_fp16, var_9606_cast_fp16))[name = tensor("op_9690_cast_fp16")]; + tensor var_9692_equation_0 = const()[name = tensor("op_9692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9692_cast_fp16 = einsum(equation = var_9692_equation_0, values = (var_9208_cast_fp16, var_9607_cast_fp16))[name = tensor("op_9692_cast_fp16")]; + tensor var_9694_equation_0 = const()[name = tensor("op_9694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9694_cast_fp16 = einsum(equation = var_9694_equation_0, values = (var_9208_cast_fp16, var_9608_cast_fp16))[name = tensor("op_9694_cast_fp16")]; + tensor var_9696_equation_0 = const()[name = tensor("op_9696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9696_cast_fp16 = einsum(equation = var_9696_equation_0, values = (var_9208_cast_fp16, var_9609_cast_fp16))[name = tensor("op_9696_cast_fp16")]; + tensor var_9698_equation_0 = const()[name = tensor("op_9698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9698_cast_fp16 = einsum(equation = var_9698_equation_0, values = (var_9208_cast_fp16, var_9610_cast_fp16))[name = tensor("op_9698_cast_fp16")]; + tensor var_9700_equation_0 = const()[name = tensor("op_9700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9700_cast_fp16 = einsum(equation = var_9700_equation_0, values = (var_9212_cast_fp16, var_9611_cast_fp16))[name = tensor("op_9700_cast_fp16")]; + tensor var_9702_equation_0 = const()[name = tensor("op_9702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9702_cast_fp16 = einsum(equation = var_9702_equation_0, values = (var_9212_cast_fp16, var_9612_cast_fp16))[name = tensor("op_9702_cast_fp16")]; + tensor var_9704_equation_0 = const()[name = tensor("op_9704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9704_cast_fp16 = einsum(equation = var_9704_equation_0, values = (var_9212_cast_fp16, var_9613_cast_fp16))[name = tensor("op_9704_cast_fp16")]; + tensor var_9706_equation_0 = const()[name = tensor("op_9706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9706_cast_fp16 = einsum(equation = var_9706_equation_0, values = (var_9212_cast_fp16, var_9614_cast_fp16))[name = tensor("op_9706_cast_fp16")]; + tensor var_9708_equation_0 = const()[name = tensor("op_9708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9708_cast_fp16 = einsum(equation = var_9708_equation_0, values = (var_9216_cast_fp16, var_9615_cast_fp16))[name = tensor("op_9708_cast_fp16")]; + tensor var_9710_equation_0 = const()[name = tensor("op_9710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9710_cast_fp16 = einsum(equation = var_9710_equation_0, values = (var_9216_cast_fp16, var_9616_cast_fp16))[name = tensor("op_9710_cast_fp16")]; + tensor var_9712_equation_0 = const()[name = tensor("op_9712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9712_cast_fp16 = einsum(equation = var_9712_equation_0, values = (var_9216_cast_fp16, var_9617_cast_fp16))[name = tensor("op_9712_cast_fp16")]; + tensor var_9714_equation_0 = const()[name = tensor("op_9714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9714_cast_fp16 = einsum(equation = var_9714_equation_0, values = (var_9216_cast_fp16, var_9618_cast_fp16))[name = tensor("op_9714_cast_fp16")]; + tensor var_9716_equation_0 = const()[name = tensor("op_9716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9716_cast_fp16 = einsum(equation = var_9716_equation_0, values = (var_9220_cast_fp16, var_9619_cast_fp16))[name = tensor("op_9716_cast_fp16")]; + tensor var_9718_equation_0 = const()[name = tensor("op_9718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9718_cast_fp16 = einsum(equation = var_9718_equation_0, values = (var_9220_cast_fp16, var_9620_cast_fp16))[name = tensor("op_9718_cast_fp16")]; + tensor var_9720_equation_0 = const()[name = tensor("op_9720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9720_cast_fp16 = einsum(equation = var_9720_equation_0, values = (var_9220_cast_fp16, var_9621_cast_fp16))[name = tensor("op_9720_cast_fp16")]; + tensor var_9722_equation_0 = const()[name = tensor("op_9722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9722_cast_fp16 = einsum(equation = var_9722_equation_0, values = (var_9220_cast_fp16, var_9622_cast_fp16))[name = tensor("op_9722_cast_fp16")]; + tensor var_9724_equation_0 = const()[name = tensor("op_9724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9724_cast_fp16 = einsum(equation = var_9724_equation_0, values = (var_9224_cast_fp16, var_9623_cast_fp16))[name = tensor("op_9724_cast_fp16")]; + tensor var_9726_equation_0 = const()[name = tensor("op_9726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9726_cast_fp16 = einsum(equation = var_9726_equation_0, values = (var_9224_cast_fp16, var_9624_cast_fp16))[name = tensor("op_9726_cast_fp16")]; + tensor var_9728_equation_0 = const()[name = tensor("op_9728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9728_cast_fp16 = einsum(equation = var_9728_equation_0, values = (var_9224_cast_fp16, var_9625_cast_fp16))[name = tensor("op_9728_cast_fp16")]; + tensor var_9730_equation_0 = const()[name = tensor("op_9730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9730_cast_fp16 = einsum(equation = var_9730_equation_0, values = (var_9224_cast_fp16, var_9626_cast_fp16))[name = tensor("op_9730_cast_fp16")]; + tensor var_9732_equation_0 = const()[name = tensor("op_9732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9732_cast_fp16 = einsum(equation = var_9732_equation_0, values = (var_9228_cast_fp16, var_9627_cast_fp16))[name = tensor("op_9732_cast_fp16")]; + tensor var_9734_equation_0 = const()[name = tensor("op_9734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9734_cast_fp16 = einsum(equation = var_9734_equation_0, values = (var_9228_cast_fp16, var_9628_cast_fp16))[name = tensor("op_9734_cast_fp16")]; + tensor var_9736_equation_0 = const()[name = tensor("op_9736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9736_cast_fp16 = einsum(equation = var_9736_equation_0, values = (var_9228_cast_fp16, var_9629_cast_fp16))[name = tensor("op_9736_cast_fp16")]; + tensor var_9738_equation_0 = const()[name = tensor("op_9738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9738_cast_fp16 = einsum(equation = var_9738_equation_0, values = (var_9228_cast_fp16, var_9630_cast_fp16))[name = tensor("op_9738_cast_fp16")]; + tensor var_9740_equation_0 = const()[name = tensor("op_9740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9740_cast_fp16 = einsum(equation = var_9740_equation_0, values = (var_9232_cast_fp16, var_9631_cast_fp16))[name = tensor("op_9740_cast_fp16")]; + tensor var_9742_equation_0 = const()[name = tensor("op_9742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9742_cast_fp16 = einsum(equation = var_9742_equation_0, values = (var_9232_cast_fp16, var_9632_cast_fp16))[name = tensor("op_9742_cast_fp16")]; + tensor var_9744_equation_0 = const()[name = tensor("op_9744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9744_cast_fp16 = einsum(equation = var_9744_equation_0, values = (var_9232_cast_fp16, var_9633_cast_fp16))[name = tensor("op_9744_cast_fp16")]; + tensor var_9746_equation_0 = const()[name = tensor("op_9746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9746_cast_fp16 = einsum(equation = var_9746_equation_0, values = (var_9232_cast_fp16, var_9634_cast_fp16))[name = tensor("op_9746_cast_fp16")]; + tensor var_9748_equation_0 = const()[name = tensor("op_9748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9748_cast_fp16 = einsum(equation = var_9748_equation_0, values = (var_9236_cast_fp16, var_9635_cast_fp16))[name = tensor("op_9748_cast_fp16")]; + tensor var_9750_equation_0 = const()[name = tensor("op_9750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9750_cast_fp16 = einsum(equation = var_9750_equation_0, values = (var_9236_cast_fp16, var_9636_cast_fp16))[name = tensor("op_9750_cast_fp16")]; + tensor var_9752_equation_0 = const()[name = tensor("op_9752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9752_cast_fp16 = einsum(equation = var_9752_equation_0, values = (var_9236_cast_fp16, var_9637_cast_fp16))[name = tensor("op_9752_cast_fp16")]; + tensor var_9754_equation_0 = const()[name = tensor("op_9754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9754_cast_fp16 = einsum(equation = var_9754_equation_0, values = (var_9236_cast_fp16, var_9638_cast_fp16))[name = tensor("op_9754_cast_fp16")]; + tensor var_9756_equation_0 = const()[name = tensor("op_9756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9756_cast_fp16 = einsum(equation = var_9756_equation_0, values = (var_9240_cast_fp16, var_9639_cast_fp16))[name = tensor("op_9756_cast_fp16")]; + tensor var_9758_equation_0 = const()[name = tensor("op_9758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9758_cast_fp16 = einsum(equation = var_9758_equation_0, values = (var_9240_cast_fp16, var_9640_cast_fp16))[name = tensor("op_9758_cast_fp16")]; + tensor var_9760_equation_0 = const()[name = tensor("op_9760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9760_cast_fp16 = einsum(equation = var_9760_equation_0, values = (var_9240_cast_fp16, var_9641_cast_fp16))[name = tensor("op_9760_cast_fp16")]; + tensor var_9762_equation_0 = const()[name = tensor("op_9762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9762_cast_fp16 = einsum(equation = var_9762_equation_0, values = (var_9240_cast_fp16, var_9642_cast_fp16))[name = tensor("op_9762_cast_fp16")]; + tensor var_9764_equation_0 = const()[name = tensor("op_9764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9764_cast_fp16 = einsum(equation = var_9764_equation_0, values = (var_9244_cast_fp16, var_9643_cast_fp16))[name = tensor("op_9764_cast_fp16")]; + tensor var_9766_equation_0 = const()[name = tensor("op_9766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9766_cast_fp16 = einsum(equation = var_9766_equation_0, values = (var_9244_cast_fp16, var_9644_cast_fp16))[name = tensor("op_9766_cast_fp16")]; + tensor var_9768_equation_0 = const()[name = tensor("op_9768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9768_cast_fp16 = einsum(equation = var_9768_equation_0, values = (var_9244_cast_fp16, var_9645_cast_fp16))[name = tensor("op_9768_cast_fp16")]; + tensor var_9770_equation_0 = const()[name = tensor("op_9770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9770_cast_fp16 = einsum(equation = var_9770_equation_0, values = (var_9244_cast_fp16, var_9646_cast_fp16))[name = tensor("op_9770_cast_fp16")]; + tensor var_9772_equation_0 = const()[name = tensor("op_9772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9772_cast_fp16 = einsum(equation = var_9772_equation_0, values = (var_9248_cast_fp16, var_9647_cast_fp16))[name = tensor("op_9772_cast_fp16")]; + tensor var_9774_equation_0 = const()[name = tensor("op_9774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9774_cast_fp16 = einsum(equation = var_9774_equation_0, values = (var_9248_cast_fp16, var_9648_cast_fp16))[name = tensor("op_9774_cast_fp16")]; + tensor var_9776_equation_0 = const()[name = tensor("op_9776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9776_cast_fp16 = einsum(equation = var_9776_equation_0, values = (var_9248_cast_fp16, var_9649_cast_fp16))[name = tensor("op_9776_cast_fp16")]; + tensor var_9778_equation_0 = const()[name = tensor("op_9778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9778_cast_fp16 = einsum(equation = var_9778_equation_0, values = (var_9248_cast_fp16, var_9650_cast_fp16))[name = tensor("op_9778_cast_fp16")]; + tensor var_9780_equation_0 = const()[name = tensor("op_9780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9780_cast_fp16 = einsum(equation = var_9780_equation_0, values = (var_9252_cast_fp16, var_9651_cast_fp16))[name = tensor("op_9780_cast_fp16")]; + tensor var_9782_equation_0 = const()[name = tensor("op_9782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9782_cast_fp16 = einsum(equation = var_9782_equation_0, values = (var_9252_cast_fp16, var_9652_cast_fp16))[name = tensor("op_9782_cast_fp16")]; + tensor var_9784_equation_0 = const()[name = tensor("op_9784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9784_cast_fp16 = einsum(equation = var_9784_equation_0, values = (var_9252_cast_fp16, var_9653_cast_fp16))[name = tensor("op_9784_cast_fp16")]; + tensor var_9786_equation_0 = const()[name = tensor("op_9786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9786_cast_fp16 = einsum(equation = var_9786_equation_0, values = (var_9252_cast_fp16, var_9654_cast_fp16))[name = tensor("op_9786_cast_fp16")]; + tensor var_9788_equation_0 = const()[name = tensor("op_9788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9788_cast_fp16 = einsum(equation = var_9788_equation_0, values = (var_9256_cast_fp16, var_9655_cast_fp16))[name = tensor("op_9788_cast_fp16")]; + tensor var_9790_equation_0 = const()[name = tensor("op_9790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9790_cast_fp16 = einsum(equation = var_9790_equation_0, values = (var_9256_cast_fp16, var_9656_cast_fp16))[name = tensor("op_9790_cast_fp16")]; + tensor var_9792_equation_0 = const()[name = tensor("op_9792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9792_cast_fp16 = einsum(equation = var_9792_equation_0, values = (var_9256_cast_fp16, var_9657_cast_fp16))[name = tensor("op_9792_cast_fp16")]; + tensor var_9794_equation_0 = const()[name = tensor("op_9794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9794_cast_fp16 = einsum(equation = var_9794_equation_0, values = (var_9256_cast_fp16, var_9658_cast_fp16))[name = tensor("op_9794_cast_fp16")]; + tensor var_9796_equation_0 = const()[name = tensor("op_9796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9796_cast_fp16 = einsum(equation = var_9796_equation_0, values = (var_9260_cast_fp16, var_9659_cast_fp16))[name = tensor("op_9796_cast_fp16")]; + tensor var_9798_equation_0 = const()[name = tensor("op_9798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9798_cast_fp16 = einsum(equation = var_9798_equation_0, values = (var_9260_cast_fp16, var_9660_cast_fp16))[name = tensor("op_9798_cast_fp16")]; + tensor var_9800_equation_0 = const()[name = tensor("op_9800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9800_cast_fp16 = einsum(equation = var_9800_equation_0, values = (var_9260_cast_fp16, var_9661_cast_fp16))[name = tensor("op_9800_cast_fp16")]; + tensor var_9802_equation_0 = const()[name = tensor("op_9802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9802_cast_fp16 = einsum(equation = var_9802_equation_0, values = (var_9260_cast_fp16, var_9662_cast_fp16))[name = tensor("op_9802_cast_fp16")]; + tensor var_9804_equation_0 = const()[name = tensor("op_9804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9804_cast_fp16 = einsum(equation = var_9804_equation_0, values = (var_9264_cast_fp16, var_9663_cast_fp16))[name = tensor("op_9804_cast_fp16")]; + tensor var_9806_equation_0 = const()[name = tensor("op_9806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9806_cast_fp16 = einsum(equation = var_9806_equation_0, values = (var_9264_cast_fp16, var_9664_cast_fp16))[name = tensor("op_9806_cast_fp16")]; + tensor var_9808_equation_0 = const()[name = tensor("op_9808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9808_cast_fp16 = einsum(equation = var_9808_equation_0, values = (var_9264_cast_fp16, var_9665_cast_fp16))[name = tensor("op_9808_cast_fp16")]; + tensor var_9810_equation_0 = const()[name = tensor("op_9810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9810_cast_fp16 = einsum(equation = var_9810_equation_0, values = (var_9264_cast_fp16, var_9666_cast_fp16))[name = tensor("op_9810_cast_fp16")]; + tensor var_9812_equation_0 = const()[name = tensor("op_9812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9812_cast_fp16 = einsum(equation = var_9812_equation_0, values = (var_9268_cast_fp16, var_9667_cast_fp16))[name = tensor("op_9812_cast_fp16")]; + tensor var_9814_equation_0 = const()[name = tensor("op_9814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9814_cast_fp16 = einsum(equation = var_9814_equation_0, values = (var_9268_cast_fp16, var_9668_cast_fp16))[name = tensor("op_9814_cast_fp16")]; + tensor var_9816_equation_0 = const()[name = tensor("op_9816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9816_cast_fp16 = einsum(equation = var_9816_equation_0, values = (var_9268_cast_fp16, var_9669_cast_fp16))[name = tensor("op_9816_cast_fp16")]; + tensor var_9818_equation_0 = const()[name = tensor("op_9818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9818_cast_fp16 = einsum(equation = var_9818_equation_0, values = (var_9268_cast_fp16, var_9670_cast_fp16))[name = tensor("op_9818_cast_fp16")]; + tensor var_9820_equation_0 = const()[name = tensor("op_9820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9820_cast_fp16 = einsum(equation = var_9820_equation_0, values = (var_9272_cast_fp16, var_9671_cast_fp16))[name = tensor("op_9820_cast_fp16")]; + tensor var_9822_equation_0 = const()[name = tensor("op_9822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9822_cast_fp16 = einsum(equation = var_9822_equation_0, values = (var_9272_cast_fp16, var_9672_cast_fp16))[name = tensor("op_9822_cast_fp16")]; + tensor var_9824_equation_0 = const()[name = tensor("op_9824_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9824_cast_fp16 = einsum(equation = var_9824_equation_0, values = (var_9272_cast_fp16, var_9673_cast_fp16))[name = tensor("op_9824_cast_fp16")]; + tensor var_9826_equation_0 = const()[name = tensor("op_9826_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9826_cast_fp16 = einsum(equation = var_9826_equation_0, values = (var_9272_cast_fp16, var_9674_cast_fp16))[name = tensor("op_9826_cast_fp16")]; + tensor var_9828_equation_0 = const()[name = tensor("op_9828_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9828_cast_fp16 = einsum(equation = var_9828_equation_0, values = (var_9276_cast_fp16, var_9675_cast_fp16))[name = tensor("op_9828_cast_fp16")]; + tensor var_9830_equation_0 = const()[name = tensor("op_9830_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9830_cast_fp16 = einsum(equation = var_9830_equation_0, values = (var_9276_cast_fp16, var_9676_cast_fp16))[name = tensor("op_9830_cast_fp16")]; + tensor var_9832_equation_0 = const()[name = tensor("op_9832_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9832_cast_fp16 = einsum(equation = var_9832_equation_0, values = (var_9276_cast_fp16, var_9677_cast_fp16))[name = tensor("op_9832_cast_fp16")]; + tensor var_9834_equation_0 = const()[name = tensor("op_9834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9834_cast_fp16 = einsum(equation = var_9834_equation_0, values = (var_9276_cast_fp16, var_9678_cast_fp16))[name = tensor("op_9834_cast_fp16")]; + tensor var_9836_equation_0 = const()[name = tensor("op_9836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9836_cast_fp16 = einsum(equation = var_9836_equation_0, values = (var_9280_cast_fp16, var_9679_cast_fp16))[name = tensor("op_9836_cast_fp16")]; + tensor var_9838_equation_0 = const()[name = tensor("op_9838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9838_cast_fp16 = einsum(equation = var_9838_equation_0, values = (var_9280_cast_fp16, var_9680_cast_fp16))[name = tensor("op_9838_cast_fp16")]; + tensor var_9840_equation_0 = const()[name = tensor("op_9840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9840_cast_fp16 = einsum(equation = var_9840_equation_0, values = (var_9280_cast_fp16, var_9681_cast_fp16))[name = tensor("op_9840_cast_fp16")]; + tensor var_9842_equation_0 = const()[name = tensor("op_9842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_9842_cast_fp16 = einsum(equation = var_9842_equation_0, values = (var_9280_cast_fp16, var_9682_cast_fp16))[name = tensor("op_9842_cast_fp16")]; + tensor var_9844_interleave_0 = const()[name = tensor("op_9844_interleave_0"), val = tensor(false)]; + tensor var_9844_cast_fp16 = concat(axis = var_8349, interleave = var_9844_interleave_0, values = (var_9684_cast_fp16, var_9686_cast_fp16, var_9688_cast_fp16, var_9690_cast_fp16))[name = tensor("op_9844_cast_fp16")]; + tensor var_9846_interleave_0 = const()[name = tensor("op_9846_interleave_0"), val = tensor(false)]; + tensor var_9846_cast_fp16 = concat(axis = var_8349, interleave = var_9846_interleave_0, values = (var_9692_cast_fp16, var_9694_cast_fp16, var_9696_cast_fp16, var_9698_cast_fp16))[name = tensor("op_9846_cast_fp16")]; + tensor var_9848_interleave_0 = const()[name = tensor("op_9848_interleave_0"), val = tensor(false)]; + tensor var_9848_cast_fp16 = concat(axis = var_8349, interleave = var_9848_interleave_0, values = (var_9700_cast_fp16, var_9702_cast_fp16, var_9704_cast_fp16, var_9706_cast_fp16))[name = tensor("op_9848_cast_fp16")]; + tensor var_9850_interleave_0 = const()[name = tensor("op_9850_interleave_0"), val = tensor(false)]; + tensor var_9850_cast_fp16 = concat(axis = var_8349, interleave = var_9850_interleave_0, values = (var_9708_cast_fp16, var_9710_cast_fp16, var_9712_cast_fp16, var_9714_cast_fp16))[name = tensor("op_9850_cast_fp16")]; + tensor var_9852_interleave_0 = const()[name = tensor("op_9852_interleave_0"), val = tensor(false)]; + tensor var_9852_cast_fp16 = concat(axis = var_8349, interleave = var_9852_interleave_0, values = (var_9716_cast_fp16, var_9718_cast_fp16, var_9720_cast_fp16, var_9722_cast_fp16))[name = tensor("op_9852_cast_fp16")]; + tensor var_9854_interleave_0 = const()[name = tensor("op_9854_interleave_0"), val = tensor(false)]; + tensor var_9854_cast_fp16 = concat(axis = var_8349, interleave = var_9854_interleave_0, values = (var_9724_cast_fp16, var_9726_cast_fp16, var_9728_cast_fp16, var_9730_cast_fp16))[name = tensor("op_9854_cast_fp16")]; + tensor var_9856_interleave_0 = const()[name = tensor("op_9856_interleave_0"), val = tensor(false)]; + tensor var_9856_cast_fp16 = concat(axis = var_8349, interleave = var_9856_interleave_0, values = (var_9732_cast_fp16, var_9734_cast_fp16, var_9736_cast_fp16, var_9738_cast_fp16))[name = tensor("op_9856_cast_fp16")]; + tensor var_9858_interleave_0 = const()[name = tensor("op_9858_interleave_0"), val = tensor(false)]; + tensor var_9858_cast_fp16 = concat(axis = var_8349, interleave = var_9858_interleave_0, values = (var_9740_cast_fp16, var_9742_cast_fp16, var_9744_cast_fp16, var_9746_cast_fp16))[name = tensor("op_9858_cast_fp16")]; + tensor var_9860_interleave_0 = const()[name = tensor("op_9860_interleave_0"), val = tensor(false)]; + tensor var_9860_cast_fp16 = concat(axis = var_8349, interleave = var_9860_interleave_0, values = (var_9748_cast_fp16, var_9750_cast_fp16, var_9752_cast_fp16, var_9754_cast_fp16))[name = tensor("op_9860_cast_fp16")]; + tensor var_9862_interleave_0 = const()[name = tensor("op_9862_interleave_0"), val = tensor(false)]; + tensor var_9862_cast_fp16 = concat(axis = var_8349, interleave = var_9862_interleave_0, values = (var_9756_cast_fp16, var_9758_cast_fp16, var_9760_cast_fp16, var_9762_cast_fp16))[name = tensor("op_9862_cast_fp16")]; + tensor var_9864_interleave_0 = const()[name = tensor("op_9864_interleave_0"), val = tensor(false)]; + tensor var_9864_cast_fp16 = concat(axis = var_8349, interleave = var_9864_interleave_0, values = (var_9764_cast_fp16, var_9766_cast_fp16, var_9768_cast_fp16, var_9770_cast_fp16))[name = tensor("op_9864_cast_fp16")]; + tensor var_9866_interleave_0 = const()[name = tensor("op_9866_interleave_0"), val = tensor(false)]; + tensor var_9866_cast_fp16 = concat(axis = var_8349, interleave = var_9866_interleave_0, values = (var_9772_cast_fp16, var_9774_cast_fp16, var_9776_cast_fp16, var_9778_cast_fp16))[name = tensor("op_9866_cast_fp16")]; + tensor var_9868_interleave_0 = const()[name = tensor("op_9868_interleave_0"), val = tensor(false)]; + tensor var_9868_cast_fp16 = concat(axis = var_8349, interleave = var_9868_interleave_0, values = (var_9780_cast_fp16, var_9782_cast_fp16, var_9784_cast_fp16, var_9786_cast_fp16))[name = tensor("op_9868_cast_fp16")]; + tensor var_9870_interleave_0 = const()[name = tensor("op_9870_interleave_0"), val = tensor(false)]; + tensor var_9870_cast_fp16 = concat(axis = var_8349, interleave = var_9870_interleave_0, values = (var_9788_cast_fp16, var_9790_cast_fp16, var_9792_cast_fp16, var_9794_cast_fp16))[name = tensor("op_9870_cast_fp16")]; + tensor var_9872_interleave_0 = const()[name = tensor("op_9872_interleave_0"), val = tensor(false)]; + tensor var_9872_cast_fp16 = concat(axis = var_8349, interleave = var_9872_interleave_0, values = (var_9796_cast_fp16, var_9798_cast_fp16, var_9800_cast_fp16, var_9802_cast_fp16))[name = tensor("op_9872_cast_fp16")]; + tensor var_9874_interleave_0 = const()[name = tensor("op_9874_interleave_0"), val = tensor(false)]; + tensor var_9874_cast_fp16 = concat(axis = var_8349, interleave = var_9874_interleave_0, values = (var_9804_cast_fp16, var_9806_cast_fp16, var_9808_cast_fp16, var_9810_cast_fp16))[name = tensor("op_9874_cast_fp16")]; + tensor var_9876_interleave_0 = const()[name = tensor("op_9876_interleave_0"), val = tensor(false)]; + tensor var_9876_cast_fp16 = concat(axis = var_8349, interleave = var_9876_interleave_0, values = (var_9812_cast_fp16, var_9814_cast_fp16, var_9816_cast_fp16, var_9818_cast_fp16))[name = tensor("op_9876_cast_fp16")]; + tensor var_9878_interleave_0 = const()[name = tensor("op_9878_interleave_0"), val = tensor(false)]; + tensor var_9878_cast_fp16 = concat(axis = var_8349, interleave = var_9878_interleave_0, values = (var_9820_cast_fp16, var_9822_cast_fp16, var_9824_cast_fp16, var_9826_cast_fp16))[name = tensor("op_9878_cast_fp16")]; + tensor var_9880_interleave_0 = const()[name = tensor("op_9880_interleave_0"), val = tensor(false)]; + tensor var_9880_cast_fp16 = concat(axis = var_8349, interleave = var_9880_interleave_0, values = (var_9828_cast_fp16, var_9830_cast_fp16, var_9832_cast_fp16, var_9834_cast_fp16))[name = tensor("op_9880_cast_fp16")]; + tensor var_9882_interleave_0 = const()[name = tensor("op_9882_interleave_0"), val = tensor(false)]; + tensor var_9882_cast_fp16 = concat(axis = var_8349, interleave = var_9882_interleave_0, values = (var_9836_cast_fp16, var_9838_cast_fp16, var_9840_cast_fp16, var_9842_cast_fp16))[name = tensor("op_9882_cast_fp16")]; + tensor input_107_interleave_0 = const()[name = tensor("input_107_interleave_0"), val = tensor(false)]; + tensor input_107_cast_fp16 = concat(axis = var_8374, interleave = input_107_interleave_0, values = (var_9844_cast_fp16, var_9846_cast_fp16, var_9848_cast_fp16, var_9850_cast_fp16, var_9852_cast_fp16, var_9854_cast_fp16, var_9856_cast_fp16, var_9858_cast_fp16, var_9860_cast_fp16, var_9862_cast_fp16, var_9864_cast_fp16, var_9866_cast_fp16, var_9868_cast_fp16, var_9870_cast_fp16, var_9872_cast_fp16, var_9874_cast_fp16, var_9876_cast_fp16, var_9878_cast_fp16, var_9880_cast_fp16, var_9882_cast_fp16))[name = tensor("input_107_cast_fp16")]; + tensor var_9890 = const()[name = tensor("op_9890"), val = tensor([1, 1])]; + tensor var_9892 = const()[name = tensor("op_9892"), val = tensor([1, 1])]; + tensor pretrained_out_67_pad_type_0 = const()[name = tensor("pretrained_out_67_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_67_pad_0 = const()[name = tensor("pretrained_out_67_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70025152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70844416))), name = tensor("layers_5_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_5_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70844544)))]; + tensor pretrained_out_67_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_9892, groups = var_8374, pad = pretrained_out_67_pad_0, pad_type = pretrained_out_67_pad_type_0, strides = var_9890, weight = layers_5_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_107_cast_fp16)[name = tensor("pretrained_out_67_cast_fp16")]; + tensor var_9896 = const()[name = tensor("op_9896"), val = tensor([1, 1])]; + tensor var_9898 = const()[name = tensor("op_9898"), val = tensor([1, 1])]; + tensor input_109_pad_type_0 = const()[name = tensor("input_109_pad_type_0"), val = tensor("custom")]; + tensor input_109_pad_0 = const()[name = tensor("input_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70847168)))]; + tensor input_109_cast_fp16 = conv(dilations = var_9898, groups = var_8374, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = var_9896, weight = layers_5_self_attn_o_proj_loraA_weight_to_fp16, x = input_107_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor var_9902 = const()[name = tensor("op_9902"), val = tensor([1, 1])]; + tensor var_9904 = const()[name = tensor("op_9904"), val = tensor([1, 1])]; + tensor lora_out_133_pad_type_0 = const()[name = tensor("lora_out_133_pad_type_0"), val = tensor("custom")]; + tensor lora_out_133_pad_0 = const()[name = tensor("lora_out_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_135_weight_0_to_fp16 = const()[name = tensor("lora_out_135_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70888192)))]; + tensor lora_out_135_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_9904, groups = var_8374, pad = lora_out_133_pad_0, pad_type = lora_out_133_pad_type_0, strides = var_9902, weight = lora_out_135_weight_0_to_fp16, x = input_109_cast_fp16)[name = tensor("lora_out_135_cast_fp16")]; + tensor obj_23_cast_fp16 = add(x = pretrained_out_67_cast_fp16, y = lora_out_135_cast_fp16)[name = tensor("obj_23_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor var_9913 = const()[name = tensor("op_9913"), val = tensor([1])]; + tensor channels_mean_23_cast_fp16 = reduce_mean(axes = var_9913, keep_dims = var_8375, x = inputs_23_cast_fp16)[name = tensor("channels_mean_23_cast_fp16")]; + tensor zero_mean_23_cast_fp16 = sub(x = inputs_23_cast_fp16, y = channels_mean_23_cast_fp16)[name = tensor("zero_mean_23_cast_fp16")]; + tensor zero_mean_sq_23_cast_fp16 = mul(x = zero_mean_23_cast_fp16, y = zero_mean_23_cast_fp16)[name = tensor("zero_mean_sq_23_cast_fp16")]; + tensor var_9917 = const()[name = tensor("op_9917"), val = tensor([1])]; + tensor var_9918_cast_fp16 = reduce_mean(axes = var_9917, keep_dims = var_8375, x = zero_mean_sq_23_cast_fp16)[name = tensor("op_9918_cast_fp16")]; + tensor var_9919_to_fp16 = const()[name = tensor("op_9919_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_9920_cast_fp16 = add(x = var_9918_cast_fp16, y = var_9919_to_fp16)[name = tensor("op_9920_cast_fp16")]; + tensor denom_23_epsilon_0 = const()[name = tensor("denom_23_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_23_cast_fp16 = rsqrt(epsilon = denom_23_epsilon_0, x = var_9920_cast_fp16)[name = tensor("denom_23_cast_fp16")]; + tensor out_23_cast_fp16 = mul(x = zero_mean_23_cast_fp16, y = denom_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; + tensor input_111_gamma_0_to_fp16 = const()[name = tensor("input_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70929216)))]; + tensor input_111_beta_0_to_fp16 = const()[name = tensor("input_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70931840)))]; + tensor input_111_epsilon_0_to_fp16 = const()[name = tensor("input_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_111_cast_fp16 = batch_norm(beta = input_111_beta_0_to_fp16, epsilon = input_111_epsilon_0_to_fp16, gamma = input_111_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor var_9934 = const()[name = tensor("op_9934"), val = tensor([1, 1])]; + tensor var_9936 = const()[name = tensor("op_9936"), val = tensor([1, 1])]; + tensor pretrained_out_69_pad_type_0 = const()[name = tensor("pretrained_out_69_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_69_pad_0 = const()[name = tensor("pretrained_out_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70934464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74211328))), name = tensor("layers_5_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_5_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_5_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74211456)))]; + tensor pretrained_out_69_cast_fp16 = conv(bias = layers_5_fc1_pretrained_bias_to_fp16, dilations = var_9936, groups = var_8374, pad = pretrained_out_69_pad_0, pad_type = pretrained_out_69_pad_type_0, strides = var_9934, weight = layers_5_fc1_pretrained_weight_to_fp16_palettized, x = input_111_cast_fp16)[name = tensor("pretrained_out_69_cast_fp16")]; + tensor var_9940 = const()[name = tensor("op_9940"), val = tensor([1, 1])]; + tensor var_9942 = const()[name = tensor("op_9942"), val = tensor([1, 1])]; + tensor input_113_pad_type_0 = const()[name = tensor("input_113_pad_type_0"), val = tensor("custom")]; + tensor input_113_pad_0 = const()[name = tensor("input_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_5_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74221760)))]; + tensor input_113_cast_fp16 = conv(dilations = var_9942, groups = var_8374, pad = input_113_pad_0, pad_type = input_113_pad_type_0, strides = var_9940, weight = layers_5_fc1_loraA_weight_to_fp16, x = input_111_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor var_9946 = const()[name = tensor("op_9946"), val = tensor([1, 1])]; + tensor var_9948 = const()[name = tensor("op_9948"), val = tensor([1, 1])]; + tensor lora_out_137_pad_type_0 = const()[name = tensor("lora_out_137_pad_type_0"), val = tensor("custom")]; + tensor lora_out_137_pad_0 = const()[name = tensor("lora_out_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_139_weight_0_to_fp16 = const()[name = tensor("lora_out_139_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74262784)))]; + tensor lora_out_139_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_9948, groups = var_8374, pad = lora_out_137_pad_0, pad_type = lora_out_137_pad_type_0, strides = var_9946, weight = lora_out_139_weight_0_to_fp16, x = input_113_cast_fp16)[name = tensor("lora_out_139_cast_fp16")]; + tensor input_115_cast_fp16 = add(x = pretrained_out_69_cast_fp16, y = lora_out_139_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor input_117_mode_0 = const()[name = tensor("input_117_mode_0"), val = tensor("EXACT")]; + tensor input_117_cast_fp16 = gelu(mode = input_117_mode_0, x = input_115_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor var_9960 = const()[name = tensor("op_9960"), val = tensor([1, 1])]; + tensor var_9962 = const()[name = tensor("op_9962"), val = tensor([1, 1])]; + tensor pretrained_out_71_pad_type_0 = const()[name = tensor("pretrained_out_71_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_71_pad_0 = const()[name = tensor("pretrained_out_71_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74426688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77703552))), name = tensor("layers_5_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_5_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_5_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77703680)))]; + tensor pretrained_out_71_cast_fp16 = conv(bias = layers_5_fc2_pretrained_bias_to_fp16, dilations = var_9962, groups = var_8374, pad = pretrained_out_71_pad_0, pad_type = pretrained_out_71_pad_type_0, strides = var_9960, weight = layers_5_fc2_pretrained_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = tensor("pretrained_out_71_cast_fp16")]; + tensor var_9966 = const()[name = tensor("op_9966"), val = tensor([1, 1])]; + tensor var_9968 = const()[name = tensor("op_9968"), val = tensor([1, 1])]; + tensor input_119_pad_type_0 = const()[name = tensor("input_119_pad_type_0"), val = tensor("custom")]; + tensor input_119_pad_0 = const()[name = tensor("input_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_5_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_5_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77706304)))]; + tensor input_119_cast_fp16 = conv(dilations = var_9968, groups = var_8374, pad = input_119_pad_0, pad_type = input_119_pad_type_0, strides = var_9966, weight = layers_5_fc2_loraA_weight_to_fp16, x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor var_9972 = const()[name = tensor("op_9972"), val = tensor([1, 1])]; + tensor var_9974 = const()[name = tensor("op_9974"), val = tensor([1, 1])]; + tensor lora_out_141_pad_type_0 = const()[name = tensor("lora_out_141_pad_type_0"), val = tensor("custom")]; + tensor lora_out_141_pad_0 = const()[name = tensor("lora_out_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_143_weight_0_to_fp16 = const()[name = tensor("lora_out_143_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77870208)))]; + tensor lora_out_143_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_9974, groups = var_8374, pad = lora_out_141_pad_0, pad_type = lora_out_141_pad_type_0, strides = var_9972, weight = lora_out_143_weight_0_to_fp16, x = input_119_cast_fp16)[name = tensor("lora_out_143_cast_fp16")]; + tensor hidden_states_15_cast_fp16 = add(x = pretrained_out_71_cast_fp16, y = lora_out_143_cast_fp16)[name = tensor("hidden_states_15_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor var_9984 = const()[name = tensor("op_9984"), val = tensor(3)]; + tensor var_10009 = const()[name = tensor("op_10009"), val = tensor(1)]; + tensor var_10010 = const()[name = tensor("op_10010"), val = tensor(true)]; + tensor var_10020 = const()[name = tensor("op_10020"), val = tensor([1])]; + tensor channels_mean_25_cast_fp16 = reduce_mean(axes = var_10020, keep_dims = var_10010, x = inputs_25_cast_fp16)[name = tensor("channels_mean_25_cast_fp16")]; + tensor zero_mean_25_cast_fp16 = sub(x = inputs_25_cast_fp16, y = channels_mean_25_cast_fp16)[name = tensor("zero_mean_25_cast_fp16")]; + tensor zero_mean_sq_25_cast_fp16 = mul(x = zero_mean_25_cast_fp16, y = zero_mean_25_cast_fp16)[name = tensor("zero_mean_sq_25_cast_fp16")]; + tensor var_10024 = const()[name = tensor("op_10024"), val = tensor([1])]; + tensor var_10025_cast_fp16 = reduce_mean(axes = var_10024, keep_dims = var_10010, x = zero_mean_sq_25_cast_fp16)[name = tensor("op_10025_cast_fp16")]; + tensor var_10026_to_fp16 = const()[name = tensor("op_10026_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_10027_cast_fp16 = add(x = var_10025_cast_fp16, y = var_10026_to_fp16)[name = tensor("op_10027_cast_fp16")]; + tensor denom_25_epsilon_0 = const()[name = tensor("denom_25_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_25_cast_fp16 = rsqrt(epsilon = denom_25_epsilon_0, x = var_10027_cast_fp16)[name = tensor("denom_25_cast_fp16")]; + tensor out_25_cast_fp16 = mul(x = zero_mean_25_cast_fp16, y = denom_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; + tensor obj_25_gamma_0_to_fp16 = const()[name = tensor("obj_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77911232)))]; + tensor obj_25_beta_0_to_fp16 = const()[name = tensor("obj_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77913856)))]; + tensor obj_25_epsilon_0_to_fp16 = const()[name = tensor("obj_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor("obj_25_cast_fp16")]; + tensor var_10045 = const()[name = tensor("op_10045"), val = tensor([1, 1])]; + tensor var_10047 = const()[name = tensor("op_10047"), val = tensor([1, 1])]; + tensor pretrained_out_73_pad_type_0 = const()[name = tensor("pretrained_out_73_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_73_pad_0 = const()[name = tensor("pretrained_out_73_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77916480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78735744))), name = tensor("layers_6_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_6_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78735872)))]; + tensor pretrained_out_73_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_10047, groups = var_10009, pad = pretrained_out_73_pad_0, pad_type = pretrained_out_73_pad_type_0, strides = var_10045, weight = layers_6_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor("pretrained_out_73_cast_fp16")]; + tensor var_10051 = const()[name = tensor("op_10051"), val = tensor([1, 1])]; + tensor var_10053 = const()[name = tensor("op_10053"), val = tensor([1, 1])]; + tensor input_121_pad_type_0 = const()[name = tensor("input_121_pad_type_0"), val = tensor("custom")]; + tensor input_121_pad_0 = const()[name = tensor("input_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78738496)))]; + tensor input_121_cast_fp16 = conv(dilations = var_10053, groups = var_10009, pad = input_121_pad_0, pad_type = input_121_pad_type_0, strides = var_10051, weight = layers_6_self_attn_q_proj_loraA_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor var_10057 = const()[name = tensor("op_10057"), val = tensor([1, 1])]; + tensor var_10059 = const()[name = tensor("op_10059"), val = tensor([1, 1])]; + tensor lora_out_145_pad_type_0 = const()[name = tensor("lora_out_145_pad_type_0"), val = tensor("custom")]; + tensor lora_out_145_pad_0 = const()[name = tensor("lora_out_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_147_weight_0_to_fp16 = const()[name = tensor("lora_out_147_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78779520)))]; + tensor lora_out_147_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_10059, groups = var_10009, pad = lora_out_145_pad_0, pad_type = lora_out_145_pad_type_0, strides = var_10057, weight = lora_out_147_weight_0_to_fp16, x = input_121_cast_fp16)[name = tensor("lora_out_147_cast_fp16")]; + tensor query_13_cast_fp16 = add(x = pretrained_out_73_cast_fp16, y = lora_out_147_cast_fp16)[name = tensor("query_13_cast_fp16")]; + tensor var_10069 = const()[name = tensor("op_10069"), val = tensor([1, 1])]; + tensor var_10071 = const()[name = tensor("op_10071"), val = tensor([1, 1])]; + tensor pretrained_out_75_pad_type_0 = const()[name = tensor("pretrained_out_75_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_75_pad_0 = const()[name = tensor("pretrained_out_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78820544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79639808))), name = tensor("layers_6_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_75_cast_fp16 = conv(dilations = var_10071, groups = var_10009, pad = pretrained_out_75_pad_0, pad_type = pretrained_out_75_pad_type_0, strides = var_10069, weight = layers_6_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor("pretrained_out_75_cast_fp16")]; + tensor var_10075 = const()[name = tensor("op_10075"), val = tensor([1, 1])]; + tensor var_10077 = const()[name = tensor("op_10077"), val = tensor([1, 1])]; + tensor input_123_pad_type_0 = const()[name = tensor("input_123_pad_type_0"), val = tensor("custom")]; + tensor input_123_pad_0 = const()[name = tensor("input_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79639936)))]; + tensor input_123_cast_fp16 = conv(dilations = var_10077, groups = var_10009, pad = input_123_pad_0, pad_type = input_123_pad_type_0, strides = var_10075, weight = layers_6_self_attn_k_proj_loraA_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor var_10081 = const()[name = tensor("op_10081"), val = tensor([1, 1])]; + tensor var_10083 = const()[name = tensor("op_10083"), val = tensor([1, 1])]; + tensor lora_out_149_pad_type_0 = const()[name = tensor("lora_out_149_pad_type_0"), val = tensor("custom")]; + tensor lora_out_149_pad_0 = const()[name = tensor("lora_out_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_151_weight_0_to_fp16 = const()[name = tensor("lora_out_151_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79680960)))]; + tensor lora_out_151_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_10083, groups = var_10009, pad = lora_out_149_pad_0, pad_type = lora_out_149_pad_type_0, strides = var_10081, weight = lora_out_151_weight_0_to_fp16, x = input_123_cast_fp16)[name = tensor("lora_out_151_cast_fp16")]; + tensor key_13_cast_fp16 = add(x = pretrained_out_75_cast_fp16, y = lora_out_151_cast_fp16)[name = tensor("key_13_cast_fp16")]; + tensor var_10094 = const()[name = tensor("op_10094"), val = tensor([1, 1])]; + tensor var_10096 = const()[name = tensor("op_10096"), val = tensor([1, 1])]; + tensor pretrained_out_77_pad_type_0 = const()[name = tensor("pretrained_out_77_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_77_pad_0 = const()[name = tensor("pretrained_out_77_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79721984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80541248))), name = tensor("layers_6_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_6_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80541376)))]; + tensor pretrained_out_77_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_10096, groups = var_10009, pad = pretrained_out_77_pad_0, pad_type = pretrained_out_77_pad_type_0, strides = var_10094, weight = layers_6_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor("pretrained_out_77_cast_fp16")]; + tensor var_10100 = const()[name = tensor("op_10100"), val = tensor([1, 1])]; + tensor var_10102 = const()[name = tensor("op_10102"), val = tensor([1, 1])]; + tensor input_125_pad_type_0 = const()[name = tensor("input_125_pad_type_0"), val = tensor("custom")]; + tensor input_125_pad_0 = const()[name = tensor("input_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80544000)))]; + tensor input_125_cast_fp16 = conv(dilations = var_10102, groups = var_10009, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = var_10100, weight = layers_6_self_attn_v_proj_loraA_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor var_10106 = const()[name = tensor("op_10106"), val = tensor([1, 1])]; + tensor var_10108 = const()[name = tensor("op_10108"), val = tensor([1, 1])]; + tensor lora_out_153_pad_type_0 = const()[name = tensor("lora_out_153_pad_type_0"), val = tensor("custom")]; + tensor lora_out_153_pad_0 = const()[name = tensor("lora_out_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_155_weight_0_to_fp16 = const()[name = tensor("lora_out_155_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80585024)))]; + tensor lora_out_155_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_10108, groups = var_10009, pad = lora_out_153_pad_0, pad_type = lora_out_153_pad_type_0, strides = var_10106, weight = lora_out_155_weight_0_to_fp16, x = input_125_cast_fp16)[name = tensor("lora_out_155_cast_fp16")]; + tensor value_13_cast_fp16 = add(x = pretrained_out_77_cast_fp16, y = lora_out_155_cast_fp16)[name = tensor("value_13_cast_fp16")]; + tensor var_10118_begin_0 = const()[name = tensor("op_10118_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10118_end_0 = const()[name = tensor("op_10118_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10118_end_mask_0 = const()[name = tensor("op_10118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10118_cast_fp16 = slice_by_index(begin = var_10118_begin_0, end = var_10118_end_0, end_mask = var_10118_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10118_cast_fp16")]; + tensor var_10122_begin_0 = const()[name = tensor("op_10122_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10122_end_0 = const()[name = tensor("op_10122_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10122_end_mask_0 = const()[name = tensor("op_10122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10122_cast_fp16 = slice_by_index(begin = var_10122_begin_0, end = var_10122_end_0, end_mask = var_10122_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10122_cast_fp16")]; + tensor var_10126_begin_0 = const()[name = tensor("op_10126_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10126_end_0 = const()[name = tensor("op_10126_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10126_end_mask_0 = const()[name = tensor("op_10126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10126_cast_fp16 = slice_by_index(begin = var_10126_begin_0, end = var_10126_end_0, end_mask = var_10126_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10126_cast_fp16")]; + tensor var_10130_begin_0 = const()[name = tensor("op_10130_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10130_end_0 = const()[name = tensor("op_10130_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10130_end_mask_0 = const()[name = tensor("op_10130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10130_cast_fp16 = slice_by_index(begin = var_10130_begin_0, end = var_10130_end_0, end_mask = var_10130_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10130_cast_fp16")]; + tensor var_10134_begin_0 = const()[name = tensor("op_10134_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10134_end_0 = const()[name = tensor("op_10134_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10134_end_mask_0 = const()[name = tensor("op_10134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10134_cast_fp16 = slice_by_index(begin = var_10134_begin_0, end = var_10134_end_0, end_mask = var_10134_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10134_cast_fp16")]; + tensor var_10138_begin_0 = const()[name = tensor("op_10138_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10138_end_0 = const()[name = tensor("op_10138_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10138_end_mask_0 = const()[name = tensor("op_10138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10138_cast_fp16 = slice_by_index(begin = var_10138_begin_0, end = var_10138_end_0, end_mask = var_10138_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10138_cast_fp16")]; + tensor var_10142_begin_0 = const()[name = tensor("op_10142_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10142_end_0 = const()[name = tensor("op_10142_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10142_end_mask_0 = const()[name = tensor("op_10142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10142_cast_fp16 = slice_by_index(begin = var_10142_begin_0, end = var_10142_end_0, end_mask = var_10142_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10142_cast_fp16")]; + tensor var_10146_begin_0 = const()[name = tensor("op_10146_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10146_end_0 = const()[name = tensor("op_10146_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10146_end_mask_0 = const()[name = tensor("op_10146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10146_cast_fp16 = slice_by_index(begin = var_10146_begin_0, end = var_10146_end_0, end_mask = var_10146_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10146_cast_fp16")]; + tensor var_10150_begin_0 = const()[name = tensor("op_10150_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10150_end_0 = const()[name = tensor("op_10150_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10150_end_mask_0 = const()[name = tensor("op_10150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10150_cast_fp16 = slice_by_index(begin = var_10150_begin_0, end = var_10150_end_0, end_mask = var_10150_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10150_cast_fp16")]; + tensor var_10154_begin_0 = const()[name = tensor("op_10154_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10154_end_0 = const()[name = tensor("op_10154_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10154_end_mask_0 = const()[name = tensor("op_10154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10154_cast_fp16 = slice_by_index(begin = var_10154_begin_0, end = var_10154_end_0, end_mask = var_10154_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10154_cast_fp16")]; + tensor var_10158_begin_0 = const()[name = tensor("op_10158_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10158_end_0 = const()[name = tensor("op_10158_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10158_end_mask_0 = const()[name = tensor("op_10158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10158_cast_fp16 = slice_by_index(begin = var_10158_begin_0, end = var_10158_end_0, end_mask = var_10158_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10158_cast_fp16")]; + tensor var_10162_begin_0 = const()[name = tensor("op_10162_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10162_end_0 = const()[name = tensor("op_10162_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10162_end_mask_0 = const()[name = tensor("op_10162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10162_cast_fp16 = slice_by_index(begin = var_10162_begin_0, end = var_10162_end_0, end_mask = var_10162_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10162_cast_fp16")]; + tensor var_10166_begin_0 = const()[name = tensor("op_10166_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10166_end_0 = const()[name = tensor("op_10166_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10166_end_mask_0 = const()[name = tensor("op_10166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10166_cast_fp16 = slice_by_index(begin = var_10166_begin_0, end = var_10166_end_0, end_mask = var_10166_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10166_cast_fp16")]; + tensor var_10170_begin_0 = const()[name = tensor("op_10170_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10170_end_0 = const()[name = tensor("op_10170_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_10170_end_mask_0 = const()[name = tensor("op_10170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10170_cast_fp16 = slice_by_index(begin = var_10170_begin_0, end = var_10170_end_0, end_mask = var_10170_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10170_cast_fp16")]; + tensor var_10174_begin_0 = const()[name = tensor("op_10174_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10174_end_0 = const()[name = tensor("op_10174_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_10174_end_mask_0 = const()[name = tensor("op_10174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10174_cast_fp16 = slice_by_index(begin = var_10174_begin_0, end = var_10174_end_0, end_mask = var_10174_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10174_cast_fp16")]; + tensor var_10178_begin_0 = const()[name = tensor("op_10178_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10178_end_0 = const()[name = tensor("op_10178_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_10178_end_mask_0 = const()[name = tensor("op_10178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10178_cast_fp16 = slice_by_index(begin = var_10178_begin_0, end = var_10178_end_0, end_mask = var_10178_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10178_cast_fp16")]; + tensor var_10182_begin_0 = const()[name = tensor("op_10182_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10182_end_0 = const()[name = tensor("op_10182_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_10182_end_mask_0 = const()[name = tensor("op_10182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10182_cast_fp16 = slice_by_index(begin = var_10182_begin_0, end = var_10182_end_0, end_mask = var_10182_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10182_cast_fp16")]; + tensor var_10186_begin_0 = const()[name = tensor("op_10186_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10186_end_0 = const()[name = tensor("op_10186_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_10186_end_mask_0 = const()[name = tensor("op_10186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10186_cast_fp16 = slice_by_index(begin = var_10186_begin_0, end = var_10186_end_0, end_mask = var_10186_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10186_cast_fp16")]; + tensor var_10190_begin_0 = const()[name = tensor("op_10190_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10190_end_0 = const()[name = tensor("op_10190_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_10190_end_mask_0 = const()[name = tensor("op_10190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10190_cast_fp16 = slice_by_index(begin = var_10190_begin_0, end = var_10190_end_0, end_mask = var_10190_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10190_cast_fp16")]; + tensor var_10194_begin_0 = const()[name = tensor("op_10194_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10194_end_0 = const()[name = tensor("op_10194_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_10194_end_mask_0 = const()[name = tensor("op_10194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10194_cast_fp16 = slice_by_index(begin = var_10194_begin_0, end = var_10194_end_0, end_mask = var_10194_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_10194_cast_fp16")]; + tensor var_10203_begin_0 = const()[name = tensor("op_10203_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10203_end_0 = const()[name = tensor("op_10203_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10203_end_mask_0 = const()[name = tensor("op_10203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10203_cast_fp16 = slice_by_index(begin = var_10203_begin_0, end = var_10203_end_0, end_mask = var_10203_end_mask_0, x = var_10118_cast_fp16)[name = tensor("op_10203_cast_fp16")]; + tensor var_10210_begin_0 = const()[name = tensor("op_10210_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10210_end_0 = const()[name = tensor("op_10210_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10210_end_mask_0 = const()[name = tensor("op_10210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10210_cast_fp16 = slice_by_index(begin = var_10210_begin_0, end = var_10210_end_0, end_mask = var_10210_end_mask_0, x = var_10118_cast_fp16)[name = tensor("op_10210_cast_fp16")]; + tensor var_10217_begin_0 = const()[name = tensor("op_10217_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10217_end_0 = const()[name = tensor("op_10217_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10217_end_mask_0 = const()[name = tensor("op_10217_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10217_cast_fp16 = slice_by_index(begin = var_10217_begin_0, end = var_10217_end_0, end_mask = var_10217_end_mask_0, x = var_10118_cast_fp16)[name = tensor("op_10217_cast_fp16")]; + tensor var_10224_begin_0 = const()[name = tensor("op_10224_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10224_end_0 = const()[name = tensor("op_10224_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10224_end_mask_0 = const()[name = tensor("op_10224_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10224_cast_fp16 = slice_by_index(begin = var_10224_begin_0, end = var_10224_end_0, end_mask = var_10224_end_mask_0, x = var_10118_cast_fp16)[name = tensor("op_10224_cast_fp16")]; + tensor var_10231_begin_0 = const()[name = tensor("op_10231_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10231_end_0 = const()[name = tensor("op_10231_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10231_end_mask_0 = const()[name = tensor("op_10231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10231_cast_fp16 = slice_by_index(begin = var_10231_begin_0, end = var_10231_end_0, end_mask = var_10231_end_mask_0, x = var_10122_cast_fp16)[name = tensor("op_10231_cast_fp16")]; + tensor var_10238_begin_0 = const()[name = tensor("op_10238_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10238_end_0 = const()[name = tensor("op_10238_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10238_end_mask_0 = const()[name = tensor("op_10238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10238_cast_fp16 = slice_by_index(begin = var_10238_begin_0, end = var_10238_end_0, end_mask = var_10238_end_mask_0, x = var_10122_cast_fp16)[name = tensor("op_10238_cast_fp16")]; + tensor var_10245_begin_0 = const()[name = tensor("op_10245_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10245_end_0 = const()[name = tensor("op_10245_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10245_end_mask_0 = const()[name = tensor("op_10245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10245_cast_fp16 = slice_by_index(begin = var_10245_begin_0, end = var_10245_end_0, end_mask = var_10245_end_mask_0, x = var_10122_cast_fp16)[name = tensor("op_10245_cast_fp16")]; + tensor var_10252_begin_0 = const()[name = tensor("op_10252_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10252_end_0 = const()[name = tensor("op_10252_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10252_end_mask_0 = const()[name = tensor("op_10252_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10252_cast_fp16 = slice_by_index(begin = var_10252_begin_0, end = var_10252_end_0, end_mask = var_10252_end_mask_0, x = var_10122_cast_fp16)[name = tensor("op_10252_cast_fp16")]; + tensor var_10259_begin_0 = const()[name = tensor("op_10259_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10259_end_0 = const()[name = tensor("op_10259_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10259_end_mask_0 = const()[name = tensor("op_10259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10259_cast_fp16 = slice_by_index(begin = var_10259_begin_0, end = var_10259_end_0, end_mask = var_10259_end_mask_0, x = var_10126_cast_fp16)[name = tensor("op_10259_cast_fp16")]; + tensor var_10266_begin_0 = const()[name = tensor("op_10266_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10266_end_0 = const()[name = tensor("op_10266_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10266_end_mask_0 = const()[name = tensor("op_10266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10266_cast_fp16 = slice_by_index(begin = var_10266_begin_0, end = var_10266_end_0, end_mask = var_10266_end_mask_0, x = var_10126_cast_fp16)[name = tensor("op_10266_cast_fp16")]; + tensor var_10273_begin_0 = const()[name = tensor("op_10273_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10273_end_0 = const()[name = tensor("op_10273_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10273_end_mask_0 = const()[name = tensor("op_10273_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10273_cast_fp16 = slice_by_index(begin = var_10273_begin_0, end = var_10273_end_0, end_mask = var_10273_end_mask_0, x = var_10126_cast_fp16)[name = tensor("op_10273_cast_fp16")]; + tensor var_10280_begin_0 = const()[name = tensor("op_10280_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10280_end_0 = const()[name = tensor("op_10280_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10280_end_mask_0 = const()[name = tensor("op_10280_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10280_cast_fp16 = slice_by_index(begin = var_10280_begin_0, end = var_10280_end_0, end_mask = var_10280_end_mask_0, x = var_10126_cast_fp16)[name = tensor("op_10280_cast_fp16")]; + tensor var_10287_begin_0 = const()[name = tensor("op_10287_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10287_end_0 = const()[name = tensor("op_10287_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10287_end_mask_0 = const()[name = tensor("op_10287_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10287_cast_fp16 = slice_by_index(begin = var_10287_begin_0, end = var_10287_end_0, end_mask = var_10287_end_mask_0, x = var_10130_cast_fp16)[name = tensor("op_10287_cast_fp16")]; + tensor var_10294_begin_0 = const()[name = tensor("op_10294_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10294_end_0 = const()[name = tensor("op_10294_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10294_end_mask_0 = const()[name = tensor("op_10294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10294_cast_fp16 = slice_by_index(begin = var_10294_begin_0, end = var_10294_end_0, end_mask = var_10294_end_mask_0, x = var_10130_cast_fp16)[name = tensor("op_10294_cast_fp16")]; + tensor var_10301_begin_0 = const()[name = tensor("op_10301_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10301_end_0 = const()[name = tensor("op_10301_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10301_end_mask_0 = const()[name = tensor("op_10301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10301_cast_fp16 = slice_by_index(begin = var_10301_begin_0, end = var_10301_end_0, end_mask = var_10301_end_mask_0, x = var_10130_cast_fp16)[name = tensor("op_10301_cast_fp16")]; + tensor var_10308_begin_0 = const()[name = tensor("op_10308_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10308_end_0 = const()[name = tensor("op_10308_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10308_end_mask_0 = const()[name = tensor("op_10308_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10308_cast_fp16 = slice_by_index(begin = var_10308_begin_0, end = var_10308_end_0, end_mask = var_10308_end_mask_0, x = var_10130_cast_fp16)[name = tensor("op_10308_cast_fp16")]; + tensor var_10315_begin_0 = const()[name = tensor("op_10315_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10315_end_0 = const()[name = tensor("op_10315_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10315_end_mask_0 = const()[name = tensor("op_10315_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10315_cast_fp16 = slice_by_index(begin = var_10315_begin_0, end = var_10315_end_0, end_mask = var_10315_end_mask_0, x = var_10134_cast_fp16)[name = tensor("op_10315_cast_fp16")]; + tensor var_10322_begin_0 = const()[name = tensor("op_10322_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10322_end_0 = const()[name = tensor("op_10322_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10322_end_mask_0 = const()[name = tensor("op_10322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10322_cast_fp16 = slice_by_index(begin = var_10322_begin_0, end = var_10322_end_0, end_mask = var_10322_end_mask_0, x = var_10134_cast_fp16)[name = tensor("op_10322_cast_fp16")]; + tensor var_10329_begin_0 = const()[name = tensor("op_10329_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10329_end_0 = const()[name = tensor("op_10329_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10329_end_mask_0 = const()[name = tensor("op_10329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10329_cast_fp16 = slice_by_index(begin = var_10329_begin_0, end = var_10329_end_0, end_mask = var_10329_end_mask_0, x = var_10134_cast_fp16)[name = tensor("op_10329_cast_fp16")]; + tensor var_10336_begin_0 = const()[name = tensor("op_10336_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10336_end_0 = const()[name = tensor("op_10336_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10336_end_mask_0 = const()[name = tensor("op_10336_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10336_cast_fp16 = slice_by_index(begin = var_10336_begin_0, end = var_10336_end_0, end_mask = var_10336_end_mask_0, x = var_10134_cast_fp16)[name = tensor("op_10336_cast_fp16")]; + tensor var_10343_begin_0 = const()[name = tensor("op_10343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10343_end_0 = const()[name = tensor("op_10343_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10343_end_mask_0 = const()[name = tensor("op_10343_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10343_cast_fp16 = slice_by_index(begin = var_10343_begin_0, end = var_10343_end_0, end_mask = var_10343_end_mask_0, x = var_10138_cast_fp16)[name = tensor("op_10343_cast_fp16")]; + tensor var_10350_begin_0 = const()[name = tensor("op_10350_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10350_end_0 = const()[name = tensor("op_10350_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10350_end_mask_0 = const()[name = tensor("op_10350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10350_cast_fp16 = slice_by_index(begin = var_10350_begin_0, end = var_10350_end_0, end_mask = var_10350_end_mask_0, x = var_10138_cast_fp16)[name = tensor("op_10350_cast_fp16")]; + tensor var_10357_begin_0 = const()[name = tensor("op_10357_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10357_end_0 = const()[name = tensor("op_10357_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10357_end_mask_0 = const()[name = tensor("op_10357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10357_cast_fp16 = slice_by_index(begin = var_10357_begin_0, end = var_10357_end_0, end_mask = var_10357_end_mask_0, x = var_10138_cast_fp16)[name = tensor("op_10357_cast_fp16")]; + tensor var_10364_begin_0 = const()[name = tensor("op_10364_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10364_end_0 = const()[name = tensor("op_10364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10364_end_mask_0 = const()[name = tensor("op_10364_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10364_cast_fp16 = slice_by_index(begin = var_10364_begin_0, end = var_10364_end_0, end_mask = var_10364_end_mask_0, x = var_10138_cast_fp16)[name = tensor("op_10364_cast_fp16")]; + tensor var_10371_begin_0 = const()[name = tensor("op_10371_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10371_end_0 = const()[name = tensor("op_10371_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10371_end_mask_0 = const()[name = tensor("op_10371_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10371_cast_fp16 = slice_by_index(begin = var_10371_begin_0, end = var_10371_end_0, end_mask = var_10371_end_mask_0, x = var_10142_cast_fp16)[name = tensor("op_10371_cast_fp16")]; + tensor var_10378_begin_0 = const()[name = tensor("op_10378_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10378_end_0 = const()[name = tensor("op_10378_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10378_end_mask_0 = const()[name = tensor("op_10378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10378_cast_fp16 = slice_by_index(begin = var_10378_begin_0, end = var_10378_end_0, end_mask = var_10378_end_mask_0, x = var_10142_cast_fp16)[name = tensor("op_10378_cast_fp16")]; + tensor var_10385_begin_0 = const()[name = tensor("op_10385_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10385_end_0 = const()[name = tensor("op_10385_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10385_end_mask_0 = const()[name = tensor("op_10385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10385_cast_fp16 = slice_by_index(begin = var_10385_begin_0, end = var_10385_end_0, end_mask = var_10385_end_mask_0, x = var_10142_cast_fp16)[name = tensor("op_10385_cast_fp16")]; + tensor var_10392_begin_0 = const()[name = tensor("op_10392_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10392_end_0 = const()[name = tensor("op_10392_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10392_end_mask_0 = const()[name = tensor("op_10392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10392_cast_fp16 = slice_by_index(begin = var_10392_begin_0, end = var_10392_end_0, end_mask = var_10392_end_mask_0, x = var_10142_cast_fp16)[name = tensor("op_10392_cast_fp16")]; + tensor var_10399_begin_0 = const()[name = tensor("op_10399_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10399_end_0 = const()[name = tensor("op_10399_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10399_end_mask_0 = const()[name = tensor("op_10399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10399_cast_fp16 = slice_by_index(begin = var_10399_begin_0, end = var_10399_end_0, end_mask = var_10399_end_mask_0, x = var_10146_cast_fp16)[name = tensor("op_10399_cast_fp16")]; + tensor var_10406_begin_0 = const()[name = tensor("op_10406_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10406_end_0 = const()[name = tensor("op_10406_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10406_end_mask_0 = const()[name = tensor("op_10406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10406_cast_fp16 = slice_by_index(begin = var_10406_begin_0, end = var_10406_end_0, end_mask = var_10406_end_mask_0, x = var_10146_cast_fp16)[name = tensor("op_10406_cast_fp16")]; + tensor var_10413_begin_0 = const()[name = tensor("op_10413_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10413_end_0 = const()[name = tensor("op_10413_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10413_end_mask_0 = const()[name = tensor("op_10413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10413_cast_fp16 = slice_by_index(begin = var_10413_begin_0, end = var_10413_end_0, end_mask = var_10413_end_mask_0, x = var_10146_cast_fp16)[name = tensor("op_10413_cast_fp16")]; + tensor var_10420_begin_0 = const()[name = tensor("op_10420_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10420_end_0 = const()[name = tensor("op_10420_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10420_end_mask_0 = const()[name = tensor("op_10420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10420_cast_fp16 = slice_by_index(begin = var_10420_begin_0, end = var_10420_end_0, end_mask = var_10420_end_mask_0, x = var_10146_cast_fp16)[name = tensor("op_10420_cast_fp16")]; + tensor var_10427_begin_0 = const()[name = tensor("op_10427_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10427_end_0 = const()[name = tensor("op_10427_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10427_end_mask_0 = const()[name = tensor("op_10427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10427_cast_fp16 = slice_by_index(begin = var_10427_begin_0, end = var_10427_end_0, end_mask = var_10427_end_mask_0, x = var_10150_cast_fp16)[name = tensor("op_10427_cast_fp16")]; + tensor var_10434_begin_0 = const()[name = tensor("op_10434_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10434_end_0 = const()[name = tensor("op_10434_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10434_end_mask_0 = const()[name = tensor("op_10434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10434_cast_fp16 = slice_by_index(begin = var_10434_begin_0, end = var_10434_end_0, end_mask = var_10434_end_mask_0, x = var_10150_cast_fp16)[name = tensor("op_10434_cast_fp16")]; + tensor var_10441_begin_0 = const()[name = tensor("op_10441_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10441_end_0 = const()[name = tensor("op_10441_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10441_end_mask_0 = const()[name = tensor("op_10441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10441_cast_fp16 = slice_by_index(begin = var_10441_begin_0, end = var_10441_end_0, end_mask = var_10441_end_mask_0, x = var_10150_cast_fp16)[name = tensor("op_10441_cast_fp16")]; + tensor var_10448_begin_0 = const()[name = tensor("op_10448_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10448_end_0 = const()[name = tensor("op_10448_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10448_end_mask_0 = const()[name = tensor("op_10448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10448_cast_fp16 = slice_by_index(begin = var_10448_begin_0, end = var_10448_end_0, end_mask = var_10448_end_mask_0, x = var_10150_cast_fp16)[name = tensor("op_10448_cast_fp16")]; + tensor var_10455_begin_0 = const()[name = tensor("op_10455_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10455_end_0 = const()[name = tensor("op_10455_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10455_end_mask_0 = const()[name = tensor("op_10455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10455_cast_fp16 = slice_by_index(begin = var_10455_begin_0, end = var_10455_end_0, end_mask = var_10455_end_mask_0, x = var_10154_cast_fp16)[name = tensor("op_10455_cast_fp16")]; + tensor var_10462_begin_0 = const()[name = tensor("op_10462_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10462_end_0 = const()[name = tensor("op_10462_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10462_end_mask_0 = const()[name = tensor("op_10462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10462_cast_fp16 = slice_by_index(begin = var_10462_begin_0, end = var_10462_end_0, end_mask = var_10462_end_mask_0, x = var_10154_cast_fp16)[name = tensor("op_10462_cast_fp16")]; + tensor var_10469_begin_0 = const()[name = tensor("op_10469_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10469_end_0 = const()[name = tensor("op_10469_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10469_end_mask_0 = const()[name = tensor("op_10469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10469_cast_fp16 = slice_by_index(begin = var_10469_begin_0, end = var_10469_end_0, end_mask = var_10469_end_mask_0, x = var_10154_cast_fp16)[name = tensor("op_10469_cast_fp16")]; + tensor var_10476_begin_0 = const()[name = tensor("op_10476_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10476_end_0 = const()[name = tensor("op_10476_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10476_end_mask_0 = const()[name = tensor("op_10476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10476_cast_fp16 = slice_by_index(begin = var_10476_begin_0, end = var_10476_end_0, end_mask = var_10476_end_mask_0, x = var_10154_cast_fp16)[name = tensor("op_10476_cast_fp16")]; + tensor var_10483_begin_0 = const()[name = tensor("op_10483_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10483_end_0 = const()[name = tensor("op_10483_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10483_end_mask_0 = const()[name = tensor("op_10483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10483_cast_fp16 = slice_by_index(begin = var_10483_begin_0, end = var_10483_end_0, end_mask = var_10483_end_mask_0, x = var_10158_cast_fp16)[name = tensor("op_10483_cast_fp16")]; + tensor var_10490_begin_0 = const()[name = tensor("op_10490_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10490_end_0 = const()[name = tensor("op_10490_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10490_end_mask_0 = const()[name = tensor("op_10490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10490_cast_fp16 = slice_by_index(begin = var_10490_begin_0, end = var_10490_end_0, end_mask = var_10490_end_mask_0, x = var_10158_cast_fp16)[name = tensor("op_10490_cast_fp16")]; + tensor var_10497_begin_0 = const()[name = tensor("op_10497_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10497_end_0 = const()[name = tensor("op_10497_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10497_end_mask_0 = const()[name = tensor("op_10497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10497_cast_fp16 = slice_by_index(begin = var_10497_begin_0, end = var_10497_end_0, end_mask = var_10497_end_mask_0, x = var_10158_cast_fp16)[name = tensor("op_10497_cast_fp16")]; + tensor var_10504_begin_0 = const()[name = tensor("op_10504_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10504_end_0 = const()[name = tensor("op_10504_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10504_end_mask_0 = const()[name = tensor("op_10504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10504_cast_fp16 = slice_by_index(begin = var_10504_begin_0, end = var_10504_end_0, end_mask = var_10504_end_mask_0, x = var_10158_cast_fp16)[name = tensor("op_10504_cast_fp16")]; + tensor var_10511_begin_0 = const()[name = tensor("op_10511_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10511_end_0 = const()[name = tensor("op_10511_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10511_end_mask_0 = const()[name = tensor("op_10511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10511_cast_fp16 = slice_by_index(begin = var_10511_begin_0, end = var_10511_end_0, end_mask = var_10511_end_mask_0, x = var_10162_cast_fp16)[name = tensor("op_10511_cast_fp16")]; + tensor var_10518_begin_0 = const()[name = tensor("op_10518_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10518_end_0 = const()[name = tensor("op_10518_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10518_end_mask_0 = const()[name = tensor("op_10518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10518_cast_fp16 = slice_by_index(begin = var_10518_begin_0, end = var_10518_end_0, end_mask = var_10518_end_mask_0, x = var_10162_cast_fp16)[name = tensor("op_10518_cast_fp16")]; + tensor var_10525_begin_0 = const()[name = tensor("op_10525_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10525_end_0 = const()[name = tensor("op_10525_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10525_end_mask_0 = const()[name = tensor("op_10525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10525_cast_fp16 = slice_by_index(begin = var_10525_begin_0, end = var_10525_end_0, end_mask = var_10525_end_mask_0, x = var_10162_cast_fp16)[name = tensor("op_10525_cast_fp16")]; + tensor var_10532_begin_0 = const()[name = tensor("op_10532_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10532_end_0 = const()[name = tensor("op_10532_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10532_end_mask_0 = const()[name = tensor("op_10532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10532_cast_fp16 = slice_by_index(begin = var_10532_begin_0, end = var_10532_end_0, end_mask = var_10532_end_mask_0, x = var_10162_cast_fp16)[name = tensor("op_10532_cast_fp16")]; + tensor var_10539_begin_0 = const()[name = tensor("op_10539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10539_end_0 = const()[name = tensor("op_10539_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10539_end_mask_0 = const()[name = tensor("op_10539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10539_cast_fp16 = slice_by_index(begin = var_10539_begin_0, end = var_10539_end_0, end_mask = var_10539_end_mask_0, x = var_10166_cast_fp16)[name = tensor("op_10539_cast_fp16")]; + tensor var_10546_begin_0 = const()[name = tensor("op_10546_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10546_end_0 = const()[name = tensor("op_10546_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10546_end_mask_0 = const()[name = tensor("op_10546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10546_cast_fp16 = slice_by_index(begin = var_10546_begin_0, end = var_10546_end_0, end_mask = var_10546_end_mask_0, x = var_10166_cast_fp16)[name = tensor("op_10546_cast_fp16")]; + tensor var_10553_begin_0 = const()[name = tensor("op_10553_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10553_end_0 = const()[name = tensor("op_10553_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10553_end_mask_0 = const()[name = tensor("op_10553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10553_cast_fp16 = slice_by_index(begin = var_10553_begin_0, end = var_10553_end_0, end_mask = var_10553_end_mask_0, x = var_10166_cast_fp16)[name = tensor("op_10553_cast_fp16")]; + tensor var_10560_begin_0 = const()[name = tensor("op_10560_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10560_end_0 = const()[name = tensor("op_10560_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10560_end_mask_0 = const()[name = tensor("op_10560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10560_cast_fp16 = slice_by_index(begin = var_10560_begin_0, end = var_10560_end_0, end_mask = var_10560_end_mask_0, x = var_10166_cast_fp16)[name = tensor("op_10560_cast_fp16")]; + tensor var_10567_begin_0 = const()[name = tensor("op_10567_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10567_end_0 = const()[name = tensor("op_10567_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10567_end_mask_0 = const()[name = tensor("op_10567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10567_cast_fp16 = slice_by_index(begin = var_10567_begin_0, end = var_10567_end_0, end_mask = var_10567_end_mask_0, x = var_10170_cast_fp16)[name = tensor("op_10567_cast_fp16")]; + tensor var_10574_begin_0 = const()[name = tensor("op_10574_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10574_end_0 = const()[name = tensor("op_10574_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10574_end_mask_0 = const()[name = tensor("op_10574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10574_cast_fp16 = slice_by_index(begin = var_10574_begin_0, end = var_10574_end_0, end_mask = var_10574_end_mask_0, x = var_10170_cast_fp16)[name = tensor("op_10574_cast_fp16")]; + tensor var_10581_begin_0 = const()[name = tensor("op_10581_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10581_end_0 = const()[name = tensor("op_10581_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10581_end_mask_0 = const()[name = tensor("op_10581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10581_cast_fp16 = slice_by_index(begin = var_10581_begin_0, end = var_10581_end_0, end_mask = var_10581_end_mask_0, x = var_10170_cast_fp16)[name = tensor("op_10581_cast_fp16")]; + tensor var_10588_begin_0 = const()[name = tensor("op_10588_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10588_end_0 = const()[name = tensor("op_10588_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10588_end_mask_0 = const()[name = tensor("op_10588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10588_cast_fp16 = slice_by_index(begin = var_10588_begin_0, end = var_10588_end_0, end_mask = var_10588_end_mask_0, x = var_10170_cast_fp16)[name = tensor("op_10588_cast_fp16")]; + tensor var_10595_begin_0 = const()[name = tensor("op_10595_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10595_end_0 = const()[name = tensor("op_10595_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10595_end_mask_0 = const()[name = tensor("op_10595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10595_cast_fp16 = slice_by_index(begin = var_10595_begin_0, end = var_10595_end_0, end_mask = var_10595_end_mask_0, x = var_10174_cast_fp16)[name = tensor("op_10595_cast_fp16")]; + tensor var_10602_begin_0 = const()[name = tensor("op_10602_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10602_end_0 = const()[name = tensor("op_10602_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10602_end_mask_0 = const()[name = tensor("op_10602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10602_cast_fp16 = slice_by_index(begin = var_10602_begin_0, end = var_10602_end_0, end_mask = var_10602_end_mask_0, x = var_10174_cast_fp16)[name = tensor("op_10602_cast_fp16")]; + tensor var_10609_begin_0 = const()[name = tensor("op_10609_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10609_end_0 = const()[name = tensor("op_10609_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10609_end_mask_0 = const()[name = tensor("op_10609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10609_cast_fp16 = slice_by_index(begin = var_10609_begin_0, end = var_10609_end_0, end_mask = var_10609_end_mask_0, x = var_10174_cast_fp16)[name = tensor("op_10609_cast_fp16")]; + tensor var_10616_begin_0 = const()[name = tensor("op_10616_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10616_end_0 = const()[name = tensor("op_10616_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10616_end_mask_0 = const()[name = tensor("op_10616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10616_cast_fp16 = slice_by_index(begin = var_10616_begin_0, end = var_10616_end_0, end_mask = var_10616_end_mask_0, x = var_10174_cast_fp16)[name = tensor("op_10616_cast_fp16")]; + tensor var_10623_begin_0 = const()[name = tensor("op_10623_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10623_end_0 = const()[name = tensor("op_10623_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10623_end_mask_0 = const()[name = tensor("op_10623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10623_cast_fp16 = slice_by_index(begin = var_10623_begin_0, end = var_10623_end_0, end_mask = var_10623_end_mask_0, x = var_10178_cast_fp16)[name = tensor("op_10623_cast_fp16")]; + tensor var_10630_begin_0 = const()[name = tensor("op_10630_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10630_end_0 = const()[name = tensor("op_10630_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10630_end_mask_0 = const()[name = tensor("op_10630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10630_cast_fp16 = slice_by_index(begin = var_10630_begin_0, end = var_10630_end_0, end_mask = var_10630_end_mask_0, x = var_10178_cast_fp16)[name = tensor("op_10630_cast_fp16")]; + tensor var_10637_begin_0 = const()[name = tensor("op_10637_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10637_end_0 = const()[name = tensor("op_10637_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10637_end_mask_0 = const()[name = tensor("op_10637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10637_cast_fp16 = slice_by_index(begin = var_10637_begin_0, end = var_10637_end_0, end_mask = var_10637_end_mask_0, x = var_10178_cast_fp16)[name = tensor("op_10637_cast_fp16")]; + tensor var_10644_begin_0 = const()[name = tensor("op_10644_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10644_end_0 = const()[name = tensor("op_10644_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10644_end_mask_0 = const()[name = tensor("op_10644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10644_cast_fp16 = slice_by_index(begin = var_10644_begin_0, end = var_10644_end_0, end_mask = var_10644_end_mask_0, x = var_10178_cast_fp16)[name = tensor("op_10644_cast_fp16")]; + tensor var_10651_begin_0 = const()[name = tensor("op_10651_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10651_end_0 = const()[name = tensor("op_10651_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10651_end_mask_0 = const()[name = tensor("op_10651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10651_cast_fp16 = slice_by_index(begin = var_10651_begin_0, end = var_10651_end_0, end_mask = var_10651_end_mask_0, x = var_10182_cast_fp16)[name = tensor("op_10651_cast_fp16")]; + tensor var_10658_begin_0 = const()[name = tensor("op_10658_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10658_end_0 = const()[name = tensor("op_10658_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10658_end_mask_0 = const()[name = tensor("op_10658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10658_cast_fp16 = slice_by_index(begin = var_10658_begin_0, end = var_10658_end_0, end_mask = var_10658_end_mask_0, x = var_10182_cast_fp16)[name = tensor("op_10658_cast_fp16")]; + tensor var_10665_begin_0 = const()[name = tensor("op_10665_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10665_end_0 = const()[name = tensor("op_10665_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10665_end_mask_0 = const()[name = tensor("op_10665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10665_cast_fp16 = slice_by_index(begin = var_10665_begin_0, end = var_10665_end_0, end_mask = var_10665_end_mask_0, x = var_10182_cast_fp16)[name = tensor("op_10665_cast_fp16")]; + tensor var_10672_begin_0 = const()[name = tensor("op_10672_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10672_end_0 = const()[name = tensor("op_10672_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10672_end_mask_0 = const()[name = tensor("op_10672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10672_cast_fp16 = slice_by_index(begin = var_10672_begin_0, end = var_10672_end_0, end_mask = var_10672_end_mask_0, x = var_10182_cast_fp16)[name = tensor("op_10672_cast_fp16")]; + tensor var_10679_begin_0 = const()[name = tensor("op_10679_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10679_end_0 = const()[name = tensor("op_10679_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10679_end_mask_0 = const()[name = tensor("op_10679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10679_cast_fp16 = slice_by_index(begin = var_10679_begin_0, end = var_10679_end_0, end_mask = var_10679_end_mask_0, x = var_10186_cast_fp16)[name = tensor("op_10679_cast_fp16")]; + tensor var_10686_begin_0 = const()[name = tensor("op_10686_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10686_end_0 = const()[name = tensor("op_10686_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10686_end_mask_0 = const()[name = tensor("op_10686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10686_cast_fp16 = slice_by_index(begin = var_10686_begin_0, end = var_10686_end_0, end_mask = var_10686_end_mask_0, x = var_10186_cast_fp16)[name = tensor("op_10686_cast_fp16")]; + tensor var_10693_begin_0 = const()[name = tensor("op_10693_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10693_end_0 = const()[name = tensor("op_10693_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10693_end_mask_0 = const()[name = tensor("op_10693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10693_cast_fp16 = slice_by_index(begin = var_10693_begin_0, end = var_10693_end_0, end_mask = var_10693_end_mask_0, x = var_10186_cast_fp16)[name = tensor("op_10693_cast_fp16")]; + tensor var_10700_begin_0 = const()[name = tensor("op_10700_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10700_end_0 = const()[name = tensor("op_10700_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10700_end_mask_0 = const()[name = tensor("op_10700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10700_cast_fp16 = slice_by_index(begin = var_10700_begin_0, end = var_10700_end_0, end_mask = var_10700_end_mask_0, x = var_10186_cast_fp16)[name = tensor("op_10700_cast_fp16")]; + tensor var_10707_begin_0 = const()[name = tensor("op_10707_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10707_end_0 = const()[name = tensor("op_10707_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10707_end_mask_0 = const()[name = tensor("op_10707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10707_cast_fp16 = slice_by_index(begin = var_10707_begin_0, end = var_10707_end_0, end_mask = var_10707_end_mask_0, x = var_10190_cast_fp16)[name = tensor("op_10707_cast_fp16")]; + tensor var_10714_begin_0 = const()[name = tensor("op_10714_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10714_end_0 = const()[name = tensor("op_10714_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10714_end_mask_0 = const()[name = tensor("op_10714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10714_cast_fp16 = slice_by_index(begin = var_10714_begin_0, end = var_10714_end_0, end_mask = var_10714_end_mask_0, x = var_10190_cast_fp16)[name = tensor("op_10714_cast_fp16")]; + tensor var_10721_begin_0 = const()[name = tensor("op_10721_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10721_end_0 = const()[name = tensor("op_10721_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10721_end_mask_0 = const()[name = tensor("op_10721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10721_cast_fp16 = slice_by_index(begin = var_10721_begin_0, end = var_10721_end_0, end_mask = var_10721_end_mask_0, x = var_10190_cast_fp16)[name = tensor("op_10721_cast_fp16")]; + tensor var_10728_begin_0 = const()[name = tensor("op_10728_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10728_end_0 = const()[name = tensor("op_10728_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10728_end_mask_0 = const()[name = tensor("op_10728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10728_cast_fp16 = slice_by_index(begin = var_10728_begin_0, end = var_10728_end_0, end_mask = var_10728_end_mask_0, x = var_10190_cast_fp16)[name = tensor("op_10728_cast_fp16")]; + tensor var_10735_begin_0 = const()[name = tensor("op_10735_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10735_end_0 = const()[name = tensor("op_10735_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_10735_end_mask_0 = const()[name = tensor("op_10735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10735_cast_fp16 = slice_by_index(begin = var_10735_begin_0, end = var_10735_end_0, end_mask = var_10735_end_mask_0, x = var_10194_cast_fp16)[name = tensor("op_10735_cast_fp16")]; + tensor var_10742_begin_0 = const()[name = tensor("op_10742_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_10742_end_0 = const()[name = tensor("op_10742_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_10742_end_mask_0 = const()[name = tensor("op_10742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10742_cast_fp16 = slice_by_index(begin = var_10742_begin_0, end = var_10742_end_0, end_mask = var_10742_end_mask_0, x = var_10194_cast_fp16)[name = tensor("op_10742_cast_fp16")]; + tensor var_10749_begin_0 = const()[name = tensor("op_10749_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_10749_end_0 = const()[name = tensor("op_10749_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_10749_end_mask_0 = const()[name = tensor("op_10749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10749_cast_fp16 = slice_by_index(begin = var_10749_begin_0, end = var_10749_end_0, end_mask = var_10749_end_mask_0, x = var_10194_cast_fp16)[name = tensor("op_10749_cast_fp16")]; + tensor var_10756_begin_0 = const()[name = tensor("op_10756_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_10756_end_0 = const()[name = tensor("op_10756_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10756_end_mask_0 = const()[name = tensor("op_10756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10756_cast_fp16 = slice_by_index(begin = var_10756_begin_0, end = var_10756_end_0, end_mask = var_10756_end_mask_0, x = var_10194_cast_fp16)[name = tensor("op_10756_cast_fp16")]; + tensor k_13_perm_0 = const()[name = tensor("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_10761_begin_0 = const()[name = tensor("op_10761_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10761_end_0 = const()[name = tensor("op_10761_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_10761_end_mask_0 = const()[name = tensor("op_10761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_25 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = tensor("transpose_25")]; + tensor var_10761_cast_fp16 = slice_by_index(begin = var_10761_begin_0, end = var_10761_end_0, end_mask = var_10761_end_mask_0, x = transpose_25)[name = tensor("op_10761_cast_fp16")]; + tensor var_10765_begin_0 = const()[name = tensor("op_10765_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_10765_end_0 = const()[name = tensor("op_10765_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_10765_end_mask_0 = const()[name = tensor("op_10765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10765_cast_fp16 = slice_by_index(begin = var_10765_begin_0, end = var_10765_end_0, end_mask = var_10765_end_mask_0, x = transpose_25)[name = tensor("op_10765_cast_fp16")]; + tensor var_10769_begin_0 = const()[name = tensor("op_10769_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_10769_end_0 = const()[name = tensor("op_10769_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_10769_end_mask_0 = const()[name = tensor("op_10769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10769_cast_fp16 = slice_by_index(begin = var_10769_begin_0, end = var_10769_end_0, end_mask = var_10769_end_mask_0, x = transpose_25)[name = tensor("op_10769_cast_fp16")]; + tensor var_10773_begin_0 = const()[name = tensor("op_10773_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_10773_end_0 = const()[name = tensor("op_10773_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_10773_end_mask_0 = const()[name = tensor("op_10773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10773_cast_fp16 = slice_by_index(begin = var_10773_begin_0, end = var_10773_end_0, end_mask = var_10773_end_mask_0, x = transpose_25)[name = tensor("op_10773_cast_fp16")]; + tensor var_10777_begin_0 = const()[name = tensor("op_10777_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_10777_end_0 = const()[name = tensor("op_10777_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_10777_end_mask_0 = const()[name = tensor("op_10777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10777_cast_fp16 = slice_by_index(begin = var_10777_begin_0, end = var_10777_end_0, end_mask = var_10777_end_mask_0, x = transpose_25)[name = tensor("op_10777_cast_fp16")]; + tensor var_10781_begin_0 = const()[name = tensor("op_10781_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_10781_end_0 = const()[name = tensor("op_10781_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_10781_end_mask_0 = const()[name = tensor("op_10781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10781_cast_fp16 = slice_by_index(begin = var_10781_begin_0, end = var_10781_end_0, end_mask = var_10781_end_mask_0, x = transpose_25)[name = tensor("op_10781_cast_fp16")]; + tensor var_10785_begin_0 = const()[name = tensor("op_10785_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_10785_end_0 = const()[name = tensor("op_10785_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_10785_end_mask_0 = const()[name = tensor("op_10785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10785_cast_fp16 = slice_by_index(begin = var_10785_begin_0, end = var_10785_end_0, end_mask = var_10785_end_mask_0, x = transpose_25)[name = tensor("op_10785_cast_fp16")]; + tensor var_10789_begin_0 = const()[name = tensor("op_10789_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_10789_end_0 = const()[name = tensor("op_10789_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_10789_end_mask_0 = const()[name = tensor("op_10789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10789_cast_fp16 = slice_by_index(begin = var_10789_begin_0, end = var_10789_end_0, end_mask = var_10789_end_mask_0, x = transpose_25)[name = tensor("op_10789_cast_fp16")]; + tensor var_10793_begin_0 = const()[name = tensor("op_10793_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_10793_end_0 = const()[name = tensor("op_10793_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_10793_end_mask_0 = const()[name = tensor("op_10793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10793_cast_fp16 = slice_by_index(begin = var_10793_begin_0, end = var_10793_end_0, end_mask = var_10793_end_mask_0, x = transpose_25)[name = tensor("op_10793_cast_fp16")]; + tensor var_10797_begin_0 = const()[name = tensor("op_10797_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_10797_end_0 = const()[name = tensor("op_10797_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_10797_end_mask_0 = const()[name = tensor("op_10797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10797_cast_fp16 = slice_by_index(begin = var_10797_begin_0, end = var_10797_end_0, end_mask = var_10797_end_mask_0, x = transpose_25)[name = tensor("op_10797_cast_fp16")]; + tensor var_10801_begin_0 = const()[name = tensor("op_10801_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_10801_end_0 = const()[name = tensor("op_10801_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_10801_end_mask_0 = const()[name = tensor("op_10801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10801_cast_fp16 = slice_by_index(begin = var_10801_begin_0, end = var_10801_end_0, end_mask = var_10801_end_mask_0, x = transpose_25)[name = tensor("op_10801_cast_fp16")]; + tensor var_10805_begin_0 = const()[name = tensor("op_10805_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_10805_end_0 = const()[name = tensor("op_10805_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_10805_end_mask_0 = const()[name = tensor("op_10805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10805_cast_fp16 = slice_by_index(begin = var_10805_begin_0, end = var_10805_end_0, end_mask = var_10805_end_mask_0, x = transpose_25)[name = tensor("op_10805_cast_fp16")]; + tensor var_10809_begin_0 = const()[name = tensor("op_10809_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_10809_end_0 = const()[name = tensor("op_10809_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_10809_end_mask_0 = const()[name = tensor("op_10809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10809_cast_fp16 = slice_by_index(begin = var_10809_begin_0, end = var_10809_end_0, end_mask = var_10809_end_mask_0, x = transpose_25)[name = tensor("op_10809_cast_fp16")]; + tensor var_10813_begin_0 = const()[name = tensor("op_10813_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_10813_end_0 = const()[name = tensor("op_10813_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_10813_end_mask_0 = const()[name = tensor("op_10813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10813_cast_fp16 = slice_by_index(begin = var_10813_begin_0, end = var_10813_end_0, end_mask = var_10813_end_mask_0, x = transpose_25)[name = tensor("op_10813_cast_fp16")]; + tensor var_10817_begin_0 = const()[name = tensor("op_10817_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_10817_end_0 = const()[name = tensor("op_10817_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_10817_end_mask_0 = const()[name = tensor("op_10817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10817_cast_fp16 = slice_by_index(begin = var_10817_begin_0, end = var_10817_end_0, end_mask = var_10817_end_mask_0, x = transpose_25)[name = tensor("op_10817_cast_fp16")]; + tensor var_10821_begin_0 = const()[name = tensor("op_10821_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_10821_end_0 = const()[name = tensor("op_10821_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_10821_end_mask_0 = const()[name = tensor("op_10821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10821_cast_fp16 = slice_by_index(begin = var_10821_begin_0, end = var_10821_end_0, end_mask = var_10821_end_mask_0, x = transpose_25)[name = tensor("op_10821_cast_fp16")]; + tensor var_10825_begin_0 = const()[name = tensor("op_10825_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_10825_end_0 = const()[name = tensor("op_10825_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_10825_end_mask_0 = const()[name = tensor("op_10825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10825_cast_fp16 = slice_by_index(begin = var_10825_begin_0, end = var_10825_end_0, end_mask = var_10825_end_mask_0, x = transpose_25)[name = tensor("op_10825_cast_fp16")]; + tensor var_10829_begin_0 = const()[name = tensor("op_10829_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_10829_end_0 = const()[name = tensor("op_10829_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_10829_end_mask_0 = const()[name = tensor("op_10829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10829_cast_fp16 = slice_by_index(begin = var_10829_begin_0, end = var_10829_end_0, end_mask = var_10829_end_mask_0, x = transpose_25)[name = tensor("op_10829_cast_fp16")]; + tensor var_10833_begin_0 = const()[name = tensor("op_10833_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_10833_end_0 = const()[name = tensor("op_10833_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_10833_end_mask_0 = const()[name = tensor("op_10833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10833_cast_fp16 = slice_by_index(begin = var_10833_begin_0, end = var_10833_end_0, end_mask = var_10833_end_mask_0, x = transpose_25)[name = tensor("op_10833_cast_fp16")]; + tensor var_10837_begin_0 = const()[name = tensor("op_10837_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_10837_end_0 = const()[name = tensor("op_10837_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_10837_end_mask_0 = const()[name = tensor("op_10837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_10837_cast_fp16 = slice_by_index(begin = var_10837_begin_0, end = var_10837_end_0, end_mask = var_10837_end_mask_0, x = transpose_25)[name = tensor("op_10837_cast_fp16")]; + tensor var_10839_begin_0 = const()[name = tensor("op_10839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_10839_end_0 = const()[name = tensor("op_10839_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_10839_end_mask_0 = const()[name = tensor("op_10839_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10839_cast_fp16 = slice_by_index(begin = var_10839_begin_0, end = var_10839_end_0, end_mask = var_10839_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10839_cast_fp16")]; + tensor var_10843_begin_0 = const()[name = tensor("op_10843_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_10843_end_0 = const()[name = tensor("op_10843_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_10843_end_mask_0 = const()[name = tensor("op_10843_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10843_cast_fp16 = slice_by_index(begin = var_10843_begin_0, end = var_10843_end_0, end_mask = var_10843_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10843_cast_fp16")]; + tensor var_10847_begin_0 = const()[name = tensor("op_10847_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_10847_end_0 = const()[name = tensor("op_10847_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_10847_end_mask_0 = const()[name = tensor("op_10847_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10847_cast_fp16 = slice_by_index(begin = var_10847_begin_0, end = var_10847_end_0, end_mask = var_10847_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10847_cast_fp16")]; + tensor var_10851_begin_0 = const()[name = tensor("op_10851_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_10851_end_0 = const()[name = tensor("op_10851_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_10851_end_mask_0 = const()[name = tensor("op_10851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10851_cast_fp16 = slice_by_index(begin = var_10851_begin_0, end = var_10851_end_0, end_mask = var_10851_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10851_cast_fp16")]; + tensor var_10855_begin_0 = const()[name = tensor("op_10855_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_10855_end_0 = const()[name = tensor("op_10855_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_10855_end_mask_0 = const()[name = tensor("op_10855_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10855_cast_fp16 = slice_by_index(begin = var_10855_begin_0, end = var_10855_end_0, end_mask = var_10855_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10855_cast_fp16")]; + tensor var_10859_begin_0 = const()[name = tensor("op_10859_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_10859_end_0 = const()[name = tensor("op_10859_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_10859_end_mask_0 = const()[name = tensor("op_10859_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10859_cast_fp16 = slice_by_index(begin = var_10859_begin_0, end = var_10859_end_0, end_mask = var_10859_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10859_cast_fp16")]; + tensor var_10863_begin_0 = const()[name = tensor("op_10863_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_10863_end_0 = const()[name = tensor("op_10863_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_10863_end_mask_0 = const()[name = tensor("op_10863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10863_cast_fp16 = slice_by_index(begin = var_10863_begin_0, end = var_10863_end_0, end_mask = var_10863_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10863_cast_fp16")]; + tensor var_10867_begin_0 = const()[name = tensor("op_10867_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_10867_end_0 = const()[name = tensor("op_10867_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_10867_end_mask_0 = const()[name = tensor("op_10867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10867_cast_fp16 = slice_by_index(begin = var_10867_begin_0, end = var_10867_end_0, end_mask = var_10867_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10867_cast_fp16")]; + tensor var_10871_begin_0 = const()[name = tensor("op_10871_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_10871_end_0 = const()[name = tensor("op_10871_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_10871_end_mask_0 = const()[name = tensor("op_10871_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10871_cast_fp16 = slice_by_index(begin = var_10871_begin_0, end = var_10871_end_0, end_mask = var_10871_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10871_cast_fp16")]; + tensor var_10875_begin_0 = const()[name = tensor("op_10875_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_10875_end_0 = const()[name = tensor("op_10875_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_10875_end_mask_0 = const()[name = tensor("op_10875_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10875_cast_fp16 = slice_by_index(begin = var_10875_begin_0, end = var_10875_end_0, end_mask = var_10875_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10875_cast_fp16")]; + tensor var_10879_begin_0 = const()[name = tensor("op_10879_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_10879_end_0 = const()[name = tensor("op_10879_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_10879_end_mask_0 = const()[name = tensor("op_10879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10879_cast_fp16 = slice_by_index(begin = var_10879_begin_0, end = var_10879_end_0, end_mask = var_10879_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10879_cast_fp16")]; + tensor var_10883_begin_0 = const()[name = tensor("op_10883_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_10883_end_0 = const()[name = tensor("op_10883_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_10883_end_mask_0 = const()[name = tensor("op_10883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10883_cast_fp16 = slice_by_index(begin = var_10883_begin_0, end = var_10883_end_0, end_mask = var_10883_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10883_cast_fp16")]; + tensor var_10887_begin_0 = const()[name = tensor("op_10887_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_10887_end_0 = const()[name = tensor("op_10887_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_10887_end_mask_0 = const()[name = tensor("op_10887_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10887_cast_fp16 = slice_by_index(begin = var_10887_begin_0, end = var_10887_end_0, end_mask = var_10887_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10887_cast_fp16")]; + tensor var_10891_begin_0 = const()[name = tensor("op_10891_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_10891_end_0 = const()[name = tensor("op_10891_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_10891_end_mask_0 = const()[name = tensor("op_10891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10891_cast_fp16 = slice_by_index(begin = var_10891_begin_0, end = var_10891_end_0, end_mask = var_10891_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10891_cast_fp16")]; + tensor var_10895_begin_0 = const()[name = tensor("op_10895_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_10895_end_0 = const()[name = tensor("op_10895_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_10895_end_mask_0 = const()[name = tensor("op_10895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10895_cast_fp16 = slice_by_index(begin = var_10895_begin_0, end = var_10895_end_0, end_mask = var_10895_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10895_cast_fp16")]; + tensor var_10899_begin_0 = const()[name = tensor("op_10899_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_10899_end_0 = const()[name = tensor("op_10899_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_10899_end_mask_0 = const()[name = tensor("op_10899_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10899_cast_fp16 = slice_by_index(begin = var_10899_begin_0, end = var_10899_end_0, end_mask = var_10899_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10899_cast_fp16")]; + tensor var_10903_begin_0 = const()[name = tensor("op_10903_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_10903_end_0 = const()[name = tensor("op_10903_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_10903_end_mask_0 = const()[name = tensor("op_10903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10903_cast_fp16 = slice_by_index(begin = var_10903_begin_0, end = var_10903_end_0, end_mask = var_10903_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10903_cast_fp16")]; + tensor var_10907_begin_0 = const()[name = tensor("op_10907_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_10907_end_0 = const()[name = tensor("op_10907_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_10907_end_mask_0 = const()[name = tensor("op_10907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10907_cast_fp16 = slice_by_index(begin = var_10907_begin_0, end = var_10907_end_0, end_mask = var_10907_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10907_cast_fp16")]; + tensor var_10911_begin_0 = const()[name = tensor("op_10911_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_10911_end_0 = const()[name = tensor("op_10911_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_10911_end_mask_0 = const()[name = tensor("op_10911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10911_cast_fp16 = slice_by_index(begin = var_10911_begin_0, end = var_10911_end_0, end_mask = var_10911_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10911_cast_fp16")]; + tensor var_10915_begin_0 = const()[name = tensor("op_10915_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_10915_end_0 = const()[name = tensor("op_10915_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_10915_end_mask_0 = const()[name = tensor("op_10915_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_10915_cast_fp16 = slice_by_index(begin = var_10915_begin_0, end = var_10915_end_0, end_mask = var_10915_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_10915_cast_fp16")]; + tensor var_10919_equation_0 = const()[name = tensor("op_10919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10919_cast_fp16 = einsum(equation = var_10919_equation_0, values = (var_10761_cast_fp16, var_10203_cast_fp16))[name = tensor("op_10919_cast_fp16")]; + tensor var_10920_to_fp16 = const()[name = tensor("op_10920_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_961_cast_fp16 = mul(x = var_10919_cast_fp16, y = var_10920_to_fp16)[name = tensor("aw_chunk_961_cast_fp16")]; + tensor var_10923_equation_0 = const()[name = tensor("op_10923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10923_cast_fp16 = einsum(equation = var_10923_equation_0, values = (var_10761_cast_fp16, var_10210_cast_fp16))[name = tensor("op_10923_cast_fp16")]; + tensor var_10924_to_fp16 = const()[name = tensor("op_10924_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_963_cast_fp16 = mul(x = var_10923_cast_fp16, y = var_10924_to_fp16)[name = tensor("aw_chunk_963_cast_fp16")]; + tensor var_10927_equation_0 = const()[name = tensor("op_10927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10927_cast_fp16 = einsum(equation = var_10927_equation_0, values = (var_10761_cast_fp16, var_10217_cast_fp16))[name = tensor("op_10927_cast_fp16")]; + tensor var_10928_to_fp16 = const()[name = tensor("op_10928_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_965_cast_fp16 = mul(x = var_10927_cast_fp16, y = var_10928_to_fp16)[name = tensor("aw_chunk_965_cast_fp16")]; + tensor var_10931_equation_0 = const()[name = tensor("op_10931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10931_cast_fp16 = einsum(equation = var_10931_equation_0, values = (var_10761_cast_fp16, var_10224_cast_fp16))[name = tensor("op_10931_cast_fp16")]; + tensor var_10932_to_fp16 = const()[name = tensor("op_10932_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_967_cast_fp16 = mul(x = var_10931_cast_fp16, y = var_10932_to_fp16)[name = tensor("aw_chunk_967_cast_fp16")]; + tensor var_10935_equation_0 = const()[name = tensor("op_10935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10935_cast_fp16 = einsum(equation = var_10935_equation_0, values = (var_10765_cast_fp16, var_10231_cast_fp16))[name = tensor("op_10935_cast_fp16")]; + tensor var_10936_to_fp16 = const()[name = tensor("op_10936_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_969_cast_fp16 = mul(x = var_10935_cast_fp16, y = var_10936_to_fp16)[name = tensor("aw_chunk_969_cast_fp16")]; + tensor var_10939_equation_0 = const()[name = tensor("op_10939_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10939_cast_fp16 = einsum(equation = var_10939_equation_0, values = (var_10765_cast_fp16, var_10238_cast_fp16))[name = tensor("op_10939_cast_fp16")]; + tensor var_10940_to_fp16 = const()[name = tensor("op_10940_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_971_cast_fp16 = mul(x = var_10939_cast_fp16, y = var_10940_to_fp16)[name = tensor("aw_chunk_971_cast_fp16")]; + tensor var_10943_equation_0 = const()[name = tensor("op_10943_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10943_cast_fp16 = einsum(equation = var_10943_equation_0, values = (var_10765_cast_fp16, var_10245_cast_fp16))[name = tensor("op_10943_cast_fp16")]; + tensor var_10944_to_fp16 = const()[name = tensor("op_10944_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_973_cast_fp16 = mul(x = var_10943_cast_fp16, y = var_10944_to_fp16)[name = tensor("aw_chunk_973_cast_fp16")]; + tensor var_10947_equation_0 = const()[name = tensor("op_10947_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10947_cast_fp16 = einsum(equation = var_10947_equation_0, values = (var_10765_cast_fp16, var_10252_cast_fp16))[name = tensor("op_10947_cast_fp16")]; + tensor var_10948_to_fp16 = const()[name = tensor("op_10948_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_975_cast_fp16 = mul(x = var_10947_cast_fp16, y = var_10948_to_fp16)[name = tensor("aw_chunk_975_cast_fp16")]; + tensor var_10951_equation_0 = const()[name = tensor("op_10951_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10951_cast_fp16 = einsum(equation = var_10951_equation_0, values = (var_10769_cast_fp16, var_10259_cast_fp16))[name = tensor("op_10951_cast_fp16")]; + tensor var_10952_to_fp16 = const()[name = tensor("op_10952_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_977_cast_fp16 = mul(x = var_10951_cast_fp16, y = var_10952_to_fp16)[name = tensor("aw_chunk_977_cast_fp16")]; + tensor var_10955_equation_0 = const()[name = tensor("op_10955_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10955_cast_fp16 = einsum(equation = var_10955_equation_0, values = (var_10769_cast_fp16, var_10266_cast_fp16))[name = tensor("op_10955_cast_fp16")]; + tensor var_10956_to_fp16 = const()[name = tensor("op_10956_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_979_cast_fp16 = mul(x = var_10955_cast_fp16, y = var_10956_to_fp16)[name = tensor("aw_chunk_979_cast_fp16")]; + tensor var_10959_equation_0 = const()[name = tensor("op_10959_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10959_cast_fp16 = einsum(equation = var_10959_equation_0, values = (var_10769_cast_fp16, var_10273_cast_fp16))[name = tensor("op_10959_cast_fp16")]; + tensor var_10960_to_fp16 = const()[name = tensor("op_10960_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_981_cast_fp16 = mul(x = var_10959_cast_fp16, y = var_10960_to_fp16)[name = tensor("aw_chunk_981_cast_fp16")]; + tensor var_10963_equation_0 = const()[name = tensor("op_10963_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10963_cast_fp16 = einsum(equation = var_10963_equation_0, values = (var_10769_cast_fp16, var_10280_cast_fp16))[name = tensor("op_10963_cast_fp16")]; + tensor var_10964_to_fp16 = const()[name = tensor("op_10964_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_983_cast_fp16 = mul(x = var_10963_cast_fp16, y = var_10964_to_fp16)[name = tensor("aw_chunk_983_cast_fp16")]; + tensor var_10967_equation_0 = const()[name = tensor("op_10967_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10967_cast_fp16 = einsum(equation = var_10967_equation_0, values = (var_10773_cast_fp16, var_10287_cast_fp16))[name = tensor("op_10967_cast_fp16")]; + tensor var_10968_to_fp16 = const()[name = tensor("op_10968_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_985_cast_fp16 = mul(x = var_10967_cast_fp16, y = var_10968_to_fp16)[name = tensor("aw_chunk_985_cast_fp16")]; + tensor var_10971_equation_0 = const()[name = tensor("op_10971_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10971_cast_fp16 = einsum(equation = var_10971_equation_0, values = (var_10773_cast_fp16, var_10294_cast_fp16))[name = tensor("op_10971_cast_fp16")]; + tensor var_10972_to_fp16 = const()[name = tensor("op_10972_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_987_cast_fp16 = mul(x = var_10971_cast_fp16, y = var_10972_to_fp16)[name = tensor("aw_chunk_987_cast_fp16")]; + tensor var_10975_equation_0 = const()[name = tensor("op_10975_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10975_cast_fp16 = einsum(equation = var_10975_equation_0, values = (var_10773_cast_fp16, var_10301_cast_fp16))[name = tensor("op_10975_cast_fp16")]; + tensor var_10976_to_fp16 = const()[name = tensor("op_10976_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_989_cast_fp16 = mul(x = var_10975_cast_fp16, y = var_10976_to_fp16)[name = tensor("aw_chunk_989_cast_fp16")]; + tensor var_10979_equation_0 = const()[name = tensor("op_10979_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10979_cast_fp16 = einsum(equation = var_10979_equation_0, values = (var_10773_cast_fp16, var_10308_cast_fp16))[name = tensor("op_10979_cast_fp16")]; + tensor var_10980_to_fp16 = const()[name = tensor("op_10980_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_991_cast_fp16 = mul(x = var_10979_cast_fp16, y = var_10980_to_fp16)[name = tensor("aw_chunk_991_cast_fp16")]; + tensor var_10983_equation_0 = const()[name = tensor("op_10983_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10983_cast_fp16 = einsum(equation = var_10983_equation_0, values = (var_10777_cast_fp16, var_10315_cast_fp16))[name = tensor("op_10983_cast_fp16")]; + tensor var_10984_to_fp16 = const()[name = tensor("op_10984_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_993_cast_fp16 = mul(x = var_10983_cast_fp16, y = var_10984_to_fp16)[name = tensor("aw_chunk_993_cast_fp16")]; + tensor var_10987_equation_0 = const()[name = tensor("op_10987_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10987_cast_fp16 = einsum(equation = var_10987_equation_0, values = (var_10777_cast_fp16, var_10322_cast_fp16))[name = tensor("op_10987_cast_fp16")]; + tensor var_10988_to_fp16 = const()[name = tensor("op_10988_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_995_cast_fp16 = mul(x = var_10987_cast_fp16, y = var_10988_to_fp16)[name = tensor("aw_chunk_995_cast_fp16")]; + tensor var_10991_equation_0 = const()[name = tensor("op_10991_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10991_cast_fp16 = einsum(equation = var_10991_equation_0, values = (var_10777_cast_fp16, var_10329_cast_fp16))[name = tensor("op_10991_cast_fp16")]; + tensor var_10992_to_fp16 = const()[name = tensor("op_10992_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_997_cast_fp16 = mul(x = var_10991_cast_fp16, y = var_10992_to_fp16)[name = tensor("aw_chunk_997_cast_fp16")]; + tensor var_10995_equation_0 = const()[name = tensor("op_10995_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10995_cast_fp16 = einsum(equation = var_10995_equation_0, values = (var_10777_cast_fp16, var_10336_cast_fp16))[name = tensor("op_10995_cast_fp16")]; + tensor var_10996_to_fp16 = const()[name = tensor("op_10996_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_999_cast_fp16 = mul(x = var_10995_cast_fp16, y = var_10996_to_fp16)[name = tensor("aw_chunk_999_cast_fp16")]; + tensor var_10999_equation_0 = const()[name = tensor("op_10999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_10999_cast_fp16 = einsum(equation = var_10999_equation_0, values = (var_10781_cast_fp16, var_10343_cast_fp16))[name = tensor("op_10999_cast_fp16")]; + tensor var_11000_to_fp16 = const()[name = tensor("op_11000_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1001_cast_fp16 = mul(x = var_10999_cast_fp16, y = var_11000_to_fp16)[name = tensor("aw_chunk_1001_cast_fp16")]; + tensor var_11003_equation_0 = const()[name = tensor("op_11003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11003_cast_fp16 = einsum(equation = var_11003_equation_0, values = (var_10781_cast_fp16, var_10350_cast_fp16))[name = tensor("op_11003_cast_fp16")]; + tensor var_11004_to_fp16 = const()[name = tensor("op_11004_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1003_cast_fp16 = mul(x = var_11003_cast_fp16, y = var_11004_to_fp16)[name = tensor("aw_chunk_1003_cast_fp16")]; + tensor var_11007_equation_0 = const()[name = tensor("op_11007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11007_cast_fp16 = einsum(equation = var_11007_equation_0, values = (var_10781_cast_fp16, var_10357_cast_fp16))[name = tensor("op_11007_cast_fp16")]; + tensor var_11008_to_fp16 = const()[name = tensor("op_11008_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1005_cast_fp16 = mul(x = var_11007_cast_fp16, y = var_11008_to_fp16)[name = tensor("aw_chunk_1005_cast_fp16")]; + tensor var_11011_equation_0 = const()[name = tensor("op_11011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11011_cast_fp16 = einsum(equation = var_11011_equation_0, values = (var_10781_cast_fp16, var_10364_cast_fp16))[name = tensor("op_11011_cast_fp16")]; + tensor var_11012_to_fp16 = const()[name = tensor("op_11012_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1007_cast_fp16 = mul(x = var_11011_cast_fp16, y = var_11012_to_fp16)[name = tensor("aw_chunk_1007_cast_fp16")]; + tensor var_11015_equation_0 = const()[name = tensor("op_11015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11015_cast_fp16 = einsum(equation = var_11015_equation_0, values = (var_10785_cast_fp16, var_10371_cast_fp16))[name = tensor("op_11015_cast_fp16")]; + tensor var_11016_to_fp16 = const()[name = tensor("op_11016_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1009_cast_fp16 = mul(x = var_11015_cast_fp16, y = var_11016_to_fp16)[name = tensor("aw_chunk_1009_cast_fp16")]; + tensor var_11019_equation_0 = const()[name = tensor("op_11019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11019_cast_fp16 = einsum(equation = var_11019_equation_0, values = (var_10785_cast_fp16, var_10378_cast_fp16))[name = tensor("op_11019_cast_fp16")]; + tensor var_11020_to_fp16 = const()[name = tensor("op_11020_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1011_cast_fp16 = mul(x = var_11019_cast_fp16, y = var_11020_to_fp16)[name = tensor("aw_chunk_1011_cast_fp16")]; + tensor var_11023_equation_0 = const()[name = tensor("op_11023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11023_cast_fp16 = einsum(equation = var_11023_equation_0, values = (var_10785_cast_fp16, var_10385_cast_fp16))[name = tensor("op_11023_cast_fp16")]; + tensor var_11024_to_fp16 = const()[name = tensor("op_11024_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1013_cast_fp16 = mul(x = var_11023_cast_fp16, y = var_11024_to_fp16)[name = tensor("aw_chunk_1013_cast_fp16")]; + tensor var_11027_equation_0 = const()[name = tensor("op_11027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11027_cast_fp16 = einsum(equation = var_11027_equation_0, values = (var_10785_cast_fp16, var_10392_cast_fp16))[name = tensor("op_11027_cast_fp16")]; + tensor var_11028_to_fp16 = const()[name = tensor("op_11028_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1015_cast_fp16 = mul(x = var_11027_cast_fp16, y = var_11028_to_fp16)[name = tensor("aw_chunk_1015_cast_fp16")]; + tensor var_11031_equation_0 = const()[name = tensor("op_11031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11031_cast_fp16 = einsum(equation = var_11031_equation_0, values = (var_10789_cast_fp16, var_10399_cast_fp16))[name = tensor("op_11031_cast_fp16")]; + tensor var_11032_to_fp16 = const()[name = tensor("op_11032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1017_cast_fp16 = mul(x = var_11031_cast_fp16, y = var_11032_to_fp16)[name = tensor("aw_chunk_1017_cast_fp16")]; + tensor var_11035_equation_0 = const()[name = tensor("op_11035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11035_cast_fp16 = einsum(equation = var_11035_equation_0, values = (var_10789_cast_fp16, var_10406_cast_fp16))[name = tensor("op_11035_cast_fp16")]; + tensor var_11036_to_fp16 = const()[name = tensor("op_11036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1019_cast_fp16 = mul(x = var_11035_cast_fp16, y = var_11036_to_fp16)[name = tensor("aw_chunk_1019_cast_fp16")]; + tensor var_11039_equation_0 = const()[name = tensor("op_11039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11039_cast_fp16 = einsum(equation = var_11039_equation_0, values = (var_10789_cast_fp16, var_10413_cast_fp16))[name = tensor("op_11039_cast_fp16")]; + tensor var_11040_to_fp16 = const()[name = tensor("op_11040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1021_cast_fp16 = mul(x = var_11039_cast_fp16, y = var_11040_to_fp16)[name = tensor("aw_chunk_1021_cast_fp16")]; + tensor var_11043_equation_0 = const()[name = tensor("op_11043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11043_cast_fp16 = einsum(equation = var_11043_equation_0, values = (var_10789_cast_fp16, var_10420_cast_fp16))[name = tensor("op_11043_cast_fp16")]; + tensor var_11044_to_fp16 = const()[name = tensor("op_11044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1023_cast_fp16 = mul(x = var_11043_cast_fp16, y = var_11044_to_fp16)[name = tensor("aw_chunk_1023_cast_fp16")]; + tensor var_11047_equation_0 = const()[name = tensor("op_11047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11047_cast_fp16 = einsum(equation = var_11047_equation_0, values = (var_10793_cast_fp16, var_10427_cast_fp16))[name = tensor("op_11047_cast_fp16")]; + tensor var_11048_to_fp16 = const()[name = tensor("op_11048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1025_cast_fp16 = mul(x = var_11047_cast_fp16, y = var_11048_to_fp16)[name = tensor("aw_chunk_1025_cast_fp16")]; + tensor var_11051_equation_0 = const()[name = tensor("op_11051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11051_cast_fp16 = einsum(equation = var_11051_equation_0, values = (var_10793_cast_fp16, var_10434_cast_fp16))[name = tensor("op_11051_cast_fp16")]; + tensor var_11052_to_fp16 = const()[name = tensor("op_11052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1027_cast_fp16 = mul(x = var_11051_cast_fp16, y = var_11052_to_fp16)[name = tensor("aw_chunk_1027_cast_fp16")]; + tensor var_11055_equation_0 = const()[name = tensor("op_11055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11055_cast_fp16 = einsum(equation = var_11055_equation_0, values = (var_10793_cast_fp16, var_10441_cast_fp16))[name = tensor("op_11055_cast_fp16")]; + tensor var_11056_to_fp16 = const()[name = tensor("op_11056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1029_cast_fp16 = mul(x = var_11055_cast_fp16, y = var_11056_to_fp16)[name = tensor("aw_chunk_1029_cast_fp16")]; + tensor var_11059_equation_0 = const()[name = tensor("op_11059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11059_cast_fp16 = einsum(equation = var_11059_equation_0, values = (var_10793_cast_fp16, var_10448_cast_fp16))[name = tensor("op_11059_cast_fp16")]; + tensor var_11060_to_fp16 = const()[name = tensor("op_11060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1031_cast_fp16 = mul(x = var_11059_cast_fp16, y = var_11060_to_fp16)[name = tensor("aw_chunk_1031_cast_fp16")]; + tensor var_11063_equation_0 = const()[name = tensor("op_11063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11063_cast_fp16 = einsum(equation = var_11063_equation_0, values = (var_10797_cast_fp16, var_10455_cast_fp16))[name = tensor("op_11063_cast_fp16")]; + tensor var_11064_to_fp16 = const()[name = tensor("op_11064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1033_cast_fp16 = mul(x = var_11063_cast_fp16, y = var_11064_to_fp16)[name = tensor("aw_chunk_1033_cast_fp16")]; + tensor var_11067_equation_0 = const()[name = tensor("op_11067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11067_cast_fp16 = einsum(equation = var_11067_equation_0, values = (var_10797_cast_fp16, var_10462_cast_fp16))[name = tensor("op_11067_cast_fp16")]; + tensor var_11068_to_fp16 = const()[name = tensor("op_11068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1035_cast_fp16 = mul(x = var_11067_cast_fp16, y = var_11068_to_fp16)[name = tensor("aw_chunk_1035_cast_fp16")]; + tensor var_11071_equation_0 = const()[name = tensor("op_11071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11071_cast_fp16 = einsum(equation = var_11071_equation_0, values = (var_10797_cast_fp16, var_10469_cast_fp16))[name = tensor("op_11071_cast_fp16")]; + tensor var_11072_to_fp16 = const()[name = tensor("op_11072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1037_cast_fp16 = mul(x = var_11071_cast_fp16, y = var_11072_to_fp16)[name = tensor("aw_chunk_1037_cast_fp16")]; + tensor var_11075_equation_0 = const()[name = tensor("op_11075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11075_cast_fp16 = einsum(equation = var_11075_equation_0, values = (var_10797_cast_fp16, var_10476_cast_fp16))[name = tensor("op_11075_cast_fp16")]; + tensor var_11076_to_fp16 = const()[name = tensor("op_11076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1039_cast_fp16 = mul(x = var_11075_cast_fp16, y = var_11076_to_fp16)[name = tensor("aw_chunk_1039_cast_fp16")]; + tensor var_11079_equation_0 = const()[name = tensor("op_11079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11079_cast_fp16 = einsum(equation = var_11079_equation_0, values = (var_10801_cast_fp16, var_10483_cast_fp16))[name = tensor("op_11079_cast_fp16")]; + tensor var_11080_to_fp16 = const()[name = tensor("op_11080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1041_cast_fp16 = mul(x = var_11079_cast_fp16, y = var_11080_to_fp16)[name = tensor("aw_chunk_1041_cast_fp16")]; + tensor var_11083_equation_0 = const()[name = tensor("op_11083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11083_cast_fp16 = einsum(equation = var_11083_equation_0, values = (var_10801_cast_fp16, var_10490_cast_fp16))[name = tensor("op_11083_cast_fp16")]; + tensor var_11084_to_fp16 = const()[name = tensor("op_11084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1043_cast_fp16 = mul(x = var_11083_cast_fp16, y = var_11084_to_fp16)[name = tensor("aw_chunk_1043_cast_fp16")]; + tensor var_11087_equation_0 = const()[name = tensor("op_11087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11087_cast_fp16 = einsum(equation = var_11087_equation_0, values = (var_10801_cast_fp16, var_10497_cast_fp16))[name = tensor("op_11087_cast_fp16")]; + tensor var_11088_to_fp16 = const()[name = tensor("op_11088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1045_cast_fp16 = mul(x = var_11087_cast_fp16, y = var_11088_to_fp16)[name = tensor("aw_chunk_1045_cast_fp16")]; + tensor var_11091_equation_0 = const()[name = tensor("op_11091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11091_cast_fp16 = einsum(equation = var_11091_equation_0, values = (var_10801_cast_fp16, var_10504_cast_fp16))[name = tensor("op_11091_cast_fp16")]; + tensor var_11092_to_fp16 = const()[name = tensor("op_11092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1047_cast_fp16 = mul(x = var_11091_cast_fp16, y = var_11092_to_fp16)[name = tensor("aw_chunk_1047_cast_fp16")]; + tensor var_11095_equation_0 = const()[name = tensor("op_11095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11095_cast_fp16 = einsum(equation = var_11095_equation_0, values = (var_10805_cast_fp16, var_10511_cast_fp16))[name = tensor("op_11095_cast_fp16")]; + tensor var_11096_to_fp16 = const()[name = tensor("op_11096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1049_cast_fp16 = mul(x = var_11095_cast_fp16, y = var_11096_to_fp16)[name = tensor("aw_chunk_1049_cast_fp16")]; + tensor var_11099_equation_0 = const()[name = tensor("op_11099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11099_cast_fp16 = einsum(equation = var_11099_equation_0, values = (var_10805_cast_fp16, var_10518_cast_fp16))[name = tensor("op_11099_cast_fp16")]; + tensor var_11100_to_fp16 = const()[name = tensor("op_11100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1051_cast_fp16 = mul(x = var_11099_cast_fp16, y = var_11100_to_fp16)[name = tensor("aw_chunk_1051_cast_fp16")]; + tensor var_11103_equation_0 = const()[name = tensor("op_11103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11103_cast_fp16 = einsum(equation = var_11103_equation_0, values = (var_10805_cast_fp16, var_10525_cast_fp16))[name = tensor("op_11103_cast_fp16")]; + tensor var_11104_to_fp16 = const()[name = tensor("op_11104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1053_cast_fp16 = mul(x = var_11103_cast_fp16, y = var_11104_to_fp16)[name = tensor("aw_chunk_1053_cast_fp16")]; + tensor var_11107_equation_0 = const()[name = tensor("op_11107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11107_cast_fp16 = einsum(equation = var_11107_equation_0, values = (var_10805_cast_fp16, var_10532_cast_fp16))[name = tensor("op_11107_cast_fp16")]; + tensor var_11108_to_fp16 = const()[name = tensor("op_11108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1055_cast_fp16 = mul(x = var_11107_cast_fp16, y = var_11108_to_fp16)[name = tensor("aw_chunk_1055_cast_fp16")]; + tensor var_11111_equation_0 = const()[name = tensor("op_11111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11111_cast_fp16 = einsum(equation = var_11111_equation_0, values = (var_10809_cast_fp16, var_10539_cast_fp16))[name = tensor("op_11111_cast_fp16")]; + tensor var_11112_to_fp16 = const()[name = tensor("op_11112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1057_cast_fp16 = mul(x = var_11111_cast_fp16, y = var_11112_to_fp16)[name = tensor("aw_chunk_1057_cast_fp16")]; + tensor var_11115_equation_0 = const()[name = tensor("op_11115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11115_cast_fp16 = einsum(equation = var_11115_equation_0, values = (var_10809_cast_fp16, var_10546_cast_fp16))[name = tensor("op_11115_cast_fp16")]; + tensor var_11116_to_fp16 = const()[name = tensor("op_11116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1059_cast_fp16 = mul(x = var_11115_cast_fp16, y = var_11116_to_fp16)[name = tensor("aw_chunk_1059_cast_fp16")]; + tensor var_11119_equation_0 = const()[name = tensor("op_11119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11119_cast_fp16 = einsum(equation = var_11119_equation_0, values = (var_10809_cast_fp16, var_10553_cast_fp16))[name = tensor("op_11119_cast_fp16")]; + tensor var_11120_to_fp16 = const()[name = tensor("op_11120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1061_cast_fp16 = mul(x = var_11119_cast_fp16, y = var_11120_to_fp16)[name = tensor("aw_chunk_1061_cast_fp16")]; + tensor var_11123_equation_0 = const()[name = tensor("op_11123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11123_cast_fp16 = einsum(equation = var_11123_equation_0, values = (var_10809_cast_fp16, var_10560_cast_fp16))[name = tensor("op_11123_cast_fp16")]; + tensor var_11124_to_fp16 = const()[name = tensor("op_11124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1063_cast_fp16 = mul(x = var_11123_cast_fp16, y = var_11124_to_fp16)[name = tensor("aw_chunk_1063_cast_fp16")]; + tensor var_11127_equation_0 = const()[name = tensor("op_11127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11127_cast_fp16 = einsum(equation = var_11127_equation_0, values = (var_10813_cast_fp16, var_10567_cast_fp16))[name = tensor("op_11127_cast_fp16")]; + tensor var_11128_to_fp16 = const()[name = tensor("op_11128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1065_cast_fp16 = mul(x = var_11127_cast_fp16, y = var_11128_to_fp16)[name = tensor("aw_chunk_1065_cast_fp16")]; + tensor var_11131_equation_0 = const()[name = tensor("op_11131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11131_cast_fp16 = einsum(equation = var_11131_equation_0, values = (var_10813_cast_fp16, var_10574_cast_fp16))[name = tensor("op_11131_cast_fp16")]; + tensor var_11132_to_fp16 = const()[name = tensor("op_11132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1067_cast_fp16 = mul(x = var_11131_cast_fp16, y = var_11132_to_fp16)[name = tensor("aw_chunk_1067_cast_fp16")]; + tensor var_11135_equation_0 = const()[name = tensor("op_11135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11135_cast_fp16 = einsum(equation = var_11135_equation_0, values = (var_10813_cast_fp16, var_10581_cast_fp16))[name = tensor("op_11135_cast_fp16")]; + tensor var_11136_to_fp16 = const()[name = tensor("op_11136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1069_cast_fp16 = mul(x = var_11135_cast_fp16, y = var_11136_to_fp16)[name = tensor("aw_chunk_1069_cast_fp16")]; + tensor var_11139_equation_0 = const()[name = tensor("op_11139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11139_cast_fp16 = einsum(equation = var_11139_equation_0, values = (var_10813_cast_fp16, var_10588_cast_fp16))[name = tensor("op_11139_cast_fp16")]; + tensor var_11140_to_fp16 = const()[name = tensor("op_11140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1071_cast_fp16 = mul(x = var_11139_cast_fp16, y = var_11140_to_fp16)[name = tensor("aw_chunk_1071_cast_fp16")]; + tensor var_11143_equation_0 = const()[name = tensor("op_11143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11143_cast_fp16 = einsum(equation = var_11143_equation_0, values = (var_10817_cast_fp16, var_10595_cast_fp16))[name = tensor("op_11143_cast_fp16")]; + tensor var_11144_to_fp16 = const()[name = tensor("op_11144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1073_cast_fp16 = mul(x = var_11143_cast_fp16, y = var_11144_to_fp16)[name = tensor("aw_chunk_1073_cast_fp16")]; + tensor var_11147_equation_0 = const()[name = tensor("op_11147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11147_cast_fp16 = einsum(equation = var_11147_equation_0, values = (var_10817_cast_fp16, var_10602_cast_fp16))[name = tensor("op_11147_cast_fp16")]; + tensor var_11148_to_fp16 = const()[name = tensor("op_11148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1075_cast_fp16 = mul(x = var_11147_cast_fp16, y = var_11148_to_fp16)[name = tensor("aw_chunk_1075_cast_fp16")]; + tensor var_11151_equation_0 = const()[name = tensor("op_11151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11151_cast_fp16 = einsum(equation = var_11151_equation_0, values = (var_10817_cast_fp16, var_10609_cast_fp16))[name = tensor("op_11151_cast_fp16")]; + tensor var_11152_to_fp16 = const()[name = tensor("op_11152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1077_cast_fp16 = mul(x = var_11151_cast_fp16, y = var_11152_to_fp16)[name = tensor("aw_chunk_1077_cast_fp16")]; + tensor var_11155_equation_0 = const()[name = tensor("op_11155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11155_cast_fp16 = einsum(equation = var_11155_equation_0, values = (var_10817_cast_fp16, var_10616_cast_fp16))[name = tensor("op_11155_cast_fp16")]; + tensor var_11156_to_fp16 = const()[name = tensor("op_11156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1079_cast_fp16 = mul(x = var_11155_cast_fp16, y = var_11156_to_fp16)[name = tensor("aw_chunk_1079_cast_fp16")]; + tensor var_11159_equation_0 = const()[name = tensor("op_11159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11159_cast_fp16 = einsum(equation = var_11159_equation_0, values = (var_10821_cast_fp16, var_10623_cast_fp16))[name = tensor("op_11159_cast_fp16")]; + tensor var_11160_to_fp16 = const()[name = tensor("op_11160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1081_cast_fp16 = mul(x = var_11159_cast_fp16, y = var_11160_to_fp16)[name = tensor("aw_chunk_1081_cast_fp16")]; + tensor var_11163_equation_0 = const()[name = tensor("op_11163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11163_cast_fp16 = einsum(equation = var_11163_equation_0, values = (var_10821_cast_fp16, var_10630_cast_fp16))[name = tensor("op_11163_cast_fp16")]; + tensor var_11164_to_fp16 = const()[name = tensor("op_11164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1083_cast_fp16 = mul(x = var_11163_cast_fp16, y = var_11164_to_fp16)[name = tensor("aw_chunk_1083_cast_fp16")]; + tensor var_11167_equation_0 = const()[name = tensor("op_11167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11167_cast_fp16 = einsum(equation = var_11167_equation_0, values = (var_10821_cast_fp16, var_10637_cast_fp16))[name = tensor("op_11167_cast_fp16")]; + tensor var_11168_to_fp16 = const()[name = tensor("op_11168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1085_cast_fp16 = mul(x = var_11167_cast_fp16, y = var_11168_to_fp16)[name = tensor("aw_chunk_1085_cast_fp16")]; + tensor var_11171_equation_0 = const()[name = tensor("op_11171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11171_cast_fp16 = einsum(equation = var_11171_equation_0, values = (var_10821_cast_fp16, var_10644_cast_fp16))[name = tensor("op_11171_cast_fp16")]; + tensor var_11172_to_fp16 = const()[name = tensor("op_11172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1087_cast_fp16 = mul(x = var_11171_cast_fp16, y = var_11172_to_fp16)[name = tensor("aw_chunk_1087_cast_fp16")]; + tensor var_11175_equation_0 = const()[name = tensor("op_11175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11175_cast_fp16 = einsum(equation = var_11175_equation_0, values = (var_10825_cast_fp16, var_10651_cast_fp16))[name = tensor("op_11175_cast_fp16")]; + tensor var_11176_to_fp16 = const()[name = tensor("op_11176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1089_cast_fp16 = mul(x = var_11175_cast_fp16, y = var_11176_to_fp16)[name = tensor("aw_chunk_1089_cast_fp16")]; + tensor var_11179_equation_0 = const()[name = tensor("op_11179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11179_cast_fp16 = einsum(equation = var_11179_equation_0, values = (var_10825_cast_fp16, var_10658_cast_fp16))[name = tensor("op_11179_cast_fp16")]; + tensor var_11180_to_fp16 = const()[name = tensor("op_11180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1091_cast_fp16 = mul(x = var_11179_cast_fp16, y = var_11180_to_fp16)[name = tensor("aw_chunk_1091_cast_fp16")]; + tensor var_11183_equation_0 = const()[name = tensor("op_11183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11183_cast_fp16 = einsum(equation = var_11183_equation_0, values = (var_10825_cast_fp16, var_10665_cast_fp16))[name = tensor("op_11183_cast_fp16")]; + tensor var_11184_to_fp16 = const()[name = tensor("op_11184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1093_cast_fp16 = mul(x = var_11183_cast_fp16, y = var_11184_to_fp16)[name = tensor("aw_chunk_1093_cast_fp16")]; + tensor var_11187_equation_0 = const()[name = tensor("op_11187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11187_cast_fp16 = einsum(equation = var_11187_equation_0, values = (var_10825_cast_fp16, var_10672_cast_fp16))[name = tensor("op_11187_cast_fp16")]; + tensor var_11188_to_fp16 = const()[name = tensor("op_11188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1095_cast_fp16 = mul(x = var_11187_cast_fp16, y = var_11188_to_fp16)[name = tensor("aw_chunk_1095_cast_fp16")]; + tensor var_11191_equation_0 = const()[name = tensor("op_11191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11191_cast_fp16 = einsum(equation = var_11191_equation_0, values = (var_10829_cast_fp16, var_10679_cast_fp16))[name = tensor("op_11191_cast_fp16")]; + tensor var_11192_to_fp16 = const()[name = tensor("op_11192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1097_cast_fp16 = mul(x = var_11191_cast_fp16, y = var_11192_to_fp16)[name = tensor("aw_chunk_1097_cast_fp16")]; + tensor var_11195_equation_0 = const()[name = tensor("op_11195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11195_cast_fp16 = einsum(equation = var_11195_equation_0, values = (var_10829_cast_fp16, var_10686_cast_fp16))[name = tensor("op_11195_cast_fp16")]; + tensor var_11196_to_fp16 = const()[name = tensor("op_11196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1099_cast_fp16 = mul(x = var_11195_cast_fp16, y = var_11196_to_fp16)[name = tensor("aw_chunk_1099_cast_fp16")]; + tensor var_11199_equation_0 = const()[name = tensor("op_11199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11199_cast_fp16 = einsum(equation = var_11199_equation_0, values = (var_10829_cast_fp16, var_10693_cast_fp16))[name = tensor("op_11199_cast_fp16")]; + tensor var_11200_to_fp16 = const()[name = tensor("op_11200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1101_cast_fp16 = mul(x = var_11199_cast_fp16, y = var_11200_to_fp16)[name = tensor("aw_chunk_1101_cast_fp16")]; + tensor var_11203_equation_0 = const()[name = tensor("op_11203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11203_cast_fp16 = einsum(equation = var_11203_equation_0, values = (var_10829_cast_fp16, var_10700_cast_fp16))[name = tensor("op_11203_cast_fp16")]; + tensor var_11204_to_fp16 = const()[name = tensor("op_11204_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1103_cast_fp16 = mul(x = var_11203_cast_fp16, y = var_11204_to_fp16)[name = tensor("aw_chunk_1103_cast_fp16")]; + tensor var_11207_equation_0 = const()[name = tensor("op_11207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11207_cast_fp16 = einsum(equation = var_11207_equation_0, values = (var_10833_cast_fp16, var_10707_cast_fp16))[name = tensor("op_11207_cast_fp16")]; + tensor var_11208_to_fp16 = const()[name = tensor("op_11208_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1105_cast_fp16 = mul(x = var_11207_cast_fp16, y = var_11208_to_fp16)[name = tensor("aw_chunk_1105_cast_fp16")]; + tensor var_11211_equation_0 = const()[name = tensor("op_11211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11211_cast_fp16 = einsum(equation = var_11211_equation_0, values = (var_10833_cast_fp16, var_10714_cast_fp16))[name = tensor("op_11211_cast_fp16")]; + tensor var_11212_to_fp16 = const()[name = tensor("op_11212_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1107_cast_fp16 = mul(x = var_11211_cast_fp16, y = var_11212_to_fp16)[name = tensor("aw_chunk_1107_cast_fp16")]; + tensor var_11215_equation_0 = const()[name = tensor("op_11215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11215_cast_fp16 = einsum(equation = var_11215_equation_0, values = (var_10833_cast_fp16, var_10721_cast_fp16))[name = tensor("op_11215_cast_fp16")]; + tensor var_11216_to_fp16 = const()[name = tensor("op_11216_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1109_cast_fp16 = mul(x = var_11215_cast_fp16, y = var_11216_to_fp16)[name = tensor("aw_chunk_1109_cast_fp16")]; + tensor var_11219_equation_0 = const()[name = tensor("op_11219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11219_cast_fp16 = einsum(equation = var_11219_equation_0, values = (var_10833_cast_fp16, var_10728_cast_fp16))[name = tensor("op_11219_cast_fp16")]; + tensor var_11220_to_fp16 = const()[name = tensor("op_11220_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1111_cast_fp16 = mul(x = var_11219_cast_fp16, y = var_11220_to_fp16)[name = tensor("aw_chunk_1111_cast_fp16")]; + tensor var_11223_equation_0 = const()[name = tensor("op_11223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11223_cast_fp16 = einsum(equation = var_11223_equation_0, values = (var_10837_cast_fp16, var_10735_cast_fp16))[name = tensor("op_11223_cast_fp16")]; + tensor var_11224_to_fp16 = const()[name = tensor("op_11224_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1113_cast_fp16 = mul(x = var_11223_cast_fp16, y = var_11224_to_fp16)[name = tensor("aw_chunk_1113_cast_fp16")]; + tensor var_11227_equation_0 = const()[name = tensor("op_11227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11227_cast_fp16 = einsum(equation = var_11227_equation_0, values = (var_10837_cast_fp16, var_10742_cast_fp16))[name = tensor("op_11227_cast_fp16")]; + tensor var_11228_to_fp16 = const()[name = tensor("op_11228_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1115_cast_fp16 = mul(x = var_11227_cast_fp16, y = var_11228_to_fp16)[name = tensor("aw_chunk_1115_cast_fp16")]; + tensor var_11231_equation_0 = const()[name = tensor("op_11231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11231_cast_fp16 = einsum(equation = var_11231_equation_0, values = (var_10837_cast_fp16, var_10749_cast_fp16))[name = tensor("op_11231_cast_fp16")]; + tensor var_11232_to_fp16 = const()[name = tensor("op_11232_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1117_cast_fp16 = mul(x = var_11231_cast_fp16, y = var_11232_to_fp16)[name = tensor("aw_chunk_1117_cast_fp16")]; + tensor var_11235_equation_0 = const()[name = tensor("op_11235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_11235_cast_fp16 = einsum(equation = var_11235_equation_0, values = (var_10837_cast_fp16, var_10756_cast_fp16))[name = tensor("op_11235_cast_fp16")]; + tensor var_11236_to_fp16 = const()[name = tensor("op_11236_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1119_cast_fp16 = mul(x = var_11235_cast_fp16, y = var_11236_to_fp16)[name = tensor("aw_chunk_1119_cast_fp16")]; + tensor var_11238_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_961_cast_fp16)[name = tensor("op_11238_cast_fp16")]; + tensor var_11239_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_963_cast_fp16)[name = tensor("op_11239_cast_fp16")]; + tensor var_11240_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_965_cast_fp16)[name = tensor("op_11240_cast_fp16")]; + tensor var_11241_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_967_cast_fp16)[name = tensor("op_11241_cast_fp16")]; + tensor var_11242_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_969_cast_fp16)[name = tensor("op_11242_cast_fp16")]; + tensor var_11243_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_971_cast_fp16)[name = tensor("op_11243_cast_fp16")]; + tensor var_11244_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_973_cast_fp16)[name = tensor("op_11244_cast_fp16")]; + tensor var_11245_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_975_cast_fp16)[name = tensor("op_11245_cast_fp16")]; + tensor var_11246_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_977_cast_fp16)[name = tensor("op_11246_cast_fp16")]; + tensor var_11247_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_979_cast_fp16)[name = tensor("op_11247_cast_fp16")]; + tensor var_11248_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_981_cast_fp16)[name = tensor("op_11248_cast_fp16")]; + tensor var_11249_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_983_cast_fp16)[name = tensor("op_11249_cast_fp16")]; + tensor var_11250_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_985_cast_fp16)[name = tensor("op_11250_cast_fp16")]; + tensor var_11251_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_987_cast_fp16)[name = tensor("op_11251_cast_fp16")]; + tensor var_11252_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_989_cast_fp16)[name = tensor("op_11252_cast_fp16")]; + tensor var_11253_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_991_cast_fp16)[name = tensor("op_11253_cast_fp16")]; + tensor var_11254_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_993_cast_fp16)[name = tensor("op_11254_cast_fp16")]; + tensor var_11255_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_995_cast_fp16)[name = tensor("op_11255_cast_fp16")]; + tensor var_11256_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_997_cast_fp16)[name = tensor("op_11256_cast_fp16")]; + tensor var_11257_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_999_cast_fp16)[name = tensor("op_11257_cast_fp16")]; + tensor var_11258_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1001_cast_fp16)[name = tensor("op_11258_cast_fp16")]; + tensor var_11259_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1003_cast_fp16)[name = tensor("op_11259_cast_fp16")]; + tensor var_11260_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1005_cast_fp16)[name = tensor("op_11260_cast_fp16")]; + tensor var_11261_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1007_cast_fp16)[name = tensor("op_11261_cast_fp16")]; + tensor var_11262_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1009_cast_fp16)[name = tensor("op_11262_cast_fp16")]; + tensor var_11263_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1011_cast_fp16)[name = tensor("op_11263_cast_fp16")]; + tensor var_11264_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1013_cast_fp16)[name = tensor("op_11264_cast_fp16")]; + tensor var_11265_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1015_cast_fp16)[name = tensor("op_11265_cast_fp16")]; + tensor var_11266_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1017_cast_fp16)[name = tensor("op_11266_cast_fp16")]; + tensor var_11267_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1019_cast_fp16)[name = tensor("op_11267_cast_fp16")]; + tensor var_11268_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1021_cast_fp16)[name = tensor("op_11268_cast_fp16")]; + tensor var_11269_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1023_cast_fp16)[name = tensor("op_11269_cast_fp16")]; + tensor var_11270_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1025_cast_fp16)[name = tensor("op_11270_cast_fp16")]; + tensor var_11271_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1027_cast_fp16)[name = tensor("op_11271_cast_fp16")]; + tensor var_11272_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1029_cast_fp16)[name = tensor("op_11272_cast_fp16")]; + tensor var_11273_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1031_cast_fp16)[name = tensor("op_11273_cast_fp16")]; + tensor var_11274_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1033_cast_fp16)[name = tensor("op_11274_cast_fp16")]; + tensor var_11275_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1035_cast_fp16)[name = tensor("op_11275_cast_fp16")]; + tensor var_11276_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1037_cast_fp16)[name = tensor("op_11276_cast_fp16")]; + tensor var_11277_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1039_cast_fp16)[name = tensor("op_11277_cast_fp16")]; + tensor var_11278_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1041_cast_fp16)[name = tensor("op_11278_cast_fp16")]; + tensor var_11279_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1043_cast_fp16)[name = tensor("op_11279_cast_fp16")]; + tensor var_11280_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1045_cast_fp16)[name = tensor("op_11280_cast_fp16")]; + tensor var_11281_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1047_cast_fp16)[name = tensor("op_11281_cast_fp16")]; + tensor var_11282_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1049_cast_fp16)[name = tensor("op_11282_cast_fp16")]; + tensor var_11283_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1051_cast_fp16)[name = tensor("op_11283_cast_fp16")]; + tensor var_11284_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1053_cast_fp16)[name = tensor("op_11284_cast_fp16")]; + tensor var_11285_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1055_cast_fp16)[name = tensor("op_11285_cast_fp16")]; + tensor var_11286_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1057_cast_fp16)[name = tensor("op_11286_cast_fp16")]; + tensor var_11287_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1059_cast_fp16)[name = tensor("op_11287_cast_fp16")]; + tensor var_11288_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1061_cast_fp16)[name = tensor("op_11288_cast_fp16")]; + tensor var_11289_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1063_cast_fp16)[name = tensor("op_11289_cast_fp16")]; + tensor var_11290_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1065_cast_fp16)[name = tensor("op_11290_cast_fp16")]; + tensor var_11291_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1067_cast_fp16)[name = tensor("op_11291_cast_fp16")]; + tensor var_11292_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1069_cast_fp16)[name = tensor("op_11292_cast_fp16")]; + tensor var_11293_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1071_cast_fp16)[name = tensor("op_11293_cast_fp16")]; + tensor var_11294_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1073_cast_fp16)[name = tensor("op_11294_cast_fp16")]; + tensor var_11295_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1075_cast_fp16)[name = tensor("op_11295_cast_fp16")]; + tensor var_11296_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1077_cast_fp16)[name = tensor("op_11296_cast_fp16")]; + tensor var_11297_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1079_cast_fp16)[name = tensor("op_11297_cast_fp16")]; + tensor var_11298_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1081_cast_fp16)[name = tensor("op_11298_cast_fp16")]; + tensor var_11299_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1083_cast_fp16)[name = tensor("op_11299_cast_fp16")]; + tensor var_11300_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1085_cast_fp16)[name = tensor("op_11300_cast_fp16")]; + tensor var_11301_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1087_cast_fp16)[name = tensor("op_11301_cast_fp16")]; + tensor var_11302_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1089_cast_fp16)[name = tensor("op_11302_cast_fp16")]; + tensor var_11303_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1091_cast_fp16)[name = tensor("op_11303_cast_fp16")]; + tensor var_11304_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1093_cast_fp16)[name = tensor("op_11304_cast_fp16")]; + tensor var_11305_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1095_cast_fp16)[name = tensor("op_11305_cast_fp16")]; + tensor var_11306_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1097_cast_fp16)[name = tensor("op_11306_cast_fp16")]; + tensor var_11307_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1099_cast_fp16)[name = tensor("op_11307_cast_fp16")]; + tensor var_11308_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1101_cast_fp16)[name = tensor("op_11308_cast_fp16")]; + tensor var_11309_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1103_cast_fp16)[name = tensor("op_11309_cast_fp16")]; + tensor var_11310_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1105_cast_fp16)[name = tensor("op_11310_cast_fp16")]; + tensor var_11311_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1107_cast_fp16)[name = tensor("op_11311_cast_fp16")]; + tensor var_11312_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1109_cast_fp16)[name = tensor("op_11312_cast_fp16")]; + tensor var_11313_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1111_cast_fp16)[name = tensor("op_11313_cast_fp16")]; + tensor var_11314_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1113_cast_fp16)[name = tensor("op_11314_cast_fp16")]; + tensor var_11315_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1115_cast_fp16)[name = tensor("op_11315_cast_fp16")]; + tensor var_11316_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1117_cast_fp16)[name = tensor("op_11316_cast_fp16")]; + tensor var_11317_cast_fp16 = softmax(axis = var_10009, x = aw_chunk_1119_cast_fp16)[name = tensor("op_11317_cast_fp16")]; + tensor var_11319_equation_0 = const()[name = tensor("op_11319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11319_cast_fp16 = einsum(equation = var_11319_equation_0, values = (var_10839_cast_fp16, var_11238_cast_fp16))[name = tensor("op_11319_cast_fp16")]; + tensor var_11321_equation_0 = const()[name = tensor("op_11321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11321_cast_fp16 = einsum(equation = var_11321_equation_0, values = (var_10839_cast_fp16, var_11239_cast_fp16))[name = tensor("op_11321_cast_fp16")]; + tensor var_11323_equation_0 = const()[name = tensor("op_11323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11323_cast_fp16 = einsum(equation = var_11323_equation_0, values = (var_10839_cast_fp16, var_11240_cast_fp16))[name = tensor("op_11323_cast_fp16")]; + tensor var_11325_equation_0 = const()[name = tensor("op_11325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11325_cast_fp16 = einsum(equation = var_11325_equation_0, values = (var_10839_cast_fp16, var_11241_cast_fp16))[name = tensor("op_11325_cast_fp16")]; + tensor var_11327_equation_0 = const()[name = tensor("op_11327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11327_cast_fp16 = einsum(equation = var_11327_equation_0, values = (var_10843_cast_fp16, var_11242_cast_fp16))[name = tensor("op_11327_cast_fp16")]; + tensor var_11329_equation_0 = const()[name = tensor("op_11329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11329_cast_fp16 = einsum(equation = var_11329_equation_0, values = (var_10843_cast_fp16, var_11243_cast_fp16))[name = tensor("op_11329_cast_fp16")]; + tensor var_11331_equation_0 = const()[name = tensor("op_11331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11331_cast_fp16 = einsum(equation = var_11331_equation_0, values = (var_10843_cast_fp16, var_11244_cast_fp16))[name = tensor("op_11331_cast_fp16")]; + tensor var_11333_equation_0 = const()[name = tensor("op_11333_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11333_cast_fp16 = einsum(equation = var_11333_equation_0, values = (var_10843_cast_fp16, var_11245_cast_fp16))[name = tensor("op_11333_cast_fp16")]; + tensor var_11335_equation_0 = const()[name = tensor("op_11335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11335_cast_fp16 = einsum(equation = var_11335_equation_0, values = (var_10847_cast_fp16, var_11246_cast_fp16))[name = tensor("op_11335_cast_fp16")]; + tensor var_11337_equation_0 = const()[name = tensor("op_11337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11337_cast_fp16 = einsum(equation = var_11337_equation_0, values = (var_10847_cast_fp16, var_11247_cast_fp16))[name = tensor("op_11337_cast_fp16")]; + tensor var_11339_equation_0 = const()[name = tensor("op_11339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11339_cast_fp16 = einsum(equation = var_11339_equation_0, values = (var_10847_cast_fp16, var_11248_cast_fp16))[name = tensor("op_11339_cast_fp16")]; + tensor var_11341_equation_0 = const()[name = tensor("op_11341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11341_cast_fp16 = einsum(equation = var_11341_equation_0, values = (var_10847_cast_fp16, var_11249_cast_fp16))[name = tensor("op_11341_cast_fp16")]; + tensor var_11343_equation_0 = const()[name = tensor("op_11343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11343_cast_fp16 = einsum(equation = var_11343_equation_0, values = (var_10851_cast_fp16, var_11250_cast_fp16))[name = tensor("op_11343_cast_fp16")]; + tensor var_11345_equation_0 = const()[name = tensor("op_11345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11345_cast_fp16 = einsum(equation = var_11345_equation_0, values = (var_10851_cast_fp16, var_11251_cast_fp16))[name = tensor("op_11345_cast_fp16")]; + tensor var_11347_equation_0 = const()[name = tensor("op_11347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11347_cast_fp16 = einsum(equation = var_11347_equation_0, values = (var_10851_cast_fp16, var_11252_cast_fp16))[name = tensor("op_11347_cast_fp16")]; + tensor var_11349_equation_0 = const()[name = tensor("op_11349_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11349_cast_fp16 = einsum(equation = var_11349_equation_0, values = (var_10851_cast_fp16, var_11253_cast_fp16))[name = tensor("op_11349_cast_fp16")]; + tensor var_11351_equation_0 = const()[name = tensor("op_11351_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11351_cast_fp16 = einsum(equation = var_11351_equation_0, values = (var_10855_cast_fp16, var_11254_cast_fp16))[name = tensor("op_11351_cast_fp16")]; + tensor var_11353_equation_0 = const()[name = tensor("op_11353_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11353_cast_fp16 = einsum(equation = var_11353_equation_0, values = (var_10855_cast_fp16, var_11255_cast_fp16))[name = tensor("op_11353_cast_fp16")]; + tensor var_11355_equation_0 = const()[name = tensor("op_11355_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11355_cast_fp16 = einsum(equation = var_11355_equation_0, values = (var_10855_cast_fp16, var_11256_cast_fp16))[name = tensor("op_11355_cast_fp16")]; + tensor var_11357_equation_0 = const()[name = tensor("op_11357_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11357_cast_fp16 = einsum(equation = var_11357_equation_0, values = (var_10855_cast_fp16, var_11257_cast_fp16))[name = tensor("op_11357_cast_fp16")]; + tensor var_11359_equation_0 = const()[name = tensor("op_11359_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11359_cast_fp16 = einsum(equation = var_11359_equation_0, values = (var_10859_cast_fp16, var_11258_cast_fp16))[name = tensor("op_11359_cast_fp16")]; + tensor var_11361_equation_0 = const()[name = tensor("op_11361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11361_cast_fp16 = einsum(equation = var_11361_equation_0, values = (var_10859_cast_fp16, var_11259_cast_fp16))[name = tensor("op_11361_cast_fp16")]; + tensor var_11363_equation_0 = const()[name = tensor("op_11363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11363_cast_fp16 = einsum(equation = var_11363_equation_0, values = (var_10859_cast_fp16, var_11260_cast_fp16))[name = tensor("op_11363_cast_fp16")]; + tensor var_11365_equation_0 = const()[name = tensor("op_11365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11365_cast_fp16 = einsum(equation = var_11365_equation_0, values = (var_10859_cast_fp16, var_11261_cast_fp16))[name = tensor("op_11365_cast_fp16")]; + tensor var_11367_equation_0 = const()[name = tensor("op_11367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11367_cast_fp16 = einsum(equation = var_11367_equation_0, values = (var_10863_cast_fp16, var_11262_cast_fp16))[name = tensor("op_11367_cast_fp16")]; + tensor var_11369_equation_0 = const()[name = tensor("op_11369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11369_cast_fp16 = einsum(equation = var_11369_equation_0, values = (var_10863_cast_fp16, var_11263_cast_fp16))[name = tensor("op_11369_cast_fp16")]; + tensor var_11371_equation_0 = const()[name = tensor("op_11371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11371_cast_fp16 = einsum(equation = var_11371_equation_0, values = (var_10863_cast_fp16, var_11264_cast_fp16))[name = tensor("op_11371_cast_fp16")]; + tensor var_11373_equation_0 = const()[name = tensor("op_11373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11373_cast_fp16 = einsum(equation = var_11373_equation_0, values = (var_10863_cast_fp16, var_11265_cast_fp16))[name = tensor("op_11373_cast_fp16")]; + tensor var_11375_equation_0 = const()[name = tensor("op_11375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11375_cast_fp16 = einsum(equation = var_11375_equation_0, values = (var_10867_cast_fp16, var_11266_cast_fp16))[name = tensor("op_11375_cast_fp16")]; + tensor var_11377_equation_0 = const()[name = tensor("op_11377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11377_cast_fp16 = einsum(equation = var_11377_equation_0, values = (var_10867_cast_fp16, var_11267_cast_fp16))[name = tensor("op_11377_cast_fp16")]; + tensor var_11379_equation_0 = const()[name = tensor("op_11379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11379_cast_fp16 = einsum(equation = var_11379_equation_0, values = (var_10867_cast_fp16, var_11268_cast_fp16))[name = tensor("op_11379_cast_fp16")]; + tensor var_11381_equation_0 = const()[name = tensor("op_11381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11381_cast_fp16 = einsum(equation = var_11381_equation_0, values = (var_10867_cast_fp16, var_11269_cast_fp16))[name = tensor("op_11381_cast_fp16")]; + tensor var_11383_equation_0 = const()[name = tensor("op_11383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11383_cast_fp16 = einsum(equation = var_11383_equation_0, values = (var_10871_cast_fp16, var_11270_cast_fp16))[name = tensor("op_11383_cast_fp16")]; + tensor var_11385_equation_0 = const()[name = tensor("op_11385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11385_cast_fp16 = einsum(equation = var_11385_equation_0, values = (var_10871_cast_fp16, var_11271_cast_fp16))[name = tensor("op_11385_cast_fp16")]; + tensor var_11387_equation_0 = const()[name = tensor("op_11387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11387_cast_fp16 = einsum(equation = var_11387_equation_0, values = (var_10871_cast_fp16, var_11272_cast_fp16))[name = tensor("op_11387_cast_fp16")]; + tensor var_11389_equation_0 = const()[name = tensor("op_11389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11389_cast_fp16 = einsum(equation = var_11389_equation_0, values = (var_10871_cast_fp16, var_11273_cast_fp16))[name = tensor("op_11389_cast_fp16")]; + tensor var_11391_equation_0 = const()[name = tensor("op_11391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11391_cast_fp16 = einsum(equation = var_11391_equation_0, values = (var_10875_cast_fp16, var_11274_cast_fp16))[name = tensor("op_11391_cast_fp16")]; + tensor var_11393_equation_0 = const()[name = tensor("op_11393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11393_cast_fp16 = einsum(equation = var_11393_equation_0, values = (var_10875_cast_fp16, var_11275_cast_fp16))[name = tensor("op_11393_cast_fp16")]; + tensor var_11395_equation_0 = const()[name = tensor("op_11395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11395_cast_fp16 = einsum(equation = var_11395_equation_0, values = (var_10875_cast_fp16, var_11276_cast_fp16))[name = tensor("op_11395_cast_fp16")]; + tensor var_11397_equation_0 = const()[name = tensor("op_11397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11397_cast_fp16 = einsum(equation = var_11397_equation_0, values = (var_10875_cast_fp16, var_11277_cast_fp16))[name = tensor("op_11397_cast_fp16")]; + tensor var_11399_equation_0 = const()[name = tensor("op_11399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11399_cast_fp16 = einsum(equation = var_11399_equation_0, values = (var_10879_cast_fp16, var_11278_cast_fp16))[name = tensor("op_11399_cast_fp16")]; + tensor var_11401_equation_0 = const()[name = tensor("op_11401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11401_cast_fp16 = einsum(equation = var_11401_equation_0, values = (var_10879_cast_fp16, var_11279_cast_fp16))[name = tensor("op_11401_cast_fp16")]; + tensor var_11403_equation_0 = const()[name = tensor("op_11403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11403_cast_fp16 = einsum(equation = var_11403_equation_0, values = (var_10879_cast_fp16, var_11280_cast_fp16))[name = tensor("op_11403_cast_fp16")]; + tensor var_11405_equation_0 = const()[name = tensor("op_11405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11405_cast_fp16 = einsum(equation = var_11405_equation_0, values = (var_10879_cast_fp16, var_11281_cast_fp16))[name = tensor("op_11405_cast_fp16")]; + tensor var_11407_equation_0 = const()[name = tensor("op_11407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11407_cast_fp16 = einsum(equation = var_11407_equation_0, values = (var_10883_cast_fp16, var_11282_cast_fp16))[name = tensor("op_11407_cast_fp16")]; + tensor var_11409_equation_0 = const()[name = tensor("op_11409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11409_cast_fp16 = einsum(equation = var_11409_equation_0, values = (var_10883_cast_fp16, var_11283_cast_fp16))[name = tensor("op_11409_cast_fp16")]; + tensor var_11411_equation_0 = const()[name = tensor("op_11411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11411_cast_fp16 = einsum(equation = var_11411_equation_0, values = (var_10883_cast_fp16, var_11284_cast_fp16))[name = tensor("op_11411_cast_fp16")]; + tensor var_11413_equation_0 = const()[name = tensor("op_11413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11413_cast_fp16 = einsum(equation = var_11413_equation_0, values = (var_10883_cast_fp16, var_11285_cast_fp16))[name = tensor("op_11413_cast_fp16")]; + tensor var_11415_equation_0 = const()[name = tensor("op_11415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11415_cast_fp16 = einsum(equation = var_11415_equation_0, values = (var_10887_cast_fp16, var_11286_cast_fp16))[name = tensor("op_11415_cast_fp16")]; + tensor var_11417_equation_0 = const()[name = tensor("op_11417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11417_cast_fp16 = einsum(equation = var_11417_equation_0, values = (var_10887_cast_fp16, var_11287_cast_fp16))[name = tensor("op_11417_cast_fp16")]; + tensor var_11419_equation_0 = const()[name = tensor("op_11419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11419_cast_fp16 = einsum(equation = var_11419_equation_0, values = (var_10887_cast_fp16, var_11288_cast_fp16))[name = tensor("op_11419_cast_fp16")]; + tensor var_11421_equation_0 = const()[name = tensor("op_11421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11421_cast_fp16 = einsum(equation = var_11421_equation_0, values = (var_10887_cast_fp16, var_11289_cast_fp16))[name = tensor("op_11421_cast_fp16")]; + tensor var_11423_equation_0 = const()[name = tensor("op_11423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11423_cast_fp16 = einsum(equation = var_11423_equation_0, values = (var_10891_cast_fp16, var_11290_cast_fp16))[name = tensor("op_11423_cast_fp16")]; + tensor var_11425_equation_0 = const()[name = tensor("op_11425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11425_cast_fp16 = einsum(equation = var_11425_equation_0, values = (var_10891_cast_fp16, var_11291_cast_fp16))[name = tensor("op_11425_cast_fp16")]; + tensor var_11427_equation_0 = const()[name = tensor("op_11427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11427_cast_fp16 = einsum(equation = var_11427_equation_0, values = (var_10891_cast_fp16, var_11292_cast_fp16))[name = tensor("op_11427_cast_fp16")]; + tensor var_11429_equation_0 = const()[name = tensor("op_11429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11429_cast_fp16 = einsum(equation = var_11429_equation_0, values = (var_10891_cast_fp16, var_11293_cast_fp16))[name = tensor("op_11429_cast_fp16")]; + tensor var_11431_equation_0 = const()[name = tensor("op_11431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11431_cast_fp16 = einsum(equation = var_11431_equation_0, values = (var_10895_cast_fp16, var_11294_cast_fp16))[name = tensor("op_11431_cast_fp16")]; + tensor var_11433_equation_0 = const()[name = tensor("op_11433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11433_cast_fp16 = einsum(equation = var_11433_equation_0, values = (var_10895_cast_fp16, var_11295_cast_fp16))[name = tensor("op_11433_cast_fp16")]; + tensor var_11435_equation_0 = const()[name = tensor("op_11435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11435_cast_fp16 = einsum(equation = var_11435_equation_0, values = (var_10895_cast_fp16, var_11296_cast_fp16))[name = tensor("op_11435_cast_fp16")]; + tensor var_11437_equation_0 = const()[name = tensor("op_11437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11437_cast_fp16 = einsum(equation = var_11437_equation_0, values = (var_10895_cast_fp16, var_11297_cast_fp16))[name = tensor("op_11437_cast_fp16")]; + tensor var_11439_equation_0 = const()[name = tensor("op_11439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11439_cast_fp16 = einsum(equation = var_11439_equation_0, values = (var_10899_cast_fp16, var_11298_cast_fp16))[name = tensor("op_11439_cast_fp16")]; + tensor var_11441_equation_0 = const()[name = tensor("op_11441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11441_cast_fp16 = einsum(equation = var_11441_equation_0, values = (var_10899_cast_fp16, var_11299_cast_fp16))[name = tensor("op_11441_cast_fp16")]; + tensor var_11443_equation_0 = const()[name = tensor("op_11443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11443_cast_fp16 = einsum(equation = var_11443_equation_0, values = (var_10899_cast_fp16, var_11300_cast_fp16))[name = tensor("op_11443_cast_fp16")]; + tensor var_11445_equation_0 = const()[name = tensor("op_11445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11445_cast_fp16 = einsum(equation = var_11445_equation_0, values = (var_10899_cast_fp16, var_11301_cast_fp16))[name = tensor("op_11445_cast_fp16")]; + tensor var_11447_equation_0 = const()[name = tensor("op_11447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11447_cast_fp16 = einsum(equation = var_11447_equation_0, values = (var_10903_cast_fp16, var_11302_cast_fp16))[name = tensor("op_11447_cast_fp16")]; + tensor var_11449_equation_0 = const()[name = tensor("op_11449_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11449_cast_fp16 = einsum(equation = var_11449_equation_0, values = (var_10903_cast_fp16, var_11303_cast_fp16))[name = tensor("op_11449_cast_fp16")]; + tensor var_11451_equation_0 = const()[name = tensor("op_11451_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11451_cast_fp16 = einsum(equation = var_11451_equation_0, values = (var_10903_cast_fp16, var_11304_cast_fp16))[name = tensor("op_11451_cast_fp16")]; + tensor var_11453_equation_0 = const()[name = tensor("op_11453_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11453_cast_fp16 = einsum(equation = var_11453_equation_0, values = (var_10903_cast_fp16, var_11305_cast_fp16))[name = tensor("op_11453_cast_fp16")]; + tensor var_11455_equation_0 = const()[name = tensor("op_11455_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11455_cast_fp16 = einsum(equation = var_11455_equation_0, values = (var_10907_cast_fp16, var_11306_cast_fp16))[name = tensor("op_11455_cast_fp16")]; + tensor var_11457_equation_0 = const()[name = tensor("op_11457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11457_cast_fp16 = einsum(equation = var_11457_equation_0, values = (var_10907_cast_fp16, var_11307_cast_fp16))[name = tensor("op_11457_cast_fp16")]; + tensor var_11459_equation_0 = const()[name = tensor("op_11459_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11459_cast_fp16 = einsum(equation = var_11459_equation_0, values = (var_10907_cast_fp16, var_11308_cast_fp16))[name = tensor("op_11459_cast_fp16")]; + tensor var_11461_equation_0 = const()[name = tensor("op_11461_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11461_cast_fp16 = einsum(equation = var_11461_equation_0, values = (var_10907_cast_fp16, var_11309_cast_fp16))[name = tensor("op_11461_cast_fp16")]; + tensor var_11463_equation_0 = const()[name = tensor("op_11463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11463_cast_fp16 = einsum(equation = var_11463_equation_0, values = (var_10911_cast_fp16, var_11310_cast_fp16))[name = tensor("op_11463_cast_fp16")]; + tensor var_11465_equation_0 = const()[name = tensor("op_11465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11465_cast_fp16 = einsum(equation = var_11465_equation_0, values = (var_10911_cast_fp16, var_11311_cast_fp16))[name = tensor("op_11465_cast_fp16")]; + tensor var_11467_equation_0 = const()[name = tensor("op_11467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11467_cast_fp16 = einsum(equation = var_11467_equation_0, values = (var_10911_cast_fp16, var_11312_cast_fp16))[name = tensor("op_11467_cast_fp16")]; + tensor var_11469_equation_0 = const()[name = tensor("op_11469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11469_cast_fp16 = einsum(equation = var_11469_equation_0, values = (var_10911_cast_fp16, var_11313_cast_fp16))[name = tensor("op_11469_cast_fp16")]; + tensor var_11471_equation_0 = const()[name = tensor("op_11471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11471_cast_fp16 = einsum(equation = var_11471_equation_0, values = (var_10915_cast_fp16, var_11314_cast_fp16))[name = tensor("op_11471_cast_fp16")]; + tensor var_11473_equation_0 = const()[name = tensor("op_11473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11473_cast_fp16 = einsum(equation = var_11473_equation_0, values = (var_10915_cast_fp16, var_11315_cast_fp16))[name = tensor("op_11473_cast_fp16")]; + tensor var_11475_equation_0 = const()[name = tensor("op_11475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11475_cast_fp16 = einsum(equation = var_11475_equation_0, values = (var_10915_cast_fp16, var_11316_cast_fp16))[name = tensor("op_11475_cast_fp16")]; + tensor var_11477_equation_0 = const()[name = tensor("op_11477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_11477_cast_fp16 = einsum(equation = var_11477_equation_0, values = (var_10915_cast_fp16, var_11317_cast_fp16))[name = tensor("op_11477_cast_fp16")]; + tensor var_11479_interleave_0 = const()[name = tensor("op_11479_interleave_0"), val = tensor(false)]; + tensor var_11479_cast_fp16 = concat(axis = var_9984, interleave = var_11479_interleave_0, values = (var_11319_cast_fp16, var_11321_cast_fp16, var_11323_cast_fp16, var_11325_cast_fp16))[name = tensor("op_11479_cast_fp16")]; + tensor var_11481_interleave_0 = const()[name = tensor("op_11481_interleave_0"), val = tensor(false)]; + tensor var_11481_cast_fp16 = concat(axis = var_9984, interleave = var_11481_interleave_0, values = (var_11327_cast_fp16, var_11329_cast_fp16, var_11331_cast_fp16, var_11333_cast_fp16))[name = tensor("op_11481_cast_fp16")]; + tensor var_11483_interleave_0 = const()[name = tensor("op_11483_interleave_0"), val = tensor(false)]; + tensor var_11483_cast_fp16 = concat(axis = var_9984, interleave = var_11483_interleave_0, values = (var_11335_cast_fp16, var_11337_cast_fp16, var_11339_cast_fp16, var_11341_cast_fp16))[name = tensor("op_11483_cast_fp16")]; + tensor var_11485_interleave_0 = const()[name = tensor("op_11485_interleave_0"), val = tensor(false)]; + tensor var_11485_cast_fp16 = concat(axis = var_9984, interleave = var_11485_interleave_0, values = (var_11343_cast_fp16, var_11345_cast_fp16, var_11347_cast_fp16, var_11349_cast_fp16))[name = tensor("op_11485_cast_fp16")]; + tensor var_11487_interleave_0 = const()[name = tensor("op_11487_interleave_0"), val = tensor(false)]; + tensor var_11487_cast_fp16 = concat(axis = var_9984, interleave = var_11487_interleave_0, values = (var_11351_cast_fp16, var_11353_cast_fp16, var_11355_cast_fp16, var_11357_cast_fp16))[name = tensor("op_11487_cast_fp16")]; + tensor var_11489_interleave_0 = const()[name = tensor("op_11489_interleave_0"), val = tensor(false)]; + tensor var_11489_cast_fp16 = concat(axis = var_9984, interleave = var_11489_interleave_0, values = (var_11359_cast_fp16, var_11361_cast_fp16, var_11363_cast_fp16, var_11365_cast_fp16))[name = tensor("op_11489_cast_fp16")]; + tensor var_11491_interleave_0 = const()[name = tensor("op_11491_interleave_0"), val = tensor(false)]; + tensor var_11491_cast_fp16 = concat(axis = var_9984, interleave = var_11491_interleave_0, values = (var_11367_cast_fp16, var_11369_cast_fp16, var_11371_cast_fp16, var_11373_cast_fp16))[name = tensor("op_11491_cast_fp16")]; + tensor var_11493_interleave_0 = const()[name = tensor("op_11493_interleave_0"), val = tensor(false)]; + tensor var_11493_cast_fp16 = concat(axis = var_9984, interleave = var_11493_interleave_0, values = (var_11375_cast_fp16, var_11377_cast_fp16, var_11379_cast_fp16, var_11381_cast_fp16))[name = tensor("op_11493_cast_fp16")]; + tensor var_11495_interleave_0 = const()[name = tensor("op_11495_interleave_0"), val = tensor(false)]; + tensor var_11495_cast_fp16 = concat(axis = var_9984, interleave = var_11495_interleave_0, values = (var_11383_cast_fp16, var_11385_cast_fp16, var_11387_cast_fp16, var_11389_cast_fp16))[name = tensor("op_11495_cast_fp16")]; + tensor var_11497_interleave_0 = const()[name = tensor("op_11497_interleave_0"), val = tensor(false)]; + tensor var_11497_cast_fp16 = concat(axis = var_9984, interleave = var_11497_interleave_0, values = (var_11391_cast_fp16, var_11393_cast_fp16, var_11395_cast_fp16, var_11397_cast_fp16))[name = tensor("op_11497_cast_fp16")]; + tensor var_11499_interleave_0 = const()[name = tensor("op_11499_interleave_0"), val = tensor(false)]; + tensor var_11499_cast_fp16 = concat(axis = var_9984, interleave = var_11499_interleave_0, values = (var_11399_cast_fp16, var_11401_cast_fp16, var_11403_cast_fp16, var_11405_cast_fp16))[name = tensor("op_11499_cast_fp16")]; + tensor var_11501_interleave_0 = const()[name = tensor("op_11501_interleave_0"), val = tensor(false)]; + tensor var_11501_cast_fp16 = concat(axis = var_9984, interleave = var_11501_interleave_0, values = (var_11407_cast_fp16, var_11409_cast_fp16, var_11411_cast_fp16, var_11413_cast_fp16))[name = tensor("op_11501_cast_fp16")]; + tensor var_11503_interleave_0 = const()[name = tensor("op_11503_interleave_0"), val = tensor(false)]; + tensor var_11503_cast_fp16 = concat(axis = var_9984, interleave = var_11503_interleave_0, values = (var_11415_cast_fp16, var_11417_cast_fp16, var_11419_cast_fp16, var_11421_cast_fp16))[name = tensor("op_11503_cast_fp16")]; + tensor var_11505_interleave_0 = const()[name = tensor("op_11505_interleave_0"), val = tensor(false)]; + tensor var_11505_cast_fp16 = concat(axis = var_9984, interleave = var_11505_interleave_0, values = (var_11423_cast_fp16, var_11425_cast_fp16, var_11427_cast_fp16, var_11429_cast_fp16))[name = tensor("op_11505_cast_fp16")]; + tensor var_11507_interleave_0 = const()[name = tensor("op_11507_interleave_0"), val = tensor(false)]; + tensor var_11507_cast_fp16 = concat(axis = var_9984, interleave = var_11507_interleave_0, values = (var_11431_cast_fp16, var_11433_cast_fp16, var_11435_cast_fp16, var_11437_cast_fp16))[name = tensor("op_11507_cast_fp16")]; + tensor var_11509_interleave_0 = const()[name = tensor("op_11509_interleave_0"), val = tensor(false)]; + tensor var_11509_cast_fp16 = concat(axis = var_9984, interleave = var_11509_interleave_0, values = (var_11439_cast_fp16, var_11441_cast_fp16, var_11443_cast_fp16, var_11445_cast_fp16))[name = tensor("op_11509_cast_fp16")]; + tensor var_11511_interleave_0 = const()[name = tensor("op_11511_interleave_0"), val = tensor(false)]; + tensor var_11511_cast_fp16 = concat(axis = var_9984, interleave = var_11511_interleave_0, values = (var_11447_cast_fp16, var_11449_cast_fp16, var_11451_cast_fp16, var_11453_cast_fp16))[name = tensor("op_11511_cast_fp16")]; + tensor var_11513_interleave_0 = const()[name = tensor("op_11513_interleave_0"), val = tensor(false)]; + tensor var_11513_cast_fp16 = concat(axis = var_9984, interleave = var_11513_interleave_0, values = (var_11455_cast_fp16, var_11457_cast_fp16, var_11459_cast_fp16, var_11461_cast_fp16))[name = tensor("op_11513_cast_fp16")]; + tensor var_11515_interleave_0 = const()[name = tensor("op_11515_interleave_0"), val = tensor(false)]; + tensor var_11515_cast_fp16 = concat(axis = var_9984, interleave = var_11515_interleave_0, values = (var_11463_cast_fp16, var_11465_cast_fp16, var_11467_cast_fp16, var_11469_cast_fp16))[name = tensor("op_11515_cast_fp16")]; + tensor var_11517_interleave_0 = const()[name = tensor("op_11517_interleave_0"), val = tensor(false)]; + tensor var_11517_cast_fp16 = concat(axis = var_9984, interleave = var_11517_interleave_0, values = (var_11471_cast_fp16, var_11473_cast_fp16, var_11475_cast_fp16, var_11477_cast_fp16))[name = tensor("op_11517_cast_fp16")]; + tensor input_127_interleave_0 = const()[name = tensor("input_127_interleave_0"), val = tensor(false)]; + tensor input_127_cast_fp16 = concat(axis = var_10009, interleave = input_127_interleave_0, values = (var_11479_cast_fp16, var_11481_cast_fp16, var_11483_cast_fp16, var_11485_cast_fp16, var_11487_cast_fp16, var_11489_cast_fp16, var_11491_cast_fp16, var_11493_cast_fp16, var_11495_cast_fp16, var_11497_cast_fp16, var_11499_cast_fp16, var_11501_cast_fp16, var_11503_cast_fp16, var_11505_cast_fp16, var_11507_cast_fp16, var_11509_cast_fp16, var_11511_cast_fp16, var_11513_cast_fp16, var_11515_cast_fp16, var_11517_cast_fp16))[name = tensor("input_127_cast_fp16")]; + tensor var_11525 = const()[name = tensor("op_11525"), val = tensor([1, 1])]; + tensor var_11527 = const()[name = tensor("op_11527"), val = tensor([1, 1])]; + tensor pretrained_out_79_pad_type_0 = const()[name = tensor("pretrained_out_79_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_79_pad_0 = const()[name = tensor("pretrained_out_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80626048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81445312))), name = tensor("layers_6_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_6_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81445440)))]; + tensor pretrained_out_79_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_11527, groups = var_10009, pad = pretrained_out_79_pad_0, pad_type = pretrained_out_79_pad_type_0, strides = var_11525, weight = layers_6_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor("pretrained_out_79_cast_fp16")]; + tensor var_11531 = const()[name = tensor("op_11531"), val = tensor([1, 1])]; + tensor var_11533 = const()[name = tensor("op_11533"), val = tensor([1, 1])]; + tensor input_129_pad_type_0 = const()[name = tensor("input_129_pad_type_0"), val = tensor("custom")]; + tensor input_129_pad_0 = const()[name = tensor("input_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81448064)))]; + tensor input_129_cast_fp16 = conv(dilations = var_11533, groups = var_10009, pad = input_129_pad_0, pad_type = input_129_pad_type_0, strides = var_11531, weight = layers_6_self_attn_o_proj_loraA_weight_to_fp16, x = input_127_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor var_11537 = const()[name = tensor("op_11537"), val = tensor([1, 1])]; + tensor var_11539 = const()[name = tensor("op_11539"), val = tensor([1, 1])]; + tensor lora_out_157_pad_type_0 = const()[name = tensor("lora_out_157_pad_type_0"), val = tensor("custom")]; + tensor lora_out_157_pad_0 = const()[name = tensor("lora_out_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_159_weight_0_to_fp16 = const()[name = tensor("lora_out_159_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81489088)))]; + tensor lora_out_159_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_11539, groups = var_10009, pad = lora_out_157_pad_0, pad_type = lora_out_157_pad_type_0, strides = var_11537, weight = lora_out_159_weight_0_to_fp16, x = input_129_cast_fp16)[name = tensor("lora_out_159_cast_fp16")]; + tensor obj_27_cast_fp16 = add(x = pretrained_out_79_cast_fp16, y = lora_out_159_cast_fp16)[name = tensor("obj_27_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor var_11548 = const()[name = tensor("op_11548"), val = tensor([1])]; + tensor channels_mean_27_cast_fp16 = reduce_mean(axes = var_11548, keep_dims = var_10010, x = inputs_27_cast_fp16)[name = tensor("channels_mean_27_cast_fp16")]; + tensor zero_mean_27_cast_fp16 = sub(x = inputs_27_cast_fp16, y = channels_mean_27_cast_fp16)[name = tensor("zero_mean_27_cast_fp16")]; + tensor zero_mean_sq_27_cast_fp16 = mul(x = zero_mean_27_cast_fp16, y = zero_mean_27_cast_fp16)[name = tensor("zero_mean_sq_27_cast_fp16")]; + tensor var_11552 = const()[name = tensor("op_11552"), val = tensor([1])]; + tensor var_11553_cast_fp16 = reduce_mean(axes = var_11552, keep_dims = var_10010, x = zero_mean_sq_27_cast_fp16)[name = tensor("op_11553_cast_fp16")]; + tensor var_11554_to_fp16 = const()[name = tensor("op_11554_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11555_cast_fp16 = add(x = var_11553_cast_fp16, y = var_11554_to_fp16)[name = tensor("op_11555_cast_fp16")]; + tensor denom_27_epsilon_0 = const()[name = tensor("denom_27_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_27_cast_fp16 = rsqrt(epsilon = denom_27_epsilon_0, x = var_11555_cast_fp16)[name = tensor("denom_27_cast_fp16")]; + tensor out_27_cast_fp16 = mul(x = zero_mean_27_cast_fp16, y = denom_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; + tensor input_131_gamma_0_to_fp16 = const()[name = tensor("input_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81530112)))]; + tensor input_131_beta_0_to_fp16 = const()[name = tensor("input_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81532736)))]; + tensor input_131_epsilon_0_to_fp16 = const()[name = tensor("input_131_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor var_11569 = const()[name = tensor("op_11569"), val = tensor([1, 1])]; + tensor var_11571 = const()[name = tensor("op_11571"), val = tensor([1, 1])]; + tensor pretrained_out_81_pad_type_0 = const()[name = tensor("pretrained_out_81_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_81_pad_0 = const()[name = tensor("pretrained_out_81_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81535360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84812224))), name = tensor("layers_6_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_6_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_6_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84812352)))]; + tensor pretrained_out_81_cast_fp16 = conv(bias = layers_6_fc1_pretrained_bias_to_fp16, dilations = var_11571, groups = var_10009, pad = pretrained_out_81_pad_0, pad_type = pretrained_out_81_pad_type_0, strides = var_11569, weight = layers_6_fc1_pretrained_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = tensor("pretrained_out_81_cast_fp16")]; + tensor var_11575 = const()[name = tensor("op_11575"), val = tensor([1, 1])]; + tensor var_11577 = const()[name = tensor("op_11577"), val = tensor([1, 1])]; + tensor input_133_pad_type_0 = const()[name = tensor("input_133_pad_type_0"), val = tensor("custom")]; + tensor input_133_pad_0 = const()[name = tensor("input_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_6_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84822656)))]; + tensor input_133_cast_fp16 = conv(dilations = var_11577, groups = var_10009, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = var_11575, weight = layers_6_fc1_loraA_weight_to_fp16, x = input_131_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor var_11581 = const()[name = tensor("op_11581"), val = tensor([1, 1])]; + tensor var_11583 = const()[name = tensor("op_11583"), val = tensor([1, 1])]; + tensor lora_out_161_pad_type_0 = const()[name = tensor("lora_out_161_pad_type_0"), val = tensor("custom")]; + tensor lora_out_161_pad_0 = const()[name = tensor("lora_out_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_163_weight_0_to_fp16 = const()[name = tensor("lora_out_163_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84863680)))]; + tensor lora_out_163_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_11583, groups = var_10009, pad = lora_out_161_pad_0, pad_type = lora_out_161_pad_type_0, strides = var_11581, weight = lora_out_163_weight_0_to_fp16, x = input_133_cast_fp16)[name = tensor("lora_out_163_cast_fp16")]; + tensor input_135_cast_fp16 = add(x = pretrained_out_81_cast_fp16, y = lora_out_163_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor input_137_mode_0 = const()[name = tensor("input_137_mode_0"), val = tensor("EXACT")]; + tensor input_137_cast_fp16 = gelu(mode = input_137_mode_0, x = input_135_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor var_11595 = const()[name = tensor("op_11595"), val = tensor([1, 1])]; + tensor var_11597 = const()[name = tensor("op_11597"), val = tensor([1, 1])]; + tensor pretrained_out_83_pad_type_0 = const()[name = tensor("pretrained_out_83_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_83_pad_0 = const()[name = tensor("pretrained_out_83_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85027584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88304448))), name = tensor("layers_6_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_6_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_6_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88304576)))]; + tensor pretrained_out_83_cast_fp16 = conv(bias = layers_6_fc2_pretrained_bias_to_fp16, dilations = var_11597, groups = var_10009, pad = pretrained_out_83_pad_0, pad_type = pretrained_out_83_pad_type_0, strides = var_11595, weight = layers_6_fc2_pretrained_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("pretrained_out_83_cast_fp16")]; + tensor var_11601 = const()[name = tensor("op_11601"), val = tensor([1, 1])]; + tensor var_11603 = const()[name = tensor("op_11603"), val = tensor([1, 1])]; + tensor input_139_pad_type_0 = const()[name = tensor("input_139_pad_type_0"), val = tensor("custom")]; + tensor input_139_pad_0 = const()[name = tensor("input_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_6_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_6_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88307200)))]; + tensor input_139_cast_fp16 = conv(dilations = var_11603, groups = var_10009, pad = input_139_pad_0, pad_type = input_139_pad_type_0, strides = var_11601, weight = layers_6_fc2_loraA_weight_to_fp16, x = input_137_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor var_11607 = const()[name = tensor("op_11607"), val = tensor([1, 1])]; + tensor var_11609 = const()[name = tensor("op_11609"), val = tensor([1, 1])]; + tensor lora_out_165_pad_type_0 = const()[name = tensor("lora_out_165_pad_type_0"), val = tensor("custom")]; + tensor lora_out_165_pad_0 = const()[name = tensor("lora_out_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_167_weight_0_to_fp16 = const()[name = tensor("lora_out_167_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88471104)))]; + tensor lora_out_167_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_11609, groups = var_10009, pad = lora_out_165_pad_0, pad_type = lora_out_165_pad_type_0, strides = var_11607, weight = lora_out_167_weight_0_to_fp16, x = input_139_cast_fp16)[name = tensor("lora_out_167_cast_fp16")]; + tensor hidden_states_17_cast_fp16 = add(x = pretrained_out_83_cast_fp16, y = lora_out_167_cast_fp16)[name = tensor("hidden_states_17_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor var_11619 = const()[name = tensor("op_11619"), val = tensor(3)]; + tensor var_11644 = const()[name = tensor("op_11644"), val = tensor(1)]; + tensor var_11645 = const()[name = tensor("op_11645"), val = tensor(true)]; + tensor var_11655 = const()[name = tensor("op_11655"), val = tensor([1])]; + tensor channels_mean_29_cast_fp16 = reduce_mean(axes = var_11655, keep_dims = var_11645, x = inputs_29_cast_fp16)[name = tensor("channels_mean_29_cast_fp16")]; + tensor zero_mean_29_cast_fp16 = sub(x = inputs_29_cast_fp16, y = channels_mean_29_cast_fp16)[name = tensor("zero_mean_29_cast_fp16")]; + tensor zero_mean_sq_29_cast_fp16 = mul(x = zero_mean_29_cast_fp16, y = zero_mean_29_cast_fp16)[name = tensor("zero_mean_sq_29_cast_fp16")]; + tensor var_11659 = const()[name = tensor("op_11659"), val = tensor([1])]; + tensor var_11660_cast_fp16 = reduce_mean(axes = var_11659, keep_dims = var_11645, x = zero_mean_sq_29_cast_fp16)[name = tensor("op_11660_cast_fp16")]; + tensor var_11661_to_fp16 = const()[name = tensor("op_11661_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_11662_cast_fp16 = add(x = var_11660_cast_fp16, y = var_11661_to_fp16)[name = tensor("op_11662_cast_fp16")]; + tensor denom_29_epsilon_0 = const()[name = tensor("denom_29_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_29_cast_fp16 = rsqrt(epsilon = denom_29_epsilon_0, x = var_11662_cast_fp16)[name = tensor("denom_29_cast_fp16")]; + tensor out_29_cast_fp16 = mul(x = zero_mean_29_cast_fp16, y = denom_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; + tensor obj_29_gamma_0_to_fp16 = const()[name = tensor("obj_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88512128)))]; + tensor obj_29_beta_0_to_fp16 = const()[name = tensor("obj_29_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88514752)))]; + tensor obj_29_epsilon_0_to_fp16 = const()[name = tensor("obj_29_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor("obj_29_cast_fp16")]; + tensor var_11680 = const()[name = tensor("op_11680"), val = tensor([1, 1])]; + tensor var_11682 = const()[name = tensor("op_11682"), val = tensor([1, 1])]; + tensor pretrained_out_85_pad_type_0 = const()[name = tensor("pretrained_out_85_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_85_pad_0 = const()[name = tensor("pretrained_out_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88517376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89336640))), name = tensor("layers_7_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_7_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89336768)))]; + tensor pretrained_out_85_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_11682, groups = var_11644, pad = pretrained_out_85_pad_0, pad_type = pretrained_out_85_pad_type_0, strides = var_11680, weight = layers_7_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("pretrained_out_85_cast_fp16")]; + tensor var_11686 = const()[name = tensor("op_11686"), val = tensor([1, 1])]; + tensor var_11688 = const()[name = tensor("op_11688"), val = tensor([1, 1])]; + tensor input_141_pad_type_0 = const()[name = tensor("input_141_pad_type_0"), val = tensor("custom")]; + tensor input_141_pad_0 = const()[name = tensor("input_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89339392)))]; + tensor input_141_cast_fp16 = conv(dilations = var_11688, groups = var_11644, pad = input_141_pad_0, pad_type = input_141_pad_type_0, strides = var_11686, weight = layers_7_self_attn_q_proj_loraA_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor var_11692 = const()[name = tensor("op_11692"), val = tensor([1, 1])]; + tensor var_11694 = const()[name = tensor("op_11694"), val = tensor([1, 1])]; + tensor lora_out_169_pad_type_0 = const()[name = tensor("lora_out_169_pad_type_0"), val = tensor("custom")]; + tensor lora_out_169_pad_0 = const()[name = tensor("lora_out_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_171_weight_0_to_fp16 = const()[name = tensor("lora_out_171_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89380416)))]; + tensor lora_out_171_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_11694, groups = var_11644, pad = lora_out_169_pad_0, pad_type = lora_out_169_pad_type_0, strides = var_11692, weight = lora_out_171_weight_0_to_fp16, x = input_141_cast_fp16)[name = tensor("lora_out_171_cast_fp16")]; + tensor query_15_cast_fp16 = add(x = pretrained_out_85_cast_fp16, y = lora_out_171_cast_fp16)[name = tensor("query_15_cast_fp16")]; + tensor var_11704 = const()[name = tensor("op_11704"), val = tensor([1, 1])]; + tensor var_11706 = const()[name = tensor("op_11706"), val = tensor([1, 1])]; + tensor pretrained_out_87_pad_type_0 = const()[name = tensor("pretrained_out_87_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_87_pad_0 = const()[name = tensor("pretrained_out_87_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89421440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90240704))), name = tensor("layers_7_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_87_cast_fp16 = conv(dilations = var_11706, groups = var_11644, pad = pretrained_out_87_pad_0, pad_type = pretrained_out_87_pad_type_0, strides = var_11704, weight = layers_7_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("pretrained_out_87_cast_fp16")]; + tensor var_11710 = const()[name = tensor("op_11710"), val = tensor([1, 1])]; + tensor var_11712 = const()[name = tensor("op_11712"), val = tensor([1, 1])]; + tensor input_143_pad_type_0 = const()[name = tensor("input_143_pad_type_0"), val = tensor("custom")]; + tensor input_143_pad_0 = const()[name = tensor("input_143_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90240832)))]; + tensor input_143_cast_fp16 = conv(dilations = var_11712, groups = var_11644, pad = input_143_pad_0, pad_type = input_143_pad_type_0, strides = var_11710, weight = layers_7_self_attn_k_proj_loraA_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor var_11716 = const()[name = tensor("op_11716"), val = tensor([1, 1])]; + tensor var_11718 = const()[name = tensor("op_11718"), val = tensor([1, 1])]; + tensor lora_out_173_pad_type_0 = const()[name = tensor("lora_out_173_pad_type_0"), val = tensor("custom")]; + tensor lora_out_173_pad_0 = const()[name = tensor("lora_out_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_175_weight_0_to_fp16 = const()[name = tensor("lora_out_175_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90281856)))]; + tensor lora_out_175_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_11718, groups = var_11644, pad = lora_out_173_pad_0, pad_type = lora_out_173_pad_type_0, strides = var_11716, weight = lora_out_175_weight_0_to_fp16, x = input_143_cast_fp16)[name = tensor("lora_out_175_cast_fp16")]; + tensor key_15_cast_fp16 = add(x = pretrained_out_87_cast_fp16, y = lora_out_175_cast_fp16)[name = tensor("key_15_cast_fp16")]; + tensor var_11729 = const()[name = tensor("op_11729"), val = tensor([1, 1])]; + tensor var_11731 = const()[name = tensor("op_11731"), val = tensor([1, 1])]; + tensor pretrained_out_89_pad_type_0 = const()[name = tensor("pretrained_out_89_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_89_pad_0 = const()[name = tensor("pretrained_out_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90322880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91142144))), name = tensor("layers_7_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_7_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91142272)))]; + tensor pretrained_out_89_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_11731, groups = var_11644, pad = pretrained_out_89_pad_0, pad_type = pretrained_out_89_pad_type_0, strides = var_11729, weight = layers_7_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor("pretrained_out_89_cast_fp16")]; + tensor var_11735 = const()[name = tensor("op_11735"), val = tensor([1, 1])]; + tensor var_11737 = const()[name = tensor("op_11737"), val = tensor([1, 1])]; + tensor input_145_pad_type_0 = const()[name = tensor("input_145_pad_type_0"), val = tensor("custom")]; + tensor input_145_pad_0 = const()[name = tensor("input_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91144896)))]; + tensor input_145_cast_fp16 = conv(dilations = var_11737, groups = var_11644, pad = input_145_pad_0, pad_type = input_145_pad_type_0, strides = var_11735, weight = layers_7_self_attn_v_proj_loraA_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor var_11741 = const()[name = tensor("op_11741"), val = tensor([1, 1])]; + tensor var_11743 = const()[name = tensor("op_11743"), val = tensor([1, 1])]; + tensor lora_out_177_pad_type_0 = const()[name = tensor("lora_out_177_pad_type_0"), val = tensor("custom")]; + tensor lora_out_177_pad_0 = const()[name = tensor("lora_out_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_179_weight_0_to_fp16 = const()[name = tensor("lora_out_179_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91185920)))]; + tensor lora_out_179_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_11743, groups = var_11644, pad = lora_out_177_pad_0, pad_type = lora_out_177_pad_type_0, strides = var_11741, weight = lora_out_179_weight_0_to_fp16, x = input_145_cast_fp16)[name = tensor("lora_out_179_cast_fp16")]; + tensor value_15_cast_fp16 = add(x = pretrained_out_89_cast_fp16, y = lora_out_179_cast_fp16)[name = tensor("value_15_cast_fp16")]; + tensor var_11753_begin_0 = const()[name = tensor("op_11753_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11753_end_0 = const()[name = tensor("op_11753_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11753_end_mask_0 = const()[name = tensor("op_11753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11753_cast_fp16 = slice_by_index(begin = var_11753_begin_0, end = var_11753_end_0, end_mask = var_11753_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11753_cast_fp16")]; + tensor var_11757_begin_0 = const()[name = tensor("op_11757_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_11757_end_0 = const()[name = tensor("op_11757_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_11757_end_mask_0 = const()[name = tensor("op_11757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11757_cast_fp16 = slice_by_index(begin = var_11757_begin_0, end = var_11757_end_0, end_mask = var_11757_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11757_cast_fp16")]; + tensor var_11761_begin_0 = const()[name = tensor("op_11761_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_11761_end_0 = const()[name = tensor("op_11761_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_11761_end_mask_0 = const()[name = tensor("op_11761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11761_cast_fp16 = slice_by_index(begin = var_11761_begin_0, end = var_11761_end_0, end_mask = var_11761_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11761_cast_fp16")]; + tensor var_11765_begin_0 = const()[name = tensor("op_11765_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_11765_end_0 = const()[name = tensor("op_11765_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_11765_end_mask_0 = const()[name = tensor("op_11765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11765_cast_fp16 = slice_by_index(begin = var_11765_begin_0, end = var_11765_end_0, end_mask = var_11765_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11765_cast_fp16")]; + tensor var_11769_begin_0 = const()[name = tensor("op_11769_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_11769_end_0 = const()[name = tensor("op_11769_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_11769_end_mask_0 = const()[name = tensor("op_11769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11769_cast_fp16 = slice_by_index(begin = var_11769_begin_0, end = var_11769_end_0, end_mask = var_11769_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11769_cast_fp16")]; + tensor var_11773_begin_0 = const()[name = tensor("op_11773_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_11773_end_0 = const()[name = tensor("op_11773_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_11773_end_mask_0 = const()[name = tensor("op_11773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11773_cast_fp16 = slice_by_index(begin = var_11773_begin_0, end = var_11773_end_0, end_mask = var_11773_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11773_cast_fp16")]; + tensor var_11777_begin_0 = const()[name = tensor("op_11777_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_11777_end_0 = const()[name = tensor("op_11777_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_11777_end_mask_0 = const()[name = tensor("op_11777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11777_cast_fp16 = slice_by_index(begin = var_11777_begin_0, end = var_11777_end_0, end_mask = var_11777_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11777_cast_fp16")]; + tensor var_11781_begin_0 = const()[name = tensor("op_11781_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_11781_end_0 = const()[name = tensor("op_11781_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_11781_end_mask_0 = const()[name = tensor("op_11781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11781_cast_fp16 = slice_by_index(begin = var_11781_begin_0, end = var_11781_end_0, end_mask = var_11781_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11781_cast_fp16")]; + tensor var_11785_begin_0 = const()[name = tensor("op_11785_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_11785_end_0 = const()[name = tensor("op_11785_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_11785_end_mask_0 = const()[name = tensor("op_11785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11785_cast_fp16 = slice_by_index(begin = var_11785_begin_0, end = var_11785_end_0, end_mask = var_11785_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11785_cast_fp16")]; + tensor var_11789_begin_0 = const()[name = tensor("op_11789_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_11789_end_0 = const()[name = tensor("op_11789_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_11789_end_mask_0 = const()[name = tensor("op_11789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11789_cast_fp16 = slice_by_index(begin = var_11789_begin_0, end = var_11789_end_0, end_mask = var_11789_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11789_cast_fp16")]; + tensor var_11793_begin_0 = const()[name = tensor("op_11793_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_11793_end_0 = const()[name = tensor("op_11793_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_11793_end_mask_0 = const()[name = tensor("op_11793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11793_cast_fp16 = slice_by_index(begin = var_11793_begin_0, end = var_11793_end_0, end_mask = var_11793_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11793_cast_fp16")]; + tensor var_11797_begin_0 = const()[name = tensor("op_11797_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_11797_end_0 = const()[name = tensor("op_11797_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_11797_end_mask_0 = const()[name = tensor("op_11797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11797_cast_fp16 = slice_by_index(begin = var_11797_begin_0, end = var_11797_end_0, end_mask = var_11797_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11797_cast_fp16")]; + tensor var_11801_begin_0 = const()[name = tensor("op_11801_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_11801_end_0 = const()[name = tensor("op_11801_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_11801_end_mask_0 = const()[name = tensor("op_11801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11801_cast_fp16 = slice_by_index(begin = var_11801_begin_0, end = var_11801_end_0, end_mask = var_11801_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11801_cast_fp16")]; + tensor var_11805_begin_0 = const()[name = tensor("op_11805_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_11805_end_0 = const()[name = tensor("op_11805_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_11805_end_mask_0 = const()[name = tensor("op_11805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11805_cast_fp16 = slice_by_index(begin = var_11805_begin_0, end = var_11805_end_0, end_mask = var_11805_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11805_cast_fp16")]; + tensor var_11809_begin_0 = const()[name = tensor("op_11809_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_11809_end_0 = const()[name = tensor("op_11809_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_11809_end_mask_0 = const()[name = tensor("op_11809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11809_cast_fp16 = slice_by_index(begin = var_11809_begin_0, end = var_11809_end_0, end_mask = var_11809_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11809_cast_fp16")]; + tensor var_11813_begin_0 = const()[name = tensor("op_11813_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_11813_end_0 = const()[name = tensor("op_11813_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_11813_end_mask_0 = const()[name = tensor("op_11813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11813_cast_fp16 = slice_by_index(begin = var_11813_begin_0, end = var_11813_end_0, end_mask = var_11813_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11813_cast_fp16")]; + tensor var_11817_begin_0 = const()[name = tensor("op_11817_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_11817_end_0 = const()[name = tensor("op_11817_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_11817_end_mask_0 = const()[name = tensor("op_11817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11817_cast_fp16 = slice_by_index(begin = var_11817_begin_0, end = var_11817_end_0, end_mask = var_11817_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11817_cast_fp16")]; + tensor var_11821_begin_0 = const()[name = tensor("op_11821_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_11821_end_0 = const()[name = tensor("op_11821_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_11821_end_mask_0 = const()[name = tensor("op_11821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11821_cast_fp16 = slice_by_index(begin = var_11821_begin_0, end = var_11821_end_0, end_mask = var_11821_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11821_cast_fp16")]; + tensor var_11825_begin_0 = const()[name = tensor("op_11825_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_11825_end_0 = const()[name = tensor("op_11825_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_11825_end_mask_0 = const()[name = tensor("op_11825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11825_cast_fp16 = slice_by_index(begin = var_11825_begin_0, end = var_11825_end_0, end_mask = var_11825_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11825_cast_fp16")]; + tensor var_11829_begin_0 = const()[name = tensor("op_11829_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_11829_end_0 = const()[name = tensor("op_11829_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_11829_end_mask_0 = const()[name = tensor("op_11829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_11829_cast_fp16 = slice_by_index(begin = var_11829_begin_0, end = var_11829_end_0, end_mask = var_11829_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_11829_cast_fp16")]; + tensor var_11838_begin_0 = const()[name = tensor("op_11838_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11838_end_0 = const()[name = tensor("op_11838_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11838_end_mask_0 = const()[name = tensor("op_11838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11838_cast_fp16 = slice_by_index(begin = var_11838_begin_0, end = var_11838_end_0, end_mask = var_11838_end_mask_0, x = var_11753_cast_fp16)[name = tensor("op_11838_cast_fp16")]; + tensor var_11845_begin_0 = const()[name = tensor("op_11845_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11845_end_0 = const()[name = tensor("op_11845_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11845_end_mask_0 = const()[name = tensor("op_11845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11845_cast_fp16 = slice_by_index(begin = var_11845_begin_0, end = var_11845_end_0, end_mask = var_11845_end_mask_0, x = var_11753_cast_fp16)[name = tensor("op_11845_cast_fp16")]; + tensor var_11852_begin_0 = const()[name = tensor("op_11852_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11852_end_0 = const()[name = tensor("op_11852_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11852_end_mask_0 = const()[name = tensor("op_11852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11852_cast_fp16 = slice_by_index(begin = var_11852_begin_0, end = var_11852_end_0, end_mask = var_11852_end_mask_0, x = var_11753_cast_fp16)[name = tensor("op_11852_cast_fp16")]; + tensor var_11859_begin_0 = const()[name = tensor("op_11859_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11859_end_0 = const()[name = tensor("op_11859_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11859_end_mask_0 = const()[name = tensor("op_11859_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11859_cast_fp16 = slice_by_index(begin = var_11859_begin_0, end = var_11859_end_0, end_mask = var_11859_end_mask_0, x = var_11753_cast_fp16)[name = tensor("op_11859_cast_fp16")]; + tensor var_11866_begin_0 = const()[name = tensor("op_11866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11866_end_0 = const()[name = tensor("op_11866_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11866_end_mask_0 = const()[name = tensor("op_11866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11866_cast_fp16 = slice_by_index(begin = var_11866_begin_0, end = var_11866_end_0, end_mask = var_11866_end_mask_0, x = var_11757_cast_fp16)[name = tensor("op_11866_cast_fp16")]; + tensor var_11873_begin_0 = const()[name = tensor("op_11873_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11873_end_0 = const()[name = tensor("op_11873_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11873_end_mask_0 = const()[name = tensor("op_11873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11873_cast_fp16 = slice_by_index(begin = var_11873_begin_0, end = var_11873_end_0, end_mask = var_11873_end_mask_0, x = var_11757_cast_fp16)[name = tensor("op_11873_cast_fp16")]; + tensor var_11880_begin_0 = const()[name = tensor("op_11880_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11880_end_0 = const()[name = tensor("op_11880_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11880_end_mask_0 = const()[name = tensor("op_11880_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11880_cast_fp16 = slice_by_index(begin = var_11880_begin_0, end = var_11880_end_0, end_mask = var_11880_end_mask_0, x = var_11757_cast_fp16)[name = tensor("op_11880_cast_fp16")]; + tensor var_11887_begin_0 = const()[name = tensor("op_11887_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11887_end_0 = const()[name = tensor("op_11887_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11887_end_mask_0 = const()[name = tensor("op_11887_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11887_cast_fp16 = slice_by_index(begin = var_11887_begin_0, end = var_11887_end_0, end_mask = var_11887_end_mask_0, x = var_11757_cast_fp16)[name = tensor("op_11887_cast_fp16")]; + tensor var_11894_begin_0 = const()[name = tensor("op_11894_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11894_end_0 = const()[name = tensor("op_11894_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11894_end_mask_0 = const()[name = tensor("op_11894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11894_cast_fp16 = slice_by_index(begin = var_11894_begin_0, end = var_11894_end_0, end_mask = var_11894_end_mask_0, x = var_11761_cast_fp16)[name = tensor("op_11894_cast_fp16")]; + tensor var_11901_begin_0 = const()[name = tensor("op_11901_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11901_end_0 = const()[name = tensor("op_11901_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11901_end_mask_0 = const()[name = tensor("op_11901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11901_cast_fp16 = slice_by_index(begin = var_11901_begin_0, end = var_11901_end_0, end_mask = var_11901_end_mask_0, x = var_11761_cast_fp16)[name = tensor("op_11901_cast_fp16")]; + tensor var_11908_begin_0 = const()[name = tensor("op_11908_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11908_end_0 = const()[name = tensor("op_11908_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11908_end_mask_0 = const()[name = tensor("op_11908_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11908_cast_fp16 = slice_by_index(begin = var_11908_begin_0, end = var_11908_end_0, end_mask = var_11908_end_mask_0, x = var_11761_cast_fp16)[name = tensor("op_11908_cast_fp16")]; + tensor var_11915_begin_0 = const()[name = tensor("op_11915_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11915_end_0 = const()[name = tensor("op_11915_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11915_end_mask_0 = const()[name = tensor("op_11915_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11915_cast_fp16 = slice_by_index(begin = var_11915_begin_0, end = var_11915_end_0, end_mask = var_11915_end_mask_0, x = var_11761_cast_fp16)[name = tensor("op_11915_cast_fp16")]; + tensor var_11922_begin_0 = const()[name = tensor("op_11922_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11922_end_0 = const()[name = tensor("op_11922_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11922_end_mask_0 = const()[name = tensor("op_11922_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11922_cast_fp16 = slice_by_index(begin = var_11922_begin_0, end = var_11922_end_0, end_mask = var_11922_end_mask_0, x = var_11765_cast_fp16)[name = tensor("op_11922_cast_fp16")]; + tensor var_11929_begin_0 = const()[name = tensor("op_11929_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11929_end_0 = const()[name = tensor("op_11929_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11929_end_mask_0 = const()[name = tensor("op_11929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11929_cast_fp16 = slice_by_index(begin = var_11929_begin_0, end = var_11929_end_0, end_mask = var_11929_end_mask_0, x = var_11765_cast_fp16)[name = tensor("op_11929_cast_fp16")]; + tensor var_11936_begin_0 = const()[name = tensor("op_11936_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11936_end_0 = const()[name = tensor("op_11936_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11936_end_mask_0 = const()[name = tensor("op_11936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11936_cast_fp16 = slice_by_index(begin = var_11936_begin_0, end = var_11936_end_0, end_mask = var_11936_end_mask_0, x = var_11765_cast_fp16)[name = tensor("op_11936_cast_fp16")]; + tensor var_11943_begin_0 = const()[name = tensor("op_11943_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11943_end_0 = const()[name = tensor("op_11943_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11943_end_mask_0 = const()[name = tensor("op_11943_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11943_cast_fp16 = slice_by_index(begin = var_11943_begin_0, end = var_11943_end_0, end_mask = var_11943_end_mask_0, x = var_11765_cast_fp16)[name = tensor("op_11943_cast_fp16")]; + tensor var_11950_begin_0 = const()[name = tensor("op_11950_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11950_end_0 = const()[name = tensor("op_11950_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11950_end_mask_0 = const()[name = tensor("op_11950_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11950_cast_fp16 = slice_by_index(begin = var_11950_begin_0, end = var_11950_end_0, end_mask = var_11950_end_mask_0, x = var_11769_cast_fp16)[name = tensor("op_11950_cast_fp16")]; + tensor var_11957_begin_0 = const()[name = tensor("op_11957_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11957_end_0 = const()[name = tensor("op_11957_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11957_end_mask_0 = const()[name = tensor("op_11957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11957_cast_fp16 = slice_by_index(begin = var_11957_begin_0, end = var_11957_end_0, end_mask = var_11957_end_mask_0, x = var_11769_cast_fp16)[name = tensor("op_11957_cast_fp16")]; + tensor var_11964_begin_0 = const()[name = tensor("op_11964_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11964_end_0 = const()[name = tensor("op_11964_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11964_end_mask_0 = const()[name = tensor("op_11964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11964_cast_fp16 = slice_by_index(begin = var_11964_begin_0, end = var_11964_end_0, end_mask = var_11964_end_mask_0, x = var_11769_cast_fp16)[name = tensor("op_11964_cast_fp16")]; + tensor var_11971_begin_0 = const()[name = tensor("op_11971_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11971_end_0 = const()[name = tensor("op_11971_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11971_end_mask_0 = const()[name = tensor("op_11971_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11971_cast_fp16 = slice_by_index(begin = var_11971_begin_0, end = var_11971_end_0, end_mask = var_11971_end_mask_0, x = var_11769_cast_fp16)[name = tensor("op_11971_cast_fp16")]; + tensor var_11978_begin_0 = const()[name = tensor("op_11978_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_11978_end_0 = const()[name = tensor("op_11978_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_11978_end_mask_0 = const()[name = tensor("op_11978_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11978_cast_fp16 = slice_by_index(begin = var_11978_begin_0, end = var_11978_end_0, end_mask = var_11978_end_mask_0, x = var_11773_cast_fp16)[name = tensor("op_11978_cast_fp16")]; + tensor var_11985_begin_0 = const()[name = tensor("op_11985_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_11985_end_0 = const()[name = tensor("op_11985_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_11985_end_mask_0 = const()[name = tensor("op_11985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11985_cast_fp16 = slice_by_index(begin = var_11985_begin_0, end = var_11985_end_0, end_mask = var_11985_end_mask_0, x = var_11773_cast_fp16)[name = tensor("op_11985_cast_fp16")]; + tensor var_11992_begin_0 = const()[name = tensor("op_11992_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_11992_end_0 = const()[name = tensor("op_11992_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_11992_end_mask_0 = const()[name = tensor("op_11992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11992_cast_fp16 = slice_by_index(begin = var_11992_begin_0, end = var_11992_end_0, end_mask = var_11992_end_mask_0, x = var_11773_cast_fp16)[name = tensor("op_11992_cast_fp16")]; + tensor var_11999_begin_0 = const()[name = tensor("op_11999_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_11999_end_0 = const()[name = tensor("op_11999_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_11999_end_mask_0 = const()[name = tensor("op_11999_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_11999_cast_fp16 = slice_by_index(begin = var_11999_begin_0, end = var_11999_end_0, end_mask = var_11999_end_mask_0, x = var_11773_cast_fp16)[name = tensor("op_11999_cast_fp16")]; + tensor var_12006_begin_0 = const()[name = tensor("op_12006_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12006_end_0 = const()[name = tensor("op_12006_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12006_end_mask_0 = const()[name = tensor("op_12006_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12006_cast_fp16 = slice_by_index(begin = var_12006_begin_0, end = var_12006_end_0, end_mask = var_12006_end_mask_0, x = var_11777_cast_fp16)[name = tensor("op_12006_cast_fp16")]; + tensor var_12013_begin_0 = const()[name = tensor("op_12013_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12013_end_0 = const()[name = tensor("op_12013_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12013_end_mask_0 = const()[name = tensor("op_12013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12013_cast_fp16 = slice_by_index(begin = var_12013_begin_0, end = var_12013_end_0, end_mask = var_12013_end_mask_0, x = var_11777_cast_fp16)[name = tensor("op_12013_cast_fp16")]; + tensor var_12020_begin_0 = const()[name = tensor("op_12020_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12020_end_0 = const()[name = tensor("op_12020_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12020_end_mask_0 = const()[name = tensor("op_12020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12020_cast_fp16 = slice_by_index(begin = var_12020_begin_0, end = var_12020_end_0, end_mask = var_12020_end_mask_0, x = var_11777_cast_fp16)[name = tensor("op_12020_cast_fp16")]; + tensor var_12027_begin_0 = const()[name = tensor("op_12027_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12027_end_0 = const()[name = tensor("op_12027_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12027_end_mask_0 = const()[name = tensor("op_12027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12027_cast_fp16 = slice_by_index(begin = var_12027_begin_0, end = var_12027_end_0, end_mask = var_12027_end_mask_0, x = var_11777_cast_fp16)[name = tensor("op_12027_cast_fp16")]; + tensor var_12034_begin_0 = const()[name = tensor("op_12034_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12034_end_0 = const()[name = tensor("op_12034_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12034_end_mask_0 = const()[name = tensor("op_12034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12034_cast_fp16 = slice_by_index(begin = var_12034_begin_0, end = var_12034_end_0, end_mask = var_12034_end_mask_0, x = var_11781_cast_fp16)[name = tensor("op_12034_cast_fp16")]; + tensor var_12041_begin_0 = const()[name = tensor("op_12041_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12041_end_0 = const()[name = tensor("op_12041_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12041_end_mask_0 = const()[name = tensor("op_12041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12041_cast_fp16 = slice_by_index(begin = var_12041_begin_0, end = var_12041_end_0, end_mask = var_12041_end_mask_0, x = var_11781_cast_fp16)[name = tensor("op_12041_cast_fp16")]; + tensor var_12048_begin_0 = const()[name = tensor("op_12048_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12048_end_0 = const()[name = tensor("op_12048_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12048_end_mask_0 = const()[name = tensor("op_12048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12048_cast_fp16 = slice_by_index(begin = var_12048_begin_0, end = var_12048_end_0, end_mask = var_12048_end_mask_0, x = var_11781_cast_fp16)[name = tensor("op_12048_cast_fp16")]; + tensor var_12055_begin_0 = const()[name = tensor("op_12055_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12055_end_0 = const()[name = tensor("op_12055_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12055_end_mask_0 = const()[name = tensor("op_12055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12055_cast_fp16 = slice_by_index(begin = var_12055_begin_0, end = var_12055_end_0, end_mask = var_12055_end_mask_0, x = var_11781_cast_fp16)[name = tensor("op_12055_cast_fp16")]; + tensor var_12062_begin_0 = const()[name = tensor("op_12062_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12062_end_0 = const()[name = tensor("op_12062_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12062_end_mask_0 = const()[name = tensor("op_12062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12062_cast_fp16 = slice_by_index(begin = var_12062_begin_0, end = var_12062_end_0, end_mask = var_12062_end_mask_0, x = var_11785_cast_fp16)[name = tensor("op_12062_cast_fp16")]; + tensor var_12069_begin_0 = const()[name = tensor("op_12069_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12069_end_0 = const()[name = tensor("op_12069_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12069_end_mask_0 = const()[name = tensor("op_12069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12069_cast_fp16 = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = var_11785_cast_fp16)[name = tensor("op_12069_cast_fp16")]; + tensor var_12076_begin_0 = const()[name = tensor("op_12076_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12076_end_0 = const()[name = tensor("op_12076_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12076_end_mask_0 = const()[name = tensor("op_12076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12076_cast_fp16 = slice_by_index(begin = var_12076_begin_0, end = var_12076_end_0, end_mask = var_12076_end_mask_0, x = var_11785_cast_fp16)[name = tensor("op_12076_cast_fp16")]; + tensor var_12083_begin_0 = const()[name = tensor("op_12083_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12083_end_0 = const()[name = tensor("op_12083_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12083_end_mask_0 = const()[name = tensor("op_12083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12083_cast_fp16 = slice_by_index(begin = var_12083_begin_0, end = var_12083_end_0, end_mask = var_12083_end_mask_0, x = var_11785_cast_fp16)[name = tensor("op_12083_cast_fp16")]; + tensor var_12090_begin_0 = const()[name = tensor("op_12090_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12090_end_0 = const()[name = tensor("op_12090_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12090_end_mask_0 = const()[name = tensor("op_12090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12090_cast_fp16 = slice_by_index(begin = var_12090_begin_0, end = var_12090_end_0, end_mask = var_12090_end_mask_0, x = var_11789_cast_fp16)[name = tensor("op_12090_cast_fp16")]; + tensor var_12097_begin_0 = const()[name = tensor("op_12097_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12097_end_0 = const()[name = tensor("op_12097_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12097_end_mask_0 = const()[name = tensor("op_12097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12097_cast_fp16 = slice_by_index(begin = var_12097_begin_0, end = var_12097_end_0, end_mask = var_12097_end_mask_0, x = var_11789_cast_fp16)[name = tensor("op_12097_cast_fp16")]; + tensor var_12104_begin_0 = const()[name = tensor("op_12104_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12104_end_0 = const()[name = tensor("op_12104_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12104_end_mask_0 = const()[name = tensor("op_12104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12104_cast_fp16 = slice_by_index(begin = var_12104_begin_0, end = var_12104_end_0, end_mask = var_12104_end_mask_0, x = var_11789_cast_fp16)[name = tensor("op_12104_cast_fp16")]; + tensor var_12111_begin_0 = const()[name = tensor("op_12111_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12111_end_0 = const()[name = tensor("op_12111_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12111_end_mask_0 = const()[name = tensor("op_12111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12111_cast_fp16 = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = var_11789_cast_fp16)[name = tensor("op_12111_cast_fp16")]; + tensor var_12118_begin_0 = const()[name = tensor("op_12118_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12118_end_0 = const()[name = tensor("op_12118_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12118_end_mask_0 = const()[name = tensor("op_12118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12118_cast_fp16 = slice_by_index(begin = var_12118_begin_0, end = var_12118_end_0, end_mask = var_12118_end_mask_0, x = var_11793_cast_fp16)[name = tensor("op_12118_cast_fp16")]; + tensor var_12125_begin_0 = const()[name = tensor("op_12125_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12125_end_0 = const()[name = tensor("op_12125_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12125_end_mask_0 = const()[name = tensor("op_12125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12125_cast_fp16 = slice_by_index(begin = var_12125_begin_0, end = var_12125_end_0, end_mask = var_12125_end_mask_0, x = var_11793_cast_fp16)[name = tensor("op_12125_cast_fp16")]; + tensor var_12132_begin_0 = const()[name = tensor("op_12132_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12132_end_0 = const()[name = tensor("op_12132_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12132_end_mask_0 = const()[name = tensor("op_12132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12132_cast_fp16 = slice_by_index(begin = var_12132_begin_0, end = var_12132_end_0, end_mask = var_12132_end_mask_0, x = var_11793_cast_fp16)[name = tensor("op_12132_cast_fp16")]; + tensor var_12139_begin_0 = const()[name = tensor("op_12139_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12139_end_0 = const()[name = tensor("op_12139_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12139_end_mask_0 = const()[name = tensor("op_12139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12139_cast_fp16 = slice_by_index(begin = var_12139_begin_0, end = var_12139_end_0, end_mask = var_12139_end_mask_0, x = var_11793_cast_fp16)[name = tensor("op_12139_cast_fp16")]; + tensor var_12146_begin_0 = const()[name = tensor("op_12146_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12146_end_0 = const()[name = tensor("op_12146_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12146_end_mask_0 = const()[name = tensor("op_12146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12146_cast_fp16 = slice_by_index(begin = var_12146_begin_0, end = var_12146_end_0, end_mask = var_12146_end_mask_0, x = var_11797_cast_fp16)[name = tensor("op_12146_cast_fp16")]; + tensor var_12153_begin_0 = const()[name = tensor("op_12153_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12153_end_0 = const()[name = tensor("op_12153_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12153_end_mask_0 = const()[name = tensor("op_12153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12153_cast_fp16 = slice_by_index(begin = var_12153_begin_0, end = var_12153_end_0, end_mask = var_12153_end_mask_0, x = var_11797_cast_fp16)[name = tensor("op_12153_cast_fp16")]; + tensor var_12160_begin_0 = const()[name = tensor("op_12160_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12160_end_0 = const()[name = tensor("op_12160_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12160_end_mask_0 = const()[name = tensor("op_12160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12160_cast_fp16 = slice_by_index(begin = var_12160_begin_0, end = var_12160_end_0, end_mask = var_12160_end_mask_0, x = var_11797_cast_fp16)[name = tensor("op_12160_cast_fp16")]; + tensor var_12167_begin_0 = const()[name = tensor("op_12167_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12167_end_0 = const()[name = tensor("op_12167_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12167_end_mask_0 = const()[name = tensor("op_12167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12167_cast_fp16 = slice_by_index(begin = var_12167_begin_0, end = var_12167_end_0, end_mask = var_12167_end_mask_0, x = var_11797_cast_fp16)[name = tensor("op_12167_cast_fp16")]; + tensor var_12174_begin_0 = const()[name = tensor("op_12174_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12174_end_0 = const()[name = tensor("op_12174_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12174_end_mask_0 = const()[name = tensor("op_12174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12174_cast_fp16 = slice_by_index(begin = var_12174_begin_0, end = var_12174_end_0, end_mask = var_12174_end_mask_0, x = var_11801_cast_fp16)[name = tensor("op_12174_cast_fp16")]; + tensor var_12181_begin_0 = const()[name = tensor("op_12181_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12181_end_0 = const()[name = tensor("op_12181_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12181_end_mask_0 = const()[name = tensor("op_12181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12181_cast_fp16 = slice_by_index(begin = var_12181_begin_0, end = var_12181_end_0, end_mask = var_12181_end_mask_0, x = var_11801_cast_fp16)[name = tensor("op_12181_cast_fp16")]; + tensor var_12188_begin_0 = const()[name = tensor("op_12188_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12188_end_0 = const()[name = tensor("op_12188_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12188_end_mask_0 = const()[name = tensor("op_12188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12188_cast_fp16 = slice_by_index(begin = var_12188_begin_0, end = var_12188_end_0, end_mask = var_12188_end_mask_0, x = var_11801_cast_fp16)[name = tensor("op_12188_cast_fp16")]; + tensor var_12195_begin_0 = const()[name = tensor("op_12195_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12195_end_0 = const()[name = tensor("op_12195_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12195_end_mask_0 = const()[name = tensor("op_12195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12195_cast_fp16 = slice_by_index(begin = var_12195_begin_0, end = var_12195_end_0, end_mask = var_12195_end_mask_0, x = var_11801_cast_fp16)[name = tensor("op_12195_cast_fp16")]; + tensor var_12202_begin_0 = const()[name = tensor("op_12202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12202_end_0 = const()[name = tensor("op_12202_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12202_end_mask_0 = const()[name = tensor("op_12202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12202_cast_fp16 = slice_by_index(begin = var_12202_begin_0, end = var_12202_end_0, end_mask = var_12202_end_mask_0, x = var_11805_cast_fp16)[name = tensor("op_12202_cast_fp16")]; + tensor var_12209_begin_0 = const()[name = tensor("op_12209_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12209_end_0 = const()[name = tensor("op_12209_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12209_end_mask_0 = const()[name = tensor("op_12209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12209_cast_fp16 = slice_by_index(begin = var_12209_begin_0, end = var_12209_end_0, end_mask = var_12209_end_mask_0, x = var_11805_cast_fp16)[name = tensor("op_12209_cast_fp16")]; + tensor var_12216_begin_0 = const()[name = tensor("op_12216_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12216_end_0 = const()[name = tensor("op_12216_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12216_end_mask_0 = const()[name = tensor("op_12216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12216_cast_fp16 = slice_by_index(begin = var_12216_begin_0, end = var_12216_end_0, end_mask = var_12216_end_mask_0, x = var_11805_cast_fp16)[name = tensor("op_12216_cast_fp16")]; + tensor var_12223_begin_0 = const()[name = tensor("op_12223_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12223_end_0 = const()[name = tensor("op_12223_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12223_end_mask_0 = const()[name = tensor("op_12223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12223_cast_fp16 = slice_by_index(begin = var_12223_begin_0, end = var_12223_end_0, end_mask = var_12223_end_mask_0, x = var_11805_cast_fp16)[name = tensor("op_12223_cast_fp16")]; + tensor var_12230_begin_0 = const()[name = tensor("op_12230_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12230_end_0 = const()[name = tensor("op_12230_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12230_end_mask_0 = const()[name = tensor("op_12230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12230_cast_fp16 = slice_by_index(begin = var_12230_begin_0, end = var_12230_end_0, end_mask = var_12230_end_mask_0, x = var_11809_cast_fp16)[name = tensor("op_12230_cast_fp16")]; + tensor var_12237_begin_0 = const()[name = tensor("op_12237_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12237_end_0 = const()[name = tensor("op_12237_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12237_end_mask_0 = const()[name = tensor("op_12237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12237_cast_fp16 = slice_by_index(begin = var_12237_begin_0, end = var_12237_end_0, end_mask = var_12237_end_mask_0, x = var_11809_cast_fp16)[name = tensor("op_12237_cast_fp16")]; + tensor var_12244_begin_0 = const()[name = tensor("op_12244_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12244_end_0 = const()[name = tensor("op_12244_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12244_end_mask_0 = const()[name = tensor("op_12244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12244_cast_fp16 = slice_by_index(begin = var_12244_begin_0, end = var_12244_end_0, end_mask = var_12244_end_mask_0, x = var_11809_cast_fp16)[name = tensor("op_12244_cast_fp16")]; + tensor var_12251_begin_0 = const()[name = tensor("op_12251_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12251_end_0 = const()[name = tensor("op_12251_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12251_end_mask_0 = const()[name = tensor("op_12251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12251_cast_fp16 = slice_by_index(begin = var_12251_begin_0, end = var_12251_end_0, end_mask = var_12251_end_mask_0, x = var_11809_cast_fp16)[name = tensor("op_12251_cast_fp16")]; + tensor var_12258_begin_0 = const()[name = tensor("op_12258_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12258_end_0 = const()[name = tensor("op_12258_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12258_end_mask_0 = const()[name = tensor("op_12258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12258_cast_fp16 = slice_by_index(begin = var_12258_begin_0, end = var_12258_end_0, end_mask = var_12258_end_mask_0, x = var_11813_cast_fp16)[name = tensor("op_12258_cast_fp16")]; + tensor var_12265_begin_0 = const()[name = tensor("op_12265_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12265_end_0 = const()[name = tensor("op_12265_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12265_end_mask_0 = const()[name = tensor("op_12265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12265_cast_fp16 = slice_by_index(begin = var_12265_begin_0, end = var_12265_end_0, end_mask = var_12265_end_mask_0, x = var_11813_cast_fp16)[name = tensor("op_12265_cast_fp16")]; + tensor var_12272_begin_0 = const()[name = tensor("op_12272_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12272_end_0 = const()[name = tensor("op_12272_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12272_end_mask_0 = const()[name = tensor("op_12272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12272_cast_fp16 = slice_by_index(begin = var_12272_begin_0, end = var_12272_end_0, end_mask = var_12272_end_mask_0, x = var_11813_cast_fp16)[name = tensor("op_12272_cast_fp16")]; + tensor var_12279_begin_0 = const()[name = tensor("op_12279_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12279_end_0 = const()[name = tensor("op_12279_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12279_end_mask_0 = const()[name = tensor("op_12279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12279_cast_fp16 = slice_by_index(begin = var_12279_begin_0, end = var_12279_end_0, end_mask = var_12279_end_mask_0, x = var_11813_cast_fp16)[name = tensor("op_12279_cast_fp16")]; + tensor var_12286_begin_0 = const()[name = tensor("op_12286_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12286_end_0 = const()[name = tensor("op_12286_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12286_end_mask_0 = const()[name = tensor("op_12286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12286_cast_fp16 = slice_by_index(begin = var_12286_begin_0, end = var_12286_end_0, end_mask = var_12286_end_mask_0, x = var_11817_cast_fp16)[name = tensor("op_12286_cast_fp16")]; + tensor var_12293_begin_0 = const()[name = tensor("op_12293_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12293_end_0 = const()[name = tensor("op_12293_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12293_end_mask_0 = const()[name = tensor("op_12293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12293_cast_fp16 = slice_by_index(begin = var_12293_begin_0, end = var_12293_end_0, end_mask = var_12293_end_mask_0, x = var_11817_cast_fp16)[name = tensor("op_12293_cast_fp16")]; + tensor var_12300_begin_0 = const()[name = tensor("op_12300_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12300_end_0 = const()[name = tensor("op_12300_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12300_end_mask_0 = const()[name = tensor("op_12300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12300_cast_fp16 = slice_by_index(begin = var_12300_begin_0, end = var_12300_end_0, end_mask = var_12300_end_mask_0, x = var_11817_cast_fp16)[name = tensor("op_12300_cast_fp16")]; + tensor var_12307_begin_0 = const()[name = tensor("op_12307_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12307_end_0 = const()[name = tensor("op_12307_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12307_end_mask_0 = const()[name = tensor("op_12307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12307_cast_fp16 = slice_by_index(begin = var_12307_begin_0, end = var_12307_end_0, end_mask = var_12307_end_mask_0, x = var_11817_cast_fp16)[name = tensor("op_12307_cast_fp16")]; + tensor var_12314_begin_0 = const()[name = tensor("op_12314_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12314_end_0 = const()[name = tensor("op_12314_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12314_end_mask_0 = const()[name = tensor("op_12314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12314_cast_fp16 = slice_by_index(begin = var_12314_begin_0, end = var_12314_end_0, end_mask = var_12314_end_mask_0, x = var_11821_cast_fp16)[name = tensor("op_12314_cast_fp16")]; + tensor var_12321_begin_0 = const()[name = tensor("op_12321_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12321_end_0 = const()[name = tensor("op_12321_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12321_end_mask_0 = const()[name = tensor("op_12321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12321_cast_fp16 = slice_by_index(begin = var_12321_begin_0, end = var_12321_end_0, end_mask = var_12321_end_mask_0, x = var_11821_cast_fp16)[name = tensor("op_12321_cast_fp16")]; + tensor var_12328_begin_0 = const()[name = tensor("op_12328_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12328_end_0 = const()[name = tensor("op_12328_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12328_end_mask_0 = const()[name = tensor("op_12328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12328_cast_fp16 = slice_by_index(begin = var_12328_begin_0, end = var_12328_end_0, end_mask = var_12328_end_mask_0, x = var_11821_cast_fp16)[name = tensor("op_12328_cast_fp16")]; + tensor var_12335_begin_0 = const()[name = tensor("op_12335_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12335_end_0 = const()[name = tensor("op_12335_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12335_end_mask_0 = const()[name = tensor("op_12335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12335_cast_fp16 = slice_by_index(begin = var_12335_begin_0, end = var_12335_end_0, end_mask = var_12335_end_mask_0, x = var_11821_cast_fp16)[name = tensor("op_12335_cast_fp16")]; + tensor var_12342_begin_0 = const()[name = tensor("op_12342_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12342_end_0 = const()[name = tensor("op_12342_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12342_end_mask_0 = const()[name = tensor("op_12342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12342_cast_fp16 = slice_by_index(begin = var_12342_begin_0, end = var_12342_end_0, end_mask = var_12342_end_mask_0, x = var_11825_cast_fp16)[name = tensor("op_12342_cast_fp16")]; + tensor var_12349_begin_0 = const()[name = tensor("op_12349_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12349_end_0 = const()[name = tensor("op_12349_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12349_end_mask_0 = const()[name = tensor("op_12349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12349_cast_fp16 = slice_by_index(begin = var_12349_begin_0, end = var_12349_end_0, end_mask = var_12349_end_mask_0, x = var_11825_cast_fp16)[name = tensor("op_12349_cast_fp16")]; + tensor var_12356_begin_0 = const()[name = tensor("op_12356_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12356_end_0 = const()[name = tensor("op_12356_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12356_end_mask_0 = const()[name = tensor("op_12356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12356_cast_fp16 = slice_by_index(begin = var_12356_begin_0, end = var_12356_end_0, end_mask = var_12356_end_mask_0, x = var_11825_cast_fp16)[name = tensor("op_12356_cast_fp16")]; + tensor var_12363_begin_0 = const()[name = tensor("op_12363_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12363_end_0 = const()[name = tensor("op_12363_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12363_end_mask_0 = const()[name = tensor("op_12363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12363_cast_fp16 = slice_by_index(begin = var_12363_begin_0, end = var_12363_end_0, end_mask = var_12363_end_mask_0, x = var_11825_cast_fp16)[name = tensor("op_12363_cast_fp16")]; + tensor var_12370_begin_0 = const()[name = tensor("op_12370_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12370_end_0 = const()[name = tensor("op_12370_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_12370_end_mask_0 = const()[name = tensor("op_12370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12370_cast_fp16 = slice_by_index(begin = var_12370_begin_0, end = var_12370_end_0, end_mask = var_12370_end_mask_0, x = var_11829_cast_fp16)[name = tensor("op_12370_cast_fp16")]; + tensor var_12377_begin_0 = const()[name = tensor("op_12377_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_12377_end_0 = const()[name = tensor("op_12377_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_12377_end_mask_0 = const()[name = tensor("op_12377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12377_cast_fp16 = slice_by_index(begin = var_12377_begin_0, end = var_12377_end_0, end_mask = var_12377_end_mask_0, x = var_11829_cast_fp16)[name = tensor("op_12377_cast_fp16")]; + tensor var_12384_begin_0 = const()[name = tensor("op_12384_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_12384_end_0 = const()[name = tensor("op_12384_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_12384_end_mask_0 = const()[name = tensor("op_12384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12384_cast_fp16 = slice_by_index(begin = var_12384_begin_0, end = var_12384_end_0, end_mask = var_12384_end_mask_0, x = var_11829_cast_fp16)[name = tensor("op_12384_cast_fp16")]; + tensor var_12391_begin_0 = const()[name = tensor("op_12391_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_12391_end_0 = const()[name = tensor("op_12391_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12391_end_mask_0 = const()[name = tensor("op_12391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12391_cast_fp16 = slice_by_index(begin = var_12391_begin_0, end = var_12391_end_0, end_mask = var_12391_end_mask_0, x = var_11829_cast_fp16)[name = tensor("op_12391_cast_fp16")]; + tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_12396_begin_0 = const()[name = tensor("op_12396_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12396_end_0 = const()[name = tensor("op_12396_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_12396_end_mask_0 = const()[name = tensor("op_12396_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_24 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = tensor("transpose_24")]; + tensor var_12396_cast_fp16 = slice_by_index(begin = var_12396_begin_0, end = var_12396_end_0, end_mask = var_12396_end_mask_0, x = transpose_24)[name = tensor("op_12396_cast_fp16")]; + tensor var_12400_begin_0 = const()[name = tensor("op_12400_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_12400_end_0 = const()[name = tensor("op_12400_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_12400_end_mask_0 = const()[name = tensor("op_12400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12400_cast_fp16 = slice_by_index(begin = var_12400_begin_0, end = var_12400_end_0, end_mask = var_12400_end_mask_0, x = transpose_24)[name = tensor("op_12400_cast_fp16")]; + tensor var_12404_begin_0 = const()[name = tensor("op_12404_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_12404_end_0 = const()[name = tensor("op_12404_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_12404_end_mask_0 = const()[name = tensor("op_12404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12404_cast_fp16 = slice_by_index(begin = var_12404_begin_0, end = var_12404_end_0, end_mask = var_12404_end_mask_0, x = transpose_24)[name = tensor("op_12404_cast_fp16")]; + tensor var_12408_begin_0 = const()[name = tensor("op_12408_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_12408_end_0 = const()[name = tensor("op_12408_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_12408_end_mask_0 = const()[name = tensor("op_12408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12408_cast_fp16 = slice_by_index(begin = var_12408_begin_0, end = var_12408_end_0, end_mask = var_12408_end_mask_0, x = transpose_24)[name = tensor("op_12408_cast_fp16")]; + tensor var_12412_begin_0 = const()[name = tensor("op_12412_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_12412_end_0 = const()[name = tensor("op_12412_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_12412_end_mask_0 = const()[name = tensor("op_12412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12412_cast_fp16 = slice_by_index(begin = var_12412_begin_0, end = var_12412_end_0, end_mask = var_12412_end_mask_0, x = transpose_24)[name = tensor("op_12412_cast_fp16")]; + tensor var_12416_begin_0 = const()[name = tensor("op_12416_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_12416_end_0 = const()[name = tensor("op_12416_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_12416_end_mask_0 = const()[name = tensor("op_12416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12416_cast_fp16 = slice_by_index(begin = var_12416_begin_0, end = var_12416_end_0, end_mask = var_12416_end_mask_0, x = transpose_24)[name = tensor("op_12416_cast_fp16")]; + tensor var_12420_begin_0 = const()[name = tensor("op_12420_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_12420_end_0 = const()[name = tensor("op_12420_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_12420_end_mask_0 = const()[name = tensor("op_12420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12420_cast_fp16 = slice_by_index(begin = var_12420_begin_0, end = var_12420_end_0, end_mask = var_12420_end_mask_0, x = transpose_24)[name = tensor("op_12420_cast_fp16")]; + tensor var_12424_begin_0 = const()[name = tensor("op_12424_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_12424_end_0 = const()[name = tensor("op_12424_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_12424_end_mask_0 = const()[name = tensor("op_12424_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12424_cast_fp16 = slice_by_index(begin = var_12424_begin_0, end = var_12424_end_0, end_mask = var_12424_end_mask_0, x = transpose_24)[name = tensor("op_12424_cast_fp16")]; + tensor var_12428_begin_0 = const()[name = tensor("op_12428_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_12428_end_0 = const()[name = tensor("op_12428_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_12428_end_mask_0 = const()[name = tensor("op_12428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12428_cast_fp16 = slice_by_index(begin = var_12428_begin_0, end = var_12428_end_0, end_mask = var_12428_end_mask_0, x = transpose_24)[name = tensor("op_12428_cast_fp16")]; + tensor var_12432_begin_0 = const()[name = tensor("op_12432_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_12432_end_0 = const()[name = tensor("op_12432_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_12432_end_mask_0 = const()[name = tensor("op_12432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12432_cast_fp16 = slice_by_index(begin = var_12432_begin_0, end = var_12432_end_0, end_mask = var_12432_end_mask_0, x = transpose_24)[name = tensor("op_12432_cast_fp16")]; + tensor var_12436_begin_0 = const()[name = tensor("op_12436_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_12436_end_0 = const()[name = tensor("op_12436_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_12436_end_mask_0 = const()[name = tensor("op_12436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12436_cast_fp16 = slice_by_index(begin = var_12436_begin_0, end = var_12436_end_0, end_mask = var_12436_end_mask_0, x = transpose_24)[name = tensor("op_12436_cast_fp16")]; + tensor var_12440_begin_0 = const()[name = tensor("op_12440_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_12440_end_0 = const()[name = tensor("op_12440_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_12440_end_mask_0 = const()[name = tensor("op_12440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12440_cast_fp16 = slice_by_index(begin = var_12440_begin_0, end = var_12440_end_0, end_mask = var_12440_end_mask_0, x = transpose_24)[name = tensor("op_12440_cast_fp16")]; + tensor var_12444_begin_0 = const()[name = tensor("op_12444_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_12444_end_0 = const()[name = tensor("op_12444_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_12444_end_mask_0 = const()[name = tensor("op_12444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12444_cast_fp16 = slice_by_index(begin = var_12444_begin_0, end = var_12444_end_0, end_mask = var_12444_end_mask_0, x = transpose_24)[name = tensor("op_12444_cast_fp16")]; + tensor var_12448_begin_0 = const()[name = tensor("op_12448_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_12448_end_0 = const()[name = tensor("op_12448_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_12448_end_mask_0 = const()[name = tensor("op_12448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12448_cast_fp16 = slice_by_index(begin = var_12448_begin_0, end = var_12448_end_0, end_mask = var_12448_end_mask_0, x = transpose_24)[name = tensor("op_12448_cast_fp16")]; + tensor var_12452_begin_0 = const()[name = tensor("op_12452_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_12452_end_0 = const()[name = tensor("op_12452_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_12452_end_mask_0 = const()[name = tensor("op_12452_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12452_cast_fp16 = slice_by_index(begin = var_12452_begin_0, end = var_12452_end_0, end_mask = var_12452_end_mask_0, x = transpose_24)[name = tensor("op_12452_cast_fp16")]; + tensor var_12456_begin_0 = const()[name = tensor("op_12456_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_12456_end_0 = const()[name = tensor("op_12456_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_12456_end_mask_0 = const()[name = tensor("op_12456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12456_cast_fp16 = slice_by_index(begin = var_12456_begin_0, end = var_12456_end_0, end_mask = var_12456_end_mask_0, x = transpose_24)[name = tensor("op_12456_cast_fp16")]; + tensor var_12460_begin_0 = const()[name = tensor("op_12460_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_12460_end_0 = const()[name = tensor("op_12460_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_12460_end_mask_0 = const()[name = tensor("op_12460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12460_cast_fp16 = slice_by_index(begin = var_12460_begin_0, end = var_12460_end_0, end_mask = var_12460_end_mask_0, x = transpose_24)[name = tensor("op_12460_cast_fp16")]; + tensor var_12464_begin_0 = const()[name = tensor("op_12464_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_12464_end_0 = const()[name = tensor("op_12464_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_12464_end_mask_0 = const()[name = tensor("op_12464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12464_cast_fp16 = slice_by_index(begin = var_12464_begin_0, end = var_12464_end_0, end_mask = var_12464_end_mask_0, x = transpose_24)[name = tensor("op_12464_cast_fp16")]; + tensor var_12468_begin_0 = const()[name = tensor("op_12468_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_12468_end_0 = const()[name = tensor("op_12468_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_12468_end_mask_0 = const()[name = tensor("op_12468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12468_cast_fp16 = slice_by_index(begin = var_12468_begin_0, end = var_12468_end_0, end_mask = var_12468_end_mask_0, x = transpose_24)[name = tensor("op_12468_cast_fp16")]; + tensor var_12472_begin_0 = const()[name = tensor("op_12472_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_12472_end_0 = const()[name = tensor("op_12472_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_12472_end_mask_0 = const()[name = tensor("op_12472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_12472_cast_fp16 = slice_by_index(begin = var_12472_begin_0, end = var_12472_end_0, end_mask = var_12472_end_mask_0, x = transpose_24)[name = tensor("op_12472_cast_fp16")]; + tensor var_12474_begin_0 = const()[name = tensor("op_12474_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_12474_end_0 = const()[name = tensor("op_12474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_12474_end_mask_0 = const()[name = tensor("op_12474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12474_cast_fp16 = slice_by_index(begin = var_12474_begin_0, end = var_12474_end_0, end_mask = var_12474_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12474_cast_fp16")]; + tensor var_12478_begin_0 = const()[name = tensor("op_12478_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_12478_end_0 = const()[name = tensor("op_12478_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_12478_end_mask_0 = const()[name = tensor("op_12478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12478_cast_fp16 = slice_by_index(begin = var_12478_begin_0, end = var_12478_end_0, end_mask = var_12478_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12478_cast_fp16")]; + tensor var_12482_begin_0 = const()[name = tensor("op_12482_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_12482_end_0 = const()[name = tensor("op_12482_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_12482_end_mask_0 = const()[name = tensor("op_12482_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12482_cast_fp16 = slice_by_index(begin = var_12482_begin_0, end = var_12482_end_0, end_mask = var_12482_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12482_cast_fp16")]; + tensor var_12486_begin_0 = const()[name = tensor("op_12486_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_12486_end_0 = const()[name = tensor("op_12486_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_12486_end_mask_0 = const()[name = tensor("op_12486_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12486_cast_fp16 = slice_by_index(begin = var_12486_begin_0, end = var_12486_end_0, end_mask = var_12486_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12486_cast_fp16")]; + tensor var_12490_begin_0 = const()[name = tensor("op_12490_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_12490_end_0 = const()[name = tensor("op_12490_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_12490_end_mask_0 = const()[name = tensor("op_12490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12490_cast_fp16 = slice_by_index(begin = var_12490_begin_0, end = var_12490_end_0, end_mask = var_12490_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12490_cast_fp16")]; + tensor var_12494_begin_0 = const()[name = tensor("op_12494_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_12494_end_0 = const()[name = tensor("op_12494_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_12494_end_mask_0 = const()[name = tensor("op_12494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12494_cast_fp16 = slice_by_index(begin = var_12494_begin_0, end = var_12494_end_0, end_mask = var_12494_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12494_cast_fp16")]; + tensor var_12498_begin_0 = const()[name = tensor("op_12498_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_12498_end_0 = const()[name = tensor("op_12498_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_12498_end_mask_0 = const()[name = tensor("op_12498_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12498_cast_fp16 = slice_by_index(begin = var_12498_begin_0, end = var_12498_end_0, end_mask = var_12498_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12498_cast_fp16")]; + tensor var_12502_begin_0 = const()[name = tensor("op_12502_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_12502_end_0 = const()[name = tensor("op_12502_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_12502_end_mask_0 = const()[name = tensor("op_12502_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12502_cast_fp16 = slice_by_index(begin = var_12502_begin_0, end = var_12502_end_0, end_mask = var_12502_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12502_cast_fp16")]; + tensor var_12506_begin_0 = const()[name = tensor("op_12506_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_12506_end_0 = const()[name = tensor("op_12506_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_12506_end_mask_0 = const()[name = tensor("op_12506_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12506_cast_fp16 = slice_by_index(begin = var_12506_begin_0, end = var_12506_end_0, end_mask = var_12506_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12506_cast_fp16")]; + tensor var_12510_begin_0 = const()[name = tensor("op_12510_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_12510_end_0 = const()[name = tensor("op_12510_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_12510_end_mask_0 = const()[name = tensor("op_12510_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12510_cast_fp16 = slice_by_index(begin = var_12510_begin_0, end = var_12510_end_0, end_mask = var_12510_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12510_cast_fp16")]; + tensor var_12514_begin_0 = const()[name = tensor("op_12514_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_12514_end_0 = const()[name = tensor("op_12514_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_12514_end_mask_0 = const()[name = tensor("op_12514_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12514_cast_fp16 = slice_by_index(begin = var_12514_begin_0, end = var_12514_end_0, end_mask = var_12514_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12514_cast_fp16")]; + tensor var_12518_begin_0 = const()[name = tensor("op_12518_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_12518_end_0 = const()[name = tensor("op_12518_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_12518_end_mask_0 = const()[name = tensor("op_12518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12518_cast_fp16 = slice_by_index(begin = var_12518_begin_0, end = var_12518_end_0, end_mask = var_12518_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12518_cast_fp16")]; + tensor var_12522_begin_0 = const()[name = tensor("op_12522_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_12522_end_0 = const()[name = tensor("op_12522_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_12522_end_mask_0 = const()[name = tensor("op_12522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12522_cast_fp16 = slice_by_index(begin = var_12522_begin_0, end = var_12522_end_0, end_mask = var_12522_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12522_cast_fp16")]; + tensor var_12526_begin_0 = const()[name = tensor("op_12526_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_12526_end_0 = const()[name = tensor("op_12526_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_12526_end_mask_0 = const()[name = tensor("op_12526_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12526_cast_fp16 = slice_by_index(begin = var_12526_begin_0, end = var_12526_end_0, end_mask = var_12526_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12526_cast_fp16")]; + tensor var_12530_begin_0 = const()[name = tensor("op_12530_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_12530_end_0 = const()[name = tensor("op_12530_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_12530_end_mask_0 = const()[name = tensor("op_12530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12530_cast_fp16 = slice_by_index(begin = var_12530_begin_0, end = var_12530_end_0, end_mask = var_12530_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12530_cast_fp16")]; + tensor var_12534_begin_0 = const()[name = tensor("op_12534_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_12534_end_0 = const()[name = tensor("op_12534_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_12534_end_mask_0 = const()[name = tensor("op_12534_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12534_cast_fp16 = slice_by_index(begin = var_12534_begin_0, end = var_12534_end_0, end_mask = var_12534_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12534_cast_fp16")]; + tensor var_12538_begin_0 = const()[name = tensor("op_12538_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_12538_end_0 = const()[name = tensor("op_12538_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_12538_end_mask_0 = const()[name = tensor("op_12538_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12538_cast_fp16 = slice_by_index(begin = var_12538_begin_0, end = var_12538_end_0, end_mask = var_12538_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12538_cast_fp16")]; + tensor var_12542_begin_0 = const()[name = tensor("op_12542_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_12542_end_0 = const()[name = tensor("op_12542_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_12542_end_mask_0 = const()[name = tensor("op_12542_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12542_cast_fp16 = slice_by_index(begin = var_12542_begin_0, end = var_12542_end_0, end_mask = var_12542_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12542_cast_fp16")]; + tensor var_12546_begin_0 = const()[name = tensor("op_12546_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_12546_end_0 = const()[name = tensor("op_12546_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_12546_end_mask_0 = const()[name = tensor("op_12546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12546_cast_fp16 = slice_by_index(begin = var_12546_begin_0, end = var_12546_end_0, end_mask = var_12546_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12546_cast_fp16")]; + tensor var_12550_begin_0 = const()[name = tensor("op_12550_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_12550_end_0 = const()[name = tensor("op_12550_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_12550_end_mask_0 = const()[name = tensor("op_12550_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_12550_cast_fp16 = slice_by_index(begin = var_12550_begin_0, end = var_12550_end_0, end_mask = var_12550_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_12550_cast_fp16")]; + tensor var_12554_equation_0 = const()[name = tensor("op_12554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12554_cast_fp16 = einsum(equation = var_12554_equation_0, values = (var_12396_cast_fp16, var_11838_cast_fp16))[name = tensor("op_12554_cast_fp16")]; + tensor var_12555_to_fp16 = const()[name = tensor("op_12555_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1121_cast_fp16 = mul(x = var_12554_cast_fp16, y = var_12555_to_fp16)[name = tensor("aw_chunk_1121_cast_fp16")]; + tensor var_12558_equation_0 = const()[name = tensor("op_12558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12558_cast_fp16 = einsum(equation = var_12558_equation_0, values = (var_12396_cast_fp16, var_11845_cast_fp16))[name = tensor("op_12558_cast_fp16")]; + tensor var_12559_to_fp16 = const()[name = tensor("op_12559_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1123_cast_fp16 = mul(x = var_12558_cast_fp16, y = var_12559_to_fp16)[name = tensor("aw_chunk_1123_cast_fp16")]; + tensor var_12562_equation_0 = const()[name = tensor("op_12562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12562_cast_fp16 = einsum(equation = var_12562_equation_0, values = (var_12396_cast_fp16, var_11852_cast_fp16))[name = tensor("op_12562_cast_fp16")]; + tensor var_12563_to_fp16 = const()[name = tensor("op_12563_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1125_cast_fp16 = mul(x = var_12562_cast_fp16, y = var_12563_to_fp16)[name = tensor("aw_chunk_1125_cast_fp16")]; + tensor var_12566_equation_0 = const()[name = tensor("op_12566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12566_cast_fp16 = einsum(equation = var_12566_equation_0, values = (var_12396_cast_fp16, var_11859_cast_fp16))[name = tensor("op_12566_cast_fp16")]; + tensor var_12567_to_fp16 = const()[name = tensor("op_12567_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1127_cast_fp16 = mul(x = var_12566_cast_fp16, y = var_12567_to_fp16)[name = tensor("aw_chunk_1127_cast_fp16")]; + tensor var_12570_equation_0 = const()[name = tensor("op_12570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12570_cast_fp16 = einsum(equation = var_12570_equation_0, values = (var_12400_cast_fp16, var_11866_cast_fp16))[name = tensor("op_12570_cast_fp16")]; + tensor var_12571_to_fp16 = const()[name = tensor("op_12571_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1129_cast_fp16 = mul(x = var_12570_cast_fp16, y = var_12571_to_fp16)[name = tensor("aw_chunk_1129_cast_fp16")]; + tensor var_12574_equation_0 = const()[name = tensor("op_12574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12574_cast_fp16 = einsum(equation = var_12574_equation_0, values = (var_12400_cast_fp16, var_11873_cast_fp16))[name = tensor("op_12574_cast_fp16")]; + tensor var_12575_to_fp16 = const()[name = tensor("op_12575_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1131_cast_fp16 = mul(x = var_12574_cast_fp16, y = var_12575_to_fp16)[name = tensor("aw_chunk_1131_cast_fp16")]; + tensor var_12578_equation_0 = const()[name = tensor("op_12578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12578_cast_fp16 = einsum(equation = var_12578_equation_0, values = (var_12400_cast_fp16, var_11880_cast_fp16))[name = tensor("op_12578_cast_fp16")]; + tensor var_12579_to_fp16 = const()[name = tensor("op_12579_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1133_cast_fp16 = mul(x = var_12578_cast_fp16, y = var_12579_to_fp16)[name = tensor("aw_chunk_1133_cast_fp16")]; + tensor var_12582_equation_0 = const()[name = tensor("op_12582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12582_cast_fp16 = einsum(equation = var_12582_equation_0, values = (var_12400_cast_fp16, var_11887_cast_fp16))[name = tensor("op_12582_cast_fp16")]; + tensor var_12583_to_fp16 = const()[name = tensor("op_12583_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1135_cast_fp16 = mul(x = var_12582_cast_fp16, y = var_12583_to_fp16)[name = tensor("aw_chunk_1135_cast_fp16")]; + tensor var_12586_equation_0 = const()[name = tensor("op_12586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12586_cast_fp16 = einsum(equation = var_12586_equation_0, values = (var_12404_cast_fp16, var_11894_cast_fp16))[name = tensor("op_12586_cast_fp16")]; + tensor var_12587_to_fp16 = const()[name = tensor("op_12587_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1137_cast_fp16 = mul(x = var_12586_cast_fp16, y = var_12587_to_fp16)[name = tensor("aw_chunk_1137_cast_fp16")]; + tensor var_12590_equation_0 = const()[name = tensor("op_12590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12590_cast_fp16 = einsum(equation = var_12590_equation_0, values = (var_12404_cast_fp16, var_11901_cast_fp16))[name = tensor("op_12590_cast_fp16")]; + tensor var_12591_to_fp16 = const()[name = tensor("op_12591_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1139_cast_fp16 = mul(x = var_12590_cast_fp16, y = var_12591_to_fp16)[name = tensor("aw_chunk_1139_cast_fp16")]; + tensor var_12594_equation_0 = const()[name = tensor("op_12594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12594_cast_fp16 = einsum(equation = var_12594_equation_0, values = (var_12404_cast_fp16, var_11908_cast_fp16))[name = tensor("op_12594_cast_fp16")]; + tensor var_12595_to_fp16 = const()[name = tensor("op_12595_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1141_cast_fp16 = mul(x = var_12594_cast_fp16, y = var_12595_to_fp16)[name = tensor("aw_chunk_1141_cast_fp16")]; + tensor var_12598_equation_0 = const()[name = tensor("op_12598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12598_cast_fp16 = einsum(equation = var_12598_equation_0, values = (var_12404_cast_fp16, var_11915_cast_fp16))[name = tensor("op_12598_cast_fp16")]; + tensor var_12599_to_fp16 = const()[name = tensor("op_12599_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1143_cast_fp16 = mul(x = var_12598_cast_fp16, y = var_12599_to_fp16)[name = tensor("aw_chunk_1143_cast_fp16")]; + tensor var_12602_equation_0 = const()[name = tensor("op_12602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12602_cast_fp16 = einsum(equation = var_12602_equation_0, values = (var_12408_cast_fp16, var_11922_cast_fp16))[name = tensor("op_12602_cast_fp16")]; + tensor var_12603_to_fp16 = const()[name = tensor("op_12603_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1145_cast_fp16 = mul(x = var_12602_cast_fp16, y = var_12603_to_fp16)[name = tensor("aw_chunk_1145_cast_fp16")]; + tensor var_12606_equation_0 = const()[name = tensor("op_12606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12606_cast_fp16 = einsum(equation = var_12606_equation_0, values = (var_12408_cast_fp16, var_11929_cast_fp16))[name = tensor("op_12606_cast_fp16")]; + tensor var_12607_to_fp16 = const()[name = tensor("op_12607_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1147_cast_fp16 = mul(x = var_12606_cast_fp16, y = var_12607_to_fp16)[name = tensor("aw_chunk_1147_cast_fp16")]; + tensor var_12610_equation_0 = const()[name = tensor("op_12610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12610_cast_fp16 = einsum(equation = var_12610_equation_0, values = (var_12408_cast_fp16, var_11936_cast_fp16))[name = tensor("op_12610_cast_fp16")]; + tensor var_12611_to_fp16 = const()[name = tensor("op_12611_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1149_cast_fp16 = mul(x = var_12610_cast_fp16, y = var_12611_to_fp16)[name = tensor("aw_chunk_1149_cast_fp16")]; + tensor var_12614_equation_0 = const()[name = tensor("op_12614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12614_cast_fp16 = einsum(equation = var_12614_equation_0, values = (var_12408_cast_fp16, var_11943_cast_fp16))[name = tensor("op_12614_cast_fp16")]; + tensor var_12615_to_fp16 = const()[name = tensor("op_12615_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1151_cast_fp16 = mul(x = var_12614_cast_fp16, y = var_12615_to_fp16)[name = tensor("aw_chunk_1151_cast_fp16")]; + tensor var_12618_equation_0 = const()[name = tensor("op_12618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12618_cast_fp16 = einsum(equation = var_12618_equation_0, values = (var_12412_cast_fp16, var_11950_cast_fp16))[name = tensor("op_12618_cast_fp16")]; + tensor var_12619_to_fp16 = const()[name = tensor("op_12619_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1153_cast_fp16 = mul(x = var_12618_cast_fp16, y = var_12619_to_fp16)[name = tensor("aw_chunk_1153_cast_fp16")]; + tensor var_12622_equation_0 = const()[name = tensor("op_12622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12622_cast_fp16 = einsum(equation = var_12622_equation_0, values = (var_12412_cast_fp16, var_11957_cast_fp16))[name = tensor("op_12622_cast_fp16")]; + tensor var_12623_to_fp16 = const()[name = tensor("op_12623_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1155_cast_fp16 = mul(x = var_12622_cast_fp16, y = var_12623_to_fp16)[name = tensor("aw_chunk_1155_cast_fp16")]; + tensor var_12626_equation_0 = const()[name = tensor("op_12626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12626_cast_fp16 = einsum(equation = var_12626_equation_0, values = (var_12412_cast_fp16, var_11964_cast_fp16))[name = tensor("op_12626_cast_fp16")]; + tensor var_12627_to_fp16 = const()[name = tensor("op_12627_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1157_cast_fp16 = mul(x = var_12626_cast_fp16, y = var_12627_to_fp16)[name = tensor("aw_chunk_1157_cast_fp16")]; + tensor var_12630_equation_0 = const()[name = tensor("op_12630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12630_cast_fp16 = einsum(equation = var_12630_equation_0, values = (var_12412_cast_fp16, var_11971_cast_fp16))[name = tensor("op_12630_cast_fp16")]; + tensor var_12631_to_fp16 = const()[name = tensor("op_12631_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1159_cast_fp16 = mul(x = var_12630_cast_fp16, y = var_12631_to_fp16)[name = tensor("aw_chunk_1159_cast_fp16")]; + tensor var_12634_equation_0 = const()[name = tensor("op_12634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12634_cast_fp16 = einsum(equation = var_12634_equation_0, values = (var_12416_cast_fp16, var_11978_cast_fp16))[name = tensor("op_12634_cast_fp16")]; + tensor var_12635_to_fp16 = const()[name = tensor("op_12635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1161_cast_fp16 = mul(x = var_12634_cast_fp16, y = var_12635_to_fp16)[name = tensor("aw_chunk_1161_cast_fp16")]; + tensor var_12638_equation_0 = const()[name = tensor("op_12638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12638_cast_fp16 = einsum(equation = var_12638_equation_0, values = (var_12416_cast_fp16, var_11985_cast_fp16))[name = tensor("op_12638_cast_fp16")]; + tensor var_12639_to_fp16 = const()[name = tensor("op_12639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1163_cast_fp16 = mul(x = var_12638_cast_fp16, y = var_12639_to_fp16)[name = tensor("aw_chunk_1163_cast_fp16")]; + tensor var_12642_equation_0 = const()[name = tensor("op_12642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12642_cast_fp16 = einsum(equation = var_12642_equation_0, values = (var_12416_cast_fp16, var_11992_cast_fp16))[name = tensor("op_12642_cast_fp16")]; + tensor var_12643_to_fp16 = const()[name = tensor("op_12643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1165_cast_fp16 = mul(x = var_12642_cast_fp16, y = var_12643_to_fp16)[name = tensor("aw_chunk_1165_cast_fp16")]; + tensor var_12646_equation_0 = const()[name = tensor("op_12646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12646_cast_fp16 = einsum(equation = var_12646_equation_0, values = (var_12416_cast_fp16, var_11999_cast_fp16))[name = tensor("op_12646_cast_fp16")]; + tensor var_12647_to_fp16 = const()[name = tensor("op_12647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1167_cast_fp16 = mul(x = var_12646_cast_fp16, y = var_12647_to_fp16)[name = tensor("aw_chunk_1167_cast_fp16")]; + tensor var_12650_equation_0 = const()[name = tensor("op_12650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12650_cast_fp16 = einsum(equation = var_12650_equation_0, values = (var_12420_cast_fp16, var_12006_cast_fp16))[name = tensor("op_12650_cast_fp16")]; + tensor var_12651_to_fp16 = const()[name = tensor("op_12651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1169_cast_fp16 = mul(x = var_12650_cast_fp16, y = var_12651_to_fp16)[name = tensor("aw_chunk_1169_cast_fp16")]; + tensor var_12654_equation_0 = const()[name = tensor("op_12654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12654_cast_fp16 = einsum(equation = var_12654_equation_0, values = (var_12420_cast_fp16, var_12013_cast_fp16))[name = tensor("op_12654_cast_fp16")]; + tensor var_12655_to_fp16 = const()[name = tensor("op_12655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1171_cast_fp16 = mul(x = var_12654_cast_fp16, y = var_12655_to_fp16)[name = tensor("aw_chunk_1171_cast_fp16")]; + tensor var_12658_equation_0 = const()[name = tensor("op_12658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12658_cast_fp16 = einsum(equation = var_12658_equation_0, values = (var_12420_cast_fp16, var_12020_cast_fp16))[name = tensor("op_12658_cast_fp16")]; + tensor var_12659_to_fp16 = const()[name = tensor("op_12659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1173_cast_fp16 = mul(x = var_12658_cast_fp16, y = var_12659_to_fp16)[name = tensor("aw_chunk_1173_cast_fp16")]; + tensor var_12662_equation_0 = const()[name = tensor("op_12662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12662_cast_fp16 = einsum(equation = var_12662_equation_0, values = (var_12420_cast_fp16, var_12027_cast_fp16))[name = tensor("op_12662_cast_fp16")]; + tensor var_12663_to_fp16 = const()[name = tensor("op_12663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1175_cast_fp16 = mul(x = var_12662_cast_fp16, y = var_12663_to_fp16)[name = tensor("aw_chunk_1175_cast_fp16")]; + tensor var_12666_equation_0 = const()[name = tensor("op_12666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12666_cast_fp16 = einsum(equation = var_12666_equation_0, values = (var_12424_cast_fp16, var_12034_cast_fp16))[name = tensor("op_12666_cast_fp16")]; + tensor var_12667_to_fp16 = const()[name = tensor("op_12667_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1177_cast_fp16 = mul(x = var_12666_cast_fp16, y = var_12667_to_fp16)[name = tensor("aw_chunk_1177_cast_fp16")]; + tensor var_12670_equation_0 = const()[name = tensor("op_12670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12670_cast_fp16 = einsum(equation = var_12670_equation_0, values = (var_12424_cast_fp16, var_12041_cast_fp16))[name = tensor("op_12670_cast_fp16")]; + tensor var_12671_to_fp16 = const()[name = tensor("op_12671_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1179_cast_fp16 = mul(x = var_12670_cast_fp16, y = var_12671_to_fp16)[name = tensor("aw_chunk_1179_cast_fp16")]; + tensor var_12674_equation_0 = const()[name = tensor("op_12674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12674_cast_fp16 = einsum(equation = var_12674_equation_0, values = (var_12424_cast_fp16, var_12048_cast_fp16))[name = tensor("op_12674_cast_fp16")]; + tensor var_12675_to_fp16 = const()[name = tensor("op_12675_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1181_cast_fp16 = mul(x = var_12674_cast_fp16, y = var_12675_to_fp16)[name = tensor("aw_chunk_1181_cast_fp16")]; + tensor var_12678_equation_0 = const()[name = tensor("op_12678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12678_cast_fp16 = einsum(equation = var_12678_equation_0, values = (var_12424_cast_fp16, var_12055_cast_fp16))[name = tensor("op_12678_cast_fp16")]; + tensor var_12679_to_fp16 = const()[name = tensor("op_12679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1183_cast_fp16 = mul(x = var_12678_cast_fp16, y = var_12679_to_fp16)[name = tensor("aw_chunk_1183_cast_fp16")]; + tensor var_12682_equation_0 = const()[name = tensor("op_12682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12682_cast_fp16 = einsum(equation = var_12682_equation_0, values = (var_12428_cast_fp16, var_12062_cast_fp16))[name = tensor("op_12682_cast_fp16")]; + tensor var_12683_to_fp16 = const()[name = tensor("op_12683_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1185_cast_fp16 = mul(x = var_12682_cast_fp16, y = var_12683_to_fp16)[name = tensor("aw_chunk_1185_cast_fp16")]; + tensor var_12686_equation_0 = const()[name = tensor("op_12686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12686_cast_fp16 = einsum(equation = var_12686_equation_0, values = (var_12428_cast_fp16, var_12069_cast_fp16))[name = tensor("op_12686_cast_fp16")]; + tensor var_12687_to_fp16 = const()[name = tensor("op_12687_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1187_cast_fp16 = mul(x = var_12686_cast_fp16, y = var_12687_to_fp16)[name = tensor("aw_chunk_1187_cast_fp16")]; + tensor var_12690_equation_0 = const()[name = tensor("op_12690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12690_cast_fp16 = einsum(equation = var_12690_equation_0, values = (var_12428_cast_fp16, var_12076_cast_fp16))[name = tensor("op_12690_cast_fp16")]; + tensor var_12691_to_fp16 = const()[name = tensor("op_12691_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1189_cast_fp16 = mul(x = var_12690_cast_fp16, y = var_12691_to_fp16)[name = tensor("aw_chunk_1189_cast_fp16")]; + tensor var_12694_equation_0 = const()[name = tensor("op_12694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12694_cast_fp16 = einsum(equation = var_12694_equation_0, values = (var_12428_cast_fp16, var_12083_cast_fp16))[name = tensor("op_12694_cast_fp16")]; + tensor var_12695_to_fp16 = const()[name = tensor("op_12695_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1191_cast_fp16 = mul(x = var_12694_cast_fp16, y = var_12695_to_fp16)[name = tensor("aw_chunk_1191_cast_fp16")]; + tensor var_12698_equation_0 = const()[name = tensor("op_12698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12698_cast_fp16 = einsum(equation = var_12698_equation_0, values = (var_12432_cast_fp16, var_12090_cast_fp16))[name = tensor("op_12698_cast_fp16")]; + tensor var_12699_to_fp16 = const()[name = tensor("op_12699_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1193_cast_fp16 = mul(x = var_12698_cast_fp16, y = var_12699_to_fp16)[name = tensor("aw_chunk_1193_cast_fp16")]; + tensor var_12702_equation_0 = const()[name = tensor("op_12702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12702_cast_fp16 = einsum(equation = var_12702_equation_0, values = (var_12432_cast_fp16, var_12097_cast_fp16))[name = tensor("op_12702_cast_fp16")]; + tensor var_12703_to_fp16 = const()[name = tensor("op_12703_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1195_cast_fp16 = mul(x = var_12702_cast_fp16, y = var_12703_to_fp16)[name = tensor("aw_chunk_1195_cast_fp16")]; + tensor var_12706_equation_0 = const()[name = tensor("op_12706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12706_cast_fp16 = einsum(equation = var_12706_equation_0, values = (var_12432_cast_fp16, var_12104_cast_fp16))[name = tensor("op_12706_cast_fp16")]; + tensor var_12707_to_fp16 = const()[name = tensor("op_12707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1197_cast_fp16 = mul(x = var_12706_cast_fp16, y = var_12707_to_fp16)[name = tensor("aw_chunk_1197_cast_fp16")]; + tensor var_12710_equation_0 = const()[name = tensor("op_12710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12710_cast_fp16 = einsum(equation = var_12710_equation_0, values = (var_12432_cast_fp16, var_12111_cast_fp16))[name = tensor("op_12710_cast_fp16")]; + tensor var_12711_to_fp16 = const()[name = tensor("op_12711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1199_cast_fp16 = mul(x = var_12710_cast_fp16, y = var_12711_to_fp16)[name = tensor("aw_chunk_1199_cast_fp16")]; + tensor var_12714_equation_0 = const()[name = tensor("op_12714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12714_cast_fp16 = einsum(equation = var_12714_equation_0, values = (var_12436_cast_fp16, var_12118_cast_fp16))[name = tensor("op_12714_cast_fp16")]; + tensor var_12715_to_fp16 = const()[name = tensor("op_12715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1201_cast_fp16 = mul(x = var_12714_cast_fp16, y = var_12715_to_fp16)[name = tensor("aw_chunk_1201_cast_fp16")]; + tensor var_12718_equation_0 = const()[name = tensor("op_12718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12718_cast_fp16 = einsum(equation = var_12718_equation_0, values = (var_12436_cast_fp16, var_12125_cast_fp16))[name = tensor("op_12718_cast_fp16")]; + tensor var_12719_to_fp16 = const()[name = tensor("op_12719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1203_cast_fp16 = mul(x = var_12718_cast_fp16, y = var_12719_to_fp16)[name = tensor("aw_chunk_1203_cast_fp16")]; + tensor var_12722_equation_0 = const()[name = tensor("op_12722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12722_cast_fp16 = einsum(equation = var_12722_equation_0, values = (var_12436_cast_fp16, var_12132_cast_fp16))[name = tensor("op_12722_cast_fp16")]; + tensor var_12723_to_fp16 = const()[name = tensor("op_12723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1205_cast_fp16 = mul(x = var_12722_cast_fp16, y = var_12723_to_fp16)[name = tensor("aw_chunk_1205_cast_fp16")]; + tensor var_12726_equation_0 = const()[name = tensor("op_12726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12726_cast_fp16 = einsum(equation = var_12726_equation_0, values = (var_12436_cast_fp16, var_12139_cast_fp16))[name = tensor("op_12726_cast_fp16")]; + tensor var_12727_to_fp16 = const()[name = tensor("op_12727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1207_cast_fp16 = mul(x = var_12726_cast_fp16, y = var_12727_to_fp16)[name = tensor("aw_chunk_1207_cast_fp16")]; + tensor var_12730_equation_0 = const()[name = tensor("op_12730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12730_cast_fp16 = einsum(equation = var_12730_equation_0, values = (var_12440_cast_fp16, var_12146_cast_fp16))[name = tensor("op_12730_cast_fp16")]; + tensor var_12731_to_fp16 = const()[name = tensor("op_12731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1209_cast_fp16 = mul(x = var_12730_cast_fp16, y = var_12731_to_fp16)[name = tensor("aw_chunk_1209_cast_fp16")]; + tensor var_12734_equation_0 = const()[name = tensor("op_12734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12734_cast_fp16 = einsum(equation = var_12734_equation_0, values = (var_12440_cast_fp16, var_12153_cast_fp16))[name = tensor("op_12734_cast_fp16")]; + tensor var_12735_to_fp16 = const()[name = tensor("op_12735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1211_cast_fp16 = mul(x = var_12734_cast_fp16, y = var_12735_to_fp16)[name = tensor("aw_chunk_1211_cast_fp16")]; + tensor var_12738_equation_0 = const()[name = tensor("op_12738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12738_cast_fp16 = einsum(equation = var_12738_equation_0, values = (var_12440_cast_fp16, var_12160_cast_fp16))[name = tensor("op_12738_cast_fp16")]; + tensor var_12739_to_fp16 = const()[name = tensor("op_12739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1213_cast_fp16 = mul(x = var_12738_cast_fp16, y = var_12739_to_fp16)[name = tensor("aw_chunk_1213_cast_fp16")]; + tensor var_12742_equation_0 = const()[name = tensor("op_12742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12742_cast_fp16 = einsum(equation = var_12742_equation_0, values = (var_12440_cast_fp16, var_12167_cast_fp16))[name = tensor("op_12742_cast_fp16")]; + tensor var_12743_to_fp16 = const()[name = tensor("op_12743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1215_cast_fp16 = mul(x = var_12742_cast_fp16, y = var_12743_to_fp16)[name = tensor("aw_chunk_1215_cast_fp16")]; + tensor var_12746_equation_0 = const()[name = tensor("op_12746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12746_cast_fp16 = einsum(equation = var_12746_equation_0, values = (var_12444_cast_fp16, var_12174_cast_fp16))[name = tensor("op_12746_cast_fp16")]; + tensor var_12747_to_fp16 = const()[name = tensor("op_12747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1217_cast_fp16 = mul(x = var_12746_cast_fp16, y = var_12747_to_fp16)[name = tensor("aw_chunk_1217_cast_fp16")]; + tensor var_12750_equation_0 = const()[name = tensor("op_12750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12750_cast_fp16 = einsum(equation = var_12750_equation_0, values = (var_12444_cast_fp16, var_12181_cast_fp16))[name = tensor("op_12750_cast_fp16")]; + tensor var_12751_to_fp16 = const()[name = tensor("op_12751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1219_cast_fp16 = mul(x = var_12750_cast_fp16, y = var_12751_to_fp16)[name = tensor("aw_chunk_1219_cast_fp16")]; + tensor var_12754_equation_0 = const()[name = tensor("op_12754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12754_cast_fp16 = einsum(equation = var_12754_equation_0, values = (var_12444_cast_fp16, var_12188_cast_fp16))[name = tensor("op_12754_cast_fp16")]; + tensor var_12755_to_fp16 = const()[name = tensor("op_12755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1221_cast_fp16 = mul(x = var_12754_cast_fp16, y = var_12755_to_fp16)[name = tensor("aw_chunk_1221_cast_fp16")]; + tensor var_12758_equation_0 = const()[name = tensor("op_12758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12758_cast_fp16 = einsum(equation = var_12758_equation_0, values = (var_12444_cast_fp16, var_12195_cast_fp16))[name = tensor("op_12758_cast_fp16")]; + tensor var_12759_to_fp16 = const()[name = tensor("op_12759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1223_cast_fp16 = mul(x = var_12758_cast_fp16, y = var_12759_to_fp16)[name = tensor("aw_chunk_1223_cast_fp16")]; + tensor var_12762_equation_0 = const()[name = tensor("op_12762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12762_cast_fp16 = einsum(equation = var_12762_equation_0, values = (var_12448_cast_fp16, var_12202_cast_fp16))[name = tensor("op_12762_cast_fp16")]; + tensor var_12763_to_fp16 = const()[name = tensor("op_12763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1225_cast_fp16 = mul(x = var_12762_cast_fp16, y = var_12763_to_fp16)[name = tensor("aw_chunk_1225_cast_fp16")]; + tensor var_12766_equation_0 = const()[name = tensor("op_12766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12766_cast_fp16 = einsum(equation = var_12766_equation_0, values = (var_12448_cast_fp16, var_12209_cast_fp16))[name = tensor("op_12766_cast_fp16")]; + tensor var_12767_to_fp16 = const()[name = tensor("op_12767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1227_cast_fp16 = mul(x = var_12766_cast_fp16, y = var_12767_to_fp16)[name = tensor("aw_chunk_1227_cast_fp16")]; + tensor var_12770_equation_0 = const()[name = tensor("op_12770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12770_cast_fp16 = einsum(equation = var_12770_equation_0, values = (var_12448_cast_fp16, var_12216_cast_fp16))[name = tensor("op_12770_cast_fp16")]; + tensor var_12771_to_fp16 = const()[name = tensor("op_12771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1229_cast_fp16 = mul(x = var_12770_cast_fp16, y = var_12771_to_fp16)[name = tensor("aw_chunk_1229_cast_fp16")]; + tensor var_12774_equation_0 = const()[name = tensor("op_12774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12774_cast_fp16 = einsum(equation = var_12774_equation_0, values = (var_12448_cast_fp16, var_12223_cast_fp16))[name = tensor("op_12774_cast_fp16")]; + tensor var_12775_to_fp16 = const()[name = tensor("op_12775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1231_cast_fp16 = mul(x = var_12774_cast_fp16, y = var_12775_to_fp16)[name = tensor("aw_chunk_1231_cast_fp16")]; + tensor var_12778_equation_0 = const()[name = tensor("op_12778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12778_cast_fp16 = einsum(equation = var_12778_equation_0, values = (var_12452_cast_fp16, var_12230_cast_fp16))[name = tensor("op_12778_cast_fp16")]; + tensor var_12779_to_fp16 = const()[name = tensor("op_12779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1233_cast_fp16 = mul(x = var_12778_cast_fp16, y = var_12779_to_fp16)[name = tensor("aw_chunk_1233_cast_fp16")]; + tensor var_12782_equation_0 = const()[name = tensor("op_12782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12782_cast_fp16 = einsum(equation = var_12782_equation_0, values = (var_12452_cast_fp16, var_12237_cast_fp16))[name = tensor("op_12782_cast_fp16")]; + tensor var_12783_to_fp16 = const()[name = tensor("op_12783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1235_cast_fp16 = mul(x = var_12782_cast_fp16, y = var_12783_to_fp16)[name = tensor("aw_chunk_1235_cast_fp16")]; + tensor var_12786_equation_0 = const()[name = tensor("op_12786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12786_cast_fp16 = einsum(equation = var_12786_equation_0, values = (var_12452_cast_fp16, var_12244_cast_fp16))[name = tensor("op_12786_cast_fp16")]; + tensor var_12787_to_fp16 = const()[name = tensor("op_12787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1237_cast_fp16 = mul(x = var_12786_cast_fp16, y = var_12787_to_fp16)[name = tensor("aw_chunk_1237_cast_fp16")]; + tensor var_12790_equation_0 = const()[name = tensor("op_12790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12790_cast_fp16 = einsum(equation = var_12790_equation_0, values = (var_12452_cast_fp16, var_12251_cast_fp16))[name = tensor("op_12790_cast_fp16")]; + tensor var_12791_to_fp16 = const()[name = tensor("op_12791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1239_cast_fp16 = mul(x = var_12790_cast_fp16, y = var_12791_to_fp16)[name = tensor("aw_chunk_1239_cast_fp16")]; + tensor var_12794_equation_0 = const()[name = tensor("op_12794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12794_cast_fp16 = einsum(equation = var_12794_equation_0, values = (var_12456_cast_fp16, var_12258_cast_fp16))[name = tensor("op_12794_cast_fp16")]; + tensor var_12795_to_fp16 = const()[name = tensor("op_12795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1241_cast_fp16 = mul(x = var_12794_cast_fp16, y = var_12795_to_fp16)[name = tensor("aw_chunk_1241_cast_fp16")]; + tensor var_12798_equation_0 = const()[name = tensor("op_12798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12798_cast_fp16 = einsum(equation = var_12798_equation_0, values = (var_12456_cast_fp16, var_12265_cast_fp16))[name = tensor("op_12798_cast_fp16")]; + tensor var_12799_to_fp16 = const()[name = tensor("op_12799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1243_cast_fp16 = mul(x = var_12798_cast_fp16, y = var_12799_to_fp16)[name = tensor("aw_chunk_1243_cast_fp16")]; + tensor var_12802_equation_0 = const()[name = tensor("op_12802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12802_cast_fp16 = einsum(equation = var_12802_equation_0, values = (var_12456_cast_fp16, var_12272_cast_fp16))[name = tensor("op_12802_cast_fp16")]; + tensor var_12803_to_fp16 = const()[name = tensor("op_12803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1245_cast_fp16 = mul(x = var_12802_cast_fp16, y = var_12803_to_fp16)[name = tensor("aw_chunk_1245_cast_fp16")]; + tensor var_12806_equation_0 = const()[name = tensor("op_12806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12806_cast_fp16 = einsum(equation = var_12806_equation_0, values = (var_12456_cast_fp16, var_12279_cast_fp16))[name = tensor("op_12806_cast_fp16")]; + tensor var_12807_to_fp16 = const()[name = tensor("op_12807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1247_cast_fp16 = mul(x = var_12806_cast_fp16, y = var_12807_to_fp16)[name = tensor("aw_chunk_1247_cast_fp16")]; + tensor var_12810_equation_0 = const()[name = tensor("op_12810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12810_cast_fp16 = einsum(equation = var_12810_equation_0, values = (var_12460_cast_fp16, var_12286_cast_fp16))[name = tensor("op_12810_cast_fp16")]; + tensor var_12811_to_fp16 = const()[name = tensor("op_12811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1249_cast_fp16 = mul(x = var_12810_cast_fp16, y = var_12811_to_fp16)[name = tensor("aw_chunk_1249_cast_fp16")]; + tensor var_12814_equation_0 = const()[name = tensor("op_12814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12814_cast_fp16 = einsum(equation = var_12814_equation_0, values = (var_12460_cast_fp16, var_12293_cast_fp16))[name = tensor("op_12814_cast_fp16")]; + tensor var_12815_to_fp16 = const()[name = tensor("op_12815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1251_cast_fp16 = mul(x = var_12814_cast_fp16, y = var_12815_to_fp16)[name = tensor("aw_chunk_1251_cast_fp16")]; + tensor var_12818_equation_0 = const()[name = tensor("op_12818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12818_cast_fp16 = einsum(equation = var_12818_equation_0, values = (var_12460_cast_fp16, var_12300_cast_fp16))[name = tensor("op_12818_cast_fp16")]; + tensor var_12819_to_fp16 = const()[name = tensor("op_12819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1253_cast_fp16 = mul(x = var_12818_cast_fp16, y = var_12819_to_fp16)[name = tensor("aw_chunk_1253_cast_fp16")]; + tensor var_12822_equation_0 = const()[name = tensor("op_12822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12822_cast_fp16 = einsum(equation = var_12822_equation_0, values = (var_12460_cast_fp16, var_12307_cast_fp16))[name = tensor("op_12822_cast_fp16")]; + tensor var_12823_to_fp16 = const()[name = tensor("op_12823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1255_cast_fp16 = mul(x = var_12822_cast_fp16, y = var_12823_to_fp16)[name = tensor("aw_chunk_1255_cast_fp16")]; + tensor var_12826_equation_0 = const()[name = tensor("op_12826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12826_cast_fp16 = einsum(equation = var_12826_equation_0, values = (var_12464_cast_fp16, var_12314_cast_fp16))[name = tensor("op_12826_cast_fp16")]; + tensor var_12827_to_fp16 = const()[name = tensor("op_12827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1257_cast_fp16 = mul(x = var_12826_cast_fp16, y = var_12827_to_fp16)[name = tensor("aw_chunk_1257_cast_fp16")]; + tensor var_12830_equation_0 = const()[name = tensor("op_12830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12830_cast_fp16 = einsum(equation = var_12830_equation_0, values = (var_12464_cast_fp16, var_12321_cast_fp16))[name = tensor("op_12830_cast_fp16")]; + tensor var_12831_to_fp16 = const()[name = tensor("op_12831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1259_cast_fp16 = mul(x = var_12830_cast_fp16, y = var_12831_to_fp16)[name = tensor("aw_chunk_1259_cast_fp16")]; + tensor var_12834_equation_0 = const()[name = tensor("op_12834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12834_cast_fp16 = einsum(equation = var_12834_equation_0, values = (var_12464_cast_fp16, var_12328_cast_fp16))[name = tensor("op_12834_cast_fp16")]; + tensor var_12835_to_fp16 = const()[name = tensor("op_12835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1261_cast_fp16 = mul(x = var_12834_cast_fp16, y = var_12835_to_fp16)[name = tensor("aw_chunk_1261_cast_fp16")]; + tensor var_12838_equation_0 = const()[name = tensor("op_12838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12838_cast_fp16 = einsum(equation = var_12838_equation_0, values = (var_12464_cast_fp16, var_12335_cast_fp16))[name = tensor("op_12838_cast_fp16")]; + tensor var_12839_to_fp16 = const()[name = tensor("op_12839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1263_cast_fp16 = mul(x = var_12838_cast_fp16, y = var_12839_to_fp16)[name = tensor("aw_chunk_1263_cast_fp16")]; + tensor var_12842_equation_0 = const()[name = tensor("op_12842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12842_cast_fp16 = einsum(equation = var_12842_equation_0, values = (var_12468_cast_fp16, var_12342_cast_fp16))[name = tensor("op_12842_cast_fp16")]; + tensor var_12843_to_fp16 = const()[name = tensor("op_12843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1265_cast_fp16 = mul(x = var_12842_cast_fp16, y = var_12843_to_fp16)[name = tensor("aw_chunk_1265_cast_fp16")]; + tensor var_12846_equation_0 = const()[name = tensor("op_12846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12846_cast_fp16 = einsum(equation = var_12846_equation_0, values = (var_12468_cast_fp16, var_12349_cast_fp16))[name = tensor("op_12846_cast_fp16")]; + tensor var_12847_to_fp16 = const()[name = tensor("op_12847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1267_cast_fp16 = mul(x = var_12846_cast_fp16, y = var_12847_to_fp16)[name = tensor("aw_chunk_1267_cast_fp16")]; + tensor var_12850_equation_0 = const()[name = tensor("op_12850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12850_cast_fp16 = einsum(equation = var_12850_equation_0, values = (var_12468_cast_fp16, var_12356_cast_fp16))[name = tensor("op_12850_cast_fp16")]; + tensor var_12851_to_fp16 = const()[name = tensor("op_12851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1269_cast_fp16 = mul(x = var_12850_cast_fp16, y = var_12851_to_fp16)[name = tensor("aw_chunk_1269_cast_fp16")]; + tensor var_12854_equation_0 = const()[name = tensor("op_12854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12854_cast_fp16 = einsum(equation = var_12854_equation_0, values = (var_12468_cast_fp16, var_12363_cast_fp16))[name = tensor("op_12854_cast_fp16")]; + tensor var_12855_to_fp16 = const()[name = tensor("op_12855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1271_cast_fp16 = mul(x = var_12854_cast_fp16, y = var_12855_to_fp16)[name = tensor("aw_chunk_1271_cast_fp16")]; + tensor var_12858_equation_0 = const()[name = tensor("op_12858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12858_cast_fp16 = einsum(equation = var_12858_equation_0, values = (var_12472_cast_fp16, var_12370_cast_fp16))[name = tensor("op_12858_cast_fp16")]; + tensor var_12859_to_fp16 = const()[name = tensor("op_12859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1273_cast_fp16 = mul(x = var_12858_cast_fp16, y = var_12859_to_fp16)[name = tensor("aw_chunk_1273_cast_fp16")]; + tensor var_12862_equation_0 = const()[name = tensor("op_12862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12862_cast_fp16 = einsum(equation = var_12862_equation_0, values = (var_12472_cast_fp16, var_12377_cast_fp16))[name = tensor("op_12862_cast_fp16")]; + tensor var_12863_to_fp16 = const()[name = tensor("op_12863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1275_cast_fp16 = mul(x = var_12862_cast_fp16, y = var_12863_to_fp16)[name = tensor("aw_chunk_1275_cast_fp16")]; + tensor var_12866_equation_0 = const()[name = tensor("op_12866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12866_cast_fp16 = einsum(equation = var_12866_equation_0, values = (var_12472_cast_fp16, var_12384_cast_fp16))[name = tensor("op_12866_cast_fp16")]; + tensor var_12867_to_fp16 = const()[name = tensor("op_12867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1277_cast_fp16 = mul(x = var_12866_cast_fp16, y = var_12867_to_fp16)[name = tensor("aw_chunk_1277_cast_fp16")]; + tensor var_12870_equation_0 = const()[name = tensor("op_12870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_12870_cast_fp16 = einsum(equation = var_12870_equation_0, values = (var_12472_cast_fp16, var_12391_cast_fp16))[name = tensor("op_12870_cast_fp16")]; + tensor var_12871_to_fp16 = const()[name = tensor("op_12871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1279_cast_fp16 = mul(x = var_12870_cast_fp16, y = var_12871_to_fp16)[name = tensor("aw_chunk_1279_cast_fp16")]; + tensor var_12873_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1121_cast_fp16)[name = tensor("op_12873_cast_fp16")]; + tensor var_12874_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1123_cast_fp16)[name = tensor("op_12874_cast_fp16")]; + tensor var_12875_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1125_cast_fp16)[name = tensor("op_12875_cast_fp16")]; + tensor var_12876_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1127_cast_fp16)[name = tensor("op_12876_cast_fp16")]; + tensor var_12877_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1129_cast_fp16)[name = tensor("op_12877_cast_fp16")]; + tensor var_12878_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1131_cast_fp16)[name = tensor("op_12878_cast_fp16")]; + tensor var_12879_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1133_cast_fp16)[name = tensor("op_12879_cast_fp16")]; + tensor var_12880_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1135_cast_fp16)[name = tensor("op_12880_cast_fp16")]; + tensor var_12881_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1137_cast_fp16)[name = tensor("op_12881_cast_fp16")]; + tensor var_12882_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1139_cast_fp16)[name = tensor("op_12882_cast_fp16")]; + tensor var_12883_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1141_cast_fp16)[name = tensor("op_12883_cast_fp16")]; + tensor var_12884_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1143_cast_fp16)[name = tensor("op_12884_cast_fp16")]; + tensor var_12885_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1145_cast_fp16)[name = tensor("op_12885_cast_fp16")]; + tensor var_12886_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1147_cast_fp16)[name = tensor("op_12886_cast_fp16")]; + tensor var_12887_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1149_cast_fp16)[name = tensor("op_12887_cast_fp16")]; + tensor var_12888_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1151_cast_fp16)[name = tensor("op_12888_cast_fp16")]; + tensor var_12889_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1153_cast_fp16)[name = tensor("op_12889_cast_fp16")]; + tensor var_12890_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1155_cast_fp16)[name = tensor("op_12890_cast_fp16")]; + tensor var_12891_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1157_cast_fp16)[name = tensor("op_12891_cast_fp16")]; + tensor var_12892_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1159_cast_fp16)[name = tensor("op_12892_cast_fp16")]; + tensor var_12893_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1161_cast_fp16)[name = tensor("op_12893_cast_fp16")]; + tensor var_12894_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1163_cast_fp16)[name = tensor("op_12894_cast_fp16")]; + tensor var_12895_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1165_cast_fp16)[name = tensor("op_12895_cast_fp16")]; + tensor var_12896_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1167_cast_fp16)[name = tensor("op_12896_cast_fp16")]; + tensor var_12897_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1169_cast_fp16)[name = tensor("op_12897_cast_fp16")]; + tensor var_12898_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1171_cast_fp16)[name = tensor("op_12898_cast_fp16")]; + tensor var_12899_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1173_cast_fp16)[name = tensor("op_12899_cast_fp16")]; + tensor var_12900_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1175_cast_fp16)[name = tensor("op_12900_cast_fp16")]; + tensor var_12901_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1177_cast_fp16)[name = tensor("op_12901_cast_fp16")]; + tensor var_12902_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1179_cast_fp16)[name = tensor("op_12902_cast_fp16")]; + tensor var_12903_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1181_cast_fp16)[name = tensor("op_12903_cast_fp16")]; + tensor var_12904_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1183_cast_fp16)[name = tensor("op_12904_cast_fp16")]; + tensor var_12905_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1185_cast_fp16)[name = tensor("op_12905_cast_fp16")]; + tensor var_12906_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1187_cast_fp16)[name = tensor("op_12906_cast_fp16")]; + tensor var_12907_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1189_cast_fp16)[name = tensor("op_12907_cast_fp16")]; + tensor var_12908_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1191_cast_fp16)[name = tensor("op_12908_cast_fp16")]; + tensor var_12909_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1193_cast_fp16)[name = tensor("op_12909_cast_fp16")]; + tensor var_12910_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1195_cast_fp16)[name = tensor("op_12910_cast_fp16")]; + tensor var_12911_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1197_cast_fp16)[name = tensor("op_12911_cast_fp16")]; + tensor var_12912_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1199_cast_fp16)[name = tensor("op_12912_cast_fp16")]; + tensor var_12913_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1201_cast_fp16)[name = tensor("op_12913_cast_fp16")]; + tensor var_12914_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1203_cast_fp16)[name = tensor("op_12914_cast_fp16")]; + tensor var_12915_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1205_cast_fp16)[name = tensor("op_12915_cast_fp16")]; + tensor var_12916_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1207_cast_fp16)[name = tensor("op_12916_cast_fp16")]; + tensor var_12917_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1209_cast_fp16)[name = tensor("op_12917_cast_fp16")]; + tensor var_12918_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1211_cast_fp16)[name = tensor("op_12918_cast_fp16")]; + tensor var_12919_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1213_cast_fp16)[name = tensor("op_12919_cast_fp16")]; + tensor var_12920_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1215_cast_fp16)[name = tensor("op_12920_cast_fp16")]; + tensor var_12921_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1217_cast_fp16)[name = tensor("op_12921_cast_fp16")]; + tensor var_12922_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1219_cast_fp16)[name = tensor("op_12922_cast_fp16")]; + tensor var_12923_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1221_cast_fp16)[name = tensor("op_12923_cast_fp16")]; + tensor var_12924_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1223_cast_fp16)[name = tensor("op_12924_cast_fp16")]; + tensor var_12925_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1225_cast_fp16)[name = tensor("op_12925_cast_fp16")]; + tensor var_12926_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1227_cast_fp16)[name = tensor("op_12926_cast_fp16")]; + tensor var_12927_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1229_cast_fp16)[name = tensor("op_12927_cast_fp16")]; + tensor var_12928_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1231_cast_fp16)[name = tensor("op_12928_cast_fp16")]; + tensor var_12929_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1233_cast_fp16)[name = tensor("op_12929_cast_fp16")]; + tensor var_12930_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1235_cast_fp16)[name = tensor("op_12930_cast_fp16")]; + tensor var_12931_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1237_cast_fp16)[name = tensor("op_12931_cast_fp16")]; + tensor var_12932_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1239_cast_fp16)[name = tensor("op_12932_cast_fp16")]; + tensor var_12933_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1241_cast_fp16)[name = tensor("op_12933_cast_fp16")]; + tensor var_12934_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1243_cast_fp16)[name = tensor("op_12934_cast_fp16")]; + tensor var_12935_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1245_cast_fp16)[name = tensor("op_12935_cast_fp16")]; + tensor var_12936_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1247_cast_fp16)[name = tensor("op_12936_cast_fp16")]; + tensor var_12937_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1249_cast_fp16)[name = tensor("op_12937_cast_fp16")]; + tensor var_12938_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1251_cast_fp16)[name = tensor("op_12938_cast_fp16")]; + tensor var_12939_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1253_cast_fp16)[name = tensor("op_12939_cast_fp16")]; + tensor var_12940_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1255_cast_fp16)[name = tensor("op_12940_cast_fp16")]; + tensor var_12941_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1257_cast_fp16)[name = tensor("op_12941_cast_fp16")]; + tensor var_12942_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1259_cast_fp16)[name = tensor("op_12942_cast_fp16")]; + tensor var_12943_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1261_cast_fp16)[name = tensor("op_12943_cast_fp16")]; + tensor var_12944_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1263_cast_fp16)[name = tensor("op_12944_cast_fp16")]; + tensor var_12945_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1265_cast_fp16)[name = tensor("op_12945_cast_fp16")]; + tensor var_12946_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1267_cast_fp16)[name = tensor("op_12946_cast_fp16")]; + tensor var_12947_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1269_cast_fp16)[name = tensor("op_12947_cast_fp16")]; + tensor var_12948_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1271_cast_fp16)[name = tensor("op_12948_cast_fp16")]; + tensor var_12949_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1273_cast_fp16)[name = tensor("op_12949_cast_fp16")]; + tensor var_12950_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1275_cast_fp16)[name = tensor("op_12950_cast_fp16")]; + tensor var_12951_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1277_cast_fp16)[name = tensor("op_12951_cast_fp16")]; + tensor var_12952_cast_fp16 = softmax(axis = var_11644, x = aw_chunk_1279_cast_fp16)[name = tensor("op_12952_cast_fp16")]; + tensor var_12954_equation_0 = const()[name = tensor("op_12954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12954_cast_fp16 = einsum(equation = var_12954_equation_0, values = (var_12474_cast_fp16, var_12873_cast_fp16))[name = tensor("op_12954_cast_fp16")]; + tensor var_12956_equation_0 = const()[name = tensor("op_12956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12956_cast_fp16 = einsum(equation = var_12956_equation_0, values = (var_12474_cast_fp16, var_12874_cast_fp16))[name = tensor("op_12956_cast_fp16")]; + tensor var_12958_equation_0 = const()[name = tensor("op_12958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12958_cast_fp16 = einsum(equation = var_12958_equation_0, values = (var_12474_cast_fp16, var_12875_cast_fp16))[name = tensor("op_12958_cast_fp16")]; + tensor var_12960_equation_0 = const()[name = tensor("op_12960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12960_cast_fp16 = einsum(equation = var_12960_equation_0, values = (var_12474_cast_fp16, var_12876_cast_fp16))[name = tensor("op_12960_cast_fp16")]; + tensor var_12962_equation_0 = const()[name = tensor("op_12962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12962_cast_fp16 = einsum(equation = var_12962_equation_0, values = (var_12478_cast_fp16, var_12877_cast_fp16))[name = tensor("op_12962_cast_fp16")]; + tensor var_12964_equation_0 = const()[name = tensor("op_12964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12964_cast_fp16 = einsum(equation = var_12964_equation_0, values = (var_12478_cast_fp16, var_12878_cast_fp16))[name = tensor("op_12964_cast_fp16")]; + tensor var_12966_equation_0 = const()[name = tensor("op_12966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12966_cast_fp16 = einsum(equation = var_12966_equation_0, values = (var_12478_cast_fp16, var_12879_cast_fp16))[name = tensor("op_12966_cast_fp16")]; + tensor var_12968_equation_0 = const()[name = tensor("op_12968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12968_cast_fp16 = einsum(equation = var_12968_equation_0, values = (var_12478_cast_fp16, var_12880_cast_fp16))[name = tensor("op_12968_cast_fp16")]; + tensor var_12970_equation_0 = const()[name = tensor("op_12970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12970_cast_fp16 = einsum(equation = var_12970_equation_0, values = (var_12482_cast_fp16, var_12881_cast_fp16))[name = tensor("op_12970_cast_fp16")]; + tensor var_12972_equation_0 = const()[name = tensor("op_12972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12972_cast_fp16 = einsum(equation = var_12972_equation_0, values = (var_12482_cast_fp16, var_12882_cast_fp16))[name = tensor("op_12972_cast_fp16")]; + tensor var_12974_equation_0 = const()[name = tensor("op_12974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12974_cast_fp16 = einsum(equation = var_12974_equation_0, values = (var_12482_cast_fp16, var_12883_cast_fp16))[name = tensor("op_12974_cast_fp16")]; + tensor var_12976_equation_0 = const()[name = tensor("op_12976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12976_cast_fp16 = einsum(equation = var_12976_equation_0, values = (var_12482_cast_fp16, var_12884_cast_fp16))[name = tensor("op_12976_cast_fp16")]; + tensor var_12978_equation_0 = const()[name = tensor("op_12978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12978_cast_fp16 = einsum(equation = var_12978_equation_0, values = (var_12486_cast_fp16, var_12885_cast_fp16))[name = tensor("op_12978_cast_fp16")]; + tensor var_12980_equation_0 = const()[name = tensor("op_12980_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12980_cast_fp16 = einsum(equation = var_12980_equation_0, values = (var_12486_cast_fp16, var_12886_cast_fp16))[name = tensor("op_12980_cast_fp16")]; + tensor var_12982_equation_0 = const()[name = tensor("op_12982_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12982_cast_fp16 = einsum(equation = var_12982_equation_0, values = (var_12486_cast_fp16, var_12887_cast_fp16))[name = tensor("op_12982_cast_fp16")]; + tensor var_12984_equation_0 = const()[name = tensor("op_12984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12984_cast_fp16 = einsum(equation = var_12984_equation_0, values = (var_12486_cast_fp16, var_12888_cast_fp16))[name = tensor("op_12984_cast_fp16")]; + tensor var_12986_equation_0 = const()[name = tensor("op_12986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12986_cast_fp16 = einsum(equation = var_12986_equation_0, values = (var_12490_cast_fp16, var_12889_cast_fp16))[name = tensor("op_12986_cast_fp16")]; + tensor var_12988_equation_0 = const()[name = tensor("op_12988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12988_cast_fp16 = einsum(equation = var_12988_equation_0, values = (var_12490_cast_fp16, var_12890_cast_fp16))[name = tensor("op_12988_cast_fp16")]; + tensor var_12990_equation_0 = const()[name = tensor("op_12990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12990_cast_fp16 = einsum(equation = var_12990_equation_0, values = (var_12490_cast_fp16, var_12891_cast_fp16))[name = tensor("op_12990_cast_fp16")]; + tensor var_12992_equation_0 = const()[name = tensor("op_12992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12992_cast_fp16 = einsum(equation = var_12992_equation_0, values = (var_12490_cast_fp16, var_12892_cast_fp16))[name = tensor("op_12992_cast_fp16")]; + tensor var_12994_equation_0 = const()[name = tensor("op_12994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12994_cast_fp16 = einsum(equation = var_12994_equation_0, values = (var_12494_cast_fp16, var_12893_cast_fp16))[name = tensor("op_12994_cast_fp16")]; + tensor var_12996_equation_0 = const()[name = tensor("op_12996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12996_cast_fp16 = einsum(equation = var_12996_equation_0, values = (var_12494_cast_fp16, var_12894_cast_fp16))[name = tensor("op_12996_cast_fp16")]; + tensor var_12998_equation_0 = const()[name = tensor("op_12998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_12998_cast_fp16 = einsum(equation = var_12998_equation_0, values = (var_12494_cast_fp16, var_12895_cast_fp16))[name = tensor("op_12998_cast_fp16")]; + tensor var_13000_equation_0 = const()[name = tensor("op_13000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13000_cast_fp16 = einsum(equation = var_13000_equation_0, values = (var_12494_cast_fp16, var_12896_cast_fp16))[name = tensor("op_13000_cast_fp16")]; + tensor var_13002_equation_0 = const()[name = tensor("op_13002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13002_cast_fp16 = einsum(equation = var_13002_equation_0, values = (var_12498_cast_fp16, var_12897_cast_fp16))[name = tensor("op_13002_cast_fp16")]; + tensor var_13004_equation_0 = const()[name = tensor("op_13004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13004_cast_fp16 = einsum(equation = var_13004_equation_0, values = (var_12498_cast_fp16, var_12898_cast_fp16))[name = tensor("op_13004_cast_fp16")]; + tensor var_13006_equation_0 = const()[name = tensor("op_13006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13006_cast_fp16 = einsum(equation = var_13006_equation_0, values = (var_12498_cast_fp16, var_12899_cast_fp16))[name = tensor("op_13006_cast_fp16")]; + tensor var_13008_equation_0 = const()[name = tensor("op_13008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13008_cast_fp16 = einsum(equation = var_13008_equation_0, values = (var_12498_cast_fp16, var_12900_cast_fp16))[name = tensor("op_13008_cast_fp16")]; + tensor var_13010_equation_0 = const()[name = tensor("op_13010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13010_cast_fp16 = einsum(equation = var_13010_equation_0, values = (var_12502_cast_fp16, var_12901_cast_fp16))[name = tensor("op_13010_cast_fp16")]; + tensor var_13012_equation_0 = const()[name = tensor("op_13012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13012_cast_fp16 = einsum(equation = var_13012_equation_0, values = (var_12502_cast_fp16, var_12902_cast_fp16))[name = tensor("op_13012_cast_fp16")]; + tensor var_13014_equation_0 = const()[name = tensor("op_13014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13014_cast_fp16 = einsum(equation = var_13014_equation_0, values = (var_12502_cast_fp16, var_12903_cast_fp16))[name = tensor("op_13014_cast_fp16")]; + tensor var_13016_equation_0 = const()[name = tensor("op_13016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13016_cast_fp16 = einsum(equation = var_13016_equation_0, values = (var_12502_cast_fp16, var_12904_cast_fp16))[name = tensor("op_13016_cast_fp16")]; + tensor var_13018_equation_0 = const()[name = tensor("op_13018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13018_cast_fp16 = einsum(equation = var_13018_equation_0, values = (var_12506_cast_fp16, var_12905_cast_fp16))[name = tensor("op_13018_cast_fp16")]; + tensor var_13020_equation_0 = const()[name = tensor("op_13020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13020_cast_fp16 = einsum(equation = var_13020_equation_0, values = (var_12506_cast_fp16, var_12906_cast_fp16))[name = tensor("op_13020_cast_fp16")]; + tensor var_13022_equation_0 = const()[name = tensor("op_13022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13022_cast_fp16 = einsum(equation = var_13022_equation_0, values = (var_12506_cast_fp16, var_12907_cast_fp16))[name = tensor("op_13022_cast_fp16")]; + tensor var_13024_equation_0 = const()[name = tensor("op_13024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13024_cast_fp16 = einsum(equation = var_13024_equation_0, values = (var_12506_cast_fp16, var_12908_cast_fp16))[name = tensor("op_13024_cast_fp16")]; + tensor var_13026_equation_0 = const()[name = tensor("op_13026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13026_cast_fp16 = einsum(equation = var_13026_equation_0, values = (var_12510_cast_fp16, var_12909_cast_fp16))[name = tensor("op_13026_cast_fp16")]; + tensor var_13028_equation_0 = const()[name = tensor("op_13028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13028_cast_fp16 = einsum(equation = var_13028_equation_0, values = (var_12510_cast_fp16, var_12910_cast_fp16))[name = tensor("op_13028_cast_fp16")]; + tensor var_13030_equation_0 = const()[name = tensor("op_13030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13030_cast_fp16 = einsum(equation = var_13030_equation_0, values = (var_12510_cast_fp16, var_12911_cast_fp16))[name = tensor("op_13030_cast_fp16")]; + tensor var_13032_equation_0 = const()[name = tensor("op_13032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13032_cast_fp16 = einsum(equation = var_13032_equation_0, values = (var_12510_cast_fp16, var_12912_cast_fp16))[name = tensor("op_13032_cast_fp16")]; + tensor var_13034_equation_0 = const()[name = tensor("op_13034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13034_cast_fp16 = einsum(equation = var_13034_equation_0, values = (var_12514_cast_fp16, var_12913_cast_fp16))[name = tensor("op_13034_cast_fp16")]; + tensor var_13036_equation_0 = const()[name = tensor("op_13036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13036_cast_fp16 = einsum(equation = var_13036_equation_0, values = (var_12514_cast_fp16, var_12914_cast_fp16))[name = tensor("op_13036_cast_fp16")]; + tensor var_13038_equation_0 = const()[name = tensor("op_13038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13038_cast_fp16 = einsum(equation = var_13038_equation_0, values = (var_12514_cast_fp16, var_12915_cast_fp16))[name = tensor("op_13038_cast_fp16")]; + tensor var_13040_equation_0 = const()[name = tensor("op_13040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13040_cast_fp16 = einsum(equation = var_13040_equation_0, values = (var_12514_cast_fp16, var_12916_cast_fp16))[name = tensor("op_13040_cast_fp16")]; + tensor var_13042_equation_0 = const()[name = tensor("op_13042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13042_cast_fp16 = einsum(equation = var_13042_equation_0, values = (var_12518_cast_fp16, var_12917_cast_fp16))[name = tensor("op_13042_cast_fp16")]; + tensor var_13044_equation_0 = const()[name = tensor("op_13044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13044_cast_fp16 = einsum(equation = var_13044_equation_0, values = (var_12518_cast_fp16, var_12918_cast_fp16))[name = tensor("op_13044_cast_fp16")]; + tensor var_13046_equation_0 = const()[name = tensor("op_13046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13046_cast_fp16 = einsum(equation = var_13046_equation_0, values = (var_12518_cast_fp16, var_12919_cast_fp16))[name = tensor("op_13046_cast_fp16")]; + tensor var_13048_equation_0 = const()[name = tensor("op_13048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13048_cast_fp16 = einsum(equation = var_13048_equation_0, values = (var_12518_cast_fp16, var_12920_cast_fp16))[name = tensor("op_13048_cast_fp16")]; + tensor var_13050_equation_0 = const()[name = tensor("op_13050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13050_cast_fp16 = einsum(equation = var_13050_equation_0, values = (var_12522_cast_fp16, var_12921_cast_fp16))[name = tensor("op_13050_cast_fp16")]; + tensor var_13052_equation_0 = const()[name = tensor("op_13052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13052_cast_fp16 = einsum(equation = var_13052_equation_0, values = (var_12522_cast_fp16, var_12922_cast_fp16))[name = tensor("op_13052_cast_fp16")]; + tensor var_13054_equation_0 = const()[name = tensor("op_13054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13054_cast_fp16 = einsum(equation = var_13054_equation_0, values = (var_12522_cast_fp16, var_12923_cast_fp16))[name = tensor("op_13054_cast_fp16")]; + tensor var_13056_equation_0 = const()[name = tensor("op_13056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13056_cast_fp16 = einsum(equation = var_13056_equation_0, values = (var_12522_cast_fp16, var_12924_cast_fp16))[name = tensor("op_13056_cast_fp16")]; + tensor var_13058_equation_0 = const()[name = tensor("op_13058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13058_cast_fp16 = einsum(equation = var_13058_equation_0, values = (var_12526_cast_fp16, var_12925_cast_fp16))[name = tensor("op_13058_cast_fp16")]; + tensor var_13060_equation_0 = const()[name = tensor("op_13060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13060_cast_fp16 = einsum(equation = var_13060_equation_0, values = (var_12526_cast_fp16, var_12926_cast_fp16))[name = tensor("op_13060_cast_fp16")]; + tensor var_13062_equation_0 = const()[name = tensor("op_13062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13062_cast_fp16 = einsum(equation = var_13062_equation_0, values = (var_12526_cast_fp16, var_12927_cast_fp16))[name = tensor("op_13062_cast_fp16")]; + tensor var_13064_equation_0 = const()[name = tensor("op_13064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13064_cast_fp16 = einsum(equation = var_13064_equation_0, values = (var_12526_cast_fp16, var_12928_cast_fp16))[name = tensor("op_13064_cast_fp16")]; + tensor var_13066_equation_0 = const()[name = tensor("op_13066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13066_cast_fp16 = einsum(equation = var_13066_equation_0, values = (var_12530_cast_fp16, var_12929_cast_fp16))[name = tensor("op_13066_cast_fp16")]; + tensor var_13068_equation_0 = const()[name = tensor("op_13068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13068_cast_fp16 = einsum(equation = var_13068_equation_0, values = (var_12530_cast_fp16, var_12930_cast_fp16))[name = tensor("op_13068_cast_fp16")]; + tensor var_13070_equation_0 = const()[name = tensor("op_13070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13070_cast_fp16 = einsum(equation = var_13070_equation_0, values = (var_12530_cast_fp16, var_12931_cast_fp16))[name = tensor("op_13070_cast_fp16")]; + tensor var_13072_equation_0 = const()[name = tensor("op_13072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13072_cast_fp16 = einsum(equation = var_13072_equation_0, values = (var_12530_cast_fp16, var_12932_cast_fp16))[name = tensor("op_13072_cast_fp16")]; + tensor var_13074_equation_0 = const()[name = tensor("op_13074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13074_cast_fp16 = einsum(equation = var_13074_equation_0, values = (var_12534_cast_fp16, var_12933_cast_fp16))[name = tensor("op_13074_cast_fp16")]; + tensor var_13076_equation_0 = const()[name = tensor("op_13076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13076_cast_fp16 = einsum(equation = var_13076_equation_0, values = (var_12534_cast_fp16, var_12934_cast_fp16))[name = tensor("op_13076_cast_fp16")]; + tensor var_13078_equation_0 = const()[name = tensor("op_13078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13078_cast_fp16 = einsum(equation = var_13078_equation_0, values = (var_12534_cast_fp16, var_12935_cast_fp16))[name = tensor("op_13078_cast_fp16")]; + tensor var_13080_equation_0 = const()[name = tensor("op_13080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13080_cast_fp16 = einsum(equation = var_13080_equation_0, values = (var_12534_cast_fp16, var_12936_cast_fp16))[name = tensor("op_13080_cast_fp16")]; + tensor var_13082_equation_0 = const()[name = tensor("op_13082_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13082_cast_fp16 = einsum(equation = var_13082_equation_0, values = (var_12538_cast_fp16, var_12937_cast_fp16))[name = tensor("op_13082_cast_fp16")]; + tensor var_13084_equation_0 = const()[name = tensor("op_13084_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13084_cast_fp16 = einsum(equation = var_13084_equation_0, values = (var_12538_cast_fp16, var_12938_cast_fp16))[name = tensor("op_13084_cast_fp16")]; + tensor var_13086_equation_0 = const()[name = tensor("op_13086_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13086_cast_fp16 = einsum(equation = var_13086_equation_0, values = (var_12538_cast_fp16, var_12939_cast_fp16))[name = tensor("op_13086_cast_fp16")]; + tensor var_13088_equation_0 = const()[name = tensor("op_13088_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13088_cast_fp16 = einsum(equation = var_13088_equation_0, values = (var_12538_cast_fp16, var_12940_cast_fp16))[name = tensor("op_13088_cast_fp16")]; + tensor var_13090_equation_0 = const()[name = tensor("op_13090_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13090_cast_fp16 = einsum(equation = var_13090_equation_0, values = (var_12542_cast_fp16, var_12941_cast_fp16))[name = tensor("op_13090_cast_fp16")]; + tensor var_13092_equation_0 = const()[name = tensor("op_13092_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13092_cast_fp16 = einsum(equation = var_13092_equation_0, values = (var_12542_cast_fp16, var_12942_cast_fp16))[name = tensor("op_13092_cast_fp16")]; + tensor var_13094_equation_0 = const()[name = tensor("op_13094_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13094_cast_fp16 = einsum(equation = var_13094_equation_0, values = (var_12542_cast_fp16, var_12943_cast_fp16))[name = tensor("op_13094_cast_fp16")]; + tensor var_13096_equation_0 = const()[name = tensor("op_13096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13096_cast_fp16 = einsum(equation = var_13096_equation_0, values = (var_12542_cast_fp16, var_12944_cast_fp16))[name = tensor("op_13096_cast_fp16")]; + tensor var_13098_equation_0 = const()[name = tensor("op_13098_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13098_cast_fp16 = einsum(equation = var_13098_equation_0, values = (var_12546_cast_fp16, var_12945_cast_fp16))[name = tensor("op_13098_cast_fp16")]; + tensor var_13100_equation_0 = const()[name = tensor("op_13100_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13100_cast_fp16 = einsum(equation = var_13100_equation_0, values = (var_12546_cast_fp16, var_12946_cast_fp16))[name = tensor("op_13100_cast_fp16")]; + tensor var_13102_equation_0 = const()[name = tensor("op_13102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13102_cast_fp16 = einsum(equation = var_13102_equation_0, values = (var_12546_cast_fp16, var_12947_cast_fp16))[name = tensor("op_13102_cast_fp16")]; + tensor var_13104_equation_0 = const()[name = tensor("op_13104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13104_cast_fp16 = einsum(equation = var_13104_equation_0, values = (var_12546_cast_fp16, var_12948_cast_fp16))[name = tensor("op_13104_cast_fp16")]; + tensor var_13106_equation_0 = const()[name = tensor("op_13106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13106_cast_fp16 = einsum(equation = var_13106_equation_0, values = (var_12550_cast_fp16, var_12949_cast_fp16))[name = tensor("op_13106_cast_fp16")]; + tensor var_13108_equation_0 = const()[name = tensor("op_13108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13108_cast_fp16 = einsum(equation = var_13108_equation_0, values = (var_12550_cast_fp16, var_12950_cast_fp16))[name = tensor("op_13108_cast_fp16")]; + tensor var_13110_equation_0 = const()[name = tensor("op_13110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13110_cast_fp16 = einsum(equation = var_13110_equation_0, values = (var_12550_cast_fp16, var_12951_cast_fp16))[name = tensor("op_13110_cast_fp16")]; + tensor var_13112_equation_0 = const()[name = tensor("op_13112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_13112_cast_fp16 = einsum(equation = var_13112_equation_0, values = (var_12550_cast_fp16, var_12952_cast_fp16))[name = tensor("op_13112_cast_fp16")]; + tensor var_13114_interleave_0 = const()[name = tensor("op_13114_interleave_0"), val = tensor(false)]; + tensor var_13114_cast_fp16 = concat(axis = var_11619, interleave = var_13114_interleave_0, values = (var_12954_cast_fp16, var_12956_cast_fp16, var_12958_cast_fp16, var_12960_cast_fp16))[name = tensor("op_13114_cast_fp16")]; + tensor var_13116_interleave_0 = const()[name = tensor("op_13116_interleave_0"), val = tensor(false)]; + tensor var_13116_cast_fp16 = concat(axis = var_11619, interleave = var_13116_interleave_0, values = (var_12962_cast_fp16, var_12964_cast_fp16, var_12966_cast_fp16, var_12968_cast_fp16))[name = tensor("op_13116_cast_fp16")]; + tensor var_13118_interleave_0 = const()[name = tensor("op_13118_interleave_0"), val = tensor(false)]; + tensor var_13118_cast_fp16 = concat(axis = var_11619, interleave = var_13118_interleave_0, values = (var_12970_cast_fp16, var_12972_cast_fp16, var_12974_cast_fp16, var_12976_cast_fp16))[name = tensor("op_13118_cast_fp16")]; + tensor var_13120_interleave_0 = const()[name = tensor("op_13120_interleave_0"), val = tensor(false)]; + tensor var_13120_cast_fp16 = concat(axis = var_11619, interleave = var_13120_interleave_0, values = (var_12978_cast_fp16, var_12980_cast_fp16, var_12982_cast_fp16, var_12984_cast_fp16))[name = tensor("op_13120_cast_fp16")]; + tensor var_13122_interleave_0 = const()[name = tensor("op_13122_interleave_0"), val = tensor(false)]; + tensor var_13122_cast_fp16 = concat(axis = var_11619, interleave = var_13122_interleave_0, values = (var_12986_cast_fp16, var_12988_cast_fp16, var_12990_cast_fp16, var_12992_cast_fp16))[name = tensor("op_13122_cast_fp16")]; + tensor var_13124_interleave_0 = const()[name = tensor("op_13124_interleave_0"), val = tensor(false)]; + tensor var_13124_cast_fp16 = concat(axis = var_11619, interleave = var_13124_interleave_0, values = (var_12994_cast_fp16, var_12996_cast_fp16, var_12998_cast_fp16, var_13000_cast_fp16))[name = tensor("op_13124_cast_fp16")]; + tensor var_13126_interleave_0 = const()[name = tensor("op_13126_interleave_0"), val = tensor(false)]; + tensor var_13126_cast_fp16 = concat(axis = var_11619, interleave = var_13126_interleave_0, values = (var_13002_cast_fp16, var_13004_cast_fp16, var_13006_cast_fp16, var_13008_cast_fp16))[name = tensor("op_13126_cast_fp16")]; + tensor var_13128_interleave_0 = const()[name = tensor("op_13128_interleave_0"), val = tensor(false)]; + tensor var_13128_cast_fp16 = concat(axis = var_11619, interleave = var_13128_interleave_0, values = (var_13010_cast_fp16, var_13012_cast_fp16, var_13014_cast_fp16, var_13016_cast_fp16))[name = tensor("op_13128_cast_fp16")]; + tensor var_13130_interleave_0 = const()[name = tensor("op_13130_interleave_0"), val = tensor(false)]; + tensor var_13130_cast_fp16 = concat(axis = var_11619, interleave = var_13130_interleave_0, values = (var_13018_cast_fp16, var_13020_cast_fp16, var_13022_cast_fp16, var_13024_cast_fp16))[name = tensor("op_13130_cast_fp16")]; + tensor var_13132_interleave_0 = const()[name = tensor("op_13132_interleave_0"), val = tensor(false)]; + tensor var_13132_cast_fp16 = concat(axis = var_11619, interleave = var_13132_interleave_0, values = (var_13026_cast_fp16, var_13028_cast_fp16, var_13030_cast_fp16, var_13032_cast_fp16))[name = tensor("op_13132_cast_fp16")]; + tensor var_13134_interleave_0 = const()[name = tensor("op_13134_interleave_0"), val = tensor(false)]; + tensor var_13134_cast_fp16 = concat(axis = var_11619, interleave = var_13134_interleave_0, values = (var_13034_cast_fp16, var_13036_cast_fp16, var_13038_cast_fp16, var_13040_cast_fp16))[name = tensor("op_13134_cast_fp16")]; + tensor var_13136_interleave_0 = const()[name = tensor("op_13136_interleave_0"), val = tensor(false)]; + tensor var_13136_cast_fp16 = concat(axis = var_11619, interleave = var_13136_interleave_0, values = (var_13042_cast_fp16, var_13044_cast_fp16, var_13046_cast_fp16, var_13048_cast_fp16))[name = tensor("op_13136_cast_fp16")]; + tensor var_13138_interleave_0 = const()[name = tensor("op_13138_interleave_0"), val = tensor(false)]; + tensor var_13138_cast_fp16 = concat(axis = var_11619, interleave = var_13138_interleave_0, values = (var_13050_cast_fp16, var_13052_cast_fp16, var_13054_cast_fp16, var_13056_cast_fp16))[name = tensor("op_13138_cast_fp16")]; + tensor var_13140_interleave_0 = const()[name = tensor("op_13140_interleave_0"), val = tensor(false)]; + tensor var_13140_cast_fp16 = concat(axis = var_11619, interleave = var_13140_interleave_0, values = (var_13058_cast_fp16, var_13060_cast_fp16, var_13062_cast_fp16, var_13064_cast_fp16))[name = tensor("op_13140_cast_fp16")]; + tensor var_13142_interleave_0 = const()[name = tensor("op_13142_interleave_0"), val = tensor(false)]; + tensor var_13142_cast_fp16 = concat(axis = var_11619, interleave = var_13142_interleave_0, values = (var_13066_cast_fp16, var_13068_cast_fp16, var_13070_cast_fp16, var_13072_cast_fp16))[name = tensor("op_13142_cast_fp16")]; + tensor var_13144_interleave_0 = const()[name = tensor("op_13144_interleave_0"), val = tensor(false)]; + tensor var_13144_cast_fp16 = concat(axis = var_11619, interleave = var_13144_interleave_0, values = (var_13074_cast_fp16, var_13076_cast_fp16, var_13078_cast_fp16, var_13080_cast_fp16))[name = tensor("op_13144_cast_fp16")]; + tensor var_13146_interleave_0 = const()[name = tensor("op_13146_interleave_0"), val = tensor(false)]; + tensor var_13146_cast_fp16 = concat(axis = var_11619, interleave = var_13146_interleave_0, values = (var_13082_cast_fp16, var_13084_cast_fp16, var_13086_cast_fp16, var_13088_cast_fp16))[name = tensor("op_13146_cast_fp16")]; + tensor var_13148_interleave_0 = const()[name = tensor("op_13148_interleave_0"), val = tensor(false)]; + tensor var_13148_cast_fp16 = concat(axis = var_11619, interleave = var_13148_interleave_0, values = (var_13090_cast_fp16, var_13092_cast_fp16, var_13094_cast_fp16, var_13096_cast_fp16))[name = tensor("op_13148_cast_fp16")]; + tensor var_13150_interleave_0 = const()[name = tensor("op_13150_interleave_0"), val = tensor(false)]; + tensor var_13150_cast_fp16 = concat(axis = var_11619, interleave = var_13150_interleave_0, values = (var_13098_cast_fp16, var_13100_cast_fp16, var_13102_cast_fp16, var_13104_cast_fp16))[name = tensor("op_13150_cast_fp16")]; + tensor var_13152_interleave_0 = const()[name = tensor("op_13152_interleave_0"), val = tensor(false)]; + tensor var_13152_cast_fp16 = concat(axis = var_11619, interleave = var_13152_interleave_0, values = (var_13106_cast_fp16, var_13108_cast_fp16, var_13110_cast_fp16, var_13112_cast_fp16))[name = tensor("op_13152_cast_fp16")]; + tensor input_147_interleave_0 = const()[name = tensor("input_147_interleave_0"), val = tensor(false)]; + tensor input_147_cast_fp16 = concat(axis = var_11644, interleave = input_147_interleave_0, values = (var_13114_cast_fp16, var_13116_cast_fp16, var_13118_cast_fp16, var_13120_cast_fp16, var_13122_cast_fp16, var_13124_cast_fp16, var_13126_cast_fp16, var_13128_cast_fp16, var_13130_cast_fp16, var_13132_cast_fp16, var_13134_cast_fp16, var_13136_cast_fp16, var_13138_cast_fp16, var_13140_cast_fp16, var_13142_cast_fp16, var_13144_cast_fp16, var_13146_cast_fp16, var_13148_cast_fp16, var_13150_cast_fp16, var_13152_cast_fp16))[name = tensor("input_147_cast_fp16")]; + tensor var_13160 = const()[name = tensor("op_13160"), val = tensor([1, 1])]; + tensor var_13162 = const()[name = tensor("op_13162"), val = tensor([1, 1])]; + tensor pretrained_out_91_pad_type_0 = const()[name = tensor("pretrained_out_91_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_91_pad_0 = const()[name = tensor("pretrained_out_91_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91226944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92046208))), name = tensor("layers_7_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_7_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92046336)))]; + tensor pretrained_out_91_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_13162, groups = var_11644, pad = pretrained_out_91_pad_0, pad_type = pretrained_out_91_pad_type_0, strides = var_13160, weight = layers_7_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor("pretrained_out_91_cast_fp16")]; + tensor var_13166 = const()[name = tensor("op_13166"), val = tensor([1, 1])]; + tensor var_13168 = const()[name = tensor("op_13168"), val = tensor([1, 1])]; + tensor input_149_pad_type_0 = const()[name = tensor("input_149_pad_type_0"), val = tensor("custom")]; + tensor input_149_pad_0 = const()[name = tensor("input_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92048960)))]; + tensor input_149_cast_fp16 = conv(dilations = var_13168, groups = var_11644, pad = input_149_pad_0, pad_type = input_149_pad_type_0, strides = var_13166, weight = layers_7_self_attn_o_proj_loraA_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor var_13172 = const()[name = tensor("op_13172"), val = tensor([1, 1])]; + tensor var_13174 = const()[name = tensor("op_13174"), val = tensor([1, 1])]; + tensor lora_out_181_pad_type_0 = const()[name = tensor("lora_out_181_pad_type_0"), val = tensor("custom")]; + tensor lora_out_181_pad_0 = const()[name = tensor("lora_out_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_183_weight_0_to_fp16 = const()[name = tensor("lora_out_183_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92089984)))]; + tensor lora_out_183_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_13174, groups = var_11644, pad = lora_out_181_pad_0, pad_type = lora_out_181_pad_type_0, strides = var_13172, weight = lora_out_183_weight_0_to_fp16, x = input_149_cast_fp16)[name = tensor("lora_out_183_cast_fp16")]; + tensor obj_31_cast_fp16 = add(x = pretrained_out_91_cast_fp16, y = lora_out_183_cast_fp16)[name = tensor("obj_31_cast_fp16")]; + tensor inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor var_13183 = const()[name = tensor("op_13183"), val = tensor([1])]; + tensor channels_mean_31_cast_fp16 = reduce_mean(axes = var_13183, keep_dims = var_11645, x = inputs_31_cast_fp16)[name = tensor("channels_mean_31_cast_fp16")]; + tensor zero_mean_31_cast_fp16 = sub(x = inputs_31_cast_fp16, y = channels_mean_31_cast_fp16)[name = tensor("zero_mean_31_cast_fp16")]; + tensor zero_mean_sq_31_cast_fp16 = mul(x = zero_mean_31_cast_fp16, y = zero_mean_31_cast_fp16)[name = tensor("zero_mean_sq_31_cast_fp16")]; + tensor var_13187 = const()[name = tensor("op_13187"), val = tensor([1])]; + tensor var_13188_cast_fp16 = reduce_mean(axes = var_13187, keep_dims = var_11645, x = zero_mean_sq_31_cast_fp16)[name = tensor("op_13188_cast_fp16")]; + tensor var_13189_to_fp16 = const()[name = tensor("op_13189_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_13190_cast_fp16 = add(x = var_13188_cast_fp16, y = var_13189_to_fp16)[name = tensor("op_13190_cast_fp16")]; + tensor denom_31_epsilon_0 = const()[name = tensor("denom_31_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_31_cast_fp16 = rsqrt(epsilon = denom_31_epsilon_0, x = var_13190_cast_fp16)[name = tensor("denom_31_cast_fp16")]; + tensor out_31_cast_fp16 = mul(x = zero_mean_31_cast_fp16, y = denom_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; + tensor input_151_gamma_0_to_fp16 = const()[name = tensor("input_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92131008)))]; + tensor input_151_beta_0_to_fp16 = const()[name = tensor("input_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92133632)))]; + tensor input_151_epsilon_0_to_fp16 = const()[name = tensor("input_151_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_151_cast_fp16 = batch_norm(beta = input_151_beta_0_to_fp16, epsilon = input_151_epsilon_0_to_fp16, gamma = input_151_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor var_13204 = const()[name = tensor("op_13204"), val = tensor([1, 1])]; + tensor var_13206 = const()[name = tensor("op_13206"), val = tensor([1, 1])]; + tensor pretrained_out_93_pad_type_0 = const()[name = tensor("pretrained_out_93_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_93_pad_0 = const()[name = tensor("pretrained_out_93_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92136256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95413120))), name = tensor("layers_7_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_7_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_7_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95413248)))]; + tensor pretrained_out_93_cast_fp16 = conv(bias = layers_7_fc1_pretrained_bias_to_fp16, dilations = var_13206, groups = var_11644, pad = pretrained_out_93_pad_0, pad_type = pretrained_out_93_pad_type_0, strides = var_13204, weight = layers_7_fc1_pretrained_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("pretrained_out_93_cast_fp16")]; + tensor var_13210 = const()[name = tensor("op_13210"), val = tensor([1, 1])]; + tensor var_13212 = const()[name = tensor("op_13212"), val = tensor([1, 1])]; + tensor input_153_pad_type_0 = const()[name = tensor("input_153_pad_type_0"), val = tensor("custom")]; + tensor input_153_pad_0 = const()[name = tensor("input_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_7_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95423552)))]; + tensor input_153_cast_fp16 = conv(dilations = var_13212, groups = var_11644, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = var_13210, weight = layers_7_fc1_loraA_weight_to_fp16, x = input_151_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor var_13216 = const()[name = tensor("op_13216"), val = tensor([1, 1])]; + tensor var_13218 = const()[name = tensor("op_13218"), val = tensor([1, 1])]; + tensor lora_out_185_pad_type_0 = const()[name = tensor("lora_out_185_pad_type_0"), val = tensor("custom")]; + tensor lora_out_185_pad_0 = const()[name = tensor("lora_out_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_187_weight_0_to_fp16 = const()[name = tensor("lora_out_187_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95464576)))]; + tensor lora_out_187_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_13218, groups = var_11644, pad = lora_out_185_pad_0, pad_type = lora_out_185_pad_type_0, strides = var_13216, weight = lora_out_187_weight_0_to_fp16, x = input_153_cast_fp16)[name = tensor("lora_out_187_cast_fp16")]; + tensor input_155_cast_fp16 = add(x = pretrained_out_93_cast_fp16, y = lora_out_187_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor input_157_mode_0 = const()[name = tensor("input_157_mode_0"), val = tensor("EXACT")]; + tensor input_157_cast_fp16 = gelu(mode = input_157_mode_0, x = input_155_cast_fp16)[name = tensor("input_157_cast_fp16")]; + tensor var_13230 = const()[name = tensor("op_13230"), val = tensor([1, 1])]; + tensor var_13232 = const()[name = tensor("op_13232"), val = tensor([1, 1])]; + tensor pretrained_out_95_pad_type_0 = const()[name = tensor("pretrained_out_95_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_95_pad_0 = const()[name = tensor("pretrained_out_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95628480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98905344))), name = tensor("layers_7_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_7_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_7_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98905472)))]; + tensor pretrained_out_95_cast_fp16 = conv(bias = layers_7_fc2_pretrained_bias_to_fp16, dilations = var_13232, groups = var_11644, pad = pretrained_out_95_pad_0, pad_type = pretrained_out_95_pad_type_0, strides = var_13230, weight = layers_7_fc2_pretrained_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor("pretrained_out_95_cast_fp16")]; + tensor var_13236 = const()[name = tensor("op_13236"), val = tensor([1, 1])]; + tensor var_13238 = const()[name = tensor("op_13238"), val = tensor([1, 1])]; + tensor input_159_pad_type_0 = const()[name = tensor("input_159_pad_type_0"), val = tensor("custom")]; + tensor input_159_pad_0 = const()[name = tensor("input_159_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_7_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_7_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98908096)))]; + tensor input_159_cast_fp16 = conv(dilations = var_13238, groups = var_11644, pad = input_159_pad_0, pad_type = input_159_pad_type_0, strides = var_13236, weight = layers_7_fc2_loraA_weight_to_fp16, x = input_157_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor var_13242 = const()[name = tensor("op_13242"), val = tensor([1, 1])]; + tensor var_13244 = const()[name = tensor("op_13244"), val = tensor([1, 1])]; + tensor lora_out_189_pad_type_0 = const()[name = tensor("lora_out_189_pad_type_0"), val = tensor("custom")]; + tensor lora_out_189_pad_0 = const()[name = tensor("lora_out_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_191_weight_0_to_fp16 = const()[name = tensor("lora_out_191_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99072000)))]; + tensor lora_out_191_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_13244, groups = var_11644, pad = lora_out_189_pad_0, pad_type = lora_out_189_pad_type_0, strides = var_13242, weight = lora_out_191_weight_0_to_fp16, x = input_159_cast_fp16)[name = tensor("lora_out_191_cast_fp16")]; + tensor hidden_states_19_cast_fp16 = add(x = pretrained_out_95_cast_fp16, y = lora_out_191_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor var_13254 = const()[name = tensor("op_13254"), val = tensor(3)]; + tensor var_13279 = const()[name = tensor("op_13279"), val = tensor(1)]; + tensor var_13280 = const()[name = tensor("op_13280"), val = tensor(true)]; + tensor var_13290 = const()[name = tensor("op_13290"), val = tensor([1])]; + tensor channels_mean_33_cast_fp16 = reduce_mean(axes = var_13290, keep_dims = var_13280, x = inputs_33_cast_fp16)[name = tensor("channels_mean_33_cast_fp16")]; + tensor zero_mean_33_cast_fp16 = sub(x = inputs_33_cast_fp16, y = channels_mean_33_cast_fp16)[name = tensor("zero_mean_33_cast_fp16")]; + tensor zero_mean_sq_33_cast_fp16 = mul(x = zero_mean_33_cast_fp16, y = zero_mean_33_cast_fp16)[name = tensor("zero_mean_sq_33_cast_fp16")]; + tensor var_13294 = const()[name = tensor("op_13294"), val = tensor([1])]; + tensor var_13295_cast_fp16 = reduce_mean(axes = var_13294, keep_dims = var_13280, x = zero_mean_sq_33_cast_fp16)[name = tensor("op_13295_cast_fp16")]; + tensor var_13296_to_fp16 = const()[name = tensor("op_13296_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_13297_cast_fp16 = add(x = var_13295_cast_fp16, y = var_13296_to_fp16)[name = tensor("op_13297_cast_fp16")]; + tensor denom_33_epsilon_0 = const()[name = tensor("denom_33_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_33_cast_fp16 = rsqrt(epsilon = denom_33_epsilon_0, x = var_13297_cast_fp16)[name = tensor("denom_33_cast_fp16")]; + tensor out_33_cast_fp16 = mul(x = zero_mean_33_cast_fp16, y = denom_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; + tensor obj_33_gamma_0_to_fp16 = const()[name = tensor("obj_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99113024)))]; + tensor obj_33_beta_0_to_fp16 = const()[name = tensor("obj_33_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99115648)))]; + tensor obj_33_epsilon_0_to_fp16 = const()[name = tensor("obj_33_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor("obj_33_cast_fp16")]; + tensor var_13315 = const()[name = tensor("op_13315"), val = tensor([1, 1])]; + tensor var_13317 = const()[name = tensor("op_13317"), val = tensor([1, 1])]; + tensor pretrained_out_97_pad_type_0 = const()[name = tensor("pretrained_out_97_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_97_pad_0 = const()[name = tensor("pretrained_out_97_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99118272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99937536))), name = tensor("layers_8_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_8_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99937664)))]; + tensor pretrained_out_97_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_13317, groups = var_13279, pad = pretrained_out_97_pad_0, pad_type = pretrained_out_97_pad_type_0, strides = var_13315, weight = layers_8_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor("pretrained_out_97_cast_fp16")]; + tensor var_13321 = const()[name = tensor("op_13321"), val = tensor([1, 1])]; + tensor var_13323 = const()[name = tensor("op_13323"), val = tensor([1, 1])]; + tensor input_161_pad_type_0 = const()[name = tensor("input_161_pad_type_0"), val = tensor("custom")]; + tensor input_161_pad_0 = const()[name = tensor("input_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99940288)))]; + tensor input_161_cast_fp16 = conv(dilations = var_13323, groups = var_13279, pad = input_161_pad_0, pad_type = input_161_pad_type_0, strides = var_13321, weight = layers_8_self_attn_q_proj_loraA_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor var_13327 = const()[name = tensor("op_13327"), val = tensor([1, 1])]; + tensor var_13329 = const()[name = tensor("op_13329"), val = tensor([1, 1])]; + tensor lora_out_193_pad_type_0 = const()[name = tensor("lora_out_193_pad_type_0"), val = tensor("custom")]; + tensor lora_out_193_pad_0 = const()[name = tensor("lora_out_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_195_weight_0_to_fp16 = const()[name = tensor("lora_out_195_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99981312)))]; + tensor lora_out_195_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_13329, groups = var_13279, pad = lora_out_193_pad_0, pad_type = lora_out_193_pad_type_0, strides = var_13327, weight = lora_out_195_weight_0_to_fp16, x = input_161_cast_fp16)[name = tensor("lora_out_195_cast_fp16")]; + tensor query_17_cast_fp16 = add(x = pretrained_out_97_cast_fp16, y = lora_out_195_cast_fp16)[name = tensor("query_17_cast_fp16")]; + tensor var_13339 = const()[name = tensor("op_13339"), val = tensor([1, 1])]; + tensor var_13341 = const()[name = tensor("op_13341"), val = tensor([1, 1])]; + tensor pretrained_out_99_pad_type_0 = const()[name = tensor("pretrained_out_99_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_99_pad_0 = const()[name = tensor("pretrained_out_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100022336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100841600))), name = tensor("layers_8_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_99_cast_fp16 = conv(dilations = var_13341, groups = var_13279, pad = pretrained_out_99_pad_0, pad_type = pretrained_out_99_pad_type_0, strides = var_13339, weight = layers_8_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor("pretrained_out_99_cast_fp16")]; + tensor var_13345 = const()[name = tensor("op_13345"), val = tensor([1, 1])]; + tensor var_13347 = const()[name = tensor("op_13347"), val = tensor([1, 1])]; + tensor input_163_pad_type_0 = const()[name = tensor("input_163_pad_type_0"), val = tensor("custom")]; + tensor input_163_pad_0 = const()[name = tensor("input_163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100841728)))]; + tensor input_163_cast_fp16 = conv(dilations = var_13347, groups = var_13279, pad = input_163_pad_0, pad_type = input_163_pad_type_0, strides = var_13345, weight = layers_8_self_attn_k_proj_loraA_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor var_13351 = const()[name = tensor("op_13351"), val = tensor([1, 1])]; + tensor var_13353 = const()[name = tensor("op_13353"), val = tensor([1, 1])]; + tensor lora_out_197_pad_type_0 = const()[name = tensor("lora_out_197_pad_type_0"), val = tensor("custom")]; + tensor lora_out_197_pad_0 = const()[name = tensor("lora_out_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_199_weight_0_to_fp16 = const()[name = tensor("lora_out_199_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100882752)))]; + tensor lora_out_199_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_13353, groups = var_13279, pad = lora_out_197_pad_0, pad_type = lora_out_197_pad_type_0, strides = var_13351, weight = lora_out_199_weight_0_to_fp16, x = input_163_cast_fp16)[name = tensor("lora_out_199_cast_fp16")]; + tensor key_17_cast_fp16 = add(x = pretrained_out_99_cast_fp16, y = lora_out_199_cast_fp16)[name = tensor("key_17_cast_fp16")]; + tensor var_13364 = const()[name = tensor("op_13364"), val = tensor([1, 1])]; + tensor var_13366 = const()[name = tensor("op_13366"), val = tensor([1, 1])]; + tensor pretrained_out_101_pad_type_0 = const()[name = tensor("pretrained_out_101_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_101_pad_0 = const()[name = tensor("pretrained_out_101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100923776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101743040))), name = tensor("layers_8_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_8_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101743168)))]; + tensor pretrained_out_101_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_13366, groups = var_13279, pad = pretrained_out_101_pad_0, pad_type = pretrained_out_101_pad_type_0, strides = var_13364, weight = layers_8_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor("pretrained_out_101_cast_fp16")]; + tensor var_13370 = const()[name = tensor("op_13370"), val = tensor([1, 1])]; + tensor var_13372 = const()[name = tensor("op_13372"), val = tensor([1, 1])]; + tensor input_165_pad_type_0 = const()[name = tensor("input_165_pad_type_0"), val = tensor("custom")]; + tensor input_165_pad_0 = const()[name = tensor("input_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101745792)))]; + tensor input_165_cast_fp16 = conv(dilations = var_13372, groups = var_13279, pad = input_165_pad_0, pad_type = input_165_pad_type_0, strides = var_13370, weight = layers_8_self_attn_v_proj_loraA_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor var_13376 = const()[name = tensor("op_13376"), val = tensor([1, 1])]; + tensor var_13378 = const()[name = tensor("op_13378"), val = tensor([1, 1])]; + tensor lora_out_201_pad_type_0 = const()[name = tensor("lora_out_201_pad_type_0"), val = tensor("custom")]; + tensor lora_out_201_pad_0 = const()[name = tensor("lora_out_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_203_weight_0_to_fp16 = const()[name = tensor("lora_out_203_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101786816)))]; + tensor lora_out_203_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_13378, groups = var_13279, pad = lora_out_201_pad_0, pad_type = lora_out_201_pad_type_0, strides = var_13376, weight = lora_out_203_weight_0_to_fp16, x = input_165_cast_fp16)[name = tensor("lora_out_203_cast_fp16")]; + tensor value_17_cast_fp16 = add(x = pretrained_out_101_cast_fp16, y = lora_out_203_cast_fp16)[name = tensor("value_17_cast_fp16")]; + tensor var_13388_begin_0 = const()[name = tensor("op_13388_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13388_end_0 = const()[name = tensor("op_13388_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13388_end_mask_0 = const()[name = tensor("op_13388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13388_cast_fp16 = slice_by_index(begin = var_13388_begin_0, end = var_13388_end_0, end_mask = var_13388_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13388_cast_fp16")]; + tensor var_13392_begin_0 = const()[name = tensor("op_13392_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_13392_end_0 = const()[name = tensor("op_13392_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_13392_end_mask_0 = const()[name = tensor("op_13392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13392_cast_fp16 = slice_by_index(begin = var_13392_begin_0, end = var_13392_end_0, end_mask = var_13392_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13392_cast_fp16")]; + tensor var_13396_begin_0 = const()[name = tensor("op_13396_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_13396_end_0 = const()[name = tensor("op_13396_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_13396_end_mask_0 = const()[name = tensor("op_13396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13396_cast_fp16 = slice_by_index(begin = var_13396_begin_0, end = var_13396_end_0, end_mask = var_13396_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13396_cast_fp16")]; + tensor var_13400_begin_0 = const()[name = tensor("op_13400_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_13400_end_0 = const()[name = tensor("op_13400_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_13400_end_mask_0 = const()[name = tensor("op_13400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13400_cast_fp16 = slice_by_index(begin = var_13400_begin_0, end = var_13400_end_0, end_mask = var_13400_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13400_cast_fp16")]; + tensor var_13404_begin_0 = const()[name = tensor("op_13404_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_13404_end_0 = const()[name = tensor("op_13404_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_13404_end_mask_0 = const()[name = tensor("op_13404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13404_cast_fp16 = slice_by_index(begin = var_13404_begin_0, end = var_13404_end_0, end_mask = var_13404_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13404_cast_fp16")]; + tensor var_13408_begin_0 = const()[name = tensor("op_13408_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_13408_end_0 = const()[name = tensor("op_13408_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_13408_end_mask_0 = const()[name = tensor("op_13408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13408_cast_fp16 = slice_by_index(begin = var_13408_begin_0, end = var_13408_end_0, end_mask = var_13408_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13408_cast_fp16")]; + tensor var_13412_begin_0 = const()[name = tensor("op_13412_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_13412_end_0 = const()[name = tensor("op_13412_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_13412_end_mask_0 = const()[name = tensor("op_13412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13412_cast_fp16 = slice_by_index(begin = var_13412_begin_0, end = var_13412_end_0, end_mask = var_13412_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13412_cast_fp16")]; + tensor var_13416_begin_0 = const()[name = tensor("op_13416_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_13416_end_0 = const()[name = tensor("op_13416_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_13416_end_mask_0 = const()[name = tensor("op_13416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13416_cast_fp16 = slice_by_index(begin = var_13416_begin_0, end = var_13416_end_0, end_mask = var_13416_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13416_cast_fp16")]; + tensor var_13420_begin_0 = const()[name = tensor("op_13420_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_13420_end_0 = const()[name = tensor("op_13420_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_13420_end_mask_0 = const()[name = tensor("op_13420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13420_cast_fp16 = slice_by_index(begin = var_13420_begin_0, end = var_13420_end_0, end_mask = var_13420_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13420_cast_fp16")]; + tensor var_13424_begin_0 = const()[name = tensor("op_13424_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_13424_end_0 = const()[name = tensor("op_13424_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_13424_end_mask_0 = const()[name = tensor("op_13424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13424_cast_fp16 = slice_by_index(begin = var_13424_begin_0, end = var_13424_end_0, end_mask = var_13424_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13424_cast_fp16")]; + tensor var_13428_begin_0 = const()[name = tensor("op_13428_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_13428_end_0 = const()[name = tensor("op_13428_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_13428_end_mask_0 = const()[name = tensor("op_13428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13428_cast_fp16 = slice_by_index(begin = var_13428_begin_0, end = var_13428_end_0, end_mask = var_13428_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13428_cast_fp16")]; + tensor var_13432_begin_0 = const()[name = tensor("op_13432_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_13432_end_0 = const()[name = tensor("op_13432_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_13432_end_mask_0 = const()[name = tensor("op_13432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13432_cast_fp16 = slice_by_index(begin = var_13432_begin_0, end = var_13432_end_0, end_mask = var_13432_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13432_cast_fp16")]; + tensor var_13436_begin_0 = const()[name = tensor("op_13436_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_13436_end_0 = const()[name = tensor("op_13436_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_13436_end_mask_0 = const()[name = tensor("op_13436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13436_cast_fp16 = slice_by_index(begin = var_13436_begin_0, end = var_13436_end_0, end_mask = var_13436_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13436_cast_fp16")]; + tensor var_13440_begin_0 = const()[name = tensor("op_13440_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_13440_end_0 = const()[name = tensor("op_13440_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_13440_end_mask_0 = const()[name = tensor("op_13440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13440_cast_fp16 = slice_by_index(begin = var_13440_begin_0, end = var_13440_end_0, end_mask = var_13440_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13440_cast_fp16")]; + tensor var_13444_begin_0 = const()[name = tensor("op_13444_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_13444_end_0 = const()[name = tensor("op_13444_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_13444_end_mask_0 = const()[name = tensor("op_13444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13444_cast_fp16 = slice_by_index(begin = var_13444_begin_0, end = var_13444_end_0, end_mask = var_13444_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13444_cast_fp16")]; + tensor var_13448_begin_0 = const()[name = tensor("op_13448_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_13448_end_0 = const()[name = tensor("op_13448_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_13448_end_mask_0 = const()[name = tensor("op_13448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13448_cast_fp16 = slice_by_index(begin = var_13448_begin_0, end = var_13448_end_0, end_mask = var_13448_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13448_cast_fp16")]; + tensor var_13452_begin_0 = const()[name = tensor("op_13452_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_13452_end_0 = const()[name = tensor("op_13452_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_13452_end_mask_0 = const()[name = tensor("op_13452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13452_cast_fp16 = slice_by_index(begin = var_13452_begin_0, end = var_13452_end_0, end_mask = var_13452_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13452_cast_fp16")]; + tensor var_13456_begin_0 = const()[name = tensor("op_13456_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_13456_end_0 = const()[name = tensor("op_13456_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_13456_end_mask_0 = const()[name = tensor("op_13456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13456_cast_fp16 = slice_by_index(begin = var_13456_begin_0, end = var_13456_end_0, end_mask = var_13456_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13456_cast_fp16")]; + tensor var_13460_begin_0 = const()[name = tensor("op_13460_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_13460_end_0 = const()[name = tensor("op_13460_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_13460_end_mask_0 = const()[name = tensor("op_13460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13460_cast_fp16 = slice_by_index(begin = var_13460_begin_0, end = var_13460_end_0, end_mask = var_13460_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13460_cast_fp16")]; + tensor var_13464_begin_0 = const()[name = tensor("op_13464_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_13464_end_0 = const()[name = tensor("op_13464_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_13464_end_mask_0 = const()[name = tensor("op_13464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_13464_cast_fp16 = slice_by_index(begin = var_13464_begin_0, end = var_13464_end_0, end_mask = var_13464_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_13464_cast_fp16")]; + tensor var_13473_begin_0 = const()[name = tensor("op_13473_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13473_end_0 = const()[name = tensor("op_13473_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13473_end_mask_0 = const()[name = tensor("op_13473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13473_cast_fp16 = slice_by_index(begin = var_13473_begin_0, end = var_13473_end_0, end_mask = var_13473_end_mask_0, x = var_13388_cast_fp16)[name = tensor("op_13473_cast_fp16")]; + tensor var_13480_begin_0 = const()[name = tensor("op_13480_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13480_end_0 = const()[name = tensor("op_13480_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13480_end_mask_0 = const()[name = tensor("op_13480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13480_cast_fp16 = slice_by_index(begin = var_13480_begin_0, end = var_13480_end_0, end_mask = var_13480_end_mask_0, x = var_13388_cast_fp16)[name = tensor("op_13480_cast_fp16")]; + tensor var_13487_begin_0 = const()[name = tensor("op_13487_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13487_end_0 = const()[name = tensor("op_13487_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13487_end_mask_0 = const()[name = tensor("op_13487_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13487_cast_fp16 = slice_by_index(begin = var_13487_begin_0, end = var_13487_end_0, end_mask = var_13487_end_mask_0, x = var_13388_cast_fp16)[name = tensor("op_13487_cast_fp16")]; + tensor var_13494_begin_0 = const()[name = tensor("op_13494_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13494_end_0 = const()[name = tensor("op_13494_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13494_end_mask_0 = const()[name = tensor("op_13494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13494_cast_fp16 = slice_by_index(begin = var_13494_begin_0, end = var_13494_end_0, end_mask = var_13494_end_mask_0, x = var_13388_cast_fp16)[name = tensor("op_13494_cast_fp16")]; + tensor var_13501_begin_0 = const()[name = tensor("op_13501_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13501_end_0 = const()[name = tensor("op_13501_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13501_end_mask_0 = const()[name = tensor("op_13501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13501_cast_fp16 = slice_by_index(begin = var_13501_begin_0, end = var_13501_end_0, end_mask = var_13501_end_mask_0, x = var_13392_cast_fp16)[name = tensor("op_13501_cast_fp16")]; + tensor var_13508_begin_0 = const()[name = tensor("op_13508_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13508_end_0 = const()[name = tensor("op_13508_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13508_end_mask_0 = const()[name = tensor("op_13508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13508_cast_fp16 = slice_by_index(begin = var_13508_begin_0, end = var_13508_end_0, end_mask = var_13508_end_mask_0, x = var_13392_cast_fp16)[name = tensor("op_13508_cast_fp16")]; + tensor var_13515_begin_0 = const()[name = tensor("op_13515_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13515_end_0 = const()[name = tensor("op_13515_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13515_end_mask_0 = const()[name = tensor("op_13515_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13515_cast_fp16 = slice_by_index(begin = var_13515_begin_0, end = var_13515_end_0, end_mask = var_13515_end_mask_0, x = var_13392_cast_fp16)[name = tensor("op_13515_cast_fp16")]; + tensor var_13522_begin_0 = const()[name = tensor("op_13522_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13522_end_0 = const()[name = tensor("op_13522_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13522_end_mask_0 = const()[name = tensor("op_13522_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13522_cast_fp16 = slice_by_index(begin = var_13522_begin_0, end = var_13522_end_0, end_mask = var_13522_end_mask_0, x = var_13392_cast_fp16)[name = tensor("op_13522_cast_fp16")]; + tensor var_13529_begin_0 = const()[name = tensor("op_13529_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13529_end_0 = const()[name = tensor("op_13529_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13529_end_mask_0 = const()[name = tensor("op_13529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13529_cast_fp16 = slice_by_index(begin = var_13529_begin_0, end = var_13529_end_0, end_mask = var_13529_end_mask_0, x = var_13396_cast_fp16)[name = tensor("op_13529_cast_fp16")]; + tensor var_13536_begin_0 = const()[name = tensor("op_13536_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13536_end_0 = const()[name = tensor("op_13536_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13536_end_mask_0 = const()[name = tensor("op_13536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13536_cast_fp16 = slice_by_index(begin = var_13536_begin_0, end = var_13536_end_0, end_mask = var_13536_end_mask_0, x = var_13396_cast_fp16)[name = tensor("op_13536_cast_fp16")]; + tensor var_13543_begin_0 = const()[name = tensor("op_13543_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13543_end_0 = const()[name = tensor("op_13543_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13543_end_mask_0 = const()[name = tensor("op_13543_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13543_cast_fp16 = slice_by_index(begin = var_13543_begin_0, end = var_13543_end_0, end_mask = var_13543_end_mask_0, x = var_13396_cast_fp16)[name = tensor("op_13543_cast_fp16")]; + tensor var_13550_begin_0 = const()[name = tensor("op_13550_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13550_end_0 = const()[name = tensor("op_13550_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13550_end_mask_0 = const()[name = tensor("op_13550_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13550_cast_fp16 = slice_by_index(begin = var_13550_begin_0, end = var_13550_end_0, end_mask = var_13550_end_mask_0, x = var_13396_cast_fp16)[name = tensor("op_13550_cast_fp16")]; + tensor var_13557_begin_0 = const()[name = tensor("op_13557_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13557_end_0 = const()[name = tensor("op_13557_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13557_end_mask_0 = const()[name = tensor("op_13557_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13557_cast_fp16 = slice_by_index(begin = var_13557_begin_0, end = var_13557_end_0, end_mask = var_13557_end_mask_0, x = var_13400_cast_fp16)[name = tensor("op_13557_cast_fp16")]; + tensor var_13564_begin_0 = const()[name = tensor("op_13564_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13564_end_0 = const()[name = tensor("op_13564_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13564_end_mask_0 = const()[name = tensor("op_13564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13564_cast_fp16 = slice_by_index(begin = var_13564_begin_0, end = var_13564_end_0, end_mask = var_13564_end_mask_0, x = var_13400_cast_fp16)[name = tensor("op_13564_cast_fp16")]; + tensor var_13571_begin_0 = const()[name = tensor("op_13571_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13571_end_0 = const()[name = tensor("op_13571_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13571_end_mask_0 = const()[name = tensor("op_13571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13571_cast_fp16 = slice_by_index(begin = var_13571_begin_0, end = var_13571_end_0, end_mask = var_13571_end_mask_0, x = var_13400_cast_fp16)[name = tensor("op_13571_cast_fp16")]; + tensor var_13578_begin_0 = const()[name = tensor("op_13578_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13578_end_0 = const()[name = tensor("op_13578_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13578_end_mask_0 = const()[name = tensor("op_13578_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13578_cast_fp16 = slice_by_index(begin = var_13578_begin_0, end = var_13578_end_0, end_mask = var_13578_end_mask_0, x = var_13400_cast_fp16)[name = tensor("op_13578_cast_fp16")]; + tensor var_13585_begin_0 = const()[name = tensor("op_13585_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13585_end_0 = const()[name = tensor("op_13585_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13585_end_mask_0 = const()[name = tensor("op_13585_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13585_cast_fp16 = slice_by_index(begin = var_13585_begin_0, end = var_13585_end_0, end_mask = var_13585_end_mask_0, x = var_13404_cast_fp16)[name = tensor("op_13585_cast_fp16")]; + tensor var_13592_begin_0 = const()[name = tensor("op_13592_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13592_end_0 = const()[name = tensor("op_13592_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13592_end_mask_0 = const()[name = tensor("op_13592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13592_cast_fp16 = slice_by_index(begin = var_13592_begin_0, end = var_13592_end_0, end_mask = var_13592_end_mask_0, x = var_13404_cast_fp16)[name = tensor("op_13592_cast_fp16")]; + tensor var_13599_begin_0 = const()[name = tensor("op_13599_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13599_end_0 = const()[name = tensor("op_13599_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13599_end_mask_0 = const()[name = tensor("op_13599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13599_cast_fp16 = slice_by_index(begin = var_13599_begin_0, end = var_13599_end_0, end_mask = var_13599_end_mask_0, x = var_13404_cast_fp16)[name = tensor("op_13599_cast_fp16")]; + tensor var_13606_begin_0 = const()[name = tensor("op_13606_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13606_end_0 = const()[name = tensor("op_13606_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13606_end_mask_0 = const()[name = tensor("op_13606_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13606_cast_fp16 = slice_by_index(begin = var_13606_begin_0, end = var_13606_end_0, end_mask = var_13606_end_mask_0, x = var_13404_cast_fp16)[name = tensor("op_13606_cast_fp16")]; + tensor var_13613_begin_0 = const()[name = tensor("op_13613_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13613_end_0 = const()[name = tensor("op_13613_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13613_end_mask_0 = const()[name = tensor("op_13613_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13613_cast_fp16 = slice_by_index(begin = var_13613_begin_0, end = var_13613_end_0, end_mask = var_13613_end_mask_0, x = var_13408_cast_fp16)[name = tensor("op_13613_cast_fp16")]; + tensor var_13620_begin_0 = const()[name = tensor("op_13620_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13620_end_0 = const()[name = tensor("op_13620_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13620_end_mask_0 = const()[name = tensor("op_13620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13620_cast_fp16 = slice_by_index(begin = var_13620_begin_0, end = var_13620_end_0, end_mask = var_13620_end_mask_0, x = var_13408_cast_fp16)[name = tensor("op_13620_cast_fp16")]; + tensor var_13627_begin_0 = const()[name = tensor("op_13627_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13627_end_0 = const()[name = tensor("op_13627_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13627_end_mask_0 = const()[name = tensor("op_13627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13627_cast_fp16 = slice_by_index(begin = var_13627_begin_0, end = var_13627_end_0, end_mask = var_13627_end_mask_0, x = var_13408_cast_fp16)[name = tensor("op_13627_cast_fp16")]; + tensor var_13634_begin_0 = const()[name = tensor("op_13634_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13634_end_0 = const()[name = tensor("op_13634_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13634_end_mask_0 = const()[name = tensor("op_13634_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13634_cast_fp16 = slice_by_index(begin = var_13634_begin_0, end = var_13634_end_0, end_mask = var_13634_end_mask_0, x = var_13408_cast_fp16)[name = tensor("op_13634_cast_fp16")]; + tensor var_13641_begin_0 = const()[name = tensor("op_13641_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13641_end_0 = const()[name = tensor("op_13641_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13641_end_mask_0 = const()[name = tensor("op_13641_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13641_cast_fp16 = slice_by_index(begin = var_13641_begin_0, end = var_13641_end_0, end_mask = var_13641_end_mask_0, x = var_13412_cast_fp16)[name = tensor("op_13641_cast_fp16")]; + tensor var_13648_begin_0 = const()[name = tensor("op_13648_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13648_end_0 = const()[name = tensor("op_13648_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13648_end_mask_0 = const()[name = tensor("op_13648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13648_cast_fp16 = slice_by_index(begin = var_13648_begin_0, end = var_13648_end_0, end_mask = var_13648_end_mask_0, x = var_13412_cast_fp16)[name = tensor("op_13648_cast_fp16")]; + tensor var_13655_begin_0 = const()[name = tensor("op_13655_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13655_end_0 = const()[name = tensor("op_13655_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13655_end_mask_0 = const()[name = tensor("op_13655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13655_cast_fp16 = slice_by_index(begin = var_13655_begin_0, end = var_13655_end_0, end_mask = var_13655_end_mask_0, x = var_13412_cast_fp16)[name = tensor("op_13655_cast_fp16")]; + tensor var_13662_begin_0 = const()[name = tensor("op_13662_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13662_end_0 = const()[name = tensor("op_13662_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13662_end_mask_0 = const()[name = tensor("op_13662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13662_cast_fp16 = slice_by_index(begin = var_13662_begin_0, end = var_13662_end_0, end_mask = var_13662_end_mask_0, x = var_13412_cast_fp16)[name = tensor("op_13662_cast_fp16")]; + tensor var_13669_begin_0 = const()[name = tensor("op_13669_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13669_end_0 = const()[name = tensor("op_13669_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13669_end_mask_0 = const()[name = tensor("op_13669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13669_cast_fp16 = slice_by_index(begin = var_13669_begin_0, end = var_13669_end_0, end_mask = var_13669_end_mask_0, x = var_13416_cast_fp16)[name = tensor("op_13669_cast_fp16")]; + tensor var_13676_begin_0 = const()[name = tensor("op_13676_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13676_end_0 = const()[name = tensor("op_13676_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13676_end_mask_0 = const()[name = tensor("op_13676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13676_cast_fp16 = slice_by_index(begin = var_13676_begin_0, end = var_13676_end_0, end_mask = var_13676_end_mask_0, x = var_13416_cast_fp16)[name = tensor("op_13676_cast_fp16")]; + tensor var_13683_begin_0 = const()[name = tensor("op_13683_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13683_end_0 = const()[name = tensor("op_13683_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13683_end_mask_0 = const()[name = tensor("op_13683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13683_cast_fp16 = slice_by_index(begin = var_13683_begin_0, end = var_13683_end_0, end_mask = var_13683_end_mask_0, x = var_13416_cast_fp16)[name = tensor("op_13683_cast_fp16")]; + tensor var_13690_begin_0 = const()[name = tensor("op_13690_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13690_end_0 = const()[name = tensor("op_13690_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13690_end_mask_0 = const()[name = tensor("op_13690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13690_cast_fp16 = slice_by_index(begin = var_13690_begin_0, end = var_13690_end_0, end_mask = var_13690_end_mask_0, x = var_13416_cast_fp16)[name = tensor("op_13690_cast_fp16")]; + tensor var_13697_begin_0 = const()[name = tensor("op_13697_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13697_end_0 = const()[name = tensor("op_13697_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13697_end_mask_0 = const()[name = tensor("op_13697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13697_cast_fp16 = slice_by_index(begin = var_13697_begin_0, end = var_13697_end_0, end_mask = var_13697_end_mask_0, x = var_13420_cast_fp16)[name = tensor("op_13697_cast_fp16")]; + tensor var_13704_begin_0 = const()[name = tensor("op_13704_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13704_end_0 = const()[name = tensor("op_13704_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13704_end_mask_0 = const()[name = tensor("op_13704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13704_cast_fp16 = slice_by_index(begin = var_13704_begin_0, end = var_13704_end_0, end_mask = var_13704_end_mask_0, x = var_13420_cast_fp16)[name = tensor("op_13704_cast_fp16")]; + tensor var_13711_begin_0 = const()[name = tensor("op_13711_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13711_end_0 = const()[name = tensor("op_13711_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13711_end_mask_0 = const()[name = tensor("op_13711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13711_cast_fp16 = slice_by_index(begin = var_13711_begin_0, end = var_13711_end_0, end_mask = var_13711_end_mask_0, x = var_13420_cast_fp16)[name = tensor("op_13711_cast_fp16")]; + tensor var_13718_begin_0 = const()[name = tensor("op_13718_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13718_end_0 = const()[name = tensor("op_13718_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13718_end_mask_0 = const()[name = tensor("op_13718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13718_cast_fp16 = slice_by_index(begin = var_13718_begin_0, end = var_13718_end_0, end_mask = var_13718_end_mask_0, x = var_13420_cast_fp16)[name = tensor("op_13718_cast_fp16")]; + tensor var_13725_begin_0 = const()[name = tensor("op_13725_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13725_end_0 = const()[name = tensor("op_13725_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13725_end_mask_0 = const()[name = tensor("op_13725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13725_cast_fp16 = slice_by_index(begin = var_13725_begin_0, end = var_13725_end_0, end_mask = var_13725_end_mask_0, x = var_13424_cast_fp16)[name = tensor("op_13725_cast_fp16")]; + tensor var_13732_begin_0 = const()[name = tensor("op_13732_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13732_end_0 = const()[name = tensor("op_13732_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13732_end_mask_0 = const()[name = tensor("op_13732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13732_cast_fp16 = slice_by_index(begin = var_13732_begin_0, end = var_13732_end_0, end_mask = var_13732_end_mask_0, x = var_13424_cast_fp16)[name = tensor("op_13732_cast_fp16")]; + tensor var_13739_begin_0 = const()[name = tensor("op_13739_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13739_end_0 = const()[name = tensor("op_13739_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13739_end_mask_0 = const()[name = tensor("op_13739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13739_cast_fp16 = slice_by_index(begin = var_13739_begin_0, end = var_13739_end_0, end_mask = var_13739_end_mask_0, x = var_13424_cast_fp16)[name = tensor("op_13739_cast_fp16")]; + tensor var_13746_begin_0 = const()[name = tensor("op_13746_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13746_end_0 = const()[name = tensor("op_13746_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13746_end_mask_0 = const()[name = tensor("op_13746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13746_cast_fp16 = slice_by_index(begin = var_13746_begin_0, end = var_13746_end_0, end_mask = var_13746_end_mask_0, x = var_13424_cast_fp16)[name = tensor("op_13746_cast_fp16")]; + tensor var_13753_begin_0 = const()[name = tensor("op_13753_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13753_end_0 = const()[name = tensor("op_13753_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13753_end_mask_0 = const()[name = tensor("op_13753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13753_cast_fp16 = slice_by_index(begin = var_13753_begin_0, end = var_13753_end_0, end_mask = var_13753_end_mask_0, x = var_13428_cast_fp16)[name = tensor("op_13753_cast_fp16")]; + tensor var_13760_begin_0 = const()[name = tensor("op_13760_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13760_end_0 = const()[name = tensor("op_13760_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13760_end_mask_0 = const()[name = tensor("op_13760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13760_cast_fp16 = slice_by_index(begin = var_13760_begin_0, end = var_13760_end_0, end_mask = var_13760_end_mask_0, x = var_13428_cast_fp16)[name = tensor("op_13760_cast_fp16")]; + tensor var_13767_begin_0 = const()[name = tensor("op_13767_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13767_end_0 = const()[name = tensor("op_13767_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13767_end_mask_0 = const()[name = tensor("op_13767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13767_cast_fp16 = slice_by_index(begin = var_13767_begin_0, end = var_13767_end_0, end_mask = var_13767_end_mask_0, x = var_13428_cast_fp16)[name = tensor("op_13767_cast_fp16")]; + tensor var_13774_begin_0 = const()[name = tensor("op_13774_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13774_end_0 = const()[name = tensor("op_13774_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13774_end_mask_0 = const()[name = tensor("op_13774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13774_cast_fp16 = slice_by_index(begin = var_13774_begin_0, end = var_13774_end_0, end_mask = var_13774_end_mask_0, x = var_13428_cast_fp16)[name = tensor("op_13774_cast_fp16")]; + tensor var_13781_begin_0 = const()[name = tensor("op_13781_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13781_end_0 = const()[name = tensor("op_13781_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13781_end_mask_0 = const()[name = tensor("op_13781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13781_cast_fp16 = slice_by_index(begin = var_13781_begin_0, end = var_13781_end_0, end_mask = var_13781_end_mask_0, x = var_13432_cast_fp16)[name = tensor("op_13781_cast_fp16")]; + tensor var_13788_begin_0 = const()[name = tensor("op_13788_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13788_end_0 = const()[name = tensor("op_13788_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13788_end_mask_0 = const()[name = tensor("op_13788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13788_cast_fp16 = slice_by_index(begin = var_13788_begin_0, end = var_13788_end_0, end_mask = var_13788_end_mask_0, x = var_13432_cast_fp16)[name = tensor("op_13788_cast_fp16")]; + tensor var_13795_begin_0 = const()[name = tensor("op_13795_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13795_end_0 = const()[name = tensor("op_13795_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13795_end_mask_0 = const()[name = tensor("op_13795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13795_cast_fp16 = slice_by_index(begin = var_13795_begin_0, end = var_13795_end_0, end_mask = var_13795_end_mask_0, x = var_13432_cast_fp16)[name = tensor("op_13795_cast_fp16")]; + tensor var_13802_begin_0 = const()[name = tensor("op_13802_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13802_end_0 = const()[name = tensor("op_13802_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13802_end_mask_0 = const()[name = tensor("op_13802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13802_cast_fp16 = slice_by_index(begin = var_13802_begin_0, end = var_13802_end_0, end_mask = var_13802_end_mask_0, x = var_13432_cast_fp16)[name = tensor("op_13802_cast_fp16")]; + tensor var_13809_begin_0 = const()[name = tensor("op_13809_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13809_end_0 = const()[name = tensor("op_13809_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13809_end_mask_0 = const()[name = tensor("op_13809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13809_cast_fp16 = slice_by_index(begin = var_13809_begin_0, end = var_13809_end_0, end_mask = var_13809_end_mask_0, x = var_13436_cast_fp16)[name = tensor("op_13809_cast_fp16")]; + tensor var_13816_begin_0 = const()[name = tensor("op_13816_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13816_end_0 = const()[name = tensor("op_13816_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13816_end_mask_0 = const()[name = tensor("op_13816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13816_cast_fp16 = slice_by_index(begin = var_13816_begin_0, end = var_13816_end_0, end_mask = var_13816_end_mask_0, x = var_13436_cast_fp16)[name = tensor("op_13816_cast_fp16")]; + tensor var_13823_begin_0 = const()[name = tensor("op_13823_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13823_end_0 = const()[name = tensor("op_13823_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13823_end_mask_0 = const()[name = tensor("op_13823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13823_cast_fp16 = slice_by_index(begin = var_13823_begin_0, end = var_13823_end_0, end_mask = var_13823_end_mask_0, x = var_13436_cast_fp16)[name = tensor("op_13823_cast_fp16")]; + tensor var_13830_begin_0 = const()[name = tensor("op_13830_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13830_end_0 = const()[name = tensor("op_13830_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13830_end_mask_0 = const()[name = tensor("op_13830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13830_cast_fp16 = slice_by_index(begin = var_13830_begin_0, end = var_13830_end_0, end_mask = var_13830_end_mask_0, x = var_13436_cast_fp16)[name = tensor("op_13830_cast_fp16")]; + tensor var_13837_begin_0 = const()[name = tensor("op_13837_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13837_end_0 = const()[name = tensor("op_13837_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13837_end_mask_0 = const()[name = tensor("op_13837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13837_cast_fp16 = slice_by_index(begin = var_13837_begin_0, end = var_13837_end_0, end_mask = var_13837_end_mask_0, x = var_13440_cast_fp16)[name = tensor("op_13837_cast_fp16")]; + tensor var_13844_begin_0 = const()[name = tensor("op_13844_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13844_end_0 = const()[name = tensor("op_13844_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13844_end_mask_0 = const()[name = tensor("op_13844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13844_cast_fp16 = slice_by_index(begin = var_13844_begin_0, end = var_13844_end_0, end_mask = var_13844_end_mask_0, x = var_13440_cast_fp16)[name = tensor("op_13844_cast_fp16")]; + tensor var_13851_begin_0 = const()[name = tensor("op_13851_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13851_end_0 = const()[name = tensor("op_13851_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13851_end_mask_0 = const()[name = tensor("op_13851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13851_cast_fp16 = slice_by_index(begin = var_13851_begin_0, end = var_13851_end_0, end_mask = var_13851_end_mask_0, x = var_13440_cast_fp16)[name = tensor("op_13851_cast_fp16")]; + tensor var_13858_begin_0 = const()[name = tensor("op_13858_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13858_end_0 = const()[name = tensor("op_13858_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13858_end_mask_0 = const()[name = tensor("op_13858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13858_cast_fp16 = slice_by_index(begin = var_13858_begin_0, end = var_13858_end_0, end_mask = var_13858_end_mask_0, x = var_13440_cast_fp16)[name = tensor("op_13858_cast_fp16")]; + tensor var_13865_begin_0 = const()[name = tensor("op_13865_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13865_end_0 = const()[name = tensor("op_13865_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13865_end_mask_0 = const()[name = tensor("op_13865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13865_cast_fp16 = slice_by_index(begin = var_13865_begin_0, end = var_13865_end_0, end_mask = var_13865_end_mask_0, x = var_13444_cast_fp16)[name = tensor("op_13865_cast_fp16")]; + tensor var_13872_begin_0 = const()[name = tensor("op_13872_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13872_end_0 = const()[name = tensor("op_13872_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13872_end_mask_0 = const()[name = tensor("op_13872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13872_cast_fp16 = slice_by_index(begin = var_13872_begin_0, end = var_13872_end_0, end_mask = var_13872_end_mask_0, x = var_13444_cast_fp16)[name = tensor("op_13872_cast_fp16")]; + tensor var_13879_begin_0 = const()[name = tensor("op_13879_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13879_end_0 = const()[name = tensor("op_13879_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13879_end_mask_0 = const()[name = tensor("op_13879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13879_cast_fp16 = slice_by_index(begin = var_13879_begin_0, end = var_13879_end_0, end_mask = var_13879_end_mask_0, x = var_13444_cast_fp16)[name = tensor("op_13879_cast_fp16")]; + tensor var_13886_begin_0 = const()[name = tensor("op_13886_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13886_end_0 = const()[name = tensor("op_13886_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13886_end_mask_0 = const()[name = tensor("op_13886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13886_cast_fp16 = slice_by_index(begin = var_13886_begin_0, end = var_13886_end_0, end_mask = var_13886_end_mask_0, x = var_13444_cast_fp16)[name = tensor("op_13886_cast_fp16")]; + tensor var_13893_begin_0 = const()[name = tensor("op_13893_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13893_end_0 = const()[name = tensor("op_13893_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13893_end_mask_0 = const()[name = tensor("op_13893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13893_cast_fp16 = slice_by_index(begin = var_13893_begin_0, end = var_13893_end_0, end_mask = var_13893_end_mask_0, x = var_13448_cast_fp16)[name = tensor("op_13893_cast_fp16")]; + tensor var_13900_begin_0 = const()[name = tensor("op_13900_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13900_end_0 = const()[name = tensor("op_13900_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13900_end_mask_0 = const()[name = tensor("op_13900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13900_cast_fp16 = slice_by_index(begin = var_13900_begin_0, end = var_13900_end_0, end_mask = var_13900_end_mask_0, x = var_13448_cast_fp16)[name = tensor("op_13900_cast_fp16")]; + tensor var_13907_begin_0 = const()[name = tensor("op_13907_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13907_end_0 = const()[name = tensor("op_13907_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13907_end_mask_0 = const()[name = tensor("op_13907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13907_cast_fp16 = slice_by_index(begin = var_13907_begin_0, end = var_13907_end_0, end_mask = var_13907_end_mask_0, x = var_13448_cast_fp16)[name = tensor("op_13907_cast_fp16")]; + tensor var_13914_begin_0 = const()[name = tensor("op_13914_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13914_end_0 = const()[name = tensor("op_13914_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13914_end_mask_0 = const()[name = tensor("op_13914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13914_cast_fp16 = slice_by_index(begin = var_13914_begin_0, end = var_13914_end_0, end_mask = var_13914_end_mask_0, x = var_13448_cast_fp16)[name = tensor("op_13914_cast_fp16")]; + tensor var_13921_begin_0 = const()[name = tensor("op_13921_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13921_end_0 = const()[name = tensor("op_13921_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13921_end_mask_0 = const()[name = tensor("op_13921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13921_cast_fp16 = slice_by_index(begin = var_13921_begin_0, end = var_13921_end_0, end_mask = var_13921_end_mask_0, x = var_13452_cast_fp16)[name = tensor("op_13921_cast_fp16")]; + tensor var_13928_begin_0 = const()[name = tensor("op_13928_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13928_end_0 = const()[name = tensor("op_13928_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13928_end_mask_0 = const()[name = tensor("op_13928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13928_cast_fp16 = slice_by_index(begin = var_13928_begin_0, end = var_13928_end_0, end_mask = var_13928_end_mask_0, x = var_13452_cast_fp16)[name = tensor("op_13928_cast_fp16")]; + tensor var_13935_begin_0 = const()[name = tensor("op_13935_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13935_end_0 = const()[name = tensor("op_13935_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13935_end_mask_0 = const()[name = tensor("op_13935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13935_cast_fp16 = slice_by_index(begin = var_13935_begin_0, end = var_13935_end_0, end_mask = var_13935_end_mask_0, x = var_13452_cast_fp16)[name = tensor("op_13935_cast_fp16")]; + tensor var_13942_begin_0 = const()[name = tensor("op_13942_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13942_end_0 = const()[name = tensor("op_13942_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13942_end_mask_0 = const()[name = tensor("op_13942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13942_cast_fp16 = slice_by_index(begin = var_13942_begin_0, end = var_13942_end_0, end_mask = var_13942_end_mask_0, x = var_13452_cast_fp16)[name = tensor("op_13942_cast_fp16")]; + tensor var_13949_begin_0 = const()[name = tensor("op_13949_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13949_end_0 = const()[name = tensor("op_13949_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13949_end_mask_0 = const()[name = tensor("op_13949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13949_cast_fp16 = slice_by_index(begin = var_13949_begin_0, end = var_13949_end_0, end_mask = var_13949_end_mask_0, x = var_13456_cast_fp16)[name = tensor("op_13949_cast_fp16")]; + tensor var_13956_begin_0 = const()[name = tensor("op_13956_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13956_end_0 = const()[name = tensor("op_13956_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13956_end_mask_0 = const()[name = tensor("op_13956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13956_cast_fp16 = slice_by_index(begin = var_13956_begin_0, end = var_13956_end_0, end_mask = var_13956_end_mask_0, x = var_13456_cast_fp16)[name = tensor("op_13956_cast_fp16")]; + tensor var_13963_begin_0 = const()[name = tensor("op_13963_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13963_end_0 = const()[name = tensor("op_13963_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13963_end_mask_0 = const()[name = tensor("op_13963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13963_cast_fp16 = slice_by_index(begin = var_13963_begin_0, end = var_13963_end_0, end_mask = var_13963_end_mask_0, x = var_13456_cast_fp16)[name = tensor("op_13963_cast_fp16")]; + tensor var_13970_begin_0 = const()[name = tensor("op_13970_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13970_end_0 = const()[name = tensor("op_13970_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13970_end_mask_0 = const()[name = tensor("op_13970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13970_cast_fp16 = slice_by_index(begin = var_13970_begin_0, end = var_13970_end_0, end_mask = var_13970_end_mask_0, x = var_13456_cast_fp16)[name = tensor("op_13970_cast_fp16")]; + tensor var_13977_begin_0 = const()[name = tensor("op_13977_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_13977_end_0 = const()[name = tensor("op_13977_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_13977_end_mask_0 = const()[name = tensor("op_13977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13977_cast_fp16 = slice_by_index(begin = var_13977_begin_0, end = var_13977_end_0, end_mask = var_13977_end_mask_0, x = var_13460_cast_fp16)[name = tensor("op_13977_cast_fp16")]; + tensor var_13984_begin_0 = const()[name = tensor("op_13984_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_13984_end_0 = const()[name = tensor("op_13984_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_13984_end_mask_0 = const()[name = tensor("op_13984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13984_cast_fp16 = slice_by_index(begin = var_13984_begin_0, end = var_13984_end_0, end_mask = var_13984_end_mask_0, x = var_13460_cast_fp16)[name = tensor("op_13984_cast_fp16")]; + tensor var_13991_begin_0 = const()[name = tensor("op_13991_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_13991_end_0 = const()[name = tensor("op_13991_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_13991_end_mask_0 = const()[name = tensor("op_13991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13991_cast_fp16 = slice_by_index(begin = var_13991_begin_0, end = var_13991_end_0, end_mask = var_13991_end_mask_0, x = var_13460_cast_fp16)[name = tensor("op_13991_cast_fp16")]; + tensor var_13998_begin_0 = const()[name = tensor("op_13998_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_13998_end_0 = const()[name = tensor("op_13998_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_13998_end_mask_0 = const()[name = tensor("op_13998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_13998_cast_fp16 = slice_by_index(begin = var_13998_begin_0, end = var_13998_end_0, end_mask = var_13998_end_mask_0, x = var_13460_cast_fp16)[name = tensor("op_13998_cast_fp16")]; + tensor var_14005_begin_0 = const()[name = tensor("op_14005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14005_end_0 = const()[name = tensor("op_14005_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_14005_end_mask_0 = const()[name = tensor("op_14005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14005_cast_fp16 = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = var_13464_cast_fp16)[name = tensor("op_14005_cast_fp16")]; + tensor var_14012_begin_0 = const()[name = tensor("op_14012_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_14012_end_0 = const()[name = tensor("op_14012_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_14012_end_mask_0 = const()[name = tensor("op_14012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14012_cast_fp16 = slice_by_index(begin = var_14012_begin_0, end = var_14012_end_0, end_mask = var_14012_end_mask_0, x = var_13464_cast_fp16)[name = tensor("op_14012_cast_fp16")]; + tensor var_14019_begin_0 = const()[name = tensor("op_14019_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_14019_end_0 = const()[name = tensor("op_14019_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_14019_end_mask_0 = const()[name = tensor("op_14019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14019_cast_fp16 = slice_by_index(begin = var_14019_begin_0, end = var_14019_end_0, end_mask = var_14019_end_mask_0, x = var_13464_cast_fp16)[name = tensor("op_14019_cast_fp16")]; + tensor var_14026_begin_0 = const()[name = tensor("op_14026_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_14026_end_0 = const()[name = tensor("op_14026_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14026_end_mask_0 = const()[name = tensor("op_14026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14026_cast_fp16 = slice_by_index(begin = var_14026_begin_0, end = var_14026_end_0, end_mask = var_14026_end_mask_0, x = var_13464_cast_fp16)[name = tensor("op_14026_cast_fp16")]; + tensor k_17_perm_0 = const()[name = tensor("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_14031_begin_0 = const()[name = tensor("op_14031_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14031_end_0 = const()[name = tensor("op_14031_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_14031_end_mask_0 = const()[name = tensor("op_14031_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_23 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = tensor("transpose_23")]; + tensor var_14031_cast_fp16 = slice_by_index(begin = var_14031_begin_0, end = var_14031_end_0, end_mask = var_14031_end_mask_0, x = transpose_23)[name = tensor("op_14031_cast_fp16")]; + tensor var_14035_begin_0 = const()[name = tensor("op_14035_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_14035_end_0 = const()[name = tensor("op_14035_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_14035_end_mask_0 = const()[name = tensor("op_14035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14035_cast_fp16 = slice_by_index(begin = var_14035_begin_0, end = var_14035_end_0, end_mask = var_14035_end_mask_0, x = transpose_23)[name = tensor("op_14035_cast_fp16")]; + tensor var_14039_begin_0 = const()[name = tensor("op_14039_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_14039_end_0 = const()[name = tensor("op_14039_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_14039_end_mask_0 = const()[name = tensor("op_14039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14039_cast_fp16 = slice_by_index(begin = var_14039_begin_0, end = var_14039_end_0, end_mask = var_14039_end_mask_0, x = transpose_23)[name = tensor("op_14039_cast_fp16")]; + tensor var_14043_begin_0 = const()[name = tensor("op_14043_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_14043_end_0 = const()[name = tensor("op_14043_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_14043_end_mask_0 = const()[name = tensor("op_14043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14043_cast_fp16 = slice_by_index(begin = var_14043_begin_0, end = var_14043_end_0, end_mask = var_14043_end_mask_0, x = transpose_23)[name = tensor("op_14043_cast_fp16")]; + tensor var_14047_begin_0 = const()[name = tensor("op_14047_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_14047_end_0 = const()[name = tensor("op_14047_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_14047_end_mask_0 = const()[name = tensor("op_14047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14047_cast_fp16 = slice_by_index(begin = var_14047_begin_0, end = var_14047_end_0, end_mask = var_14047_end_mask_0, x = transpose_23)[name = tensor("op_14047_cast_fp16")]; + tensor var_14051_begin_0 = const()[name = tensor("op_14051_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_14051_end_0 = const()[name = tensor("op_14051_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_14051_end_mask_0 = const()[name = tensor("op_14051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14051_cast_fp16 = slice_by_index(begin = var_14051_begin_0, end = var_14051_end_0, end_mask = var_14051_end_mask_0, x = transpose_23)[name = tensor("op_14051_cast_fp16")]; + tensor var_14055_begin_0 = const()[name = tensor("op_14055_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_14055_end_0 = const()[name = tensor("op_14055_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_14055_end_mask_0 = const()[name = tensor("op_14055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14055_cast_fp16 = slice_by_index(begin = var_14055_begin_0, end = var_14055_end_0, end_mask = var_14055_end_mask_0, x = transpose_23)[name = tensor("op_14055_cast_fp16")]; + tensor var_14059_begin_0 = const()[name = tensor("op_14059_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_14059_end_0 = const()[name = tensor("op_14059_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_14059_end_mask_0 = const()[name = tensor("op_14059_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14059_cast_fp16 = slice_by_index(begin = var_14059_begin_0, end = var_14059_end_0, end_mask = var_14059_end_mask_0, x = transpose_23)[name = tensor("op_14059_cast_fp16")]; + tensor var_14063_begin_0 = const()[name = tensor("op_14063_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_14063_end_0 = const()[name = tensor("op_14063_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_14063_end_mask_0 = const()[name = tensor("op_14063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14063_cast_fp16 = slice_by_index(begin = var_14063_begin_0, end = var_14063_end_0, end_mask = var_14063_end_mask_0, x = transpose_23)[name = tensor("op_14063_cast_fp16")]; + tensor var_14067_begin_0 = const()[name = tensor("op_14067_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_14067_end_0 = const()[name = tensor("op_14067_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_14067_end_mask_0 = const()[name = tensor("op_14067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14067_cast_fp16 = slice_by_index(begin = var_14067_begin_0, end = var_14067_end_0, end_mask = var_14067_end_mask_0, x = transpose_23)[name = tensor("op_14067_cast_fp16")]; + tensor var_14071_begin_0 = const()[name = tensor("op_14071_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_14071_end_0 = const()[name = tensor("op_14071_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_14071_end_mask_0 = const()[name = tensor("op_14071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14071_cast_fp16 = slice_by_index(begin = var_14071_begin_0, end = var_14071_end_0, end_mask = var_14071_end_mask_0, x = transpose_23)[name = tensor("op_14071_cast_fp16")]; + tensor var_14075_begin_0 = const()[name = tensor("op_14075_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_14075_end_0 = const()[name = tensor("op_14075_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_14075_end_mask_0 = const()[name = tensor("op_14075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14075_cast_fp16 = slice_by_index(begin = var_14075_begin_0, end = var_14075_end_0, end_mask = var_14075_end_mask_0, x = transpose_23)[name = tensor("op_14075_cast_fp16")]; + tensor var_14079_begin_0 = const()[name = tensor("op_14079_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_14079_end_0 = const()[name = tensor("op_14079_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_14079_end_mask_0 = const()[name = tensor("op_14079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14079_cast_fp16 = slice_by_index(begin = var_14079_begin_0, end = var_14079_end_0, end_mask = var_14079_end_mask_0, x = transpose_23)[name = tensor("op_14079_cast_fp16")]; + tensor var_14083_begin_0 = const()[name = tensor("op_14083_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_14083_end_0 = const()[name = tensor("op_14083_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_14083_end_mask_0 = const()[name = tensor("op_14083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14083_cast_fp16 = slice_by_index(begin = var_14083_begin_0, end = var_14083_end_0, end_mask = var_14083_end_mask_0, x = transpose_23)[name = tensor("op_14083_cast_fp16")]; + tensor var_14087_begin_0 = const()[name = tensor("op_14087_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_14087_end_0 = const()[name = tensor("op_14087_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_14087_end_mask_0 = const()[name = tensor("op_14087_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14087_cast_fp16 = slice_by_index(begin = var_14087_begin_0, end = var_14087_end_0, end_mask = var_14087_end_mask_0, x = transpose_23)[name = tensor("op_14087_cast_fp16")]; + tensor var_14091_begin_0 = const()[name = tensor("op_14091_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_14091_end_0 = const()[name = tensor("op_14091_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_14091_end_mask_0 = const()[name = tensor("op_14091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14091_cast_fp16 = slice_by_index(begin = var_14091_begin_0, end = var_14091_end_0, end_mask = var_14091_end_mask_0, x = transpose_23)[name = tensor("op_14091_cast_fp16")]; + tensor var_14095_begin_0 = const()[name = tensor("op_14095_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_14095_end_0 = const()[name = tensor("op_14095_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_14095_end_mask_0 = const()[name = tensor("op_14095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14095_cast_fp16 = slice_by_index(begin = var_14095_begin_0, end = var_14095_end_0, end_mask = var_14095_end_mask_0, x = transpose_23)[name = tensor("op_14095_cast_fp16")]; + tensor var_14099_begin_0 = const()[name = tensor("op_14099_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_14099_end_0 = const()[name = tensor("op_14099_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_14099_end_mask_0 = const()[name = tensor("op_14099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14099_cast_fp16 = slice_by_index(begin = var_14099_begin_0, end = var_14099_end_0, end_mask = var_14099_end_mask_0, x = transpose_23)[name = tensor("op_14099_cast_fp16")]; + tensor var_14103_begin_0 = const()[name = tensor("op_14103_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_14103_end_0 = const()[name = tensor("op_14103_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_14103_end_mask_0 = const()[name = tensor("op_14103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14103_cast_fp16 = slice_by_index(begin = var_14103_begin_0, end = var_14103_end_0, end_mask = var_14103_end_mask_0, x = transpose_23)[name = tensor("op_14103_cast_fp16")]; + tensor var_14107_begin_0 = const()[name = tensor("op_14107_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_14107_end_0 = const()[name = tensor("op_14107_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_14107_end_mask_0 = const()[name = tensor("op_14107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_14107_cast_fp16 = slice_by_index(begin = var_14107_begin_0, end = var_14107_end_0, end_mask = var_14107_end_mask_0, x = transpose_23)[name = tensor("op_14107_cast_fp16")]; + tensor var_14109_begin_0 = const()[name = tensor("op_14109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_14109_end_0 = const()[name = tensor("op_14109_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_14109_end_mask_0 = const()[name = tensor("op_14109_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14109_cast_fp16 = slice_by_index(begin = var_14109_begin_0, end = var_14109_end_0, end_mask = var_14109_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14109_cast_fp16")]; + tensor var_14113_begin_0 = const()[name = tensor("op_14113_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_14113_end_0 = const()[name = tensor("op_14113_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_14113_end_mask_0 = const()[name = tensor("op_14113_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14113_cast_fp16 = slice_by_index(begin = var_14113_begin_0, end = var_14113_end_0, end_mask = var_14113_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14113_cast_fp16")]; + tensor var_14117_begin_0 = const()[name = tensor("op_14117_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_14117_end_0 = const()[name = tensor("op_14117_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_14117_end_mask_0 = const()[name = tensor("op_14117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14117_cast_fp16 = slice_by_index(begin = var_14117_begin_0, end = var_14117_end_0, end_mask = var_14117_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14117_cast_fp16")]; + tensor var_14121_begin_0 = const()[name = tensor("op_14121_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_14121_end_0 = const()[name = tensor("op_14121_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_14121_end_mask_0 = const()[name = tensor("op_14121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14121_cast_fp16 = slice_by_index(begin = var_14121_begin_0, end = var_14121_end_0, end_mask = var_14121_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14121_cast_fp16")]; + tensor var_14125_begin_0 = const()[name = tensor("op_14125_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_14125_end_0 = const()[name = tensor("op_14125_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_14125_end_mask_0 = const()[name = tensor("op_14125_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14125_cast_fp16 = slice_by_index(begin = var_14125_begin_0, end = var_14125_end_0, end_mask = var_14125_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14125_cast_fp16")]; + tensor var_14129_begin_0 = const()[name = tensor("op_14129_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_14129_end_0 = const()[name = tensor("op_14129_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_14129_end_mask_0 = const()[name = tensor("op_14129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14129_cast_fp16 = slice_by_index(begin = var_14129_begin_0, end = var_14129_end_0, end_mask = var_14129_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14129_cast_fp16")]; + tensor var_14133_begin_0 = const()[name = tensor("op_14133_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_14133_end_0 = const()[name = tensor("op_14133_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_14133_end_mask_0 = const()[name = tensor("op_14133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14133_cast_fp16 = slice_by_index(begin = var_14133_begin_0, end = var_14133_end_0, end_mask = var_14133_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14133_cast_fp16")]; + tensor var_14137_begin_0 = const()[name = tensor("op_14137_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_14137_end_0 = const()[name = tensor("op_14137_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_14137_end_mask_0 = const()[name = tensor("op_14137_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14137_cast_fp16 = slice_by_index(begin = var_14137_begin_0, end = var_14137_end_0, end_mask = var_14137_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14137_cast_fp16")]; + tensor var_14141_begin_0 = const()[name = tensor("op_14141_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_14141_end_0 = const()[name = tensor("op_14141_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_14141_end_mask_0 = const()[name = tensor("op_14141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14141_cast_fp16 = slice_by_index(begin = var_14141_begin_0, end = var_14141_end_0, end_mask = var_14141_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14141_cast_fp16")]; + tensor var_14145_begin_0 = const()[name = tensor("op_14145_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_14145_end_0 = const()[name = tensor("op_14145_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_14145_end_mask_0 = const()[name = tensor("op_14145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14145_cast_fp16 = slice_by_index(begin = var_14145_begin_0, end = var_14145_end_0, end_mask = var_14145_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14145_cast_fp16")]; + tensor var_14149_begin_0 = const()[name = tensor("op_14149_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_14149_end_0 = const()[name = tensor("op_14149_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_14149_end_mask_0 = const()[name = tensor("op_14149_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14149_cast_fp16 = slice_by_index(begin = var_14149_begin_0, end = var_14149_end_0, end_mask = var_14149_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14149_cast_fp16")]; + tensor var_14153_begin_0 = const()[name = tensor("op_14153_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_14153_end_0 = const()[name = tensor("op_14153_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_14153_end_mask_0 = const()[name = tensor("op_14153_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14153_cast_fp16 = slice_by_index(begin = var_14153_begin_0, end = var_14153_end_0, end_mask = var_14153_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14153_cast_fp16")]; + tensor var_14157_begin_0 = const()[name = tensor("op_14157_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_14157_end_0 = const()[name = tensor("op_14157_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_14157_end_mask_0 = const()[name = tensor("op_14157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14157_cast_fp16 = slice_by_index(begin = var_14157_begin_0, end = var_14157_end_0, end_mask = var_14157_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14157_cast_fp16")]; + tensor var_14161_begin_0 = const()[name = tensor("op_14161_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_14161_end_0 = const()[name = tensor("op_14161_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_14161_end_mask_0 = const()[name = tensor("op_14161_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14161_cast_fp16 = slice_by_index(begin = var_14161_begin_0, end = var_14161_end_0, end_mask = var_14161_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14161_cast_fp16")]; + tensor var_14165_begin_0 = const()[name = tensor("op_14165_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_14165_end_0 = const()[name = tensor("op_14165_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_14165_end_mask_0 = const()[name = tensor("op_14165_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14165_cast_fp16 = slice_by_index(begin = var_14165_begin_0, end = var_14165_end_0, end_mask = var_14165_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14165_cast_fp16")]; + tensor var_14169_begin_0 = const()[name = tensor("op_14169_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_14169_end_0 = const()[name = tensor("op_14169_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_14169_end_mask_0 = const()[name = tensor("op_14169_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14169_cast_fp16 = slice_by_index(begin = var_14169_begin_0, end = var_14169_end_0, end_mask = var_14169_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14169_cast_fp16")]; + tensor var_14173_begin_0 = const()[name = tensor("op_14173_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_14173_end_0 = const()[name = tensor("op_14173_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_14173_end_mask_0 = const()[name = tensor("op_14173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14173_cast_fp16 = slice_by_index(begin = var_14173_begin_0, end = var_14173_end_0, end_mask = var_14173_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14173_cast_fp16")]; + tensor var_14177_begin_0 = const()[name = tensor("op_14177_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_14177_end_0 = const()[name = tensor("op_14177_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_14177_end_mask_0 = const()[name = tensor("op_14177_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14177_cast_fp16 = slice_by_index(begin = var_14177_begin_0, end = var_14177_end_0, end_mask = var_14177_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14177_cast_fp16")]; + tensor var_14181_begin_0 = const()[name = tensor("op_14181_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_14181_end_0 = const()[name = tensor("op_14181_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_14181_end_mask_0 = const()[name = tensor("op_14181_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14181_cast_fp16 = slice_by_index(begin = var_14181_begin_0, end = var_14181_end_0, end_mask = var_14181_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14181_cast_fp16")]; + tensor var_14185_begin_0 = const()[name = tensor("op_14185_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_14185_end_0 = const()[name = tensor("op_14185_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_14185_end_mask_0 = const()[name = tensor("op_14185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_14185_cast_fp16 = slice_by_index(begin = var_14185_begin_0, end = var_14185_end_0, end_mask = var_14185_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_14185_cast_fp16")]; + tensor var_14189_equation_0 = const()[name = tensor("op_14189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14189_cast_fp16 = einsum(equation = var_14189_equation_0, values = (var_14031_cast_fp16, var_13473_cast_fp16))[name = tensor("op_14189_cast_fp16")]; + tensor var_14190_to_fp16 = const()[name = tensor("op_14190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1281_cast_fp16 = mul(x = var_14189_cast_fp16, y = var_14190_to_fp16)[name = tensor("aw_chunk_1281_cast_fp16")]; + tensor var_14193_equation_0 = const()[name = tensor("op_14193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14193_cast_fp16 = einsum(equation = var_14193_equation_0, values = (var_14031_cast_fp16, var_13480_cast_fp16))[name = tensor("op_14193_cast_fp16")]; + tensor var_14194_to_fp16 = const()[name = tensor("op_14194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1283_cast_fp16 = mul(x = var_14193_cast_fp16, y = var_14194_to_fp16)[name = tensor("aw_chunk_1283_cast_fp16")]; + tensor var_14197_equation_0 = const()[name = tensor("op_14197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14197_cast_fp16 = einsum(equation = var_14197_equation_0, values = (var_14031_cast_fp16, var_13487_cast_fp16))[name = tensor("op_14197_cast_fp16")]; + tensor var_14198_to_fp16 = const()[name = tensor("op_14198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1285_cast_fp16 = mul(x = var_14197_cast_fp16, y = var_14198_to_fp16)[name = tensor("aw_chunk_1285_cast_fp16")]; + tensor var_14201_equation_0 = const()[name = tensor("op_14201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14201_cast_fp16 = einsum(equation = var_14201_equation_0, values = (var_14031_cast_fp16, var_13494_cast_fp16))[name = tensor("op_14201_cast_fp16")]; + tensor var_14202_to_fp16 = const()[name = tensor("op_14202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1287_cast_fp16 = mul(x = var_14201_cast_fp16, y = var_14202_to_fp16)[name = tensor("aw_chunk_1287_cast_fp16")]; + tensor var_14205_equation_0 = const()[name = tensor("op_14205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14205_cast_fp16 = einsum(equation = var_14205_equation_0, values = (var_14035_cast_fp16, var_13501_cast_fp16))[name = tensor("op_14205_cast_fp16")]; + tensor var_14206_to_fp16 = const()[name = tensor("op_14206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1289_cast_fp16 = mul(x = var_14205_cast_fp16, y = var_14206_to_fp16)[name = tensor("aw_chunk_1289_cast_fp16")]; + tensor var_14209_equation_0 = const()[name = tensor("op_14209_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14209_cast_fp16 = einsum(equation = var_14209_equation_0, values = (var_14035_cast_fp16, var_13508_cast_fp16))[name = tensor("op_14209_cast_fp16")]; + tensor var_14210_to_fp16 = const()[name = tensor("op_14210_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1291_cast_fp16 = mul(x = var_14209_cast_fp16, y = var_14210_to_fp16)[name = tensor("aw_chunk_1291_cast_fp16")]; + tensor var_14213_equation_0 = const()[name = tensor("op_14213_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14213_cast_fp16 = einsum(equation = var_14213_equation_0, values = (var_14035_cast_fp16, var_13515_cast_fp16))[name = tensor("op_14213_cast_fp16")]; + tensor var_14214_to_fp16 = const()[name = tensor("op_14214_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1293_cast_fp16 = mul(x = var_14213_cast_fp16, y = var_14214_to_fp16)[name = tensor("aw_chunk_1293_cast_fp16")]; + tensor var_14217_equation_0 = const()[name = tensor("op_14217_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14217_cast_fp16 = einsum(equation = var_14217_equation_0, values = (var_14035_cast_fp16, var_13522_cast_fp16))[name = tensor("op_14217_cast_fp16")]; + tensor var_14218_to_fp16 = const()[name = tensor("op_14218_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1295_cast_fp16 = mul(x = var_14217_cast_fp16, y = var_14218_to_fp16)[name = tensor("aw_chunk_1295_cast_fp16")]; + tensor var_14221_equation_0 = const()[name = tensor("op_14221_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14221_cast_fp16 = einsum(equation = var_14221_equation_0, values = (var_14039_cast_fp16, var_13529_cast_fp16))[name = tensor("op_14221_cast_fp16")]; + tensor var_14222_to_fp16 = const()[name = tensor("op_14222_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1297_cast_fp16 = mul(x = var_14221_cast_fp16, y = var_14222_to_fp16)[name = tensor("aw_chunk_1297_cast_fp16")]; + tensor var_14225_equation_0 = const()[name = tensor("op_14225_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14225_cast_fp16 = einsum(equation = var_14225_equation_0, values = (var_14039_cast_fp16, var_13536_cast_fp16))[name = tensor("op_14225_cast_fp16")]; + tensor var_14226_to_fp16 = const()[name = tensor("op_14226_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1299_cast_fp16 = mul(x = var_14225_cast_fp16, y = var_14226_to_fp16)[name = tensor("aw_chunk_1299_cast_fp16")]; + tensor var_14229_equation_0 = const()[name = tensor("op_14229_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14229_cast_fp16 = einsum(equation = var_14229_equation_0, values = (var_14039_cast_fp16, var_13543_cast_fp16))[name = tensor("op_14229_cast_fp16")]; + tensor var_14230_to_fp16 = const()[name = tensor("op_14230_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1301_cast_fp16 = mul(x = var_14229_cast_fp16, y = var_14230_to_fp16)[name = tensor("aw_chunk_1301_cast_fp16")]; + tensor var_14233_equation_0 = const()[name = tensor("op_14233_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14233_cast_fp16 = einsum(equation = var_14233_equation_0, values = (var_14039_cast_fp16, var_13550_cast_fp16))[name = tensor("op_14233_cast_fp16")]; + tensor var_14234_to_fp16 = const()[name = tensor("op_14234_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1303_cast_fp16 = mul(x = var_14233_cast_fp16, y = var_14234_to_fp16)[name = tensor("aw_chunk_1303_cast_fp16")]; + tensor var_14237_equation_0 = const()[name = tensor("op_14237_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14237_cast_fp16 = einsum(equation = var_14237_equation_0, values = (var_14043_cast_fp16, var_13557_cast_fp16))[name = tensor("op_14237_cast_fp16")]; + tensor var_14238_to_fp16 = const()[name = tensor("op_14238_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1305_cast_fp16 = mul(x = var_14237_cast_fp16, y = var_14238_to_fp16)[name = tensor("aw_chunk_1305_cast_fp16")]; + tensor var_14241_equation_0 = const()[name = tensor("op_14241_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14241_cast_fp16 = einsum(equation = var_14241_equation_0, values = (var_14043_cast_fp16, var_13564_cast_fp16))[name = tensor("op_14241_cast_fp16")]; + tensor var_14242_to_fp16 = const()[name = tensor("op_14242_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1307_cast_fp16 = mul(x = var_14241_cast_fp16, y = var_14242_to_fp16)[name = tensor("aw_chunk_1307_cast_fp16")]; + tensor var_14245_equation_0 = const()[name = tensor("op_14245_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14245_cast_fp16 = einsum(equation = var_14245_equation_0, values = (var_14043_cast_fp16, var_13571_cast_fp16))[name = tensor("op_14245_cast_fp16")]; + tensor var_14246_to_fp16 = const()[name = tensor("op_14246_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1309_cast_fp16 = mul(x = var_14245_cast_fp16, y = var_14246_to_fp16)[name = tensor("aw_chunk_1309_cast_fp16")]; + tensor var_14249_equation_0 = const()[name = tensor("op_14249_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14249_cast_fp16 = einsum(equation = var_14249_equation_0, values = (var_14043_cast_fp16, var_13578_cast_fp16))[name = tensor("op_14249_cast_fp16")]; + tensor var_14250_to_fp16 = const()[name = tensor("op_14250_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1311_cast_fp16 = mul(x = var_14249_cast_fp16, y = var_14250_to_fp16)[name = tensor("aw_chunk_1311_cast_fp16")]; + tensor var_14253_equation_0 = const()[name = tensor("op_14253_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14253_cast_fp16 = einsum(equation = var_14253_equation_0, values = (var_14047_cast_fp16, var_13585_cast_fp16))[name = tensor("op_14253_cast_fp16")]; + tensor var_14254_to_fp16 = const()[name = tensor("op_14254_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1313_cast_fp16 = mul(x = var_14253_cast_fp16, y = var_14254_to_fp16)[name = tensor("aw_chunk_1313_cast_fp16")]; + tensor var_14257_equation_0 = const()[name = tensor("op_14257_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14257_cast_fp16 = einsum(equation = var_14257_equation_0, values = (var_14047_cast_fp16, var_13592_cast_fp16))[name = tensor("op_14257_cast_fp16")]; + tensor var_14258_to_fp16 = const()[name = tensor("op_14258_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1315_cast_fp16 = mul(x = var_14257_cast_fp16, y = var_14258_to_fp16)[name = tensor("aw_chunk_1315_cast_fp16")]; + tensor var_14261_equation_0 = const()[name = tensor("op_14261_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14261_cast_fp16 = einsum(equation = var_14261_equation_0, values = (var_14047_cast_fp16, var_13599_cast_fp16))[name = tensor("op_14261_cast_fp16")]; + tensor var_14262_to_fp16 = const()[name = tensor("op_14262_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1317_cast_fp16 = mul(x = var_14261_cast_fp16, y = var_14262_to_fp16)[name = tensor("aw_chunk_1317_cast_fp16")]; + tensor var_14265_equation_0 = const()[name = tensor("op_14265_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14265_cast_fp16 = einsum(equation = var_14265_equation_0, values = (var_14047_cast_fp16, var_13606_cast_fp16))[name = tensor("op_14265_cast_fp16")]; + tensor var_14266_to_fp16 = const()[name = tensor("op_14266_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1319_cast_fp16 = mul(x = var_14265_cast_fp16, y = var_14266_to_fp16)[name = tensor("aw_chunk_1319_cast_fp16")]; + tensor var_14269_equation_0 = const()[name = tensor("op_14269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14269_cast_fp16 = einsum(equation = var_14269_equation_0, values = (var_14051_cast_fp16, var_13613_cast_fp16))[name = tensor("op_14269_cast_fp16")]; + tensor var_14270_to_fp16 = const()[name = tensor("op_14270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1321_cast_fp16 = mul(x = var_14269_cast_fp16, y = var_14270_to_fp16)[name = tensor("aw_chunk_1321_cast_fp16")]; + tensor var_14273_equation_0 = const()[name = tensor("op_14273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14273_cast_fp16 = einsum(equation = var_14273_equation_0, values = (var_14051_cast_fp16, var_13620_cast_fp16))[name = tensor("op_14273_cast_fp16")]; + tensor var_14274_to_fp16 = const()[name = tensor("op_14274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1323_cast_fp16 = mul(x = var_14273_cast_fp16, y = var_14274_to_fp16)[name = tensor("aw_chunk_1323_cast_fp16")]; + tensor var_14277_equation_0 = const()[name = tensor("op_14277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14277_cast_fp16 = einsum(equation = var_14277_equation_0, values = (var_14051_cast_fp16, var_13627_cast_fp16))[name = tensor("op_14277_cast_fp16")]; + tensor var_14278_to_fp16 = const()[name = tensor("op_14278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1325_cast_fp16 = mul(x = var_14277_cast_fp16, y = var_14278_to_fp16)[name = tensor("aw_chunk_1325_cast_fp16")]; + tensor var_14281_equation_0 = const()[name = tensor("op_14281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14281_cast_fp16 = einsum(equation = var_14281_equation_0, values = (var_14051_cast_fp16, var_13634_cast_fp16))[name = tensor("op_14281_cast_fp16")]; + tensor var_14282_to_fp16 = const()[name = tensor("op_14282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1327_cast_fp16 = mul(x = var_14281_cast_fp16, y = var_14282_to_fp16)[name = tensor("aw_chunk_1327_cast_fp16")]; + tensor var_14285_equation_0 = const()[name = tensor("op_14285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14285_cast_fp16 = einsum(equation = var_14285_equation_0, values = (var_14055_cast_fp16, var_13641_cast_fp16))[name = tensor("op_14285_cast_fp16")]; + tensor var_14286_to_fp16 = const()[name = tensor("op_14286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1329_cast_fp16 = mul(x = var_14285_cast_fp16, y = var_14286_to_fp16)[name = tensor("aw_chunk_1329_cast_fp16")]; + tensor var_14289_equation_0 = const()[name = tensor("op_14289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14289_cast_fp16 = einsum(equation = var_14289_equation_0, values = (var_14055_cast_fp16, var_13648_cast_fp16))[name = tensor("op_14289_cast_fp16")]; + tensor var_14290_to_fp16 = const()[name = tensor("op_14290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1331_cast_fp16 = mul(x = var_14289_cast_fp16, y = var_14290_to_fp16)[name = tensor("aw_chunk_1331_cast_fp16")]; + tensor var_14293_equation_0 = const()[name = tensor("op_14293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14293_cast_fp16 = einsum(equation = var_14293_equation_0, values = (var_14055_cast_fp16, var_13655_cast_fp16))[name = tensor("op_14293_cast_fp16")]; + tensor var_14294_to_fp16 = const()[name = tensor("op_14294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1333_cast_fp16 = mul(x = var_14293_cast_fp16, y = var_14294_to_fp16)[name = tensor("aw_chunk_1333_cast_fp16")]; + tensor var_14297_equation_0 = const()[name = tensor("op_14297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14297_cast_fp16 = einsum(equation = var_14297_equation_0, values = (var_14055_cast_fp16, var_13662_cast_fp16))[name = tensor("op_14297_cast_fp16")]; + tensor var_14298_to_fp16 = const()[name = tensor("op_14298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1335_cast_fp16 = mul(x = var_14297_cast_fp16, y = var_14298_to_fp16)[name = tensor("aw_chunk_1335_cast_fp16")]; + tensor var_14301_equation_0 = const()[name = tensor("op_14301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14301_cast_fp16 = einsum(equation = var_14301_equation_0, values = (var_14059_cast_fp16, var_13669_cast_fp16))[name = tensor("op_14301_cast_fp16")]; + tensor var_14302_to_fp16 = const()[name = tensor("op_14302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1337_cast_fp16 = mul(x = var_14301_cast_fp16, y = var_14302_to_fp16)[name = tensor("aw_chunk_1337_cast_fp16")]; + tensor var_14305_equation_0 = const()[name = tensor("op_14305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14305_cast_fp16 = einsum(equation = var_14305_equation_0, values = (var_14059_cast_fp16, var_13676_cast_fp16))[name = tensor("op_14305_cast_fp16")]; + tensor var_14306_to_fp16 = const()[name = tensor("op_14306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1339_cast_fp16 = mul(x = var_14305_cast_fp16, y = var_14306_to_fp16)[name = tensor("aw_chunk_1339_cast_fp16")]; + tensor var_14309_equation_0 = const()[name = tensor("op_14309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14309_cast_fp16 = einsum(equation = var_14309_equation_0, values = (var_14059_cast_fp16, var_13683_cast_fp16))[name = tensor("op_14309_cast_fp16")]; + tensor var_14310_to_fp16 = const()[name = tensor("op_14310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1341_cast_fp16 = mul(x = var_14309_cast_fp16, y = var_14310_to_fp16)[name = tensor("aw_chunk_1341_cast_fp16")]; + tensor var_14313_equation_0 = const()[name = tensor("op_14313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14313_cast_fp16 = einsum(equation = var_14313_equation_0, values = (var_14059_cast_fp16, var_13690_cast_fp16))[name = tensor("op_14313_cast_fp16")]; + tensor var_14314_to_fp16 = const()[name = tensor("op_14314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1343_cast_fp16 = mul(x = var_14313_cast_fp16, y = var_14314_to_fp16)[name = tensor("aw_chunk_1343_cast_fp16")]; + tensor var_14317_equation_0 = const()[name = tensor("op_14317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14317_cast_fp16 = einsum(equation = var_14317_equation_0, values = (var_14063_cast_fp16, var_13697_cast_fp16))[name = tensor("op_14317_cast_fp16")]; + tensor var_14318_to_fp16 = const()[name = tensor("op_14318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1345_cast_fp16 = mul(x = var_14317_cast_fp16, y = var_14318_to_fp16)[name = tensor("aw_chunk_1345_cast_fp16")]; + tensor var_14321_equation_0 = const()[name = tensor("op_14321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14321_cast_fp16 = einsum(equation = var_14321_equation_0, values = (var_14063_cast_fp16, var_13704_cast_fp16))[name = tensor("op_14321_cast_fp16")]; + tensor var_14322_to_fp16 = const()[name = tensor("op_14322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1347_cast_fp16 = mul(x = var_14321_cast_fp16, y = var_14322_to_fp16)[name = tensor("aw_chunk_1347_cast_fp16")]; + tensor var_14325_equation_0 = const()[name = tensor("op_14325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14325_cast_fp16 = einsum(equation = var_14325_equation_0, values = (var_14063_cast_fp16, var_13711_cast_fp16))[name = tensor("op_14325_cast_fp16")]; + tensor var_14326_to_fp16 = const()[name = tensor("op_14326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1349_cast_fp16 = mul(x = var_14325_cast_fp16, y = var_14326_to_fp16)[name = tensor("aw_chunk_1349_cast_fp16")]; + tensor var_14329_equation_0 = const()[name = tensor("op_14329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14329_cast_fp16 = einsum(equation = var_14329_equation_0, values = (var_14063_cast_fp16, var_13718_cast_fp16))[name = tensor("op_14329_cast_fp16")]; + tensor var_14330_to_fp16 = const()[name = tensor("op_14330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1351_cast_fp16 = mul(x = var_14329_cast_fp16, y = var_14330_to_fp16)[name = tensor("aw_chunk_1351_cast_fp16")]; + tensor var_14333_equation_0 = const()[name = tensor("op_14333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14333_cast_fp16 = einsum(equation = var_14333_equation_0, values = (var_14067_cast_fp16, var_13725_cast_fp16))[name = tensor("op_14333_cast_fp16")]; + tensor var_14334_to_fp16 = const()[name = tensor("op_14334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1353_cast_fp16 = mul(x = var_14333_cast_fp16, y = var_14334_to_fp16)[name = tensor("aw_chunk_1353_cast_fp16")]; + tensor var_14337_equation_0 = const()[name = tensor("op_14337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14337_cast_fp16 = einsum(equation = var_14337_equation_0, values = (var_14067_cast_fp16, var_13732_cast_fp16))[name = tensor("op_14337_cast_fp16")]; + tensor var_14338_to_fp16 = const()[name = tensor("op_14338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1355_cast_fp16 = mul(x = var_14337_cast_fp16, y = var_14338_to_fp16)[name = tensor("aw_chunk_1355_cast_fp16")]; + tensor var_14341_equation_0 = const()[name = tensor("op_14341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14341_cast_fp16 = einsum(equation = var_14341_equation_0, values = (var_14067_cast_fp16, var_13739_cast_fp16))[name = tensor("op_14341_cast_fp16")]; + tensor var_14342_to_fp16 = const()[name = tensor("op_14342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1357_cast_fp16 = mul(x = var_14341_cast_fp16, y = var_14342_to_fp16)[name = tensor("aw_chunk_1357_cast_fp16")]; + tensor var_14345_equation_0 = const()[name = tensor("op_14345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14345_cast_fp16 = einsum(equation = var_14345_equation_0, values = (var_14067_cast_fp16, var_13746_cast_fp16))[name = tensor("op_14345_cast_fp16")]; + tensor var_14346_to_fp16 = const()[name = tensor("op_14346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1359_cast_fp16 = mul(x = var_14345_cast_fp16, y = var_14346_to_fp16)[name = tensor("aw_chunk_1359_cast_fp16")]; + tensor var_14349_equation_0 = const()[name = tensor("op_14349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14349_cast_fp16 = einsum(equation = var_14349_equation_0, values = (var_14071_cast_fp16, var_13753_cast_fp16))[name = tensor("op_14349_cast_fp16")]; + tensor var_14350_to_fp16 = const()[name = tensor("op_14350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1361_cast_fp16 = mul(x = var_14349_cast_fp16, y = var_14350_to_fp16)[name = tensor("aw_chunk_1361_cast_fp16")]; + tensor var_14353_equation_0 = const()[name = tensor("op_14353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14353_cast_fp16 = einsum(equation = var_14353_equation_0, values = (var_14071_cast_fp16, var_13760_cast_fp16))[name = tensor("op_14353_cast_fp16")]; + tensor var_14354_to_fp16 = const()[name = tensor("op_14354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1363_cast_fp16 = mul(x = var_14353_cast_fp16, y = var_14354_to_fp16)[name = tensor("aw_chunk_1363_cast_fp16")]; + tensor var_14357_equation_0 = const()[name = tensor("op_14357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14357_cast_fp16 = einsum(equation = var_14357_equation_0, values = (var_14071_cast_fp16, var_13767_cast_fp16))[name = tensor("op_14357_cast_fp16")]; + tensor var_14358_to_fp16 = const()[name = tensor("op_14358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1365_cast_fp16 = mul(x = var_14357_cast_fp16, y = var_14358_to_fp16)[name = tensor("aw_chunk_1365_cast_fp16")]; + tensor var_14361_equation_0 = const()[name = tensor("op_14361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14361_cast_fp16 = einsum(equation = var_14361_equation_0, values = (var_14071_cast_fp16, var_13774_cast_fp16))[name = tensor("op_14361_cast_fp16")]; + tensor var_14362_to_fp16 = const()[name = tensor("op_14362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1367_cast_fp16 = mul(x = var_14361_cast_fp16, y = var_14362_to_fp16)[name = tensor("aw_chunk_1367_cast_fp16")]; + tensor var_14365_equation_0 = const()[name = tensor("op_14365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14365_cast_fp16 = einsum(equation = var_14365_equation_0, values = (var_14075_cast_fp16, var_13781_cast_fp16))[name = tensor("op_14365_cast_fp16")]; + tensor var_14366_to_fp16 = const()[name = tensor("op_14366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1369_cast_fp16 = mul(x = var_14365_cast_fp16, y = var_14366_to_fp16)[name = tensor("aw_chunk_1369_cast_fp16")]; + tensor var_14369_equation_0 = const()[name = tensor("op_14369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14369_cast_fp16 = einsum(equation = var_14369_equation_0, values = (var_14075_cast_fp16, var_13788_cast_fp16))[name = tensor("op_14369_cast_fp16")]; + tensor var_14370_to_fp16 = const()[name = tensor("op_14370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1371_cast_fp16 = mul(x = var_14369_cast_fp16, y = var_14370_to_fp16)[name = tensor("aw_chunk_1371_cast_fp16")]; + tensor var_14373_equation_0 = const()[name = tensor("op_14373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14373_cast_fp16 = einsum(equation = var_14373_equation_0, values = (var_14075_cast_fp16, var_13795_cast_fp16))[name = tensor("op_14373_cast_fp16")]; + tensor var_14374_to_fp16 = const()[name = tensor("op_14374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1373_cast_fp16 = mul(x = var_14373_cast_fp16, y = var_14374_to_fp16)[name = tensor("aw_chunk_1373_cast_fp16")]; + tensor var_14377_equation_0 = const()[name = tensor("op_14377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14377_cast_fp16 = einsum(equation = var_14377_equation_0, values = (var_14075_cast_fp16, var_13802_cast_fp16))[name = tensor("op_14377_cast_fp16")]; + tensor var_14378_to_fp16 = const()[name = tensor("op_14378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1375_cast_fp16 = mul(x = var_14377_cast_fp16, y = var_14378_to_fp16)[name = tensor("aw_chunk_1375_cast_fp16")]; + tensor var_14381_equation_0 = const()[name = tensor("op_14381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14381_cast_fp16 = einsum(equation = var_14381_equation_0, values = (var_14079_cast_fp16, var_13809_cast_fp16))[name = tensor("op_14381_cast_fp16")]; + tensor var_14382_to_fp16 = const()[name = tensor("op_14382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1377_cast_fp16 = mul(x = var_14381_cast_fp16, y = var_14382_to_fp16)[name = tensor("aw_chunk_1377_cast_fp16")]; + tensor var_14385_equation_0 = const()[name = tensor("op_14385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14385_cast_fp16 = einsum(equation = var_14385_equation_0, values = (var_14079_cast_fp16, var_13816_cast_fp16))[name = tensor("op_14385_cast_fp16")]; + tensor var_14386_to_fp16 = const()[name = tensor("op_14386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1379_cast_fp16 = mul(x = var_14385_cast_fp16, y = var_14386_to_fp16)[name = tensor("aw_chunk_1379_cast_fp16")]; + tensor var_14389_equation_0 = const()[name = tensor("op_14389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14389_cast_fp16 = einsum(equation = var_14389_equation_0, values = (var_14079_cast_fp16, var_13823_cast_fp16))[name = tensor("op_14389_cast_fp16")]; + tensor var_14390_to_fp16 = const()[name = tensor("op_14390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1381_cast_fp16 = mul(x = var_14389_cast_fp16, y = var_14390_to_fp16)[name = tensor("aw_chunk_1381_cast_fp16")]; + tensor var_14393_equation_0 = const()[name = tensor("op_14393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14393_cast_fp16 = einsum(equation = var_14393_equation_0, values = (var_14079_cast_fp16, var_13830_cast_fp16))[name = tensor("op_14393_cast_fp16")]; + tensor var_14394_to_fp16 = const()[name = tensor("op_14394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1383_cast_fp16 = mul(x = var_14393_cast_fp16, y = var_14394_to_fp16)[name = tensor("aw_chunk_1383_cast_fp16")]; + tensor var_14397_equation_0 = const()[name = tensor("op_14397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14397_cast_fp16 = einsum(equation = var_14397_equation_0, values = (var_14083_cast_fp16, var_13837_cast_fp16))[name = tensor("op_14397_cast_fp16")]; + tensor var_14398_to_fp16 = const()[name = tensor("op_14398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1385_cast_fp16 = mul(x = var_14397_cast_fp16, y = var_14398_to_fp16)[name = tensor("aw_chunk_1385_cast_fp16")]; + tensor var_14401_equation_0 = const()[name = tensor("op_14401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14401_cast_fp16 = einsum(equation = var_14401_equation_0, values = (var_14083_cast_fp16, var_13844_cast_fp16))[name = tensor("op_14401_cast_fp16")]; + tensor var_14402_to_fp16 = const()[name = tensor("op_14402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1387_cast_fp16 = mul(x = var_14401_cast_fp16, y = var_14402_to_fp16)[name = tensor("aw_chunk_1387_cast_fp16")]; + tensor var_14405_equation_0 = const()[name = tensor("op_14405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14405_cast_fp16 = einsum(equation = var_14405_equation_0, values = (var_14083_cast_fp16, var_13851_cast_fp16))[name = tensor("op_14405_cast_fp16")]; + tensor var_14406_to_fp16 = const()[name = tensor("op_14406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1389_cast_fp16 = mul(x = var_14405_cast_fp16, y = var_14406_to_fp16)[name = tensor("aw_chunk_1389_cast_fp16")]; + tensor var_14409_equation_0 = const()[name = tensor("op_14409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14409_cast_fp16 = einsum(equation = var_14409_equation_0, values = (var_14083_cast_fp16, var_13858_cast_fp16))[name = tensor("op_14409_cast_fp16")]; + tensor var_14410_to_fp16 = const()[name = tensor("op_14410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1391_cast_fp16 = mul(x = var_14409_cast_fp16, y = var_14410_to_fp16)[name = tensor("aw_chunk_1391_cast_fp16")]; + tensor var_14413_equation_0 = const()[name = tensor("op_14413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14413_cast_fp16 = einsum(equation = var_14413_equation_0, values = (var_14087_cast_fp16, var_13865_cast_fp16))[name = tensor("op_14413_cast_fp16")]; + tensor var_14414_to_fp16 = const()[name = tensor("op_14414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1393_cast_fp16 = mul(x = var_14413_cast_fp16, y = var_14414_to_fp16)[name = tensor("aw_chunk_1393_cast_fp16")]; + tensor var_14417_equation_0 = const()[name = tensor("op_14417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14417_cast_fp16 = einsum(equation = var_14417_equation_0, values = (var_14087_cast_fp16, var_13872_cast_fp16))[name = tensor("op_14417_cast_fp16")]; + tensor var_14418_to_fp16 = const()[name = tensor("op_14418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1395_cast_fp16 = mul(x = var_14417_cast_fp16, y = var_14418_to_fp16)[name = tensor("aw_chunk_1395_cast_fp16")]; + tensor var_14421_equation_0 = const()[name = tensor("op_14421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14421_cast_fp16 = einsum(equation = var_14421_equation_0, values = (var_14087_cast_fp16, var_13879_cast_fp16))[name = tensor("op_14421_cast_fp16")]; + tensor var_14422_to_fp16 = const()[name = tensor("op_14422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1397_cast_fp16 = mul(x = var_14421_cast_fp16, y = var_14422_to_fp16)[name = tensor("aw_chunk_1397_cast_fp16")]; + tensor var_14425_equation_0 = const()[name = tensor("op_14425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14425_cast_fp16 = einsum(equation = var_14425_equation_0, values = (var_14087_cast_fp16, var_13886_cast_fp16))[name = tensor("op_14425_cast_fp16")]; + tensor var_14426_to_fp16 = const()[name = tensor("op_14426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1399_cast_fp16 = mul(x = var_14425_cast_fp16, y = var_14426_to_fp16)[name = tensor("aw_chunk_1399_cast_fp16")]; + tensor var_14429_equation_0 = const()[name = tensor("op_14429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14429_cast_fp16 = einsum(equation = var_14429_equation_0, values = (var_14091_cast_fp16, var_13893_cast_fp16))[name = tensor("op_14429_cast_fp16")]; + tensor var_14430_to_fp16 = const()[name = tensor("op_14430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1401_cast_fp16 = mul(x = var_14429_cast_fp16, y = var_14430_to_fp16)[name = tensor("aw_chunk_1401_cast_fp16")]; + tensor var_14433_equation_0 = const()[name = tensor("op_14433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14433_cast_fp16 = einsum(equation = var_14433_equation_0, values = (var_14091_cast_fp16, var_13900_cast_fp16))[name = tensor("op_14433_cast_fp16")]; + tensor var_14434_to_fp16 = const()[name = tensor("op_14434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1403_cast_fp16 = mul(x = var_14433_cast_fp16, y = var_14434_to_fp16)[name = tensor("aw_chunk_1403_cast_fp16")]; + tensor var_14437_equation_0 = const()[name = tensor("op_14437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14437_cast_fp16 = einsum(equation = var_14437_equation_0, values = (var_14091_cast_fp16, var_13907_cast_fp16))[name = tensor("op_14437_cast_fp16")]; + tensor var_14438_to_fp16 = const()[name = tensor("op_14438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1405_cast_fp16 = mul(x = var_14437_cast_fp16, y = var_14438_to_fp16)[name = tensor("aw_chunk_1405_cast_fp16")]; + tensor var_14441_equation_0 = const()[name = tensor("op_14441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14441_cast_fp16 = einsum(equation = var_14441_equation_0, values = (var_14091_cast_fp16, var_13914_cast_fp16))[name = tensor("op_14441_cast_fp16")]; + tensor var_14442_to_fp16 = const()[name = tensor("op_14442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1407_cast_fp16 = mul(x = var_14441_cast_fp16, y = var_14442_to_fp16)[name = tensor("aw_chunk_1407_cast_fp16")]; + tensor var_14445_equation_0 = const()[name = tensor("op_14445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14445_cast_fp16 = einsum(equation = var_14445_equation_0, values = (var_14095_cast_fp16, var_13921_cast_fp16))[name = tensor("op_14445_cast_fp16")]; + tensor var_14446_to_fp16 = const()[name = tensor("op_14446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1409_cast_fp16 = mul(x = var_14445_cast_fp16, y = var_14446_to_fp16)[name = tensor("aw_chunk_1409_cast_fp16")]; + tensor var_14449_equation_0 = const()[name = tensor("op_14449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14449_cast_fp16 = einsum(equation = var_14449_equation_0, values = (var_14095_cast_fp16, var_13928_cast_fp16))[name = tensor("op_14449_cast_fp16")]; + tensor var_14450_to_fp16 = const()[name = tensor("op_14450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1411_cast_fp16 = mul(x = var_14449_cast_fp16, y = var_14450_to_fp16)[name = tensor("aw_chunk_1411_cast_fp16")]; + tensor var_14453_equation_0 = const()[name = tensor("op_14453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14453_cast_fp16 = einsum(equation = var_14453_equation_0, values = (var_14095_cast_fp16, var_13935_cast_fp16))[name = tensor("op_14453_cast_fp16")]; + tensor var_14454_to_fp16 = const()[name = tensor("op_14454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1413_cast_fp16 = mul(x = var_14453_cast_fp16, y = var_14454_to_fp16)[name = tensor("aw_chunk_1413_cast_fp16")]; + tensor var_14457_equation_0 = const()[name = tensor("op_14457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14457_cast_fp16 = einsum(equation = var_14457_equation_0, values = (var_14095_cast_fp16, var_13942_cast_fp16))[name = tensor("op_14457_cast_fp16")]; + tensor var_14458_to_fp16 = const()[name = tensor("op_14458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1415_cast_fp16 = mul(x = var_14457_cast_fp16, y = var_14458_to_fp16)[name = tensor("aw_chunk_1415_cast_fp16")]; + tensor var_14461_equation_0 = const()[name = tensor("op_14461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14461_cast_fp16 = einsum(equation = var_14461_equation_0, values = (var_14099_cast_fp16, var_13949_cast_fp16))[name = tensor("op_14461_cast_fp16")]; + tensor var_14462_to_fp16 = const()[name = tensor("op_14462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1417_cast_fp16 = mul(x = var_14461_cast_fp16, y = var_14462_to_fp16)[name = tensor("aw_chunk_1417_cast_fp16")]; + tensor var_14465_equation_0 = const()[name = tensor("op_14465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14465_cast_fp16 = einsum(equation = var_14465_equation_0, values = (var_14099_cast_fp16, var_13956_cast_fp16))[name = tensor("op_14465_cast_fp16")]; + tensor var_14466_to_fp16 = const()[name = tensor("op_14466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1419_cast_fp16 = mul(x = var_14465_cast_fp16, y = var_14466_to_fp16)[name = tensor("aw_chunk_1419_cast_fp16")]; + tensor var_14469_equation_0 = const()[name = tensor("op_14469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14469_cast_fp16 = einsum(equation = var_14469_equation_0, values = (var_14099_cast_fp16, var_13963_cast_fp16))[name = tensor("op_14469_cast_fp16")]; + tensor var_14470_to_fp16 = const()[name = tensor("op_14470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1421_cast_fp16 = mul(x = var_14469_cast_fp16, y = var_14470_to_fp16)[name = tensor("aw_chunk_1421_cast_fp16")]; + tensor var_14473_equation_0 = const()[name = tensor("op_14473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14473_cast_fp16 = einsum(equation = var_14473_equation_0, values = (var_14099_cast_fp16, var_13970_cast_fp16))[name = tensor("op_14473_cast_fp16")]; + tensor var_14474_to_fp16 = const()[name = tensor("op_14474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1423_cast_fp16 = mul(x = var_14473_cast_fp16, y = var_14474_to_fp16)[name = tensor("aw_chunk_1423_cast_fp16")]; + tensor var_14477_equation_0 = const()[name = tensor("op_14477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14477_cast_fp16 = einsum(equation = var_14477_equation_0, values = (var_14103_cast_fp16, var_13977_cast_fp16))[name = tensor("op_14477_cast_fp16")]; + tensor var_14478_to_fp16 = const()[name = tensor("op_14478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1425_cast_fp16 = mul(x = var_14477_cast_fp16, y = var_14478_to_fp16)[name = tensor("aw_chunk_1425_cast_fp16")]; + tensor var_14481_equation_0 = const()[name = tensor("op_14481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14481_cast_fp16 = einsum(equation = var_14481_equation_0, values = (var_14103_cast_fp16, var_13984_cast_fp16))[name = tensor("op_14481_cast_fp16")]; + tensor var_14482_to_fp16 = const()[name = tensor("op_14482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1427_cast_fp16 = mul(x = var_14481_cast_fp16, y = var_14482_to_fp16)[name = tensor("aw_chunk_1427_cast_fp16")]; + tensor var_14485_equation_0 = const()[name = tensor("op_14485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14485_cast_fp16 = einsum(equation = var_14485_equation_0, values = (var_14103_cast_fp16, var_13991_cast_fp16))[name = tensor("op_14485_cast_fp16")]; + tensor var_14486_to_fp16 = const()[name = tensor("op_14486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1429_cast_fp16 = mul(x = var_14485_cast_fp16, y = var_14486_to_fp16)[name = tensor("aw_chunk_1429_cast_fp16")]; + tensor var_14489_equation_0 = const()[name = tensor("op_14489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14489_cast_fp16 = einsum(equation = var_14489_equation_0, values = (var_14103_cast_fp16, var_13998_cast_fp16))[name = tensor("op_14489_cast_fp16")]; + tensor var_14490_to_fp16 = const()[name = tensor("op_14490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1431_cast_fp16 = mul(x = var_14489_cast_fp16, y = var_14490_to_fp16)[name = tensor("aw_chunk_1431_cast_fp16")]; + tensor var_14493_equation_0 = const()[name = tensor("op_14493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14493_cast_fp16 = einsum(equation = var_14493_equation_0, values = (var_14107_cast_fp16, var_14005_cast_fp16))[name = tensor("op_14493_cast_fp16")]; + tensor var_14494_to_fp16 = const()[name = tensor("op_14494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1433_cast_fp16 = mul(x = var_14493_cast_fp16, y = var_14494_to_fp16)[name = tensor("aw_chunk_1433_cast_fp16")]; + tensor var_14497_equation_0 = const()[name = tensor("op_14497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14497_cast_fp16 = einsum(equation = var_14497_equation_0, values = (var_14107_cast_fp16, var_14012_cast_fp16))[name = tensor("op_14497_cast_fp16")]; + tensor var_14498_to_fp16 = const()[name = tensor("op_14498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1435_cast_fp16 = mul(x = var_14497_cast_fp16, y = var_14498_to_fp16)[name = tensor("aw_chunk_1435_cast_fp16")]; + tensor var_14501_equation_0 = const()[name = tensor("op_14501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14501_cast_fp16 = einsum(equation = var_14501_equation_0, values = (var_14107_cast_fp16, var_14019_cast_fp16))[name = tensor("op_14501_cast_fp16")]; + tensor var_14502_to_fp16 = const()[name = tensor("op_14502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1437_cast_fp16 = mul(x = var_14501_cast_fp16, y = var_14502_to_fp16)[name = tensor("aw_chunk_1437_cast_fp16")]; + tensor var_14505_equation_0 = const()[name = tensor("op_14505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_14505_cast_fp16 = einsum(equation = var_14505_equation_0, values = (var_14107_cast_fp16, var_14026_cast_fp16))[name = tensor("op_14505_cast_fp16")]; + tensor var_14506_to_fp16 = const()[name = tensor("op_14506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1439_cast_fp16 = mul(x = var_14505_cast_fp16, y = var_14506_to_fp16)[name = tensor("aw_chunk_1439_cast_fp16")]; + tensor var_14508_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1281_cast_fp16)[name = tensor("op_14508_cast_fp16")]; + tensor var_14509_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1283_cast_fp16)[name = tensor("op_14509_cast_fp16")]; + tensor var_14510_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1285_cast_fp16)[name = tensor("op_14510_cast_fp16")]; + tensor var_14511_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1287_cast_fp16)[name = tensor("op_14511_cast_fp16")]; + tensor var_14512_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1289_cast_fp16)[name = tensor("op_14512_cast_fp16")]; + tensor var_14513_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1291_cast_fp16)[name = tensor("op_14513_cast_fp16")]; + tensor var_14514_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1293_cast_fp16)[name = tensor("op_14514_cast_fp16")]; + tensor var_14515_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1295_cast_fp16)[name = tensor("op_14515_cast_fp16")]; + tensor var_14516_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1297_cast_fp16)[name = tensor("op_14516_cast_fp16")]; + tensor var_14517_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1299_cast_fp16)[name = tensor("op_14517_cast_fp16")]; + tensor var_14518_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1301_cast_fp16)[name = tensor("op_14518_cast_fp16")]; + tensor var_14519_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1303_cast_fp16)[name = tensor("op_14519_cast_fp16")]; + tensor var_14520_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1305_cast_fp16)[name = tensor("op_14520_cast_fp16")]; + tensor var_14521_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1307_cast_fp16)[name = tensor("op_14521_cast_fp16")]; + tensor var_14522_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1309_cast_fp16)[name = tensor("op_14522_cast_fp16")]; + tensor var_14523_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1311_cast_fp16)[name = tensor("op_14523_cast_fp16")]; + tensor var_14524_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1313_cast_fp16)[name = tensor("op_14524_cast_fp16")]; + tensor var_14525_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1315_cast_fp16)[name = tensor("op_14525_cast_fp16")]; + tensor var_14526_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1317_cast_fp16)[name = tensor("op_14526_cast_fp16")]; + tensor var_14527_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1319_cast_fp16)[name = tensor("op_14527_cast_fp16")]; + tensor var_14528_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1321_cast_fp16)[name = tensor("op_14528_cast_fp16")]; + tensor var_14529_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1323_cast_fp16)[name = tensor("op_14529_cast_fp16")]; + tensor var_14530_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1325_cast_fp16)[name = tensor("op_14530_cast_fp16")]; + tensor var_14531_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1327_cast_fp16)[name = tensor("op_14531_cast_fp16")]; + tensor var_14532_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1329_cast_fp16)[name = tensor("op_14532_cast_fp16")]; + tensor var_14533_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1331_cast_fp16)[name = tensor("op_14533_cast_fp16")]; + tensor var_14534_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1333_cast_fp16)[name = tensor("op_14534_cast_fp16")]; + tensor var_14535_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1335_cast_fp16)[name = tensor("op_14535_cast_fp16")]; + tensor var_14536_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1337_cast_fp16)[name = tensor("op_14536_cast_fp16")]; + tensor var_14537_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1339_cast_fp16)[name = tensor("op_14537_cast_fp16")]; + tensor var_14538_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1341_cast_fp16)[name = tensor("op_14538_cast_fp16")]; + tensor var_14539_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1343_cast_fp16)[name = tensor("op_14539_cast_fp16")]; + tensor var_14540_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1345_cast_fp16)[name = tensor("op_14540_cast_fp16")]; + tensor var_14541_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1347_cast_fp16)[name = tensor("op_14541_cast_fp16")]; + tensor var_14542_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1349_cast_fp16)[name = tensor("op_14542_cast_fp16")]; + tensor var_14543_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1351_cast_fp16)[name = tensor("op_14543_cast_fp16")]; + tensor var_14544_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1353_cast_fp16)[name = tensor("op_14544_cast_fp16")]; + tensor var_14545_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1355_cast_fp16)[name = tensor("op_14545_cast_fp16")]; + tensor var_14546_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1357_cast_fp16)[name = tensor("op_14546_cast_fp16")]; + tensor var_14547_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1359_cast_fp16)[name = tensor("op_14547_cast_fp16")]; + tensor var_14548_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1361_cast_fp16)[name = tensor("op_14548_cast_fp16")]; + tensor var_14549_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1363_cast_fp16)[name = tensor("op_14549_cast_fp16")]; + tensor var_14550_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1365_cast_fp16)[name = tensor("op_14550_cast_fp16")]; + tensor var_14551_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1367_cast_fp16)[name = tensor("op_14551_cast_fp16")]; + tensor var_14552_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1369_cast_fp16)[name = tensor("op_14552_cast_fp16")]; + tensor var_14553_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1371_cast_fp16)[name = tensor("op_14553_cast_fp16")]; + tensor var_14554_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1373_cast_fp16)[name = tensor("op_14554_cast_fp16")]; + tensor var_14555_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1375_cast_fp16)[name = tensor("op_14555_cast_fp16")]; + tensor var_14556_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1377_cast_fp16)[name = tensor("op_14556_cast_fp16")]; + tensor var_14557_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1379_cast_fp16)[name = tensor("op_14557_cast_fp16")]; + tensor var_14558_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1381_cast_fp16)[name = tensor("op_14558_cast_fp16")]; + tensor var_14559_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1383_cast_fp16)[name = tensor("op_14559_cast_fp16")]; + tensor var_14560_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1385_cast_fp16)[name = tensor("op_14560_cast_fp16")]; + tensor var_14561_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1387_cast_fp16)[name = tensor("op_14561_cast_fp16")]; + tensor var_14562_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1389_cast_fp16)[name = tensor("op_14562_cast_fp16")]; + tensor var_14563_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1391_cast_fp16)[name = tensor("op_14563_cast_fp16")]; + tensor var_14564_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1393_cast_fp16)[name = tensor("op_14564_cast_fp16")]; + tensor var_14565_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1395_cast_fp16)[name = tensor("op_14565_cast_fp16")]; + tensor var_14566_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1397_cast_fp16)[name = tensor("op_14566_cast_fp16")]; + tensor var_14567_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1399_cast_fp16)[name = tensor("op_14567_cast_fp16")]; + tensor var_14568_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1401_cast_fp16)[name = tensor("op_14568_cast_fp16")]; + tensor var_14569_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1403_cast_fp16)[name = tensor("op_14569_cast_fp16")]; + tensor var_14570_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1405_cast_fp16)[name = tensor("op_14570_cast_fp16")]; + tensor var_14571_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1407_cast_fp16)[name = tensor("op_14571_cast_fp16")]; + tensor var_14572_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1409_cast_fp16)[name = tensor("op_14572_cast_fp16")]; + tensor var_14573_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1411_cast_fp16)[name = tensor("op_14573_cast_fp16")]; + tensor var_14574_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1413_cast_fp16)[name = tensor("op_14574_cast_fp16")]; + tensor var_14575_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1415_cast_fp16)[name = tensor("op_14575_cast_fp16")]; + tensor var_14576_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1417_cast_fp16)[name = tensor("op_14576_cast_fp16")]; + tensor var_14577_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1419_cast_fp16)[name = tensor("op_14577_cast_fp16")]; + tensor var_14578_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1421_cast_fp16)[name = tensor("op_14578_cast_fp16")]; + tensor var_14579_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1423_cast_fp16)[name = tensor("op_14579_cast_fp16")]; + tensor var_14580_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1425_cast_fp16)[name = tensor("op_14580_cast_fp16")]; + tensor var_14581_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1427_cast_fp16)[name = tensor("op_14581_cast_fp16")]; + tensor var_14582_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1429_cast_fp16)[name = tensor("op_14582_cast_fp16")]; + tensor var_14583_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1431_cast_fp16)[name = tensor("op_14583_cast_fp16")]; + tensor var_14584_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1433_cast_fp16)[name = tensor("op_14584_cast_fp16")]; + tensor var_14585_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1435_cast_fp16)[name = tensor("op_14585_cast_fp16")]; + tensor var_14586_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1437_cast_fp16)[name = tensor("op_14586_cast_fp16")]; + tensor var_14587_cast_fp16 = softmax(axis = var_13279, x = aw_chunk_1439_cast_fp16)[name = tensor("op_14587_cast_fp16")]; + tensor var_14589_equation_0 = const()[name = tensor("op_14589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14589_cast_fp16 = einsum(equation = var_14589_equation_0, values = (var_14109_cast_fp16, var_14508_cast_fp16))[name = tensor("op_14589_cast_fp16")]; + tensor var_14591_equation_0 = const()[name = tensor("op_14591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14591_cast_fp16 = einsum(equation = var_14591_equation_0, values = (var_14109_cast_fp16, var_14509_cast_fp16))[name = tensor("op_14591_cast_fp16")]; + tensor var_14593_equation_0 = const()[name = tensor("op_14593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14593_cast_fp16 = einsum(equation = var_14593_equation_0, values = (var_14109_cast_fp16, var_14510_cast_fp16))[name = tensor("op_14593_cast_fp16")]; + tensor var_14595_equation_0 = const()[name = tensor("op_14595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14595_cast_fp16 = einsum(equation = var_14595_equation_0, values = (var_14109_cast_fp16, var_14511_cast_fp16))[name = tensor("op_14595_cast_fp16")]; + tensor var_14597_equation_0 = const()[name = tensor("op_14597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14597_cast_fp16 = einsum(equation = var_14597_equation_0, values = (var_14113_cast_fp16, var_14512_cast_fp16))[name = tensor("op_14597_cast_fp16")]; + tensor var_14599_equation_0 = const()[name = tensor("op_14599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14599_cast_fp16 = einsum(equation = var_14599_equation_0, values = (var_14113_cast_fp16, var_14513_cast_fp16))[name = tensor("op_14599_cast_fp16")]; + tensor var_14601_equation_0 = const()[name = tensor("op_14601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14601_cast_fp16 = einsum(equation = var_14601_equation_0, values = (var_14113_cast_fp16, var_14514_cast_fp16))[name = tensor("op_14601_cast_fp16")]; + tensor var_14603_equation_0 = const()[name = tensor("op_14603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14603_cast_fp16 = einsum(equation = var_14603_equation_0, values = (var_14113_cast_fp16, var_14515_cast_fp16))[name = tensor("op_14603_cast_fp16")]; + tensor var_14605_equation_0 = const()[name = tensor("op_14605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14605_cast_fp16 = einsum(equation = var_14605_equation_0, values = (var_14117_cast_fp16, var_14516_cast_fp16))[name = tensor("op_14605_cast_fp16")]; + tensor var_14607_equation_0 = const()[name = tensor("op_14607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14607_cast_fp16 = einsum(equation = var_14607_equation_0, values = (var_14117_cast_fp16, var_14517_cast_fp16))[name = tensor("op_14607_cast_fp16")]; + tensor var_14609_equation_0 = const()[name = tensor("op_14609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14609_cast_fp16 = einsum(equation = var_14609_equation_0, values = (var_14117_cast_fp16, var_14518_cast_fp16))[name = tensor("op_14609_cast_fp16")]; + tensor var_14611_equation_0 = const()[name = tensor("op_14611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14611_cast_fp16 = einsum(equation = var_14611_equation_0, values = (var_14117_cast_fp16, var_14519_cast_fp16))[name = tensor("op_14611_cast_fp16")]; + tensor var_14613_equation_0 = const()[name = tensor("op_14613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14613_cast_fp16 = einsum(equation = var_14613_equation_0, values = (var_14121_cast_fp16, var_14520_cast_fp16))[name = tensor("op_14613_cast_fp16")]; + tensor var_14615_equation_0 = const()[name = tensor("op_14615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14615_cast_fp16 = einsum(equation = var_14615_equation_0, values = (var_14121_cast_fp16, var_14521_cast_fp16))[name = tensor("op_14615_cast_fp16")]; + tensor var_14617_equation_0 = const()[name = tensor("op_14617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14617_cast_fp16 = einsum(equation = var_14617_equation_0, values = (var_14121_cast_fp16, var_14522_cast_fp16))[name = tensor("op_14617_cast_fp16")]; + tensor var_14619_equation_0 = const()[name = tensor("op_14619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14619_cast_fp16 = einsum(equation = var_14619_equation_0, values = (var_14121_cast_fp16, var_14523_cast_fp16))[name = tensor("op_14619_cast_fp16")]; + tensor var_14621_equation_0 = const()[name = tensor("op_14621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14621_cast_fp16 = einsum(equation = var_14621_equation_0, values = (var_14125_cast_fp16, var_14524_cast_fp16))[name = tensor("op_14621_cast_fp16")]; + tensor var_14623_equation_0 = const()[name = tensor("op_14623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14623_cast_fp16 = einsum(equation = var_14623_equation_0, values = (var_14125_cast_fp16, var_14525_cast_fp16))[name = tensor("op_14623_cast_fp16")]; + tensor var_14625_equation_0 = const()[name = tensor("op_14625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14625_cast_fp16 = einsum(equation = var_14625_equation_0, values = (var_14125_cast_fp16, var_14526_cast_fp16))[name = tensor("op_14625_cast_fp16")]; + tensor var_14627_equation_0 = const()[name = tensor("op_14627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14627_cast_fp16 = einsum(equation = var_14627_equation_0, values = (var_14125_cast_fp16, var_14527_cast_fp16))[name = tensor("op_14627_cast_fp16")]; + tensor var_14629_equation_0 = const()[name = tensor("op_14629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14629_cast_fp16 = einsum(equation = var_14629_equation_0, values = (var_14129_cast_fp16, var_14528_cast_fp16))[name = tensor("op_14629_cast_fp16")]; + tensor var_14631_equation_0 = const()[name = tensor("op_14631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14631_cast_fp16 = einsum(equation = var_14631_equation_0, values = (var_14129_cast_fp16, var_14529_cast_fp16))[name = tensor("op_14631_cast_fp16")]; + tensor var_14633_equation_0 = const()[name = tensor("op_14633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14633_cast_fp16 = einsum(equation = var_14633_equation_0, values = (var_14129_cast_fp16, var_14530_cast_fp16))[name = tensor("op_14633_cast_fp16")]; + tensor var_14635_equation_0 = const()[name = tensor("op_14635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14635_cast_fp16 = einsum(equation = var_14635_equation_0, values = (var_14129_cast_fp16, var_14531_cast_fp16))[name = tensor("op_14635_cast_fp16")]; + tensor var_14637_equation_0 = const()[name = tensor("op_14637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14637_cast_fp16 = einsum(equation = var_14637_equation_0, values = (var_14133_cast_fp16, var_14532_cast_fp16))[name = tensor("op_14637_cast_fp16")]; + tensor var_14639_equation_0 = const()[name = tensor("op_14639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14639_cast_fp16 = einsum(equation = var_14639_equation_0, values = (var_14133_cast_fp16, var_14533_cast_fp16))[name = tensor("op_14639_cast_fp16")]; + tensor var_14641_equation_0 = const()[name = tensor("op_14641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14641_cast_fp16 = einsum(equation = var_14641_equation_0, values = (var_14133_cast_fp16, var_14534_cast_fp16))[name = tensor("op_14641_cast_fp16")]; + tensor var_14643_equation_0 = const()[name = tensor("op_14643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14643_cast_fp16 = einsum(equation = var_14643_equation_0, values = (var_14133_cast_fp16, var_14535_cast_fp16))[name = tensor("op_14643_cast_fp16")]; + tensor var_14645_equation_0 = const()[name = tensor("op_14645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14645_cast_fp16 = einsum(equation = var_14645_equation_0, values = (var_14137_cast_fp16, var_14536_cast_fp16))[name = tensor("op_14645_cast_fp16")]; + tensor var_14647_equation_0 = const()[name = tensor("op_14647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14647_cast_fp16 = einsum(equation = var_14647_equation_0, values = (var_14137_cast_fp16, var_14537_cast_fp16))[name = tensor("op_14647_cast_fp16")]; + tensor var_14649_equation_0 = const()[name = tensor("op_14649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14649_cast_fp16 = einsum(equation = var_14649_equation_0, values = (var_14137_cast_fp16, var_14538_cast_fp16))[name = tensor("op_14649_cast_fp16")]; + tensor var_14651_equation_0 = const()[name = tensor("op_14651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14651_cast_fp16 = einsum(equation = var_14651_equation_0, values = (var_14137_cast_fp16, var_14539_cast_fp16))[name = tensor("op_14651_cast_fp16")]; + tensor var_14653_equation_0 = const()[name = tensor("op_14653_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14653_cast_fp16 = einsum(equation = var_14653_equation_0, values = (var_14141_cast_fp16, var_14540_cast_fp16))[name = tensor("op_14653_cast_fp16")]; + tensor var_14655_equation_0 = const()[name = tensor("op_14655_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14655_cast_fp16 = einsum(equation = var_14655_equation_0, values = (var_14141_cast_fp16, var_14541_cast_fp16))[name = tensor("op_14655_cast_fp16")]; + tensor var_14657_equation_0 = const()[name = tensor("op_14657_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14657_cast_fp16 = einsum(equation = var_14657_equation_0, values = (var_14141_cast_fp16, var_14542_cast_fp16))[name = tensor("op_14657_cast_fp16")]; + tensor var_14659_equation_0 = const()[name = tensor("op_14659_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14659_cast_fp16 = einsum(equation = var_14659_equation_0, values = (var_14141_cast_fp16, var_14543_cast_fp16))[name = tensor("op_14659_cast_fp16")]; + tensor var_14661_equation_0 = const()[name = tensor("op_14661_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14661_cast_fp16 = einsum(equation = var_14661_equation_0, values = (var_14145_cast_fp16, var_14544_cast_fp16))[name = tensor("op_14661_cast_fp16")]; + tensor var_14663_equation_0 = const()[name = tensor("op_14663_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14663_cast_fp16 = einsum(equation = var_14663_equation_0, values = (var_14145_cast_fp16, var_14545_cast_fp16))[name = tensor("op_14663_cast_fp16")]; + tensor var_14665_equation_0 = const()[name = tensor("op_14665_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14665_cast_fp16 = einsum(equation = var_14665_equation_0, values = (var_14145_cast_fp16, var_14546_cast_fp16))[name = tensor("op_14665_cast_fp16")]; + tensor var_14667_equation_0 = const()[name = tensor("op_14667_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14667_cast_fp16 = einsum(equation = var_14667_equation_0, values = (var_14145_cast_fp16, var_14547_cast_fp16))[name = tensor("op_14667_cast_fp16")]; + tensor var_14669_equation_0 = const()[name = tensor("op_14669_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14669_cast_fp16 = einsum(equation = var_14669_equation_0, values = (var_14149_cast_fp16, var_14548_cast_fp16))[name = tensor("op_14669_cast_fp16")]; + tensor var_14671_equation_0 = const()[name = tensor("op_14671_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14671_cast_fp16 = einsum(equation = var_14671_equation_0, values = (var_14149_cast_fp16, var_14549_cast_fp16))[name = tensor("op_14671_cast_fp16")]; + tensor var_14673_equation_0 = const()[name = tensor("op_14673_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14673_cast_fp16 = einsum(equation = var_14673_equation_0, values = (var_14149_cast_fp16, var_14550_cast_fp16))[name = tensor("op_14673_cast_fp16")]; + tensor var_14675_equation_0 = const()[name = tensor("op_14675_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14675_cast_fp16 = einsum(equation = var_14675_equation_0, values = (var_14149_cast_fp16, var_14551_cast_fp16))[name = tensor("op_14675_cast_fp16")]; + tensor var_14677_equation_0 = const()[name = tensor("op_14677_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14677_cast_fp16 = einsum(equation = var_14677_equation_0, values = (var_14153_cast_fp16, var_14552_cast_fp16))[name = tensor("op_14677_cast_fp16")]; + tensor var_14679_equation_0 = const()[name = tensor("op_14679_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14679_cast_fp16 = einsum(equation = var_14679_equation_0, values = (var_14153_cast_fp16, var_14553_cast_fp16))[name = tensor("op_14679_cast_fp16")]; + tensor var_14681_equation_0 = const()[name = tensor("op_14681_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14681_cast_fp16 = einsum(equation = var_14681_equation_0, values = (var_14153_cast_fp16, var_14554_cast_fp16))[name = tensor("op_14681_cast_fp16")]; + tensor var_14683_equation_0 = const()[name = tensor("op_14683_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14683_cast_fp16 = einsum(equation = var_14683_equation_0, values = (var_14153_cast_fp16, var_14555_cast_fp16))[name = tensor("op_14683_cast_fp16")]; + tensor var_14685_equation_0 = const()[name = tensor("op_14685_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14685_cast_fp16 = einsum(equation = var_14685_equation_0, values = (var_14157_cast_fp16, var_14556_cast_fp16))[name = tensor("op_14685_cast_fp16")]; + tensor var_14687_equation_0 = const()[name = tensor("op_14687_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14687_cast_fp16 = einsum(equation = var_14687_equation_0, values = (var_14157_cast_fp16, var_14557_cast_fp16))[name = tensor("op_14687_cast_fp16")]; + tensor var_14689_equation_0 = const()[name = tensor("op_14689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14689_cast_fp16 = einsum(equation = var_14689_equation_0, values = (var_14157_cast_fp16, var_14558_cast_fp16))[name = tensor("op_14689_cast_fp16")]; + tensor var_14691_equation_0 = const()[name = tensor("op_14691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14691_cast_fp16 = einsum(equation = var_14691_equation_0, values = (var_14157_cast_fp16, var_14559_cast_fp16))[name = tensor("op_14691_cast_fp16")]; + tensor var_14693_equation_0 = const()[name = tensor("op_14693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14693_cast_fp16 = einsum(equation = var_14693_equation_0, values = (var_14161_cast_fp16, var_14560_cast_fp16))[name = tensor("op_14693_cast_fp16")]; + tensor var_14695_equation_0 = const()[name = tensor("op_14695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14695_cast_fp16 = einsum(equation = var_14695_equation_0, values = (var_14161_cast_fp16, var_14561_cast_fp16))[name = tensor("op_14695_cast_fp16")]; + tensor var_14697_equation_0 = const()[name = tensor("op_14697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14697_cast_fp16 = einsum(equation = var_14697_equation_0, values = (var_14161_cast_fp16, var_14562_cast_fp16))[name = tensor("op_14697_cast_fp16")]; + tensor var_14699_equation_0 = const()[name = tensor("op_14699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14699_cast_fp16 = einsum(equation = var_14699_equation_0, values = (var_14161_cast_fp16, var_14563_cast_fp16))[name = tensor("op_14699_cast_fp16")]; + tensor var_14701_equation_0 = const()[name = tensor("op_14701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14701_cast_fp16 = einsum(equation = var_14701_equation_0, values = (var_14165_cast_fp16, var_14564_cast_fp16))[name = tensor("op_14701_cast_fp16")]; + tensor var_14703_equation_0 = const()[name = tensor("op_14703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14703_cast_fp16 = einsum(equation = var_14703_equation_0, values = (var_14165_cast_fp16, var_14565_cast_fp16))[name = tensor("op_14703_cast_fp16")]; + tensor var_14705_equation_0 = const()[name = tensor("op_14705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14705_cast_fp16 = einsum(equation = var_14705_equation_0, values = (var_14165_cast_fp16, var_14566_cast_fp16))[name = tensor("op_14705_cast_fp16")]; + tensor var_14707_equation_0 = const()[name = tensor("op_14707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14707_cast_fp16 = einsum(equation = var_14707_equation_0, values = (var_14165_cast_fp16, var_14567_cast_fp16))[name = tensor("op_14707_cast_fp16")]; + tensor var_14709_equation_0 = const()[name = tensor("op_14709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14709_cast_fp16 = einsum(equation = var_14709_equation_0, values = (var_14169_cast_fp16, var_14568_cast_fp16))[name = tensor("op_14709_cast_fp16")]; + tensor var_14711_equation_0 = const()[name = tensor("op_14711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14711_cast_fp16 = einsum(equation = var_14711_equation_0, values = (var_14169_cast_fp16, var_14569_cast_fp16))[name = tensor("op_14711_cast_fp16")]; + tensor var_14713_equation_0 = const()[name = tensor("op_14713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14713_cast_fp16 = einsum(equation = var_14713_equation_0, values = (var_14169_cast_fp16, var_14570_cast_fp16))[name = tensor("op_14713_cast_fp16")]; + tensor var_14715_equation_0 = const()[name = tensor("op_14715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14715_cast_fp16 = einsum(equation = var_14715_equation_0, values = (var_14169_cast_fp16, var_14571_cast_fp16))[name = tensor("op_14715_cast_fp16")]; + tensor var_14717_equation_0 = const()[name = tensor("op_14717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14717_cast_fp16 = einsum(equation = var_14717_equation_0, values = (var_14173_cast_fp16, var_14572_cast_fp16))[name = tensor("op_14717_cast_fp16")]; + tensor var_14719_equation_0 = const()[name = tensor("op_14719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14719_cast_fp16 = einsum(equation = var_14719_equation_0, values = (var_14173_cast_fp16, var_14573_cast_fp16))[name = tensor("op_14719_cast_fp16")]; + tensor var_14721_equation_0 = const()[name = tensor("op_14721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14721_cast_fp16 = einsum(equation = var_14721_equation_0, values = (var_14173_cast_fp16, var_14574_cast_fp16))[name = tensor("op_14721_cast_fp16")]; + tensor var_14723_equation_0 = const()[name = tensor("op_14723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14723_cast_fp16 = einsum(equation = var_14723_equation_0, values = (var_14173_cast_fp16, var_14575_cast_fp16))[name = tensor("op_14723_cast_fp16")]; + tensor var_14725_equation_0 = const()[name = tensor("op_14725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14725_cast_fp16 = einsum(equation = var_14725_equation_0, values = (var_14177_cast_fp16, var_14576_cast_fp16))[name = tensor("op_14725_cast_fp16")]; + tensor var_14727_equation_0 = const()[name = tensor("op_14727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14727_cast_fp16 = einsum(equation = var_14727_equation_0, values = (var_14177_cast_fp16, var_14577_cast_fp16))[name = tensor("op_14727_cast_fp16")]; + tensor var_14729_equation_0 = const()[name = tensor("op_14729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14729_cast_fp16 = einsum(equation = var_14729_equation_0, values = (var_14177_cast_fp16, var_14578_cast_fp16))[name = tensor("op_14729_cast_fp16")]; + tensor var_14731_equation_0 = const()[name = tensor("op_14731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14731_cast_fp16 = einsum(equation = var_14731_equation_0, values = (var_14177_cast_fp16, var_14579_cast_fp16))[name = tensor("op_14731_cast_fp16")]; + tensor var_14733_equation_0 = const()[name = tensor("op_14733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14733_cast_fp16 = einsum(equation = var_14733_equation_0, values = (var_14181_cast_fp16, var_14580_cast_fp16))[name = tensor("op_14733_cast_fp16")]; + tensor var_14735_equation_0 = const()[name = tensor("op_14735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14735_cast_fp16 = einsum(equation = var_14735_equation_0, values = (var_14181_cast_fp16, var_14581_cast_fp16))[name = tensor("op_14735_cast_fp16")]; + tensor var_14737_equation_0 = const()[name = tensor("op_14737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14737_cast_fp16 = einsum(equation = var_14737_equation_0, values = (var_14181_cast_fp16, var_14582_cast_fp16))[name = tensor("op_14737_cast_fp16")]; + tensor var_14739_equation_0 = const()[name = tensor("op_14739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14739_cast_fp16 = einsum(equation = var_14739_equation_0, values = (var_14181_cast_fp16, var_14583_cast_fp16))[name = tensor("op_14739_cast_fp16")]; + tensor var_14741_equation_0 = const()[name = tensor("op_14741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14741_cast_fp16 = einsum(equation = var_14741_equation_0, values = (var_14185_cast_fp16, var_14584_cast_fp16))[name = tensor("op_14741_cast_fp16")]; + tensor var_14743_equation_0 = const()[name = tensor("op_14743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14743_cast_fp16 = einsum(equation = var_14743_equation_0, values = (var_14185_cast_fp16, var_14585_cast_fp16))[name = tensor("op_14743_cast_fp16")]; + tensor var_14745_equation_0 = const()[name = tensor("op_14745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14745_cast_fp16 = einsum(equation = var_14745_equation_0, values = (var_14185_cast_fp16, var_14586_cast_fp16))[name = tensor("op_14745_cast_fp16")]; + tensor var_14747_equation_0 = const()[name = tensor("op_14747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_14747_cast_fp16 = einsum(equation = var_14747_equation_0, values = (var_14185_cast_fp16, var_14587_cast_fp16))[name = tensor("op_14747_cast_fp16")]; + tensor var_14749_interleave_0 = const()[name = tensor("op_14749_interleave_0"), val = tensor(false)]; + tensor var_14749_cast_fp16 = concat(axis = var_13254, interleave = var_14749_interleave_0, values = (var_14589_cast_fp16, var_14591_cast_fp16, var_14593_cast_fp16, var_14595_cast_fp16))[name = tensor("op_14749_cast_fp16")]; + tensor var_14751_interleave_0 = const()[name = tensor("op_14751_interleave_0"), val = tensor(false)]; + tensor var_14751_cast_fp16 = concat(axis = var_13254, interleave = var_14751_interleave_0, values = (var_14597_cast_fp16, var_14599_cast_fp16, var_14601_cast_fp16, var_14603_cast_fp16))[name = tensor("op_14751_cast_fp16")]; + tensor var_14753_interleave_0 = const()[name = tensor("op_14753_interleave_0"), val = tensor(false)]; + tensor var_14753_cast_fp16 = concat(axis = var_13254, interleave = var_14753_interleave_0, values = (var_14605_cast_fp16, var_14607_cast_fp16, var_14609_cast_fp16, var_14611_cast_fp16))[name = tensor("op_14753_cast_fp16")]; + tensor var_14755_interleave_0 = const()[name = tensor("op_14755_interleave_0"), val = tensor(false)]; + tensor var_14755_cast_fp16 = concat(axis = var_13254, interleave = var_14755_interleave_0, values = (var_14613_cast_fp16, var_14615_cast_fp16, var_14617_cast_fp16, var_14619_cast_fp16))[name = tensor("op_14755_cast_fp16")]; + tensor var_14757_interleave_0 = const()[name = tensor("op_14757_interleave_0"), val = tensor(false)]; + tensor var_14757_cast_fp16 = concat(axis = var_13254, interleave = var_14757_interleave_0, values = (var_14621_cast_fp16, var_14623_cast_fp16, var_14625_cast_fp16, var_14627_cast_fp16))[name = tensor("op_14757_cast_fp16")]; + tensor var_14759_interleave_0 = const()[name = tensor("op_14759_interleave_0"), val = tensor(false)]; + tensor var_14759_cast_fp16 = concat(axis = var_13254, interleave = var_14759_interleave_0, values = (var_14629_cast_fp16, var_14631_cast_fp16, var_14633_cast_fp16, var_14635_cast_fp16))[name = tensor("op_14759_cast_fp16")]; + tensor var_14761_interleave_0 = const()[name = tensor("op_14761_interleave_0"), val = tensor(false)]; + tensor var_14761_cast_fp16 = concat(axis = var_13254, interleave = var_14761_interleave_0, values = (var_14637_cast_fp16, var_14639_cast_fp16, var_14641_cast_fp16, var_14643_cast_fp16))[name = tensor("op_14761_cast_fp16")]; + tensor var_14763_interleave_0 = const()[name = tensor("op_14763_interleave_0"), val = tensor(false)]; + tensor var_14763_cast_fp16 = concat(axis = var_13254, interleave = var_14763_interleave_0, values = (var_14645_cast_fp16, var_14647_cast_fp16, var_14649_cast_fp16, var_14651_cast_fp16))[name = tensor("op_14763_cast_fp16")]; + tensor var_14765_interleave_0 = const()[name = tensor("op_14765_interleave_0"), val = tensor(false)]; + tensor var_14765_cast_fp16 = concat(axis = var_13254, interleave = var_14765_interleave_0, values = (var_14653_cast_fp16, var_14655_cast_fp16, var_14657_cast_fp16, var_14659_cast_fp16))[name = tensor("op_14765_cast_fp16")]; + tensor var_14767_interleave_0 = const()[name = tensor("op_14767_interleave_0"), val = tensor(false)]; + tensor var_14767_cast_fp16 = concat(axis = var_13254, interleave = var_14767_interleave_0, values = (var_14661_cast_fp16, var_14663_cast_fp16, var_14665_cast_fp16, var_14667_cast_fp16))[name = tensor("op_14767_cast_fp16")]; + tensor var_14769_interleave_0 = const()[name = tensor("op_14769_interleave_0"), val = tensor(false)]; + tensor var_14769_cast_fp16 = concat(axis = var_13254, interleave = var_14769_interleave_0, values = (var_14669_cast_fp16, var_14671_cast_fp16, var_14673_cast_fp16, var_14675_cast_fp16))[name = tensor("op_14769_cast_fp16")]; + tensor var_14771_interleave_0 = const()[name = tensor("op_14771_interleave_0"), val = tensor(false)]; + tensor var_14771_cast_fp16 = concat(axis = var_13254, interleave = var_14771_interleave_0, values = (var_14677_cast_fp16, var_14679_cast_fp16, var_14681_cast_fp16, var_14683_cast_fp16))[name = tensor("op_14771_cast_fp16")]; + tensor var_14773_interleave_0 = const()[name = tensor("op_14773_interleave_0"), val = tensor(false)]; + tensor var_14773_cast_fp16 = concat(axis = var_13254, interleave = var_14773_interleave_0, values = (var_14685_cast_fp16, var_14687_cast_fp16, var_14689_cast_fp16, var_14691_cast_fp16))[name = tensor("op_14773_cast_fp16")]; + tensor var_14775_interleave_0 = const()[name = tensor("op_14775_interleave_0"), val = tensor(false)]; + tensor var_14775_cast_fp16 = concat(axis = var_13254, interleave = var_14775_interleave_0, values = (var_14693_cast_fp16, var_14695_cast_fp16, var_14697_cast_fp16, var_14699_cast_fp16))[name = tensor("op_14775_cast_fp16")]; + tensor var_14777_interleave_0 = const()[name = tensor("op_14777_interleave_0"), val = tensor(false)]; + tensor var_14777_cast_fp16 = concat(axis = var_13254, interleave = var_14777_interleave_0, values = (var_14701_cast_fp16, var_14703_cast_fp16, var_14705_cast_fp16, var_14707_cast_fp16))[name = tensor("op_14777_cast_fp16")]; + tensor var_14779_interleave_0 = const()[name = tensor("op_14779_interleave_0"), val = tensor(false)]; + tensor var_14779_cast_fp16 = concat(axis = var_13254, interleave = var_14779_interleave_0, values = (var_14709_cast_fp16, var_14711_cast_fp16, var_14713_cast_fp16, var_14715_cast_fp16))[name = tensor("op_14779_cast_fp16")]; + tensor var_14781_interleave_0 = const()[name = tensor("op_14781_interleave_0"), val = tensor(false)]; + tensor var_14781_cast_fp16 = concat(axis = var_13254, interleave = var_14781_interleave_0, values = (var_14717_cast_fp16, var_14719_cast_fp16, var_14721_cast_fp16, var_14723_cast_fp16))[name = tensor("op_14781_cast_fp16")]; + tensor var_14783_interleave_0 = const()[name = tensor("op_14783_interleave_0"), val = tensor(false)]; + tensor var_14783_cast_fp16 = concat(axis = var_13254, interleave = var_14783_interleave_0, values = (var_14725_cast_fp16, var_14727_cast_fp16, var_14729_cast_fp16, var_14731_cast_fp16))[name = tensor("op_14783_cast_fp16")]; + tensor var_14785_interleave_0 = const()[name = tensor("op_14785_interleave_0"), val = tensor(false)]; + tensor var_14785_cast_fp16 = concat(axis = var_13254, interleave = var_14785_interleave_0, values = (var_14733_cast_fp16, var_14735_cast_fp16, var_14737_cast_fp16, var_14739_cast_fp16))[name = tensor("op_14785_cast_fp16")]; + tensor var_14787_interleave_0 = const()[name = tensor("op_14787_interleave_0"), val = tensor(false)]; + tensor var_14787_cast_fp16 = concat(axis = var_13254, interleave = var_14787_interleave_0, values = (var_14741_cast_fp16, var_14743_cast_fp16, var_14745_cast_fp16, var_14747_cast_fp16))[name = tensor("op_14787_cast_fp16")]; + tensor input_167_interleave_0 = const()[name = tensor("input_167_interleave_0"), val = tensor(false)]; + tensor input_167_cast_fp16 = concat(axis = var_13279, interleave = input_167_interleave_0, values = (var_14749_cast_fp16, var_14751_cast_fp16, var_14753_cast_fp16, var_14755_cast_fp16, var_14757_cast_fp16, var_14759_cast_fp16, var_14761_cast_fp16, var_14763_cast_fp16, var_14765_cast_fp16, var_14767_cast_fp16, var_14769_cast_fp16, var_14771_cast_fp16, var_14773_cast_fp16, var_14775_cast_fp16, var_14777_cast_fp16, var_14779_cast_fp16, var_14781_cast_fp16, var_14783_cast_fp16, var_14785_cast_fp16, var_14787_cast_fp16))[name = tensor("input_167_cast_fp16")]; + tensor var_14795 = const()[name = tensor("op_14795"), val = tensor([1, 1])]; + tensor var_14797 = const()[name = tensor("op_14797"), val = tensor([1, 1])]; + tensor pretrained_out_103_pad_type_0 = const()[name = tensor("pretrained_out_103_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_103_pad_0 = const()[name = tensor("pretrained_out_103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(101827840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102647104))), name = tensor("layers_8_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_8_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102647232)))]; + tensor pretrained_out_103_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_14797, groups = var_13279, pad = pretrained_out_103_pad_0, pad_type = pretrained_out_103_pad_type_0, strides = var_14795, weight = layers_8_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("pretrained_out_103_cast_fp16")]; + tensor var_14801 = const()[name = tensor("op_14801"), val = tensor([1, 1])]; + tensor var_14803 = const()[name = tensor("op_14803"), val = tensor([1, 1])]; + tensor input_169_pad_type_0 = const()[name = tensor("input_169_pad_type_0"), val = tensor("custom")]; + tensor input_169_pad_0 = const()[name = tensor("input_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102649856)))]; + tensor input_169_cast_fp16 = conv(dilations = var_14803, groups = var_13279, pad = input_169_pad_0, pad_type = input_169_pad_type_0, strides = var_14801, weight = layers_8_self_attn_o_proj_loraA_weight_to_fp16, x = input_167_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor var_14807 = const()[name = tensor("op_14807"), val = tensor([1, 1])]; + tensor var_14809 = const()[name = tensor("op_14809"), val = tensor([1, 1])]; + tensor lora_out_205_pad_type_0 = const()[name = tensor("lora_out_205_pad_type_0"), val = tensor("custom")]; + tensor lora_out_205_pad_0 = const()[name = tensor("lora_out_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_207_weight_0_to_fp16 = const()[name = tensor("lora_out_207_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102690880)))]; + tensor lora_out_207_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_14809, groups = var_13279, pad = lora_out_205_pad_0, pad_type = lora_out_205_pad_type_0, strides = var_14807, weight = lora_out_207_weight_0_to_fp16, x = input_169_cast_fp16)[name = tensor("lora_out_207_cast_fp16")]; + tensor obj_35_cast_fp16 = add(x = pretrained_out_103_cast_fp16, y = lora_out_207_cast_fp16)[name = tensor("obj_35_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor var_14818 = const()[name = tensor("op_14818"), val = tensor([1])]; + tensor channels_mean_35_cast_fp16 = reduce_mean(axes = var_14818, keep_dims = var_13280, x = inputs_35_cast_fp16)[name = tensor("channels_mean_35_cast_fp16")]; + tensor zero_mean_35_cast_fp16 = sub(x = inputs_35_cast_fp16, y = channels_mean_35_cast_fp16)[name = tensor("zero_mean_35_cast_fp16")]; + tensor zero_mean_sq_35_cast_fp16 = mul(x = zero_mean_35_cast_fp16, y = zero_mean_35_cast_fp16)[name = tensor("zero_mean_sq_35_cast_fp16")]; + tensor var_14822 = const()[name = tensor("op_14822"), val = tensor([1])]; + tensor var_14823_cast_fp16 = reduce_mean(axes = var_14822, keep_dims = var_13280, x = zero_mean_sq_35_cast_fp16)[name = tensor("op_14823_cast_fp16")]; + tensor var_14824_to_fp16 = const()[name = tensor("op_14824_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_14825_cast_fp16 = add(x = var_14823_cast_fp16, y = var_14824_to_fp16)[name = tensor("op_14825_cast_fp16")]; + tensor denom_35_epsilon_0 = const()[name = tensor("denom_35_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_35_cast_fp16 = rsqrt(epsilon = denom_35_epsilon_0, x = var_14825_cast_fp16)[name = tensor("denom_35_cast_fp16")]; + tensor out_35_cast_fp16 = mul(x = zero_mean_35_cast_fp16, y = denom_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; + tensor input_171_gamma_0_to_fp16 = const()[name = tensor("input_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102731904)))]; + tensor input_171_beta_0_to_fp16 = const()[name = tensor("input_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102734528)))]; + tensor input_171_epsilon_0_to_fp16 = const()[name = tensor("input_171_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_171_cast_fp16 = batch_norm(beta = input_171_beta_0_to_fp16, epsilon = input_171_epsilon_0_to_fp16, gamma = input_171_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor var_14839 = const()[name = tensor("op_14839"), val = tensor([1, 1])]; + tensor var_14841 = const()[name = tensor("op_14841"), val = tensor([1, 1])]; + tensor pretrained_out_105_pad_type_0 = const()[name = tensor("pretrained_out_105_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_105_pad_0 = const()[name = tensor("pretrained_out_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102737152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106014016))), name = tensor("layers_8_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_8_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_8_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106014144)))]; + tensor pretrained_out_105_cast_fp16 = conv(bias = layers_8_fc1_pretrained_bias_to_fp16, dilations = var_14841, groups = var_13279, pad = pretrained_out_105_pad_0, pad_type = pretrained_out_105_pad_type_0, strides = var_14839, weight = layers_8_fc1_pretrained_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("pretrained_out_105_cast_fp16")]; + tensor var_14845 = const()[name = tensor("op_14845"), val = tensor([1, 1])]; + tensor var_14847 = const()[name = tensor("op_14847"), val = tensor([1, 1])]; + tensor input_173_pad_type_0 = const()[name = tensor("input_173_pad_type_0"), val = tensor("custom")]; + tensor input_173_pad_0 = const()[name = tensor("input_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_8_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106024448)))]; + tensor input_173_cast_fp16 = conv(dilations = var_14847, groups = var_13279, pad = input_173_pad_0, pad_type = input_173_pad_type_0, strides = var_14845, weight = layers_8_fc1_loraA_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor var_14851 = const()[name = tensor("op_14851"), val = tensor([1, 1])]; + tensor var_14853 = const()[name = tensor("op_14853"), val = tensor([1, 1])]; + tensor lora_out_209_pad_type_0 = const()[name = tensor("lora_out_209_pad_type_0"), val = tensor("custom")]; + tensor lora_out_209_pad_0 = const()[name = tensor("lora_out_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_211_weight_0_to_fp16 = const()[name = tensor("lora_out_211_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106065472)))]; + tensor lora_out_211_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_14853, groups = var_13279, pad = lora_out_209_pad_0, pad_type = lora_out_209_pad_type_0, strides = var_14851, weight = lora_out_211_weight_0_to_fp16, x = input_173_cast_fp16)[name = tensor("lora_out_211_cast_fp16")]; + tensor input_175_cast_fp16 = add(x = pretrained_out_105_cast_fp16, y = lora_out_211_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor input_177_mode_0 = const()[name = tensor("input_177_mode_0"), val = tensor("EXACT")]; + tensor input_177_cast_fp16 = gelu(mode = input_177_mode_0, x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor var_14865 = const()[name = tensor("op_14865"), val = tensor([1, 1])]; + tensor var_14867 = const()[name = tensor("op_14867"), val = tensor([1, 1])]; + tensor pretrained_out_107_pad_type_0 = const()[name = tensor("pretrained_out_107_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_107_pad_0 = const()[name = tensor("pretrained_out_107_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106229376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109506240))), name = tensor("layers_8_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_8_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_8_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109506368)))]; + tensor pretrained_out_107_cast_fp16 = conv(bias = layers_8_fc2_pretrained_bias_to_fp16, dilations = var_14867, groups = var_13279, pad = pretrained_out_107_pad_0, pad_type = pretrained_out_107_pad_type_0, strides = var_14865, weight = layers_8_fc2_pretrained_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("pretrained_out_107_cast_fp16")]; + tensor var_14871 = const()[name = tensor("op_14871"), val = tensor([1, 1])]; + tensor var_14873 = const()[name = tensor("op_14873"), val = tensor([1, 1])]; + tensor input_179_pad_type_0 = const()[name = tensor("input_179_pad_type_0"), val = tensor("custom")]; + tensor input_179_pad_0 = const()[name = tensor("input_179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_8_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_8_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109508992)))]; + tensor input_179_cast_fp16 = conv(dilations = var_14873, groups = var_13279, pad = input_179_pad_0, pad_type = input_179_pad_type_0, strides = var_14871, weight = layers_8_fc2_loraA_weight_to_fp16, x = input_177_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor var_14877 = const()[name = tensor("op_14877"), val = tensor([1, 1])]; + tensor var_14879 = const()[name = tensor("op_14879"), val = tensor([1, 1])]; + tensor lora_out_213_pad_type_0 = const()[name = tensor("lora_out_213_pad_type_0"), val = tensor("custom")]; + tensor lora_out_213_pad_0 = const()[name = tensor("lora_out_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_215_weight_0_to_fp16 = const()[name = tensor("lora_out_215_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109672896)))]; + tensor lora_out_215_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_14879, groups = var_13279, pad = lora_out_213_pad_0, pad_type = lora_out_213_pad_type_0, strides = var_14877, weight = lora_out_215_weight_0_to_fp16, x = input_179_cast_fp16)[name = tensor("lora_out_215_cast_fp16")]; + tensor hidden_states_21_cast_fp16 = add(x = pretrained_out_107_cast_fp16, y = lora_out_215_cast_fp16)[name = tensor("hidden_states_21_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor var_14889 = const()[name = tensor("op_14889"), val = tensor(3)]; + tensor var_14914 = const()[name = tensor("op_14914"), val = tensor(1)]; + tensor var_14915 = const()[name = tensor("op_14915"), val = tensor(true)]; + tensor var_14925 = const()[name = tensor("op_14925"), val = tensor([1])]; + tensor channels_mean_37_cast_fp16 = reduce_mean(axes = var_14925, keep_dims = var_14915, x = inputs_37_cast_fp16)[name = tensor("channels_mean_37_cast_fp16")]; + tensor zero_mean_37_cast_fp16 = sub(x = inputs_37_cast_fp16, y = channels_mean_37_cast_fp16)[name = tensor("zero_mean_37_cast_fp16")]; + tensor zero_mean_sq_37_cast_fp16 = mul(x = zero_mean_37_cast_fp16, y = zero_mean_37_cast_fp16)[name = tensor("zero_mean_sq_37_cast_fp16")]; + tensor var_14929 = const()[name = tensor("op_14929"), val = tensor([1])]; + tensor var_14930_cast_fp16 = reduce_mean(axes = var_14929, keep_dims = var_14915, x = zero_mean_sq_37_cast_fp16)[name = tensor("op_14930_cast_fp16")]; + tensor var_14931_to_fp16 = const()[name = tensor("op_14931_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_14932_cast_fp16 = add(x = var_14930_cast_fp16, y = var_14931_to_fp16)[name = tensor("op_14932_cast_fp16")]; + tensor denom_37_epsilon_0 = const()[name = tensor("denom_37_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_37_cast_fp16 = rsqrt(epsilon = denom_37_epsilon_0, x = var_14932_cast_fp16)[name = tensor("denom_37_cast_fp16")]; + tensor out_37_cast_fp16 = mul(x = zero_mean_37_cast_fp16, y = denom_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; + tensor obj_37_gamma_0_to_fp16 = const()[name = tensor("obj_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109713920)))]; + tensor obj_37_beta_0_to_fp16 = const()[name = tensor("obj_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109716544)))]; + tensor obj_37_epsilon_0_to_fp16 = const()[name = tensor("obj_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor("obj_37_cast_fp16")]; + tensor var_14950 = const()[name = tensor("op_14950"), val = tensor([1, 1])]; + tensor var_14952 = const()[name = tensor("op_14952"), val = tensor([1, 1])]; + tensor pretrained_out_109_pad_type_0 = const()[name = tensor("pretrained_out_109_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_109_pad_0 = const()[name = tensor("pretrained_out_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109719168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110538432))), name = tensor("layers_9_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_9_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110538560)))]; + tensor pretrained_out_109_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_14952, groups = var_14914, pad = pretrained_out_109_pad_0, pad_type = pretrained_out_109_pad_type_0, strides = var_14950, weight = layers_9_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor("pretrained_out_109_cast_fp16")]; + tensor var_14956 = const()[name = tensor("op_14956"), val = tensor([1, 1])]; + tensor var_14958 = const()[name = tensor("op_14958"), val = tensor([1, 1])]; + tensor input_181_pad_type_0 = const()[name = tensor("input_181_pad_type_0"), val = tensor("custom")]; + tensor input_181_pad_0 = const()[name = tensor("input_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110541184)))]; + tensor input_181_cast_fp16 = conv(dilations = var_14958, groups = var_14914, pad = input_181_pad_0, pad_type = input_181_pad_type_0, strides = var_14956, weight = layers_9_self_attn_q_proj_loraA_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor var_14962 = const()[name = tensor("op_14962"), val = tensor([1, 1])]; + tensor var_14964 = const()[name = tensor("op_14964"), val = tensor([1, 1])]; + tensor lora_out_217_pad_type_0 = const()[name = tensor("lora_out_217_pad_type_0"), val = tensor("custom")]; + tensor lora_out_217_pad_0 = const()[name = tensor("lora_out_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_219_weight_0_to_fp16 = const()[name = tensor("lora_out_219_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110582208)))]; + tensor lora_out_219_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_14964, groups = var_14914, pad = lora_out_217_pad_0, pad_type = lora_out_217_pad_type_0, strides = var_14962, weight = lora_out_219_weight_0_to_fp16, x = input_181_cast_fp16)[name = tensor("lora_out_219_cast_fp16")]; + tensor query_19_cast_fp16 = add(x = pretrained_out_109_cast_fp16, y = lora_out_219_cast_fp16)[name = tensor("query_19_cast_fp16")]; + tensor var_14974 = const()[name = tensor("op_14974"), val = tensor([1, 1])]; + tensor var_14976 = const()[name = tensor("op_14976"), val = tensor([1, 1])]; + tensor pretrained_out_111_pad_type_0 = const()[name = tensor("pretrained_out_111_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_111_pad_0 = const()[name = tensor("pretrained_out_111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110623232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111442496))), name = tensor("layers_9_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_111_cast_fp16 = conv(dilations = var_14976, groups = var_14914, pad = pretrained_out_111_pad_0, pad_type = pretrained_out_111_pad_type_0, strides = var_14974, weight = layers_9_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor("pretrained_out_111_cast_fp16")]; + tensor var_14980 = const()[name = tensor("op_14980"), val = tensor([1, 1])]; + tensor var_14982 = const()[name = tensor("op_14982"), val = tensor([1, 1])]; + tensor input_183_pad_type_0 = const()[name = tensor("input_183_pad_type_0"), val = tensor("custom")]; + tensor input_183_pad_0 = const()[name = tensor("input_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111442624)))]; + tensor input_183_cast_fp16 = conv(dilations = var_14982, groups = var_14914, pad = input_183_pad_0, pad_type = input_183_pad_type_0, strides = var_14980, weight = layers_9_self_attn_k_proj_loraA_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor var_14986 = const()[name = tensor("op_14986"), val = tensor([1, 1])]; + tensor var_14988 = const()[name = tensor("op_14988"), val = tensor([1, 1])]; + tensor lora_out_221_pad_type_0 = const()[name = tensor("lora_out_221_pad_type_0"), val = tensor("custom")]; + tensor lora_out_221_pad_0 = const()[name = tensor("lora_out_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_223_weight_0_to_fp16 = const()[name = tensor("lora_out_223_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111483648)))]; + tensor lora_out_223_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_14988, groups = var_14914, pad = lora_out_221_pad_0, pad_type = lora_out_221_pad_type_0, strides = var_14986, weight = lora_out_223_weight_0_to_fp16, x = input_183_cast_fp16)[name = tensor("lora_out_223_cast_fp16")]; + tensor key_19_cast_fp16 = add(x = pretrained_out_111_cast_fp16, y = lora_out_223_cast_fp16)[name = tensor("key_19_cast_fp16")]; + tensor var_14999 = const()[name = tensor("op_14999"), val = tensor([1, 1])]; + tensor var_15001 = const()[name = tensor("op_15001"), val = tensor([1, 1])]; + tensor pretrained_out_113_pad_type_0 = const()[name = tensor("pretrained_out_113_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_113_pad_0 = const()[name = tensor("pretrained_out_113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111524672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112343936))), name = tensor("layers_9_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_9_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112344064)))]; + tensor pretrained_out_113_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_15001, groups = var_14914, pad = pretrained_out_113_pad_0, pad_type = pretrained_out_113_pad_type_0, strides = var_14999, weight = layers_9_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor("pretrained_out_113_cast_fp16")]; + tensor var_15005 = const()[name = tensor("op_15005"), val = tensor([1, 1])]; + tensor var_15007 = const()[name = tensor("op_15007"), val = tensor([1, 1])]; + tensor input_185_pad_type_0 = const()[name = tensor("input_185_pad_type_0"), val = tensor("custom")]; + tensor input_185_pad_0 = const()[name = tensor("input_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112346688)))]; + tensor input_185_cast_fp16 = conv(dilations = var_15007, groups = var_14914, pad = input_185_pad_0, pad_type = input_185_pad_type_0, strides = var_15005, weight = layers_9_self_attn_v_proj_loraA_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor var_15011 = const()[name = tensor("op_15011"), val = tensor([1, 1])]; + tensor var_15013 = const()[name = tensor("op_15013"), val = tensor([1, 1])]; + tensor lora_out_225_pad_type_0 = const()[name = tensor("lora_out_225_pad_type_0"), val = tensor("custom")]; + tensor lora_out_225_pad_0 = const()[name = tensor("lora_out_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_227_weight_0_to_fp16 = const()[name = tensor("lora_out_227_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112387712)))]; + tensor lora_out_227_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_15013, groups = var_14914, pad = lora_out_225_pad_0, pad_type = lora_out_225_pad_type_0, strides = var_15011, weight = lora_out_227_weight_0_to_fp16, x = input_185_cast_fp16)[name = tensor("lora_out_227_cast_fp16")]; + tensor value_19_cast_fp16 = add(x = pretrained_out_113_cast_fp16, y = lora_out_227_cast_fp16)[name = tensor("value_19_cast_fp16")]; + tensor var_15023_begin_0 = const()[name = tensor("op_15023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15023_end_0 = const()[name = tensor("op_15023_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15023_end_mask_0 = const()[name = tensor("op_15023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15023_cast_fp16 = slice_by_index(begin = var_15023_begin_0, end = var_15023_end_0, end_mask = var_15023_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15023_cast_fp16")]; + tensor var_15027_begin_0 = const()[name = tensor("op_15027_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15027_end_0 = const()[name = tensor("op_15027_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_15027_end_mask_0 = const()[name = tensor("op_15027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15027_cast_fp16 = slice_by_index(begin = var_15027_begin_0, end = var_15027_end_0, end_mask = var_15027_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15027_cast_fp16")]; + tensor var_15031_begin_0 = const()[name = tensor("op_15031_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15031_end_0 = const()[name = tensor("op_15031_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_15031_end_mask_0 = const()[name = tensor("op_15031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15031_cast_fp16 = slice_by_index(begin = var_15031_begin_0, end = var_15031_end_0, end_mask = var_15031_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15031_cast_fp16")]; + tensor var_15035_begin_0 = const()[name = tensor("op_15035_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15035_end_0 = const()[name = tensor("op_15035_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_15035_end_mask_0 = const()[name = tensor("op_15035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15035_cast_fp16 = slice_by_index(begin = var_15035_begin_0, end = var_15035_end_0, end_mask = var_15035_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15035_cast_fp16")]; + tensor var_15039_begin_0 = const()[name = tensor("op_15039_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15039_end_0 = const()[name = tensor("op_15039_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_15039_end_mask_0 = const()[name = tensor("op_15039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15039_cast_fp16 = slice_by_index(begin = var_15039_begin_0, end = var_15039_end_0, end_mask = var_15039_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15039_cast_fp16")]; + tensor var_15043_begin_0 = const()[name = tensor("op_15043_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15043_end_0 = const()[name = tensor("op_15043_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_15043_end_mask_0 = const()[name = tensor("op_15043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15043_cast_fp16 = slice_by_index(begin = var_15043_begin_0, end = var_15043_end_0, end_mask = var_15043_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15043_cast_fp16")]; + tensor var_15047_begin_0 = const()[name = tensor("op_15047_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15047_end_0 = const()[name = tensor("op_15047_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_15047_end_mask_0 = const()[name = tensor("op_15047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15047_cast_fp16 = slice_by_index(begin = var_15047_begin_0, end = var_15047_end_0, end_mask = var_15047_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15047_cast_fp16")]; + tensor var_15051_begin_0 = const()[name = tensor("op_15051_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15051_end_0 = const()[name = tensor("op_15051_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_15051_end_mask_0 = const()[name = tensor("op_15051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15051_cast_fp16 = slice_by_index(begin = var_15051_begin_0, end = var_15051_end_0, end_mask = var_15051_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15051_cast_fp16")]; + tensor var_15055_begin_0 = const()[name = tensor("op_15055_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15055_end_0 = const()[name = tensor("op_15055_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_15055_end_mask_0 = const()[name = tensor("op_15055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15055_cast_fp16 = slice_by_index(begin = var_15055_begin_0, end = var_15055_end_0, end_mask = var_15055_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15055_cast_fp16")]; + tensor var_15059_begin_0 = const()[name = tensor("op_15059_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15059_end_0 = const()[name = tensor("op_15059_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_15059_end_mask_0 = const()[name = tensor("op_15059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15059_cast_fp16 = slice_by_index(begin = var_15059_begin_0, end = var_15059_end_0, end_mask = var_15059_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15059_cast_fp16")]; + tensor var_15063_begin_0 = const()[name = tensor("op_15063_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15063_end_0 = const()[name = tensor("op_15063_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_15063_end_mask_0 = const()[name = tensor("op_15063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15063_cast_fp16 = slice_by_index(begin = var_15063_begin_0, end = var_15063_end_0, end_mask = var_15063_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15063_cast_fp16")]; + tensor var_15067_begin_0 = const()[name = tensor("op_15067_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15067_end_0 = const()[name = tensor("op_15067_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_15067_end_mask_0 = const()[name = tensor("op_15067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15067_cast_fp16 = slice_by_index(begin = var_15067_begin_0, end = var_15067_end_0, end_mask = var_15067_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15067_cast_fp16")]; + tensor var_15071_begin_0 = const()[name = tensor("op_15071_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15071_end_0 = const()[name = tensor("op_15071_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_15071_end_mask_0 = const()[name = tensor("op_15071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15071_cast_fp16 = slice_by_index(begin = var_15071_begin_0, end = var_15071_end_0, end_mask = var_15071_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15071_cast_fp16")]; + tensor var_15075_begin_0 = const()[name = tensor("op_15075_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15075_end_0 = const()[name = tensor("op_15075_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_15075_end_mask_0 = const()[name = tensor("op_15075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15075_cast_fp16 = slice_by_index(begin = var_15075_begin_0, end = var_15075_end_0, end_mask = var_15075_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15075_cast_fp16")]; + tensor var_15079_begin_0 = const()[name = tensor("op_15079_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15079_end_0 = const()[name = tensor("op_15079_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_15079_end_mask_0 = const()[name = tensor("op_15079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15079_cast_fp16 = slice_by_index(begin = var_15079_begin_0, end = var_15079_end_0, end_mask = var_15079_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15079_cast_fp16")]; + tensor var_15083_begin_0 = const()[name = tensor("op_15083_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15083_end_0 = const()[name = tensor("op_15083_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_15083_end_mask_0 = const()[name = tensor("op_15083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15083_cast_fp16 = slice_by_index(begin = var_15083_begin_0, end = var_15083_end_0, end_mask = var_15083_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15083_cast_fp16")]; + tensor var_15087_begin_0 = const()[name = tensor("op_15087_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15087_end_0 = const()[name = tensor("op_15087_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_15087_end_mask_0 = const()[name = tensor("op_15087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15087_cast_fp16 = slice_by_index(begin = var_15087_begin_0, end = var_15087_end_0, end_mask = var_15087_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15087_cast_fp16")]; + tensor var_15091_begin_0 = const()[name = tensor("op_15091_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15091_end_0 = const()[name = tensor("op_15091_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_15091_end_mask_0 = const()[name = tensor("op_15091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15091_cast_fp16 = slice_by_index(begin = var_15091_begin_0, end = var_15091_end_0, end_mask = var_15091_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15091_cast_fp16")]; + tensor var_15095_begin_0 = const()[name = tensor("op_15095_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15095_end_0 = const()[name = tensor("op_15095_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_15095_end_mask_0 = const()[name = tensor("op_15095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15095_cast_fp16 = slice_by_index(begin = var_15095_begin_0, end = var_15095_end_0, end_mask = var_15095_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15095_cast_fp16")]; + tensor var_15099_begin_0 = const()[name = tensor("op_15099_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15099_end_0 = const()[name = tensor("op_15099_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_15099_end_mask_0 = const()[name = tensor("op_15099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15099_cast_fp16 = slice_by_index(begin = var_15099_begin_0, end = var_15099_end_0, end_mask = var_15099_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_15099_cast_fp16")]; + tensor var_15108_begin_0 = const()[name = tensor("op_15108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15108_end_0 = const()[name = tensor("op_15108_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15108_end_mask_0 = const()[name = tensor("op_15108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15108_cast_fp16 = slice_by_index(begin = var_15108_begin_0, end = var_15108_end_0, end_mask = var_15108_end_mask_0, x = var_15023_cast_fp16)[name = tensor("op_15108_cast_fp16")]; + tensor var_15115_begin_0 = const()[name = tensor("op_15115_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15115_end_0 = const()[name = tensor("op_15115_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15115_end_mask_0 = const()[name = tensor("op_15115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15115_cast_fp16 = slice_by_index(begin = var_15115_begin_0, end = var_15115_end_0, end_mask = var_15115_end_mask_0, x = var_15023_cast_fp16)[name = tensor("op_15115_cast_fp16")]; + tensor var_15122_begin_0 = const()[name = tensor("op_15122_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15122_end_0 = const()[name = tensor("op_15122_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15122_end_mask_0 = const()[name = tensor("op_15122_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15122_cast_fp16 = slice_by_index(begin = var_15122_begin_0, end = var_15122_end_0, end_mask = var_15122_end_mask_0, x = var_15023_cast_fp16)[name = tensor("op_15122_cast_fp16")]; + tensor var_15129_begin_0 = const()[name = tensor("op_15129_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15129_end_0 = const()[name = tensor("op_15129_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15129_end_mask_0 = const()[name = tensor("op_15129_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15129_cast_fp16 = slice_by_index(begin = var_15129_begin_0, end = var_15129_end_0, end_mask = var_15129_end_mask_0, x = var_15023_cast_fp16)[name = tensor("op_15129_cast_fp16")]; + tensor var_15136_begin_0 = const()[name = tensor("op_15136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15136_end_0 = const()[name = tensor("op_15136_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15136_end_mask_0 = const()[name = tensor("op_15136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15136_cast_fp16 = slice_by_index(begin = var_15136_begin_0, end = var_15136_end_0, end_mask = var_15136_end_mask_0, x = var_15027_cast_fp16)[name = tensor("op_15136_cast_fp16")]; + tensor var_15143_begin_0 = const()[name = tensor("op_15143_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15143_end_0 = const()[name = tensor("op_15143_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15143_end_mask_0 = const()[name = tensor("op_15143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15143_cast_fp16 = slice_by_index(begin = var_15143_begin_0, end = var_15143_end_0, end_mask = var_15143_end_mask_0, x = var_15027_cast_fp16)[name = tensor("op_15143_cast_fp16")]; + tensor var_15150_begin_0 = const()[name = tensor("op_15150_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15150_end_0 = const()[name = tensor("op_15150_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15150_end_mask_0 = const()[name = tensor("op_15150_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15150_cast_fp16 = slice_by_index(begin = var_15150_begin_0, end = var_15150_end_0, end_mask = var_15150_end_mask_0, x = var_15027_cast_fp16)[name = tensor("op_15150_cast_fp16")]; + tensor var_15157_begin_0 = const()[name = tensor("op_15157_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15157_end_0 = const()[name = tensor("op_15157_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15157_end_mask_0 = const()[name = tensor("op_15157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15157_cast_fp16 = slice_by_index(begin = var_15157_begin_0, end = var_15157_end_0, end_mask = var_15157_end_mask_0, x = var_15027_cast_fp16)[name = tensor("op_15157_cast_fp16")]; + tensor var_15164_begin_0 = const()[name = tensor("op_15164_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15164_end_0 = const()[name = tensor("op_15164_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15164_end_mask_0 = const()[name = tensor("op_15164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15164_cast_fp16 = slice_by_index(begin = var_15164_begin_0, end = var_15164_end_0, end_mask = var_15164_end_mask_0, x = var_15031_cast_fp16)[name = tensor("op_15164_cast_fp16")]; + tensor var_15171_begin_0 = const()[name = tensor("op_15171_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15171_end_0 = const()[name = tensor("op_15171_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15171_end_mask_0 = const()[name = tensor("op_15171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15171_cast_fp16 = slice_by_index(begin = var_15171_begin_0, end = var_15171_end_0, end_mask = var_15171_end_mask_0, x = var_15031_cast_fp16)[name = tensor("op_15171_cast_fp16")]; + tensor var_15178_begin_0 = const()[name = tensor("op_15178_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15178_end_0 = const()[name = tensor("op_15178_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15178_end_mask_0 = const()[name = tensor("op_15178_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15178_cast_fp16 = slice_by_index(begin = var_15178_begin_0, end = var_15178_end_0, end_mask = var_15178_end_mask_0, x = var_15031_cast_fp16)[name = tensor("op_15178_cast_fp16")]; + tensor var_15185_begin_0 = const()[name = tensor("op_15185_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15185_end_0 = const()[name = tensor("op_15185_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15185_end_mask_0 = const()[name = tensor("op_15185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15185_cast_fp16 = slice_by_index(begin = var_15185_begin_0, end = var_15185_end_0, end_mask = var_15185_end_mask_0, x = var_15031_cast_fp16)[name = tensor("op_15185_cast_fp16")]; + tensor var_15192_begin_0 = const()[name = tensor("op_15192_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15192_end_0 = const()[name = tensor("op_15192_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15192_end_mask_0 = const()[name = tensor("op_15192_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15192_cast_fp16 = slice_by_index(begin = var_15192_begin_0, end = var_15192_end_0, end_mask = var_15192_end_mask_0, x = var_15035_cast_fp16)[name = tensor("op_15192_cast_fp16")]; + tensor var_15199_begin_0 = const()[name = tensor("op_15199_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15199_end_0 = const()[name = tensor("op_15199_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15199_end_mask_0 = const()[name = tensor("op_15199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15199_cast_fp16 = slice_by_index(begin = var_15199_begin_0, end = var_15199_end_0, end_mask = var_15199_end_mask_0, x = var_15035_cast_fp16)[name = tensor("op_15199_cast_fp16")]; + tensor var_15206_begin_0 = const()[name = tensor("op_15206_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15206_end_0 = const()[name = tensor("op_15206_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15206_end_mask_0 = const()[name = tensor("op_15206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15206_cast_fp16 = slice_by_index(begin = var_15206_begin_0, end = var_15206_end_0, end_mask = var_15206_end_mask_0, x = var_15035_cast_fp16)[name = tensor("op_15206_cast_fp16")]; + tensor var_15213_begin_0 = const()[name = tensor("op_15213_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15213_end_0 = const()[name = tensor("op_15213_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15213_end_mask_0 = const()[name = tensor("op_15213_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15213_cast_fp16 = slice_by_index(begin = var_15213_begin_0, end = var_15213_end_0, end_mask = var_15213_end_mask_0, x = var_15035_cast_fp16)[name = tensor("op_15213_cast_fp16")]; + tensor var_15220_begin_0 = const()[name = tensor("op_15220_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15220_end_0 = const()[name = tensor("op_15220_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15220_end_mask_0 = const()[name = tensor("op_15220_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15220_cast_fp16 = slice_by_index(begin = var_15220_begin_0, end = var_15220_end_0, end_mask = var_15220_end_mask_0, x = var_15039_cast_fp16)[name = tensor("op_15220_cast_fp16")]; + tensor var_15227_begin_0 = const()[name = tensor("op_15227_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15227_end_0 = const()[name = tensor("op_15227_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15227_end_mask_0 = const()[name = tensor("op_15227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15227_cast_fp16 = slice_by_index(begin = var_15227_begin_0, end = var_15227_end_0, end_mask = var_15227_end_mask_0, x = var_15039_cast_fp16)[name = tensor("op_15227_cast_fp16")]; + tensor var_15234_begin_0 = const()[name = tensor("op_15234_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15234_end_0 = const()[name = tensor("op_15234_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15234_end_mask_0 = const()[name = tensor("op_15234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15234_cast_fp16 = slice_by_index(begin = var_15234_begin_0, end = var_15234_end_0, end_mask = var_15234_end_mask_0, x = var_15039_cast_fp16)[name = tensor("op_15234_cast_fp16")]; + tensor var_15241_begin_0 = const()[name = tensor("op_15241_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15241_end_0 = const()[name = tensor("op_15241_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15241_end_mask_0 = const()[name = tensor("op_15241_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15241_cast_fp16 = slice_by_index(begin = var_15241_begin_0, end = var_15241_end_0, end_mask = var_15241_end_mask_0, x = var_15039_cast_fp16)[name = tensor("op_15241_cast_fp16")]; + tensor var_15248_begin_0 = const()[name = tensor("op_15248_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15248_end_0 = const()[name = tensor("op_15248_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15248_end_mask_0 = const()[name = tensor("op_15248_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15248_cast_fp16 = slice_by_index(begin = var_15248_begin_0, end = var_15248_end_0, end_mask = var_15248_end_mask_0, x = var_15043_cast_fp16)[name = tensor("op_15248_cast_fp16")]; + tensor var_15255_begin_0 = const()[name = tensor("op_15255_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15255_end_0 = const()[name = tensor("op_15255_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15255_end_mask_0 = const()[name = tensor("op_15255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15255_cast_fp16 = slice_by_index(begin = var_15255_begin_0, end = var_15255_end_0, end_mask = var_15255_end_mask_0, x = var_15043_cast_fp16)[name = tensor("op_15255_cast_fp16")]; + tensor var_15262_begin_0 = const()[name = tensor("op_15262_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15262_end_0 = const()[name = tensor("op_15262_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15262_end_mask_0 = const()[name = tensor("op_15262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15262_cast_fp16 = slice_by_index(begin = var_15262_begin_0, end = var_15262_end_0, end_mask = var_15262_end_mask_0, x = var_15043_cast_fp16)[name = tensor("op_15262_cast_fp16")]; + tensor var_15269_begin_0 = const()[name = tensor("op_15269_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15269_end_0 = const()[name = tensor("op_15269_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15269_end_mask_0 = const()[name = tensor("op_15269_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15269_cast_fp16 = slice_by_index(begin = var_15269_begin_0, end = var_15269_end_0, end_mask = var_15269_end_mask_0, x = var_15043_cast_fp16)[name = tensor("op_15269_cast_fp16")]; + tensor var_15276_begin_0 = const()[name = tensor("op_15276_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15276_end_0 = const()[name = tensor("op_15276_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15276_end_mask_0 = const()[name = tensor("op_15276_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15276_cast_fp16 = slice_by_index(begin = var_15276_begin_0, end = var_15276_end_0, end_mask = var_15276_end_mask_0, x = var_15047_cast_fp16)[name = tensor("op_15276_cast_fp16")]; + tensor var_15283_begin_0 = const()[name = tensor("op_15283_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15283_end_0 = const()[name = tensor("op_15283_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15283_end_mask_0 = const()[name = tensor("op_15283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15283_cast_fp16 = slice_by_index(begin = var_15283_begin_0, end = var_15283_end_0, end_mask = var_15283_end_mask_0, x = var_15047_cast_fp16)[name = tensor("op_15283_cast_fp16")]; + tensor var_15290_begin_0 = const()[name = tensor("op_15290_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15290_end_0 = const()[name = tensor("op_15290_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15290_end_mask_0 = const()[name = tensor("op_15290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15290_cast_fp16 = slice_by_index(begin = var_15290_begin_0, end = var_15290_end_0, end_mask = var_15290_end_mask_0, x = var_15047_cast_fp16)[name = tensor("op_15290_cast_fp16")]; + tensor var_15297_begin_0 = const()[name = tensor("op_15297_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15297_end_0 = const()[name = tensor("op_15297_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15297_end_mask_0 = const()[name = tensor("op_15297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15297_cast_fp16 = slice_by_index(begin = var_15297_begin_0, end = var_15297_end_0, end_mask = var_15297_end_mask_0, x = var_15047_cast_fp16)[name = tensor("op_15297_cast_fp16")]; + tensor var_15304_begin_0 = const()[name = tensor("op_15304_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15304_end_0 = const()[name = tensor("op_15304_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15304_end_mask_0 = const()[name = tensor("op_15304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15304_cast_fp16 = slice_by_index(begin = var_15304_begin_0, end = var_15304_end_0, end_mask = var_15304_end_mask_0, x = var_15051_cast_fp16)[name = tensor("op_15304_cast_fp16")]; + tensor var_15311_begin_0 = const()[name = tensor("op_15311_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15311_end_0 = const()[name = tensor("op_15311_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15311_end_mask_0 = const()[name = tensor("op_15311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15311_cast_fp16 = slice_by_index(begin = var_15311_begin_0, end = var_15311_end_0, end_mask = var_15311_end_mask_0, x = var_15051_cast_fp16)[name = tensor("op_15311_cast_fp16")]; + tensor var_15318_begin_0 = const()[name = tensor("op_15318_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15318_end_0 = const()[name = tensor("op_15318_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15318_end_mask_0 = const()[name = tensor("op_15318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15318_cast_fp16 = slice_by_index(begin = var_15318_begin_0, end = var_15318_end_0, end_mask = var_15318_end_mask_0, x = var_15051_cast_fp16)[name = tensor("op_15318_cast_fp16")]; + tensor var_15325_begin_0 = const()[name = tensor("op_15325_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15325_end_0 = const()[name = tensor("op_15325_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15325_end_mask_0 = const()[name = tensor("op_15325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15325_cast_fp16 = slice_by_index(begin = var_15325_begin_0, end = var_15325_end_0, end_mask = var_15325_end_mask_0, x = var_15051_cast_fp16)[name = tensor("op_15325_cast_fp16")]; + tensor var_15332_begin_0 = const()[name = tensor("op_15332_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15332_end_0 = const()[name = tensor("op_15332_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15332_end_mask_0 = const()[name = tensor("op_15332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15332_cast_fp16 = slice_by_index(begin = var_15332_begin_0, end = var_15332_end_0, end_mask = var_15332_end_mask_0, x = var_15055_cast_fp16)[name = tensor("op_15332_cast_fp16")]; + tensor var_15339_begin_0 = const()[name = tensor("op_15339_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15339_end_0 = const()[name = tensor("op_15339_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15339_end_mask_0 = const()[name = tensor("op_15339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15339_cast_fp16 = slice_by_index(begin = var_15339_begin_0, end = var_15339_end_0, end_mask = var_15339_end_mask_0, x = var_15055_cast_fp16)[name = tensor("op_15339_cast_fp16")]; + tensor var_15346_begin_0 = const()[name = tensor("op_15346_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15346_end_0 = const()[name = tensor("op_15346_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15346_end_mask_0 = const()[name = tensor("op_15346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15346_cast_fp16 = slice_by_index(begin = var_15346_begin_0, end = var_15346_end_0, end_mask = var_15346_end_mask_0, x = var_15055_cast_fp16)[name = tensor("op_15346_cast_fp16")]; + tensor var_15353_begin_0 = const()[name = tensor("op_15353_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15353_end_0 = const()[name = tensor("op_15353_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15353_end_mask_0 = const()[name = tensor("op_15353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15353_cast_fp16 = slice_by_index(begin = var_15353_begin_0, end = var_15353_end_0, end_mask = var_15353_end_mask_0, x = var_15055_cast_fp16)[name = tensor("op_15353_cast_fp16")]; + tensor var_15360_begin_0 = const()[name = tensor("op_15360_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15360_end_0 = const()[name = tensor("op_15360_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15360_end_mask_0 = const()[name = tensor("op_15360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15360_cast_fp16 = slice_by_index(begin = var_15360_begin_0, end = var_15360_end_0, end_mask = var_15360_end_mask_0, x = var_15059_cast_fp16)[name = tensor("op_15360_cast_fp16")]; + tensor var_15367_begin_0 = const()[name = tensor("op_15367_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15367_end_0 = const()[name = tensor("op_15367_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15367_end_mask_0 = const()[name = tensor("op_15367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15367_cast_fp16 = slice_by_index(begin = var_15367_begin_0, end = var_15367_end_0, end_mask = var_15367_end_mask_0, x = var_15059_cast_fp16)[name = tensor("op_15367_cast_fp16")]; + tensor var_15374_begin_0 = const()[name = tensor("op_15374_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15374_end_0 = const()[name = tensor("op_15374_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15374_end_mask_0 = const()[name = tensor("op_15374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15374_cast_fp16 = slice_by_index(begin = var_15374_begin_0, end = var_15374_end_0, end_mask = var_15374_end_mask_0, x = var_15059_cast_fp16)[name = tensor("op_15374_cast_fp16")]; + tensor var_15381_begin_0 = const()[name = tensor("op_15381_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15381_end_0 = const()[name = tensor("op_15381_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15381_end_mask_0 = const()[name = tensor("op_15381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15381_cast_fp16 = slice_by_index(begin = var_15381_begin_0, end = var_15381_end_0, end_mask = var_15381_end_mask_0, x = var_15059_cast_fp16)[name = tensor("op_15381_cast_fp16")]; + tensor var_15388_begin_0 = const()[name = tensor("op_15388_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15388_end_0 = const()[name = tensor("op_15388_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15388_end_mask_0 = const()[name = tensor("op_15388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15388_cast_fp16 = slice_by_index(begin = var_15388_begin_0, end = var_15388_end_0, end_mask = var_15388_end_mask_0, x = var_15063_cast_fp16)[name = tensor("op_15388_cast_fp16")]; + tensor var_15395_begin_0 = const()[name = tensor("op_15395_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15395_end_0 = const()[name = tensor("op_15395_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15395_end_mask_0 = const()[name = tensor("op_15395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15395_cast_fp16 = slice_by_index(begin = var_15395_begin_0, end = var_15395_end_0, end_mask = var_15395_end_mask_0, x = var_15063_cast_fp16)[name = tensor("op_15395_cast_fp16")]; + tensor var_15402_begin_0 = const()[name = tensor("op_15402_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15402_end_0 = const()[name = tensor("op_15402_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15402_end_mask_0 = const()[name = tensor("op_15402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15402_cast_fp16 = slice_by_index(begin = var_15402_begin_0, end = var_15402_end_0, end_mask = var_15402_end_mask_0, x = var_15063_cast_fp16)[name = tensor("op_15402_cast_fp16")]; + tensor var_15409_begin_0 = const()[name = tensor("op_15409_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15409_end_0 = const()[name = tensor("op_15409_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15409_end_mask_0 = const()[name = tensor("op_15409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15409_cast_fp16 = slice_by_index(begin = var_15409_begin_0, end = var_15409_end_0, end_mask = var_15409_end_mask_0, x = var_15063_cast_fp16)[name = tensor("op_15409_cast_fp16")]; + tensor var_15416_begin_0 = const()[name = tensor("op_15416_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15416_end_0 = const()[name = tensor("op_15416_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15416_end_mask_0 = const()[name = tensor("op_15416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15416_cast_fp16 = slice_by_index(begin = var_15416_begin_0, end = var_15416_end_0, end_mask = var_15416_end_mask_0, x = var_15067_cast_fp16)[name = tensor("op_15416_cast_fp16")]; + tensor var_15423_begin_0 = const()[name = tensor("op_15423_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15423_end_0 = const()[name = tensor("op_15423_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15423_end_mask_0 = const()[name = tensor("op_15423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15423_cast_fp16 = slice_by_index(begin = var_15423_begin_0, end = var_15423_end_0, end_mask = var_15423_end_mask_0, x = var_15067_cast_fp16)[name = tensor("op_15423_cast_fp16")]; + tensor var_15430_begin_0 = const()[name = tensor("op_15430_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15430_end_0 = const()[name = tensor("op_15430_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15430_end_mask_0 = const()[name = tensor("op_15430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15430_cast_fp16 = slice_by_index(begin = var_15430_begin_0, end = var_15430_end_0, end_mask = var_15430_end_mask_0, x = var_15067_cast_fp16)[name = tensor("op_15430_cast_fp16")]; + tensor var_15437_begin_0 = const()[name = tensor("op_15437_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15437_end_0 = const()[name = tensor("op_15437_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15437_end_mask_0 = const()[name = tensor("op_15437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15437_cast_fp16 = slice_by_index(begin = var_15437_begin_0, end = var_15437_end_0, end_mask = var_15437_end_mask_0, x = var_15067_cast_fp16)[name = tensor("op_15437_cast_fp16")]; + tensor var_15444_begin_0 = const()[name = tensor("op_15444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15444_end_0 = const()[name = tensor("op_15444_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15444_end_mask_0 = const()[name = tensor("op_15444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15444_cast_fp16 = slice_by_index(begin = var_15444_begin_0, end = var_15444_end_0, end_mask = var_15444_end_mask_0, x = var_15071_cast_fp16)[name = tensor("op_15444_cast_fp16")]; + tensor var_15451_begin_0 = const()[name = tensor("op_15451_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15451_end_0 = const()[name = tensor("op_15451_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15451_end_mask_0 = const()[name = tensor("op_15451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15451_cast_fp16 = slice_by_index(begin = var_15451_begin_0, end = var_15451_end_0, end_mask = var_15451_end_mask_0, x = var_15071_cast_fp16)[name = tensor("op_15451_cast_fp16")]; + tensor var_15458_begin_0 = const()[name = tensor("op_15458_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15458_end_0 = const()[name = tensor("op_15458_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15458_end_mask_0 = const()[name = tensor("op_15458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15458_cast_fp16 = slice_by_index(begin = var_15458_begin_0, end = var_15458_end_0, end_mask = var_15458_end_mask_0, x = var_15071_cast_fp16)[name = tensor("op_15458_cast_fp16")]; + tensor var_15465_begin_0 = const()[name = tensor("op_15465_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15465_end_0 = const()[name = tensor("op_15465_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15465_end_mask_0 = const()[name = tensor("op_15465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15465_cast_fp16 = slice_by_index(begin = var_15465_begin_0, end = var_15465_end_0, end_mask = var_15465_end_mask_0, x = var_15071_cast_fp16)[name = tensor("op_15465_cast_fp16")]; + tensor var_15472_begin_0 = const()[name = tensor("op_15472_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15472_end_0 = const()[name = tensor("op_15472_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15472_end_mask_0 = const()[name = tensor("op_15472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15472_cast_fp16 = slice_by_index(begin = var_15472_begin_0, end = var_15472_end_0, end_mask = var_15472_end_mask_0, x = var_15075_cast_fp16)[name = tensor("op_15472_cast_fp16")]; + tensor var_15479_begin_0 = const()[name = tensor("op_15479_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15479_end_0 = const()[name = tensor("op_15479_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15479_end_mask_0 = const()[name = tensor("op_15479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15479_cast_fp16 = slice_by_index(begin = var_15479_begin_0, end = var_15479_end_0, end_mask = var_15479_end_mask_0, x = var_15075_cast_fp16)[name = tensor("op_15479_cast_fp16")]; + tensor var_15486_begin_0 = const()[name = tensor("op_15486_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15486_end_0 = const()[name = tensor("op_15486_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15486_end_mask_0 = const()[name = tensor("op_15486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15486_cast_fp16 = slice_by_index(begin = var_15486_begin_0, end = var_15486_end_0, end_mask = var_15486_end_mask_0, x = var_15075_cast_fp16)[name = tensor("op_15486_cast_fp16")]; + tensor var_15493_begin_0 = const()[name = tensor("op_15493_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15493_end_0 = const()[name = tensor("op_15493_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15493_end_mask_0 = const()[name = tensor("op_15493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15493_cast_fp16 = slice_by_index(begin = var_15493_begin_0, end = var_15493_end_0, end_mask = var_15493_end_mask_0, x = var_15075_cast_fp16)[name = tensor("op_15493_cast_fp16")]; + tensor var_15500_begin_0 = const()[name = tensor("op_15500_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15500_end_0 = const()[name = tensor("op_15500_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15500_end_mask_0 = const()[name = tensor("op_15500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15500_cast_fp16 = slice_by_index(begin = var_15500_begin_0, end = var_15500_end_0, end_mask = var_15500_end_mask_0, x = var_15079_cast_fp16)[name = tensor("op_15500_cast_fp16")]; + tensor var_15507_begin_0 = const()[name = tensor("op_15507_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15507_end_0 = const()[name = tensor("op_15507_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15507_end_mask_0 = const()[name = tensor("op_15507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15507_cast_fp16 = slice_by_index(begin = var_15507_begin_0, end = var_15507_end_0, end_mask = var_15507_end_mask_0, x = var_15079_cast_fp16)[name = tensor("op_15507_cast_fp16")]; + tensor var_15514_begin_0 = const()[name = tensor("op_15514_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15514_end_0 = const()[name = tensor("op_15514_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15514_end_mask_0 = const()[name = tensor("op_15514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15514_cast_fp16 = slice_by_index(begin = var_15514_begin_0, end = var_15514_end_0, end_mask = var_15514_end_mask_0, x = var_15079_cast_fp16)[name = tensor("op_15514_cast_fp16")]; + tensor var_15521_begin_0 = const()[name = tensor("op_15521_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15521_end_0 = const()[name = tensor("op_15521_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15521_end_mask_0 = const()[name = tensor("op_15521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15521_cast_fp16 = slice_by_index(begin = var_15521_begin_0, end = var_15521_end_0, end_mask = var_15521_end_mask_0, x = var_15079_cast_fp16)[name = tensor("op_15521_cast_fp16")]; + tensor var_15528_begin_0 = const()[name = tensor("op_15528_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15528_end_0 = const()[name = tensor("op_15528_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15528_end_mask_0 = const()[name = tensor("op_15528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15528_cast_fp16 = slice_by_index(begin = var_15528_begin_0, end = var_15528_end_0, end_mask = var_15528_end_mask_0, x = var_15083_cast_fp16)[name = tensor("op_15528_cast_fp16")]; + tensor var_15535_begin_0 = const()[name = tensor("op_15535_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15535_end_0 = const()[name = tensor("op_15535_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15535_end_mask_0 = const()[name = tensor("op_15535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15535_cast_fp16 = slice_by_index(begin = var_15535_begin_0, end = var_15535_end_0, end_mask = var_15535_end_mask_0, x = var_15083_cast_fp16)[name = tensor("op_15535_cast_fp16")]; + tensor var_15542_begin_0 = const()[name = tensor("op_15542_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15542_end_0 = const()[name = tensor("op_15542_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15542_end_mask_0 = const()[name = tensor("op_15542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15542_cast_fp16 = slice_by_index(begin = var_15542_begin_0, end = var_15542_end_0, end_mask = var_15542_end_mask_0, x = var_15083_cast_fp16)[name = tensor("op_15542_cast_fp16")]; + tensor var_15549_begin_0 = const()[name = tensor("op_15549_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15549_end_0 = const()[name = tensor("op_15549_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15549_end_mask_0 = const()[name = tensor("op_15549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15549_cast_fp16 = slice_by_index(begin = var_15549_begin_0, end = var_15549_end_0, end_mask = var_15549_end_mask_0, x = var_15083_cast_fp16)[name = tensor("op_15549_cast_fp16")]; + tensor var_15556_begin_0 = const()[name = tensor("op_15556_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15556_end_0 = const()[name = tensor("op_15556_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15556_end_mask_0 = const()[name = tensor("op_15556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15556_cast_fp16 = slice_by_index(begin = var_15556_begin_0, end = var_15556_end_0, end_mask = var_15556_end_mask_0, x = var_15087_cast_fp16)[name = tensor("op_15556_cast_fp16")]; + tensor var_15563_begin_0 = const()[name = tensor("op_15563_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15563_end_0 = const()[name = tensor("op_15563_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15563_end_mask_0 = const()[name = tensor("op_15563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15563_cast_fp16 = slice_by_index(begin = var_15563_begin_0, end = var_15563_end_0, end_mask = var_15563_end_mask_0, x = var_15087_cast_fp16)[name = tensor("op_15563_cast_fp16")]; + tensor var_15570_begin_0 = const()[name = tensor("op_15570_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15570_end_0 = const()[name = tensor("op_15570_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15570_end_mask_0 = const()[name = tensor("op_15570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15570_cast_fp16 = slice_by_index(begin = var_15570_begin_0, end = var_15570_end_0, end_mask = var_15570_end_mask_0, x = var_15087_cast_fp16)[name = tensor("op_15570_cast_fp16")]; + tensor var_15577_begin_0 = const()[name = tensor("op_15577_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15577_end_0 = const()[name = tensor("op_15577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15577_end_mask_0 = const()[name = tensor("op_15577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15577_cast_fp16 = slice_by_index(begin = var_15577_begin_0, end = var_15577_end_0, end_mask = var_15577_end_mask_0, x = var_15087_cast_fp16)[name = tensor("op_15577_cast_fp16")]; + tensor var_15584_begin_0 = const()[name = tensor("op_15584_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15584_end_0 = const()[name = tensor("op_15584_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15584_end_mask_0 = const()[name = tensor("op_15584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15584_cast_fp16 = slice_by_index(begin = var_15584_begin_0, end = var_15584_end_0, end_mask = var_15584_end_mask_0, x = var_15091_cast_fp16)[name = tensor("op_15584_cast_fp16")]; + tensor var_15591_begin_0 = const()[name = tensor("op_15591_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15591_end_0 = const()[name = tensor("op_15591_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15591_end_mask_0 = const()[name = tensor("op_15591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15591_cast_fp16 = slice_by_index(begin = var_15591_begin_0, end = var_15591_end_0, end_mask = var_15591_end_mask_0, x = var_15091_cast_fp16)[name = tensor("op_15591_cast_fp16")]; + tensor var_15598_begin_0 = const()[name = tensor("op_15598_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15598_end_0 = const()[name = tensor("op_15598_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15598_end_mask_0 = const()[name = tensor("op_15598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15598_cast_fp16 = slice_by_index(begin = var_15598_begin_0, end = var_15598_end_0, end_mask = var_15598_end_mask_0, x = var_15091_cast_fp16)[name = tensor("op_15598_cast_fp16")]; + tensor var_15605_begin_0 = const()[name = tensor("op_15605_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15605_end_0 = const()[name = tensor("op_15605_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15605_end_mask_0 = const()[name = tensor("op_15605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15605_cast_fp16 = slice_by_index(begin = var_15605_begin_0, end = var_15605_end_0, end_mask = var_15605_end_mask_0, x = var_15091_cast_fp16)[name = tensor("op_15605_cast_fp16")]; + tensor var_15612_begin_0 = const()[name = tensor("op_15612_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15612_end_0 = const()[name = tensor("op_15612_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15612_end_mask_0 = const()[name = tensor("op_15612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15612_cast_fp16 = slice_by_index(begin = var_15612_begin_0, end = var_15612_end_0, end_mask = var_15612_end_mask_0, x = var_15095_cast_fp16)[name = tensor("op_15612_cast_fp16")]; + tensor var_15619_begin_0 = const()[name = tensor("op_15619_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15619_end_0 = const()[name = tensor("op_15619_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15619_end_mask_0 = const()[name = tensor("op_15619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15619_cast_fp16 = slice_by_index(begin = var_15619_begin_0, end = var_15619_end_0, end_mask = var_15619_end_mask_0, x = var_15095_cast_fp16)[name = tensor("op_15619_cast_fp16")]; + tensor var_15626_begin_0 = const()[name = tensor("op_15626_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15626_end_0 = const()[name = tensor("op_15626_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15626_end_mask_0 = const()[name = tensor("op_15626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15626_cast_fp16 = slice_by_index(begin = var_15626_begin_0, end = var_15626_end_0, end_mask = var_15626_end_mask_0, x = var_15095_cast_fp16)[name = tensor("op_15626_cast_fp16")]; + tensor var_15633_begin_0 = const()[name = tensor("op_15633_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15633_end_0 = const()[name = tensor("op_15633_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15633_end_mask_0 = const()[name = tensor("op_15633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15633_cast_fp16 = slice_by_index(begin = var_15633_begin_0, end = var_15633_end_0, end_mask = var_15633_end_mask_0, x = var_15095_cast_fp16)[name = tensor("op_15633_cast_fp16")]; + tensor var_15640_begin_0 = const()[name = tensor("op_15640_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15640_end_0 = const()[name = tensor("op_15640_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_15640_end_mask_0 = const()[name = tensor("op_15640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15640_cast_fp16 = slice_by_index(begin = var_15640_begin_0, end = var_15640_end_0, end_mask = var_15640_end_mask_0, x = var_15099_cast_fp16)[name = tensor("op_15640_cast_fp16")]; + tensor var_15647_begin_0 = const()[name = tensor("op_15647_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_15647_end_0 = const()[name = tensor("op_15647_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_15647_end_mask_0 = const()[name = tensor("op_15647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15647_cast_fp16 = slice_by_index(begin = var_15647_begin_0, end = var_15647_end_0, end_mask = var_15647_end_mask_0, x = var_15099_cast_fp16)[name = tensor("op_15647_cast_fp16")]; + tensor var_15654_begin_0 = const()[name = tensor("op_15654_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_15654_end_0 = const()[name = tensor("op_15654_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_15654_end_mask_0 = const()[name = tensor("op_15654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15654_cast_fp16 = slice_by_index(begin = var_15654_begin_0, end = var_15654_end_0, end_mask = var_15654_end_mask_0, x = var_15099_cast_fp16)[name = tensor("op_15654_cast_fp16")]; + tensor var_15661_begin_0 = const()[name = tensor("op_15661_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_15661_end_0 = const()[name = tensor("op_15661_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15661_end_mask_0 = const()[name = tensor("op_15661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15661_cast_fp16 = slice_by_index(begin = var_15661_begin_0, end = var_15661_end_0, end_mask = var_15661_end_mask_0, x = var_15099_cast_fp16)[name = tensor("op_15661_cast_fp16")]; + tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_15666_begin_0 = const()[name = tensor("op_15666_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15666_end_0 = const()[name = tensor("op_15666_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_15666_end_mask_0 = const()[name = tensor("op_15666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_22 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = tensor("transpose_22")]; + tensor var_15666_cast_fp16 = slice_by_index(begin = var_15666_begin_0, end = var_15666_end_0, end_mask = var_15666_end_mask_0, x = transpose_22)[name = tensor("op_15666_cast_fp16")]; + tensor var_15670_begin_0 = const()[name = tensor("op_15670_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_15670_end_0 = const()[name = tensor("op_15670_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_15670_end_mask_0 = const()[name = tensor("op_15670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15670_cast_fp16 = slice_by_index(begin = var_15670_begin_0, end = var_15670_end_0, end_mask = var_15670_end_mask_0, x = transpose_22)[name = tensor("op_15670_cast_fp16")]; + tensor var_15674_begin_0 = const()[name = tensor("op_15674_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_15674_end_0 = const()[name = tensor("op_15674_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_15674_end_mask_0 = const()[name = tensor("op_15674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15674_cast_fp16 = slice_by_index(begin = var_15674_begin_0, end = var_15674_end_0, end_mask = var_15674_end_mask_0, x = transpose_22)[name = tensor("op_15674_cast_fp16")]; + tensor var_15678_begin_0 = const()[name = tensor("op_15678_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_15678_end_0 = const()[name = tensor("op_15678_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_15678_end_mask_0 = const()[name = tensor("op_15678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15678_cast_fp16 = slice_by_index(begin = var_15678_begin_0, end = var_15678_end_0, end_mask = var_15678_end_mask_0, x = transpose_22)[name = tensor("op_15678_cast_fp16")]; + tensor var_15682_begin_0 = const()[name = tensor("op_15682_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_15682_end_0 = const()[name = tensor("op_15682_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_15682_end_mask_0 = const()[name = tensor("op_15682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15682_cast_fp16 = slice_by_index(begin = var_15682_begin_0, end = var_15682_end_0, end_mask = var_15682_end_mask_0, x = transpose_22)[name = tensor("op_15682_cast_fp16")]; + tensor var_15686_begin_0 = const()[name = tensor("op_15686_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_15686_end_0 = const()[name = tensor("op_15686_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_15686_end_mask_0 = const()[name = tensor("op_15686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15686_cast_fp16 = slice_by_index(begin = var_15686_begin_0, end = var_15686_end_0, end_mask = var_15686_end_mask_0, x = transpose_22)[name = tensor("op_15686_cast_fp16")]; + tensor var_15690_begin_0 = const()[name = tensor("op_15690_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_15690_end_0 = const()[name = tensor("op_15690_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_15690_end_mask_0 = const()[name = tensor("op_15690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15690_cast_fp16 = slice_by_index(begin = var_15690_begin_0, end = var_15690_end_0, end_mask = var_15690_end_mask_0, x = transpose_22)[name = tensor("op_15690_cast_fp16")]; + tensor var_15694_begin_0 = const()[name = tensor("op_15694_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_15694_end_0 = const()[name = tensor("op_15694_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_15694_end_mask_0 = const()[name = tensor("op_15694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15694_cast_fp16 = slice_by_index(begin = var_15694_begin_0, end = var_15694_end_0, end_mask = var_15694_end_mask_0, x = transpose_22)[name = tensor("op_15694_cast_fp16")]; + tensor var_15698_begin_0 = const()[name = tensor("op_15698_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_15698_end_0 = const()[name = tensor("op_15698_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_15698_end_mask_0 = const()[name = tensor("op_15698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15698_cast_fp16 = slice_by_index(begin = var_15698_begin_0, end = var_15698_end_0, end_mask = var_15698_end_mask_0, x = transpose_22)[name = tensor("op_15698_cast_fp16")]; + tensor var_15702_begin_0 = const()[name = tensor("op_15702_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_15702_end_0 = const()[name = tensor("op_15702_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_15702_end_mask_0 = const()[name = tensor("op_15702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15702_cast_fp16 = slice_by_index(begin = var_15702_begin_0, end = var_15702_end_0, end_mask = var_15702_end_mask_0, x = transpose_22)[name = tensor("op_15702_cast_fp16")]; + tensor var_15706_begin_0 = const()[name = tensor("op_15706_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_15706_end_0 = const()[name = tensor("op_15706_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_15706_end_mask_0 = const()[name = tensor("op_15706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15706_cast_fp16 = slice_by_index(begin = var_15706_begin_0, end = var_15706_end_0, end_mask = var_15706_end_mask_0, x = transpose_22)[name = tensor("op_15706_cast_fp16")]; + tensor var_15710_begin_0 = const()[name = tensor("op_15710_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_15710_end_0 = const()[name = tensor("op_15710_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_15710_end_mask_0 = const()[name = tensor("op_15710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15710_cast_fp16 = slice_by_index(begin = var_15710_begin_0, end = var_15710_end_0, end_mask = var_15710_end_mask_0, x = transpose_22)[name = tensor("op_15710_cast_fp16")]; + tensor var_15714_begin_0 = const()[name = tensor("op_15714_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_15714_end_0 = const()[name = tensor("op_15714_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_15714_end_mask_0 = const()[name = tensor("op_15714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15714_cast_fp16 = slice_by_index(begin = var_15714_begin_0, end = var_15714_end_0, end_mask = var_15714_end_mask_0, x = transpose_22)[name = tensor("op_15714_cast_fp16")]; + tensor var_15718_begin_0 = const()[name = tensor("op_15718_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_15718_end_0 = const()[name = tensor("op_15718_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_15718_end_mask_0 = const()[name = tensor("op_15718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15718_cast_fp16 = slice_by_index(begin = var_15718_begin_0, end = var_15718_end_0, end_mask = var_15718_end_mask_0, x = transpose_22)[name = tensor("op_15718_cast_fp16")]; + tensor var_15722_begin_0 = const()[name = tensor("op_15722_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_15722_end_0 = const()[name = tensor("op_15722_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_15722_end_mask_0 = const()[name = tensor("op_15722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15722_cast_fp16 = slice_by_index(begin = var_15722_begin_0, end = var_15722_end_0, end_mask = var_15722_end_mask_0, x = transpose_22)[name = tensor("op_15722_cast_fp16")]; + tensor var_15726_begin_0 = const()[name = tensor("op_15726_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_15726_end_0 = const()[name = tensor("op_15726_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_15726_end_mask_0 = const()[name = tensor("op_15726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15726_cast_fp16 = slice_by_index(begin = var_15726_begin_0, end = var_15726_end_0, end_mask = var_15726_end_mask_0, x = transpose_22)[name = tensor("op_15726_cast_fp16")]; + tensor var_15730_begin_0 = const()[name = tensor("op_15730_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_15730_end_0 = const()[name = tensor("op_15730_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_15730_end_mask_0 = const()[name = tensor("op_15730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15730_cast_fp16 = slice_by_index(begin = var_15730_begin_0, end = var_15730_end_0, end_mask = var_15730_end_mask_0, x = transpose_22)[name = tensor("op_15730_cast_fp16")]; + tensor var_15734_begin_0 = const()[name = tensor("op_15734_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_15734_end_0 = const()[name = tensor("op_15734_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_15734_end_mask_0 = const()[name = tensor("op_15734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15734_cast_fp16 = slice_by_index(begin = var_15734_begin_0, end = var_15734_end_0, end_mask = var_15734_end_mask_0, x = transpose_22)[name = tensor("op_15734_cast_fp16")]; + tensor var_15738_begin_0 = const()[name = tensor("op_15738_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_15738_end_0 = const()[name = tensor("op_15738_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_15738_end_mask_0 = const()[name = tensor("op_15738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15738_cast_fp16 = slice_by_index(begin = var_15738_begin_0, end = var_15738_end_0, end_mask = var_15738_end_mask_0, x = transpose_22)[name = tensor("op_15738_cast_fp16")]; + tensor var_15742_begin_0 = const()[name = tensor("op_15742_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_15742_end_0 = const()[name = tensor("op_15742_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_15742_end_mask_0 = const()[name = tensor("op_15742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_15742_cast_fp16 = slice_by_index(begin = var_15742_begin_0, end = var_15742_end_0, end_mask = var_15742_end_mask_0, x = transpose_22)[name = tensor("op_15742_cast_fp16")]; + tensor var_15744_begin_0 = const()[name = tensor("op_15744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_15744_end_0 = const()[name = tensor("op_15744_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_15744_end_mask_0 = const()[name = tensor("op_15744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15744_cast_fp16 = slice_by_index(begin = var_15744_begin_0, end = var_15744_end_0, end_mask = var_15744_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15744_cast_fp16")]; + tensor var_15748_begin_0 = const()[name = tensor("op_15748_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_15748_end_0 = const()[name = tensor("op_15748_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_15748_end_mask_0 = const()[name = tensor("op_15748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15748_cast_fp16 = slice_by_index(begin = var_15748_begin_0, end = var_15748_end_0, end_mask = var_15748_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15748_cast_fp16")]; + tensor var_15752_begin_0 = const()[name = tensor("op_15752_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_15752_end_0 = const()[name = tensor("op_15752_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_15752_end_mask_0 = const()[name = tensor("op_15752_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15752_cast_fp16 = slice_by_index(begin = var_15752_begin_0, end = var_15752_end_0, end_mask = var_15752_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15752_cast_fp16")]; + tensor var_15756_begin_0 = const()[name = tensor("op_15756_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_15756_end_0 = const()[name = tensor("op_15756_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_15756_end_mask_0 = const()[name = tensor("op_15756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15756_cast_fp16 = slice_by_index(begin = var_15756_begin_0, end = var_15756_end_0, end_mask = var_15756_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15756_cast_fp16")]; + tensor var_15760_begin_0 = const()[name = tensor("op_15760_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_15760_end_0 = const()[name = tensor("op_15760_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_15760_end_mask_0 = const()[name = tensor("op_15760_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15760_cast_fp16 = slice_by_index(begin = var_15760_begin_0, end = var_15760_end_0, end_mask = var_15760_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15760_cast_fp16")]; + tensor var_15764_begin_0 = const()[name = tensor("op_15764_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_15764_end_0 = const()[name = tensor("op_15764_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_15764_end_mask_0 = const()[name = tensor("op_15764_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15764_cast_fp16 = slice_by_index(begin = var_15764_begin_0, end = var_15764_end_0, end_mask = var_15764_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15764_cast_fp16")]; + tensor var_15768_begin_0 = const()[name = tensor("op_15768_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_15768_end_0 = const()[name = tensor("op_15768_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_15768_end_mask_0 = const()[name = tensor("op_15768_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15768_cast_fp16 = slice_by_index(begin = var_15768_begin_0, end = var_15768_end_0, end_mask = var_15768_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15768_cast_fp16")]; + tensor var_15772_begin_0 = const()[name = tensor("op_15772_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_15772_end_0 = const()[name = tensor("op_15772_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_15772_end_mask_0 = const()[name = tensor("op_15772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15772_cast_fp16 = slice_by_index(begin = var_15772_begin_0, end = var_15772_end_0, end_mask = var_15772_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15772_cast_fp16")]; + tensor var_15776_begin_0 = const()[name = tensor("op_15776_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_15776_end_0 = const()[name = tensor("op_15776_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_15776_end_mask_0 = const()[name = tensor("op_15776_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15776_cast_fp16 = slice_by_index(begin = var_15776_begin_0, end = var_15776_end_0, end_mask = var_15776_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15776_cast_fp16")]; + tensor var_15780_begin_0 = const()[name = tensor("op_15780_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_15780_end_0 = const()[name = tensor("op_15780_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_15780_end_mask_0 = const()[name = tensor("op_15780_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15780_cast_fp16 = slice_by_index(begin = var_15780_begin_0, end = var_15780_end_0, end_mask = var_15780_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15780_cast_fp16")]; + tensor var_15784_begin_0 = const()[name = tensor("op_15784_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_15784_end_0 = const()[name = tensor("op_15784_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_15784_end_mask_0 = const()[name = tensor("op_15784_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15784_cast_fp16 = slice_by_index(begin = var_15784_begin_0, end = var_15784_end_0, end_mask = var_15784_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15784_cast_fp16")]; + tensor var_15788_begin_0 = const()[name = tensor("op_15788_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_15788_end_0 = const()[name = tensor("op_15788_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_15788_end_mask_0 = const()[name = tensor("op_15788_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15788_cast_fp16 = slice_by_index(begin = var_15788_begin_0, end = var_15788_end_0, end_mask = var_15788_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15788_cast_fp16")]; + tensor var_15792_begin_0 = const()[name = tensor("op_15792_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_15792_end_0 = const()[name = tensor("op_15792_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_15792_end_mask_0 = const()[name = tensor("op_15792_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15792_cast_fp16 = slice_by_index(begin = var_15792_begin_0, end = var_15792_end_0, end_mask = var_15792_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15792_cast_fp16")]; + tensor var_15796_begin_0 = const()[name = tensor("op_15796_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_15796_end_0 = const()[name = tensor("op_15796_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_15796_end_mask_0 = const()[name = tensor("op_15796_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15796_cast_fp16 = slice_by_index(begin = var_15796_begin_0, end = var_15796_end_0, end_mask = var_15796_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15796_cast_fp16")]; + tensor var_15800_begin_0 = const()[name = tensor("op_15800_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_15800_end_0 = const()[name = tensor("op_15800_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_15800_end_mask_0 = const()[name = tensor("op_15800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15800_cast_fp16 = slice_by_index(begin = var_15800_begin_0, end = var_15800_end_0, end_mask = var_15800_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15800_cast_fp16")]; + tensor var_15804_begin_0 = const()[name = tensor("op_15804_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_15804_end_0 = const()[name = tensor("op_15804_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_15804_end_mask_0 = const()[name = tensor("op_15804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15804_cast_fp16 = slice_by_index(begin = var_15804_begin_0, end = var_15804_end_0, end_mask = var_15804_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15804_cast_fp16")]; + tensor var_15808_begin_0 = const()[name = tensor("op_15808_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_15808_end_0 = const()[name = tensor("op_15808_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_15808_end_mask_0 = const()[name = tensor("op_15808_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15808_cast_fp16 = slice_by_index(begin = var_15808_begin_0, end = var_15808_end_0, end_mask = var_15808_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15808_cast_fp16")]; + tensor var_15812_begin_0 = const()[name = tensor("op_15812_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_15812_end_0 = const()[name = tensor("op_15812_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_15812_end_mask_0 = const()[name = tensor("op_15812_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15812_cast_fp16 = slice_by_index(begin = var_15812_begin_0, end = var_15812_end_0, end_mask = var_15812_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15812_cast_fp16")]; + tensor var_15816_begin_0 = const()[name = tensor("op_15816_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_15816_end_0 = const()[name = tensor("op_15816_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_15816_end_mask_0 = const()[name = tensor("op_15816_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15816_cast_fp16 = slice_by_index(begin = var_15816_begin_0, end = var_15816_end_0, end_mask = var_15816_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15816_cast_fp16")]; + tensor var_15820_begin_0 = const()[name = tensor("op_15820_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_15820_end_0 = const()[name = tensor("op_15820_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_15820_end_mask_0 = const()[name = tensor("op_15820_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_15820_cast_fp16 = slice_by_index(begin = var_15820_begin_0, end = var_15820_end_0, end_mask = var_15820_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_15820_cast_fp16")]; + tensor var_15824_equation_0 = const()[name = tensor("op_15824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15824_cast_fp16 = einsum(equation = var_15824_equation_0, values = (var_15666_cast_fp16, var_15108_cast_fp16))[name = tensor("op_15824_cast_fp16")]; + tensor var_15825_to_fp16 = const()[name = tensor("op_15825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1441_cast_fp16 = mul(x = var_15824_cast_fp16, y = var_15825_to_fp16)[name = tensor("aw_chunk_1441_cast_fp16")]; + tensor var_15828_equation_0 = const()[name = tensor("op_15828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15828_cast_fp16 = einsum(equation = var_15828_equation_0, values = (var_15666_cast_fp16, var_15115_cast_fp16))[name = tensor("op_15828_cast_fp16")]; + tensor var_15829_to_fp16 = const()[name = tensor("op_15829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1443_cast_fp16 = mul(x = var_15828_cast_fp16, y = var_15829_to_fp16)[name = tensor("aw_chunk_1443_cast_fp16")]; + tensor var_15832_equation_0 = const()[name = tensor("op_15832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15832_cast_fp16 = einsum(equation = var_15832_equation_0, values = (var_15666_cast_fp16, var_15122_cast_fp16))[name = tensor("op_15832_cast_fp16")]; + tensor var_15833_to_fp16 = const()[name = tensor("op_15833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1445_cast_fp16 = mul(x = var_15832_cast_fp16, y = var_15833_to_fp16)[name = tensor("aw_chunk_1445_cast_fp16")]; + tensor var_15836_equation_0 = const()[name = tensor("op_15836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15836_cast_fp16 = einsum(equation = var_15836_equation_0, values = (var_15666_cast_fp16, var_15129_cast_fp16))[name = tensor("op_15836_cast_fp16")]; + tensor var_15837_to_fp16 = const()[name = tensor("op_15837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1447_cast_fp16 = mul(x = var_15836_cast_fp16, y = var_15837_to_fp16)[name = tensor("aw_chunk_1447_cast_fp16")]; + tensor var_15840_equation_0 = const()[name = tensor("op_15840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15840_cast_fp16 = einsum(equation = var_15840_equation_0, values = (var_15670_cast_fp16, var_15136_cast_fp16))[name = tensor("op_15840_cast_fp16")]; + tensor var_15841_to_fp16 = const()[name = tensor("op_15841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1449_cast_fp16 = mul(x = var_15840_cast_fp16, y = var_15841_to_fp16)[name = tensor("aw_chunk_1449_cast_fp16")]; + tensor var_15844_equation_0 = const()[name = tensor("op_15844_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15844_cast_fp16 = einsum(equation = var_15844_equation_0, values = (var_15670_cast_fp16, var_15143_cast_fp16))[name = tensor("op_15844_cast_fp16")]; + tensor var_15845_to_fp16 = const()[name = tensor("op_15845_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1451_cast_fp16 = mul(x = var_15844_cast_fp16, y = var_15845_to_fp16)[name = tensor("aw_chunk_1451_cast_fp16")]; + tensor var_15848_equation_0 = const()[name = tensor("op_15848_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15848_cast_fp16 = einsum(equation = var_15848_equation_0, values = (var_15670_cast_fp16, var_15150_cast_fp16))[name = tensor("op_15848_cast_fp16")]; + tensor var_15849_to_fp16 = const()[name = tensor("op_15849_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1453_cast_fp16 = mul(x = var_15848_cast_fp16, y = var_15849_to_fp16)[name = tensor("aw_chunk_1453_cast_fp16")]; + tensor var_15852_equation_0 = const()[name = tensor("op_15852_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15852_cast_fp16 = einsum(equation = var_15852_equation_0, values = (var_15670_cast_fp16, var_15157_cast_fp16))[name = tensor("op_15852_cast_fp16")]; + tensor var_15853_to_fp16 = const()[name = tensor("op_15853_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1455_cast_fp16 = mul(x = var_15852_cast_fp16, y = var_15853_to_fp16)[name = tensor("aw_chunk_1455_cast_fp16")]; + tensor var_15856_equation_0 = const()[name = tensor("op_15856_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15856_cast_fp16 = einsum(equation = var_15856_equation_0, values = (var_15674_cast_fp16, var_15164_cast_fp16))[name = tensor("op_15856_cast_fp16")]; + tensor var_15857_to_fp16 = const()[name = tensor("op_15857_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1457_cast_fp16 = mul(x = var_15856_cast_fp16, y = var_15857_to_fp16)[name = tensor("aw_chunk_1457_cast_fp16")]; + tensor var_15860_equation_0 = const()[name = tensor("op_15860_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15860_cast_fp16 = einsum(equation = var_15860_equation_0, values = (var_15674_cast_fp16, var_15171_cast_fp16))[name = tensor("op_15860_cast_fp16")]; + tensor var_15861_to_fp16 = const()[name = tensor("op_15861_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1459_cast_fp16 = mul(x = var_15860_cast_fp16, y = var_15861_to_fp16)[name = tensor("aw_chunk_1459_cast_fp16")]; + tensor var_15864_equation_0 = const()[name = tensor("op_15864_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15864_cast_fp16 = einsum(equation = var_15864_equation_0, values = (var_15674_cast_fp16, var_15178_cast_fp16))[name = tensor("op_15864_cast_fp16")]; + tensor var_15865_to_fp16 = const()[name = tensor("op_15865_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1461_cast_fp16 = mul(x = var_15864_cast_fp16, y = var_15865_to_fp16)[name = tensor("aw_chunk_1461_cast_fp16")]; + tensor var_15868_equation_0 = const()[name = tensor("op_15868_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15868_cast_fp16 = einsum(equation = var_15868_equation_0, values = (var_15674_cast_fp16, var_15185_cast_fp16))[name = tensor("op_15868_cast_fp16")]; + tensor var_15869_to_fp16 = const()[name = tensor("op_15869_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1463_cast_fp16 = mul(x = var_15868_cast_fp16, y = var_15869_to_fp16)[name = tensor("aw_chunk_1463_cast_fp16")]; + tensor var_15872_equation_0 = const()[name = tensor("op_15872_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15872_cast_fp16 = einsum(equation = var_15872_equation_0, values = (var_15678_cast_fp16, var_15192_cast_fp16))[name = tensor("op_15872_cast_fp16")]; + tensor var_15873_to_fp16 = const()[name = tensor("op_15873_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1465_cast_fp16 = mul(x = var_15872_cast_fp16, y = var_15873_to_fp16)[name = tensor("aw_chunk_1465_cast_fp16")]; + tensor var_15876_equation_0 = const()[name = tensor("op_15876_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15876_cast_fp16 = einsum(equation = var_15876_equation_0, values = (var_15678_cast_fp16, var_15199_cast_fp16))[name = tensor("op_15876_cast_fp16")]; + tensor var_15877_to_fp16 = const()[name = tensor("op_15877_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1467_cast_fp16 = mul(x = var_15876_cast_fp16, y = var_15877_to_fp16)[name = tensor("aw_chunk_1467_cast_fp16")]; + tensor var_15880_equation_0 = const()[name = tensor("op_15880_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15880_cast_fp16 = einsum(equation = var_15880_equation_0, values = (var_15678_cast_fp16, var_15206_cast_fp16))[name = tensor("op_15880_cast_fp16")]; + tensor var_15881_to_fp16 = const()[name = tensor("op_15881_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1469_cast_fp16 = mul(x = var_15880_cast_fp16, y = var_15881_to_fp16)[name = tensor("aw_chunk_1469_cast_fp16")]; + tensor var_15884_equation_0 = const()[name = tensor("op_15884_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15884_cast_fp16 = einsum(equation = var_15884_equation_0, values = (var_15678_cast_fp16, var_15213_cast_fp16))[name = tensor("op_15884_cast_fp16")]; + tensor var_15885_to_fp16 = const()[name = tensor("op_15885_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1471_cast_fp16 = mul(x = var_15884_cast_fp16, y = var_15885_to_fp16)[name = tensor("aw_chunk_1471_cast_fp16")]; + tensor var_15888_equation_0 = const()[name = tensor("op_15888_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15888_cast_fp16 = einsum(equation = var_15888_equation_0, values = (var_15682_cast_fp16, var_15220_cast_fp16))[name = tensor("op_15888_cast_fp16")]; + tensor var_15889_to_fp16 = const()[name = tensor("op_15889_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1473_cast_fp16 = mul(x = var_15888_cast_fp16, y = var_15889_to_fp16)[name = tensor("aw_chunk_1473_cast_fp16")]; + tensor var_15892_equation_0 = const()[name = tensor("op_15892_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15892_cast_fp16 = einsum(equation = var_15892_equation_0, values = (var_15682_cast_fp16, var_15227_cast_fp16))[name = tensor("op_15892_cast_fp16")]; + tensor var_15893_to_fp16 = const()[name = tensor("op_15893_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1475_cast_fp16 = mul(x = var_15892_cast_fp16, y = var_15893_to_fp16)[name = tensor("aw_chunk_1475_cast_fp16")]; + tensor var_15896_equation_0 = const()[name = tensor("op_15896_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15896_cast_fp16 = einsum(equation = var_15896_equation_0, values = (var_15682_cast_fp16, var_15234_cast_fp16))[name = tensor("op_15896_cast_fp16")]; + tensor var_15897_to_fp16 = const()[name = tensor("op_15897_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1477_cast_fp16 = mul(x = var_15896_cast_fp16, y = var_15897_to_fp16)[name = tensor("aw_chunk_1477_cast_fp16")]; + tensor var_15900_equation_0 = const()[name = tensor("op_15900_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15900_cast_fp16 = einsum(equation = var_15900_equation_0, values = (var_15682_cast_fp16, var_15241_cast_fp16))[name = tensor("op_15900_cast_fp16")]; + tensor var_15901_to_fp16 = const()[name = tensor("op_15901_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1479_cast_fp16 = mul(x = var_15900_cast_fp16, y = var_15901_to_fp16)[name = tensor("aw_chunk_1479_cast_fp16")]; + tensor var_15904_equation_0 = const()[name = tensor("op_15904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15904_cast_fp16 = einsum(equation = var_15904_equation_0, values = (var_15686_cast_fp16, var_15248_cast_fp16))[name = tensor("op_15904_cast_fp16")]; + tensor var_15905_to_fp16 = const()[name = tensor("op_15905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1481_cast_fp16 = mul(x = var_15904_cast_fp16, y = var_15905_to_fp16)[name = tensor("aw_chunk_1481_cast_fp16")]; + tensor var_15908_equation_0 = const()[name = tensor("op_15908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15908_cast_fp16 = einsum(equation = var_15908_equation_0, values = (var_15686_cast_fp16, var_15255_cast_fp16))[name = tensor("op_15908_cast_fp16")]; + tensor var_15909_to_fp16 = const()[name = tensor("op_15909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1483_cast_fp16 = mul(x = var_15908_cast_fp16, y = var_15909_to_fp16)[name = tensor("aw_chunk_1483_cast_fp16")]; + tensor var_15912_equation_0 = const()[name = tensor("op_15912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15912_cast_fp16 = einsum(equation = var_15912_equation_0, values = (var_15686_cast_fp16, var_15262_cast_fp16))[name = tensor("op_15912_cast_fp16")]; + tensor var_15913_to_fp16 = const()[name = tensor("op_15913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1485_cast_fp16 = mul(x = var_15912_cast_fp16, y = var_15913_to_fp16)[name = tensor("aw_chunk_1485_cast_fp16")]; + tensor var_15916_equation_0 = const()[name = tensor("op_15916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15916_cast_fp16 = einsum(equation = var_15916_equation_0, values = (var_15686_cast_fp16, var_15269_cast_fp16))[name = tensor("op_15916_cast_fp16")]; + tensor var_15917_to_fp16 = const()[name = tensor("op_15917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1487_cast_fp16 = mul(x = var_15916_cast_fp16, y = var_15917_to_fp16)[name = tensor("aw_chunk_1487_cast_fp16")]; + tensor var_15920_equation_0 = const()[name = tensor("op_15920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15920_cast_fp16 = einsum(equation = var_15920_equation_0, values = (var_15690_cast_fp16, var_15276_cast_fp16))[name = tensor("op_15920_cast_fp16")]; + tensor var_15921_to_fp16 = const()[name = tensor("op_15921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1489_cast_fp16 = mul(x = var_15920_cast_fp16, y = var_15921_to_fp16)[name = tensor("aw_chunk_1489_cast_fp16")]; + tensor var_15924_equation_0 = const()[name = tensor("op_15924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15924_cast_fp16 = einsum(equation = var_15924_equation_0, values = (var_15690_cast_fp16, var_15283_cast_fp16))[name = tensor("op_15924_cast_fp16")]; + tensor var_15925_to_fp16 = const()[name = tensor("op_15925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1491_cast_fp16 = mul(x = var_15924_cast_fp16, y = var_15925_to_fp16)[name = tensor("aw_chunk_1491_cast_fp16")]; + tensor var_15928_equation_0 = const()[name = tensor("op_15928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15928_cast_fp16 = einsum(equation = var_15928_equation_0, values = (var_15690_cast_fp16, var_15290_cast_fp16))[name = tensor("op_15928_cast_fp16")]; + tensor var_15929_to_fp16 = const()[name = tensor("op_15929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1493_cast_fp16 = mul(x = var_15928_cast_fp16, y = var_15929_to_fp16)[name = tensor("aw_chunk_1493_cast_fp16")]; + tensor var_15932_equation_0 = const()[name = tensor("op_15932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15932_cast_fp16 = einsum(equation = var_15932_equation_0, values = (var_15690_cast_fp16, var_15297_cast_fp16))[name = tensor("op_15932_cast_fp16")]; + tensor var_15933_to_fp16 = const()[name = tensor("op_15933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1495_cast_fp16 = mul(x = var_15932_cast_fp16, y = var_15933_to_fp16)[name = tensor("aw_chunk_1495_cast_fp16")]; + tensor var_15936_equation_0 = const()[name = tensor("op_15936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15936_cast_fp16 = einsum(equation = var_15936_equation_0, values = (var_15694_cast_fp16, var_15304_cast_fp16))[name = tensor("op_15936_cast_fp16")]; + tensor var_15937_to_fp16 = const()[name = tensor("op_15937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1497_cast_fp16 = mul(x = var_15936_cast_fp16, y = var_15937_to_fp16)[name = tensor("aw_chunk_1497_cast_fp16")]; + tensor var_15940_equation_0 = const()[name = tensor("op_15940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15940_cast_fp16 = einsum(equation = var_15940_equation_0, values = (var_15694_cast_fp16, var_15311_cast_fp16))[name = tensor("op_15940_cast_fp16")]; + tensor var_15941_to_fp16 = const()[name = tensor("op_15941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1499_cast_fp16 = mul(x = var_15940_cast_fp16, y = var_15941_to_fp16)[name = tensor("aw_chunk_1499_cast_fp16")]; + tensor var_15944_equation_0 = const()[name = tensor("op_15944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15944_cast_fp16 = einsum(equation = var_15944_equation_0, values = (var_15694_cast_fp16, var_15318_cast_fp16))[name = tensor("op_15944_cast_fp16")]; + tensor var_15945_to_fp16 = const()[name = tensor("op_15945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1501_cast_fp16 = mul(x = var_15944_cast_fp16, y = var_15945_to_fp16)[name = tensor("aw_chunk_1501_cast_fp16")]; + tensor var_15948_equation_0 = const()[name = tensor("op_15948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15948_cast_fp16 = einsum(equation = var_15948_equation_0, values = (var_15694_cast_fp16, var_15325_cast_fp16))[name = tensor("op_15948_cast_fp16")]; + tensor var_15949_to_fp16 = const()[name = tensor("op_15949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1503_cast_fp16 = mul(x = var_15948_cast_fp16, y = var_15949_to_fp16)[name = tensor("aw_chunk_1503_cast_fp16")]; + tensor var_15952_equation_0 = const()[name = tensor("op_15952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15952_cast_fp16 = einsum(equation = var_15952_equation_0, values = (var_15698_cast_fp16, var_15332_cast_fp16))[name = tensor("op_15952_cast_fp16")]; + tensor var_15953_to_fp16 = const()[name = tensor("op_15953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1505_cast_fp16 = mul(x = var_15952_cast_fp16, y = var_15953_to_fp16)[name = tensor("aw_chunk_1505_cast_fp16")]; + tensor var_15956_equation_0 = const()[name = tensor("op_15956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15956_cast_fp16 = einsum(equation = var_15956_equation_0, values = (var_15698_cast_fp16, var_15339_cast_fp16))[name = tensor("op_15956_cast_fp16")]; + tensor var_15957_to_fp16 = const()[name = tensor("op_15957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1507_cast_fp16 = mul(x = var_15956_cast_fp16, y = var_15957_to_fp16)[name = tensor("aw_chunk_1507_cast_fp16")]; + tensor var_15960_equation_0 = const()[name = tensor("op_15960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15960_cast_fp16 = einsum(equation = var_15960_equation_0, values = (var_15698_cast_fp16, var_15346_cast_fp16))[name = tensor("op_15960_cast_fp16")]; + tensor var_15961_to_fp16 = const()[name = tensor("op_15961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1509_cast_fp16 = mul(x = var_15960_cast_fp16, y = var_15961_to_fp16)[name = tensor("aw_chunk_1509_cast_fp16")]; + tensor var_15964_equation_0 = const()[name = tensor("op_15964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15964_cast_fp16 = einsum(equation = var_15964_equation_0, values = (var_15698_cast_fp16, var_15353_cast_fp16))[name = tensor("op_15964_cast_fp16")]; + tensor var_15965_to_fp16 = const()[name = tensor("op_15965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1511_cast_fp16 = mul(x = var_15964_cast_fp16, y = var_15965_to_fp16)[name = tensor("aw_chunk_1511_cast_fp16")]; + tensor var_15968_equation_0 = const()[name = tensor("op_15968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15968_cast_fp16 = einsum(equation = var_15968_equation_0, values = (var_15702_cast_fp16, var_15360_cast_fp16))[name = tensor("op_15968_cast_fp16")]; + tensor var_15969_to_fp16 = const()[name = tensor("op_15969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1513_cast_fp16 = mul(x = var_15968_cast_fp16, y = var_15969_to_fp16)[name = tensor("aw_chunk_1513_cast_fp16")]; + tensor var_15972_equation_0 = const()[name = tensor("op_15972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15972_cast_fp16 = einsum(equation = var_15972_equation_0, values = (var_15702_cast_fp16, var_15367_cast_fp16))[name = tensor("op_15972_cast_fp16")]; + tensor var_15973_to_fp16 = const()[name = tensor("op_15973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1515_cast_fp16 = mul(x = var_15972_cast_fp16, y = var_15973_to_fp16)[name = tensor("aw_chunk_1515_cast_fp16")]; + tensor var_15976_equation_0 = const()[name = tensor("op_15976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15976_cast_fp16 = einsum(equation = var_15976_equation_0, values = (var_15702_cast_fp16, var_15374_cast_fp16))[name = tensor("op_15976_cast_fp16")]; + tensor var_15977_to_fp16 = const()[name = tensor("op_15977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1517_cast_fp16 = mul(x = var_15976_cast_fp16, y = var_15977_to_fp16)[name = tensor("aw_chunk_1517_cast_fp16")]; + tensor var_15980_equation_0 = const()[name = tensor("op_15980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15980_cast_fp16 = einsum(equation = var_15980_equation_0, values = (var_15702_cast_fp16, var_15381_cast_fp16))[name = tensor("op_15980_cast_fp16")]; + tensor var_15981_to_fp16 = const()[name = tensor("op_15981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1519_cast_fp16 = mul(x = var_15980_cast_fp16, y = var_15981_to_fp16)[name = tensor("aw_chunk_1519_cast_fp16")]; + tensor var_15984_equation_0 = const()[name = tensor("op_15984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15984_cast_fp16 = einsum(equation = var_15984_equation_0, values = (var_15706_cast_fp16, var_15388_cast_fp16))[name = tensor("op_15984_cast_fp16")]; + tensor var_15985_to_fp16 = const()[name = tensor("op_15985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1521_cast_fp16 = mul(x = var_15984_cast_fp16, y = var_15985_to_fp16)[name = tensor("aw_chunk_1521_cast_fp16")]; + tensor var_15988_equation_0 = const()[name = tensor("op_15988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15988_cast_fp16 = einsum(equation = var_15988_equation_0, values = (var_15706_cast_fp16, var_15395_cast_fp16))[name = tensor("op_15988_cast_fp16")]; + tensor var_15989_to_fp16 = const()[name = tensor("op_15989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1523_cast_fp16 = mul(x = var_15988_cast_fp16, y = var_15989_to_fp16)[name = tensor("aw_chunk_1523_cast_fp16")]; + tensor var_15992_equation_0 = const()[name = tensor("op_15992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15992_cast_fp16 = einsum(equation = var_15992_equation_0, values = (var_15706_cast_fp16, var_15402_cast_fp16))[name = tensor("op_15992_cast_fp16")]; + tensor var_15993_to_fp16 = const()[name = tensor("op_15993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1525_cast_fp16 = mul(x = var_15992_cast_fp16, y = var_15993_to_fp16)[name = tensor("aw_chunk_1525_cast_fp16")]; + tensor var_15996_equation_0 = const()[name = tensor("op_15996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_15996_cast_fp16 = einsum(equation = var_15996_equation_0, values = (var_15706_cast_fp16, var_15409_cast_fp16))[name = tensor("op_15996_cast_fp16")]; + tensor var_15997_to_fp16 = const()[name = tensor("op_15997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1527_cast_fp16 = mul(x = var_15996_cast_fp16, y = var_15997_to_fp16)[name = tensor("aw_chunk_1527_cast_fp16")]; + tensor var_16000_equation_0 = const()[name = tensor("op_16000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16000_cast_fp16 = einsum(equation = var_16000_equation_0, values = (var_15710_cast_fp16, var_15416_cast_fp16))[name = tensor("op_16000_cast_fp16")]; + tensor var_16001_to_fp16 = const()[name = tensor("op_16001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1529_cast_fp16 = mul(x = var_16000_cast_fp16, y = var_16001_to_fp16)[name = tensor("aw_chunk_1529_cast_fp16")]; + tensor var_16004_equation_0 = const()[name = tensor("op_16004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16004_cast_fp16 = einsum(equation = var_16004_equation_0, values = (var_15710_cast_fp16, var_15423_cast_fp16))[name = tensor("op_16004_cast_fp16")]; + tensor var_16005_to_fp16 = const()[name = tensor("op_16005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1531_cast_fp16 = mul(x = var_16004_cast_fp16, y = var_16005_to_fp16)[name = tensor("aw_chunk_1531_cast_fp16")]; + tensor var_16008_equation_0 = const()[name = tensor("op_16008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16008_cast_fp16 = einsum(equation = var_16008_equation_0, values = (var_15710_cast_fp16, var_15430_cast_fp16))[name = tensor("op_16008_cast_fp16")]; + tensor var_16009_to_fp16 = const()[name = tensor("op_16009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1533_cast_fp16 = mul(x = var_16008_cast_fp16, y = var_16009_to_fp16)[name = tensor("aw_chunk_1533_cast_fp16")]; + tensor var_16012_equation_0 = const()[name = tensor("op_16012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16012_cast_fp16 = einsum(equation = var_16012_equation_0, values = (var_15710_cast_fp16, var_15437_cast_fp16))[name = tensor("op_16012_cast_fp16")]; + tensor var_16013_to_fp16 = const()[name = tensor("op_16013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1535_cast_fp16 = mul(x = var_16012_cast_fp16, y = var_16013_to_fp16)[name = tensor("aw_chunk_1535_cast_fp16")]; + tensor var_16016_equation_0 = const()[name = tensor("op_16016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16016_cast_fp16 = einsum(equation = var_16016_equation_0, values = (var_15714_cast_fp16, var_15444_cast_fp16))[name = tensor("op_16016_cast_fp16")]; + tensor var_16017_to_fp16 = const()[name = tensor("op_16017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1537_cast_fp16 = mul(x = var_16016_cast_fp16, y = var_16017_to_fp16)[name = tensor("aw_chunk_1537_cast_fp16")]; + tensor var_16020_equation_0 = const()[name = tensor("op_16020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16020_cast_fp16 = einsum(equation = var_16020_equation_0, values = (var_15714_cast_fp16, var_15451_cast_fp16))[name = tensor("op_16020_cast_fp16")]; + tensor var_16021_to_fp16 = const()[name = tensor("op_16021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1539_cast_fp16 = mul(x = var_16020_cast_fp16, y = var_16021_to_fp16)[name = tensor("aw_chunk_1539_cast_fp16")]; + tensor var_16024_equation_0 = const()[name = tensor("op_16024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16024_cast_fp16 = einsum(equation = var_16024_equation_0, values = (var_15714_cast_fp16, var_15458_cast_fp16))[name = tensor("op_16024_cast_fp16")]; + tensor var_16025_to_fp16 = const()[name = tensor("op_16025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1541_cast_fp16 = mul(x = var_16024_cast_fp16, y = var_16025_to_fp16)[name = tensor("aw_chunk_1541_cast_fp16")]; + tensor var_16028_equation_0 = const()[name = tensor("op_16028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16028_cast_fp16 = einsum(equation = var_16028_equation_0, values = (var_15714_cast_fp16, var_15465_cast_fp16))[name = tensor("op_16028_cast_fp16")]; + tensor var_16029_to_fp16 = const()[name = tensor("op_16029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1543_cast_fp16 = mul(x = var_16028_cast_fp16, y = var_16029_to_fp16)[name = tensor("aw_chunk_1543_cast_fp16")]; + tensor var_16032_equation_0 = const()[name = tensor("op_16032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16032_cast_fp16 = einsum(equation = var_16032_equation_0, values = (var_15718_cast_fp16, var_15472_cast_fp16))[name = tensor("op_16032_cast_fp16")]; + tensor var_16033_to_fp16 = const()[name = tensor("op_16033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1545_cast_fp16 = mul(x = var_16032_cast_fp16, y = var_16033_to_fp16)[name = tensor("aw_chunk_1545_cast_fp16")]; + tensor var_16036_equation_0 = const()[name = tensor("op_16036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16036_cast_fp16 = einsum(equation = var_16036_equation_0, values = (var_15718_cast_fp16, var_15479_cast_fp16))[name = tensor("op_16036_cast_fp16")]; + tensor var_16037_to_fp16 = const()[name = tensor("op_16037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1547_cast_fp16 = mul(x = var_16036_cast_fp16, y = var_16037_to_fp16)[name = tensor("aw_chunk_1547_cast_fp16")]; + tensor var_16040_equation_0 = const()[name = tensor("op_16040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16040_cast_fp16 = einsum(equation = var_16040_equation_0, values = (var_15718_cast_fp16, var_15486_cast_fp16))[name = tensor("op_16040_cast_fp16")]; + tensor var_16041_to_fp16 = const()[name = tensor("op_16041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1549_cast_fp16 = mul(x = var_16040_cast_fp16, y = var_16041_to_fp16)[name = tensor("aw_chunk_1549_cast_fp16")]; + tensor var_16044_equation_0 = const()[name = tensor("op_16044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16044_cast_fp16 = einsum(equation = var_16044_equation_0, values = (var_15718_cast_fp16, var_15493_cast_fp16))[name = tensor("op_16044_cast_fp16")]; + tensor var_16045_to_fp16 = const()[name = tensor("op_16045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1551_cast_fp16 = mul(x = var_16044_cast_fp16, y = var_16045_to_fp16)[name = tensor("aw_chunk_1551_cast_fp16")]; + tensor var_16048_equation_0 = const()[name = tensor("op_16048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16048_cast_fp16 = einsum(equation = var_16048_equation_0, values = (var_15722_cast_fp16, var_15500_cast_fp16))[name = tensor("op_16048_cast_fp16")]; + tensor var_16049_to_fp16 = const()[name = tensor("op_16049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1553_cast_fp16 = mul(x = var_16048_cast_fp16, y = var_16049_to_fp16)[name = tensor("aw_chunk_1553_cast_fp16")]; + tensor var_16052_equation_0 = const()[name = tensor("op_16052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16052_cast_fp16 = einsum(equation = var_16052_equation_0, values = (var_15722_cast_fp16, var_15507_cast_fp16))[name = tensor("op_16052_cast_fp16")]; + tensor var_16053_to_fp16 = const()[name = tensor("op_16053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1555_cast_fp16 = mul(x = var_16052_cast_fp16, y = var_16053_to_fp16)[name = tensor("aw_chunk_1555_cast_fp16")]; + tensor var_16056_equation_0 = const()[name = tensor("op_16056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16056_cast_fp16 = einsum(equation = var_16056_equation_0, values = (var_15722_cast_fp16, var_15514_cast_fp16))[name = tensor("op_16056_cast_fp16")]; + tensor var_16057_to_fp16 = const()[name = tensor("op_16057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1557_cast_fp16 = mul(x = var_16056_cast_fp16, y = var_16057_to_fp16)[name = tensor("aw_chunk_1557_cast_fp16")]; + tensor var_16060_equation_0 = const()[name = tensor("op_16060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16060_cast_fp16 = einsum(equation = var_16060_equation_0, values = (var_15722_cast_fp16, var_15521_cast_fp16))[name = tensor("op_16060_cast_fp16")]; + tensor var_16061_to_fp16 = const()[name = tensor("op_16061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1559_cast_fp16 = mul(x = var_16060_cast_fp16, y = var_16061_to_fp16)[name = tensor("aw_chunk_1559_cast_fp16")]; + tensor var_16064_equation_0 = const()[name = tensor("op_16064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16064_cast_fp16 = einsum(equation = var_16064_equation_0, values = (var_15726_cast_fp16, var_15528_cast_fp16))[name = tensor("op_16064_cast_fp16")]; + tensor var_16065_to_fp16 = const()[name = tensor("op_16065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1561_cast_fp16 = mul(x = var_16064_cast_fp16, y = var_16065_to_fp16)[name = tensor("aw_chunk_1561_cast_fp16")]; + tensor var_16068_equation_0 = const()[name = tensor("op_16068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16068_cast_fp16 = einsum(equation = var_16068_equation_0, values = (var_15726_cast_fp16, var_15535_cast_fp16))[name = tensor("op_16068_cast_fp16")]; + tensor var_16069_to_fp16 = const()[name = tensor("op_16069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1563_cast_fp16 = mul(x = var_16068_cast_fp16, y = var_16069_to_fp16)[name = tensor("aw_chunk_1563_cast_fp16")]; + tensor var_16072_equation_0 = const()[name = tensor("op_16072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16072_cast_fp16 = einsum(equation = var_16072_equation_0, values = (var_15726_cast_fp16, var_15542_cast_fp16))[name = tensor("op_16072_cast_fp16")]; + tensor var_16073_to_fp16 = const()[name = tensor("op_16073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1565_cast_fp16 = mul(x = var_16072_cast_fp16, y = var_16073_to_fp16)[name = tensor("aw_chunk_1565_cast_fp16")]; + tensor var_16076_equation_0 = const()[name = tensor("op_16076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16076_cast_fp16 = einsum(equation = var_16076_equation_0, values = (var_15726_cast_fp16, var_15549_cast_fp16))[name = tensor("op_16076_cast_fp16")]; + tensor var_16077_to_fp16 = const()[name = tensor("op_16077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1567_cast_fp16 = mul(x = var_16076_cast_fp16, y = var_16077_to_fp16)[name = tensor("aw_chunk_1567_cast_fp16")]; + tensor var_16080_equation_0 = const()[name = tensor("op_16080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16080_cast_fp16 = einsum(equation = var_16080_equation_0, values = (var_15730_cast_fp16, var_15556_cast_fp16))[name = tensor("op_16080_cast_fp16")]; + tensor var_16081_to_fp16 = const()[name = tensor("op_16081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1569_cast_fp16 = mul(x = var_16080_cast_fp16, y = var_16081_to_fp16)[name = tensor("aw_chunk_1569_cast_fp16")]; + tensor var_16084_equation_0 = const()[name = tensor("op_16084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16084_cast_fp16 = einsum(equation = var_16084_equation_0, values = (var_15730_cast_fp16, var_15563_cast_fp16))[name = tensor("op_16084_cast_fp16")]; + tensor var_16085_to_fp16 = const()[name = tensor("op_16085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1571_cast_fp16 = mul(x = var_16084_cast_fp16, y = var_16085_to_fp16)[name = tensor("aw_chunk_1571_cast_fp16")]; + tensor var_16088_equation_0 = const()[name = tensor("op_16088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16088_cast_fp16 = einsum(equation = var_16088_equation_0, values = (var_15730_cast_fp16, var_15570_cast_fp16))[name = tensor("op_16088_cast_fp16")]; + tensor var_16089_to_fp16 = const()[name = tensor("op_16089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1573_cast_fp16 = mul(x = var_16088_cast_fp16, y = var_16089_to_fp16)[name = tensor("aw_chunk_1573_cast_fp16")]; + tensor var_16092_equation_0 = const()[name = tensor("op_16092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16092_cast_fp16 = einsum(equation = var_16092_equation_0, values = (var_15730_cast_fp16, var_15577_cast_fp16))[name = tensor("op_16092_cast_fp16")]; + tensor var_16093_to_fp16 = const()[name = tensor("op_16093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1575_cast_fp16 = mul(x = var_16092_cast_fp16, y = var_16093_to_fp16)[name = tensor("aw_chunk_1575_cast_fp16")]; + tensor var_16096_equation_0 = const()[name = tensor("op_16096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16096_cast_fp16 = einsum(equation = var_16096_equation_0, values = (var_15734_cast_fp16, var_15584_cast_fp16))[name = tensor("op_16096_cast_fp16")]; + tensor var_16097_to_fp16 = const()[name = tensor("op_16097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1577_cast_fp16 = mul(x = var_16096_cast_fp16, y = var_16097_to_fp16)[name = tensor("aw_chunk_1577_cast_fp16")]; + tensor var_16100_equation_0 = const()[name = tensor("op_16100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16100_cast_fp16 = einsum(equation = var_16100_equation_0, values = (var_15734_cast_fp16, var_15591_cast_fp16))[name = tensor("op_16100_cast_fp16")]; + tensor var_16101_to_fp16 = const()[name = tensor("op_16101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1579_cast_fp16 = mul(x = var_16100_cast_fp16, y = var_16101_to_fp16)[name = tensor("aw_chunk_1579_cast_fp16")]; + tensor var_16104_equation_0 = const()[name = tensor("op_16104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16104_cast_fp16 = einsum(equation = var_16104_equation_0, values = (var_15734_cast_fp16, var_15598_cast_fp16))[name = tensor("op_16104_cast_fp16")]; + tensor var_16105_to_fp16 = const()[name = tensor("op_16105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1581_cast_fp16 = mul(x = var_16104_cast_fp16, y = var_16105_to_fp16)[name = tensor("aw_chunk_1581_cast_fp16")]; + tensor var_16108_equation_0 = const()[name = tensor("op_16108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16108_cast_fp16 = einsum(equation = var_16108_equation_0, values = (var_15734_cast_fp16, var_15605_cast_fp16))[name = tensor("op_16108_cast_fp16")]; + tensor var_16109_to_fp16 = const()[name = tensor("op_16109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1583_cast_fp16 = mul(x = var_16108_cast_fp16, y = var_16109_to_fp16)[name = tensor("aw_chunk_1583_cast_fp16")]; + tensor var_16112_equation_0 = const()[name = tensor("op_16112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16112_cast_fp16 = einsum(equation = var_16112_equation_0, values = (var_15738_cast_fp16, var_15612_cast_fp16))[name = tensor("op_16112_cast_fp16")]; + tensor var_16113_to_fp16 = const()[name = tensor("op_16113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1585_cast_fp16 = mul(x = var_16112_cast_fp16, y = var_16113_to_fp16)[name = tensor("aw_chunk_1585_cast_fp16")]; + tensor var_16116_equation_0 = const()[name = tensor("op_16116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16116_cast_fp16 = einsum(equation = var_16116_equation_0, values = (var_15738_cast_fp16, var_15619_cast_fp16))[name = tensor("op_16116_cast_fp16")]; + tensor var_16117_to_fp16 = const()[name = tensor("op_16117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1587_cast_fp16 = mul(x = var_16116_cast_fp16, y = var_16117_to_fp16)[name = tensor("aw_chunk_1587_cast_fp16")]; + tensor var_16120_equation_0 = const()[name = tensor("op_16120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16120_cast_fp16 = einsum(equation = var_16120_equation_0, values = (var_15738_cast_fp16, var_15626_cast_fp16))[name = tensor("op_16120_cast_fp16")]; + tensor var_16121_to_fp16 = const()[name = tensor("op_16121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1589_cast_fp16 = mul(x = var_16120_cast_fp16, y = var_16121_to_fp16)[name = tensor("aw_chunk_1589_cast_fp16")]; + tensor var_16124_equation_0 = const()[name = tensor("op_16124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16124_cast_fp16 = einsum(equation = var_16124_equation_0, values = (var_15738_cast_fp16, var_15633_cast_fp16))[name = tensor("op_16124_cast_fp16")]; + tensor var_16125_to_fp16 = const()[name = tensor("op_16125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1591_cast_fp16 = mul(x = var_16124_cast_fp16, y = var_16125_to_fp16)[name = tensor("aw_chunk_1591_cast_fp16")]; + tensor var_16128_equation_0 = const()[name = tensor("op_16128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16128_cast_fp16 = einsum(equation = var_16128_equation_0, values = (var_15742_cast_fp16, var_15640_cast_fp16))[name = tensor("op_16128_cast_fp16")]; + tensor var_16129_to_fp16 = const()[name = tensor("op_16129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1593_cast_fp16 = mul(x = var_16128_cast_fp16, y = var_16129_to_fp16)[name = tensor("aw_chunk_1593_cast_fp16")]; + tensor var_16132_equation_0 = const()[name = tensor("op_16132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16132_cast_fp16 = einsum(equation = var_16132_equation_0, values = (var_15742_cast_fp16, var_15647_cast_fp16))[name = tensor("op_16132_cast_fp16")]; + tensor var_16133_to_fp16 = const()[name = tensor("op_16133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1595_cast_fp16 = mul(x = var_16132_cast_fp16, y = var_16133_to_fp16)[name = tensor("aw_chunk_1595_cast_fp16")]; + tensor var_16136_equation_0 = const()[name = tensor("op_16136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16136_cast_fp16 = einsum(equation = var_16136_equation_0, values = (var_15742_cast_fp16, var_15654_cast_fp16))[name = tensor("op_16136_cast_fp16")]; + tensor var_16137_to_fp16 = const()[name = tensor("op_16137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1597_cast_fp16 = mul(x = var_16136_cast_fp16, y = var_16137_to_fp16)[name = tensor("aw_chunk_1597_cast_fp16")]; + tensor var_16140_equation_0 = const()[name = tensor("op_16140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_16140_cast_fp16 = einsum(equation = var_16140_equation_0, values = (var_15742_cast_fp16, var_15661_cast_fp16))[name = tensor("op_16140_cast_fp16")]; + tensor var_16141_to_fp16 = const()[name = tensor("op_16141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1599_cast_fp16 = mul(x = var_16140_cast_fp16, y = var_16141_to_fp16)[name = tensor("aw_chunk_1599_cast_fp16")]; + tensor var_16143_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1441_cast_fp16)[name = tensor("op_16143_cast_fp16")]; + tensor var_16144_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1443_cast_fp16)[name = tensor("op_16144_cast_fp16")]; + tensor var_16145_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1445_cast_fp16)[name = tensor("op_16145_cast_fp16")]; + tensor var_16146_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1447_cast_fp16)[name = tensor("op_16146_cast_fp16")]; + tensor var_16147_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1449_cast_fp16)[name = tensor("op_16147_cast_fp16")]; + tensor var_16148_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1451_cast_fp16)[name = tensor("op_16148_cast_fp16")]; + tensor var_16149_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1453_cast_fp16)[name = tensor("op_16149_cast_fp16")]; + tensor var_16150_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1455_cast_fp16)[name = tensor("op_16150_cast_fp16")]; + tensor var_16151_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1457_cast_fp16)[name = tensor("op_16151_cast_fp16")]; + tensor var_16152_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1459_cast_fp16)[name = tensor("op_16152_cast_fp16")]; + tensor var_16153_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1461_cast_fp16)[name = tensor("op_16153_cast_fp16")]; + tensor var_16154_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1463_cast_fp16)[name = tensor("op_16154_cast_fp16")]; + tensor var_16155_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1465_cast_fp16)[name = tensor("op_16155_cast_fp16")]; + tensor var_16156_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1467_cast_fp16)[name = tensor("op_16156_cast_fp16")]; + tensor var_16157_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1469_cast_fp16)[name = tensor("op_16157_cast_fp16")]; + tensor var_16158_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1471_cast_fp16)[name = tensor("op_16158_cast_fp16")]; + tensor var_16159_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1473_cast_fp16)[name = tensor("op_16159_cast_fp16")]; + tensor var_16160_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1475_cast_fp16)[name = tensor("op_16160_cast_fp16")]; + tensor var_16161_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1477_cast_fp16)[name = tensor("op_16161_cast_fp16")]; + tensor var_16162_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1479_cast_fp16)[name = tensor("op_16162_cast_fp16")]; + tensor var_16163_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1481_cast_fp16)[name = tensor("op_16163_cast_fp16")]; + tensor var_16164_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1483_cast_fp16)[name = tensor("op_16164_cast_fp16")]; + tensor var_16165_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1485_cast_fp16)[name = tensor("op_16165_cast_fp16")]; + tensor var_16166_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1487_cast_fp16)[name = tensor("op_16166_cast_fp16")]; + tensor var_16167_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1489_cast_fp16)[name = tensor("op_16167_cast_fp16")]; + tensor var_16168_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1491_cast_fp16)[name = tensor("op_16168_cast_fp16")]; + tensor var_16169_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1493_cast_fp16)[name = tensor("op_16169_cast_fp16")]; + tensor var_16170_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1495_cast_fp16)[name = tensor("op_16170_cast_fp16")]; + tensor var_16171_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1497_cast_fp16)[name = tensor("op_16171_cast_fp16")]; + tensor var_16172_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1499_cast_fp16)[name = tensor("op_16172_cast_fp16")]; + tensor var_16173_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1501_cast_fp16)[name = tensor("op_16173_cast_fp16")]; + tensor var_16174_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1503_cast_fp16)[name = tensor("op_16174_cast_fp16")]; + tensor var_16175_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1505_cast_fp16)[name = tensor("op_16175_cast_fp16")]; + tensor var_16176_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1507_cast_fp16)[name = tensor("op_16176_cast_fp16")]; + tensor var_16177_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1509_cast_fp16)[name = tensor("op_16177_cast_fp16")]; + tensor var_16178_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1511_cast_fp16)[name = tensor("op_16178_cast_fp16")]; + tensor var_16179_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1513_cast_fp16)[name = tensor("op_16179_cast_fp16")]; + tensor var_16180_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1515_cast_fp16)[name = tensor("op_16180_cast_fp16")]; + tensor var_16181_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1517_cast_fp16)[name = tensor("op_16181_cast_fp16")]; + tensor var_16182_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1519_cast_fp16)[name = tensor("op_16182_cast_fp16")]; + tensor var_16183_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1521_cast_fp16)[name = tensor("op_16183_cast_fp16")]; + tensor var_16184_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1523_cast_fp16)[name = tensor("op_16184_cast_fp16")]; + tensor var_16185_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1525_cast_fp16)[name = tensor("op_16185_cast_fp16")]; + tensor var_16186_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1527_cast_fp16)[name = tensor("op_16186_cast_fp16")]; + tensor var_16187_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1529_cast_fp16)[name = tensor("op_16187_cast_fp16")]; + tensor var_16188_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1531_cast_fp16)[name = tensor("op_16188_cast_fp16")]; + tensor var_16189_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1533_cast_fp16)[name = tensor("op_16189_cast_fp16")]; + tensor var_16190_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1535_cast_fp16)[name = tensor("op_16190_cast_fp16")]; + tensor var_16191_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1537_cast_fp16)[name = tensor("op_16191_cast_fp16")]; + tensor var_16192_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1539_cast_fp16)[name = tensor("op_16192_cast_fp16")]; + tensor var_16193_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1541_cast_fp16)[name = tensor("op_16193_cast_fp16")]; + tensor var_16194_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1543_cast_fp16)[name = tensor("op_16194_cast_fp16")]; + tensor var_16195_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1545_cast_fp16)[name = tensor("op_16195_cast_fp16")]; + tensor var_16196_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1547_cast_fp16)[name = tensor("op_16196_cast_fp16")]; + tensor var_16197_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1549_cast_fp16)[name = tensor("op_16197_cast_fp16")]; + tensor var_16198_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1551_cast_fp16)[name = tensor("op_16198_cast_fp16")]; + tensor var_16199_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1553_cast_fp16)[name = tensor("op_16199_cast_fp16")]; + tensor var_16200_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1555_cast_fp16)[name = tensor("op_16200_cast_fp16")]; + tensor var_16201_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1557_cast_fp16)[name = tensor("op_16201_cast_fp16")]; + tensor var_16202_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1559_cast_fp16)[name = tensor("op_16202_cast_fp16")]; + tensor var_16203_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1561_cast_fp16)[name = tensor("op_16203_cast_fp16")]; + tensor var_16204_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1563_cast_fp16)[name = tensor("op_16204_cast_fp16")]; + tensor var_16205_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1565_cast_fp16)[name = tensor("op_16205_cast_fp16")]; + tensor var_16206_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1567_cast_fp16)[name = tensor("op_16206_cast_fp16")]; + tensor var_16207_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1569_cast_fp16)[name = tensor("op_16207_cast_fp16")]; + tensor var_16208_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1571_cast_fp16)[name = tensor("op_16208_cast_fp16")]; + tensor var_16209_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1573_cast_fp16)[name = tensor("op_16209_cast_fp16")]; + tensor var_16210_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1575_cast_fp16)[name = tensor("op_16210_cast_fp16")]; + tensor var_16211_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1577_cast_fp16)[name = tensor("op_16211_cast_fp16")]; + tensor var_16212_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1579_cast_fp16)[name = tensor("op_16212_cast_fp16")]; + tensor var_16213_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1581_cast_fp16)[name = tensor("op_16213_cast_fp16")]; + tensor var_16214_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1583_cast_fp16)[name = tensor("op_16214_cast_fp16")]; + tensor var_16215_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1585_cast_fp16)[name = tensor("op_16215_cast_fp16")]; + tensor var_16216_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1587_cast_fp16)[name = tensor("op_16216_cast_fp16")]; + tensor var_16217_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1589_cast_fp16)[name = tensor("op_16217_cast_fp16")]; + tensor var_16218_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1591_cast_fp16)[name = tensor("op_16218_cast_fp16")]; + tensor var_16219_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1593_cast_fp16)[name = tensor("op_16219_cast_fp16")]; + tensor var_16220_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1595_cast_fp16)[name = tensor("op_16220_cast_fp16")]; + tensor var_16221_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1597_cast_fp16)[name = tensor("op_16221_cast_fp16")]; + tensor var_16222_cast_fp16 = softmax(axis = var_14914, x = aw_chunk_1599_cast_fp16)[name = tensor("op_16222_cast_fp16")]; + tensor var_16224_equation_0 = const()[name = tensor("op_16224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16224_cast_fp16 = einsum(equation = var_16224_equation_0, values = (var_15744_cast_fp16, var_16143_cast_fp16))[name = tensor("op_16224_cast_fp16")]; + tensor var_16226_equation_0 = const()[name = tensor("op_16226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16226_cast_fp16 = einsum(equation = var_16226_equation_0, values = (var_15744_cast_fp16, var_16144_cast_fp16))[name = tensor("op_16226_cast_fp16")]; + tensor var_16228_equation_0 = const()[name = tensor("op_16228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16228_cast_fp16 = einsum(equation = var_16228_equation_0, values = (var_15744_cast_fp16, var_16145_cast_fp16))[name = tensor("op_16228_cast_fp16")]; + tensor var_16230_equation_0 = const()[name = tensor("op_16230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16230_cast_fp16 = einsum(equation = var_16230_equation_0, values = (var_15744_cast_fp16, var_16146_cast_fp16))[name = tensor("op_16230_cast_fp16")]; + tensor var_16232_equation_0 = const()[name = tensor("op_16232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16232_cast_fp16 = einsum(equation = var_16232_equation_0, values = (var_15748_cast_fp16, var_16147_cast_fp16))[name = tensor("op_16232_cast_fp16")]; + tensor var_16234_equation_0 = const()[name = tensor("op_16234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16234_cast_fp16 = einsum(equation = var_16234_equation_0, values = (var_15748_cast_fp16, var_16148_cast_fp16))[name = tensor("op_16234_cast_fp16")]; + tensor var_16236_equation_0 = const()[name = tensor("op_16236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16236_cast_fp16 = einsum(equation = var_16236_equation_0, values = (var_15748_cast_fp16, var_16149_cast_fp16))[name = tensor("op_16236_cast_fp16")]; + tensor var_16238_equation_0 = const()[name = tensor("op_16238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16238_cast_fp16 = einsum(equation = var_16238_equation_0, values = (var_15748_cast_fp16, var_16150_cast_fp16))[name = tensor("op_16238_cast_fp16")]; + tensor var_16240_equation_0 = const()[name = tensor("op_16240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16240_cast_fp16 = einsum(equation = var_16240_equation_0, values = (var_15752_cast_fp16, var_16151_cast_fp16))[name = tensor("op_16240_cast_fp16")]; + tensor var_16242_equation_0 = const()[name = tensor("op_16242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16242_cast_fp16 = einsum(equation = var_16242_equation_0, values = (var_15752_cast_fp16, var_16152_cast_fp16))[name = tensor("op_16242_cast_fp16")]; + tensor var_16244_equation_0 = const()[name = tensor("op_16244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16244_cast_fp16 = einsum(equation = var_16244_equation_0, values = (var_15752_cast_fp16, var_16153_cast_fp16))[name = tensor("op_16244_cast_fp16")]; + tensor var_16246_equation_0 = const()[name = tensor("op_16246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16246_cast_fp16 = einsum(equation = var_16246_equation_0, values = (var_15752_cast_fp16, var_16154_cast_fp16))[name = tensor("op_16246_cast_fp16")]; + tensor var_16248_equation_0 = const()[name = tensor("op_16248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16248_cast_fp16 = einsum(equation = var_16248_equation_0, values = (var_15756_cast_fp16, var_16155_cast_fp16))[name = tensor("op_16248_cast_fp16")]; + tensor var_16250_equation_0 = const()[name = tensor("op_16250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16250_cast_fp16 = einsum(equation = var_16250_equation_0, values = (var_15756_cast_fp16, var_16156_cast_fp16))[name = tensor("op_16250_cast_fp16")]; + tensor var_16252_equation_0 = const()[name = tensor("op_16252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16252_cast_fp16 = einsum(equation = var_16252_equation_0, values = (var_15756_cast_fp16, var_16157_cast_fp16))[name = tensor("op_16252_cast_fp16")]; + tensor var_16254_equation_0 = const()[name = tensor("op_16254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16254_cast_fp16 = einsum(equation = var_16254_equation_0, values = (var_15756_cast_fp16, var_16158_cast_fp16))[name = tensor("op_16254_cast_fp16")]; + tensor var_16256_equation_0 = const()[name = tensor("op_16256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16256_cast_fp16 = einsum(equation = var_16256_equation_0, values = (var_15760_cast_fp16, var_16159_cast_fp16))[name = tensor("op_16256_cast_fp16")]; + tensor var_16258_equation_0 = const()[name = tensor("op_16258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16258_cast_fp16 = einsum(equation = var_16258_equation_0, values = (var_15760_cast_fp16, var_16160_cast_fp16))[name = tensor("op_16258_cast_fp16")]; + tensor var_16260_equation_0 = const()[name = tensor("op_16260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16260_cast_fp16 = einsum(equation = var_16260_equation_0, values = (var_15760_cast_fp16, var_16161_cast_fp16))[name = tensor("op_16260_cast_fp16")]; + tensor var_16262_equation_0 = const()[name = tensor("op_16262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16262_cast_fp16 = einsum(equation = var_16262_equation_0, values = (var_15760_cast_fp16, var_16162_cast_fp16))[name = tensor("op_16262_cast_fp16")]; + tensor var_16264_equation_0 = const()[name = tensor("op_16264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16264_cast_fp16 = einsum(equation = var_16264_equation_0, values = (var_15764_cast_fp16, var_16163_cast_fp16))[name = tensor("op_16264_cast_fp16")]; + tensor var_16266_equation_0 = const()[name = tensor("op_16266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16266_cast_fp16 = einsum(equation = var_16266_equation_0, values = (var_15764_cast_fp16, var_16164_cast_fp16))[name = tensor("op_16266_cast_fp16")]; + tensor var_16268_equation_0 = const()[name = tensor("op_16268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16268_cast_fp16 = einsum(equation = var_16268_equation_0, values = (var_15764_cast_fp16, var_16165_cast_fp16))[name = tensor("op_16268_cast_fp16")]; + tensor var_16270_equation_0 = const()[name = tensor("op_16270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16270_cast_fp16 = einsum(equation = var_16270_equation_0, values = (var_15764_cast_fp16, var_16166_cast_fp16))[name = tensor("op_16270_cast_fp16")]; + tensor var_16272_equation_0 = const()[name = tensor("op_16272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16272_cast_fp16 = einsum(equation = var_16272_equation_0, values = (var_15768_cast_fp16, var_16167_cast_fp16))[name = tensor("op_16272_cast_fp16")]; + tensor var_16274_equation_0 = const()[name = tensor("op_16274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16274_cast_fp16 = einsum(equation = var_16274_equation_0, values = (var_15768_cast_fp16, var_16168_cast_fp16))[name = tensor("op_16274_cast_fp16")]; + tensor var_16276_equation_0 = const()[name = tensor("op_16276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16276_cast_fp16 = einsum(equation = var_16276_equation_0, values = (var_15768_cast_fp16, var_16169_cast_fp16))[name = tensor("op_16276_cast_fp16")]; + tensor var_16278_equation_0 = const()[name = tensor("op_16278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16278_cast_fp16 = einsum(equation = var_16278_equation_0, values = (var_15768_cast_fp16, var_16170_cast_fp16))[name = tensor("op_16278_cast_fp16")]; + tensor var_16280_equation_0 = const()[name = tensor("op_16280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16280_cast_fp16 = einsum(equation = var_16280_equation_0, values = (var_15772_cast_fp16, var_16171_cast_fp16))[name = tensor("op_16280_cast_fp16")]; + tensor var_16282_equation_0 = const()[name = tensor("op_16282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16282_cast_fp16 = einsum(equation = var_16282_equation_0, values = (var_15772_cast_fp16, var_16172_cast_fp16))[name = tensor("op_16282_cast_fp16")]; + tensor var_16284_equation_0 = const()[name = tensor("op_16284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16284_cast_fp16 = einsum(equation = var_16284_equation_0, values = (var_15772_cast_fp16, var_16173_cast_fp16))[name = tensor("op_16284_cast_fp16")]; + tensor var_16286_equation_0 = const()[name = tensor("op_16286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16286_cast_fp16 = einsum(equation = var_16286_equation_0, values = (var_15772_cast_fp16, var_16174_cast_fp16))[name = tensor("op_16286_cast_fp16")]; + tensor var_16288_equation_0 = const()[name = tensor("op_16288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16288_cast_fp16 = einsum(equation = var_16288_equation_0, values = (var_15776_cast_fp16, var_16175_cast_fp16))[name = tensor("op_16288_cast_fp16")]; + tensor var_16290_equation_0 = const()[name = tensor("op_16290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16290_cast_fp16 = einsum(equation = var_16290_equation_0, values = (var_15776_cast_fp16, var_16176_cast_fp16))[name = tensor("op_16290_cast_fp16")]; + tensor var_16292_equation_0 = const()[name = tensor("op_16292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16292_cast_fp16 = einsum(equation = var_16292_equation_0, values = (var_15776_cast_fp16, var_16177_cast_fp16))[name = tensor("op_16292_cast_fp16")]; + tensor var_16294_equation_0 = const()[name = tensor("op_16294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16294_cast_fp16 = einsum(equation = var_16294_equation_0, values = (var_15776_cast_fp16, var_16178_cast_fp16))[name = tensor("op_16294_cast_fp16")]; + tensor var_16296_equation_0 = const()[name = tensor("op_16296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16296_cast_fp16 = einsum(equation = var_16296_equation_0, values = (var_15780_cast_fp16, var_16179_cast_fp16))[name = tensor("op_16296_cast_fp16")]; + tensor var_16298_equation_0 = const()[name = tensor("op_16298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16298_cast_fp16 = einsum(equation = var_16298_equation_0, values = (var_15780_cast_fp16, var_16180_cast_fp16))[name = tensor("op_16298_cast_fp16")]; + tensor var_16300_equation_0 = const()[name = tensor("op_16300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16300_cast_fp16 = einsum(equation = var_16300_equation_0, values = (var_15780_cast_fp16, var_16181_cast_fp16))[name = tensor("op_16300_cast_fp16")]; + tensor var_16302_equation_0 = const()[name = tensor("op_16302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16302_cast_fp16 = einsum(equation = var_16302_equation_0, values = (var_15780_cast_fp16, var_16182_cast_fp16))[name = tensor("op_16302_cast_fp16")]; + tensor var_16304_equation_0 = const()[name = tensor("op_16304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16304_cast_fp16 = einsum(equation = var_16304_equation_0, values = (var_15784_cast_fp16, var_16183_cast_fp16))[name = tensor("op_16304_cast_fp16")]; + tensor var_16306_equation_0 = const()[name = tensor("op_16306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16306_cast_fp16 = einsum(equation = var_16306_equation_0, values = (var_15784_cast_fp16, var_16184_cast_fp16))[name = tensor("op_16306_cast_fp16")]; + tensor var_16308_equation_0 = const()[name = tensor("op_16308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16308_cast_fp16 = einsum(equation = var_16308_equation_0, values = (var_15784_cast_fp16, var_16185_cast_fp16))[name = tensor("op_16308_cast_fp16")]; + tensor var_16310_equation_0 = const()[name = tensor("op_16310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16310_cast_fp16 = einsum(equation = var_16310_equation_0, values = (var_15784_cast_fp16, var_16186_cast_fp16))[name = tensor("op_16310_cast_fp16")]; + tensor var_16312_equation_0 = const()[name = tensor("op_16312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16312_cast_fp16 = einsum(equation = var_16312_equation_0, values = (var_15788_cast_fp16, var_16187_cast_fp16))[name = tensor("op_16312_cast_fp16")]; + tensor var_16314_equation_0 = const()[name = tensor("op_16314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16314_cast_fp16 = einsum(equation = var_16314_equation_0, values = (var_15788_cast_fp16, var_16188_cast_fp16))[name = tensor("op_16314_cast_fp16")]; + tensor var_16316_equation_0 = const()[name = tensor("op_16316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16316_cast_fp16 = einsum(equation = var_16316_equation_0, values = (var_15788_cast_fp16, var_16189_cast_fp16))[name = tensor("op_16316_cast_fp16")]; + tensor var_16318_equation_0 = const()[name = tensor("op_16318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16318_cast_fp16 = einsum(equation = var_16318_equation_0, values = (var_15788_cast_fp16, var_16190_cast_fp16))[name = tensor("op_16318_cast_fp16")]; + tensor var_16320_equation_0 = const()[name = tensor("op_16320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16320_cast_fp16 = einsum(equation = var_16320_equation_0, values = (var_15792_cast_fp16, var_16191_cast_fp16))[name = tensor("op_16320_cast_fp16")]; + tensor var_16322_equation_0 = const()[name = tensor("op_16322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16322_cast_fp16 = einsum(equation = var_16322_equation_0, values = (var_15792_cast_fp16, var_16192_cast_fp16))[name = tensor("op_16322_cast_fp16")]; + tensor var_16324_equation_0 = const()[name = tensor("op_16324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16324_cast_fp16 = einsum(equation = var_16324_equation_0, values = (var_15792_cast_fp16, var_16193_cast_fp16))[name = tensor("op_16324_cast_fp16")]; + tensor var_16326_equation_0 = const()[name = tensor("op_16326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16326_cast_fp16 = einsum(equation = var_16326_equation_0, values = (var_15792_cast_fp16, var_16194_cast_fp16))[name = tensor("op_16326_cast_fp16")]; + tensor var_16328_equation_0 = const()[name = tensor("op_16328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16328_cast_fp16 = einsum(equation = var_16328_equation_0, values = (var_15796_cast_fp16, var_16195_cast_fp16))[name = tensor("op_16328_cast_fp16")]; + tensor var_16330_equation_0 = const()[name = tensor("op_16330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16330_cast_fp16 = einsum(equation = var_16330_equation_0, values = (var_15796_cast_fp16, var_16196_cast_fp16))[name = tensor("op_16330_cast_fp16")]; + tensor var_16332_equation_0 = const()[name = tensor("op_16332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16332_cast_fp16 = einsum(equation = var_16332_equation_0, values = (var_15796_cast_fp16, var_16197_cast_fp16))[name = tensor("op_16332_cast_fp16")]; + tensor var_16334_equation_0 = const()[name = tensor("op_16334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16334_cast_fp16 = einsum(equation = var_16334_equation_0, values = (var_15796_cast_fp16, var_16198_cast_fp16))[name = tensor("op_16334_cast_fp16")]; + tensor var_16336_equation_0 = const()[name = tensor("op_16336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16336_cast_fp16 = einsum(equation = var_16336_equation_0, values = (var_15800_cast_fp16, var_16199_cast_fp16))[name = tensor("op_16336_cast_fp16")]; + tensor var_16338_equation_0 = const()[name = tensor("op_16338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16338_cast_fp16 = einsum(equation = var_16338_equation_0, values = (var_15800_cast_fp16, var_16200_cast_fp16))[name = tensor("op_16338_cast_fp16")]; + tensor var_16340_equation_0 = const()[name = tensor("op_16340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16340_cast_fp16 = einsum(equation = var_16340_equation_0, values = (var_15800_cast_fp16, var_16201_cast_fp16))[name = tensor("op_16340_cast_fp16")]; + tensor var_16342_equation_0 = const()[name = tensor("op_16342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16342_cast_fp16 = einsum(equation = var_16342_equation_0, values = (var_15800_cast_fp16, var_16202_cast_fp16))[name = tensor("op_16342_cast_fp16")]; + tensor var_16344_equation_0 = const()[name = tensor("op_16344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16344_cast_fp16 = einsum(equation = var_16344_equation_0, values = (var_15804_cast_fp16, var_16203_cast_fp16))[name = tensor("op_16344_cast_fp16")]; + tensor var_16346_equation_0 = const()[name = tensor("op_16346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16346_cast_fp16 = einsum(equation = var_16346_equation_0, values = (var_15804_cast_fp16, var_16204_cast_fp16))[name = tensor("op_16346_cast_fp16")]; + tensor var_16348_equation_0 = const()[name = tensor("op_16348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16348_cast_fp16 = einsum(equation = var_16348_equation_0, values = (var_15804_cast_fp16, var_16205_cast_fp16))[name = tensor("op_16348_cast_fp16")]; + tensor var_16350_equation_0 = const()[name = tensor("op_16350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16350_cast_fp16 = einsum(equation = var_16350_equation_0, values = (var_15804_cast_fp16, var_16206_cast_fp16))[name = tensor("op_16350_cast_fp16")]; + tensor var_16352_equation_0 = const()[name = tensor("op_16352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16352_cast_fp16 = einsum(equation = var_16352_equation_0, values = (var_15808_cast_fp16, var_16207_cast_fp16))[name = tensor("op_16352_cast_fp16")]; + tensor var_16354_equation_0 = const()[name = tensor("op_16354_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16354_cast_fp16 = einsum(equation = var_16354_equation_0, values = (var_15808_cast_fp16, var_16208_cast_fp16))[name = tensor("op_16354_cast_fp16")]; + tensor var_16356_equation_0 = const()[name = tensor("op_16356_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16356_cast_fp16 = einsum(equation = var_16356_equation_0, values = (var_15808_cast_fp16, var_16209_cast_fp16))[name = tensor("op_16356_cast_fp16")]; + tensor var_16358_equation_0 = const()[name = tensor("op_16358_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16358_cast_fp16 = einsum(equation = var_16358_equation_0, values = (var_15808_cast_fp16, var_16210_cast_fp16))[name = tensor("op_16358_cast_fp16")]; + tensor var_16360_equation_0 = const()[name = tensor("op_16360_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16360_cast_fp16 = einsum(equation = var_16360_equation_0, values = (var_15812_cast_fp16, var_16211_cast_fp16))[name = tensor("op_16360_cast_fp16")]; + tensor var_16362_equation_0 = const()[name = tensor("op_16362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16362_cast_fp16 = einsum(equation = var_16362_equation_0, values = (var_15812_cast_fp16, var_16212_cast_fp16))[name = tensor("op_16362_cast_fp16")]; + tensor var_16364_equation_0 = const()[name = tensor("op_16364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16364_cast_fp16 = einsum(equation = var_16364_equation_0, values = (var_15812_cast_fp16, var_16213_cast_fp16))[name = tensor("op_16364_cast_fp16")]; + tensor var_16366_equation_0 = const()[name = tensor("op_16366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16366_cast_fp16 = einsum(equation = var_16366_equation_0, values = (var_15812_cast_fp16, var_16214_cast_fp16))[name = tensor("op_16366_cast_fp16")]; + tensor var_16368_equation_0 = const()[name = tensor("op_16368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16368_cast_fp16 = einsum(equation = var_16368_equation_0, values = (var_15816_cast_fp16, var_16215_cast_fp16))[name = tensor("op_16368_cast_fp16")]; + tensor var_16370_equation_0 = const()[name = tensor("op_16370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16370_cast_fp16 = einsum(equation = var_16370_equation_0, values = (var_15816_cast_fp16, var_16216_cast_fp16))[name = tensor("op_16370_cast_fp16")]; + tensor var_16372_equation_0 = const()[name = tensor("op_16372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16372_cast_fp16 = einsum(equation = var_16372_equation_0, values = (var_15816_cast_fp16, var_16217_cast_fp16))[name = tensor("op_16372_cast_fp16")]; + tensor var_16374_equation_0 = const()[name = tensor("op_16374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16374_cast_fp16 = einsum(equation = var_16374_equation_0, values = (var_15816_cast_fp16, var_16218_cast_fp16))[name = tensor("op_16374_cast_fp16")]; + tensor var_16376_equation_0 = const()[name = tensor("op_16376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16376_cast_fp16 = einsum(equation = var_16376_equation_0, values = (var_15820_cast_fp16, var_16219_cast_fp16))[name = tensor("op_16376_cast_fp16")]; + tensor var_16378_equation_0 = const()[name = tensor("op_16378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16378_cast_fp16 = einsum(equation = var_16378_equation_0, values = (var_15820_cast_fp16, var_16220_cast_fp16))[name = tensor("op_16378_cast_fp16")]; + tensor var_16380_equation_0 = const()[name = tensor("op_16380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16380_cast_fp16 = einsum(equation = var_16380_equation_0, values = (var_15820_cast_fp16, var_16221_cast_fp16))[name = tensor("op_16380_cast_fp16")]; + tensor var_16382_equation_0 = const()[name = tensor("op_16382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_16382_cast_fp16 = einsum(equation = var_16382_equation_0, values = (var_15820_cast_fp16, var_16222_cast_fp16))[name = tensor("op_16382_cast_fp16")]; + tensor var_16384_interleave_0 = const()[name = tensor("op_16384_interleave_0"), val = tensor(false)]; + tensor var_16384_cast_fp16 = concat(axis = var_14889, interleave = var_16384_interleave_0, values = (var_16224_cast_fp16, var_16226_cast_fp16, var_16228_cast_fp16, var_16230_cast_fp16))[name = tensor("op_16384_cast_fp16")]; + tensor var_16386_interleave_0 = const()[name = tensor("op_16386_interleave_0"), val = tensor(false)]; + tensor var_16386_cast_fp16 = concat(axis = var_14889, interleave = var_16386_interleave_0, values = (var_16232_cast_fp16, var_16234_cast_fp16, var_16236_cast_fp16, var_16238_cast_fp16))[name = tensor("op_16386_cast_fp16")]; + tensor var_16388_interleave_0 = const()[name = tensor("op_16388_interleave_0"), val = tensor(false)]; + tensor var_16388_cast_fp16 = concat(axis = var_14889, interleave = var_16388_interleave_0, values = (var_16240_cast_fp16, var_16242_cast_fp16, var_16244_cast_fp16, var_16246_cast_fp16))[name = tensor("op_16388_cast_fp16")]; + tensor var_16390_interleave_0 = const()[name = tensor("op_16390_interleave_0"), val = tensor(false)]; + tensor var_16390_cast_fp16 = concat(axis = var_14889, interleave = var_16390_interleave_0, values = (var_16248_cast_fp16, var_16250_cast_fp16, var_16252_cast_fp16, var_16254_cast_fp16))[name = tensor("op_16390_cast_fp16")]; + tensor var_16392_interleave_0 = const()[name = tensor("op_16392_interleave_0"), val = tensor(false)]; + tensor var_16392_cast_fp16 = concat(axis = var_14889, interleave = var_16392_interleave_0, values = (var_16256_cast_fp16, var_16258_cast_fp16, var_16260_cast_fp16, var_16262_cast_fp16))[name = tensor("op_16392_cast_fp16")]; + tensor var_16394_interleave_0 = const()[name = tensor("op_16394_interleave_0"), val = tensor(false)]; + tensor var_16394_cast_fp16 = concat(axis = var_14889, interleave = var_16394_interleave_0, values = (var_16264_cast_fp16, var_16266_cast_fp16, var_16268_cast_fp16, var_16270_cast_fp16))[name = tensor("op_16394_cast_fp16")]; + tensor var_16396_interleave_0 = const()[name = tensor("op_16396_interleave_0"), val = tensor(false)]; + tensor var_16396_cast_fp16 = concat(axis = var_14889, interleave = var_16396_interleave_0, values = (var_16272_cast_fp16, var_16274_cast_fp16, var_16276_cast_fp16, var_16278_cast_fp16))[name = tensor("op_16396_cast_fp16")]; + tensor var_16398_interleave_0 = const()[name = tensor("op_16398_interleave_0"), val = tensor(false)]; + tensor var_16398_cast_fp16 = concat(axis = var_14889, interleave = var_16398_interleave_0, values = (var_16280_cast_fp16, var_16282_cast_fp16, var_16284_cast_fp16, var_16286_cast_fp16))[name = tensor("op_16398_cast_fp16")]; + tensor var_16400_interleave_0 = const()[name = tensor("op_16400_interleave_0"), val = tensor(false)]; + tensor var_16400_cast_fp16 = concat(axis = var_14889, interleave = var_16400_interleave_0, values = (var_16288_cast_fp16, var_16290_cast_fp16, var_16292_cast_fp16, var_16294_cast_fp16))[name = tensor("op_16400_cast_fp16")]; + tensor var_16402_interleave_0 = const()[name = tensor("op_16402_interleave_0"), val = tensor(false)]; + tensor var_16402_cast_fp16 = concat(axis = var_14889, interleave = var_16402_interleave_0, values = (var_16296_cast_fp16, var_16298_cast_fp16, var_16300_cast_fp16, var_16302_cast_fp16))[name = tensor("op_16402_cast_fp16")]; + tensor var_16404_interleave_0 = const()[name = tensor("op_16404_interleave_0"), val = tensor(false)]; + tensor var_16404_cast_fp16 = concat(axis = var_14889, interleave = var_16404_interleave_0, values = (var_16304_cast_fp16, var_16306_cast_fp16, var_16308_cast_fp16, var_16310_cast_fp16))[name = tensor("op_16404_cast_fp16")]; + tensor var_16406_interleave_0 = const()[name = tensor("op_16406_interleave_0"), val = tensor(false)]; + tensor var_16406_cast_fp16 = concat(axis = var_14889, interleave = var_16406_interleave_0, values = (var_16312_cast_fp16, var_16314_cast_fp16, var_16316_cast_fp16, var_16318_cast_fp16))[name = tensor("op_16406_cast_fp16")]; + tensor var_16408_interleave_0 = const()[name = tensor("op_16408_interleave_0"), val = tensor(false)]; + tensor var_16408_cast_fp16 = concat(axis = var_14889, interleave = var_16408_interleave_0, values = (var_16320_cast_fp16, var_16322_cast_fp16, var_16324_cast_fp16, var_16326_cast_fp16))[name = tensor("op_16408_cast_fp16")]; + tensor var_16410_interleave_0 = const()[name = tensor("op_16410_interleave_0"), val = tensor(false)]; + tensor var_16410_cast_fp16 = concat(axis = var_14889, interleave = var_16410_interleave_0, values = (var_16328_cast_fp16, var_16330_cast_fp16, var_16332_cast_fp16, var_16334_cast_fp16))[name = tensor("op_16410_cast_fp16")]; + tensor var_16412_interleave_0 = const()[name = tensor("op_16412_interleave_0"), val = tensor(false)]; + tensor var_16412_cast_fp16 = concat(axis = var_14889, interleave = var_16412_interleave_0, values = (var_16336_cast_fp16, var_16338_cast_fp16, var_16340_cast_fp16, var_16342_cast_fp16))[name = tensor("op_16412_cast_fp16")]; + tensor var_16414_interleave_0 = const()[name = tensor("op_16414_interleave_0"), val = tensor(false)]; + tensor var_16414_cast_fp16 = concat(axis = var_14889, interleave = var_16414_interleave_0, values = (var_16344_cast_fp16, var_16346_cast_fp16, var_16348_cast_fp16, var_16350_cast_fp16))[name = tensor("op_16414_cast_fp16")]; + tensor var_16416_interleave_0 = const()[name = tensor("op_16416_interleave_0"), val = tensor(false)]; + tensor var_16416_cast_fp16 = concat(axis = var_14889, interleave = var_16416_interleave_0, values = (var_16352_cast_fp16, var_16354_cast_fp16, var_16356_cast_fp16, var_16358_cast_fp16))[name = tensor("op_16416_cast_fp16")]; + tensor var_16418_interleave_0 = const()[name = tensor("op_16418_interleave_0"), val = tensor(false)]; + tensor var_16418_cast_fp16 = concat(axis = var_14889, interleave = var_16418_interleave_0, values = (var_16360_cast_fp16, var_16362_cast_fp16, var_16364_cast_fp16, var_16366_cast_fp16))[name = tensor("op_16418_cast_fp16")]; + tensor var_16420_interleave_0 = const()[name = tensor("op_16420_interleave_0"), val = tensor(false)]; + tensor var_16420_cast_fp16 = concat(axis = var_14889, interleave = var_16420_interleave_0, values = (var_16368_cast_fp16, var_16370_cast_fp16, var_16372_cast_fp16, var_16374_cast_fp16))[name = tensor("op_16420_cast_fp16")]; + tensor var_16422_interleave_0 = const()[name = tensor("op_16422_interleave_0"), val = tensor(false)]; + tensor var_16422_cast_fp16 = concat(axis = var_14889, interleave = var_16422_interleave_0, values = (var_16376_cast_fp16, var_16378_cast_fp16, var_16380_cast_fp16, var_16382_cast_fp16))[name = tensor("op_16422_cast_fp16")]; + tensor input_187_interleave_0 = const()[name = tensor("input_187_interleave_0"), val = tensor(false)]; + tensor input_187_cast_fp16 = concat(axis = var_14914, interleave = input_187_interleave_0, values = (var_16384_cast_fp16, var_16386_cast_fp16, var_16388_cast_fp16, var_16390_cast_fp16, var_16392_cast_fp16, var_16394_cast_fp16, var_16396_cast_fp16, var_16398_cast_fp16, var_16400_cast_fp16, var_16402_cast_fp16, var_16404_cast_fp16, var_16406_cast_fp16, var_16408_cast_fp16, var_16410_cast_fp16, var_16412_cast_fp16, var_16414_cast_fp16, var_16416_cast_fp16, var_16418_cast_fp16, var_16420_cast_fp16, var_16422_cast_fp16))[name = tensor("input_187_cast_fp16")]; + tensor var_16430 = const()[name = tensor("op_16430"), val = tensor([1, 1])]; + tensor var_16432 = const()[name = tensor("op_16432"), val = tensor([1, 1])]; + tensor pretrained_out_115_pad_type_0 = const()[name = tensor("pretrained_out_115_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_115_pad_0 = const()[name = tensor("pretrained_out_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112428736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113248000))), name = tensor("layers_9_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_9_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113248128)))]; + tensor pretrained_out_115_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_16432, groups = var_14914, pad = pretrained_out_115_pad_0, pad_type = pretrained_out_115_pad_type_0, strides = var_16430, weight = layers_9_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_187_cast_fp16)[name = tensor("pretrained_out_115_cast_fp16")]; + tensor var_16436 = const()[name = tensor("op_16436"), val = tensor([1, 1])]; + tensor var_16438 = const()[name = tensor("op_16438"), val = tensor([1, 1])]; + tensor input_189_pad_type_0 = const()[name = tensor("input_189_pad_type_0"), val = tensor("custom")]; + tensor input_189_pad_0 = const()[name = tensor("input_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113250752)))]; + tensor input_189_cast_fp16 = conv(dilations = var_16438, groups = var_14914, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = var_16436, weight = layers_9_self_attn_o_proj_loraA_weight_to_fp16, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor var_16442 = const()[name = tensor("op_16442"), val = tensor([1, 1])]; + tensor var_16444 = const()[name = tensor("op_16444"), val = tensor([1, 1])]; + tensor lora_out_229_pad_type_0 = const()[name = tensor("lora_out_229_pad_type_0"), val = tensor("custom")]; + tensor lora_out_229_pad_0 = const()[name = tensor("lora_out_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_231_weight_0_to_fp16 = const()[name = tensor("lora_out_231_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113291776)))]; + tensor lora_out_231_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_16444, groups = var_14914, pad = lora_out_229_pad_0, pad_type = lora_out_229_pad_type_0, strides = var_16442, weight = lora_out_231_weight_0_to_fp16, x = input_189_cast_fp16)[name = tensor("lora_out_231_cast_fp16")]; + tensor obj_39_cast_fp16 = add(x = pretrained_out_115_cast_fp16, y = lora_out_231_cast_fp16)[name = tensor("obj_39_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor var_16453 = const()[name = tensor("op_16453"), val = tensor([1])]; + tensor channels_mean_39_cast_fp16 = reduce_mean(axes = var_16453, keep_dims = var_14915, x = inputs_39_cast_fp16)[name = tensor("channels_mean_39_cast_fp16")]; + tensor zero_mean_39_cast_fp16 = sub(x = inputs_39_cast_fp16, y = channels_mean_39_cast_fp16)[name = tensor("zero_mean_39_cast_fp16")]; + tensor zero_mean_sq_39_cast_fp16 = mul(x = zero_mean_39_cast_fp16, y = zero_mean_39_cast_fp16)[name = tensor("zero_mean_sq_39_cast_fp16")]; + tensor var_16457 = const()[name = tensor("op_16457"), val = tensor([1])]; + tensor var_16458_cast_fp16 = reduce_mean(axes = var_16457, keep_dims = var_14915, x = zero_mean_sq_39_cast_fp16)[name = tensor("op_16458_cast_fp16")]; + tensor var_16459_to_fp16 = const()[name = tensor("op_16459_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_16460_cast_fp16 = add(x = var_16458_cast_fp16, y = var_16459_to_fp16)[name = tensor("op_16460_cast_fp16")]; + tensor denom_39_epsilon_0 = const()[name = tensor("denom_39_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_39_cast_fp16 = rsqrt(epsilon = denom_39_epsilon_0, x = var_16460_cast_fp16)[name = tensor("denom_39_cast_fp16")]; + tensor out_39_cast_fp16 = mul(x = zero_mean_39_cast_fp16, y = denom_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; + tensor input_191_gamma_0_to_fp16 = const()[name = tensor("input_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113332800)))]; + tensor input_191_beta_0_to_fp16 = const()[name = tensor("input_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113335424)))]; + tensor input_191_epsilon_0_to_fp16 = const()[name = tensor("input_191_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_191_cast_fp16 = batch_norm(beta = input_191_beta_0_to_fp16, epsilon = input_191_epsilon_0_to_fp16, gamma = input_191_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor var_16474 = const()[name = tensor("op_16474"), val = tensor([1, 1])]; + tensor var_16476 = const()[name = tensor("op_16476"), val = tensor([1, 1])]; + tensor pretrained_out_117_pad_type_0 = const()[name = tensor("pretrained_out_117_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_117_pad_0 = const()[name = tensor("pretrained_out_117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113338048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116614912))), name = tensor("layers_9_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_9_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_9_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116615040)))]; + tensor pretrained_out_117_cast_fp16 = conv(bias = layers_9_fc1_pretrained_bias_to_fp16, dilations = var_16476, groups = var_14914, pad = pretrained_out_117_pad_0, pad_type = pretrained_out_117_pad_type_0, strides = var_16474, weight = layers_9_fc1_pretrained_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("pretrained_out_117_cast_fp16")]; + tensor var_16480 = const()[name = tensor("op_16480"), val = tensor([1, 1])]; + tensor var_16482 = const()[name = tensor("op_16482"), val = tensor([1, 1])]; + tensor input_193_pad_type_0 = const()[name = tensor("input_193_pad_type_0"), val = tensor("custom")]; + tensor input_193_pad_0 = const()[name = tensor("input_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_9_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116625344)))]; + tensor input_193_cast_fp16 = conv(dilations = var_16482, groups = var_14914, pad = input_193_pad_0, pad_type = input_193_pad_type_0, strides = var_16480, weight = layers_9_fc1_loraA_weight_to_fp16, x = input_191_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor var_16486 = const()[name = tensor("op_16486"), val = tensor([1, 1])]; + tensor var_16488 = const()[name = tensor("op_16488"), val = tensor([1, 1])]; + tensor lora_out_233_pad_type_0 = const()[name = tensor("lora_out_233_pad_type_0"), val = tensor("custom")]; + tensor lora_out_233_pad_0 = const()[name = tensor("lora_out_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_235_weight_0_to_fp16 = const()[name = tensor("lora_out_235_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116666368)))]; + tensor lora_out_235_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_16488, groups = var_14914, pad = lora_out_233_pad_0, pad_type = lora_out_233_pad_type_0, strides = var_16486, weight = lora_out_235_weight_0_to_fp16, x = input_193_cast_fp16)[name = tensor("lora_out_235_cast_fp16")]; + tensor input_195_cast_fp16 = add(x = pretrained_out_117_cast_fp16, y = lora_out_235_cast_fp16)[name = tensor("input_195_cast_fp16")]; + tensor input_197_mode_0 = const()[name = tensor("input_197_mode_0"), val = tensor("EXACT")]; + tensor input_197_cast_fp16 = gelu(mode = input_197_mode_0, x = input_195_cast_fp16)[name = tensor("input_197_cast_fp16")]; + tensor var_16500 = const()[name = tensor("op_16500"), val = tensor([1, 1])]; + tensor var_16502 = const()[name = tensor("op_16502"), val = tensor([1, 1])]; + tensor pretrained_out_119_pad_type_0 = const()[name = tensor("pretrained_out_119_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_119_pad_0 = const()[name = tensor("pretrained_out_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116830272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120107136))), name = tensor("layers_9_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_9_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_9_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120107264)))]; + tensor pretrained_out_119_cast_fp16 = conv(bias = layers_9_fc2_pretrained_bias_to_fp16, dilations = var_16502, groups = var_14914, pad = pretrained_out_119_pad_0, pad_type = pretrained_out_119_pad_type_0, strides = var_16500, weight = layers_9_fc2_pretrained_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor("pretrained_out_119_cast_fp16")]; + tensor var_16506 = const()[name = tensor("op_16506"), val = tensor([1, 1])]; + tensor var_16508 = const()[name = tensor("op_16508"), val = tensor([1, 1])]; + tensor input_199_pad_type_0 = const()[name = tensor("input_199_pad_type_0"), val = tensor("custom")]; + tensor input_199_pad_0 = const()[name = tensor("input_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_9_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_9_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120109888)))]; + tensor input_199_cast_fp16 = conv(dilations = var_16508, groups = var_14914, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = var_16506, weight = layers_9_fc2_loraA_weight_to_fp16, x = input_197_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_16512 = const()[name = tensor("op_16512"), val = tensor([1, 1])]; + tensor var_16514 = const()[name = tensor("op_16514"), val = tensor([1, 1])]; + tensor lora_out_237_pad_type_0 = const()[name = tensor("lora_out_237_pad_type_0"), val = tensor("custom")]; + tensor lora_out_237_pad_0 = const()[name = tensor("lora_out_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_239_weight_0_to_fp16 = const()[name = tensor("lora_out_239_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120273792)))]; + tensor lora_out_239_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_16514, groups = var_14914, pad = lora_out_237_pad_0, pad_type = lora_out_237_pad_type_0, strides = var_16512, weight = lora_out_239_weight_0_to_fp16, x = input_199_cast_fp16)[name = tensor("lora_out_239_cast_fp16")]; + tensor hidden_states_23_cast_fp16 = add(x = pretrained_out_119_cast_fp16, y = lora_out_239_cast_fp16)[name = tensor("hidden_states_23_cast_fp16")]; + tensor inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor var_16524 = const()[name = tensor("op_16524"), val = tensor(3)]; + tensor var_16549 = const()[name = tensor("op_16549"), val = tensor(1)]; + tensor var_16550 = const()[name = tensor("op_16550"), val = tensor(true)]; + tensor var_16560 = const()[name = tensor("op_16560"), val = tensor([1])]; + tensor channels_mean_41_cast_fp16 = reduce_mean(axes = var_16560, keep_dims = var_16550, x = inputs_41_cast_fp16)[name = tensor("channels_mean_41_cast_fp16")]; + tensor zero_mean_41_cast_fp16 = sub(x = inputs_41_cast_fp16, y = channels_mean_41_cast_fp16)[name = tensor("zero_mean_41_cast_fp16")]; + tensor zero_mean_sq_41_cast_fp16 = mul(x = zero_mean_41_cast_fp16, y = zero_mean_41_cast_fp16)[name = tensor("zero_mean_sq_41_cast_fp16")]; + tensor var_16564 = const()[name = tensor("op_16564"), val = tensor([1])]; + tensor var_16565_cast_fp16 = reduce_mean(axes = var_16564, keep_dims = var_16550, x = zero_mean_sq_41_cast_fp16)[name = tensor("op_16565_cast_fp16")]; + tensor var_16566_to_fp16 = const()[name = tensor("op_16566_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_16567_cast_fp16 = add(x = var_16565_cast_fp16, y = var_16566_to_fp16)[name = tensor("op_16567_cast_fp16")]; + tensor denom_41_epsilon_0 = const()[name = tensor("denom_41_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_41_cast_fp16 = rsqrt(epsilon = denom_41_epsilon_0, x = var_16567_cast_fp16)[name = tensor("denom_41_cast_fp16")]; + tensor out_41_cast_fp16 = mul(x = zero_mean_41_cast_fp16, y = denom_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; + tensor obj_41_gamma_0_to_fp16 = const()[name = tensor("obj_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120314816)))]; + tensor obj_41_beta_0_to_fp16 = const()[name = tensor("obj_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120317440)))]; + tensor obj_41_epsilon_0_to_fp16 = const()[name = tensor("obj_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor("obj_41_cast_fp16")]; + tensor var_16585 = const()[name = tensor("op_16585"), val = tensor([1, 1])]; + tensor var_16587 = const()[name = tensor("op_16587"), val = tensor([1, 1])]; + tensor pretrained_out_121_pad_type_0 = const()[name = tensor("pretrained_out_121_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_121_pad_0 = const()[name = tensor("pretrained_out_121_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120320064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121139328))), name = tensor("layers_10_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_10_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121139456)))]; + tensor pretrained_out_121_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_16587, groups = var_16549, pad = pretrained_out_121_pad_0, pad_type = pretrained_out_121_pad_type_0, strides = var_16585, weight = layers_10_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor("pretrained_out_121_cast_fp16")]; + tensor var_16591 = const()[name = tensor("op_16591"), val = tensor([1, 1])]; + tensor var_16593 = const()[name = tensor("op_16593"), val = tensor([1, 1])]; + tensor input_201_pad_type_0 = const()[name = tensor("input_201_pad_type_0"), val = tensor("custom")]; + tensor input_201_pad_0 = const()[name = tensor("input_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121142080)))]; + tensor input_201_cast_fp16 = conv(dilations = var_16593, groups = var_16549, pad = input_201_pad_0, pad_type = input_201_pad_type_0, strides = var_16591, weight = layers_10_self_attn_q_proj_loraA_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("input_201_cast_fp16")]; + tensor var_16597 = const()[name = tensor("op_16597"), val = tensor([1, 1])]; + tensor var_16599 = const()[name = tensor("op_16599"), val = tensor([1, 1])]; + tensor lora_out_241_pad_type_0 = const()[name = tensor("lora_out_241_pad_type_0"), val = tensor("custom")]; + tensor lora_out_241_pad_0 = const()[name = tensor("lora_out_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_243_weight_0_to_fp16 = const()[name = tensor("lora_out_243_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121183104)))]; + tensor lora_out_243_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_16599, groups = var_16549, pad = lora_out_241_pad_0, pad_type = lora_out_241_pad_type_0, strides = var_16597, weight = lora_out_243_weight_0_to_fp16, x = input_201_cast_fp16)[name = tensor("lora_out_243_cast_fp16")]; + tensor query_21_cast_fp16 = add(x = pretrained_out_121_cast_fp16, y = lora_out_243_cast_fp16)[name = tensor("query_21_cast_fp16")]; + tensor var_16609 = const()[name = tensor("op_16609"), val = tensor([1, 1])]; + tensor var_16611 = const()[name = tensor("op_16611"), val = tensor([1, 1])]; + tensor pretrained_out_123_pad_type_0 = const()[name = tensor("pretrained_out_123_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_123_pad_0 = const()[name = tensor("pretrained_out_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121224128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122043392))), name = tensor("layers_10_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_123_cast_fp16 = conv(dilations = var_16611, groups = var_16549, pad = pretrained_out_123_pad_0, pad_type = pretrained_out_123_pad_type_0, strides = var_16609, weight = layers_10_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor("pretrained_out_123_cast_fp16")]; + tensor var_16615 = const()[name = tensor("op_16615"), val = tensor([1, 1])]; + tensor var_16617 = const()[name = tensor("op_16617"), val = tensor([1, 1])]; + tensor input_203_pad_type_0 = const()[name = tensor("input_203_pad_type_0"), val = tensor("custom")]; + tensor input_203_pad_0 = const()[name = tensor("input_203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122043520)))]; + tensor input_203_cast_fp16 = conv(dilations = var_16617, groups = var_16549, pad = input_203_pad_0, pad_type = input_203_pad_type_0, strides = var_16615, weight = layers_10_self_attn_k_proj_loraA_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("input_203_cast_fp16")]; + tensor var_16621 = const()[name = tensor("op_16621"), val = tensor([1, 1])]; + tensor var_16623 = const()[name = tensor("op_16623"), val = tensor([1, 1])]; + tensor lora_out_245_pad_type_0 = const()[name = tensor("lora_out_245_pad_type_0"), val = tensor("custom")]; + tensor lora_out_245_pad_0 = const()[name = tensor("lora_out_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_247_weight_0_to_fp16 = const()[name = tensor("lora_out_247_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122084544)))]; + tensor lora_out_247_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_16623, groups = var_16549, pad = lora_out_245_pad_0, pad_type = lora_out_245_pad_type_0, strides = var_16621, weight = lora_out_247_weight_0_to_fp16, x = input_203_cast_fp16)[name = tensor("lora_out_247_cast_fp16")]; + tensor key_21_cast_fp16 = add(x = pretrained_out_123_cast_fp16, y = lora_out_247_cast_fp16)[name = tensor("key_21_cast_fp16")]; + tensor var_16634 = const()[name = tensor("op_16634"), val = tensor([1, 1])]; + tensor var_16636 = const()[name = tensor("op_16636"), val = tensor([1, 1])]; + tensor pretrained_out_125_pad_type_0 = const()[name = tensor("pretrained_out_125_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_125_pad_0 = const()[name = tensor("pretrained_out_125_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122125568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122944832))), name = tensor("layers_10_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_10_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122944960)))]; + tensor pretrained_out_125_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_16636, groups = var_16549, pad = pretrained_out_125_pad_0, pad_type = pretrained_out_125_pad_type_0, strides = var_16634, weight = layers_10_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor("pretrained_out_125_cast_fp16")]; + tensor var_16640 = const()[name = tensor("op_16640"), val = tensor([1, 1])]; + tensor var_16642 = const()[name = tensor("op_16642"), val = tensor([1, 1])]; + tensor input_205_pad_type_0 = const()[name = tensor("input_205_pad_type_0"), val = tensor("custom")]; + tensor input_205_pad_0 = const()[name = tensor("input_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122947584)))]; + tensor input_205_cast_fp16 = conv(dilations = var_16642, groups = var_16549, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = var_16640, weight = layers_10_self_attn_v_proj_loraA_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("input_205_cast_fp16")]; + tensor var_16646 = const()[name = tensor("op_16646"), val = tensor([1, 1])]; + tensor var_16648 = const()[name = tensor("op_16648"), val = tensor([1, 1])]; + tensor lora_out_249_pad_type_0 = const()[name = tensor("lora_out_249_pad_type_0"), val = tensor("custom")]; + tensor lora_out_249_pad_0 = const()[name = tensor("lora_out_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_251_weight_0_to_fp16 = const()[name = tensor("lora_out_251_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122988608)))]; + tensor lora_out_251_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_16648, groups = var_16549, pad = lora_out_249_pad_0, pad_type = lora_out_249_pad_type_0, strides = var_16646, weight = lora_out_251_weight_0_to_fp16, x = input_205_cast_fp16)[name = tensor("lora_out_251_cast_fp16")]; + tensor value_21_cast_fp16 = add(x = pretrained_out_125_cast_fp16, y = lora_out_251_cast_fp16)[name = tensor("value_21_cast_fp16")]; + tensor var_16658_begin_0 = const()[name = tensor("op_16658_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16658_end_0 = const()[name = tensor("op_16658_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16658_end_mask_0 = const()[name = tensor("op_16658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16658_cast_fp16 = slice_by_index(begin = var_16658_begin_0, end = var_16658_end_0, end_mask = var_16658_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16658_cast_fp16")]; + tensor var_16662_begin_0 = const()[name = tensor("op_16662_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_16662_end_0 = const()[name = tensor("op_16662_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_16662_end_mask_0 = const()[name = tensor("op_16662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16662_cast_fp16 = slice_by_index(begin = var_16662_begin_0, end = var_16662_end_0, end_mask = var_16662_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16662_cast_fp16")]; + tensor var_16666_begin_0 = const()[name = tensor("op_16666_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_16666_end_0 = const()[name = tensor("op_16666_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_16666_end_mask_0 = const()[name = tensor("op_16666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16666_cast_fp16 = slice_by_index(begin = var_16666_begin_0, end = var_16666_end_0, end_mask = var_16666_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16666_cast_fp16")]; + tensor var_16670_begin_0 = const()[name = tensor("op_16670_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_16670_end_0 = const()[name = tensor("op_16670_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_16670_end_mask_0 = const()[name = tensor("op_16670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16670_cast_fp16 = slice_by_index(begin = var_16670_begin_0, end = var_16670_end_0, end_mask = var_16670_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16670_cast_fp16")]; + tensor var_16674_begin_0 = const()[name = tensor("op_16674_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_16674_end_0 = const()[name = tensor("op_16674_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_16674_end_mask_0 = const()[name = tensor("op_16674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16674_cast_fp16 = slice_by_index(begin = var_16674_begin_0, end = var_16674_end_0, end_mask = var_16674_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16674_cast_fp16")]; + tensor var_16678_begin_0 = const()[name = tensor("op_16678_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_16678_end_0 = const()[name = tensor("op_16678_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_16678_end_mask_0 = const()[name = tensor("op_16678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16678_cast_fp16 = slice_by_index(begin = var_16678_begin_0, end = var_16678_end_0, end_mask = var_16678_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16678_cast_fp16")]; + tensor var_16682_begin_0 = const()[name = tensor("op_16682_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_16682_end_0 = const()[name = tensor("op_16682_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_16682_end_mask_0 = const()[name = tensor("op_16682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16682_cast_fp16 = slice_by_index(begin = var_16682_begin_0, end = var_16682_end_0, end_mask = var_16682_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16682_cast_fp16")]; + tensor var_16686_begin_0 = const()[name = tensor("op_16686_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_16686_end_0 = const()[name = tensor("op_16686_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_16686_end_mask_0 = const()[name = tensor("op_16686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16686_cast_fp16 = slice_by_index(begin = var_16686_begin_0, end = var_16686_end_0, end_mask = var_16686_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16686_cast_fp16")]; + tensor var_16690_begin_0 = const()[name = tensor("op_16690_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_16690_end_0 = const()[name = tensor("op_16690_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_16690_end_mask_0 = const()[name = tensor("op_16690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16690_cast_fp16 = slice_by_index(begin = var_16690_begin_0, end = var_16690_end_0, end_mask = var_16690_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16690_cast_fp16")]; + tensor var_16694_begin_0 = const()[name = tensor("op_16694_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_16694_end_0 = const()[name = tensor("op_16694_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_16694_end_mask_0 = const()[name = tensor("op_16694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16694_cast_fp16 = slice_by_index(begin = var_16694_begin_0, end = var_16694_end_0, end_mask = var_16694_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16694_cast_fp16")]; + tensor var_16698_begin_0 = const()[name = tensor("op_16698_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_16698_end_0 = const()[name = tensor("op_16698_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_16698_end_mask_0 = const()[name = tensor("op_16698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16698_cast_fp16 = slice_by_index(begin = var_16698_begin_0, end = var_16698_end_0, end_mask = var_16698_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16698_cast_fp16")]; + tensor var_16702_begin_0 = const()[name = tensor("op_16702_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_16702_end_0 = const()[name = tensor("op_16702_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_16702_end_mask_0 = const()[name = tensor("op_16702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16702_cast_fp16 = slice_by_index(begin = var_16702_begin_0, end = var_16702_end_0, end_mask = var_16702_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16702_cast_fp16")]; + tensor var_16706_begin_0 = const()[name = tensor("op_16706_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_16706_end_0 = const()[name = tensor("op_16706_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_16706_end_mask_0 = const()[name = tensor("op_16706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16706_cast_fp16 = slice_by_index(begin = var_16706_begin_0, end = var_16706_end_0, end_mask = var_16706_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16706_cast_fp16")]; + tensor var_16710_begin_0 = const()[name = tensor("op_16710_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_16710_end_0 = const()[name = tensor("op_16710_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_16710_end_mask_0 = const()[name = tensor("op_16710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16710_cast_fp16 = slice_by_index(begin = var_16710_begin_0, end = var_16710_end_0, end_mask = var_16710_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16710_cast_fp16")]; + tensor var_16714_begin_0 = const()[name = tensor("op_16714_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_16714_end_0 = const()[name = tensor("op_16714_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_16714_end_mask_0 = const()[name = tensor("op_16714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16714_cast_fp16 = slice_by_index(begin = var_16714_begin_0, end = var_16714_end_0, end_mask = var_16714_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16714_cast_fp16")]; + tensor var_16718_begin_0 = const()[name = tensor("op_16718_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_16718_end_0 = const()[name = tensor("op_16718_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_16718_end_mask_0 = const()[name = tensor("op_16718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16718_cast_fp16 = slice_by_index(begin = var_16718_begin_0, end = var_16718_end_0, end_mask = var_16718_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16718_cast_fp16")]; + tensor var_16722_begin_0 = const()[name = tensor("op_16722_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_16722_end_0 = const()[name = tensor("op_16722_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_16722_end_mask_0 = const()[name = tensor("op_16722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16722_cast_fp16 = slice_by_index(begin = var_16722_begin_0, end = var_16722_end_0, end_mask = var_16722_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16722_cast_fp16")]; + tensor var_16726_begin_0 = const()[name = tensor("op_16726_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_16726_end_0 = const()[name = tensor("op_16726_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_16726_end_mask_0 = const()[name = tensor("op_16726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16726_cast_fp16 = slice_by_index(begin = var_16726_begin_0, end = var_16726_end_0, end_mask = var_16726_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16726_cast_fp16")]; + tensor var_16730_begin_0 = const()[name = tensor("op_16730_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_16730_end_0 = const()[name = tensor("op_16730_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_16730_end_mask_0 = const()[name = tensor("op_16730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16730_cast_fp16 = slice_by_index(begin = var_16730_begin_0, end = var_16730_end_0, end_mask = var_16730_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16730_cast_fp16")]; + tensor var_16734_begin_0 = const()[name = tensor("op_16734_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_16734_end_0 = const()[name = tensor("op_16734_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_16734_end_mask_0 = const()[name = tensor("op_16734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_16734_cast_fp16 = slice_by_index(begin = var_16734_begin_0, end = var_16734_end_0, end_mask = var_16734_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_16734_cast_fp16")]; + tensor var_16743_begin_0 = const()[name = tensor("op_16743_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16743_end_0 = const()[name = tensor("op_16743_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16743_end_mask_0 = const()[name = tensor("op_16743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16743_cast_fp16 = slice_by_index(begin = var_16743_begin_0, end = var_16743_end_0, end_mask = var_16743_end_mask_0, x = var_16658_cast_fp16)[name = tensor("op_16743_cast_fp16")]; + tensor var_16750_begin_0 = const()[name = tensor("op_16750_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16750_end_0 = const()[name = tensor("op_16750_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16750_end_mask_0 = const()[name = tensor("op_16750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16750_cast_fp16 = slice_by_index(begin = var_16750_begin_0, end = var_16750_end_0, end_mask = var_16750_end_mask_0, x = var_16658_cast_fp16)[name = tensor("op_16750_cast_fp16")]; + tensor var_16757_begin_0 = const()[name = tensor("op_16757_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16757_end_0 = const()[name = tensor("op_16757_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16757_end_mask_0 = const()[name = tensor("op_16757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16757_cast_fp16 = slice_by_index(begin = var_16757_begin_0, end = var_16757_end_0, end_mask = var_16757_end_mask_0, x = var_16658_cast_fp16)[name = tensor("op_16757_cast_fp16")]; + tensor var_16764_begin_0 = const()[name = tensor("op_16764_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16764_end_0 = const()[name = tensor("op_16764_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16764_end_mask_0 = const()[name = tensor("op_16764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16764_cast_fp16 = slice_by_index(begin = var_16764_begin_0, end = var_16764_end_0, end_mask = var_16764_end_mask_0, x = var_16658_cast_fp16)[name = tensor("op_16764_cast_fp16")]; + tensor var_16771_begin_0 = const()[name = tensor("op_16771_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16771_end_0 = const()[name = tensor("op_16771_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16771_end_mask_0 = const()[name = tensor("op_16771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16771_cast_fp16 = slice_by_index(begin = var_16771_begin_0, end = var_16771_end_0, end_mask = var_16771_end_mask_0, x = var_16662_cast_fp16)[name = tensor("op_16771_cast_fp16")]; + tensor var_16778_begin_0 = const()[name = tensor("op_16778_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16778_end_0 = const()[name = tensor("op_16778_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16778_end_mask_0 = const()[name = tensor("op_16778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16778_cast_fp16 = slice_by_index(begin = var_16778_begin_0, end = var_16778_end_0, end_mask = var_16778_end_mask_0, x = var_16662_cast_fp16)[name = tensor("op_16778_cast_fp16")]; + tensor var_16785_begin_0 = const()[name = tensor("op_16785_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16785_end_0 = const()[name = tensor("op_16785_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16785_end_mask_0 = const()[name = tensor("op_16785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16785_cast_fp16 = slice_by_index(begin = var_16785_begin_0, end = var_16785_end_0, end_mask = var_16785_end_mask_0, x = var_16662_cast_fp16)[name = tensor("op_16785_cast_fp16")]; + tensor var_16792_begin_0 = const()[name = tensor("op_16792_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16792_end_0 = const()[name = tensor("op_16792_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16792_end_mask_0 = const()[name = tensor("op_16792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16792_cast_fp16 = slice_by_index(begin = var_16792_begin_0, end = var_16792_end_0, end_mask = var_16792_end_mask_0, x = var_16662_cast_fp16)[name = tensor("op_16792_cast_fp16")]; + tensor var_16799_begin_0 = const()[name = tensor("op_16799_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16799_end_0 = const()[name = tensor("op_16799_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16799_end_mask_0 = const()[name = tensor("op_16799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16799_cast_fp16 = slice_by_index(begin = var_16799_begin_0, end = var_16799_end_0, end_mask = var_16799_end_mask_0, x = var_16666_cast_fp16)[name = tensor("op_16799_cast_fp16")]; + tensor var_16806_begin_0 = const()[name = tensor("op_16806_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16806_end_0 = const()[name = tensor("op_16806_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16806_end_mask_0 = const()[name = tensor("op_16806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16806_cast_fp16 = slice_by_index(begin = var_16806_begin_0, end = var_16806_end_0, end_mask = var_16806_end_mask_0, x = var_16666_cast_fp16)[name = tensor("op_16806_cast_fp16")]; + tensor var_16813_begin_0 = const()[name = tensor("op_16813_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16813_end_0 = const()[name = tensor("op_16813_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16813_end_mask_0 = const()[name = tensor("op_16813_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16813_cast_fp16 = slice_by_index(begin = var_16813_begin_0, end = var_16813_end_0, end_mask = var_16813_end_mask_0, x = var_16666_cast_fp16)[name = tensor("op_16813_cast_fp16")]; + tensor var_16820_begin_0 = const()[name = tensor("op_16820_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16820_end_0 = const()[name = tensor("op_16820_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16820_end_mask_0 = const()[name = tensor("op_16820_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16820_cast_fp16 = slice_by_index(begin = var_16820_begin_0, end = var_16820_end_0, end_mask = var_16820_end_mask_0, x = var_16666_cast_fp16)[name = tensor("op_16820_cast_fp16")]; + tensor var_16827_begin_0 = const()[name = tensor("op_16827_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16827_end_0 = const()[name = tensor("op_16827_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16827_end_mask_0 = const()[name = tensor("op_16827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16827_cast_fp16 = slice_by_index(begin = var_16827_begin_0, end = var_16827_end_0, end_mask = var_16827_end_mask_0, x = var_16670_cast_fp16)[name = tensor("op_16827_cast_fp16")]; + tensor var_16834_begin_0 = const()[name = tensor("op_16834_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16834_end_0 = const()[name = tensor("op_16834_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16834_end_mask_0 = const()[name = tensor("op_16834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16834_cast_fp16 = slice_by_index(begin = var_16834_begin_0, end = var_16834_end_0, end_mask = var_16834_end_mask_0, x = var_16670_cast_fp16)[name = tensor("op_16834_cast_fp16")]; + tensor var_16841_begin_0 = const()[name = tensor("op_16841_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16841_end_0 = const()[name = tensor("op_16841_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16841_end_mask_0 = const()[name = tensor("op_16841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16841_cast_fp16 = slice_by_index(begin = var_16841_begin_0, end = var_16841_end_0, end_mask = var_16841_end_mask_0, x = var_16670_cast_fp16)[name = tensor("op_16841_cast_fp16")]; + tensor var_16848_begin_0 = const()[name = tensor("op_16848_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16848_end_0 = const()[name = tensor("op_16848_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16848_end_mask_0 = const()[name = tensor("op_16848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16848_cast_fp16 = slice_by_index(begin = var_16848_begin_0, end = var_16848_end_0, end_mask = var_16848_end_mask_0, x = var_16670_cast_fp16)[name = tensor("op_16848_cast_fp16")]; + tensor var_16855_begin_0 = const()[name = tensor("op_16855_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16855_end_0 = const()[name = tensor("op_16855_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16855_end_mask_0 = const()[name = tensor("op_16855_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16855_cast_fp16 = slice_by_index(begin = var_16855_begin_0, end = var_16855_end_0, end_mask = var_16855_end_mask_0, x = var_16674_cast_fp16)[name = tensor("op_16855_cast_fp16")]; + tensor var_16862_begin_0 = const()[name = tensor("op_16862_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16862_end_0 = const()[name = tensor("op_16862_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16862_end_mask_0 = const()[name = tensor("op_16862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16862_cast_fp16 = slice_by_index(begin = var_16862_begin_0, end = var_16862_end_0, end_mask = var_16862_end_mask_0, x = var_16674_cast_fp16)[name = tensor("op_16862_cast_fp16")]; + tensor var_16869_begin_0 = const()[name = tensor("op_16869_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16869_end_0 = const()[name = tensor("op_16869_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16869_end_mask_0 = const()[name = tensor("op_16869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16869_cast_fp16 = slice_by_index(begin = var_16869_begin_0, end = var_16869_end_0, end_mask = var_16869_end_mask_0, x = var_16674_cast_fp16)[name = tensor("op_16869_cast_fp16")]; + tensor var_16876_begin_0 = const()[name = tensor("op_16876_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16876_end_0 = const()[name = tensor("op_16876_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16876_end_mask_0 = const()[name = tensor("op_16876_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16876_cast_fp16 = slice_by_index(begin = var_16876_begin_0, end = var_16876_end_0, end_mask = var_16876_end_mask_0, x = var_16674_cast_fp16)[name = tensor("op_16876_cast_fp16")]; + tensor var_16883_begin_0 = const()[name = tensor("op_16883_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16883_end_0 = const()[name = tensor("op_16883_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16883_end_mask_0 = const()[name = tensor("op_16883_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16883_cast_fp16 = slice_by_index(begin = var_16883_begin_0, end = var_16883_end_0, end_mask = var_16883_end_mask_0, x = var_16678_cast_fp16)[name = tensor("op_16883_cast_fp16")]; + tensor var_16890_begin_0 = const()[name = tensor("op_16890_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16890_end_0 = const()[name = tensor("op_16890_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16890_end_mask_0 = const()[name = tensor("op_16890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16890_cast_fp16 = slice_by_index(begin = var_16890_begin_0, end = var_16890_end_0, end_mask = var_16890_end_mask_0, x = var_16678_cast_fp16)[name = tensor("op_16890_cast_fp16")]; + tensor var_16897_begin_0 = const()[name = tensor("op_16897_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16897_end_0 = const()[name = tensor("op_16897_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16897_end_mask_0 = const()[name = tensor("op_16897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16897_cast_fp16 = slice_by_index(begin = var_16897_begin_0, end = var_16897_end_0, end_mask = var_16897_end_mask_0, x = var_16678_cast_fp16)[name = tensor("op_16897_cast_fp16")]; + tensor var_16904_begin_0 = const()[name = tensor("op_16904_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16904_end_0 = const()[name = tensor("op_16904_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16904_end_mask_0 = const()[name = tensor("op_16904_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16904_cast_fp16 = slice_by_index(begin = var_16904_begin_0, end = var_16904_end_0, end_mask = var_16904_end_mask_0, x = var_16678_cast_fp16)[name = tensor("op_16904_cast_fp16")]; + tensor var_16911_begin_0 = const()[name = tensor("op_16911_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16911_end_0 = const()[name = tensor("op_16911_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16911_end_mask_0 = const()[name = tensor("op_16911_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16911_cast_fp16 = slice_by_index(begin = var_16911_begin_0, end = var_16911_end_0, end_mask = var_16911_end_mask_0, x = var_16682_cast_fp16)[name = tensor("op_16911_cast_fp16")]; + tensor var_16918_begin_0 = const()[name = tensor("op_16918_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16918_end_0 = const()[name = tensor("op_16918_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16918_end_mask_0 = const()[name = tensor("op_16918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16918_cast_fp16 = slice_by_index(begin = var_16918_begin_0, end = var_16918_end_0, end_mask = var_16918_end_mask_0, x = var_16682_cast_fp16)[name = tensor("op_16918_cast_fp16")]; + tensor var_16925_begin_0 = const()[name = tensor("op_16925_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16925_end_0 = const()[name = tensor("op_16925_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16925_end_mask_0 = const()[name = tensor("op_16925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16925_cast_fp16 = slice_by_index(begin = var_16925_begin_0, end = var_16925_end_0, end_mask = var_16925_end_mask_0, x = var_16682_cast_fp16)[name = tensor("op_16925_cast_fp16")]; + tensor var_16932_begin_0 = const()[name = tensor("op_16932_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16932_end_0 = const()[name = tensor("op_16932_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16932_end_mask_0 = const()[name = tensor("op_16932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16932_cast_fp16 = slice_by_index(begin = var_16932_begin_0, end = var_16932_end_0, end_mask = var_16932_end_mask_0, x = var_16682_cast_fp16)[name = tensor("op_16932_cast_fp16")]; + tensor var_16939_begin_0 = const()[name = tensor("op_16939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16939_end_0 = const()[name = tensor("op_16939_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16939_end_mask_0 = const()[name = tensor("op_16939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16939_cast_fp16 = slice_by_index(begin = var_16939_begin_0, end = var_16939_end_0, end_mask = var_16939_end_mask_0, x = var_16686_cast_fp16)[name = tensor("op_16939_cast_fp16")]; + tensor var_16946_begin_0 = const()[name = tensor("op_16946_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16946_end_0 = const()[name = tensor("op_16946_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16946_end_mask_0 = const()[name = tensor("op_16946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16946_cast_fp16 = slice_by_index(begin = var_16946_begin_0, end = var_16946_end_0, end_mask = var_16946_end_mask_0, x = var_16686_cast_fp16)[name = tensor("op_16946_cast_fp16")]; + tensor var_16953_begin_0 = const()[name = tensor("op_16953_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16953_end_0 = const()[name = tensor("op_16953_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16953_end_mask_0 = const()[name = tensor("op_16953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16953_cast_fp16 = slice_by_index(begin = var_16953_begin_0, end = var_16953_end_0, end_mask = var_16953_end_mask_0, x = var_16686_cast_fp16)[name = tensor("op_16953_cast_fp16")]; + tensor var_16960_begin_0 = const()[name = tensor("op_16960_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16960_end_0 = const()[name = tensor("op_16960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16960_end_mask_0 = const()[name = tensor("op_16960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16960_cast_fp16 = slice_by_index(begin = var_16960_begin_0, end = var_16960_end_0, end_mask = var_16960_end_mask_0, x = var_16686_cast_fp16)[name = tensor("op_16960_cast_fp16")]; + tensor var_16967_begin_0 = const()[name = tensor("op_16967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16967_end_0 = const()[name = tensor("op_16967_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16967_end_mask_0 = const()[name = tensor("op_16967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16967_cast_fp16 = slice_by_index(begin = var_16967_begin_0, end = var_16967_end_0, end_mask = var_16967_end_mask_0, x = var_16690_cast_fp16)[name = tensor("op_16967_cast_fp16")]; + tensor var_16974_begin_0 = const()[name = tensor("op_16974_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_16974_end_0 = const()[name = tensor("op_16974_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_16974_end_mask_0 = const()[name = tensor("op_16974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16974_cast_fp16 = slice_by_index(begin = var_16974_begin_0, end = var_16974_end_0, end_mask = var_16974_end_mask_0, x = var_16690_cast_fp16)[name = tensor("op_16974_cast_fp16")]; + tensor var_16981_begin_0 = const()[name = tensor("op_16981_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_16981_end_0 = const()[name = tensor("op_16981_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_16981_end_mask_0 = const()[name = tensor("op_16981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16981_cast_fp16 = slice_by_index(begin = var_16981_begin_0, end = var_16981_end_0, end_mask = var_16981_end_mask_0, x = var_16690_cast_fp16)[name = tensor("op_16981_cast_fp16")]; + tensor var_16988_begin_0 = const()[name = tensor("op_16988_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_16988_end_0 = const()[name = tensor("op_16988_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_16988_end_mask_0 = const()[name = tensor("op_16988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16988_cast_fp16 = slice_by_index(begin = var_16988_begin_0, end = var_16988_end_0, end_mask = var_16988_end_mask_0, x = var_16690_cast_fp16)[name = tensor("op_16988_cast_fp16")]; + tensor var_16995_begin_0 = const()[name = tensor("op_16995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_16995_end_0 = const()[name = tensor("op_16995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_16995_end_mask_0 = const()[name = tensor("op_16995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_16995_cast_fp16 = slice_by_index(begin = var_16995_begin_0, end = var_16995_end_0, end_mask = var_16995_end_mask_0, x = var_16694_cast_fp16)[name = tensor("op_16995_cast_fp16")]; + tensor var_17002_begin_0 = const()[name = tensor("op_17002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17002_end_0 = const()[name = tensor("op_17002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17002_end_mask_0 = const()[name = tensor("op_17002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17002_cast_fp16 = slice_by_index(begin = var_17002_begin_0, end = var_17002_end_0, end_mask = var_17002_end_mask_0, x = var_16694_cast_fp16)[name = tensor("op_17002_cast_fp16")]; + tensor var_17009_begin_0 = const()[name = tensor("op_17009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17009_end_0 = const()[name = tensor("op_17009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17009_end_mask_0 = const()[name = tensor("op_17009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17009_cast_fp16 = slice_by_index(begin = var_17009_begin_0, end = var_17009_end_0, end_mask = var_17009_end_mask_0, x = var_16694_cast_fp16)[name = tensor("op_17009_cast_fp16")]; + tensor var_17016_begin_0 = const()[name = tensor("op_17016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17016_end_0 = const()[name = tensor("op_17016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17016_end_mask_0 = const()[name = tensor("op_17016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17016_cast_fp16 = slice_by_index(begin = var_17016_begin_0, end = var_17016_end_0, end_mask = var_17016_end_mask_0, x = var_16694_cast_fp16)[name = tensor("op_17016_cast_fp16")]; + tensor var_17023_begin_0 = const()[name = tensor("op_17023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17023_end_0 = const()[name = tensor("op_17023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17023_end_mask_0 = const()[name = tensor("op_17023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17023_cast_fp16 = slice_by_index(begin = var_17023_begin_0, end = var_17023_end_0, end_mask = var_17023_end_mask_0, x = var_16698_cast_fp16)[name = tensor("op_17023_cast_fp16")]; + tensor var_17030_begin_0 = const()[name = tensor("op_17030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17030_end_0 = const()[name = tensor("op_17030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17030_end_mask_0 = const()[name = tensor("op_17030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17030_cast_fp16 = slice_by_index(begin = var_17030_begin_0, end = var_17030_end_0, end_mask = var_17030_end_mask_0, x = var_16698_cast_fp16)[name = tensor("op_17030_cast_fp16")]; + tensor var_17037_begin_0 = const()[name = tensor("op_17037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17037_end_0 = const()[name = tensor("op_17037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17037_end_mask_0 = const()[name = tensor("op_17037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17037_cast_fp16 = slice_by_index(begin = var_17037_begin_0, end = var_17037_end_0, end_mask = var_17037_end_mask_0, x = var_16698_cast_fp16)[name = tensor("op_17037_cast_fp16")]; + tensor var_17044_begin_0 = const()[name = tensor("op_17044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17044_end_0 = const()[name = tensor("op_17044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17044_end_mask_0 = const()[name = tensor("op_17044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17044_cast_fp16 = slice_by_index(begin = var_17044_begin_0, end = var_17044_end_0, end_mask = var_17044_end_mask_0, x = var_16698_cast_fp16)[name = tensor("op_17044_cast_fp16")]; + tensor var_17051_begin_0 = const()[name = tensor("op_17051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17051_end_0 = const()[name = tensor("op_17051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17051_end_mask_0 = const()[name = tensor("op_17051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17051_cast_fp16 = slice_by_index(begin = var_17051_begin_0, end = var_17051_end_0, end_mask = var_17051_end_mask_0, x = var_16702_cast_fp16)[name = tensor("op_17051_cast_fp16")]; + tensor var_17058_begin_0 = const()[name = tensor("op_17058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17058_end_0 = const()[name = tensor("op_17058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17058_end_mask_0 = const()[name = tensor("op_17058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17058_cast_fp16 = slice_by_index(begin = var_17058_begin_0, end = var_17058_end_0, end_mask = var_17058_end_mask_0, x = var_16702_cast_fp16)[name = tensor("op_17058_cast_fp16")]; + tensor var_17065_begin_0 = const()[name = tensor("op_17065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17065_end_0 = const()[name = tensor("op_17065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17065_end_mask_0 = const()[name = tensor("op_17065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17065_cast_fp16 = slice_by_index(begin = var_17065_begin_0, end = var_17065_end_0, end_mask = var_17065_end_mask_0, x = var_16702_cast_fp16)[name = tensor("op_17065_cast_fp16")]; + tensor var_17072_begin_0 = const()[name = tensor("op_17072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17072_end_0 = const()[name = tensor("op_17072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17072_end_mask_0 = const()[name = tensor("op_17072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17072_cast_fp16 = slice_by_index(begin = var_17072_begin_0, end = var_17072_end_0, end_mask = var_17072_end_mask_0, x = var_16702_cast_fp16)[name = tensor("op_17072_cast_fp16")]; + tensor var_17079_begin_0 = const()[name = tensor("op_17079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17079_end_0 = const()[name = tensor("op_17079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17079_end_mask_0 = const()[name = tensor("op_17079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17079_cast_fp16 = slice_by_index(begin = var_17079_begin_0, end = var_17079_end_0, end_mask = var_17079_end_mask_0, x = var_16706_cast_fp16)[name = tensor("op_17079_cast_fp16")]; + tensor var_17086_begin_0 = const()[name = tensor("op_17086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17086_end_0 = const()[name = tensor("op_17086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17086_end_mask_0 = const()[name = tensor("op_17086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17086_cast_fp16 = slice_by_index(begin = var_17086_begin_0, end = var_17086_end_0, end_mask = var_17086_end_mask_0, x = var_16706_cast_fp16)[name = tensor("op_17086_cast_fp16")]; + tensor var_17093_begin_0 = const()[name = tensor("op_17093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17093_end_0 = const()[name = tensor("op_17093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17093_end_mask_0 = const()[name = tensor("op_17093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17093_cast_fp16 = slice_by_index(begin = var_17093_begin_0, end = var_17093_end_0, end_mask = var_17093_end_mask_0, x = var_16706_cast_fp16)[name = tensor("op_17093_cast_fp16")]; + tensor var_17100_begin_0 = const()[name = tensor("op_17100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17100_end_0 = const()[name = tensor("op_17100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17100_end_mask_0 = const()[name = tensor("op_17100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17100_cast_fp16 = slice_by_index(begin = var_17100_begin_0, end = var_17100_end_0, end_mask = var_17100_end_mask_0, x = var_16706_cast_fp16)[name = tensor("op_17100_cast_fp16")]; + tensor var_17107_begin_0 = const()[name = tensor("op_17107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17107_end_0 = const()[name = tensor("op_17107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17107_end_mask_0 = const()[name = tensor("op_17107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17107_cast_fp16 = slice_by_index(begin = var_17107_begin_0, end = var_17107_end_0, end_mask = var_17107_end_mask_0, x = var_16710_cast_fp16)[name = tensor("op_17107_cast_fp16")]; + tensor var_17114_begin_0 = const()[name = tensor("op_17114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17114_end_0 = const()[name = tensor("op_17114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17114_end_mask_0 = const()[name = tensor("op_17114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17114_cast_fp16 = slice_by_index(begin = var_17114_begin_0, end = var_17114_end_0, end_mask = var_17114_end_mask_0, x = var_16710_cast_fp16)[name = tensor("op_17114_cast_fp16")]; + tensor var_17121_begin_0 = const()[name = tensor("op_17121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17121_end_0 = const()[name = tensor("op_17121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17121_end_mask_0 = const()[name = tensor("op_17121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17121_cast_fp16 = slice_by_index(begin = var_17121_begin_0, end = var_17121_end_0, end_mask = var_17121_end_mask_0, x = var_16710_cast_fp16)[name = tensor("op_17121_cast_fp16")]; + tensor var_17128_begin_0 = const()[name = tensor("op_17128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17128_end_0 = const()[name = tensor("op_17128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17128_end_mask_0 = const()[name = tensor("op_17128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17128_cast_fp16 = slice_by_index(begin = var_17128_begin_0, end = var_17128_end_0, end_mask = var_17128_end_mask_0, x = var_16710_cast_fp16)[name = tensor("op_17128_cast_fp16")]; + tensor var_17135_begin_0 = const()[name = tensor("op_17135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17135_end_0 = const()[name = tensor("op_17135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17135_end_mask_0 = const()[name = tensor("op_17135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17135_cast_fp16 = slice_by_index(begin = var_17135_begin_0, end = var_17135_end_0, end_mask = var_17135_end_mask_0, x = var_16714_cast_fp16)[name = tensor("op_17135_cast_fp16")]; + tensor var_17142_begin_0 = const()[name = tensor("op_17142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17142_end_0 = const()[name = tensor("op_17142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17142_end_mask_0 = const()[name = tensor("op_17142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17142_cast_fp16 = slice_by_index(begin = var_17142_begin_0, end = var_17142_end_0, end_mask = var_17142_end_mask_0, x = var_16714_cast_fp16)[name = tensor("op_17142_cast_fp16")]; + tensor var_17149_begin_0 = const()[name = tensor("op_17149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17149_end_0 = const()[name = tensor("op_17149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17149_end_mask_0 = const()[name = tensor("op_17149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17149_cast_fp16 = slice_by_index(begin = var_17149_begin_0, end = var_17149_end_0, end_mask = var_17149_end_mask_0, x = var_16714_cast_fp16)[name = tensor("op_17149_cast_fp16")]; + tensor var_17156_begin_0 = const()[name = tensor("op_17156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17156_end_0 = const()[name = tensor("op_17156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17156_end_mask_0 = const()[name = tensor("op_17156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17156_cast_fp16 = slice_by_index(begin = var_17156_begin_0, end = var_17156_end_0, end_mask = var_17156_end_mask_0, x = var_16714_cast_fp16)[name = tensor("op_17156_cast_fp16")]; + tensor var_17163_begin_0 = const()[name = tensor("op_17163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17163_end_0 = const()[name = tensor("op_17163_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17163_end_mask_0 = const()[name = tensor("op_17163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17163_cast_fp16 = slice_by_index(begin = var_17163_begin_0, end = var_17163_end_0, end_mask = var_17163_end_mask_0, x = var_16718_cast_fp16)[name = tensor("op_17163_cast_fp16")]; + tensor var_17170_begin_0 = const()[name = tensor("op_17170_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17170_end_0 = const()[name = tensor("op_17170_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17170_end_mask_0 = const()[name = tensor("op_17170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17170_cast_fp16 = slice_by_index(begin = var_17170_begin_0, end = var_17170_end_0, end_mask = var_17170_end_mask_0, x = var_16718_cast_fp16)[name = tensor("op_17170_cast_fp16")]; + tensor var_17177_begin_0 = const()[name = tensor("op_17177_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17177_end_0 = const()[name = tensor("op_17177_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17177_end_mask_0 = const()[name = tensor("op_17177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17177_cast_fp16 = slice_by_index(begin = var_17177_begin_0, end = var_17177_end_0, end_mask = var_17177_end_mask_0, x = var_16718_cast_fp16)[name = tensor("op_17177_cast_fp16")]; + tensor var_17184_begin_0 = const()[name = tensor("op_17184_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17184_end_0 = const()[name = tensor("op_17184_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17184_end_mask_0 = const()[name = tensor("op_17184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17184_cast_fp16 = slice_by_index(begin = var_17184_begin_0, end = var_17184_end_0, end_mask = var_17184_end_mask_0, x = var_16718_cast_fp16)[name = tensor("op_17184_cast_fp16")]; + tensor var_17191_begin_0 = const()[name = tensor("op_17191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17191_end_0 = const()[name = tensor("op_17191_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17191_end_mask_0 = const()[name = tensor("op_17191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17191_cast_fp16 = slice_by_index(begin = var_17191_begin_0, end = var_17191_end_0, end_mask = var_17191_end_mask_0, x = var_16722_cast_fp16)[name = tensor("op_17191_cast_fp16")]; + tensor var_17198_begin_0 = const()[name = tensor("op_17198_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17198_end_0 = const()[name = tensor("op_17198_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17198_end_mask_0 = const()[name = tensor("op_17198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17198_cast_fp16 = slice_by_index(begin = var_17198_begin_0, end = var_17198_end_0, end_mask = var_17198_end_mask_0, x = var_16722_cast_fp16)[name = tensor("op_17198_cast_fp16")]; + tensor var_17205_begin_0 = const()[name = tensor("op_17205_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17205_end_0 = const()[name = tensor("op_17205_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17205_end_mask_0 = const()[name = tensor("op_17205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17205_cast_fp16 = slice_by_index(begin = var_17205_begin_0, end = var_17205_end_0, end_mask = var_17205_end_mask_0, x = var_16722_cast_fp16)[name = tensor("op_17205_cast_fp16")]; + tensor var_17212_begin_0 = const()[name = tensor("op_17212_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17212_end_0 = const()[name = tensor("op_17212_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17212_end_mask_0 = const()[name = tensor("op_17212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17212_cast_fp16 = slice_by_index(begin = var_17212_begin_0, end = var_17212_end_0, end_mask = var_17212_end_mask_0, x = var_16722_cast_fp16)[name = tensor("op_17212_cast_fp16")]; + tensor var_17219_begin_0 = const()[name = tensor("op_17219_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17219_end_0 = const()[name = tensor("op_17219_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17219_end_mask_0 = const()[name = tensor("op_17219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17219_cast_fp16 = slice_by_index(begin = var_17219_begin_0, end = var_17219_end_0, end_mask = var_17219_end_mask_0, x = var_16726_cast_fp16)[name = tensor("op_17219_cast_fp16")]; + tensor var_17226_begin_0 = const()[name = tensor("op_17226_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17226_end_0 = const()[name = tensor("op_17226_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17226_end_mask_0 = const()[name = tensor("op_17226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17226_cast_fp16 = slice_by_index(begin = var_17226_begin_0, end = var_17226_end_0, end_mask = var_17226_end_mask_0, x = var_16726_cast_fp16)[name = tensor("op_17226_cast_fp16")]; + tensor var_17233_begin_0 = const()[name = tensor("op_17233_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17233_end_0 = const()[name = tensor("op_17233_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17233_end_mask_0 = const()[name = tensor("op_17233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17233_cast_fp16 = slice_by_index(begin = var_17233_begin_0, end = var_17233_end_0, end_mask = var_17233_end_mask_0, x = var_16726_cast_fp16)[name = tensor("op_17233_cast_fp16")]; + tensor var_17240_begin_0 = const()[name = tensor("op_17240_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17240_end_0 = const()[name = tensor("op_17240_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17240_end_mask_0 = const()[name = tensor("op_17240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17240_cast_fp16 = slice_by_index(begin = var_17240_begin_0, end = var_17240_end_0, end_mask = var_17240_end_mask_0, x = var_16726_cast_fp16)[name = tensor("op_17240_cast_fp16")]; + tensor var_17247_begin_0 = const()[name = tensor("op_17247_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17247_end_0 = const()[name = tensor("op_17247_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17247_end_mask_0 = const()[name = tensor("op_17247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17247_cast_fp16 = slice_by_index(begin = var_17247_begin_0, end = var_17247_end_0, end_mask = var_17247_end_mask_0, x = var_16730_cast_fp16)[name = tensor("op_17247_cast_fp16")]; + tensor var_17254_begin_0 = const()[name = tensor("op_17254_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17254_end_0 = const()[name = tensor("op_17254_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17254_end_mask_0 = const()[name = tensor("op_17254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17254_cast_fp16 = slice_by_index(begin = var_17254_begin_0, end = var_17254_end_0, end_mask = var_17254_end_mask_0, x = var_16730_cast_fp16)[name = tensor("op_17254_cast_fp16")]; + tensor var_17261_begin_0 = const()[name = tensor("op_17261_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17261_end_0 = const()[name = tensor("op_17261_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17261_end_mask_0 = const()[name = tensor("op_17261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17261_cast_fp16 = slice_by_index(begin = var_17261_begin_0, end = var_17261_end_0, end_mask = var_17261_end_mask_0, x = var_16730_cast_fp16)[name = tensor("op_17261_cast_fp16")]; + tensor var_17268_begin_0 = const()[name = tensor("op_17268_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17268_end_0 = const()[name = tensor("op_17268_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17268_end_mask_0 = const()[name = tensor("op_17268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17268_cast_fp16 = slice_by_index(begin = var_17268_begin_0, end = var_17268_end_0, end_mask = var_17268_end_mask_0, x = var_16730_cast_fp16)[name = tensor("op_17268_cast_fp16")]; + tensor var_17275_begin_0 = const()[name = tensor("op_17275_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17275_end_0 = const()[name = tensor("op_17275_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_17275_end_mask_0 = const()[name = tensor("op_17275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17275_cast_fp16 = slice_by_index(begin = var_17275_begin_0, end = var_17275_end_0, end_mask = var_17275_end_mask_0, x = var_16734_cast_fp16)[name = tensor("op_17275_cast_fp16")]; + tensor var_17282_begin_0 = const()[name = tensor("op_17282_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_17282_end_0 = const()[name = tensor("op_17282_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_17282_end_mask_0 = const()[name = tensor("op_17282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17282_cast_fp16 = slice_by_index(begin = var_17282_begin_0, end = var_17282_end_0, end_mask = var_17282_end_mask_0, x = var_16734_cast_fp16)[name = tensor("op_17282_cast_fp16")]; + tensor var_17289_begin_0 = const()[name = tensor("op_17289_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_17289_end_0 = const()[name = tensor("op_17289_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_17289_end_mask_0 = const()[name = tensor("op_17289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17289_cast_fp16 = slice_by_index(begin = var_17289_begin_0, end = var_17289_end_0, end_mask = var_17289_end_mask_0, x = var_16734_cast_fp16)[name = tensor("op_17289_cast_fp16")]; + tensor var_17296_begin_0 = const()[name = tensor("op_17296_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_17296_end_0 = const()[name = tensor("op_17296_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17296_end_mask_0 = const()[name = tensor("op_17296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17296_cast_fp16 = slice_by_index(begin = var_17296_begin_0, end = var_17296_end_0, end_mask = var_17296_end_mask_0, x = var_16734_cast_fp16)[name = tensor("op_17296_cast_fp16")]; + tensor k_21_perm_0 = const()[name = tensor("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_17301_begin_0 = const()[name = tensor("op_17301_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17301_end_0 = const()[name = tensor("op_17301_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_17301_end_mask_0 = const()[name = tensor("op_17301_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_21 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = tensor("transpose_21")]; + tensor var_17301_cast_fp16 = slice_by_index(begin = var_17301_begin_0, end = var_17301_end_0, end_mask = var_17301_end_mask_0, x = transpose_21)[name = tensor("op_17301_cast_fp16")]; + tensor var_17305_begin_0 = const()[name = tensor("op_17305_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_17305_end_0 = const()[name = tensor("op_17305_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_17305_end_mask_0 = const()[name = tensor("op_17305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17305_cast_fp16 = slice_by_index(begin = var_17305_begin_0, end = var_17305_end_0, end_mask = var_17305_end_mask_0, x = transpose_21)[name = tensor("op_17305_cast_fp16")]; + tensor var_17309_begin_0 = const()[name = tensor("op_17309_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_17309_end_0 = const()[name = tensor("op_17309_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_17309_end_mask_0 = const()[name = tensor("op_17309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17309_cast_fp16 = slice_by_index(begin = var_17309_begin_0, end = var_17309_end_0, end_mask = var_17309_end_mask_0, x = transpose_21)[name = tensor("op_17309_cast_fp16")]; + tensor var_17313_begin_0 = const()[name = tensor("op_17313_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_17313_end_0 = const()[name = tensor("op_17313_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_17313_end_mask_0 = const()[name = tensor("op_17313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17313_cast_fp16 = slice_by_index(begin = var_17313_begin_0, end = var_17313_end_0, end_mask = var_17313_end_mask_0, x = transpose_21)[name = tensor("op_17313_cast_fp16")]; + tensor var_17317_begin_0 = const()[name = tensor("op_17317_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_17317_end_0 = const()[name = tensor("op_17317_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_17317_end_mask_0 = const()[name = tensor("op_17317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17317_cast_fp16 = slice_by_index(begin = var_17317_begin_0, end = var_17317_end_0, end_mask = var_17317_end_mask_0, x = transpose_21)[name = tensor("op_17317_cast_fp16")]; + tensor var_17321_begin_0 = const()[name = tensor("op_17321_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_17321_end_0 = const()[name = tensor("op_17321_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_17321_end_mask_0 = const()[name = tensor("op_17321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17321_cast_fp16 = slice_by_index(begin = var_17321_begin_0, end = var_17321_end_0, end_mask = var_17321_end_mask_0, x = transpose_21)[name = tensor("op_17321_cast_fp16")]; + tensor var_17325_begin_0 = const()[name = tensor("op_17325_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_17325_end_0 = const()[name = tensor("op_17325_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_17325_end_mask_0 = const()[name = tensor("op_17325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17325_cast_fp16 = slice_by_index(begin = var_17325_begin_0, end = var_17325_end_0, end_mask = var_17325_end_mask_0, x = transpose_21)[name = tensor("op_17325_cast_fp16")]; + tensor var_17329_begin_0 = const()[name = tensor("op_17329_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_17329_end_0 = const()[name = tensor("op_17329_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_17329_end_mask_0 = const()[name = tensor("op_17329_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17329_cast_fp16 = slice_by_index(begin = var_17329_begin_0, end = var_17329_end_0, end_mask = var_17329_end_mask_0, x = transpose_21)[name = tensor("op_17329_cast_fp16")]; + tensor var_17333_begin_0 = const()[name = tensor("op_17333_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_17333_end_0 = const()[name = tensor("op_17333_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_17333_end_mask_0 = const()[name = tensor("op_17333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17333_cast_fp16 = slice_by_index(begin = var_17333_begin_0, end = var_17333_end_0, end_mask = var_17333_end_mask_0, x = transpose_21)[name = tensor("op_17333_cast_fp16")]; + tensor var_17337_begin_0 = const()[name = tensor("op_17337_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_17337_end_0 = const()[name = tensor("op_17337_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_17337_end_mask_0 = const()[name = tensor("op_17337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17337_cast_fp16 = slice_by_index(begin = var_17337_begin_0, end = var_17337_end_0, end_mask = var_17337_end_mask_0, x = transpose_21)[name = tensor("op_17337_cast_fp16")]; + tensor var_17341_begin_0 = const()[name = tensor("op_17341_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_17341_end_0 = const()[name = tensor("op_17341_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_17341_end_mask_0 = const()[name = tensor("op_17341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17341_cast_fp16 = slice_by_index(begin = var_17341_begin_0, end = var_17341_end_0, end_mask = var_17341_end_mask_0, x = transpose_21)[name = tensor("op_17341_cast_fp16")]; + tensor var_17345_begin_0 = const()[name = tensor("op_17345_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_17345_end_0 = const()[name = tensor("op_17345_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_17345_end_mask_0 = const()[name = tensor("op_17345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17345_cast_fp16 = slice_by_index(begin = var_17345_begin_0, end = var_17345_end_0, end_mask = var_17345_end_mask_0, x = transpose_21)[name = tensor("op_17345_cast_fp16")]; + tensor var_17349_begin_0 = const()[name = tensor("op_17349_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_17349_end_0 = const()[name = tensor("op_17349_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_17349_end_mask_0 = const()[name = tensor("op_17349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17349_cast_fp16 = slice_by_index(begin = var_17349_begin_0, end = var_17349_end_0, end_mask = var_17349_end_mask_0, x = transpose_21)[name = tensor("op_17349_cast_fp16")]; + tensor var_17353_begin_0 = const()[name = tensor("op_17353_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_17353_end_0 = const()[name = tensor("op_17353_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_17353_end_mask_0 = const()[name = tensor("op_17353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17353_cast_fp16 = slice_by_index(begin = var_17353_begin_0, end = var_17353_end_0, end_mask = var_17353_end_mask_0, x = transpose_21)[name = tensor("op_17353_cast_fp16")]; + tensor var_17357_begin_0 = const()[name = tensor("op_17357_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_17357_end_0 = const()[name = tensor("op_17357_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_17357_end_mask_0 = const()[name = tensor("op_17357_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17357_cast_fp16 = slice_by_index(begin = var_17357_begin_0, end = var_17357_end_0, end_mask = var_17357_end_mask_0, x = transpose_21)[name = tensor("op_17357_cast_fp16")]; + tensor var_17361_begin_0 = const()[name = tensor("op_17361_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_17361_end_0 = const()[name = tensor("op_17361_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_17361_end_mask_0 = const()[name = tensor("op_17361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17361_cast_fp16 = slice_by_index(begin = var_17361_begin_0, end = var_17361_end_0, end_mask = var_17361_end_mask_0, x = transpose_21)[name = tensor("op_17361_cast_fp16")]; + tensor var_17365_begin_0 = const()[name = tensor("op_17365_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_17365_end_0 = const()[name = tensor("op_17365_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_17365_end_mask_0 = const()[name = tensor("op_17365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17365_cast_fp16 = slice_by_index(begin = var_17365_begin_0, end = var_17365_end_0, end_mask = var_17365_end_mask_0, x = transpose_21)[name = tensor("op_17365_cast_fp16")]; + tensor var_17369_begin_0 = const()[name = tensor("op_17369_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_17369_end_0 = const()[name = tensor("op_17369_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_17369_end_mask_0 = const()[name = tensor("op_17369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17369_cast_fp16 = slice_by_index(begin = var_17369_begin_0, end = var_17369_end_0, end_mask = var_17369_end_mask_0, x = transpose_21)[name = tensor("op_17369_cast_fp16")]; + tensor var_17373_begin_0 = const()[name = tensor("op_17373_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_17373_end_0 = const()[name = tensor("op_17373_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_17373_end_mask_0 = const()[name = tensor("op_17373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17373_cast_fp16 = slice_by_index(begin = var_17373_begin_0, end = var_17373_end_0, end_mask = var_17373_end_mask_0, x = transpose_21)[name = tensor("op_17373_cast_fp16")]; + tensor var_17377_begin_0 = const()[name = tensor("op_17377_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_17377_end_0 = const()[name = tensor("op_17377_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_17377_end_mask_0 = const()[name = tensor("op_17377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_17377_cast_fp16 = slice_by_index(begin = var_17377_begin_0, end = var_17377_end_0, end_mask = var_17377_end_mask_0, x = transpose_21)[name = tensor("op_17377_cast_fp16")]; + tensor var_17379_begin_0 = const()[name = tensor("op_17379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_17379_end_0 = const()[name = tensor("op_17379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_17379_end_mask_0 = const()[name = tensor("op_17379_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17379_cast_fp16 = slice_by_index(begin = var_17379_begin_0, end = var_17379_end_0, end_mask = var_17379_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17379_cast_fp16")]; + tensor var_17383_begin_0 = const()[name = tensor("op_17383_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_17383_end_0 = const()[name = tensor("op_17383_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_17383_end_mask_0 = const()[name = tensor("op_17383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17383_cast_fp16 = slice_by_index(begin = var_17383_begin_0, end = var_17383_end_0, end_mask = var_17383_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17383_cast_fp16")]; + tensor var_17387_begin_0 = const()[name = tensor("op_17387_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_17387_end_0 = const()[name = tensor("op_17387_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_17387_end_mask_0 = const()[name = tensor("op_17387_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17387_cast_fp16 = slice_by_index(begin = var_17387_begin_0, end = var_17387_end_0, end_mask = var_17387_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17387_cast_fp16")]; + tensor var_17391_begin_0 = const()[name = tensor("op_17391_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_17391_end_0 = const()[name = tensor("op_17391_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_17391_end_mask_0 = const()[name = tensor("op_17391_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17391_cast_fp16 = slice_by_index(begin = var_17391_begin_0, end = var_17391_end_0, end_mask = var_17391_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17391_cast_fp16")]; + tensor var_17395_begin_0 = const()[name = tensor("op_17395_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_17395_end_0 = const()[name = tensor("op_17395_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_17395_end_mask_0 = const()[name = tensor("op_17395_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17395_cast_fp16 = slice_by_index(begin = var_17395_begin_0, end = var_17395_end_0, end_mask = var_17395_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17395_cast_fp16")]; + tensor var_17399_begin_0 = const()[name = tensor("op_17399_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_17399_end_0 = const()[name = tensor("op_17399_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_17399_end_mask_0 = const()[name = tensor("op_17399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17399_cast_fp16 = slice_by_index(begin = var_17399_begin_0, end = var_17399_end_0, end_mask = var_17399_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17399_cast_fp16")]; + tensor var_17403_begin_0 = const()[name = tensor("op_17403_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_17403_end_0 = const()[name = tensor("op_17403_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_17403_end_mask_0 = const()[name = tensor("op_17403_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17403_cast_fp16 = slice_by_index(begin = var_17403_begin_0, end = var_17403_end_0, end_mask = var_17403_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17403_cast_fp16")]; + tensor var_17407_begin_0 = const()[name = tensor("op_17407_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_17407_end_0 = const()[name = tensor("op_17407_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_17407_end_mask_0 = const()[name = tensor("op_17407_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17407_cast_fp16 = slice_by_index(begin = var_17407_begin_0, end = var_17407_end_0, end_mask = var_17407_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17407_cast_fp16")]; + tensor var_17411_begin_0 = const()[name = tensor("op_17411_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_17411_end_0 = const()[name = tensor("op_17411_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_17411_end_mask_0 = const()[name = tensor("op_17411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17411_cast_fp16 = slice_by_index(begin = var_17411_begin_0, end = var_17411_end_0, end_mask = var_17411_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17411_cast_fp16")]; + tensor var_17415_begin_0 = const()[name = tensor("op_17415_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_17415_end_0 = const()[name = tensor("op_17415_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_17415_end_mask_0 = const()[name = tensor("op_17415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17415_cast_fp16 = slice_by_index(begin = var_17415_begin_0, end = var_17415_end_0, end_mask = var_17415_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17415_cast_fp16")]; + tensor var_17419_begin_0 = const()[name = tensor("op_17419_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_17419_end_0 = const()[name = tensor("op_17419_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_17419_end_mask_0 = const()[name = tensor("op_17419_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17419_cast_fp16 = slice_by_index(begin = var_17419_begin_0, end = var_17419_end_0, end_mask = var_17419_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17419_cast_fp16")]; + tensor var_17423_begin_0 = const()[name = tensor("op_17423_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_17423_end_0 = const()[name = tensor("op_17423_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_17423_end_mask_0 = const()[name = tensor("op_17423_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17423_cast_fp16 = slice_by_index(begin = var_17423_begin_0, end = var_17423_end_0, end_mask = var_17423_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17423_cast_fp16")]; + tensor var_17427_begin_0 = const()[name = tensor("op_17427_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_17427_end_0 = const()[name = tensor("op_17427_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_17427_end_mask_0 = const()[name = tensor("op_17427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17427_cast_fp16 = slice_by_index(begin = var_17427_begin_0, end = var_17427_end_0, end_mask = var_17427_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17427_cast_fp16")]; + tensor var_17431_begin_0 = const()[name = tensor("op_17431_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_17431_end_0 = const()[name = tensor("op_17431_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_17431_end_mask_0 = const()[name = tensor("op_17431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17431_cast_fp16 = slice_by_index(begin = var_17431_begin_0, end = var_17431_end_0, end_mask = var_17431_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17431_cast_fp16")]; + tensor var_17435_begin_0 = const()[name = tensor("op_17435_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_17435_end_0 = const()[name = tensor("op_17435_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_17435_end_mask_0 = const()[name = tensor("op_17435_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17435_cast_fp16 = slice_by_index(begin = var_17435_begin_0, end = var_17435_end_0, end_mask = var_17435_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17435_cast_fp16")]; + tensor var_17439_begin_0 = const()[name = tensor("op_17439_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_17439_end_0 = const()[name = tensor("op_17439_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_17439_end_mask_0 = const()[name = tensor("op_17439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17439_cast_fp16 = slice_by_index(begin = var_17439_begin_0, end = var_17439_end_0, end_mask = var_17439_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17439_cast_fp16")]; + tensor var_17443_begin_0 = const()[name = tensor("op_17443_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_17443_end_0 = const()[name = tensor("op_17443_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_17443_end_mask_0 = const()[name = tensor("op_17443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17443_cast_fp16 = slice_by_index(begin = var_17443_begin_0, end = var_17443_end_0, end_mask = var_17443_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17443_cast_fp16")]; + tensor var_17447_begin_0 = const()[name = tensor("op_17447_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_17447_end_0 = const()[name = tensor("op_17447_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_17447_end_mask_0 = const()[name = tensor("op_17447_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17447_cast_fp16 = slice_by_index(begin = var_17447_begin_0, end = var_17447_end_0, end_mask = var_17447_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17447_cast_fp16")]; + tensor var_17451_begin_0 = const()[name = tensor("op_17451_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_17451_end_0 = const()[name = tensor("op_17451_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_17451_end_mask_0 = const()[name = tensor("op_17451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17451_cast_fp16 = slice_by_index(begin = var_17451_begin_0, end = var_17451_end_0, end_mask = var_17451_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17451_cast_fp16")]; + tensor var_17455_begin_0 = const()[name = tensor("op_17455_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_17455_end_0 = const()[name = tensor("op_17455_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_17455_end_mask_0 = const()[name = tensor("op_17455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_17455_cast_fp16 = slice_by_index(begin = var_17455_begin_0, end = var_17455_end_0, end_mask = var_17455_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_17455_cast_fp16")]; + tensor var_17459_equation_0 = const()[name = tensor("op_17459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17459_cast_fp16 = einsum(equation = var_17459_equation_0, values = (var_17301_cast_fp16, var_16743_cast_fp16))[name = tensor("op_17459_cast_fp16")]; + tensor var_17460_to_fp16 = const()[name = tensor("op_17460_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1601_cast_fp16 = mul(x = var_17459_cast_fp16, y = var_17460_to_fp16)[name = tensor("aw_chunk_1601_cast_fp16")]; + tensor var_17463_equation_0 = const()[name = tensor("op_17463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17463_cast_fp16 = einsum(equation = var_17463_equation_0, values = (var_17301_cast_fp16, var_16750_cast_fp16))[name = tensor("op_17463_cast_fp16")]; + tensor var_17464_to_fp16 = const()[name = tensor("op_17464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1603_cast_fp16 = mul(x = var_17463_cast_fp16, y = var_17464_to_fp16)[name = tensor("aw_chunk_1603_cast_fp16")]; + tensor var_17467_equation_0 = const()[name = tensor("op_17467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17467_cast_fp16 = einsum(equation = var_17467_equation_0, values = (var_17301_cast_fp16, var_16757_cast_fp16))[name = tensor("op_17467_cast_fp16")]; + tensor var_17468_to_fp16 = const()[name = tensor("op_17468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1605_cast_fp16 = mul(x = var_17467_cast_fp16, y = var_17468_to_fp16)[name = tensor("aw_chunk_1605_cast_fp16")]; + tensor var_17471_equation_0 = const()[name = tensor("op_17471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17471_cast_fp16 = einsum(equation = var_17471_equation_0, values = (var_17301_cast_fp16, var_16764_cast_fp16))[name = tensor("op_17471_cast_fp16")]; + tensor var_17472_to_fp16 = const()[name = tensor("op_17472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1607_cast_fp16 = mul(x = var_17471_cast_fp16, y = var_17472_to_fp16)[name = tensor("aw_chunk_1607_cast_fp16")]; + tensor var_17475_equation_0 = const()[name = tensor("op_17475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17475_cast_fp16 = einsum(equation = var_17475_equation_0, values = (var_17305_cast_fp16, var_16771_cast_fp16))[name = tensor("op_17475_cast_fp16")]; + tensor var_17476_to_fp16 = const()[name = tensor("op_17476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1609_cast_fp16 = mul(x = var_17475_cast_fp16, y = var_17476_to_fp16)[name = tensor("aw_chunk_1609_cast_fp16")]; + tensor var_17479_equation_0 = const()[name = tensor("op_17479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17479_cast_fp16 = einsum(equation = var_17479_equation_0, values = (var_17305_cast_fp16, var_16778_cast_fp16))[name = tensor("op_17479_cast_fp16")]; + tensor var_17480_to_fp16 = const()[name = tensor("op_17480_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1611_cast_fp16 = mul(x = var_17479_cast_fp16, y = var_17480_to_fp16)[name = tensor("aw_chunk_1611_cast_fp16")]; + tensor var_17483_equation_0 = const()[name = tensor("op_17483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17483_cast_fp16 = einsum(equation = var_17483_equation_0, values = (var_17305_cast_fp16, var_16785_cast_fp16))[name = tensor("op_17483_cast_fp16")]; + tensor var_17484_to_fp16 = const()[name = tensor("op_17484_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1613_cast_fp16 = mul(x = var_17483_cast_fp16, y = var_17484_to_fp16)[name = tensor("aw_chunk_1613_cast_fp16")]; + tensor var_17487_equation_0 = const()[name = tensor("op_17487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17487_cast_fp16 = einsum(equation = var_17487_equation_0, values = (var_17305_cast_fp16, var_16792_cast_fp16))[name = tensor("op_17487_cast_fp16")]; + tensor var_17488_to_fp16 = const()[name = tensor("op_17488_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1615_cast_fp16 = mul(x = var_17487_cast_fp16, y = var_17488_to_fp16)[name = tensor("aw_chunk_1615_cast_fp16")]; + tensor var_17491_equation_0 = const()[name = tensor("op_17491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17491_cast_fp16 = einsum(equation = var_17491_equation_0, values = (var_17309_cast_fp16, var_16799_cast_fp16))[name = tensor("op_17491_cast_fp16")]; + tensor var_17492_to_fp16 = const()[name = tensor("op_17492_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1617_cast_fp16 = mul(x = var_17491_cast_fp16, y = var_17492_to_fp16)[name = tensor("aw_chunk_1617_cast_fp16")]; + tensor var_17495_equation_0 = const()[name = tensor("op_17495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17495_cast_fp16 = einsum(equation = var_17495_equation_0, values = (var_17309_cast_fp16, var_16806_cast_fp16))[name = tensor("op_17495_cast_fp16")]; + tensor var_17496_to_fp16 = const()[name = tensor("op_17496_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1619_cast_fp16 = mul(x = var_17495_cast_fp16, y = var_17496_to_fp16)[name = tensor("aw_chunk_1619_cast_fp16")]; + tensor var_17499_equation_0 = const()[name = tensor("op_17499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17499_cast_fp16 = einsum(equation = var_17499_equation_0, values = (var_17309_cast_fp16, var_16813_cast_fp16))[name = tensor("op_17499_cast_fp16")]; + tensor var_17500_to_fp16 = const()[name = tensor("op_17500_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1621_cast_fp16 = mul(x = var_17499_cast_fp16, y = var_17500_to_fp16)[name = tensor("aw_chunk_1621_cast_fp16")]; + tensor var_17503_equation_0 = const()[name = tensor("op_17503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17503_cast_fp16 = einsum(equation = var_17503_equation_0, values = (var_17309_cast_fp16, var_16820_cast_fp16))[name = tensor("op_17503_cast_fp16")]; + tensor var_17504_to_fp16 = const()[name = tensor("op_17504_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1623_cast_fp16 = mul(x = var_17503_cast_fp16, y = var_17504_to_fp16)[name = tensor("aw_chunk_1623_cast_fp16")]; + tensor var_17507_equation_0 = const()[name = tensor("op_17507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17507_cast_fp16 = einsum(equation = var_17507_equation_0, values = (var_17313_cast_fp16, var_16827_cast_fp16))[name = tensor("op_17507_cast_fp16")]; + tensor var_17508_to_fp16 = const()[name = tensor("op_17508_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1625_cast_fp16 = mul(x = var_17507_cast_fp16, y = var_17508_to_fp16)[name = tensor("aw_chunk_1625_cast_fp16")]; + tensor var_17511_equation_0 = const()[name = tensor("op_17511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17511_cast_fp16 = einsum(equation = var_17511_equation_0, values = (var_17313_cast_fp16, var_16834_cast_fp16))[name = tensor("op_17511_cast_fp16")]; + tensor var_17512_to_fp16 = const()[name = tensor("op_17512_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1627_cast_fp16 = mul(x = var_17511_cast_fp16, y = var_17512_to_fp16)[name = tensor("aw_chunk_1627_cast_fp16")]; + tensor var_17515_equation_0 = const()[name = tensor("op_17515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17515_cast_fp16 = einsum(equation = var_17515_equation_0, values = (var_17313_cast_fp16, var_16841_cast_fp16))[name = tensor("op_17515_cast_fp16")]; + tensor var_17516_to_fp16 = const()[name = tensor("op_17516_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1629_cast_fp16 = mul(x = var_17515_cast_fp16, y = var_17516_to_fp16)[name = tensor("aw_chunk_1629_cast_fp16")]; + tensor var_17519_equation_0 = const()[name = tensor("op_17519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17519_cast_fp16 = einsum(equation = var_17519_equation_0, values = (var_17313_cast_fp16, var_16848_cast_fp16))[name = tensor("op_17519_cast_fp16")]; + tensor var_17520_to_fp16 = const()[name = tensor("op_17520_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1631_cast_fp16 = mul(x = var_17519_cast_fp16, y = var_17520_to_fp16)[name = tensor("aw_chunk_1631_cast_fp16")]; + tensor var_17523_equation_0 = const()[name = tensor("op_17523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17523_cast_fp16 = einsum(equation = var_17523_equation_0, values = (var_17317_cast_fp16, var_16855_cast_fp16))[name = tensor("op_17523_cast_fp16")]; + tensor var_17524_to_fp16 = const()[name = tensor("op_17524_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1633_cast_fp16 = mul(x = var_17523_cast_fp16, y = var_17524_to_fp16)[name = tensor("aw_chunk_1633_cast_fp16")]; + tensor var_17527_equation_0 = const()[name = tensor("op_17527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17527_cast_fp16 = einsum(equation = var_17527_equation_0, values = (var_17317_cast_fp16, var_16862_cast_fp16))[name = tensor("op_17527_cast_fp16")]; + tensor var_17528_to_fp16 = const()[name = tensor("op_17528_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1635_cast_fp16 = mul(x = var_17527_cast_fp16, y = var_17528_to_fp16)[name = tensor("aw_chunk_1635_cast_fp16")]; + tensor var_17531_equation_0 = const()[name = tensor("op_17531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17531_cast_fp16 = einsum(equation = var_17531_equation_0, values = (var_17317_cast_fp16, var_16869_cast_fp16))[name = tensor("op_17531_cast_fp16")]; + tensor var_17532_to_fp16 = const()[name = tensor("op_17532_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1637_cast_fp16 = mul(x = var_17531_cast_fp16, y = var_17532_to_fp16)[name = tensor("aw_chunk_1637_cast_fp16")]; + tensor var_17535_equation_0 = const()[name = tensor("op_17535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17535_cast_fp16 = einsum(equation = var_17535_equation_0, values = (var_17317_cast_fp16, var_16876_cast_fp16))[name = tensor("op_17535_cast_fp16")]; + tensor var_17536_to_fp16 = const()[name = tensor("op_17536_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1639_cast_fp16 = mul(x = var_17535_cast_fp16, y = var_17536_to_fp16)[name = tensor("aw_chunk_1639_cast_fp16")]; + tensor var_17539_equation_0 = const()[name = tensor("op_17539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17539_cast_fp16 = einsum(equation = var_17539_equation_0, values = (var_17321_cast_fp16, var_16883_cast_fp16))[name = tensor("op_17539_cast_fp16")]; + tensor var_17540_to_fp16 = const()[name = tensor("op_17540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1641_cast_fp16 = mul(x = var_17539_cast_fp16, y = var_17540_to_fp16)[name = tensor("aw_chunk_1641_cast_fp16")]; + tensor var_17543_equation_0 = const()[name = tensor("op_17543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17543_cast_fp16 = einsum(equation = var_17543_equation_0, values = (var_17321_cast_fp16, var_16890_cast_fp16))[name = tensor("op_17543_cast_fp16")]; + tensor var_17544_to_fp16 = const()[name = tensor("op_17544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1643_cast_fp16 = mul(x = var_17543_cast_fp16, y = var_17544_to_fp16)[name = tensor("aw_chunk_1643_cast_fp16")]; + tensor var_17547_equation_0 = const()[name = tensor("op_17547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17547_cast_fp16 = einsum(equation = var_17547_equation_0, values = (var_17321_cast_fp16, var_16897_cast_fp16))[name = tensor("op_17547_cast_fp16")]; + tensor var_17548_to_fp16 = const()[name = tensor("op_17548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1645_cast_fp16 = mul(x = var_17547_cast_fp16, y = var_17548_to_fp16)[name = tensor("aw_chunk_1645_cast_fp16")]; + tensor var_17551_equation_0 = const()[name = tensor("op_17551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17551_cast_fp16 = einsum(equation = var_17551_equation_0, values = (var_17321_cast_fp16, var_16904_cast_fp16))[name = tensor("op_17551_cast_fp16")]; + tensor var_17552_to_fp16 = const()[name = tensor("op_17552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1647_cast_fp16 = mul(x = var_17551_cast_fp16, y = var_17552_to_fp16)[name = tensor("aw_chunk_1647_cast_fp16")]; + tensor var_17555_equation_0 = const()[name = tensor("op_17555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17555_cast_fp16 = einsum(equation = var_17555_equation_0, values = (var_17325_cast_fp16, var_16911_cast_fp16))[name = tensor("op_17555_cast_fp16")]; + tensor var_17556_to_fp16 = const()[name = tensor("op_17556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1649_cast_fp16 = mul(x = var_17555_cast_fp16, y = var_17556_to_fp16)[name = tensor("aw_chunk_1649_cast_fp16")]; + tensor var_17559_equation_0 = const()[name = tensor("op_17559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17559_cast_fp16 = einsum(equation = var_17559_equation_0, values = (var_17325_cast_fp16, var_16918_cast_fp16))[name = tensor("op_17559_cast_fp16")]; + tensor var_17560_to_fp16 = const()[name = tensor("op_17560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1651_cast_fp16 = mul(x = var_17559_cast_fp16, y = var_17560_to_fp16)[name = tensor("aw_chunk_1651_cast_fp16")]; + tensor var_17563_equation_0 = const()[name = tensor("op_17563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17563_cast_fp16 = einsum(equation = var_17563_equation_0, values = (var_17325_cast_fp16, var_16925_cast_fp16))[name = tensor("op_17563_cast_fp16")]; + tensor var_17564_to_fp16 = const()[name = tensor("op_17564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1653_cast_fp16 = mul(x = var_17563_cast_fp16, y = var_17564_to_fp16)[name = tensor("aw_chunk_1653_cast_fp16")]; + tensor var_17567_equation_0 = const()[name = tensor("op_17567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17567_cast_fp16 = einsum(equation = var_17567_equation_0, values = (var_17325_cast_fp16, var_16932_cast_fp16))[name = tensor("op_17567_cast_fp16")]; + tensor var_17568_to_fp16 = const()[name = tensor("op_17568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1655_cast_fp16 = mul(x = var_17567_cast_fp16, y = var_17568_to_fp16)[name = tensor("aw_chunk_1655_cast_fp16")]; + tensor var_17571_equation_0 = const()[name = tensor("op_17571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17571_cast_fp16 = einsum(equation = var_17571_equation_0, values = (var_17329_cast_fp16, var_16939_cast_fp16))[name = tensor("op_17571_cast_fp16")]; + tensor var_17572_to_fp16 = const()[name = tensor("op_17572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1657_cast_fp16 = mul(x = var_17571_cast_fp16, y = var_17572_to_fp16)[name = tensor("aw_chunk_1657_cast_fp16")]; + tensor var_17575_equation_0 = const()[name = tensor("op_17575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17575_cast_fp16 = einsum(equation = var_17575_equation_0, values = (var_17329_cast_fp16, var_16946_cast_fp16))[name = tensor("op_17575_cast_fp16")]; + tensor var_17576_to_fp16 = const()[name = tensor("op_17576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1659_cast_fp16 = mul(x = var_17575_cast_fp16, y = var_17576_to_fp16)[name = tensor("aw_chunk_1659_cast_fp16")]; + tensor var_17579_equation_0 = const()[name = tensor("op_17579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17579_cast_fp16 = einsum(equation = var_17579_equation_0, values = (var_17329_cast_fp16, var_16953_cast_fp16))[name = tensor("op_17579_cast_fp16")]; + tensor var_17580_to_fp16 = const()[name = tensor("op_17580_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1661_cast_fp16 = mul(x = var_17579_cast_fp16, y = var_17580_to_fp16)[name = tensor("aw_chunk_1661_cast_fp16")]; + tensor var_17583_equation_0 = const()[name = tensor("op_17583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17583_cast_fp16 = einsum(equation = var_17583_equation_0, values = (var_17329_cast_fp16, var_16960_cast_fp16))[name = tensor("op_17583_cast_fp16")]; + tensor var_17584_to_fp16 = const()[name = tensor("op_17584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1663_cast_fp16 = mul(x = var_17583_cast_fp16, y = var_17584_to_fp16)[name = tensor("aw_chunk_1663_cast_fp16")]; + tensor var_17587_equation_0 = const()[name = tensor("op_17587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17587_cast_fp16 = einsum(equation = var_17587_equation_0, values = (var_17333_cast_fp16, var_16967_cast_fp16))[name = tensor("op_17587_cast_fp16")]; + tensor var_17588_to_fp16 = const()[name = tensor("op_17588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1665_cast_fp16 = mul(x = var_17587_cast_fp16, y = var_17588_to_fp16)[name = tensor("aw_chunk_1665_cast_fp16")]; + tensor var_17591_equation_0 = const()[name = tensor("op_17591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17591_cast_fp16 = einsum(equation = var_17591_equation_0, values = (var_17333_cast_fp16, var_16974_cast_fp16))[name = tensor("op_17591_cast_fp16")]; + tensor var_17592_to_fp16 = const()[name = tensor("op_17592_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1667_cast_fp16 = mul(x = var_17591_cast_fp16, y = var_17592_to_fp16)[name = tensor("aw_chunk_1667_cast_fp16")]; + tensor var_17595_equation_0 = const()[name = tensor("op_17595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17595_cast_fp16 = einsum(equation = var_17595_equation_0, values = (var_17333_cast_fp16, var_16981_cast_fp16))[name = tensor("op_17595_cast_fp16")]; + tensor var_17596_to_fp16 = const()[name = tensor("op_17596_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1669_cast_fp16 = mul(x = var_17595_cast_fp16, y = var_17596_to_fp16)[name = tensor("aw_chunk_1669_cast_fp16")]; + tensor var_17599_equation_0 = const()[name = tensor("op_17599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17599_cast_fp16 = einsum(equation = var_17599_equation_0, values = (var_17333_cast_fp16, var_16988_cast_fp16))[name = tensor("op_17599_cast_fp16")]; + tensor var_17600_to_fp16 = const()[name = tensor("op_17600_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1671_cast_fp16 = mul(x = var_17599_cast_fp16, y = var_17600_to_fp16)[name = tensor("aw_chunk_1671_cast_fp16")]; + tensor var_17603_equation_0 = const()[name = tensor("op_17603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17603_cast_fp16 = einsum(equation = var_17603_equation_0, values = (var_17337_cast_fp16, var_16995_cast_fp16))[name = tensor("op_17603_cast_fp16")]; + tensor var_17604_to_fp16 = const()[name = tensor("op_17604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1673_cast_fp16 = mul(x = var_17603_cast_fp16, y = var_17604_to_fp16)[name = tensor("aw_chunk_1673_cast_fp16")]; + tensor var_17607_equation_0 = const()[name = tensor("op_17607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17607_cast_fp16 = einsum(equation = var_17607_equation_0, values = (var_17337_cast_fp16, var_17002_cast_fp16))[name = tensor("op_17607_cast_fp16")]; + tensor var_17608_to_fp16 = const()[name = tensor("op_17608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1675_cast_fp16 = mul(x = var_17607_cast_fp16, y = var_17608_to_fp16)[name = tensor("aw_chunk_1675_cast_fp16")]; + tensor var_17611_equation_0 = const()[name = tensor("op_17611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17611_cast_fp16 = einsum(equation = var_17611_equation_0, values = (var_17337_cast_fp16, var_17009_cast_fp16))[name = tensor("op_17611_cast_fp16")]; + tensor var_17612_to_fp16 = const()[name = tensor("op_17612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1677_cast_fp16 = mul(x = var_17611_cast_fp16, y = var_17612_to_fp16)[name = tensor("aw_chunk_1677_cast_fp16")]; + tensor var_17615_equation_0 = const()[name = tensor("op_17615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17615_cast_fp16 = einsum(equation = var_17615_equation_0, values = (var_17337_cast_fp16, var_17016_cast_fp16))[name = tensor("op_17615_cast_fp16")]; + tensor var_17616_to_fp16 = const()[name = tensor("op_17616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1679_cast_fp16 = mul(x = var_17615_cast_fp16, y = var_17616_to_fp16)[name = tensor("aw_chunk_1679_cast_fp16")]; + tensor var_17619_equation_0 = const()[name = tensor("op_17619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17619_cast_fp16 = einsum(equation = var_17619_equation_0, values = (var_17341_cast_fp16, var_17023_cast_fp16))[name = tensor("op_17619_cast_fp16")]; + tensor var_17620_to_fp16 = const()[name = tensor("op_17620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1681_cast_fp16 = mul(x = var_17619_cast_fp16, y = var_17620_to_fp16)[name = tensor("aw_chunk_1681_cast_fp16")]; + tensor var_17623_equation_0 = const()[name = tensor("op_17623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17623_cast_fp16 = einsum(equation = var_17623_equation_0, values = (var_17341_cast_fp16, var_17030_cast_fp16))[name = tensor("op_17623_cast_fp16")]; + tensor var_17624_to_fp16 = const()[name = tensor("op_17624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1683_cast_fp16 = mul(x = var_17623_cast_fp16, y = var_17624_to_fp16)[name = tensor("aw_chunk_1683_cast_fp16")]; + tensor var_17627_equation_0 = const()[name = tensor("op_17627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17627_cast_fp16 = einsum(equation = var_17627_equation_0, values = (var_17341_cast_fp16, var_17037_cast_fp16))[name = tensor("op_17627_cast_fp16")]; + tensor var_17628_to_fp16 = const()[name = tensor("op_17628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1685_cast_fp16 = mul(x = var_17627_cast_fp16, y = var_17628_to_fp16)[name = tensor("aw_chunk_1685_cast_fp16")]; + tensor var_17631_equation_0 = const()[name = tensor("op_17631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17631_cast_fp16 = einsum(equation = var_17631_equation_0, values = (var_17341_cast_fp16, var_17044_cast_fp16))[name = tensor("op_17631_cast_fp16")]; + tensor var_17632_to_fp16 = const()[name = tensor("op_17632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1687_cast_fp16 = mul(x = var_17631_cast_fp16, y = var_17632_to_fp16)[name = tensor("aw_chunk_1687_cast_fp16")]; + tensor var_17635_equation_0 = const()[name = tensor("op_17635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17635_cast_fp16 = einsum(equation = var_17635_equation_0, values = (var_17345_cast_fp16, var_17051_cast_fp16))[name = tensor("op_17635_cast_fp16")]; + tensor var_17636_to_fp16 = const()[name = tensor("op_17636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1689_cast_fp16 = mul(x = var_17635_cast_fp16, y = var_17636_to_fp16)[name = tensor("aw_chunk_1689_cast_fp16")]; + tensor var_17639_equation_0 = const()[name = tensor("op_17639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17639_cast_fp16 = einsum(equation = var_17639_equation_0, values = (var_17345_cast_fp16, var_17058_cast_fp16))[name = tensor("op_17639_cast_fp16")]; + tensor var_17640_to_fp16 = const()[name = tensor("op_17640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1691_cast_fp16 = mul(x = var_17639_cast_fp16, y = var_17640_to_fp16)[name = tensor("aw_chunk_1691_cast_fp16")]; + tensor var_17643_equation_0 = const()[name = tensor("op_17643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17643_cast_fp16 = einsum(equation = var_17643_equation_0, values = (var_17345_cast_fp16, var_17065_cast_fp16))[name = tensor("op_17643_cast_fp16")]; + tensor var_17644_to_fp16 = const()[name = tensor("op_17644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1693_cast_fp16 = mul(x = var_17643_cast_fp16, y = var_17644_to_fp16)[name = tensor("aw_chunk_1693_cast_fp16")]; + tensor var_17647_equation_0 = const()[name = tensor("op_17647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17647_cast_fp16 = einsum(equation = var_17647_equation_0, values = (var_17345_cast_fp16, var_17072_cast_fp16))[name = tensor("op_17647_cast_fp16")]; + tensor var_17648_to_fp16 = const()[name = tensor("op_17648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1695_cast_fp16 = mul(x = var_17647_cast_fp16, y = var_17648_to_fp16)[name = tensor("aw_chunk_1695_cast_fp16")]; + tensor var_17651_equation_0 = const()[name = tensor("op_17651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17651_cast_fp16 = einsum(equation = var_17651_equation_0, values = (var_17349_cast_fp16, var_17079_cast_fp16))[name = tensor("op_17651_cast_fp16")]; + tensor var_17652_to_fp16 = const()[name = tensor("op_17652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1697_cast_fp16 = mul(x = var_17651_cast_fp16, y = var_17652_to_fp16)[name = tensor("aw_chunk_1697_cast_fp16")]; + tensor var_17655_equation_0 = const()[name = tensor("op_17655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17655_cast_fp16 = einsum(equation = var_17655_equation_0, values = (var_17349_cast_fp16, var_17086_cast_fp16))[name = tensor("op_17655_cast_fp16")]; + tensor var_17656_to_fp16 = const()[name = tensor("op_17656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1699_cast_fp16 = mul(x = var_17655_cast_fp16, y = var_17656_to_fp16)[name = tensor("aw_chunk_1699_cast_fp16")]; + tensor var_17659_equation_0 = const()[name = tensor("op_17659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17659_cast_fp16 = einsum(equation = var_17659_equation_0, values = (var_17349_cast_fp16, var_17093_cast_fp16))[name = tensor("op_17659_cast_fp16")]; + tensor var_17660_to_fp16 = const()[name = tensor("op_17660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1701_cast_fp16 = mul(x = var_17659_cast_fp16, y = var_17660_to_fp16)[name = tensor("aw_chunk_1701_cast_fp16")]; + tensor var_17663_equation_0 = const()[name = tensor("op_17663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17663_cast_fp16 = einsum(equation = var_17663_equation_0, values = (var_17349_cast_fp16, var_17100_cast_fp16))[name = tensor("op_17663_cast_fp16")]; + tensor var_17664_to_fp16 = const()[name = tensor("op_17664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1703_cast_fp16 = mul(x = var_17663_cast_fp16, y = var_17664_to_fp16)[name = tensor("aw_chunk_1703_cast_fp16")]; + tensor var_17667_equation_0 = const()[name = tensor("op_17667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17667_cast_fp16 = einsum(equation = var_17667_equation_0, values = (var_17353_cast_fp16, var_17107_cast_fp16))[name = tensor("op_17667_cast_fp16")]; + tensor var_17668_to_fp16 = const()[name = tensor("op_17668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1705_cast_fp16 = mul(x = var_17667_cast_fp16, y = var_17668_to_fp16)[name = tensor("aw_chunk_1705_cast_fp16")]; + tensor var_17671_equation_0 = const()[name = tensor("op_17671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17671_cast_fp16 = einsum(equation = var_17671_equation_0, values = (var_17353_cast_fp16, var_17114_cast_fp16))[name = tensor("op_17671_cast_fp16")]; + tensor var_17672_to_fp16 = const()[name = tensor("op_17672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1707_cast_fp16 = mul(x = var_17671_cast_fp16, y = var_17672_to_fp16)[name = tensor("aw_chunk_1707_cast_fp16")]; + tensor var_17675_equation_0 = const()[name = tensor("op_17675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17675_cast_fp16 = einsum(equation = var_17675_equation_0, values = (var_17353_cast_fp16, var_17121_cast_fp16))[name = tensor("op_17675_cast_fp16")]; + tensor var_17676_to_fp16 = const()[name = tensor("op_17676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1709_cast_fp16 = mul(x = var_17675_cast_fp16, y = var_17676_to_fp16)[name = tensor("aw_chunk_1709_cast_fp16")]; + tensor var_17679_equation_0 = const()[name = tensor("op_17679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17679_cast_fp16 = einsum(equation = var_17679_equation_0, values = (var_17353_cast_fp16, var_17128_cast_fp16))[name = tensor("op_17679_cast_fp16")]; + tensor var_17680_to_fp16 = const()[name = tensor("op_17680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1711_cast_fp16 = mul(x = var_17679_cast_fp16, y = var_17680_to_fp16)[name = tensor("aw_chunk_1711_cast_fp16")]; + tensor var_17683_equation_0 = const()[name = tensor("op_17683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17683_cast_fp16 = einsum(equation = var_17683_equation_0, values = (var_17357_cast_fp16, var_17135_cast_fp16))[name = tensor("op_17683_cast_fp16")]; + tensor var_17684_to_fp16 = const()[name = tensor("op_17684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1713_cast_fp16 = mul(x = var_17683_cast_fp16, y = var_17684_to_fp16)[name = tensor("aw_chunk_1713_cast_fp16")]; + tensor var_17687_equation_0 = const()[name = tensor("op_17687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17687_cast_fp16 = einsum(equation = var_17687_equation_0, values = (var_17357_cast_fp16, var_17142_cast_fp16))[name = tensor("op_17687_cast_fp16")]; + tensor var_17688_to_fp16 = const()[name = tensor("op_17688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1715_cast_fp16 = mul(x = var_17687_cast_fp16, y = var_17688_to_fp16)[name = tensor("aw_chunk_1715_cast_fp16")]; + tensor var_17691_equation_0 = const()[name = tensor("op_17691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17691_cast_fp16 = einsum(equation = var_17691_equation_0, values = (var_17357_cast_fp16, var_17149_cast_fp16))[name = tensor("op_17691_cast_fp16")]; + tensor var_17692_to_fp16 = const()[name = tensor("op_17692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1717_cast_fp16 = mul(x = var_17691_cast_fp16, y = var_17692_to_fp16)[name = tensor("aw_chunk_1717_cast_fp16")]; + tensor var_17695_equation_0 = const()[name = tensor("op_17695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17695_cast_fp16 = einsum(equation = var_17695_equation_0, values = (var_17357_cast_fp16, var_17156_cast_fp16))[name = tensor("op_17695_cast_fp16")]; + tensor var_17696_to_fp16 = const()[name = tensor("op_17696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1719_cast_fp16 = mul(x = var_17695_cast_fp16, y = var_17696_to_fp16)[name = tensor("aw_chunk_1719_cast_fp16")]; + tensor var_17699_equation_0 = const()[name = tensor("op_17699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17699_cast_fp16 = einsum(equation = var_17699_equation_0, values = (var_17361_cast_fp16, var_17163_cast_fp16))[name = tensor("op_17699_cast_fp16")]; + tensor var_17700_to_fp16 = const()[name = tensor("op_17700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1721_cast_fp16 = mul(x = var_17699_cast_fp16, y = var_17700_to_fp16)[name = tensor("aw_chunk_1721_cast_fp16")]; + tensor var_17703_equation_0 = const()[name = tensor("op_17703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17703_cast_fp16 = einsum(equation = var_17703_equation_0, values = (var_17361_cast_fp16, var_17170_cast_fp16))[name = tensor("op_17703_cast_fp16")]; + tensor var_17704_to_fp16 = const()[name = tensor("op_17704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1723_cast_fp16 = mul(x = var_17703_cast_fp16, y = var_17704_to_fp16)[name = tensor("aw_chunk_1723_cast_fp16")]; + tensor var_17707_equation_0 = const()[name = tensor("op_17707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17707_cast_fp16 = einsum(equation = var_17707_equation_0, values = (var_17361_cast_fp16, var_17177_cast_fp16))[name = tensor("op_17707_cast_fp16")]; + tensor var_17708_to_fp16 = const()[name = tensor("op_17708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1725_cast_fp16 = mul(x = var_17707_cast_fp16, y = var_17708_to_fp16)[name = tensor("aw_chunk_1725_cast_fp16")]; + tensor var_17711_equation_0 = const()[name = tensor("op_17711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17711_cast_fp16 = einsum(equation = var_17711_equation_0, values = (var_17361_cast_fp16, var_17184_cast_fp16))[name = tensor("op_17711_cast_fp16")]; + tensor var_17712_to_fp16 = const()[name = tensor("op_17712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1727_cast_fp16 = mul(x = var_17711_cast_fp16, y = var_17712_to_fp16)[name = tensor("aw_chunk_1727_cast_fp16")]; + tensor var_17715_equation_0 = const()[name = tensor("op_17715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17715_cast_fp16 = einsum(equation = var_17715_equation_0, values = (var_17365_cast_fp16, var_17191_cast_fp16))[name = tensor("op_17715_cast_fp16")]; + tensor var_17716_to_fp16 = const()[name = tensor("op_17716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1729_cast_fp16 = mul(x = var_17715_cast_fp16, y = var_17716_to_fp16)[name = tensor("aw_chunk_1729_cast_fp16")]; + tensor var_17719_equation_0 = const()[name = tensor("op_17719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17719_cast_fp16 = einsum(equation = var_17719_equation_0, values = (var_17365_cast_fp16, var_17198_cast_fp16))[name = tensor("op_17719_cast_fp16")]; + tensor var_17720_to_fp16 = const()[name = tensor("op_17720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1731_cast_fp16 = mul(x = var_17719_cast_fp16, y = var_17720_to_fp16)[name = tensor("aw_chunk_1731_cast_fp16")]; + tensor var_17723_equation_0 = const()[name = tensor("op_17723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17723_cast_fp16 = einsum(equation = var_17723_equation_0, values = (var_17365_cast_fp16, var_17205_cast_fp16))[name = tensor("op_17723_cast_fp16")]; + tensor var_17724_to_fp16 = const()[name = tensor("op_17724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1733_cast_fp16 = mul(x = var_17723_cast_fp16, y = var_17724_to_fp16)[name = tensor("aw_chunk_1733_cast_fp16")]; + tensor var_17727_equation_0 = const()[name = tensor("op_17727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17727_cast_fp16 = einsum(equation = var_17727_equation_0, values = (var_17365_cast_fp16, var_17212_cast_fp16))[name = tensor("op_17727_cast_fp16")]; + tensor var_17728_to_fp16 = const()[name = tensor("op_17728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1735_cast_fp16 = mul(x = var_17727_cast_fp16, y = var_17728_to_fp16)[name = tensor("aw_chunk_1735_cast_fp16")]; + tensor var_17731_equation_0 = const()[name = tensor("op_17731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17731_cast_fp16 = einsum(equation = var_17731_equation_0, values = (var_17369_cast_fp16, var_17219_cast_fp16))[name = tensor("op_17731_cast_fp16")]; + tensor var_17732_to_fp16 = const()[name = tensor("op_17732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1737_cast_fp16 = mul(x = var_17731_cast_fp16, y = var_17732_to_fp16)[name = tensor("aw_chunk_1737_cast_fp16")]; + tensor var_17735_equation_0 = const()[name = tensor("op_17735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17735_cast_fp16 = einsum(equation = var_17735_equation_0, values = (var_17369_cast_fp16, var_17226_cast_fp16))[name = tensor("op_17735_cast_fp16")]; + tensor var_17736_to_fp16 = const()[name = tensor("op_17736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1739_cast_fp16 = mul(x = var_17735_cast_fp16, y = var_17736_to_fp16)[name = tensor("aw_chunk_1739_cast_fp16")]; + tensor var_17739_equation_0 = const()[name = tensor("op_17739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17739_cast_fp16 = einsum(equation = var_17739_equation_0, values = (var_17369_cast_fp16, var_17233_cast_fp16))[name = tensor("op_17739_cast_fp16")]; + tensor var_17740_to_fp16 = const()[name = tensor("op_17740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1741_cast_fp16 = mul(x = var_17739_cast_fp16, y = var_17740_to_fp16)[name = tensor("aw_chunk_1741_cast_fp16")]; + tensor var_17743_equation_0 = const()[name = tensor("op_17743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17743_cast_fp16 = einsum(equation = var_17743_equation_0, values = (var_17369_cast_fp16, var_17240_cast_fp16))[name = tensor("op_17743_cast_fp16")]; + tensor var_17744_to_fp16 = const()[name = tensor("op_17744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1743_cast_fp16 = mul(x = var_17743_cast_fp16, y = var_17744_to_fp16)[name = tensor("aw_chunk_1743_cast_fp16")]; + tensor var_17747_equation_0 = const()[name = tensor("op_17747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17747_cast_fp16 = einsum(equation = var_17747_equation_0, values = (var_17373_cast_fp16, var_17247_cast_fp16))[name = tensor("op_17747_cast_fp16")]; + tensor var_17748_to_fp16 = const()[name = tensor("op_17748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1745_cast_fp16 = mul(x = var_17747_cast_fp16, y = var_17748_to_fp16)[name = tensor("aw_chunk_1745_cast_fp16")]; + tensor var_17751_equation_0 = const()[name = tensor("op_17751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17751_cast_fp16 = einsum(equation = var_17751_equation_0, values = (var_17373_cast_fp16, var_17254_cast_fp16))[name = tensor("op_17751_cast_fp16")]; + tensor var_17752_to_fp16 = const()[name = tensor("op_17752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1747_cast_fp16 = mul(x = var_17751_cast_fp16, y = var_17752_to_fp16)[name = tensor("aw_chunk_1747_cast_fp16")]; + tensor var_17755_equation_0 = const()[name = tensor("op_17755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17755_cast_fp16 = einsum(equation = var_17755_equation_0, values = (var_17373_cast_fp16, var_17261_cast_fp16))[name = tensor("op_17755_cast_fp16")]; + tensor var_17756_to_fp16 = const()[name = tensor("op_17756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1749_cast_fp16 = mul(x = var_17755_cast_fp16, y = var_17756_to_fp16)[name = tensor("aw_chunk_1749_cast_fp16")]; + tensor var_17759_equation_0 = const()[name = tensor("op_17759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17759_cast_fp16 = einsum(equation = var_17759_equation_0, values = (var_17373_cast_fp16, var_17268_cast_fp16))[name = tensor("op_17759_cast_fp16")]; + tensor var_17760_to_fp16 = const()[name = tensor("op_17760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1751_cast_fp16 = mul(x = var_17759_cast_fp16, y = var_17760_to_fp16)[name = tensor("aw_chunk_1751_cast_fp16")]; + tensor var_17763_equation_0 = const()[name = tensor("op_17763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17763_cast_fp16 = einsum(equation = var_17763_equation_0, values = (var_17377_cast_fp16, var_17275_cast_fp16))[name = tensor("op_17763_cast_fp16")]; + tensor var_17764_to_fp16 = const()[name = tensor("op_17764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1753_cast_fp16 = mul(x = var_17763_cast_fp16, y = var_17764_to_fp16)[name = tensor("aw_chunk_1753_cast_fp16")]; + tensor var_17767_equation_0 = const()[name = tensor("op_17767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17767_cast_fp16 = einsum(equation = var_17767_equation_0, values = (var_17377_cast_fp16, var_17282_cast_fp16))[name = tensor("op_17767_cast_fp16")]; + tensor var_17768_to_fp16 = const()[name = tensor("op_17768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1755_cast_fp16 = mul(x = var_17767_cast_fp16, y = var_17768_to_fp16)[name = tensor("aw_chunk_1755_cast_fp16")]; + tensor var_17771_equation_0 = const()[name = tensor("op_17771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17771_cast_fp16 = einsum(equation = var_17771_equation_0, values = (var_17377_cast_fp16, var_17289_cast_fp16))[name = tensor("op_17771_cast_fp16")]; + tensor var_17772_to_fp16 = const()[name = tensor("op_17772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1757_cast_fp16 = mul(x = var_17771_cast_fp16, y = var_17772_to_fp16)[name = tensor("aw_chunk_1757_cast_fp16")]; + tensor var_17775_equation_0 = const()[name = tensor("op_17775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_17775_cast_fp16 = einsum(equation = var_17775_equation_0, values = (var_17377_cast_fp16, var_17296_cast_fp16))[name = tensor("op_17775_cast_fp16")]; + tensor var_17776_to_fp16 = const()[name = tensor("op_17776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1759_cast_fp16 = mul(x = var_17775_cast_fp16, y = var_17776_to_fp16)[name = tensor("aw_chunk_1759_cast_fp16")]; + tensor var_17778_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1601_cast_fp16)[name = tensor("op_17778_cast_fp16")]; + tensor var_17779_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1603_cast_fp16)[name = tensor("op_17779_cast_fp16")]; + tensor var_17780_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1605_cast_fp16)[name = tensor("op_17780_cast_fp16")]; + tensor var_17781_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1607_cast_fp16)[name = tensor("op_17781_cast_fp16")]; + tensor var_17782_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1609_cast_fp16)[name = tensor("op_17782_cast_fp16")]; + tensor var_17783_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1611_cast_fp16)[name = tensor("op_17783_cast_fp16")]; + tensor var_17784_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1613_cast_fp16)[name = tensor("op_17784_cast_fp16")]; + tensor var_17785_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1615_cast_fp16)[name = tensor("op_17785_cast_fp16")]; + tensor var_17786_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1617_cast_fp16)[name = tensor("op_17786_cast_fp16")]; + tensor var_17787_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1619_cast_fp16)[name = tensor("op_17787_cast_fp16")]; + tensor var_17788_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1621_cast_fp16)[name = tensor("op_17788_cast_fp16")]; + tensor var_17789_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1623_cast_fp16)[name = tensor("op_17789_cast_fp16")]; + tensor var_17790_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1625_cast_fp16)[name = tensor("op_17790_cast_fp16")]; + tensor var_17791_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1627_cast_fp16)[name = tensor("op_17791_cast_fp16")]; + tensor var_17792_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1629_cast_fp16)[name = tensor("op_17792_cast_fp16")]; + tensor var_17793_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1631_cast_fp16)[name = tensor("op_17793_cast_fp16")]; + tensor var_17794_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1633_cast_fp16)[name = tensor("op_17794_cast_fp16")]; + tensor var_17795_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1635_cast_fp16)[name = tensor("op_17795_cast_fp16")]; + tensor var_17796_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1637_cast_fp16)[name = tensor("op_17796_cast_fp16")]; + tensor var_17797_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1639_cast_fp16)[name = tensor("op_17797_cast_fp16")]; + tensor var_17798_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1641_cast_fp16)[name = tensor("op_17798_cast_fp16")]; + tensor var_17799_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1643_cast_fp16)[name = tensor("op_17799_cast_fp16")]; + tensor var_17800_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1645_cast_fp16)[name = tensor("op_17800_cast_fp16")]; + tensor var_17801_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1647_cast_fp16)[name = tensor("op_17801_cast_fp16")]; + tensor var_17802_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1649_cast_fp16)[name = tensor("op_17802_cast_fp16")]; + tensor var_17803_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1651_cast_fp16)[name = tensor("op_17803_cast_fp16")]; + tensor var_17804_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1653_cast_fp16)[name = tensor("op_17804_cast_fp16")]; + tensor var_17805_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1655_cast_fp16)[name = tensor("op_17805_cast_fp16")]; + tensor var_17806_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1657_cast_fp16)[name = tensor("op_17806_cast_fp16")]; + tensor var_17807_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1659_cast_fp16)[name = tensor("op_17807_cast_fp16")]; + tensor var_17808_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1661_cast_fp16)[name = tensor("op_17808_cast_fp16")]; + tensor var_17809_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1663_cast_fp16)[name = tensor("op_17809_cast_fp16")]; + tensor var_17810_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1665_cast_fp16)[name = tensor("op_17810_cast_fp16")]; + tensor var_17811_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1667_cast_fp16)[name = tensor("op_17811_cast_fp16")]; + tensor var_17812_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1669_cast_fp16)[name = tensor("op_17812_cast_fp16")]; + tensor var_17813_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1671_cast_fp16)[name = tensor("op_17813_cast_fp16")]; + tensor var_17814_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1673_cast_fp16)[name = tensor("op_17814_cast_fp16")]; + tensor var_17815_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1675_cast_fp16)[name = tensor("op_17815_cast_fp16")]; + tensor var_17816_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1677_cast_fp16)[name = tensor("op_17816_cast_fp16")]; + tensor var_17817_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1679_cast_fp16)[name = tensor("op_17817_cast_fp16")]; + tensor var_17818_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1681_cast_fp16)[name = tensor("op_17818_cast_fp16")]; + tensor var_17819_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1683_cast_fp16)[name = tensor("op_17819_cast_fp16")]; + tensor var_17820_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1685_cast_fp16)[name = tensor("op_17820_cast_fp16")]; + tensor var_17821_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1687_cast_fp16)[name = tensor("op_17821_cast_fp16")]; + tensor var_17822_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1689_cast_fp16)[name = tensor("op_17822_cast_fp16")]; + tensor var_17823_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1691_cast_fp16)[name = tensor("op_17823_cast_fp16")]; + tensor var_17824_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1693_cast_fp16)[name = tensor("op_17824_cast_fp16")]; + tensor var_17825_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1695_cast_fp16)[name = tensor("op_17825_cast_fp16")]; + tensor var_17826_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1697_cast_fp16)[name = tensor("op_17826_cast_fp16")]; + tensor var_17827_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1699_cast_fp16)[name = tensor("op_17827_cast_fp16")]; + tensor var_17828_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1701_cast_fp16)[name = tensor("op_17828_cast_fp16")]; + tensor var_17829_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1703_cast_fp16)[name = tensor("op_17829_cast_fp16")]; + tensor var_17830_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1705_cast_fp16)[name = tensor("op_17830_cast_fp16")]; + tensor var_17831_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1707_cast_fp16)[name = tensor("op_17831_cast_fp16")]; + tensor var_17832_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1709_cast_fp16)[name = tensor("op_17832_cast_fp16")]; + tensor var_17833_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1711_cast_fp16)[name = tensor("op_17833_cast_fp16")]; + tensor var_17834_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1713_cast_fp16)[name = tensor("op_17834_cast_fp16")]; + tensor var_17835_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1715_cast_fp16)[name = tensor("op_17835_cast_fp16")]; + tensor var_17836_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1717_cast_fp16)[name = tensor("op_17836_cast_fp16")]; + tensor var_17837_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1719_cast_fp16)[name = tensor("op_17837_cast_fp16")]; + tensor var_17838_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1721_cast_fp16)[name = tensor("op_17838_cast_fp16")]; + tensor var_17839_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1723_cast_fp16)[name = tensor("op_17839_cast_fp16")]; + tensor var_17840_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1725_cast_fp16)[name = tensor("op_17840_cast_fp16")]; + tensor var_17841_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1727_cast_fp16)[name = tensor("op_17841_cast_fp16")]; + tensor var_17842_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1729_cast_fp16)[name = tensor("op_17842_cast_fp16")]; + tensor var_17843_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1731_cast_fp16)[name = tensor("op_17843_cast_fp16")]; + tensor var_17844_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1733_cast_fp16)[name = tensor("op_17844_cast_fp16")]; + tensor var_17845_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1735_cast_fp16)[name = tensor("op_17845_cast_fp16")]; + tensor var_17846_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1737_cast_fp16)[name = tensor("op_17846_cast_fp16")]; + tensor var_17847_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1739_cast_fp16)[name = tensor("op_17847_cast_fp16")]; + tensor var_17848_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1741_cast_fp16)[name = tensor("op_17848_cast_fp16")]; + tensor var_17849_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1743_cast_fp16)[name = tensor("op_17849_cast_fp16")]; + tensor var_17850_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1745_cast_fp16)[name = tensor("op_17850_cast_fp16")]; + tensor var_17851_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1747_cast_fp16)[name = tensor("op_17851_cast_fp16")]; + tensor var_17852_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1749_cast_fp16)[name = tensor("op_17852_cast_fp16")]; + tensor var_17853_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1751_cast_fp16)[name = tensor("op_17853_cast_fp16")]; + tensor var_17854_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1753_cast_fp16)[name = tensor("op_17854_cast_fp16")]; + tensor var_17855_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1755_cast_fp16)[name = tensor("op_17855_cast_fp16")]; + tensor var_17856_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1757_cast_fp16)[name = tensor("op_17856_cast_fp16")]; + tensor var_17857_cast_fp16 = softmax(axis = var_16549, x = aw_chunk_1759_cast_fp16)[name = tensor("op_17857_cast_fp16")]; + tensor var_17859_equation_0 = const()[name = tensor("op_17859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17859_cast_fp16 = einsum(equation = var_17859_equation_0, values = (var_17379_cast_fp16, var_17778_cast_fp16))[name = tensor("op_17859_cast_fp16")]; + tensor var_17861_equation_0 = const()[name = tensor("op_17861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17861_cast_fp16 = einsum(equation = var_17861_equation_0, values = (var_17379_cast_fp16, var_17779_cast_fp16))[name = tensor("op_17861_cast_fp16")]; + tensor var_17863_equation_0 = const()[name = tensor("op_17863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17863_cast_fp16 = einsum(equation = var_17863_equation_0, values = (var_17379_cast_fp16, var_17780_cast_fp16))[name = tensor("op_17863_cast_fp16")]; + tensor var_17865_equation_0 = const()[name = tensor("op_17865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17865_cast_fp16 = einsum(equation = var_17865_equation_0, values = (var_17379_cast_fp16, var_17781_cast_fp16))[name = tensor("op_17865_cast_fp16")]; + tensor var_17867_equation_0 = const()[name = tensor("op_17867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17867_cast_fp16 = einsum(equation = var_17867_equation_0, values = (var_17383_cast_fp16, var_17782_cast_fp16))[name = tensor("op_17867_cast_fp16")]; + tensor var_17869_equation_0 = const()[name = tensor("op_17869_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17869_cast_fp16 = einsum(equation = var_17869_equation_0, values = (var_17383_cast_fp16, var_17783_cast_fp16))[name = tensor("op_17869_cast_fp16")]; + tensor var_17871_equation_0 = const()[name = tensor("op_17871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17871_cast_fp16 = einsum(equation = var_17871_equation_0, values = (var_17383_cast_fp16, var_17784_cast_fp16))[name = tensor("op_17871_cast_fp16")]; + tensor var_17873_equation_0 = const()[name = tensor("op_17873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17873_cast_fp16 = einsum(equation = var_17873_equation_0, values = (var_17383_cast_fp16, var_17785_cast_fp16))[name = tensor("op_17873_cast_fp16")]; + tensor var_17875_equation_0 = const()[name = tensor("op_17875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17875_cast_fp16 = einsum(equation = var_17875_equation_0, values = (var_17387_cast_fp16, var_17786_cast_fp16))[name = tensor("op_17875_cast_fp16")]; + tensor var_17877_equation_0 = const()[name = tensor("op_17877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17877_cast_fp16 = einsum(equation = var_17877_equation_0, values = (var_17387_cast_fp16, var_17787_cast_fp16))[name = tensor("op_17877_cast_fp16")]; + tensor var_17879_equation_0 = const()[name = tensor("op_17879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17879_cast_fp16 = einsum(equation = var_17879_equation_0, values = (var_17387_cast_fp16, var_17788_cast_fp16))[name = tensor("op_17879_cast_fp16")]; + tensor var_17881_equation_0 = const()[name = tensor("op_17881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17881_cast_fp16 = einsum(equation = var_17881_equation_0, values = (var_17387_cast_fp16, var_17789_cast_fp16))[name = tensor("op_17881_cast_fp16")]; + tensor var_17883_equation_0 = const()[name = tensor("op_17883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17883_cast_fp16 = einsum(equation = var_17883_equation_0, values = (var_17391_cast_fp16, var_17790_cast_fp16))[name = tensor("op_17883_cast_fp16")]; + tensor var_17885_equation_0 = const()[name = tensor("op_17885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17885_cast_fp16 = einsum(equation = var_17885_equation_0, values = (var_17391_cast_fp16, var_17791_cast_fp16))[name = tensor("op_17885_cast_fp16")]; + tensor var_17887_equation_0 = const()[name = tensor("op_17887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17887_cast_fp16 = einsum(equation = var_17887_equation_0, values = (var_17391_cast_fp16, var_17792_cast_fp16))[name = tensor("op_17887_cast_fp16")]; + tensor var_17889_equation_0 = const()[name = tensor("op_17889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17889_cast_fp16 = einsum(equation = var_17889_equation_0, values = (var_17391_cast_fp16, var_17793_cast_fp16))[name = tensor("op_17889_cast_fp16")]; + tensor var_17891_equation_0 = const()[name = tensor("op_17891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17891_cast_fp16 = einsum(equation = var_17891_equation_0, values = (var_17395_cast_fp16, var_17794_cast_fp16))[name = tensor("op_17891_cast_fp16")]; + tensor var_17893_equation_0 = const()[name = tensor("op_17893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17893_cast_fp16 = einsum(equation = var_17893_equation_0, values = (var_17395_cast_fp16, var_17795_cast_fp16))[name = tensor("op_17893_cast_fp16")]; + tensor var_17895_equation_0 = const()[name = tensor("op_17895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17895_cast_fp16 = einsum(equation = var_17895_equation_0, values = (var_17395_cast_fp16, var_17796_cast_fp16))[name = tensor("op_17895_cast_fp16")]; + tensor var_17897_equation_0 = const()[name = tensor("op_17897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17897_cast_fp16 = einsum(equation = var_17897_equation_0, values = (var_17395_cast_fp16, var_17797_cast_fp16))[name = tensor("op_17897_cast_fp16")]; + tensor var_17899_equation_0 = const()[name = tensor("op_17899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17899_cast_fp16 = einsum(equation = var_17899_equation_0, values = (var_17399_cast_fp16, var_17798_cast_fp16))[name = tensor("op_17899_cast_fp16")]; + tensor var_17901_equation_0 = const()[name = tensor("op_17901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17901_cast_fp16 = einsum(equation = var_17901_equation_0, values = (var_17399_cast_fp16, var_17799_cast_fp16))[name = tensor("op_17901_cast_fp16")]; + tensor var_17903_equation_0 = const()[name = tensor("op_17903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17903_cast_fp16 = einsum(equation = var_17903_equation_0, values = (var_17399_cast_fp16, var_17800_cast_fp16))[name = tensor("op_17903_cast_fp16")]; + tensor var_17905_equation_0 = const()[name = tensor("op_17905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17905_cast_fp16 = einsum(equation = var_17905_equation_0, values = (var_17399_cast_fp16, var_17801_cast_fp16))[name = tensor("op_17905_cast_fp16")]; + tensor var_17907_equation_0 = const()[name = tensor("op_17907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17907_cast_fp16 = einsum(equation = var_17907_equation_0, values = (var_17403_cast_fp16, var_17802_cast_fp16))[name = tensor("op_17907_cast_fp16")]; + tensor var_17909_equation_0 = const()[name = tensor("op_17909_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17909_cast_fp16 = einsum(equation = var_17909_equation_0, values = (var_17403_cast_fp16, var_17803_cast_fp16))[name = tensor("op_17909_cast_fp16")]; + tensor var_17911_equation_0 = const()[name = tensor("op_17911_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17911_cast_fp16 = einsum(equation = var_17911_equation_0, values = (var_17403_cast_fp16, var_17804_cast_fp16))[name = tensor("op_17911_cast_fp16")]; + tensor var_17913_equation_0 = const()[name = tensor("op_17913_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17913_cast_fp16 = einsum(equation = var_17913_equation_0, values = (var_17403_cast_fp16, var_17805_cast_fp16))[name = tensor("op_17913_cast_fp16")]; + tensor var_17915_equation_0 = const()[name = tensor("op_17915_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17915_cast_fp16 = einsum(equation = var_17915_equation_0, values = (var_17407_cast_fp16, var_17806_cast_fp16))[name = tensor("op_17915_cast_fp16")]; + tensor var_17917_equation_0 = const()[name = tensor("op_17917_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17917_cast_fp16 = einsum(equation = var_17917_equation_0, values = (var_17407_cast_fp16, var_17807_cast_fp16))[name = tensor("op_17917_cast_fp16")]; + tensor var_17919_equation_0 = const()[name = tensor("op_17919_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17919_cast_fp16 = einsum(equation = var_17919_equation_0, values = (var_17407_cast_fp16, var_17808_cast_fp16))[name = tensor("op_17919_cast_fp16")]; + tensor var_17921_equation_0 = const()[name = tensor("op_17921_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17921_cast_fp16 = einsum(equation = var_17921_equation_0, values = (var_17407_cast_fp16, var_17809_cast_fp16))[name = tensor("op_17921_cast_fp16")]; + tensor var_17923_equation_0 = const()[name = tensor("op_17923_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17923_cast_fp16 = einsum(equation = var_17923_equation_0, values = (var_17411_cast_fp16, var_17810_cast_fp16))[name = tensor("op_17923_cast_fp16")]; + tensor var_17925_equation_0 = const()[name = tensor("op_17925_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17925_cast_fp16 = einsum(equation = var_17925_equation_0, values = (var_17411_cast_fp16, var_17811_cast_fp16))[name = tensor("op_17925_cast_fp16")]; + tensor var_17927_equation_0 = const()[name = tensor("op_17927_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17927_cast_fp16 = einsum(equation = var_17927_equation_0, values = (var_17411_cast_fp16, var_17812_cast_fp16))[name = tensor("op_17927_cast_fp16")]; + tensor var_17929_equation_0 = const()[name = tensor("op_17929_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17929_cast_fp16 = einsum(equation = var_17929_equation_0, values = (var_17411_cast_fp16, var_17813_cast_fp16))[name = tensor("op_17929_cast_fp16")]; + tensor var_17931_equation_0 = const()[name = tensor("op_17931_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17931_cast_fp16 = einsum(equation = var_17931_equation_0, values = (var_17415_cast_fp16, var_17814_cast_fp16))[name = tensor("op_17931_cast_fp16")]; + tensor var_17933_equation_0 = const()[name = tensor("op_17933_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17933_cast_fp16 = einsum(equation = var_17933_equation_0, values = (var_17415_cast_fp16, var_17815_cast_fp16))[name = tensor("op_17933_cast_fp16")]; + tensor var_17935_equation_0 = const()[name = tensor("op_17935_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17935_cast_fp16 = einsum(equation = var_17935_equation_0, values = (var_17415_cast_fp16, var_17816_cast_fp16))[name = tensor("op_17935_cast_fp16")]; + tensor var_17937_equation_0 = const()[name = tensor("op_17937_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17937_cast_fp16 = einsum(equation = var_17937_equation_0, values = (var_17415_cast_fp16, var_17817_cast_fp16))[name = tensor("op_17937_cast_fp16")]; + tensor var_17939_equation_0 = const()[name = tensor("op_17939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17939_cast_fp16 = einsum(equation = var_17939_equation_0, values = (var_17419_cast_fp16, var_17818_cast_fp16))[name = tensor("op_17939_cast_fp16")]; + tensor var_17941_equation_0 = const()[name = tensor("op_17941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17941_cast_fp16 = einsum(equation = var_17941_equation_0, values = (var_17419_cast_fp16, var_17819_cast_fp16))[name = tensor("op_17941_cast_fp16")]; + tensor var_17943_equation_0 = const()[name = tensor("op_17943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17943_cast_fp16 = einsum(equation = var_17943_equation_0, values = (var_17419_cast_fp16, var_17820_cast_fp16))[name = tensor("op_17943_cast_fp16")]; + tensor var_17945_equation_0 = const()[name = tensor("op_17945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17945_cast_fp16 = einsum(equation = var_17945_equation_0, values = (var_17419_cast_fp16, var_17821_cast_fp16))[name = tensor("op_17945_cast_fp16")]; + tensor var_17947_equation_0 = const()[name = tensor("op_17947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17947_cast_fp16 = einsum(equation = var_17947_equation_0, values = (var_17423_cast_fp16, var_17822_cast_fp16))[name = tensor("op_17947_cast_fp16")]; + tensor var_17949_equation_0 = const()[name = tensor("op_17949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17949_cast_fp16 = einsum(equation = var_17949_equation_0, values = (var_17423_cast_fp16, var_17823_cast_fp16))[name = tensor("op_17949_cast_fp16")]; + tensor var_17951_equation_0 = const()[name = tensor("op_17951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17951_cast_fp16 = einsum(equation = var_17951_equation_0, values = (var_17423_cast_fp16, var_17824_cast_fp16))[name = tensor("op_17951_cast_fp16")]; + tensor var_17953_equation_0 = const()[name = tensor("op_17953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17953_cast_fp16 = einsum(equation = var_17953_equation_0, values = (var_17423_cast_fp16, var_17825_cast_fp16))[name = tensor("op_17953_cast_fp16")]; + tensor var_17955_equation_0 = const()[name = tensor("op_17955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17955_cast_fp16 = einsum(equation = var_17955_equation_0, values = (var_17427_cast_fp16, var_17826_cast_fp16))[name = tensor("op_17955_cast_fp16")]; + tensor var_17957_equation_0 = const()[name = tensor("op_17957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17957_cast_fp16 = einsum(equation = var_17957_equation_0, values = (var_17427_cast_fp16, var_17827_cast_fp16))[name = tensor("op_17957_cast_fp16")]; + tensor var_17959_equation_0 = const()[name = tensor("op_17959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17959_cast_fp16 = einsum(equation = var_17959_equation_0, values = (var_17427_cast_fp16, var_17828_cast_fp16))[name = tensor("op_17959_cast_fp16")]; + tensor var_17961_equation_0 = const()[name = tensor("op_17961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17961_cast_fp16 = einsum(equation = var_17961_equation_0, values = (var_17427_cast_fp16, var_17829_cast_fp16))[name = tensor("op_17961_cast_fp16")]; + tensor var_17963_equation_0 = const()[name = tensor("op_17963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17963_cast_fp16 = einsum(equation = var_17963_equation_0, values = (var_17431_cast_fp16, var_17830_cast_fp16))[name = tensor("op_17963_cast_fp16")]; + tensor var_17965_equation_0 = const()[name = tensor("op_17965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17965_cast_fp16 = einsum(equation = var_17965_equation_0, values = (var_17431_cast_fp16, var_17831_cast_fp16))[name = tensor("op_17965_cast_fp16")]; + tensor var_17967_equation_0 = const()[name = tensor("op_17967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17967_cast_fp16 = einsum(equation = var_17967_equation_0, values = (var_17431_cast_fp16, var_17832_cast_fp16))[name = tensor("op_17967_cast_fp16")]; + tensor var_17969_equation_0 = const()[name = tensor("op_17969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17969_cast_fp16 = einsum(equation = var_17969_equation_0, values = (var_17431_cast_fp16, var_17833_cast_fp16))[name = tensor("op_17969_cast_fp16")]; + tensor var_17971_equation_0 = const()[name = tensor("op_17971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17971_cast_fp16 = einsum(equation = var_17971_equation_0, values = (var_17435_cast_fp16, var_17834_cast_fp16))[name = tensor("op_17971_cast_fp16")]; + tensor var_17973_equation_0 = const()[name = tensor("op_17973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17973_cast_fp16 = einsum(equation = var_17973_equation_0, values = (var_17435_cast_fp16, var_17835_cast_fp16))[name = tensor("op_17973_cast_fp16")]; + tensor var_17975_equation_0 = const()[name = tensor("op_17975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17975_cast_fp16 = einsum(equation = var_17975_equation_0, values = (var_17435_cast_fp16, var_17836_cast_fp16))[name = tensor("op_17975_cast_fp16")]; + tensor var_17977_equation_0 = const()[name = tensor("op_17977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17977_cast_fp16 = einsum(equation = var_17977_equation_0, values = (var_17435_cast_fp16, var_17837_cast_fp16))[name = tensor("op_17977_cast_fp16")]; + tensor var_17979_equation_0 = const()[name = tensor("op_17979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17979_cast_fp16 = einsum(equation = var_17979_equation_0, values = (var_17439_cast_fp16, var_17838_cast_fp16))[name = tensor("op_17979_cast_fp16")]; + tensor var_17981_equation_0 = const()[name = tensor("op_17981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17981_cast_fp16 = einsum(equation = var_17981_equation_0, values = (var_17439_cast_fp16, var_17839_cast_fp16))[name = tensor("op_17981_cast_fp16")]; + tensor var_17983_equation_0 = const()[name = tensor("op_17983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17983_cast_fp16 = einsum(equation = var_17983_equation_0, values = (var_17439_cast_fp16, var_17840_cast_fp16))[name = tensor("op_17983_cast_fp16")]; + tensor var_17985_equation_0 = const()[name = tensor("op_17985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17985_cast_fp16 = einsum(equation = var_17985_equation_0, values = (var_17439_cast_fp16, var_17841_cast_fp16))[name = tensor("op_17985_cast_fp16")]; + tensor var_17987_equation_0 = const()[name = tensor("op_17987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17987_cast_fp16 = einsum(equation = var_17987_equation_0, values = (var_17443_cast_fp16, var_17842_cast_fp16))[name = tensor("op_17987_cast_fp16")]; + tensor var_17989_equation_0 = const()[name = tensor("op_17989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17989_cast_fp16 = einsum(equation = var_17989_equation_0, values = (var_17443_cast_fp16, var_17843_cast_fp16))[name = tensor("op_17989_cast_fp16")]; + tensor var_17991_equation_0 = const()[name = tensor("op_17991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17991_cast_fp16 = einsum(equation = var_17991_equation_0, values = (var_17443_cast_fp16, var_17844_cast_fp16))[name = tensor("op_17991_cast_fp16")]; + tensor var_17993_equation_0 = const()[name = tensor("op_17993_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17993_cast_fp16 = einsum(equation = var_17993_equation_0, values = (var_17443_cast_fp16, var_17845_cast_fp16))[name = tensor("op_17993_cast_fp16")]; + tensor var_17995_equation_0 = const()[name = tensor("op_17995_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17995_cast_fp16 = einsum(equation = var_17995_equation_0, values = (var_17447_cast_fp16, var_17846_cast_fp16))[name = tensor("op_17995_cast_fp16")]; + tensor var_17997_equation_0 = const()[name = tensor("op_17997_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17997_cast_fp16 = einsum(equation = var_17997_equation_0, values = (var_17447_cast_fp16, var_17847_cast_fp16))[name = tensor("op_17997_cast_fp16")]; + tensor var_17999_equation_0 = const()[name = tensor("op_17999_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_17999_cast_fp16 = einsum(equation = var_17999_equation_0, values = (var_17447_cast_fp16, var_17848_cast_fp16))[name = tensor("op_17999_cast_fp16")]; + tensor var_18001_equation_0 = const()[name = tensor("op_18001_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18001_cast_fp16 = einsum(equation = var_18001_equation_0, values = (var_17447_cast_fp16, var_17849_cast_fp16))[name = tensor("op_18001_cast_fp16")]; + tensor var_18003_equation_0 = const()[name = tensor("op_18003_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18003_cast_fp16 = einsum(equation = var_18003_equation_0, values = (var_17451_cast_fp16, var_17850_cast_fp16))[name = tensor("op_18003_cast_fp16")]; + tensor var_18005_equation_0 = const()[name = tensor("op_18005_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18005_cast_fp16 = einsum(equation = var_18005_equation_0, values = (var_17451_cast_fp16, var_17851_cast_fp16))[name = tensor("op_18005_cast_fp16")]; + tensor var_18007_equation_0 = const()[name = tensor("op_18007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18007_cast_fp16 = einsum(equation = var_18007_equation_0, values = (var_17451_cast_fp16, var_17852_cast_fp16))[name = tensor("op_18007_cast_fp16")]; + tensor var_18009_equation_0 = const()[name = tensor("op_18009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18009_cast_fp16 = einsum(equation = var_18009_equation_0, values = (var_17451_cast_fp16, var_17853_cast_fp16))[name = tensor("op_18009_cast_fp16")]; + tensor var_18011_equation_0 = const()[name = tensor("op_18011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18011_cast_fp16 = einsum(equation = var_18011_equation_0, values = (var_17455_cast_fp16, var_17854_cast_fp16))[name = tensor("op_18011_cast_fp16")]; + tensor var_18013_equation_0 = const()[name = tensor("op_18013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18013_cast_fp16 = einsum(equation = var_18013_equation_0, values = (var_17455_cast_fp16, var_17855_cast_fp16))[name = tensor("op_18013_cast_fp16")]; + tensor var_18015_equation_0 = const()[name = tensor("op_18015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18015_cast_fp16 = einsum(equation = var_18015_equation_0, values = (var_17455_cast_fp16, var_17856_cast_fp16))[name = tensor("op_18015_cast_fp16")]; + tensor var_18017_equation_0 = const()[name = tensor("op_18017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_18017_cast_fp16 = einsum(equation = var_18017_equation_0, values = (var_17455_cast_fp16, var_17857_cast_fp16))[name = tensor("op_18017_cast_fp16")]; + tensor var_18019_interleave_0 = const()[name = tensor("op_18019_interleave_0"), val = tensor(false)]; + tensor var_18019_cast_fp16 = concat(axis = var_16524, interleave = var_18019_interleave_0, values = (var_17859_cast_fp16, var_17861_cast_fp16, var_17863_cast_fp16, var_17865_cast_fp16))[name = tensor("op_18019_cast_fp16")]; + tensor var_18021_interleave_0 = const()[name = tensor("op_18021_interleave_0"), val = tensor(false)]; + tensor var_18021_cast_fp16 = concat(axis = var_16524, interleave = var_18021_interleave_0, values = (var_17867_cast_fp16, var_17869_cast_fp16, var_17871_cast_fp16, var_17873_cast_fp16))[name = tensor("op_18021_cast_fp16")]; + tensor var_18023_interleave_0 = const()[name = tensor("op_18023_interleave_0"), val = tensor(false)]; + tensor var_18023_cast_fp16 = concat(axis = var_16524, interleave = var_18023_interleave_0, values = (var_17875_cast_fp16, var_17877_cast_fp16, var_17879_cast_fp16, var_17881_cast_fp16))[name = tensor("op_18023_cast_fp16")]; + tensor var_18025_interleave_0 = const()[name = tensor("op_18025_interleave_0"), val = tensor(false)]; + tensor var_18025_cast_fp16 = concat(axis = var_16524, interleave = var_18025_interleave_0, values = (var_17883_cast_fp16, var_17885_cast_fp16, var_17887_cast_fp16, var_17889_cast_fp16))[name = tensor("op_18025_cast_fp16")]; + tensor var_18027_interleave_0 = const()[name = tensor("op_18027_interleave_0"), val = tensor(false)]; + tensor var_18027_cast_fp16 = concat(axis = var_16524, interleave = var_18027_interleave_0, values = (var_17891_cast_fp16, var_17893_cast_fp16, var_17895_cast_fp16, var_17897_cast_fp16))[name = tensor("op_18027_cast_fp16")]; + tensor var_18029_interleave_0 = const()[name = tensor("op_18029_interleave_0"), val = tensor(false)]; + tensor var_18029_cast_fp16 = concat(axis = var_16524, interleave = var_18029_interleave_0, values = (var_17899_cast_fp16, var_17901_cast_fp16, var_17903_cast_fp16, var_17905_cast_fp16))[name = tensor("op_18029_cast_fp16")]; + tensor var_18031_interleave_0 = const()[name = tensor("op_18031_interleave_0"), val = tensor(false)]; + tensor var_18031_cast_fp16 = concat(axis = var_16524, interleave = var_18031_interleave_0, values = (var_17907_cast_fp16, var_17909_cast_fp16, var_17911_cast_fp16, var_17913_cast_fp16))[name = tensor("op_18031_cast_fp16")]; + tensor var_18033_interleave_0 = const()[name = tensor("op_18033_interleave_0"), val = tensor(false)]; + tensor var_18033_cast_fp16 = concat(axis = var_16524, interleave = var_18033_interleave_0, values = (var_17915_cast_fp16, var_17917_cast_fp16, var_17919_cast_fp16, var_17921_cast_fp16))[name = tensor("op_18033_cast_fp16")]; + tensor var_18035_interleave_0 = const()[name = tensor("op_18035_interleave_0"), val = tensor(false)]; + tensor var_18035_cast_fp16 = concat(axis = var_16524, interleave = var_18035_interleave_0, values = (var_17923_cast_fp16, var_17925_cast_fp16, var_17927_cast_fp16, var_17929_cast_fp16))[name = tensor("op_18035_cast_fp16")]; + tensor var_18037_interleave_0 = const()[name = tensor("op_18037_interleave_0"), val = tensor(false)]; + tensor var_18037_cast_fp16 = concat(axis = var_16524, interleave = var_18037_interleave_0, values = (var_17931_cast_fp16, var_17933_cast_fp16, var_17935_cast_fp16, var_17937_cast_fp16))[name = tensor("op_18037_cast_fp16")]; + tensor var_18039_interleave_0 = const()[name = tensor("op_18039_interleave_0"), val = tensor(false)]; + tensor var_18039_cast_fp16 = concat(axis = var_16524, interleave = var_18039_interleave_0, values = (var_17939_cast_fp16, var_17941_cast_fp16, var_17943_cast_fp16, var_17945_cast_fp16))[name = tensor("op_18039_cast_fp16")]; + tensor var_18041_interleave_0 = const()[name = tensor("op_18041_interleave_0"), val = tensor(false)]; + tensor var_18041_cast_fp16 = concat(axis = var_16524, interleave = var_18041_interleave_0, values = (var_17947_cast_fp16, var_17949_cast_fp16, var_17951_cast_fp16, var_17953_cast_fp16))[name = tensor("op_18041_cast_fp16")]; + tensor var_18043_interleave_0 = const()[name = tensor("op_18043_interleave_0"), val = tensor(false)]; + tensor var_18043_cast_fp16 = concat(axis = var_16524, interleave = var_18043_interleave_0, values = (var_17955_cast_fp16, var_17957_cast_fp16, var_17959_cast_fp16, var_17961_cast_fp16))[name = tensor("op_18043_cast_fp16")]; + tensor var_18045_interleave_0 = const()[name = tensor("op_18045_interleave_0"), val = tensor(false)]; + tensor var_18045_cast_fp16 = concat(axis = var_16524, interleave = var_18045_interleave_0, values = (var_17963_cast_fp16, var_17965_cast_fp16, var_17967_cast_fp16, var_17969_cast_fp16))[name = tensor("op_18045_cast_fp16")]; + tensor var_18047_interleave_0 = const()[name = tensor("op_18047_interleave_0"), val = tensor(false)]; + tensor var_18047_cast_fp16 = concat(axis = var_16524, interleave = var_18047_interleave_0, values = (var_17971_cast_fp16, var_17973_cast_fp16, var_17975_cast_fp16, var_17977_cast_fp16))[name = tensor("op_18047_cast_fp16")]; + tensor var_18049_interleave_0 = const()[name = tensor("op_18049_interleave_0"), val = tensor(false)]; + tensor var_18049_cast_fp16 = concat(axis = var_16524, interleave = var_18049_interleave_0, values = (var_17979_cast_fp16, var_17981_cast_fp16, var_17983_cast_fp16, var_17985_cast_fp16))[name = tensor("op_18049_cast_fp16")]; + tensor var_18051_interleave_0 = const()[name = tensor("op_18051_interleave_0"), val = tensor(false)]; + tensor var_18051_cast_fp16 = concat(axis = var_16524, interleave = var_18051_interleave_0, values = (var_17987_cast_fp16, var_17989_cast_fp16, var_17991_cast_fp16, var_17993_cast_fp16))[name = tensor("op_18051_cast_fp16")]; + tensor var_18053_interleave_0 = const()[name = tensor("op_18053_interleave_0"), val = tensor(false)]; + tensor var_18053_cast_fp16 = concat(axis = var_16524, interleave = var_18053_interleave_0, values = (var_17995_cast_fp16, var_17997_cast_fp16, var_17999_cast_fp16, var_18001_cast_fp16))[name = tensor("op_18053_cast_fp16")]; + tensor var_18055_interleave_0 = const()[name = tensor("op_18055_interleave_0"), val = tensor(false)]; + tensor var_18055_cast_fp16 = concat(axis = var_16524, interleave = var_18055_interleave_0, values = (var_18003_cast_fp16, var_18005_cast_fp16, var_18007_cast_fp16, var_18009_cast_fp16))[name = tensor("op_18055_cast_fp16")]; + tensor var_18057_interleave_0 = const()[name = tensor("op_18057_interleave_0"), val = tensor(false)]; + tensor var_18057_cast_fp16 = concat(axis = var_16524, interleave = var_18057_interleave_0, values = (var_18011_cast_fp16, var_18013_cast_fp16, var_18015_cast_fp16, var_18017_cast_fp16))[name = tensor("op_18057_cast_fp16")]; + tensor input_207_interleave_0 = const()[name = tensor("input_207_interleave_0"), val = tensor(false)]; + tensor input_207_cast_fp16 = concat(axis = var_16549, interleave = input_207_interleave_0, values = (var_18019_cast_fp16, var_18021_cast_fp16, var_18023_cast_fp16, var_18025_cast_fp16, var_18027_cast_fp16, var_18029_cast_fp16, var_18031_cast_fp16, var_18033_cast_fp16, var_18035_cast_fp16, var_18037_cast_fp16, var_18039_cast_fp16, var_18041_cast_fp16, var_18043_cast_fp16, var_18045_cast_fp16, var_18047_cast_fp16, var_18049_cast_fp16, var_18051_cast_fp16, var_18053_cast_fp16, var_18055_cast_fp16, var_18057_cast_fp16))[name = tensor("input_207_cast_fp16")]; + tensor var_18065 = const()[name = tensor("op_18065"), val = tensor([1, 1])]; + tensor var_18067 = const()[name = tensor("op_18067"), val = tensor([1, 1])]; + tensor pretrained_out_127_pad_type_0 = const()[name = tensor("pretrained_out_127_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_127_pad_0 = const()[name = tensor("pretrained_out_127_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123029632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123848896))), name = tensor("layers_10_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_10_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123849024)))]; + tensor pretrained_out_127_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_18067, groups = var_16549, pad = pretrained_out_127_pad_0, pad_type = pretrained_out_127_pad_type_0, strides = var_18065, weight = layers_10_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor("pretrained_out_127_cast_fp16")]; + tensor var_18071 = const()[name = tensor("op_18071"), val = tensor([1, 1])]; + tensor var_18073 = const()[name = tensor("op_18073"), val = tensor([1, 1])]; + tensor input_209_pad_type_0 = const()[name = tensor("input_209_pad_type_0"), val = tensor("custom")]; + tensor input_209_pad_0 = const()[name = tensor("input_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123851648)))]; + tensor input_209_cast_fp16 = conv(dilations = var_18073, groups = var_16549, pad = input_209_pad_0, pad_type = input_209_pad_type_0, strides = var_18071, weight = layers_10_self_attn_o_proj_loraA_weight_to_fp16, x = input_207_cast_fp16)[name = tensor("input_209_cast_fp16")]; + tensor var_18077 = const()[name = tensor("op_18077"), val = tensor([1, 1])]; + tensor var_18079 = const()[name = tensor("op_18079"), val = tensor([1, 1])]; + tensor lora_out_253_pad_type_0 = const()[name = tensor("lora_out_253_pad_type_0"), val = tensor("custom")]; + tensor lora_out_253_pad_0 = const()[name = tensor("lora_out_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_255_weight_0_to_fp16 = const()[name = tensor("lora_out_255_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123892672)))]; + tensor lora_out_255_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_18079, groups = var_16549, pad = lora_out_253_pad_0, pad_type = lora_out_253_pad_type_0, strides = var_18077, weight = lora_out_255_weight_0_to_fp16, x = input_209_cast_fp16)[name = tensor("lora_out_255_cast_fp16")]; + tensor obj_43_cast_fp16 = add(x = pretrained_out_127_cast_fp16, y = lora_out_255_cast_fp16)[name = tensor("obj_43_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; + tensor var_18088 = const()[name = tensor("op_18088"), val = tensor([1])]; + tensor channels_mean_43_cast_fp16 = reduce_mean(axes = var_18088, keep_dims = var_16550, x = inputs_43_cast_fp16)[name = tensor("channels_mean_43_cast_fp16")]; + tensor zero_mean_43_cast_fp16 = sub(x = inputs_43_cast_fp16, y = channels_mean_43_cast_fp16)[name = tensor("zero_mean_43_cast_fp16")]; + tensor zero_mean_sq_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = zero_mean_43_cast_fp16)[name = tensor("zero_mean_sq_43_cast_fp16")]; + tensor var_18092 = const()[name = tensor("op_18092"), val = tensor([1])]; + tensor var_18093_cast_fp16 = reduce_mean(axes = var_18092, keep_dims = var_16550, x = zero_mean_sq_43_cast_fp16)[name = tensor("op_18093_cast_fp16")]; + tensor var_18094_to_fp16 = const()[name = tensor("op_18094_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_18095_cast_fp16 = add(x = var_18093_cast_fp16, y = var_18094_to_fp16)[name = tensor("op_18095_cast_fp16")]; + tensor denom_43_epsilon_0 = const()[name = tensor("denom_43_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_43_cast_fp16 = rsqrt(epsilon = denom_43_epsilon_0, x = var_18095_cast_fp16)[name = tensor("denom_43_cast_fp16")]; + tensor out_43_cast_fp16 = mul(x = zero_mean_43_cast_fp16, y = denom_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; + tensor input_211_gamma_0_to_fp16 = const()[name = tensor("input_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123933696)))]; + tensor input_211_beta_0_to_fp16 = const()[name = tensor("input_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123936320)))]; + tensor input_211_epsilon_0_to_fp16 = const()[name = tensor("input_211_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_211_cast_fp16 = batch_norm(beta = input_211_beta_0_to_fp16, epsilon = input_211_epsilon_0_to_fp16, gamma = input_211_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor var_18109 = const()[name = tensor("op_18109"), val = tensor([1, 1])]; + tensor var_18111 = const()[name = tensor("op_18111"), val = tensor([1, 1])]; + tensor pretrained_out_129_pad_type_0 = const()[name = tensor("pretrained_out_129_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_129_pad_0 = const()[name = tensor("pretrained_out_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123938944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127215808))), name = tensor("layers_10_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_10_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_10_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127215936)))]; + tensor pretrained_out_129_cast_fp16 = conv(bias = layers_10_fc1_pretrained_bias_to_fp16, dilations = var_18111, groups = var_16549, pad = pretrained_out_129_pad_0, pad_type = pretrained_out_129_pad_type_0, strides = var_18109, weight = layers_10_fc1_pretrained_weight_to_fp16_palettized, x = input_211_cast_fp16)[name = tensor("pretrained_out_129_cast_fp16")]; + tensor var_18115 = const()[name = tensor("op_18115"), val = tensor([1, 1])]; + tensor var_18117 = const()[name = tensor("op_18117"), val = tensor([1, 1])]; + tensor input_213_pad_type_0 = const()[name = tensor("input_213_pad_type_0"), val = tensor("custom")]; + tensor input_213_pad_0 = const()[name = tensor("input_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_10_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127226240)))]; + tensor input_213_cast_fp16 = conv(dilations = var_18117, groups = var_16549, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = var_18115, weight = layers_10_fc1_loraA_weight_to_fp16, x = input_211_cast_fp16)[name = tensor("input_213_cast_fp16")]; + tensor var_18121 = const()[name = tensor("op_18121"), val = tensor([1, 1])]; + tensor var_18123 = const()[name = tensor("op_18123"), val = tensor([1, 1])]; + tensor lora_out_257_pad_type_0 = const()[name = tensor("lora_out_257_pad_type_0"), val = tensor("custom")]; + tensor lora_out_257_pad_0 = const()[name = tensor("lora_out_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_259_weight_0_to_fp16 = const()[name = tensor("lora_out_259_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127267264)))]; + tensor lora_out_259_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_18123, groups = var_16549, pad = lora_out_257_pad_0, pad_type = lora_out_257_pad_type_0, strides = var_18121, weight = lora_out_259_weight_0_to_fp16, x = input_213_cast_fp16)[name = tensor("lora_out_259_cast_fp16")]; + tensor input_215_cast_fp16 = add(x = pretrained_out_129_cast_fp16, y = lora_out_259_cast_fp16)[name = tensor("input_215_cast_fp16")]; + tensor input_217_mode_0 = const()[name = tensor("input_217_mode_0"), val = tensor("EXACT")]; + tensor input_217_cast_fp16 = gelu(mode = input_217_mode_0, x = input_215_cast_fp16)[name = tensor("input_217_cast_fp16")]; + tensor var_18135 = const()[name = tensor("op_18135"), val = tensor([1, 1])]; + tensor var_18137 = const()[name = tensor("op_18137"), val = tensor([1, 1])]; + tensor pretrained_out_131_pad_type_0 = const()[name = tensor("pretrained_out_131_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_131_pad_0 = const()[name = tensor("pretrained_out_131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127431168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130708032))), name = tensor("layers_10_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_10_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_10_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130708160)))]; + tensor pretrained_out_131_cast_fp16 = conv(bias = layers_10_fc2_pretrained_bias_to_fp16, dilations = var_18137, groups = var_16549, pad = pretrained_out_131_pad_0, pad_type = pretrained_out_131_pad_type_0, strides = var_18135, weight = layers_10_fc2_pretrained_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor("pretrained_out_131_cast_fp16")]; + tensor var_18141 = const()[name = tensor("op_18141"), val = tensor([1, 1])]; + tensor var_18143 = const()[name = tensor("op_18143"), val = tensor([1, 1])]; + tensor input_219_pad_type_0 = const()[name = tensor("input_219_pad_type_0"), val = tensor("custom")]; + tensor input_219_pad_0 = const()[name = tensor("input_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_10_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_10_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130710784)))]; + tensor input_219_cast_fp16 = conv(dilations = var_18143, groups = var_16549, pad = input_219_pad_0, pad_type = input_219_pad_type_0, strides = var_18141, weight = layers_10_fc2_loraA_weight_to_fp16, x = input_217_cast_fp16)[name = tensor("input_219_cast_fp16")]; + tensor var_18147 = const()[name = tensor("op_18147"), val = tensor([1, 1])]; + tensor var_18149 = const()[name = tensor("op_18149"), val = tensor([1, 1])]; + tensor lora_out_261_pad_type_0 = const()[name = tensor("lora_out_261_pad_type_0"), val = tensor("custom")]; + tensor lora_out_261_pad_0 = const()[name = tensor("lora_out_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_263_weight_0_to_fp16 = const()[name = tensor("lora_out_263_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130874688)))]; + tensor lora_out_263_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_18149, groups = var_16549, pad = lora_out_261_pad_0, pad_type = lora_out_261_pad_type_0, strides = var_18147, weight = lora_out_263_weight_0_to_fp16, x = input_219_cast_fp16)[name = tensor("lora_out_263_cast_fp16")]; + tensor hidden_states_25_cast_fp16 = add(x = pretrained_out_131_cast_fp16, y = lora_out_263_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; + tensor var_18159 = const()[name = tensor("op_18159"), val = tensor(3)]; + tensor var_18184 = const()[name = tensor("op_18184"), val = tensor(1)]; + tensor var_18185 = const()[name = tensor("op_18185"), val = tensor(true)]; + tensor var_18195 = const()[name = tensor("op_18195"), val = tensor([1])]; + tensor channels_mean_45_cast_fp16 = reduce_mean(axes = var_18195, keep_dims = var_18185, x = inputs_45_cast_fp16)[name = tensor("channels_mean_45_cast_fp16")]; + tensor zero_mean_45_cast_fp16 = sub(x = inputs_45_cast_fp16, y = channels_mean_45_cast_fp16)[name = tensor("zero_mean_45_cast_fp16")]; + tensor zero_mean_sq_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = zero_mean_45_cast_fp16)[name = tensor("zero_mean_sq_45_cast_fp16")]; + tensor var_18199 = const()[name = tensor("op_18199"), val = tensor([1])]; + tensor var_18200_cast_fp16 = reduce_mean(axes = var_18199, keep_dims = var_18185, x = zero_mean_sq_45_cast_fp16)[name = tensor("op_18200_cast_fp16")]; + tensor var_18201_to_fp16 = const()[name = tensor("op_18201_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_18202_cast_fp16 = add(x = var_18200_cast_fp16, y = var_18201_to_fp16)[name = tensor("op_18202_cast_fp16")]; + tensor denom_45_epsilon_0 = const()[name = tensor("denom_45_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_45_cast_fp16 = rsqrt(epsilon = denom_45_epsilon_0, x = var_18202_cast_fp16)[name = tensor("denom_45_cast_fp16")]; + tensor out_45_cast_fp16 = mul(x = zero_mean_45_cast_fp16, y = denom_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; + tensor obj_45_gamma_0_to_fp16 = const()[name = tensor("obj_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130915712)))]; + tensor obj_45_beta_0_to_fp16 = const()[name = tensor("obj_45_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130918336)))]; + tensor obj_45_epsilon_0_to_fp16 = const()[name = tensor("obj_45_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor("obj_45_cast_fp16")]; + tensor var_18220 = const()[name = tensor("op_18220"), val = tensor([1, 1])]; + tensor var_18222 = const()[name = tensor("op_18222"), val = tensor([1, 1])]; + tensor pretrained_out_133_pad_type_0 = const()[name = tensor("pretrained_out_133_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_133_pad_0 = const()[name = tensor("pretrained_out_133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130920960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131740224))), name = tensor("layers_11_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_11_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131740352)))]; + tensor pretrained_out_133_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_18222, groups = var_18184, pad = pretrained_out_133_pad_0, pad_type = pretrained_out_133_pad_type_0, strides = var_18220, weight = layers_11_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor("pretrained_out_133_cast_fp16")]; + tensor var_18226 = const()[name = tensor("op_18226"), val = tensor([1, 1])]; + tensor var_18228 = const()[name = tensor("op_18228"), val = tensor([1, 1])]; + tensor input_221_pad_type_0 = const()[name = tensor("input_221_pad_type_0"), val = tensor("custom")]; + tensor input_221_pad_0 = const()[name = tensor("input_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131742976)))]; + tensor input_221_cast_fp16 = conv(dilations = var_18228, groups = var_18184, pad = input_221_pad_0, pad_type = input_221_pad_type_0, strides = var_18226, weight = layers_11_self_attn_q_proj_loraA_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor var_18232 = const()[name = tensor("op_18232"), val = tensor([1, 1])]; + tensor var_18234 = const()[name = tensor("op_18234"), val = tensor([1, 1])]; + tensor lora_out_265_pad_type_0 = const()[name = tensor("lora_out_265_pad_type_0"), val = tensor("custom")]; + tensor lora_out_265_pad_0 = const()[name = tensor("lora_out_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_267_weight_0_to_fp16 = const()[name = tensor("lora_out_267_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131784000)))]; + tensor lora_out_267_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_18234, groups = var_18184, pad = lora_out_265_pad_0, pad_type = lora_out_265_pad_type_0, strides = var_18232, weight = lora_out_267_weight_0_to_fp16, x = input_221_cast_fp16)[name = tensor("lora_out_267_cast_fp16")]; + tensor query_23_cast_fp16 = add(x = pretrained_out_133_cast_fp16, y = lora_out_267_cast_fp16)[name = tensor("query_23_cast_fp16")]; + tensor var_18244 = const()[name = tensor("op_18244"), val = tensor([1, 1])]; + tensor var_18246 = const()[name = tensor("op_18246"), val = tensor([1, 1])]; + tensor pretrained_out_135_pad_type_0 = const()[name = tensor("pretrained_out_135_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_135_pad_0 = const()[name = tensor("pretrained_out_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131825024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132644288))), name = tensor("layers_11_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_135_cast_fp16 = conv(dilations = var_18246, groups = var_18184, pad = pretrained_out_135_pad_0, pad_type = pretrained_out_135_pad_type_0, strides = var_18244, weight = layers_11_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor("pretrained_out_135_cast_fp16")]; + tensor var_18250 = const()[name = tensor("op_18250"), val = tensor([1, 1])]; + tensor var_18252 = const()[name = tensor("op_18252"), val = tensor([1, 1])]; + tensor input_223_pad_type_0 = const()[name = tensor("input_223_pad_type_0"), val = tensor("custom")]; + tensor input_223_pad_0 = const()[name = tensor("input_223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132644416)))]; + tensor input_223_cast_fp16 = conv(dilations = var_18252, groups = var_18184, pad = input_223_pad_0, pad_type = input_223_pad_type_0, strides = var_18250, weight = layers_11_self_attn_k_proj_loraA_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("input_223_cast_fp16")]; + tensor var_18256 = const()[name = tensor("op_18256"), val = tensor([1, 1])]; + tensor var_18258 = const()[name = tensor("op_18258"), val = tensor([1, 1])]; + tensor lora_out_269_pad_type_0 = const()[name = tensor("lora_out_269_pad_type_0"), val = tensor("custom")]; + tensor lora_out_269_pad_0 = const()[name = tensor("lora_out_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_271_weight_0_to_fp16 = const()[name = tensor("lora_out_271_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132685440)))]; + tensor lora_out_271_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_18258, groups = var_18184, pad = lora_out_269_pad_0, pad_type = lora_out_269_pad_type_0, strides = var_18256, weight = lora_out_271_weight_0_to_fp16, x = input_223_cast_fp16)[name = tensor("lora_out_271_cast_fp16")]; + tensor key_23_cast_fp16 = add(x = pretrained_out_135_cast_fp16, y = lora_out_271_cast_fp16)[name = tensor("key_23_cast_fp16")]; + tensor var_18269 = const()[name = tensor("op_18269"), val = tensor([1, 1])]; + tensor var_18271 = const()[name = tensor("op_18271"), val = tensor([1, 1])]; + tensor pretrained_out_137_pad_type_0 = const()[name = tensor("pretrained_out_137_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_137_pad_0 = const()[name = tensor("pretrained_out_137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132726464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133545728))), name = tensor("layers_11_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_11_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133545856)))]; + tensor pretrained_out_137_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_18271, groups = var_18184, pad = pretrained_out_137_pad_0, pad_type = pretrained_out_137_pad_type_0, strides = var_18269, weight = layers_11_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor("pretrained_out_137_cast_fp16")]; + tensor var_18275 = const()[name = tensor("op_18275"), val = tensor([1, 1])]; + tensor var_18277 = const()[name = tensor("op_18277"), val = tensor([1, 1])]; + tensor input_225_pad_type_0 = const()[name = tensor("input_225_pad_type_0"), val = tensor("custom")]; + tensor input_225_pad_0 = const()[name = tensor("input_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133548480)))]; + tensor input_225_cast_fp16 = conv(dilations = var_18277, groups = var_18184, pad = input_225_pad_0, pad_type = input_225_pad_type_0, strides = var_18275, weight = layers_11_self_attn_v_proj_loraA_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor var_18281 = const()[name = tensor("op_18281"), val = tensor([1, 1])]; + tensor var_18283 = const()[name = tensor("op_18283"), val = tensor([1, 1])]; + tensor lora_out_273_pad_type_0 = const()[name = tensor("lora_out_273_pad_type_0"), val = tensor("custom")]; + tensor lora_out_273_pad_0 = const()[name = tensor("lora_out_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_275_weight_0_to_fp16 = const()[name = tensor("lora_out_275_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133589504)))]; + tensor lora_out_275_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_18283, groups = var_18184, pad = lora_out_273_pad_0, pad_type = lora_out_273_pad_type_0, strides = var_18281, weight = lora_out_275_weight_0_to_fp16, x = input_225_cast_fp16)[name = tensor("lora_out_275_cast_fp16")]; + tensor value_23_cast_fp16 = add(x = pretrained_out_137_cast_fp16, y = lora_out_275_cast_fp16)[name = tensor("value_23_cast_fp16")]; + tensor var_18293_begin_0 = const()[name = tensor("op_18293_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18293_end_0 = const()[name = tensor("op_18293_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18293_end_mask_0 = const()[name = tensor("op_18293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18293_cast_fp16 = slice_by_index(begin = var_18293_begin_0, end = var_18293_end_0, end_mask = var_18293_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18293_cast_fp16")]; + tensor var_18297_begin_0 = const()[name = tensor("op_18297_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_18297_end_0 = const()[name = tensor("op_18297_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_18297_end_mask_0 = const()[name = tensor("op_18297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18297_cast_fp16 = slice_by_index(begin = var_18297_begin_0, end = var_18297_end_0, end_mask = var_18297_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18297_cast_fp16")]; + tensor var_18301_begin_0 = const()[name = tensor("op_18301_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_18301_end_0 = const()[name = tensor("op_18301_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_18301_end_mask_0 = const()[name = tensor("op_18301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18301_cast_fp16 = slice_by_index(begin = var_18301_begin_0, end = var_18301_end_0, end_mask = var_18301_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18301_cast_fp16")]; + tensor var_18305_begin_0 = const()[name = tensor("op_18305_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_18305_end_0 = const()[name = tensor("op_18305_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_18305_end_mask_0 = const()[name = tensor("op_18305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18305_cast_fp16 = slice_by_index(begin = var_18305_begin_0, end = var_18305_end_0, end_mask = var_18305_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18305_cast_fp16")]; + tensor var_18309_begin_0 = const()[name = tensor("op_18309_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_18309_end_0 = const()[name = tensor("op_18309_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_18309_end_mask_0 = const()[name = tensor("op_18309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18309_cast_fp16 = slice_by_index(begin = var_18309_begin_0, end = var_18309_end_0, end_mask = var_18309_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18309_cast_fp16")]; + tensor var_18313_begin_0 = const()[name = tensor("op_18313_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_18313_end_0 = const()[name = tensor("op_18313_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_18313_end_mask_0 = const()[name = tensor("op_18313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18313_cast_fp16 = slice_by_index(begin = var_18313_begin_0, end = var_18313_end_0, end_mask = var_18313_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18313_cast_fp16")]; + tensor var_18317_begin_0 = const()[name = tensor("op_18317_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_18317_end_0 = const()[name = tensor("op_18317_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_18317_end_mask_0 = const()[name = tensor("op_18317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18317_cast_fp16 = slice_by_index(begin = var_18317_begin_0, end = var_18317_end_0, end_mask = var_18317_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18317_cast_fp16")]; + tensor var_18321_begin_0 = const()[name = tensor("op_18321_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_18321_end_0 = const()[name = tensor("op_18321_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_18321_end_mask_0 = const()[name = tensor("op_18321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18321_cast_fp16 = slice_by_index(begin = var_18321_begin_0, end = var_18321_end_0, end_mask = var_18321_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18321_cast_fp16")]; + tensor var_18325_begin_0 = const()[name = tensor("op_18325_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_18325_end_0 = const()[name = tensor("op_18325_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_18325_end_mask_0 = const()[name = tensor("op_18325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18325_cast_fp16 = slice_by_index(begin = var_18325_begin_0, end = var_18325_end_0, end_mask = var_18325_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18325_cast_fp16")]; + tensor var_18329_begin_0 = const()[name = tensor("op_18329_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_18329_end_0 = const()[name = tensor("op_18329_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_18329_end_mask_0 = const()[name = tensor("op_18329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18329_cast_fp16 = slice_by_index(begin = var_18329_begin_0, end = var_18329_end_0, end_mask = var_18329_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18329_cast_fp16")]; + tensor var_18333_begin_0 = const()[name = tensor("op_18333_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_18333_end_0 = const()[name = tensor("op_18333_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_18333_end_mask_0 = const()[name = tensor("op_18333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18333_cast_fp16 = slice_by_index(begin = var_18333_begin_0, end = var_18333_end_0, end_mask = var_18333_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18333_cast_fp16")]; + tensor var_18337_begin_0 = const()[name = tensor("op_18337_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_18337_end_0 = const()[name = tensor("op_18337_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_18337_end_mask_0 = const()[name = tensor("op_18337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18337_cast_fp16 = slice_by_index(begin = var_18337_begin_0, end = var_18337_end_0, end_mask = var_18337_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18337_cast_fp16")]; + tensor var_18341_begin_0 = const()[name = tensor("op_18341_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_18341_end_0 = const()[name = tensor("op_18341_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_18341_end_mask_0 = const()[name = tensor("op_18341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18341_cast_fp16 = slice_by_index(begin = var_18341_begin_0, end = var_18341_end_0, end_mask = var_18341_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18341_cast_fp16")]; + tensor var_18345_begin_0 = const()[name = tensor("op_18345_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_18345_end_0 = const()[name = tensor("op_18345_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_18345_end_mask_0 = const()[name = tensor("op_18345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18345_cast_fp16 = slice_by_index(begin = var_18345_begin_0, end = var_18345_end_0, end_mask = var_18345_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18345_cast_fp16")]; + tensor var_18349_begin_0 = const()[name = tensor("op_18349_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_18349_end_0 = const()[name = tensor("op_18349_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_18349_end_mask_0 = const()[name = tensor("op_18349_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18349_cast_fp16 = slice_by_index(begin = var_18349_begin_0, end = var_18349_end_0, end_mask = var_18349_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18349_cast_fp16")]; + tensor var_18353_begin_0 = const()[name = tensor("op_18353_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_18353_end_0 = const()[name = tensor("op_18353_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_18353_end_mask_0 = const()[name = tensor("op_18353_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18353_cast_fp16 = slice_by_index(begin = var_18353_begin_0, end = var_18353_end_0, end_mask = var_18353_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18353_cast_fp16")]; + tensor var_18357_begin_0 = const()[name = tensor("op_18357_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_18357_end_0 = const()[name = tensor("op_18357_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_18357_end_mask_0 = const()[name = tensor("op_18357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18357_cast_fp16 = slice_by_index(begin = var_18357_begin_0, end = var_18357_end_0, end_mask = var_18357_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18357_cast_fp16")]; + tensor var_18361_begin_0 = const()[name = tensor("op_18361_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_18361_end_0 = const()[name = tensor("op_18361_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_18361_end_mask_0 = const()[name = tensor("op_18361_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18361_cast_fp16 = slice_by_index(begin = var_18361_begin_0, end = var_18361_end_0, end_mask = var_18361_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18361_cast_fp16")]; + tensor var_18365_begin_0 = const()[name = tensor("op_18365_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_18365_end_0 = const()[name = tensor("op_18365_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_18365_end_mask_0 = const()[name = tensor("op_18365_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18365_cast_fp16 = slice_by_index(begin = var_18365_begin_0, end = var_18365_end_0, end_mask = var_18365_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18365_cast_fp16")]; + tensor var_18369_begin_0 = const()[name = tensor("op_18369_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_18369_end_0 = const()[name = tensor("op_18369_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_18369_end_mask_0 = const()[name = tensor("op_18369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_18369_cast_fp16 = slice_by_index(begin = var_18369_begin_0, end = var_18369_end_0, end_mask = var_18369_end_mask_0, x = query_23_cast_fp16)[name = tensor("op_18369_cast_fp16")]; + tensor var_18378_begin_0 = const()[name = tensor("op_18378_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18378_end_0 = const()[name = tensor("op_18378_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18378_end_mask_0 = const()[name = tensor("op_18378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18378_cast_fp16 = slice_by_index(begin = var_18378_begin_0, end = var_18378_end_0, end_mask = var_18378_end_mask_0, x = var_18293_cast_fp16)[name = tensor("op_18378_cast_fp16")]; + tensor var_18385_begin_0 = const()[name = tensor("op_18385_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18385_end_0 = const()[name = tensor("op_18385_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18385_end_mask_0 = const()[name = tensor("op_18385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18385_cast_fp16 = slice_by_index(begin = var_18385_begin_0, end = var_18385_end_0, end_mask = var_18385_end_mask_0, x = var_18293_cast_fp16)[name = tensor("op_18385_cast_fp16")]; + tensor var_18392_begin_0 = const()[name = tensor("op_18392_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18392_end_0 = const()[name = tensor("op_18392_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18392_end_mask_0 = const()[name = tensor("op_18392_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18392_cast_fp16 = slice_by_index(begin = var_18392_begin_0, end = var_18392_end_0, end_mask = var_18392_end_mask_0, x = var_18293_cast_fp16)[name = tensor("op_18392_cast_fp16")]; + tensor var_18399_begin_0 = const()[name = tensor("op_18399_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18399_end_0 = const()[name = tensor("op_18399_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18399_end_mask_0 = const()[name = tensor("op_18399_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18399_cast_fp16 = slice_by_index(begin = var_18399_begin_0, end = var_18399_end_0, end_mask = var_18399_end_mask_0, x = var_18293_cast_fp16)[name = tensor("op_18399_cast_fp16")]; + tensor var_18406_begin_0 = const()[name = tensor("op_18406_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18406_end_0 = const()[name = tensor("op_18406_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18406_end_mask_0 = const()[name = tensor("op_18406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18406_cast_fp16 = slice_by_index(begin = var_18406_begin_0, end = var_18406_end_0, end_mask = var_18406_end_mask_0, x = var_18297_cast_fp16)[name = tensor("op_18406_cast_fp16")]; + tensor var_18413_begin_0 = const()[name = tensor("op_18413_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18413_end_0 = const()[name = tensor("op_18413_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18413_end_mask_0 = const()[name = tensor("op_18413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18413_cast_fp16 = slice_by_index(begin = var_18413_begin_0, end = var_18413_end_0, end_mask = var_18413_end_mask_0, x = var_18297_cast_fp16)[name = tensor("op_18413_cast_fp16")]; + tensor var_18420_begin_0 = const()[name = tensor("op_18420_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18420_end_0 = const()[name = tensor("op_18420_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18420_end_mask_0 = const()[name = tensor("op_18420_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18420_cast_fp16 = slice_by_index(begin = var_18420_begin_0, end = var_18420_end_0, end_mask = var_18420_end_mask_0, x = var_18297_cast_fp16)[name = tensor("op_18420_cast_fp16")]; + tensor var_18427_begin_0 = const()[name = tensor("op_18427_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18427_end_0 = const()[name = tensor("op_18427_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18427_end_mask_0 = const()[name = tensor("op_18427_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18427_cast_fp16 = slice_by_index(begin = var_18427_begin_0, end = var_18427_end_0, end_mask = var_18427_end_mask_0, x = var_18297_cast_fp16)[name = tensor("op_18427_cast_fp16")]; + tensor var_18434_begin_0 = const()[name = tensor("op_18434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18434_end_0 = const()[name = tensor("op_18434_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18434_end_mask_0 = const()[name = tensor("op_18434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18434_cast_fp16 = slice_by_index(begin = var_18434_begin_0, end = var_18434_end_0, end_mask = var_18434_end_mask_0, x = var_18301_cast_fp16)[name = tensor("op_18434_cast_fp16")]; + tensor var_18441_begin_0 = const()[name = tensor("op_18441_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18441_end_0 = const()[name = tensor("op_18441_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18441_end_mask_0 = const()[name = tensor("op_18441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18441_cast_fp16 = slice_by_index(begin = var_18441_begin_0, end = var_18441_end_0, end_mask = var_18441_end_mask_0, x = var_18301_cast_fp16)[name = tensor("op_18441_cast_fp16")]; + tensor var_18448_begin_0 = const()[name = tensor("op_18448_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18448_end_0 = const()[name = tensor("op_18448_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18448_end_mask_0 = const()[name = tensor("op_18448_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18448_cast_fp16 = slice_by_index(begin = var_18448_begin_0, end = var_18448_end_0, end_mask = var_18448_end_mask_0, x = var_18301_cast_fp16)[name = tensor("op_18448_cast_fp16")]; + tensor var_18455_begin_0 = const()[name = tensor("op_18455_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18455_end_0 = const()[name = tensor("op_18455_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18455_end_mask_0 = const()[name = tensor("op_18455_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18455_cast_fp16 = slice_by_index(begin = var_18455_begin_0, end = var_18455_end_0, end_mask = var_18455_end_mask_0, x = var_18301_cast_fp16)[name = tensor("op_18455_cast_fp16")]; + tensor var_18462_begin_0 = const()[name = tensor("op_18462_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18462_end_0 = const()[name = tensor("op_18462_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18462_end_mask_0 = const()[name = tensor("op_18462_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18462_cast_fp16 = slice_by_index(begin = var_18462_begin_0, end = var_18462_end_0, end_mask = var_18462_end_mask_0, x = var_18305_cast_fp16)[name = tensor("op_18462_cast_fp16")]; + tensor var_18469_begin_0 = const()[name = tensor("op_18469_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18469_end_0 = const()[name = tensor("op_18469_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18469_end_mask_0 = const()[name = tensor("op_18469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18469_cast_fp16 = slice_by_index(begin = var_18469_begin_0, end = var_18469_end_0, end_mask = var_18469_end_mask_0, x = var_18305_cast_fp16)[name = tensor("op_18469_cast_fp16")]; + tensor var_18476_begin_0 = const()[name = tensor("op_18476_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18476_end_0 = const()[name = tensor("op_18476_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18476_end_mask_0 = const()[name = tensor("op_18476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18476_cast_fp16 = slice_by_index(begin = var_18476_begin_0, end = var_18476_end_0, end_mask = var_18476_end_mask_0, x = var_18305_cast_fp16)[name = tensor("op_18476_cast_fp16")]; + tensor var_18483_begin_0 = const()[name = tensor("op_18483_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18483_end_0 = const()[name = tensor("op_18483_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18483_end_mask_0 = const()[name = tensor("op_18483_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18483_cast_fp16 = slice_by_index(begin = var_18483_begin_0, end = var_18483_end_0, end_mask = var_18483_end_mask_0, x = var_18305_cast_fp16)[name = tensor("op_18483_cast_fp16")]; + tensor var_18490_begin_0 = const()[name = tensor("op_18490_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18490_end_0 = const()[name = tensor("op_18490_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18490_end_mask_0 = const()[name = tensor("op_18490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18490_cast_fp16 = slice_by_index(begin = var_18490_begin_0, end = var_18490_end_0, end_mask = var_18490_end_mask_0, x = var_18309_cast_fp16)[name = tensor("op_18490_cast_fp16")]; + tensor var_18497_begin_0 = const()[name = tensor("op_18497_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18497_end_0 = const()[name = tensor("op_18497_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18497_end_mask_0 = const()[name = tensor("op_18497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18497_cast_fp16 = slice_by_index(begin = var_18497_begin_0, end = var_18497_end_0, end_mask = var_18497_end_mask_0, x = var_18309_cast_fp16)[name = tensor("op_18497_cast_fp16")]; + tensor var_18504_begin_0 = const()[name = tensor("op_18504_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18504_end_0 = const()[name = tensor("op_18504_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18504_end_mask_0 = const()[name = tensor("op_18504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18504_cast_fp16 = slice_by_index(begin = var_18504_begin_0, end = var_18504_end_0, end_mask = var_18504_end_mask_0, x = var_18309_cast_fp16)[name = tensor("op_18504_cast_fp16")]; + tensor var_18511_begin_0 = const()[name = tensor("op_18511_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18511_end_0 = const()[name = tensor("op_18511_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18511_end_mask_0 = const()[name = tensor("op_18511_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18511_cast_fp16 = slice_by_index(begin = var_18511_begin_0, end = var_18511_end_0, end_mask = var_18511_end_mask_0, x = var_18309_cast_fp16)[name = tensor("op_18511_cast_fp16")]; + tensor var_18518_begin_0 = const()[name = tensor("op_18518_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18518_end_0 = const()[name = tensor("op_18518_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18518_end_mask_0 = const()[name = tensor("op_18518_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18518_cast_fp16 = slice_by_index(begin = var_18518_begin_0, end = var_18518_end_0, end_mask = var_18518_end_mask_0, x = var_18313_cast_fp16)[name = tensor("op_18518_cast_fp16")]; + tensor var_18525_begin_0 = const()[name = tensor("op_18525_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18525_end_0 = const()[name = tensor("op_18525_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18525_end_mask_0 = const()[name = tensor("op_18525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18525_cast_fp16 = slice_by_index(begin = var_18525_begin_0, end = var_18525_end_0, end_mask = var_18525_end_mask_0, x = var_18313_cast_fp16)[name = tensor("op_18525_cast_fp16")]; + tensor var_18532_begin_0 = const()[name = tensor("op_18532_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18532_end_0 = const()[name = tensor("op_18532_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18532_end_mask_0 = const()[name = tensor("op_18532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18532_cast_fp16 = slice_by_index(begin = var_18532_begin_0, end = var_18532_end_0, end_mask = var_18532_end_mask_0, x = var_18313_cast_fp16)[name = tensor("op_18532_cast_fp16")]; + tensor var_18539_begin_0 = const()[name = tensor("op_18539_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18539_end_0 = const()[name = tensor("op_18539_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18539_end_mask_0 = const()[name = tensor("op_18539_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18539_cast_fp16 = slice_by_index(begin = var_18539_begin_0, end = var_18539_end_0, end_mask = var_18539_end_mask_0, x = var_18313_cast_fp16)[name = tensor("op_18539_cast_fp16")]; + tensor var_18546_begin_0 = const()[name = tensor("op_18546_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18546_end_0 = const()[name = tensor("op_18546_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18546_end_mask_0 = const()[name = tensor("op_18546_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18546_cast_fp16 = slice_by_index(begin = var_18546_begin_0, end = var_18546_end_0, end_mask = var_18546_end_mask_0, x = var_18317_cast_fp16)[name = tensor("op_18546_cast_fp16")]; + tensor var_18553_begin_0 = const()[name = tensor("op_18553_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18553_end_0 = const()[name = tensor("op_18553_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18553_end_mask_0 = const()[name = tensor("op_18553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18553_cast_fp16 = slice_by_index(begin = var_18553_begin_0, end = var_18553_end_0, end_mask = var_18553_end_mask_0, x = var_18317_cast_fp16)[name = tensor("op_18553_cast_fp16")]; + tensor var_18560_begin_0 = const()[name = tensor("op_18560_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18560_end_0 = const()[name = tensor("op_18560_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18560_end_mask_0 = const()[name = tensor("op_18560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18560_cast_fp16 = slice_by_index(begin = var_18560_begin_0, end = var_18560_end_0, end_mask = var_18560_end_mask_0, x = var_18317_cast_fp16)[name = tensor("op_18560_cast_fp16")]; + tensor var_18567_begin_0 = const()[name = tensor("op_18567_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18567_end_0 = const()[name = tensor("op_18567_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18567_end_mask_0 = const()[name = tensor("op_18567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18567_cast_fp16 = slice_by_index(begin = var_18567_begin_0, end = var_18567_end_0, end_mask = var_18567_end_mask_0, x = var_18317_cast_fp16)[name = tensor("op_18567_cast_fp16")]; + tensor var_18574_begin_0 = const()[name = tensor("op_18574_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18574_end_0 = const()[name = tensor("op_18574_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18574_end_mask_0 = const()[name = tensor("op_18574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18574_cast_fp16 = slice_by_index(begin = var_18574_begin_0, end = var_18574_end_0, end_mask = var_18574_end_mask_0, x = var_18321_cast_fp16)[name = tensor("op_18574_cast_fp16")]; + tensor var_18581_begin_0 = const()[name = tensor("op_18581_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18581_end_0 = const()[name = tensor("op_18581_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18581_end_mask_0 = const()[name = tensor("op_18581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18581_cast_fp16 = slice_by_index(begin = var_18581_begin_0, end = var_18581_end_0, end_mask = var_18581_end_mask_0, x = var_18321_cast_fp16)[name = tensor("op_18581_cast_fp16")]; + tensor var_18588_begin_0 = const()[name = tensor("op_18588_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18588_end_0 = const()[name = tensor("op_18588_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18588_end_mask_0 = const()[name = tensor("op_18588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18588_cast_fp16 = slice_by_index(begin = var_18588_begin_0, end = var_18588_end_0, end_mask = var_18588_end_mask_0, x = var_18321_cast_fp16)[name = tensor("op_18588_cast_fp16")]; + tensor var_18595_begin_0 = const()[name = tensor("op_18595_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18595_end_0 = const()[name = tensor("op_18595_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18595_end_mask_0 = const()[name = tensor("op_18595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18595_cast_fp16 = slice_by_index(begin = var_18595_begin_0, end = var_18595_end_0, end_mask = var_18595_end_mask_0, x = var_18321_cast_fp16)[name = tensor("op_18595_cast_fp16")]; + tensor var_18602_begin_0 = const()[name = tensor("op_18602_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18602_end_0 = const()[name = tensor("op_18602_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18602_end_mask_0 = const()[name = tensor("op_18602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18602_cast_fp16 = slice_by_index(begin = var_18602_begin_0, end = var_18602_end_0, end_mask = var_18602_end_mask_0, x = var_18325_cast_fp16)[name = tensor("op_18602_cast_fp16")]; + tensor var_18609_begin_0 = const()[name = tensor("op_18609_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18609_end_0 = const()[name = tensor("op_18609_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18609_end_mask_0 = const()[name = tensor("op_18609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18609_cast_fp16 = slice_by_index(begin = var_18609_begin_0, end = var_18609_end_0, end_mask = var_18609_end_mask_0, x = var_18325_cast_fp16)[name = tensor("op_18609_cast_fp16")]; + tensor var_18616_begin_0 = const()[name = tensor("op_18616_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18616_end_0 = const()[name = tensor("op_18616_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18616_end_mask_0 = const()[name = tensor("op_18616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18616_cast_fp16 = slice_by_index(begin = var_18616_begin_0, end = var_18616_end_0, end_mask = var_18616_end_mask_0, x = var_18325_cast_fp16)[name = tensor("op_18616_cast_fp16")]; + tensor var_18623_begin_0 = const()[name = tensor("op_18623_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18623_end_0 = const()[name = tensor("op_18623_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18623_end_mask_0 = const()[name = tensor("op_18623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18623_cast_fp16 = slice_by_index(begin = var_18623_begin_0, end = var_18623_end_0, end_mask = var_18623_end_mask_0, x = var_18325_cast_fp16)[name = tensor("op_18623_cast_fp16")]; + tensor var_18630_begin_0 = const()[name = tensor("op_18630_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18630_end_0 = const()[name = tensor("op_18630_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18630_end_mask_0 = const()[name = tensor("op_18630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18630_cast_fp16 = slice_by_index(begin = var_18630_begin_0, end = var_18630_end_0, end_mask = var_18630_end_mask_0, x = var_18329_cast_fp16)[name = tensor("op_18630_cast_fp16")]; + tensor var_18637_begin_0 = const()[name = tensor("op_18637_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18637_end_0 = const()[name = tensor("op_18637_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18637_end_mask_0 = const()[name = tensor("op_18637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18637_cast_fp16 = slice_by_index(begin = var_18637_begin_0, end = var_18637_end_0, end_mask = var_18637_end_mask_0, x = var_18329_cast_fp16)[name = tensor("op_18637_cast_fp16")]; + tensor var_18644_begin_0 = const()[name = tensor("op_18644_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18644_end_0 = const()[name = tensor("op_18644_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18644_end_mask_0 = const()[name = tensor("op_18644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18644_cast_fp16 = slice_by_index(begin = var_18644_begin_0, end = var_18644_end_0, end_mask = var_18644_end_mask_0, x = var_18329_cast_fp16)[name = tensor("op_18644_cast_fp16")]; + tensor var_18651_begin_0 = const()[name = tensor("op_18651_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18651_end_0 = const()[name = tensor("op_18651_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18651_end_mask_0 = const()[name = tensor("op_18651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18651_cast_fp16 = slice_by_index(begin = var_18651_begin_0, end = var_18651_end_0, end_mask = var_18651_end_mask_0, x = var_18329_cast_fp16)[name = tensor("op_18651_cast_fp16")]; + tensor var_18658_begin_0 = const()[name = tensor("op_18658_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18658_end_0 = const()[name = tensor("op_18658_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18658_end_mask_0 = const()[name = tensor("op_18658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18658_cast_fp16 = slice_by_index(begin = var_18658_begin_0, end = var_18658_end_0, end_mask = var_18658_end_mask_0, x = var_18333_cast_fp16)[name = tensor("op_18658_cast_fp16")]; + tensor var_18665_begin_0 = const()[name = tensor("op_18665_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18665_end_0 = const()[name = tensor("op_18665_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18665_end_mask_0 = const()[name = tensor("op_18665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18665_cast_fp16 = slice_by_index(begin = var_18665_begin_0, end = var_18665_end_0, end_mask = var_18665_end_mask_0, x = var_18333_cast_fp16)[name = tensor("op_18665_cast_fp16")]; + tensor var_18672_begin_0 = const()[name = tensor("op_18672_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18672_end_0 = const()[name = tensor("op_18672_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18672_end_mask_0 = const()[name = tensor("op_18672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18672_cast_fp16 = slice_by_index(begin = var_18672_begin_0, end = var_18672_end_0, end_mask = var_18672_end_mask_0, x = var_18333_cast_fp16)[name = tensor("op_18672_cast_fp16")]; + tensor var_18679_begin_0 = const()[name = tensor("op_18679_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18679_end_0 = const()[name = tensor("op_18679_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18679_end_mask_0 = const()[name = tensor("op_18679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18679_cast_fp16 = slice_by_index(begin = var_18679_begin_0, end = var_18679_end_0, end_mask = var_18679_end_mask_0, x = var_18333_cast_fp16)[name = tensor("op_18679_cast_fp16")]; + tensor var_18686_begin_0 = const()[name = tensor("op_18686_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18686_end_0 = const()[name = tensor("op_18686_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18686_end_mask_0 = const()[name = tensor("op_18686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18686_cast_fp16 = slice_by_index(begin = var_18686_begin_0, end = var_18686_end_0, end_mask = var_18686_end_mask_0, x = var_18337_cast_fp16)[name = tensor("op_18686_cast_fp16")]; + tensor var_18693_begin_0 = const()[name = tensor("op_18693_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18693_end_0 = const()[name = tensor("op_18693_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18693_end_mask_0 = const()[name = tensor("op_18693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18693_cast_fp16 = slice_by_index(begin = var_18693_begin_0, end = var_18693_end_0, end_mask = var_18693_end_mask_0, x = var_18337_cast_fp16)[name = tensor("op_18693_cast_fp16")]; + tensor var_18700_begin_0 = const()[name = tensor("op_18700_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18700_end_0 = const()[name = tensor("op_18700_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18700_end_mask_0 = const()[name = tensor("op_18700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18700_cast_fp16 = slice_by_index(begin = var_18700_begin_0, end = var_18700_end_0, end_mask = var_18700_end_mask_0, x = var_18337_cast_fp16)[name = tensor("op_18700_cast_fp16")]; + tensor var_18707_begin_0 = const()[name = tensor("op_18707_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18707_end_0 = const()[name = tensor("op_18707_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18707_end_mask_0 = const()[name = tensor("op_18707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18707_cast_fp16 = slice_by_index(begin = var_18707_begin_0, end = var_18707_end_0, end_mask = var_18707_end_mask_0, x = var_18337_cast_fp16)[name = tensor("op_18707_cast_fp16")]; + tensor var_18714_begin_0 = const()[name = tensor("op_18714_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18714_end_0 = const()[name = tensor("op_18714_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18714_end_mask_0 = const()[name = tensor("op_18714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18714_cast_fp16 = slice_by_index(begin = var_18714_begin_0, end = var_18714_end_0, end_mask = var_18714_end_mask_0, x = var_18341_cast_fp16)[name = tensor("op_18714_cast_fp16")]; + tensor var_18721_begin_0 = const()[name = tensor("op_18721_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18721_end_0 = const()[name = tensor("op_18721_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18721_end_mask_0 = const()[name = tensor("op_18721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18721_cast_fp16 = slice_by_index(begin = var_18721_begin_0, end = var_18721_end_0, end_mask = var_18721_end_mask_0, x = var_18341_cast_fp16)[name = tensor("op_18721_cast_fp16")]; + tensor var_18728_begin_0 = const()[name = tensor("op_18728_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18728_end_0 = const()[name = tensor("op_18728_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18728_end_mask_0 = const()[name = tensor("op_18728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18728_cast_fp16 = slice_by_index(begin = var_18728_begin_0, end = var_18728_end_0, end_mask = var_18728_end_mask_0, x = var_18341_cast_fp16)[name = tensor("op_18728_cast_fp16")]; + tensor var_18735_begin_0 = const()[name = tensor("op_18735_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18735_end_0 = const()[name = tensor("op_18735_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18735_end_mask_0 = const()[name = tensor("op_18735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18735_cast_fp16 = slice_by_index(begin = var_18735_begin_0, end = var_18735_end_0, end_mask = var_18735_end_mask_0, x = var_18341_cast_fp16)[name = tensor("op_18735_cast_fp16")]; + tensor var_18742_begin_0 = const()[name = tensor("op_18742_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18742_end_0 = const()[name = tensor("op_18742_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18742_end_mask_0 = const()[name = tensor("op_18742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18742_cast_fp16 = slice_by_index(begin = var_18742_begin_0, end = var_18742_end_0, end_mask = var_18742_end_mask_0, x = var_18345_cast_fp16)[name = tensor("op_18742_cast_fp16")]; + tensor var_18749_begin_0 = const()[name = tensor("op_18749_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18749_end_0 = const()[name = tensor("op_18749_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18749_end_mask_0 = const()[name = tensor("op_18749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18749_cast_fp16 = slice_by_index(begin = var_18749_begin_0, end = var_18749_end_0, end_mask = var_18749_end_mask_0, x = var_18345_cast_fp16)[name = tensor("op_18749_cast_fp16")]; + tensor var_18756_begin_0 = const()[name = tensor("op_18756_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18756_end_0 = const()[name = tensor("op_18756_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18756_end_mask_0 = const()[name = tensor("op_18756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18756_cast_fp16 = slice_by_index(begin = var_18756_begin_0, end = var_18756_end_0, end_mask = var_18756_end_mask_0, x = var_18345_cast_fp16)[name = tensor("op_18756_cast_fp16")]; + tensor var_18763_begin_0 = const()[name = tensor("op_18763_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18763_end_0 = const()[name = tensor("op_18763_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18763_end_mask_0 = const()[name = tensor("op_18763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18763_cast_fp16 = slice_by_index(begin = var_18763_begin_0, end = var_18763_end_0, end_mask = var_18763_end_mask_0, x = var_18345_cast_fp16)[name = tensor("op_18763_cast_fp16")]; + tensor var_18770_begin_0 = const()[name = tensor("op_18770_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18770_end_0 = const()[name = tensor("op_18770_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18770_end_mask_0 = const()[name = tensor("op_18770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18770_cast_fp16 = slice_by_index(begin = var_18770_begin_0, end = var_18770_end_0, end_mask = var_18770_end_mask_0, x = var_18349_cast_fp16)[name = tensor("op_18770_cast_fp16")]; + tensor var_18777_begin_0 = const()[name = tensor("op_18777_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18777_end_0 = const()[name = tensor("op_18777_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18777_end_mask_0 = const()[name = tensor("op_18777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18777_cast_fp16 = slice_by_index(begin = var_18777_begin_0, end = var_18777_end_0, end_mask = var_18777_end_mask_0, x = var_18349_cast_fp16)[name = tensor("op_18777_cast_fp16")]; + tensor var_18784_begin_0 = const()[name = tensor("op_18784_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18784_end_0 = const()[name = tensor("op_18784_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18784_end_mask_0 = const()[name = tensor("op_18784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18784_cast_fp16 = slice_by_index(begin = var_18784_begin_0, end = var_18784_end_0, end_mask = var_18784_end_mask_0, x = var_18349_cast_fp16)[name = tensor("op_18784_cast_fp16")]; + tensor var_18791_begin_0 = const()[name = tensor("op_18791_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18791_end_0 = const()[name = tensor("op_18791_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18791_end_mask_0 = const()[name = tensor("op_18791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18791_cast_fp16 = slice_by_index(begin = var_18791_begin_0, end = var_18791_end_0, end_mask = var_18791_end_mask_0, x = var_18349_cast_fp16)[name = tensor("op_18791_cast_fp16")]; + tensor var_18798_begin_0 = const()[name = tensor("op_18798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18798_end_0 = const()[name = tensor("op_18798_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18798_end_mask_0 = const()[name = tensor("op_18798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18798_cast_fp16 = slice_by_index(begin = var_18798_begin_0, end = var_18798_end_0, end_mask = var_18798_end_mask_0, x = var_18353_cast_fp16)[name = tensor("op_18798_cast_fp16")]; + tensor var_18805_begin_0 = const()[name = tensor("op_18805_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18805_end_0 = const()[name = tensor("op_18805_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18805_end_mask_0 = const()[name = tensor("op_18805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18805_cast_fp16 = slice_by_index(begin = var_18805_begin_0, end = var_18805_end_0, end_mask = var_18805_end_mask_0, x = var_18353_cast_fp16)[name = tensor("op_18805_cast_fp16")]; + tensor var_18812_begin_0 = const()[name = tensor("op_18812_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18812_end_0 = const()[name = tensor("op_18812_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18812_end_mask_0 = const()[name = tensor("op_18812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18812_cast_fp16 = slice_by_index(begin = var_18812_begin_0, end = var_18812_end_0, end_mask = var_18812_end_mask_0, x = var_18353_cast_fp16)[name = tensor("op_18812_cast_fp16")]; + tensor var_18819_begin_0 = const()[name = tensor("op_18819_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18819_end_0 = const()[name = tensor("op_18819_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18819_end_mask_0 = const()[name = tensor("op_18819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18819_cast_fp16 = slice_by_index(begin = var_18819_begin_0, end = var_18819_end_0, end_mask = var_18819_end_mask_0, x = var_18353_cast_fp16)[name = tensor("op_18819_cast_fp16")]; + tensor var_18826_begin_0 = const()[name = tensor("op_18826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18826_end_0 = const()[name = tensor("op_18826_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18826_end_mask_0 = const()[name = tensor("op_18826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18826_cast_fp16 = slice_by_index(begin = var_18826_begin_0, end = var_18826_end_0, end_mask = var_18826_end_mask_0, x = var_18357_cast_fp16)[name = tensor("op_18826_cast_fp16")]; + tensor var_18833_begin_0 = const()[name = tensor("op_18833_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18833_end_0 = const()[name = tensor("op_18833_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18833_end_mask_0 = const()[name = tensor("op_18833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18833_cast_fp16 = slice_by_index(begin = var_18833_begin_0, end = var_18833_end_0, end_mask = var_18833_end_mask_0, x = var_18357_cast_fp16)[name = tensor("op_18833_cast_fp16")]; + tensor var_18840_begin_0 = const()[name = tensor("op_18840_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18840_end_0 = const()[name = tensor("op_18840_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18840_end_mask_0 = const()[name = tensor("op_18840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18840_cast_fp16 = slice_by_index(begin = var_18840_begin_0, end = var_18840_end_0, end_mask = var_18840_end_mask_0, x = var_18357_cast_fp16)[name = tensor("op_18840_cast_fp16")]; + tensor var_18847_begin_0 = const()[name = tensor("op_18847_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18847_end_0 = const()[name = tensor("op_18847_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18847_end_mask_0 = const()[name = tensor("op_18847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18847_cast_fp16 = slice_by_index(begin = var_18847_begin_0, end = var_18847_end_0, end_mask = var_18847_end_mask_0, x = var_18357_cast_fp16)[name = tensor("op_18847_cast_fp16")]; + tensor var_18854_begin_0 = const()[name = tensor("op_18854_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18854_end_0 = const()[name = tensor("op_18854_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18854_end_mask_0 = const()[name = tensor("op_18854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18854_cast_fp16 = slice_by_index(begin = var_18854_begin_0, end = var_18854_end_0, end_mask = var_18854_end_mask_0, x = var_18361_cast_fp16)[name = tensor("op_18854_cast_fp16")]; + tensor var_18861_begin_0 = const()[name = tensor("op_18861_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18861_end_0 = const()[name = tensor("op_18861_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18861_end_mask_0 = const()[name = tensor("op_18861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18861_cast_fp16 = slice_by_index(begin = var_18861_begin_0, end = var_18861_end_0, end_mask = var_18861_end_mask_0, x = var_18361_cast_fp16)[name = tensor("op_18861_cast_fp16")]; + tensor var_18868_begin_0 = const()[name = tensor("op_18868_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18868_end_0 = const()[name = tensor("op_18868_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18868_end_mask_0 = const()[name = tensor("op_18868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18868_cast_fp16 = slice_by_index(begin = var_18868_begin_0, end = var_18868_end_0, end_mask = var_18868_end_mask_0, x = var_18361_cast_fp16)[name = tensor("op_18868_cast_fp16")]; + tensor var_18875_begin_0 = const()[name = tensor("op_18875_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18875_end_0 = const()[name = tensor("op_18875_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18875_end_mask_0 = const()[name = tensor("op_18875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18875_cast_fp16 = slice_by_index(begin = var_18875_begin_0, end = var_18875_end_0, end_mask = var_18875_end_mask_0, x = var_18361_cast_fp16)[name = tensor("op_18875_cast_fp16")]; + tensor var_18882_begin_0 = const()[name = tensor("op_18882_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18882_end_0 = const()[name = tensor("op_18882_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18882_end_mask_0 = const()[name = tensor("op_18882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18882_cast_fp16 = slice_by_index(begin = var_18882_begin_0, end = var_18882_end_0, end_mask = var_18882_end_mask_0, x = var_18365_cast_fp16)[name = tensor("op_18882_cast_fp16")]; + tensor var_18889_begin_0 = const()[name = tensor("op_18889_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18889_end_0 = const()[name = tensor("op_18889_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18889_end_mask_0 = const()[name = tensor("op_18889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18889_cast_fp16 = slice_by_index(begin = var_18889_begin_0, end = var_18889_end_0, end_mask = var_18889_end_mask_0, x = var_18365_cast_fp16)[name = tensor("op_18889_cast_fp16")]; + tensor var_18896_begin_0 = const()[name = tensor("op_18896_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18896_end_0 = const()[name = tensor("op_18896_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18896_end_mask_0 = const()[name = tensor("op_18896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18896_cast_fp16 = slice_by_index(begin = var_18896_begin_0, end = var_18896_end_0, end_mask = var_18896_end_mask_0, x = var_18365_cast_fp16)[name = tensor("op_18896_cast_fp16")]; + tensor var_18903_begin_0 = const()[name = tensor("op_18903_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18903_end_0 = const()[name = tensor("op_18903_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18903_end_mask_0 = const()[name = tensor("op_18903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18903_cast_fp16 = slice_by_index(begin = var_18903_begin_0, end = var_18903_end_0, end_mask = var_18903_end_mask_0, x = var_18365_cast_fp16)[name = tensor("op_18903_cast_fp16")]; + tensor var_18910_begin_0 = const()[name = tensor("op_18910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18910_end_0 = const()[name = tensor("op_18910_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_18910_end_mask_0 = const()[name = tensor("op_18910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18910_cast_fp16 = slice_by_index(begin = var_18910_begin_0, end = var_18910_end_0, end_mask = var_18910_end_mask_0, x = var_18369_cast_fp16)[name = tensor("op_18910_cast_fp16")]; + tensor var_18917_begin_0 = const()[name = tensor("op_18917_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_18917_end_0 = const()[name = tensor("op_18917_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_18917_end_mask_0 = const()[name = tensor("op_18917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18917_cast_fp16 = slice_by_index(begin = var_18917_begin_0, end = var_18917_end_0, end_mask = var_18917_end_mask_0, x = var_18369_cast_fp16)[name = tensor("op_18917_cast_fp16")]; + tensor var_18924_begin_0 = const()[name = tensor("op_18924_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_18924_end_0 = const()[name = tensor("op_18924_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_18924_end_mask_0 = const()[name = tensor("op_18924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18924_cast_fp16 = slice_by_index(begin = var_18924_begin_0, end = var_18924_end_0, end_mask = var_18924_end_mask_0, x = var_18369_cast_fp16)[name = tensor("op_18924_cast_fp16")]; + tensor var_18931_begin_0 = const()[name = tensor("op_18931_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_18931_end_0 = const()[name = tensor("op_18931_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_18931_end_mask_0 = const()[name = tensor("op_18931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18931_cast_fp16 = slice_by_index(begin = var_18931_begin_0, end = var_18931_end_0, end_mask = var_18931_end_mask_0, x = var_18369_cast_fp16)[name = tensor("op_18931_cast_fp16")]; + tensor k_23_perm_0 = const()[name = tensor("k_23_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_18936_begin_0 = const()[name = tensor("op_18936_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_18936_end_0 = const()[name = tensor("op_18936_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_18936_end_mask_0 = const()[name = tensor("op_18936_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_20 = transpose(perm = k_23_perm_0, x = key_23_cast_fp16)[name = tensor("transpose_20")]; + tensor var_18936_cast_fp16 = slice_by_index(begin = var_18936_begin_0, end = var_18936_end_0, end_mask = var_18936_end_mask_0, x = transpose_20)[name = tensor("op_18936_cast_fp16")]; + tensor var_18940_begin_0 = const()[name = tensor("op_18940_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_18940_end_0 = const()[name = tensor("op_18940_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_18940_end_mask_0 = const()[name = tensor("op_18940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18940_cast_fp16 = slice_by_index(begin = var_18940_begin_0, end = var_18940_end_0, end_mask = var_18940_end_mask_0, x = transpose_20)[name = tensor("op_18940_cast_fp16")]; + tensor var_18944_begin_0 = const()[name = tensor("op_18944_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_18944_end_0 = const()[name = tensor("op_18944_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_18944_end_mask_0 = const()[name = tensor("op_18944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18944_cast_fp16 = slice_by_index(begin = var_18944_begin_0, end = var_18944_end_0, end_mask = var_18944_end_mask_0, x = transpose_20)[name = tensor("op_18944_cast_fp16")]; + tensor var_18948_begin_0 = const()[name = tensor("op_18948_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_18948_end_0 = const()[name = tensor("op_18948_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_18948_end_mask_0 = const()[name = tensor("op_18948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18948_cast_fp16 = slice_by_index(begin = var_18948_begin_0, end = var_18948_end_0, end_mask = var_18948_end_mask_0, x = transpose_20)[name = tensor("op_18948_cast_fp16")]; + tensor var_18952_begin_0 = const()[name = tensor("op_18952_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_18952_end_0 = const()[name = tensor("op_18952_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_18952_end_mask_0 = const()[name = tensor("op_18952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18952_cast_fp16 = slice_by_index(begin = var_18952_begin_0, end = var_18952_end_0, end_mask = var_18952_end_mask_0, x = transpose_20)[name = tensor("op_18952_cast_fp16")]; + tensor var_18956_begin_0 = const()[name = tensor("op_18956_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_18956_end_0 = const()[name = tensor("op_18956_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_18956_end_mask_0 = const()[name = tensor("op_18956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18956_cast_fp16 = slice_by_index(begin = var_18956_begin_0, end = var_18956_end_0, end_mask = var_18956_end_mask_0, x = transpose_20)[name = tensor("op_18956_cast_fp16")]; + tensor var_18960_begin_0 = const()[name = tensor("op_18960_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_18960_end_0 = const()[name = tensor("op_18960_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_18960_end_mask_0 = const()[name = tensor("op_18960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18960_cast_fp16 = slice_by_index(begin = var_18960_begin_0, end = var_18960_end_0, end_mask = var_18960_end_mask_0, x = transpose_20)[name = tensor("op_18960_cast_fp16")]; + tensor var_18964_begin_0 = const()[name = tensor("op_18964_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_18964_end_0 = const()[name = tensor("op_18964_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_18964_end_mask_0 = const()[name = tensor("op_18964_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18964_cast_fp16 = slice_by_index(begin = var_18964_begin_0, end = var_18964_end_0, end_mask = var_18964_end_mask_0, x = transpose_20)[name = tensor("op_18964_cast_fp16")]; + tensor var_18968_begin_0 = const()[name = tensor("op_18968_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_18968_end_0 = const()[name = tensor("op_18968_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_18968_end_mask_0 = const()[name = tensor("op_18968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18968_cast_fp16 = slice_by_index(begin = var_18968_begin_0, end = var_18968_end_0, end_mask = var_18968_end_mask_0, x = transpose_20)[name = tensor("op_18968_cast_fp16")]; + tensor var_18972_begin_0 = const()[name = tensor("op_18972_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_18972_end_0 = const()[name = tensor("op_18972_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_18972_end_mask_0 = const()[name = tensor("op_18972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18972_cast_fp16 = slice_by_index(begin = var_18972_begin_0, end = var_18972_end_0, end_mask = var_18972_end_mask_0, x = transpose_20)[name = tensor("op_18972_cast_fp16")]; + tensor var_18976_begin_0 = const()[name = tensor("op_18976_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_18976_end_0 = const()[name = tensor("op_18976_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_18976_end_mask_0 = const()[name = tensor("op_18976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18976_cast_fp16 = slice_by_index(begin = var_18976_begin_0, end = var_18976_end_0, end_mask = var_18976_end_mask_0, x = transpose_20)[name = tensor("op_18976_cast_fp16")]; + tensor var_18980_begin_0 = const()[name = tensor("op_18980_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_18980_end_0 = const()[name = tensor("op_18980_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_18980_end_mask_0 = const()[name = tensor("op_18980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18980_cast_fp16 = slice_by_index(begin = var_18980_begin_0, end = var_18980_end_0, end_mask = var_18980_end_mask_0, x = transpose_20)[name = tensor("op_18980_cast_fp16")]; + tensor var_18984_begin_0 = const()[name = tensor("op_18984_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_18984_end_0 = const()[name = tensor("op_18984_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_18984_end_mask_0 = const()[name = tensor("op_18984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18984_cast_fp16 = slice_by_index(begin = var_18984_begin_0, end = var_18984_end_0, end_mask = var_18984_end_mask_0, x = transpose_20)[name = tensor("op_18984_cast_fp16")]; + tensor var_18988_begin_0 = const()[name = tensor("op_18988_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_18988_end_0 = const()[name = tensor("op_18988_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_18988_end_mask_0 = const()[name = tensor("op_18988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18988_cast_fp16 = slice_by_index(begin = var_18988_begin_0, end = var_18988_end_0, end_mask = var_18988_end_mask_0, x = transpose_20)[name = tensor("op_18988_cast_fp16")]; + tensor var_18992_begin_0 = const()[name = tensor("op_18992_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_18992_end_0 = const()[name = tensor("op_18992_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_18992_end_mask_0 = const()[name = tensor("op_18992_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18992_cast_fp16 = slice_by_index(begin = var_18992_begin_0, end = var_18992_end_0, end_mask = var_18992_end_mask_0, x = transpose_20)[name = tensor("op_18992_cast_fp16")]; + tensor var_18996_begin_0 = const()[name = tensor("op_18996_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_18996_end_0 = const()[name = tensor("op_18996_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_18996_end_mask_0 = const()[name = tensor("op_18996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_18996_cast_fp16 = slice_by_index(begin = var_18996_begin_0, end = var_18996_end_0, end_mask = var_18996_end_mask_0, x = transpose_20)[name = tensor("op_18996_cast_fp16")]; + tensor var_19000_begin_0 = const()[name = tensor("op_19000_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_19000_end_0 = const()[name = tensor("op_19000_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_19000_end_mask_0 = const()[name = tensor("op_19000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19000_cast_fp16 = slice_by_index(begin = var_19000_begin_0, end = var_19000_end_0, end_mask = var_19000_end_mask_0, x = transpose_20)[name = tensor("op_19000_cast_fp16")]; + tensor var_19004_begin_0 = const()[name = tensor("op_19004_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_19004_end_0 = const()[name = tensor("op_19004_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_19004_end_mask_0 = const()[name = tensor("op_19004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19004_cast_fp16 = slice_by_index(begin = var_19004_begin_0, end = var_19004_end_0, end_mask = var_19004_end_mask_0, x = transpose_20)[name = tensor("op_19004_cast_fp16")]; + tensor var_19008_begin_0 = const()[name = tensor("op_19008_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_19008_end_0 = const()[name = tensor("op_19008_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_19008_end_mask_0 = const()[name = tensor("op_19008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19008_cast_fp16 = slice_by_index(begin = var_19008_begin_0, end = var_19008_end_0, end_mask = var_19008_end_mask_0, x = transpose_20)[name = tensor("op_19008_cast_fp16")]; + tensor var_19012_begin_0 = const()[name = tensor("op_19012_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_19012_end_0 = const()[name = tensor("op_19012_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_19012_end_mask_0 = const()[name = tensor("op_19012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_19012_cast_fp16 = slice_by_index(begin = var_19012_begin_0, end = var_19012_end_0, end_mask = var_19012_end_mask_0, x = transpose_20)[name = tensor("op_19012_cast_fp16")]; + tensor var_19014_begin_0 = const()[name = tensor("op_19014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19014_end_0 = const()[name = tensor("op_19014_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19014_end_mask_0 = const()[name = tensor("op_19014_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19014_cast_fp16 = slice_by_index(begin = var_19014_begin_0, end = var_19014_end_0, end_mask = var_19014_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19014_cast_fp16")]; + tensor var_19018_begin_0 = const()[name = tensor("op_19018_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19018_end_0 = const()[name = tensor("op_19018_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_19018_end_mask_0 = const()[name = tensor("op_19018_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19018_cast_fp16 = slice_by_index(begin = var_19018_begin_0, end = var_19018_end_0, end_mask = var_19018_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19018_cast_fp16")]; + tensor var_19022_begin_0 = const()[name = tensor("op_19022_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19022_end_0 = const()[name = tensor("op_19022_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_19022_end_mask_0 = const()[name = tensor("op_19022_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19022_cast_fp16 = slice_by_index(begin = var_19022_begin_0, end = var_19022_end_0, end_mask = var_19022_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19022_cast_fp16")]; + tensor var_19026_begin_0 = const()[name = tensor("op_19026_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19026_end_0 = const()[name = tensor("op_19026_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_19026_end_mask_0 = const()[name = tensor("op_19026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19026_cast_fp16 = slice_by_index(begin = var_19026_begin_0, end = var_19026_end_0, end_mask = var_19026_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19026_cast_fp16")]; + tensor var_19030_begin_0 = const()[name = tensor("op_19030_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19030_end_0 = const()[name = tensor("op_19030_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_19030_end_mask_0 = const()[name = tensor("op_19030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19030_cast_fp16 = slice_by_index(begin = var_19030_begin_0, end = var_19030_end_0, end_mask = var_19030_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19030_cast_fp16")]; + tensor var_19034_begin_0 = const()[name = tensor("op_19034_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19034_end_0 = const()[name = tensor("op_19034_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_19034_end_mask_0 = const()[name = tensor("op_19034_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19034_cast_fp16 = slice_by_index(begin = var_19034_begin_0, end = var_19034_end_0, end_mask = var_19034_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19034_cast_fp16")]; + tensor var_19038_begin_0 = const()[name = tensor("op_19038_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19038_end_0 = const()[name = tensor("op_19038_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_19038_end_mask_0 = const()[name = tensor("op_19038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19038_cast_fp16 = slice_by_index(begin = var_19038_begin_0, end = var_19038_end_0, end_mask = var_19038_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19038_cast_fp16")]; + tensor var_19042_begin_0 = const()[name = tensor("op_19042_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19042_end_0 = const()[name = tensor("op_19042_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_19042_end_mask_0 = const()[name = tensor("op_19042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19042_cast_fp16 = slice_by_index(begin = var_19042_begin_0, end = var_19042_end_0, end_mask = var_19042_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19042_cast_fp16")]; + tensor var_19046_begin_0 = const()[name = tensor("op_19046_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19046_end_0 = const()[name = tensor("op_19046_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_19046_end_mask_0 = const()[name = tensor("op_19046_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19046_cast_fp16 = slice_by_index(begin = var_19046_begin_0, end = var_19046_end_0, end_mask = var_19046_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19046_cast_fp16")]; + tensor var_19050_begin_0 = const()[name = tensor("op_19050_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19050_end_0 = const()[name = tensor("op_19050_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_19050_end_mask_0 = const()[name = tensor("op_19050_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19050_cast_fp16 = slice_by_index(begin = var_19050_begin_0, end = var_19050_end_0, end_mask = var_19050_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19050_cast_fp16")]; + tensor var_19054_begin_0 = const()[name = tensor("op_19054_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19054_end_0 = const()[name = tensor("op_19054_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_19054_end_mask_0 = const()[name = tensor("op_19054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19054_cast_fp16 = slice_by_index(begin = var_19054_begin_0, end = var_19054_end_0, end_mask = var_19054_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19054_cast_fp16")]; + tensor var_19058_begin_0 = const()[name = tensor("op_19058_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19058_end_0 = const()[name = tensor("op_19058_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_19058_end_mask_0 = const()[name = tensor("op_19058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19058_cast_fp16 = slice_by_index(begin = var_19058_begin_0, end = var_19058_end_0, end_mask = var_19058_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19058_cast_fp16")]; + tensor var_19062_begin_0 = const()[name = tensor("op_19062_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19062_end_0 = const()[name = tensor("op_19062_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_19062_end_mask_0 = const()[name = tensor("op_19062_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19062_cast_fp16 = slice_by_index(begin = var_19062_begin_0, end = var_19062_end_0, end_mask = var_19062_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19062_cast_fp16")]; + tensor var_19066_begin_0 = const()[name = tensor("op_19066_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19066_end_0 = const()[name = tensor("op_19066_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_19066_end_mask_0 = const()[name = tensor("op_19066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19066_cast_fp16 = slice_by_index(begin = var_19066_begin_0, end = var_19066_end_0, end_mask = var_19066_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19066_cast_fp16")]; + tensor var_19070_begin_0 = const()[name = tensor("op_19070_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19070_end_0 = const()[name = tensor("op_19070_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_19070_end_mask_0 = const()[name = tensor("op_19070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19070_cast_fp16 = slice_by_index(begin = var_19070_begin_0, end = var_19070_end_0, end_mask = var_19070_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19070_cast_fp16")]; + tensor var_19074_begin_0 = const()[name = tensor("op_19074_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19074_end_0 = const()[name = tensor("op_19074_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_19074_end_mask_0 = const()[name = tensor("op_19074_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19074_cast_fp16 = slice_by_index(begin = var_19074_begin_0, end = var_19074_end_0, end_mask = var_19074_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19074_cast_fp16")]; + tensor var_19078_begin_0 = const()[name = tensor("op_19078_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19078_end_0 = const()[name = tensor("op_19078_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_19078_end_mask_0 = const()[name = tensor("op_19078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19078_cast_fp16 = slice_by_index(begin = var_19078_begin_0, end = var_19078_end_0, end_mask = var_19078_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19078_cast_fp16")]; + tensor var_19082_begin_0 = const()[name = tensor("op_19082_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19082_end_0 = const()[name = tensor("op_19082_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_19082_end_mask_0 = const()[name = tensor("op_19082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19082_cast_fp16 = slice_by_index(begin = var_19082_begin_0, end = var_19082_end_0, end_mask = var_19082_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19082_cast_fp16")]; + tensor var_19086_begin_0 = const()[name = tensor("op_19086_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_19086_end_0 = const()[name = tensor("op_19086_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_19086_end_mask_0 = const()[name = tensor("op_19086_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19086_cast_fp16 = slice_by_index(begin = var_19086_begin_0, end = var_19086_end_0, end_mask = var_19086_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19086_cast_fp16")]; + tensor var_19090_begin_0 = const()[name = tensor("op_19090_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_19090_end_0 = const()[name = tensor("op_19090_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_19090_end_mask_0 = const()[name = tensor("op_19090_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19090_cast_fp16 = slice_by_index(begin = var_19090_begin_0, end = var_19090_end_0, end_mask = var_19090_end_mask_0, x = value_23_cast_fp16)[name = tensor("op_19090_cast_fp16")]; + tensor var_19094_equation_0 = const()[name = tensor("op_19094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19094_cast_fp16 = einsum(equation = var_19094_equation_0, values = (var_18936_cast_fp16, var_18378_cast_fp16))[name = tensor("op_19094_cast_fp16")]; + tensor var_19095_to_fp16 = const()[name = tensor("op_19095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1761_cast_fp16 = mul(x = var_19094_cast_fp16, y = var_19095_to_fp16)[name = tensor("aw_chunk_1761_cast_fp16")]; + tensor var_19098_equation_0 = const()[name = tensor("op_19098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19098_cast_fp16 = einsum(equation = var_19098_equation_0, values = (var_18936_cast_fp16, var_18385_cast_fp16))[name = tensor("op_19098_cast_fp16")]; + tensor var_19099_to_fp16 = const()[name = tensor("op_19099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1763_cast_fp16 = mul(x = var_19098_cast_fp16, y = var_19099_to_fp16)[name = tensor("aw_chunk_1763_cast_fp16")]; + tensor var_19102_equation_0 = const()[name = tensor("op_19102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19102_cast_fp16 = einsum(equation = var_19102_equation_0, values = (var_18936_cast_fp16, var_18392_cast_fp16))[name = tensor("op_19102_cast_fp16")]; + tensor var_19103_to_fp16 = const()[name = tensor("op_19103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1765_cast_fp16 = mul(x = var_19102_cast_fp16, y = var_19103_to_fp16)[name = tensor("aw_chunk_1765_cast_fp16")]; + tensor var_19106_equation_0 = const()[name = tensor("op_19106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19106_cast_fp16 = einsum(equation = var_19106_equation_0, values = (var_18936_cast_fp16, var_18399_cast_fp16))[name = tensor("op_19106_cast_fp16")]; + tensor var_19107_to_fp16 = const()[name = tensor("op_19107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1767_cast_fp16 = mul(x = var_19106_cast_fp16, y = var_19107_to_fp16)[name = tensor("aw_chunk_1767_cast_fp16")]; + tensor var_19110_equation_0 = const()[name = tensor("op_19110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19110_cast_fp16 = einsum(equation = var_19110_equation_0, values = (var_18940_cast_fp16, var_18406_cast_fp16))[name = tensor("op_19110_cast_fp16")]; + tensor var_19111_to_fp16 = const()[name = tensor("op_19111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1769_cast_fp16 = mul(x = var_19110_cast_fp16, y = var_19111_to_fp16)[name = tensor("aw_chunk_1769_cast_fp16")]; + tensor var_19114_equation_0 = const()[name = tensor("op_19114_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19114_cast_fp16 = einsum(equation = var_19114_equation_0, values = (var_18940_cast_fp16, var_18413_cast_fp16))[name = tensor("op_19114_cast_fp16")]; + tensor var_19115_to_fp16 = const()[name = tensor("op_19115_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1771_cast_fp16 = mul(x = var_19114_cast_fp16, y = var_19115_to_fp16)[name = tensor("aw_chunk_1771_cast_fp16")]; + tensor var_19118_equation_0 = const()[name = tensor("op_19118_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19118_cast_fp16 = einsum(equation = var_19118_equation_0, values = (var_18940_cast_fp16, var_18420_cast_fp16))[name = tensor("op_19118_cast_fp16")]; + tensor var_19119_to_fp16 = const()[name = tensor("op_19119_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1773_cast_fp16 = mul(x = var_19118_cast_fp16, y = var_19119_to_fp16)[name = tensor("aw_chunk_1773_cast_fp16")]; + tensor var_19122_equation_0 = const()[name = tensor("op_19122_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19122_cast_fp16 = einsum(equation = var_19122_equation_0, values = (var_18940_cast_fp16, var_18427_cast_fp16))[name = tensor("op_19122_cast_fp16")]; + tensor var_19123_to_fp16 = const()[name = tensor("op_19123_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1775_cast_fp16 = mul(x = var_19122_cast_fp16, y = var_19123_to_fp16)[name = tensor("aw_chunk_1775_cast_fp16")]; + tensor var_19126_equation_0 = const()[name = tensor("op_19126_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19126_cast_fp16 = einsum(equation = var_19126_equation_0, values = (var_18944_cast_fp16, var_18434_cast_fp16))[name = tensor("op_19126_cast_fp16")]; + tensor var_19127_to_fp16 = const()[name = tensor("op_19127_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1777_cast_fp16 = mul(x = var_19126_cast_fp16, y = var_19127_to_fp16)[name = tensor("aw_chunk_1777_cast_fp16")]; + tensor var_19130_equation_0 = const()[name = tensor("op_19130_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19130_cast_fp16 = einsum(equation = var_19130_equation_0, values = (var_18944_cast_fp16, var_18441_cast_fp16))[name = tensor("op_19130_cast_fp16")]; + tensor var_19131_to_fp16 = const()[name = tensor("op_19131_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1779_cast_fp16 = mul(x = var_19130_cast_fp16, y = var_19131_to_fp16)[name = tensor("aw_chunk_1779_cast_fp16")]; + tensor var_19134_equation_0 = const()[name = tensor("op_19134_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19134_cast_fp16 = einsum(equation = var_19134_equation_0, values = (var_18944_cast_fp16, var_18448_cast_fp16))[name = tensor("op_19134_cast_fp16")]; + tensor var_19135_to_fp16 = const()[name = tensor("op_19135_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1781_cast_fp16 = mul(x = var_19134_cast_fp16, y = var_19135_to_fp16)[name = tensor("aw_chunk_1781_cast_fp16")]; + tensor var_19138_equation_0 = const()[name = tensor("op_19138_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19138_cast_fp16 = einsum(equation = var_19138_equation_0, values = (var_18944_cast_fp16, var_18455_cast_fp16))[name = tensor("op_19138_cast_fp16")]; + tensor var_19139_to_fp16 = const()[name = tensor("op_19139_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1783_cast_fp16 = mul(x = var_19138_cast_fp16, y = var_19139_to_fp16)[name = tensor("aw_chunk_1783_cast_fp16")]; + tensor var_19142_equation_0 = const()[name = tensor("op_19142_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19142_cast_fp16 = einsum(equation = var_19142_equation_0, values = (var_18948_cast_fp16, var_18462_cast_fp16))[name = tensor("op_19142_cast_fp16")]; + tensor var_19143_to_fp16 = const()[name = tensor("op_19143_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1785_cast_fp16 = mul(x = var_19142_cast_fp16, y = var_19143_to_fp16)[name = tensor("aw_chunk_1785_cast_fp16")]; + tensor var_19146_equation_0 = const()[name = tensor("op_19146_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19146_cast_fp16 = einsum(equation = var_19146_equation_0, values = (var_18948_cast_fp16, var_18469_cast_fp16))[name = tensor("op_19146_cast_fp16")]; + tensor var_19147_to_fp16 = const()[name = tensor("op_19147_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1787_cast_fp16 = mul(x = var_19146_cast_fp16, y = var_19147_to_fp16)[name = tensor("aw_chunk_1787_cast_fp16")]; + tensor var_19150_equation_0 = const()[name = tensor("op_19150_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19150_cast_fp16 = einsum(equation = var_19150_equation_0, values = (var_18948_cast_fp16, var_18476_cast_fp16))[name = tensor("op_19150_cast_fp16")]; + tensor var_19151_to_fp16 = const()[name = tensor("op_19151_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1789_cast_fp16 = mul(x = var_19150_cast_fp16, y = var_19151_to_fp16)[name = tensor("aw_chunk_1789_cast_fp16")]; + tensor var_19154_equation_0 = const()[name = tensor("op_19154_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19154_cast_fp16 = einsum(equation = var_19154_equation_0, values = (var_18948_cast_fp16, var_18483_cast_fp16))[name = tensor("op_19154_cast_fp16")]; + tensor var_19155_to_fp16 = const()[name = tensor("op_19155_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1791_cast_fp16 = mul(x = var_19154_cast_fp16, y = var_19155_to_fp16)[name = tensor("aw_chunk_1791_cast_fp16")]; + tensor var_19158_equation_0 = const()[name = tensor("op_19158_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19158_cast_fp16 = einsum(equation = var_19158_equation_0, values = (var_18952_cast_fp16, var_18490_cast_fp16))[name = tensor("op_19158_cast_fp16")]; + tensor var_19159_to_fp16 = const()[name = tensor("op_19159_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1793_cast_fp16 = mul(x = var_19158_cast_fp16, y = var_19159_to_fp16)[name = tensor("aw_chunk_1793_cast_fp16")]; + tensor var_19162_equation_0 = const()[name = tensor("op_19162_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19162_cast_fp16 = einsum(equation = var_19162_equation_0, values = (var_18952_cast_fp16, var_18497_cast_fp16))[name = tensor("op_19162_cast_fp16")]; + tensor var_19163_to_fp16 = const()[name = tensor("op_19163_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1795_cast_fp16 = mul(x = var_19162_cast_fp16, y = var_19163_to_fp16)[name = tensor("aw_chunk_1795_cast_fp16")]; + tensor var_19166_equation_0 = const()[name = tensor("op_19166_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19166_cast_fp16 = einsum(equation = var_19166_equation_0, values = (var_18952_cast_fp16, var_18504_cast_fp16))[name = tensor("op_19166_cast_fp16")]; + tensor var_19167_to_fp16 = const()[name = tensor("op_19167_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1797_cast_fp16 = mul(x = var_19166_cast_fp16, y = var_19167_to_fp16)[name = tensor("aw_chunk_1797_cast_fp16")]; + tensor var_19170_equation_0 = const()[name = tensor("op_19170_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19170_cast_fp16 = einsum(equation = var_19170_equation_0, values = (var_18952_cast_fp16, var_18511_cast_fp16))[name = tensor("op_19170_cast_fp16")]; + tensor var_19171_to_fp16 = const()[name = tensor("op_19171_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1799_cast_fp16 = mul(x = var_19170_cast_fp16, y = var_19171_to_fp16)[name = tensor("aw_chunk_1799_cast_fp16")]; + tensor var_19174_equation_0 = const()[name = tensor("op_19174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19174_cast_fp16 = einsum(equation = var_19174_equation_0, values = (var_18956_cast_fp16, var_18518_cast_fp16))[name = tensor("op_19174_cast_fp16")]; + tensor var_19175_to_fp16 = const()[name = tensor("op_19175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1801_cast_fp16 = mul(x = var_19174_cast_fp16, y = var_19175_to_fp16)[name = tensor("aw_chunk_1801_cast_fp16")]; + tensor var_19178_equation_0 = const()[name = tensor("op_19178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19178_cast_fp16 = einsum(equation = var_19178_equation_0, values = (var_18956_cast_fp16, var_18525_cast_fp16))[name = tensor("op_19178_cast_fp16")]; + tensor var_19179_to_fp16 = const()[name = tensor("op_19179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1803_cast_fp16 = mul(x = var_19178_cast_fp16, y = var_19179_to_fp16)[name = tensor("aw_chunk_1803_cast_fp16")]; + tensor var_19182_equation_0 = const()[name = tensor("op_19182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19182_cast_fp16 = einsum(equation = var_19182_equation_0, values = (var_18956_cast_fp16, var_18532_cast_fp16))[name = tensor("op_19182_cast_fp16")]; + tensor var_19183_to_fp16 = const()[name = tensor("op_19183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1805_cast_fp16 = mul(x = var_19182_cast_fp16, y = var_19183_to_fp16)[name = tensor("aw_chunk_1805_cast_fp16")]; + tensor var_19186_equation_0 = const()[name = tensor("op_19186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19186_cast_fp16 = einsum(equation = var_19186_equation_0, values = (var_18956_cast_fp16, var_18539_cast_fp16))[name = tensor("op_19186_cast_fp16")]; + tensor var_19187_to_fp16 = const()[name = tensor("op_19187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1807_cast_fp16 = mul(x = var_19186_cast_fp16, y = var_19187_to_fp16)[name = tensor("aw_chunk_1807_cast_fp16")]; + tensor var_19190_equation_0 = const()[name = tensor("op_19190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19190_cast_fp16 = einsum(equation = var_19190_equation_0, values = (var_18960_cast_fp16, var_18546_cast_fp16))[name = tensor("op_19190_cast_fp16")]; + tensor var_19191_to_fp16 = const()[name = tensor("op_19191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1809_cast_fp16 = mul(x = var_19190_cast_fp16, y = var_19191_to_fp16)[name = tensor("aw_chunk_1809_cast_fp16")]; + tensor var_19194_equation_0 = const()[name = tensor("op_19194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19194_cast_fp16 = einsum(equation = var_19194_equation_0, values = (var_18960_cast_fp16, var_18553_cast_fp16))[name = tensor("op_19194_cast_fp16")]; + tensor var_19195_to_fp16 = const()[name = tensor("op_19195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1811_cast_fp16 = mul(x = var_19194_cast_fp16, y = var_19195_to_fp16)[name = tensor("aw_chunk_1811_cast_fp16")]; + tensor var_19198_equation_0 = const()[name = tensor("op_19198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19198_cast_fp16 = einsum(equation = var_19198_equation_0, values = (var_18960_cast_fp16, var_18560_cast_fp16))[name = tensor("op_19198_cast_fp16")]; + tensor var_19199_to_fp16 = const()[name = tensor("op_19199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1813_cast_fp16 = mul(x = var_19198_cast_fp16, y = var_19199_to_fp16)[name = tensor("aw_chunk_1813_cast_fp16")]; + tensor var_19202_equation_0 = const()[name = tensor("op_19202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19202_cast_fp16 = einsum(equation = var_19202_equation_0, values = (var_18960_cast_fp16, var_18567_cast_fp16))[name = tensor("op_19202_cast_fp16")]; + tensor var_19203_to_fp16 = const()[name = tensor("op_19203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1815_cast_fp16 = mul(x = var_19202_cast_fp16, y = var_19203_to_fp16)[name = tensor("aw_chunk_1815_cast_fp16")]; + tensor var_19206_equation_0 = const()[name = tensor("op_19206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19206_cast_fp16 = einsum(equation = var_19206_equation_0, values = (var_18964_cast_fp16, var_18574_cast_fp16))[name = tensor("op_19206_cast_fp16")]; + tensor var_19207_to_fp16 = const()[name = tensor("op_19207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1817_cast_fp16 = mul(x = var_19206_cast_fp16, y = var_19207_to_fp16)[name = tensor("aw_chunk_1817_cast_fp16")]; + tensor var_19210_equation_0 = const()[name = tensor("op_19210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19210_cast_fp16 = einsum(equation = var_19210_equation_0, values = (var_18964_cast_fp16, var_18581_cast_fp16))[name = tensor("op_19210_cast_fp16")]; + tensor var_19211_to_fp16 = const()[name = tensor("op_19211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1819_cast_fp16 = mul(x = var_19210_cast_fp16, y = var_19211_to_fp16)[name = tensor("aw_chunk_1819_cast_fp16")]; + tensor var_19214_equation_0 = const()[name = tensor("op_19214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19214_cast_fp16 = einsum(equation = var_19214_equation_0, values = (var_18964_cast_fp16, var_18588_cast_fp16))[name = tensor("op_19214_cast_fp16")]; + tensor var_19215_to_fp16 = const()[name = tensor("op_19215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1821_cast_fp16 = mul(x = var_19214_cast_fp16, y = var_19215_to_fp16)[name = tensor("aw_chunk_1821_cast_fp16")]; + tensor var_19218_equation_0 = const()[name = tensor("op_19218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19218_cast_fp16 = einsum(equation = var_19218_equation_0, values = (var_18964_cast_fp16, var_18595_cast_fp16))[name = tensor("op_19218_cast_fp16")]; + tensor var_19219_to_fp16 = const()[name = tensor("op_19219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1823_cast_fp16 = mul(x = var_19218_cast_fp16, y = var_19219_to_fp16)[name = tensor("aw_chunk_1823_cast_fp16")]; + tensor var_19222_equation_0 = const()[name = tensor("op_19222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19222_cast_fp16 = einsum(equation = var_19222_equation_0, values = (var_18968_cast_fp16, var_18602_cast_fp16))[name = tensor("op_19222_cast_fp16")]; + tensor var_19223_to_fp16 = const()[name = tensor("op_19223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1825_cast_fp16 = mul(x = var_19222_cast_fp16, y = var_19223_to_fp16)[name = tensor("aw_chunk_1825_cast_fp16")]; + tensor var_19226_equation_0 = const()[name = tensor("op_19226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19226_cast_fp16 = einsum(equation = var_19226_equation_0, values = (var_18968_cast_fp16, var_18609_cast_fp16))[name = tensor("op_19226_cast_fp16")]; + tensor var_19227_to_fp16 = const()[name = tensor("op_19227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1827_cast_fp16 = mul(x = var_19226_cast_fp16, y = var_19227_to_fp16)[name = tensor("aw_chunk_1827_cast_fp16")]; + tensor var_19230_equation_0 = const()[name = tensor("op_19230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19230_cast_fp16 = einsum(equation = var_19230_equation_0, values = (var_18968_cast_fp16, var_18616_cast_fp16))[name = tensor("op_19230_cast_fp16")]; + tensor var_19231_to_fp16 = const()[name = tensor("op_19231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1829_cast_fp16 = mul(x = var_19230_cast_fp16, y = var_19231_to_fp16)[name = tensor("aw_chunk_1829_cast_fp16")]; + tensor var_19234_equation_0 = const()[name = tensor("op_19234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19234_cast_fp16 = einsum(equation = var_19234_equation_0, values = (var_18968_cast_fp16, var_18623_cast_fp16))[name = tensor("op_19234_cast_fp16")]; + tensor var_19235_to_fp16 = const()[name = tensor("op_19235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1831_cast_fp16 = mul(x = var_19234_cast_fp16, y = var_19235_to_fp16)[name = tensor("aw_chunk_1831_cast_fp16")]; + tensor var_19238_equation_0 = const()[name = tensor("op_19238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19238_cast_fp16 = einsum(equation = var_19238_equation_0, values = (var_18972_cast_fp16, var_18630_cast_fp16))[name = tensor("op_19238_cast_fp16")]; + tensor var_19239_to_fp16 = const()[name = tensor("op_19239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1833_cast_fp16 = mul(x = var_19238_cast_fp16, y = var_19239_to_fp16)[name = tensor("aw_chunk_1833_cast_fp16")]; + tensor var_19242_equation_0 = const()[name = tensor("op_19242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19242_cast_fp16 = einsum(equation = var_19242_equation_0, values = (var_18972_cast_fp16, var_18637_cast_fp16))[name = tensor("op_19242_cast_fp16")]; + tensor var_19243_to_fp16 = const()[name = tensor("op_19243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1835_cast_fp16 = mul(x = var_19242_cast_fp16, y = var_19243_to_fp16)[name = tensor("aw_chunk_1835_cast_fp16")]; + tensor var_19246_equation_0 = const()[name = tensor("op_19246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19246_cast_fp16 = einsum(equation = var_19246_equation_0, values = (var_18972_cast_fp16, var_18644_cast_fp16))[name = tensor("op_19246_cast_fp16")]; + tensor var_19247_to_fp16 = const()[name = tensor("op_19247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1837_cast_fp16 = mul(x = var_19246_cast_fp16, y = var_19247_to_fp16)[name = tensor("aw_chunk_1837_cast_fp16")]; + tensor var_19250_equation_0 = const()[name = tensor("op_19250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19250_cast_fp16 = einsum(equation = var_19250_equation_0, values = (var_18972_cast_fp16, var_18651_cast_fp16))[name = tensor("op_19250_cast_fp16")]; + tensor var_19251_to_fp16 = const()[name = tensor("op_19251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1839_cast_fp16 = mul(x = var_19250_cast_fp16, y = var_19251_to_fp16)[name = tensor("aw_chunk_1839_cast_fp16")]; + tensor var_19254_equation_0 = const()[name = tensor("op_19254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19254_cast_fp16 = einsum(equation = var_19254_equation_0, values = (var_18976_cast_fp16, var_18658_cast_fp16))[name = tensor("op_19254_cast_fp16")]; + tensor var_19255_to_fp16 = const()[name = tensor("op_19255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1841_cast_fp16 = mul(x = var_19254_cast_fp16, y = var_19255_to_fp16)[name = tensor("aw_chunk_1841_cast_fp16")]; + tensor var_19258_equation_0 = const()[name = tensor("op_19258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19258_cast_fp16 = einsum(equation = var_19258_equation_0, values = (var_18976_cast_fp16, var_18665_cast_fp16))[name = tensor("op_19258_cast_fp16")]; + tensor var_19259_to_fp16 = const()[name = tensor("op_19259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1843_cast_fp16 = mul(x = var_19258_cast_fp16, y = var_19259_to_fp16)[name = tensor("aw_chunk_1843_cast_fp16")]; + tensor var_19262_equation_0 = const()[name = tensor("op_19262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19262_cast_fp16 = einsum(equation = var_19262_equation_0, values = (var_18976_cast_fp16, var_18672_cast_fp16))[name = tensor("op_19262_cast_fp16")]; + tensor var_19263_to_fp16 = const()[name = tensor("op_19263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1845_cast_fp16 = mul(x = var_19262_cast_fp16, y = var_19263_to_fp16)[name = tensor("aw_chunk_1845_cast_fp16")]; + tensor var_19266_equation_0 = const()[name = tensor("op_19266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19266_cast_fp16 = einsum(equation = var_19266_equation_0, values = (var_18976_cast_fp16, var_18679_cast_fp16))[name = tensor("op_19266_cast_fp16")]; + tensor var_19267_to_fp16 = const()[name = tensor("op_19267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1847_cast_fp16 = mul(x = var_19266_cast_fp16, y = var_19267_to_fp16)[name = tensor("aw_chunk_1847_cast_fp16")]; + tensor var_19270_equation_0 = const()[name = tensor("op_19270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19270_cast_fp16 = einsum(equation = var_19270_equation_0, values = (var_18980_cast_fp16, var_18686_cast_fp16))[name = tensor("op_19270_cast_fp16")]; + tensor var_19271_to_fp16 = const()[name = tensor("op_19271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1849_cast_fp16 = mul(x = var_19270_cast_fp16, y = var_19271_to_fp16)[name = tensor("aw_chunk_1849_cast_fp16")]; + tensor var_19274_equation_0 = const()[name = tensor("op_19274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19274_cast_fp16 = einsum(equation = var_19274_equation_0, values = (var_18980_cast_fp16, var_18693_cast_fp16))[name = tensor("op_19274_cast_fp16")]; + tensor var_19275_to_fp16 = const()[name = tensor("op_19275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1851_cast_fp16 = mul(x = var_19274_cast_fp16, y = var_19275_to_fp16)[name = tensor("aw_chunk_1851_cast_fp16")]; + tensor var_19278_equation_0 = const()[name = tensor("op_19278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19278_cast_fp16 = einsum(equation = var_19278_equation_0, values = (var_18980_cast_fp16, var_18700_cast_fp16))[name = tensor("op_19278_cast_fp16")]; + tensor var_19279_to_fp16 = const()[name = tensor("op_19279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1853_cast_fp16 = mul(x = var_19278_cast_fp16, y = var_19279_to_fp16)[name = tensor("aw_chunk_1853_cast_fp16")]; + tensor var_19282_equation_0 = const()[name = tensor("op_19282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19282_cast_fp16 = einsum(equation = var_19282_equation_0, values = (var_18980_cast_fp16, var_18707_cast_fp16))[name = tensor("op_19282_cast_fp16")]; + tensor var_19283_to_fp16 = const()[name = tensor("op_19283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1855_cast_fp16 = mul(x = var_19282_cast_fp16, y = var_19283_to_fp16)[name = tensor("aw_chunk_1855_cast_fp16")]; + tensor var_19286_equation_0 = const()[name = tensor("op_19286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19286_cast_fp16 = einsum(equation = var_19286_equation_0, values = (var_18984_cast_fp16, var_18714_cast_fp16))[name = tensor("op_19286_cast_fp16")]; + tensor var_19287_to_fp16 = const()[name = tensor("op_19287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1857_cast_fp16 = mul(x = var_19286_cast_fp16, y = var_19287_to_fp16)[name = tensor("aw_chunk_1857_cast_fp16")]; + tensor var_19290_equation_0 = const()[name = tensor("op_19290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19290_cast_fp16 = einsum(equation = var_19290_equation_0, values = (var_18984_cast_fp16, var_18721_cast_fp16))[name = tensor("op_19290_cast_fp16")]; + tensor var_19291_to_fp16 = const()[name = tensor("op_19291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1859_cast_fp16 = mul(x = var_19290_cast_fp16, y = var_19291_to_fp16)[name = tensor("aw_chunk_1859_cast_fp16")]; + tensor var_19294_equation_0 = const()[name = tensor("op_19294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19294_cast_fp16 = einsum(equation = var_19294_equation_0, values = (var_18984_cast_fp16, var_18728_cast_fp16))[name = tensor("op_19294_cast_fp16")]; + tensor var_19295_to_fp16 = const()[name = tensor("op_19295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1861_cast_fp16 = mul(x = var_19294_cast_fp16, y = var_19295_to_fp16)[name = tensor("aw_chunk_1861_cast_fp16")]; + tensor var_19298_equation_0 = const()[name = tensor("op_19298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19298_cast_fp16 = einsum(equation = var_19298_equation_0, values = (var_18984_cast_fp16, var_18735_cast_fp16))[name = tensor("op_19298_cast_fp16")]; + tensor var_19299_to_fp16 = const()[name = tensor("op_19299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1863_cast_fp16 = mul(x = var_19298_cast_fp16, y = var_19299_to_fp16)[name = tensor("aw_chunk_1863_cast_fp16")]; + tensor var_19302_equation_0 = const()[name = tensor("op_19302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19302_cast_fp16 = einsum(equation = var_19302_equation_0, values = (var_18988_cast_fp16, var_18742_cast_fp16))[name = tensor("op_19302_cast_fp16")]; + tensor var_19303_to_fp16 = const()[name = tensor("op_19303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1865_cast_fp16 = mul(x = var_19302_cast_fp16, y = var_19303_to_fp16)[name = tensor("aw_chunk_1865_cast_fp16")]; + tensor var_19306_equation_0 = const()[name = tensor("op_19306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19306_cast_fp16 = einsum(equation = var_19306_equation_0, values = (var_18988_cast_fp16, var_18749_cast_fp16))[name = tensor("op_19306_cast_fp16")]; + tensor var_19307_to_fp16 = const()[name = tensor("op_19307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1867_cast_fp16 = mul(x = var_19306_cast_fp16, y = var_19307_to_fp16)[name = tensor("aw_chunk_1867_cast_fp16")]; + tensor var_19310_equation_0 = const()[name = tensor("op_19310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19310_cast_fp16 = einsum(equation = var_19310_equation_0, values = (var_18988_cast_fp16, var_18756_cast_fp16))[name = tensor("op_19310_cast_fp16")]; + tensor var_19311_to_fp16 = const()[name = tensor("op_19311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1869_cast_fp16 = mul(x = var_19310_cast_fp16, y = var_19311_to_fp16)[name = tensor("aw_chunk_1869_cast_fp16")]; + tensor var_19314_equation_0 = const()[name = tensor("op_19314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19314_cast_fp16 = einsum(equation = var_19314_equation_0, values = (var_18988_cast_fp16, var_18763_cast_fp16))[name = tensor("op_19314_cast_fp16")]; + tensor var_19315_to_fp16 = const()[name = tensor("op_19315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1871_cast_fp16 = mul(x = var_19314_cast_fp16, y = var_19315_to_fp16)[name = tensor("aw_chunk_1871_cast_fp16")]; + tensor var_19318_equation_0 = const()[name = tensor("op_19318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19318_cast_fp16 = einsum(equation = var_19318_equation_0, values = (var_18992_cast_fp16, var_18770_cast_fp16))[name = tensor("op_19318_cast_fp16")]; + tensor var_19319_to_fp16 = const()[name = tensor("op_19319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1873_cast_fp16 = mul(x = var_19318_cast_fp16, y = var_19319_to_fp16)[name = tensor("aw_chunk_1873_cast_fp16")]; + tensor var_19322_equation_0 = const()[name = tensor("op_19322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19322_cast_fp16 = einsum(equation = var_19322_equation_0, values = (var_18992_cast_fp16, var_18777_cast_fp16))[name = tensor("op_19322_cast_fp16")]; + tensor var_19323_to_fp16 = const()[name = tensor("op_19323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1875_cast_fp16 = mul(x = var_19322_cast_fp16, y = var_19323_to_fp16)[name = tensor("aw_chunk_1875_cast_fp16")]; + tensor var_19326_equation_0 = const()[name = tensor("op_19326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19326_cast_fp16 = einsum(equation = var_19326_equation_0, values = (var_18992_cast_fp16, var_18784_cast_fp16))[name = tensor("op_19326_cast_fp16")]; + tensor var_19327_to_fp16 = const()[name = tensor("op_19327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1877_cast_fp16 = mul(x = var_19326_cast_fp16, y = var_19327_to_fp16)[name = tensor("aw_chunk_1877_cast_fp16")]; + tensor var_19330_equation_0 = const()[name = tensor("op_19330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19330_cast_fp16 = einsum(equation = var_19330_equation_0, values = (var_18992_cast_fp16, var_18791_cast_fp16))[name = tensor("op_19330_cast_fp16")]; + tensor var_19331_to_fp16 = const()[name = tensor("op_19331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1879_cast_fp16 = mul(x = var_19330_cast_fp16, y = var_19331_to_fp16)[name = tensor("aw_chunk_1879_cast_fp16")]; + tensor var_19334_equation_0 = const()[name = tensor("op_19334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19334_cast_fp16 = einsum(equation = var_19334_equation_0, values = (var_18996_cast_fp16, var_18798_cast_fp16))[name = tensor("op_19334_cast_fp16")]; + tensor var_19335_to_fp16 = const()[name = tensor("op_19335_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1881_cast_fp16 = mul(x = var_19334_cast_fp16, y = var_19335_to_fp16)[name = tensor("aw_chunk_1881_cast_fp16")]; + tensor var_19338_equation_0 = const()[name = tensor("op_19338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19338_cast_fp16 = einsum(equation = var_19338_equation_0, values = (var_18996_cast_fp16, var_18805_cast_fp16))[name = tensor("op_19338_cast_fp16")]; + tensor var_19339_to_fp16 = const()[name = tensor("op_19339_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1883_cast_fp16 = mul(x = var_19338_cast_fp16, y = var_19339_to_fp16)[name = tensor("aw_chunk_1883_cast_fp16")]; + tensor var_19342_equation_0 = const()[name = tensor("op_19342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19342_cast_fp16 = einsum(equation = var_19342_equation_0, values = (var_18996_cast_fp16, var_18812_cast_fp16))[name = tensor("op_19342_cast_fp16")]; + tensor var_19343_to_fp16 = const()[name = tensor("op_19343_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1885_cast_fp16 = mul(x = var_19342_cast_fp16, y = var_19343_to_fp16)[name = tensor("aw_chunk_1885_cast_fp16")]; + tensor var_19346_equation_0 = const()[name = tensor("op_19346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19346_cast_fp16 = einsum(equation = var_19346_equation_0, values = (var_18996_cast_fp16, var_18819_cast_fp16))[name = tensor("op_19346_cast_fp16")]; + tensor var_19347_to_fp16 = const()[name = tensor("op_19347_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1887_cast_fp16 = mul(x = var_19346_cast_fp16, y = var_19347_to_fp16)[name = tensor("aw_chunk_1887_cast_fp16")]; + tensor var_19350_equation_0 = const()[name = tensor("op_19350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19350_cast_fp16 = einsum(equation = var_19350_equation_0, values = (var_19000_cast_fp16, var_18826_cast_fp16))[name = tensor("op_19350_cast_fp16")]; + tensor var_19351_to_fp16 = const()[name = tensor("op_19351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1889_cast_fp16 = mul(x = var_19350_cast_fp16, y = var_19351_to_fp16)[name = tensor("aw_chunk_1889_cast_fp16")]; + tensor var_19354_equation_0 = const()[name = tensor("op_19354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19354_cast_fp16 = einsum(equation = var_19354_equation_0, values = (var_19000_cast_fp16, var_18833_cast_fp16))[name = tensor("op_19354_cast_fp16")]; + tensor var_19355_to_fp16 = const()[name = tensor("op_19355_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1891_cast_fp16 = mul(x = var_19354_cast_fp16, y = var_19355_to_fp16)[name = tensor("aw_chunk_1891_cast_fp16")]; + tensor var_19358_equation_0 = const()[name = tensor("op_19358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19358_cast_fp16 = einsum(equation = var_19358_equation_0, values = (var_19000_cast_fp16, var_18840_cast_fp16))[name = tensor("op_19358_cast_fp16")]; + tensor var_19359_to_fp16 = const()[name = tensor("op_19359_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1893_cast_fp16 = mul(x = var_19358_cast_fp16, y = var_19359_to_fp16)[name = tensor("aw_chunk_1893_cast_fp16")]; + tensor var_19362_equation_0 = const()[name = tensor("op_19362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19362_cast_fp16 = einsum(equation = var_19362_equation_0, values = (var_19000_cast_fp16, var_18847_cast_fp16))[name = tensor("op_19362_cast_fp16")]; + tensor var_19363_to_fp16 = const()[name = tensor("op_19363_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1895_cast_fp16 = mul(x = var_19362_cast_fp16, y = var_19363_to_fp16)[name = tensor("aw_chunk_1895_cast_fp16")]; + tensor var_19366_equation_0 = const()[name = tensor("op_19366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19366_cast_fp16 = einsum(equation = var_19366_equation_0, values = (var_19004_cast_fp16, var_18854_cast_fp16))[name = tensor("op_19366_cast_fp16")]; + tensor var_19367_to_fp16 = const()[name = tensor("op_19367_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1897_cast_fp16 = mul(x = var_19366_cast_fp16, y = var_19367_to_fp16)[name = tensor("aw_chunk_1897_cast_fp16")]; + tensor var_19370_equation_0 = const()[name = tensor("op_19370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19370_cast_fp16 = einsum(equation = var_19370_equation_0, values = (var_19004_cast_fp16, var_18861_cast_fp16))[name = tensor("op_19370_cast_fp16")]; + tensor var_19371_to_fp16 = const()[name = tensor("op_19371_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1899_cast_fp16 = mul(x = var_19370_cast_fp16, y = var_19371_to_fp16)[name = tensor("aw_chunk_1899_cast_fp16")]; + tensor var_19374_equation_0 = const()[name = tensor("op_19374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19374_cast_fp16 = einsum(equation = var_19374_equation_0, values = (var_19004_cast_fp16, var_18868_cast_fp16))[name = tensor("op_19374_cast_fp16")]; + tensor var_19375_to_fp16 = const()[name = tensor("op_19375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1901_cast_fp16 = mul(x = var_19374_cast_fp16, y = var_19375_to_fp16)[name = tensor("aw_chunk_1901_cast_fp16")]; + tensor var_19378_equation_0 = const()[name = tensor("op_19378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19378_cast_fp16 = einsum(equation = var_19378_equation_0, values = (var_19004_cast_fp16, var_18875_cast_fp16))[name = tensor("op_19378_cast_fp16")]; + tensor var_19379_to_fp16 = const()[name = tensor("op_19379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1903_cast_fp16 = mul(x = var_19378_cast_fp16, y = var_19379_to_fp16)[name = tensor("aw_chunk_1903_cast_fp16")]; + tensor var_19382_equation_0 = const()[name = tensor("op_19382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19382_cast_fp16 = einsum(equation = var_19382_equation_0, values = (var_19008_cast_fp16, var_18882_cast_fp16))[name = tensor("op_19382_cast_fp16")]; + tensor var_19383_to_fp16 = const()[name = tensor("op_19383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1905_cast_fp16 = mul(x = var_19382_cast_fp16, y = var_19383_to_fp16)[name = tensor("aw_chunk_1905_cast_fp16")]; + tensor var_19386_equation_0 = const()[name = tensor("op_19386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19386_cast_fp16 = einsum(equation = var_19386_equation_0, values = (var_19008_cast_fp16, var_18889_cast_fp16))[name = tensor("op_19386_cast_fp16")]; + tensor var_19387_to_fp16 = const()[name = tensor("op_19387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1907_cast_fp16 = mul(x = var_19386_cast_fp16, y = var_19387_to_fp16)[name = tensor("aw_chunk_1907_cast_fp16")]; + tensor var_19390_equation_0 = const()[name = tensor("op_19390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19390_cast_fp16 = einsum(equation = var_19390_equation_0, values = (var_19008_cast_fp16, var_18896_cast_fp16))[name = tensor("op_19390_cast_fp16")]; + tensor var_19391_to_fp16 = const()[name = tensor("op_19391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1909_cast_fp16 = mul(x = var_19390_cast_fp16, y = var_19391_to_fp16)[name = tensor("aw_chunk_1909_cast_fp16")]; + tensor var_19394_equation_0 = const()[name = tensor("op_19394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19394_cast_fp16 = einsum(equation = var_19394_equation_0, values = (var_19008_cast_fp16, var_18903_cast_fp16))[name = tensor("op_19394_cast_fp16")]; + tensor var_19395_to_fp16 = const()[name = tensor("op_19395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1911_cast_fp16 = mul(x = var_19394_cast_fp16, y = var_19395_to_fp16)[name = tensor("aw_chunk_1911_cast_fp16")]; + tensor var_19398_equation_0 = const()[name = tensor("op_19398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19398_cast_fp16 = einsum(equation = var_19398_equation_0, values = (var_19012_cast_fp16, var_18910_cast_fp16))[name = tensor("op_19398_cast_fp16")]; + tensor var_19399_to_fp16 = const()[name = tensor("op_19399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1913_cast_fp16 = mul(x = var_19398_cast_fp16, y = var_19399_to_fp16)[name = tensor("aw_chunk_1913_cast_fp16")]; + tensor var_19402_equation_0 = const()[name = tensor("op_19402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19402_cast_fp16 = einsum(equation = var_19402_equation_0, values = (var_19012_cast_fp16, var_18917_cast_fp16))[name = tensor("op_19402_cast_fp16")]; + tensor var_19403_to_fp16 = const()[name = tensor("op_19403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1915_cast_fp16 = mul(x = var_19402_cast_fp16, y = var_19403_to_fp16)[name = tensor("aw_chunk_1915_cast_fp16")]; + tensor var_19406_equation_0 = const()[name = tensor("op_19406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19406_cast_fp16 = einsum(equation = var_19406_equation_0, values = (var_19012_cast_fp16, var_18924_cast_fp16))[name = tensor("op_19406_cast_fp16")]; + tensor var_19407_to_fp16 = const()[name = tensor("op_19407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1917_cast_fp16 = mul(x = var_19406_cast_fp16, y = var_19407_to_fp16)[name = tensor("aw_chunk_1917_cast_fp16")]; + tensor var_19410_equation_0 = const()[name = tensor("op_19410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_19410_cast_fp16 = einsum(equation = var_19410_equation_0, values = (var_19012_cast_fp16, var_18931_cast_fp16))[name = tensor("op_19410_cast_fp16")]; + tensor var_19411_to_fp16 = const()[name = tensor("op_19411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1919_cast_fp16 = mul(x = var_19410_cast_fp16, y = var_19411_to_fp16)[name = tensor("aw_chunk_1919_cast_fp16")]; + tensor var_19413_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1761_cast_fp16)[name = tensor("op_19413_cast_fp16")]; + tensor var_19414_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1763_cast_fp16)[name = tensor("op_19414_cast_fp16")]; + tensor var_19415_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1765_cast_fp16)[name = tensor("op_19415_cast_fp16")]; + tensor var_19416_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1767_cast_fp16)[name = tensor("op_19416_cast_fp16")]; + tensor var_19417_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1769_cast_fp16)[name = tensor("op_19417_cast_fp16")]; + tensor var_19418_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1771_cast_fp16)[name = tensor("op_19418_cast_fp16")]; + tensor var_19419_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1773_cast_fp16)[name = tensor("op_19419_cast_fp16")]; + tensor var_19420_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1775_cast_fp16)[name = tensor("op_19420_cast_fp16")]; + tensor var_19421_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1777_cast_fp16)[name = tensor("op_19421_cast_fp16")]; + tensor var_19422_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1779_cast_fp16)[name = tensor("op_19422_cast_fp16")]; + tensor var_19423_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1781_cast_fp16)[name = tensor("op_19423_cast_fp16")]; + tensor var_19424_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1783_cast_fp16)[name = tensor("op_19424_cast_fp16")]; + tensor var_19425_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1785_cast_fp16)[name = tensor("op_19425_cast_fp16")]; + tensor var_19426_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1787_cast_fp16)[name = tensor("op_19426_cast_fp16")]; + tensor var_19427_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1789_cast_fp16)[name = tensor("op_19427_cast_fp16")]; + tensor var_19428_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1791_cast_fp16)[name = tensor("op_19428_cast_fp16")]; + tensor var_19429_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1793_cast_fp16)[name = tensor("op_19429_cast_fp16")]; + tensor var_19430_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1795_cast_fp16)[name = tensor("op_19430_cast_fp16")]; + tensor var_19431_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1797_cast_fp16)[name = tensor("op_19431_cast_fp16")]; + tensor var_19432_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1799_cast_fp16)[name = tensor("op_19432_cast_fp16")]; + tensor var_19433_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1801_cast_fp16)[name = tensor("op_19433_cast_fp16")]; + tensor var_19434_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1803_cast_fp16)[name = tensor("op_19434_cast_fp16")]; + tensor var_19435_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1805_cast_fp16)[name = tensor("op_19435_cast_fp16")]; + tensor var_19436_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1807_cast_fp16)[name = tensor("op_19436_cast_fp16")]; + tensor var_19437_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1809_cast_fp16)[name = tensor("op_19437_cast_fp16")]; + tensor var_19438_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1811_cast_fp16)[name = tensor("op_19438_cast_fp16")]; + tensor var_19439_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1813_cast_fp16)[name = tensor("op_19439_cast_fp16")]; + tensor var_19440_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1815_cast_fp16)[name = tensor("op_19440_cast_fp16")]; + tensor var_19441_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1817_cast_fp16)[name = tensor("op_19441_cast_fp16")]; + tensor var_19442_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1819_cast_fp16)[name = tensor("op_19442_cast_fp16")]; + tensor var_19443_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1821_cast_fp16)[name = tensor("op_19443_cast_fp16")]; + tensor var_19444_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1823_cast_fp16)[name = tensor("op_19444_cast_fp16")]; + tensor var_19445_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1825_cast_fp16)[name = tensor("op_19445_cast_fp16")]; + tensor var_19446_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1827_cast_fp16)[name = tensor("op_19446_cast_fp16")]; + tensor var_19447_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1829_cast_fp16)[name = tensor("op_19447_cast_fp16")]; + tensor var_19448_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1831_cast_fp16)[name = tensor("op_19448_cast_fp16")]; + tensor var_19449_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1833_cast_fp16)[name = tensor("op_19449_cast_fp16")]; + tensor var_19450_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1835_cast_fp16)[name = tensor("op_19450_cast_fp16")]; + tensor var_19451_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1837_cast_fp16)[name = tensor("op_19451_cast_fp16")]; + tensor var_19452_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1839_cast_fp16)[name = tensor("op_19452_cast_fp16")]; + tensor var_19453_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1841_cast_fp16)[name = tensor("op_19453_cast_fp16")]; + tensor var_19454_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1843_cast_fp16)[name = tensor("op_19454_cast_fp16")]; + tensor var_19455_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1845_cast_fp16)[name = tensor("op_19455_cast_fp16")]; + tensor var_19456_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1847_cast_fp16)[name = tensor("op_19456_cast_fp16")]; + tensor var_19457_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1849_cast_fp16)[name = tensor("op_19457_cast_fp16")]; + tensor var_19458_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1851_cast_fp16)[name = tensor("op_19458_cast_fp16")]; + tensor var_19459_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1853_cast_fp16)[name = tensor("op_19459_cast_fp16")]; + tensor var_19460_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1855_cast_fp16)[name = tensor("op_19460_cast_fp16")]; + tensor var_19461_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1857_cast_fp16)[name = tensor("op_19461_cast_fp16")]; + tensor var_19462_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1859_cast_fp16)[name = tensor("op_19462_cast_fp16")]; + tensor var_19463_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1861_cast_fp16)[name = tensor("op_19463_cast_fp16")]; + tensor var_19464_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1863_cast_fp16)[name = tensor("op_19464_cast_fp16")]; + tensor var_19465_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1865_cast_fp16)[name = tensor("op_19465_cast_fp16")]; + tensor var_19466_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1867_cast_fp16)[name = tensor("op_19466_cast_fp16")]; + tensor var_19467_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1869_cast_fp16)[name = tensor("op_19467_cast_fp16")]; + tensor var_19468_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1871_cast_fp16)[name = tensor("op_19468_cast_fp16")]; + tensor var_19469_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1873_cast_fp16)[name = tensor("op_19469_cast_fp16")]; + tensor var_19470_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1875_cast_fp16)[name = tensor("op_19470_cast_fp16")]; + tensor var_19471_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1877_cast_fp16)[name = tensor("op_19471_cast_fp16")]; + tensor var_19472_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1879_cast_fp16)[name = tensor("op_19472_cast_fp16")]; + tensor var_19473_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1881_cast_fp16)[name = tensor("op_19473_cast_fp16")]; + tensor var_19474_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1883_cast_fp16)[name = tensor("op_19474_cast_fp16")]; + tensor var_19475_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1885_cast_fp16)[name = tensor("op_19475_cast_fp16")]; + tensor var_19476_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1887_cast_fp16)[name = tensor("op_19476_cast_fp16")]; + tensor var_19477_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1889_cast_fp16)[name = tensor("op_19477_cast_fp16")]; + tensor var_19478_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1891_cast_fp16)[name = tensor("op_19478_cast_fp16")]; + tensor var_19479_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1893_cast_fp16)[name = tensor("op_19479_cast_fp16")]; + tensor var_19480_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1895_cast_fp16)[name = tensor("op_19480_cast_fp16")]; + tensor var_19481_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1897_cast_fp16)[name = tensor("op_19481_cast_fp16")]; + tensor var_19482_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1899_cast_fp16)[name = tensor("op_19482_cast_fp16")]; + tensor var_19483_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1901_cast_fp16)[name = tensor("op_19483_cast_fp16")]; + tensor var_19484_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1903_cast_fp16)[name = tensor("op_19484_cast_fp16")]; + tensor var_19485_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1905_cast_fp16)[name = tensor("op_19485_cast_fp16")]; + tensor var_19486_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1907_cast_fp16)[name = tensor("op_19486_cast_fp16")]; + tensor var_19487_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1909_cast_fp16)[name = tensor("op_19487_cast_fp16")]; + tensor var_19488_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1911_cast_fp16)[name = tensor("op_19488_cast_fp16")]; + tensor var_19489_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1913_cast_fp16)[name = tensor("op_19489_cast_fp16")]; + tensor var_19490_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1915_cast_fp16)[name = tensor("op_19490_cast_fp16")]; + tensor var_19491_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1917_cast_fp16)[name = tensor("op_19491_cast_fp16")]; + tensor var_19492_cast_fp16 = softmax(axis = var_18184, x = aw_chunk_1919_cast_fp16)[name = tensor("op_19492_cast_fp16")]; + tensor var_19494_equation_0 = const()[name = tensor("op_19494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19494_cast_fp16 = einsum(equation = var_19494_equation_0, values = (var_19014_cast_fp16, var_19413_cast_fp16))[name = tensor("op_19494_cast_fp16")]; + tensor var_19496_equation_0 = const()[name = tensor("op_19496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19496_cast_fp16 = einsum(equation = var_19496_equation_0, values = (var_19014_cast_fp16, var_19414_cast_fp16))[name = tensor("op_19496_cast_fp16")]; + tensor var_19498_equation_0 = const()[name = tensor("op_19498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19498_cast_fp16 = einsum(equation = var_19498_equation_0, values = (var_19014_cast_fp16, var_19415_cast_fp16))[name = tensor("op_19498_cast_fp16")]; + tensor var_19500_equation_0 = const()[name = tensor("op_19500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19500_cast_fp16 = einsum(equation = var_19500_equation_0, values = (var_19014_cast_fp16, var_19416_cast_fp16))[name = tensor("op_19500_cast_fp16")]; + tensor var_19502_equation_0 = const()[name = tensor("op_19502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19502_cast_fp16 = einsum(equation = var_19502_equation_0, values = (var_19018_cast_fp16, var_19417_cast_fp16))[name = tensor("op_19502_cast_fp16")]; + tensor var_19504_equation_0 = const()[name = tensor("op_19504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19504_cast_fp16 = einsum(equation = var_19504_equation_0, values = (var_19018_cast_fp16, var_19418_cast_fp16))[name = tensor("op_19504_cast_fp16")]; + tensor var_19506_equation_0 = const()[name = tensor("op_19506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19506_cast_fp16 = einsum(equation = var_19506_equation_0, values = (var_19018_cast_fp16, var_19419_cast_fp16))[name = tensor("op_19506_cast_fp16")]; + tensor var_19508_equation_0 = const()[name = tensor("op_19508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19508_cast_fp16 = einsum(equation = var_19508_equation_0, values = (var_19018_cast_fp16, var_19420_cast_fp16))[name = tensor("op_19508_cast_fp16")]; + tensor var_19510_equation_0 = const()[name = tensor("op_19510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19510_cast_fp16 = einsum(equation = var_19510_equation_0, values = (var_19022_cast_fp16, var_19421_cast_fp16))[name = tensor("op_19510_cast_fp16")]; + tensor var_19512_equation_0 = const()[name = tensor("op_19512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19512_cast_fp16 = einsum(equation = var_19512_equation_0, values = (var_19022_cast_fp16, var_19422_cast_fp16))[name = tensor("op_19512_cast_fp16")]; + tensor var_19514_equation_0 = const()[name = tensor("op_19514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19514_cast_fp16 = einsum(equation = var_19514_equation_0, values = (var_19022_cast_fp16, var_19423_cast_fp16))[name = tensor("op_19514_cast_fp16")]; + tensor var_19516_equation_0 = const()[name = tensor("op_19516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19516_cast_fp16 = einsum(equation = var_19516_equation_0, values = (var_19022_cast_fp16, var_19424_cast_fp16))[name = tensor("op_19516_cast_fp16")]; + tensor var_19518_equation_0 = const()[name = tensor("op_19518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19518_cast_fp16 = einsum(equation = var_19518_equation_0, values = (var_19026_cast_fp16, var_19425_cast_fp16))[name = tensor("op_19518_cast_fp16")]; + tensor var_19520_equation_0 = const()[name = tensor("op_19520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19520_cast_fp16 = einsum(equation = var_19520_equation_0, values = (var_19026_cast_fp16, var_19426_cast_fp16))[name = tensor("op_19520_cast_fp16")]; + tensor var_19522_equation_0 = const()[name = tensor("op_19522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19522_cast_fp16 = einsum(equation = var_19522_equation_0, values = (var_19026_cast_fp16, var_19427_cast_fp16))[name = tensor("op_19522_cast_fp16")]; + tensor var_19524_equation_0 = const()[name = tensor("op_19524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19524_cast_fp16 = einsum(equation = var_19524_equation_0, values = (var_19026_cast_fp16, var_19428_cast_fp16))[name = tensor("op_19524_cast_fp16")]; + tensor var_19526_equation_0 = const()[name = tensor("op_19526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19526_cast_fp16 = einsum(equation = var_19526_equation_0, values = (var_19030_cast_fp16, var_19429_cast_fp16))[name = tensor("op_19526_cast_fp16")]; + tensor var_19528_equation_0 = const()[name = tensor("op_19528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19528_cast_fp16 = einsum(equation = var_19528_equation_0, values = (var_19030_cast_fp16, var_19430_cast_fp16))[name = tensor("op_19528_cast_fp16")]; + tensor var_19530_equation_0 = const()[name = tensor("op_19530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19530_cast_fp16 = einsum(equation = var_19530_equation_0, values = (var_19030_cast_fp16, var_19431_cast_fp16))[name = tensor("op_19530_cast_fp16")]; + tensor var_19532_equation_0 = const()[name = tensor("op_19532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19532_cast_fp16 = einsum(equation = var_19532_equation_0, values = (var_19030_cast_fp16, var_19432_cast_fp16))[name = tensor("op_19532_cast_fp16")]; + tensor var_19534_equation_0 = const()[name = tensor("op_19534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19534_cast_fp16 = einsum(equation = var_19534_equation_0, values = (var_19034_cast_fp16, var_19433_cast_fp16))[name = tensor("op_19534_cast_fp16")]; + tensor var_19536_equation_0 = const()[name = tensor("op_19536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19536_cast_fp16 = einsum(equation = var_19536_equation_0, values = (var_19034_cast_fp16, var_19434_cast_fp16))[name = tensor("op_19536_cast_fp16")]; + tensor var_19538_equation_0 = const()[name = tensor("op_19538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19538_cast_fp16 = einsum(equation = var_19538_equation_0, values = (var_19034_cast_fp16, var_19435_cast_fp16))[name = tensor("op_19538_cast_fp16")]; + tensor var_19540_equation_0 = const()[name = tensor("op_19540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19540_cast_fp16 = einsum(equation = var_19540_equation_0, values = (var_19034_cast_fp16, var_19436_cast_fp16))[name = tensor("op_19540_cast_fp16")]; + tensor var_19542_equation_0 = const()[name = tensor("op_19542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19542_cast_fp16 = einsum(equation = var_19542_equation_0, values = (var_19038_cast_fp16, var_19437_cast_fp16))[name = tensor("op_19542_cast_fp16")]; + tensor var_19544_equation_0 = const()[name = tensor("op_19544_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19544_cast_fp16 = einsum(equation = var_19544_equation_0, values = (var_19038_cast_fp16, var_19438_cast_fp16))[name = tensor("op_19544_cast_fp16")]; + tensor var_19546_equation_0 = const()[name = tensor("op_19546_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19546_cast_fp16 = einsum(equation = var_19546_equation_0, values = (var_19038_cast_fp16, var_19439_cast_fp16))[name = tensor("op_19546_cast_fp16")]; + tensor var_19548_equation_0 = const()[name = tensor("op_19548_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19548_cast_fp16 = einsum(equation = var_19548_equation_0, values = (var_19038_cast_fp16, var_19440_cast_fp16))[name = tensor("op_19548_cast_fp16")]; + tensor var_19550_equation_0 = const()[name = tensor("op_19550_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19550_cast_fp16 = einsum(equation = var_19550_equation_0, values = (var_19042_cast_fp16, var_19441_cast_fp16))[name = tensor("op_19550_cast_fp16")]; + tensor var_19552_equation_0 = const()[name = tensor("op_19552_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19552_cast_fp16 = einsum(equation = var_19552_equation_0, values = (var_19042_cast_fp16, var_19442_cast_fp16))[name = tensor("op_19552_cast_fp16")]; + tensor var_19554_equation_0 = const()[name = tensor("op_19554_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19554_cast_fp16 = einsum(equation = var_19554_equation_0, values = (var_19042_cast_fp16, var_19443_cast_fp16))[name = tensor("op_19554_cast_fp16")]; + tensor var_19556_equation_0 = const()[name = tensor("op_19556_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19556_cast_fp16 = einsum(equation = var_19556_equation_0, values = (var_19042_cast_fp16, var_19444_cast_fp16))[name = tensor("op_19556_cast_fp16")]; + tensor var_19558_equation_0 = const()[name = tensor("op_19558_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19558_cast_fp16 = einsum(equation = var_19558_equation_0, values = (var_19046_cast_fp16, var_19445_cast_fp16))[name = tensor("op_19558_cast_fp16")]; + tensor var_19560_equation_0 = const()[name = tensor("op_19560_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19560_cast_fp16 = einsum(equation = var_19560_equation_0, values = (var_19046_cast_fp16, var_19446_cast_fp16))[name = tensor("op_19560_cast_fp16")]; + tensor var_19562_equation_0 = const()[name = tensor("op_19562_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19562_cast_fp16 = einsum(equation = var_19562_equation_0, values = (var_19046_cast_fp16, var_19447_cast_fp16))[name = tensor("op_19562_cast_fp16")]; + tensor var_19564_equation_0 = const()[name = tensor("op_19564_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19564_cast_fp16 = einsum(equation = var_19564_equation_0, values = (var_19046_cast_fp16, var_19448_cast_fp16))[name = tensor("op_19564_cast_fp16")]; + tensor var_19566_equation_0 = const()[name = tensor("op_19566_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19566_cast_fp16 = einsum(equation = var_19566_equation_0, values = (var_19050_cast_fp16, var_19449_cast_fp16))[name = tensor("op_19566_cast_fp16")]; + tensor var_19568_equation_0 = const()[name = tensor("op_19568_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19568_cast_fp16 = einsum(equation = var_19568_equation_0, values = (var_19050_cast_fp16, var_19450_cast_fp16))[name = tensor("op_19568_cast_fp16")]; + tensor var_19570_equation_0 = const()[name = tensor("op_19570_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19570_cast_fp16 = einsum(equation = var_19570_equation_0, values = (var_19050_cast_fp16, var_19451_cast_fp16))[name = tensor("op_19570_cast_fp16")]; + tensor var_19572_equation_0 = const()[name = tensor("op_19572_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19572_cast_fp16 = einsum(equation = var_19572_equation_0, values = (var_19050_cast_fp16, var_19452_cast_fp16))[name = tensor("op_19572_cast_fp16")]; + tensor var_19574_equation_0 = const()[name = tensor("op_19574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19574_cast_fp16 = einsum(equation = var_19574_equation_0, values = (var_19054_cast_fp16, var_19453_cast_fp16))[name = tensor("op_19574_cast_fp16")]; + tensor var_19576_equation_0 = const()[name = tensor("op_19576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19576_cast_fp16 = einsum(equation = var_19576_equation_0, values = (var_19054_cast_fp16, var_19454_cast_fp16))[name = tensor("op_19576_cast_fp16")]; + tensor var_19578_equation_0 = const()[name = tensor("op_19578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19578_cast_fp16 = einsum(equation = var_19578_equation_0, values = (var_19054_cast_fp16, var_19455_cast_fp16))[name = tensor("op_19578_cast_fp16")]; + tensor var_19580_equation_0 = const()[name = tensor("op_19580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19580_cast_fp16 = einsum(equation = var_19580_equation_0, values = (var_19054_cast_fp16, var_19456_cast_fp16))[name = tensor("op_19580_cast_fp16")]; + tensor var_19582_equation_0 = const()[name = tensor("op_19582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19582_cast_fp16 = einsum(equation = var_19582_equation_0, values = (var_19058_cast_fp16, var_19457_cast_fp16))[name = tensor("op_19582_cast_fp16")]; + tensor var_19584_equation_0 = const()[name = tensor("op_19584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19584_cast_fp16 = einsum(equation = var_19584_equation_0, values = (var_19058_cast_fp16, var_19458_cast_fp16))[name = tensor("op_19584_cast_fp16")]; + tensor var_19586_equation_0 = const()[name = tensor("op_19586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19586_cast_fp16 = einsum(equation = var_19586_equation_0, values = (var_19058_cast_fp16, var_19459_cast_fp16))[name = tensor("op_19586_cast_fp16")]; + tensor var_19588_equation_0 = const()[name = tensor("op_19588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19588_cast_fp16 = einsum(equation = var_19588_equation_0, values = (var_19058_cast_fp16, var_19460_cast_fp16))[name = tensor("op_19588_cast_fp16")]; + tensor var_19590_equation_0 = const()[name = tensor("op_19590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19590_cast_fp16 = einsum(equation = var_19590_equation_0, values = (var_19062_cast_fp16, var_19461_cast_fp16))[name = tensor("op_19590_cast_fp16")]; + tensor var_19592_equation_0 = const()[name = tensor("op_19592_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19592_cast_fp16 = einsum(equation = var_19592_equation_0, values = (var_19062_cast_fp16, var_19462_cast_fp16))[name = tensor("op_19592_cast_fp16")]; + tensor var_19594_equation_0 = const()[name = tensor("op_19594_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19594_cast_fp16 = einsum(equation = var_19594_equation_0, values = (var_19062_cast_fp16, var_19463_cast_fp16))[name = tensor("op_19594_cast_fp16")]; + tensor var_19596_equation_0 = const()[name = tensor("op_19596_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19596_cast_fp16 = einsum(equation = var_19596_equation_0, values = (var_19062_cast_fp16, var_19464_cast_fp16))[name = tensor("op_19596_cast_fp16")]; + tensor var_19598_equation_0 = const()[name = tensor("op_19598_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19598_cast_fp16 = einsum(equation = var_19598_equation_0, values = (var_19066_cast_fp16, var_19465_cast_fp16))[name = tensor("op_19598_cast_fp16")]; + tensor var_19600_equation_0 = const()[name = tensor("op_19600_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19600_cast_fp16 = einsum(equation = var_19600_equation_0, values = (var_19066_cast_fp16, var_19466_cast_fp16))[name = tensor("op_19600_cast_fp16")]; + tensor var_19602_equation_0 = const()[name = tensor("op_19602_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19602_cast_fp16 = einsum(equation = var_19602_equation_0, values = (var_19066_cast_fp16, var_19467_cast_fp16))[name = tensor("op_19602_cast_fp16")]; + tensor var_19604_equation_0 = const()[name = tensor("op_19604_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19604_cast_fp16 = einsum(equation = var_19604_equation_0, values = (var_19066_cast_fp16, var_19468_cast_fp16))[name = tensor("op_19604_cast_fp16")]; + tensor var_19606_equation_0 = const()[name = tensor("op_19606_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19606_cast_fp16 = einsum(equation = var_19606_equation_0, values = (var_19070_cast_fp16, var_19469_cast_fp16))[name = tensor("op_19606_cast_fp16")]; + tensor var_19608_equation_0 = const()[name = tensor("op_19608_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19608_cast_fp16 = einsum(equation = var_19608_equation_0, values = (var_19070_cast_fp16, var_19470_cast_fp16))[name = tensor("op_19608_cast_fp16")]; + tensor var_19610_equation_0 = const()[name = tensor("op_19610_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19610_cast_fp16 = einsum(equation = var_19610_equation_0, values = (var_19070_cast_fp16, var_19471_cast_fp16))[name = tensor("op_19610_cast_fp16")]; + tensor var_19612_equation_0 = const()[name = tensor("op_19612_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19612_cast_fp16 = einsum(equation = var_19612_equation_0, values = (var_19070_cast_fp16, var_19472_cast_fp16))[name = tensor("op_19612_cast_fp16")]; + tensor var_19614_equation_0 = const()[name = tensor("op_19614_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19614_cast_fp16 = einsum(equation = var_19614_equation_0, values = (var_19074_cast_fp16, var_19473_cast_fp16))[name = tensor("op_19614_cast_fp16")]; + tensor var_19616_equation_0 = const()[name = tensor("op_19616_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19616_cast_fp16 = einsum(equation = var_19616_equation_0, values = (var_19074_cast_fp16, var_19474_cast_fp16))[name = tensor("op_19616_cast_fp16")]; + tensor var_19618_equation_0 = const()[name = tensor("op_19618_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19618_cast_fp16 = einsum(equation = var_19618_equation_0, values = (var_19074_cast_fp16, var_19475_cast_fp16))[name = tensor("op_19618_cast_fp16")]; + tensor var_19620_equation_0 = const()[name = tensor("op_19620_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19620_cast_fp16 = einsum(equation = var_19620_equation_0, values = (var_19074_cast_fp16, var_19476_cast_fp16))[name = tensor("op_19620_cast_fp16")]; + tensor var_19622_equation_0 = const()[name = tensor("op_19622_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19622_cast_fp16 = einsum(equation = var_19622_equation_0, values = (var_19078_cast_fp16, var_19477_cast_fp16))[name = tensor("op_19622_cast_fp16")]; + tensor var_19624_equation_0 = const()[name = tensor("op_19624_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19624_cast_fp16 = einsum(equation = var_19624_equation_0, values = (var_19078_cast_fp16, var_19478_cast_fp16))[name = tensor("op_19624_cast_fp16")]; + tensor var_19626_equation_0 = const()[name = tensor("op_19626_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19626_cast_fp16 = einsum(equation = var_19626_equation_0, values = (var_19078_cast_fp16, var_19479_cast_fp16))[name = tensor("op_19626_cast_fp16")]; + tensor var_19628_equation_0 = const()[name = tensor("op_19628_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19628_cast_fp16 = einsum(equation = var_19628_equation_0, values = (var_19078_cast_fp16, var_19480_cast_fp16))[name = tensor("op_19628_cast_fp16")]; + tensor var_19630_equation_0 = const()[name = tensor("op_19630_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19630_cast_fp16 = einsum(equation = var_19630_equation_0, values = (var_19082_cast_fp16, var_19481_cast_fp16))[name = tensor("op_19630_cast_fp16")]; + tensor var_19632_equation_0 = const()[name = tensor("op_19632_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19632_cast_fp16 = einsum(equation = var_19632_equation_0, values = (var_19082_cast_fp16, var_19482_cast_fp16))[name = tensor("op_19632_cast_fp16")]; + tensor var_19634_equation_0 = const()[name = tensor("op_19634_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19634_cast_fp16 = einsum(equation = var_19634_equation_0, values = (var_19082_cast_fp16, var_19483_cast_fp16))[name = tensor("op_19634_cast_fp16")]; + tensor var_19636_equation_0 = const()[name = tensor("op_19636_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19636_cast_fp16 = einsum(equation = var_19636_equation_0, values = (var_19082_cast_fp16, var_19484_cast_fp16))[name = tensor("op_19636_cast_fp16")]; + tensor var_19638_equation_0 = const()[name = tensor("op_19638_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19638_cast_fp16 = einsum(equation = var_19638_equation_0, values = (var_19086_cast_fp16, var_19485_cast_fp16))[name = tensor("op_19638_cast_fp16")]; + tensor var_19640_equation_0 = const()[name = tensor("op_19640_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19640_cast_fp16 = einsum(equation = var_19640_equation_0, values = (var_19086_cast_fp16, var_19486_cast_fp16))[name = tensor("op_19640_cast_fp16")]; + tensor var_19642_equation_0 = const()[name = tensor("op_19642_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19642_cast_fp16 = einsum(equation = var_19642_equation_0, values = (var_19086_cast_fp16, var_19487_cast_fp16))[name = tensor("op_19642_cast_fp16")]; + tensor var_19644_equation_0 = const()[name = tensor("op_19644_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19644_cast_fp16 = einsum(equation = var_19644_equation_0, values = (var_19086_cast_fp16, var_19488_cast_fp16))[name = tensor("op_19644_cast_fp16")]; + tensor var_19646_equation_0 = const()[name = tensor("op_19646_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19646_cast_fp16 = einsum(equation = var_19646_equation_0, values = (var_19090_cast_fp16, var_19489_cast_fp16))[name = tensor("op_19646_cast_fp16")]; + tensor var_19648_equation_0 = const()[name = tensor("op_19648_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19648_cast_fp16 = einsum(equation = var_19648_equation_0, values = (var_19090_cast_fp16, var_19490_cast_fp16))[name = tensor("op_19648_cast_fp16")]; + tensor var_19650_equation_0 = const()[name = tensor("op_19650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19650_cast_fp16 = einsum(equation = var_19650_equation_0, values = (var_19090_cast_fp16, var_19491_cast_fp16))[name = tensor("op_19650_cast_fp16")]; + tensor var_19652_equation_0 = const()[name = tensor("op_19652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_19652_cast_fp16 = einsum(equation = var_19652_equation_0, values = (var_19090_cast_fp16, var_19492_cast_fp16))[name = tensor("op_19652_cast_fp16")]; + tensor var_19654_interleave_0 = const()[name = tensor("op_19654_interleave_0"), val = tensor(false)]; + tensor var_19654_cast_fp16 = concat(axis = var_18159, interleave = var_19654_interleave_0, values = (var_19494_cast_fp16, var_19496_cast_fp16, var_19498_cast_fp16, var_19500_cast_fp16))[name = tensor("op_19654_cast_fp16")]; + tensor var_19656_interleave_0 = const()[name = tensor("op_19656_interleave_0"), val = tensor(false)]; + tensor var_19656_cast_fp16 = concat(axis = var_18159, interleave = var_19656_interleave_0, values = (var_19502_cast_fp16, var_19504_cast_fp16, var_19506_cast_fp16, var_19508_cast_fp16))[name = tensor("op_19656_cast_fp16")]; + tensor var_19658_interleave_0 = const()[name = tensor("op_19658_interleave_0"), val = tensor(false)]; + tensor var_19658_cast_fp16 = concat(axis = var_18159, interleave = var_19658_interleave_0, values = (var_19510_cast_fp16, var_19512_cast_fp16, var_19514_cast_fp16, var_19516_cast_fp16))[name = tensor("op_19658_cast_fp16")]; + tensor var_19660_interleave_0 = const()[name = tensor("op_19660_interleave_0"), val = tensor(false)]; + tensor var_19660_cast_fp16 = concat(axis = var_18159, interleave = var_19660_interleave_0, values = (var_19518_cast_fp16, var_19520_cast_fp16, var_19522_cast_fp16, var_19524_cast_fp16))[name = tensor("op_19660_cast_fp16")]; + tensor var_19662_interleave_0 = const()[name = tensor("op_19662_interleave_0"), val = tensor(false)]; + tensor var_19662_cast_fp16 = concat(axis = var_18159, interleave = var_19662_interleave_0, values = (var_19526_cast_fp16, var_19528_cast_fp16, var_19530_cast_fp16, var_19532_cast_fp16))[name = tensor("op_19662_cast_fp16")]; + tensor var_19664_interleave_0 = const()[name = tensor("op_19664_interleave_0"), val = tensor(false)]; + tensor var_19664_cast_fp16 = concat(axis = var_18159, interleave = var_19664_interleave_0, values = (var_19534_cast_fp16, var_19536_cast_fp16, var_19538_cast_fp16, var_19540_cast_fp16))[name = tensor("op_19664_cast_fp16")]; + tensor var_19666_interleave_0 = const()[name = tensor("op_19666_interleave_0"), val = tensor(false)]; + tensor var_19666_cast_fp16 = concat(axis = var_18159, interleave = var_19666_interleave_0, values = (var_19542_cast_fp16, var_19544_cast_fp16, var_19546_cast_fp16, var_19548_cast_fp16))[name = tensor("op_19666_cast_fp16")]; + tensor var_19668_interleave_0 = const()[name = tensor("op_19668_interleave_0"), val = tensor(false)]; + tensor var_19668_cast_fp16 = concat(axis = var_18159, interleave = var_19668_interleave_0, values = (var_19550_cast_fp16, var_19552_cast_fp16, var_19554_cast_fp16, var_19556_cast_fp16))[name = tensor("op_19668_cast_fp16")]; + tensor var_19670_interleave_0 = const()[name = tensor("op_19670_interleave_0"), val = tensor(false)]; + tensor var_19670_cast_fp16 = concat(axis = var_18159, interleave = var_19670_interleave_0, values = (var_19558_cast_fp16, var_19560_cast_fp16, var_19562_cast_fp16, var_19564_cast_fp16))[name = tensor("op_19670_cast_fp16")]; + tensor var_19672_interleave_0 = const()[name = tensor("op_19672_interleave_0"), val = tensor(false)]; + tensor var_19672_cast_fp16 = concat(axis = var_18159, interleave = var_19672_interleave_0, values = (var_19566_cast_fp16, var_19568_cast_fp16, var_19570_cast_fp16, var_19572_cast_fp16))[name = tensor("op_19672_cast_fp16")]; + tensor var_19674_interleave_0 = const()[name = tensor("op_19674_interleave_0"), val = tensor(false)]; + tensor var_19674_cast_fp16 = concat(axis = var_18159, interleave = var_19674_interleave_0, values = (var_19574_cast_fp16, var_19576_cast_fp16, var_19578_cast_fp16, var_19580_cast_fp16))[name = tensor("op_19674_cast_fp16")]; + tensor var_19676_interleave_0 = const()[name = tensor("op_19676_interleave_0"), val = tensor(false)]; + tensor var_19676_cast_fp16 = concat(axis = var_18159, interleave = var_19676_interleave_0, values = (var_19582_cast_fp16, var_19584_cast_fp16, var_19586_cast_fp16, var_19588_cast_fp16))[name = tensor("op_19676_cast_fp16")]; + tensor var_19678_interleave_0 = const()[name = tensor("op_19678_interleave_0"), val = tensor(false)]; + tensor var_19678_cast_fp16 = concat(axis = var_18159, interleave = var_19678_interleave_0, values = (var_19590_cast_fp16, var_19592_cast_fp16, var_19594_cast_fp16, var_19596_cast_fp16))[name = tensor("op_19678_cast_fp16")]; + tensor var_19680_interleave_0 = const()[name = tensor("op_19680_interleave_0"), val = tensor(false)]; + tensor var_19680_cast_fp16 = concat(axis = var_18159, interleave = var_19680_interleave_0, values = (var_19598_cast_fp16, var_19600_cast_fp16, var_19602_cast_fp16, var_19604_cast_fp16))[name = tensor("op_19680_cast_fp16")]; + tensor var_19682_interleave_0 = const()[name = tensor("op_19682_interleave_0"), val = tensor(false)]; + tensor var_19682_cast_fp16 = concat(axis = var_18159, interleave = var_19682_interleave_0, values = (var_19606_cast_fp16, var_19608_cast_fp16, var_19610_cast_fp16, var_19612_cast_fp16))[name = tensor("op_19682_cast_fp16")]; + tensor var_19684_interleave_0 = const()[name = tensor("op_19684_interleave_0"), val = tensor(false)]; + tensor var_19684_cast_fp16 = concat(axis = var_18159, interleave = var_19684_interleave_0, values = (var_19614_cast_fp16, var_19616_cast_fp16, var_19618_cast_fp16, var_19620_cast_fp16))[name = tensor("op_19684_cast_fp16")]; + tensor var_19686_interleave_0 = const()[name = tensor("op_19686_interleave_0"), val = tensor(false)]; + tensor var_19686_cast_fp16 = concat(axis = var_18159, interleave = var_19686_interleave_0, values = (var_19622_cast_fp16, var_19624_cast_fp16, var_19626_cast_fp16, var_19628_cast_fp16))[name = tensor("op_19686_cast_fp16")]; + tensor var_19688_interleave_0 = const()[name = tensor("op_19688_interleave_0"), val = tensor(false)]; + tensor var_19688_cast_fp16 = concat(axis = var_18159, interleave = var_19688_interleave_0, values = (var_19630_cast_fp16, var_19632_cast_fp16, var_19634_cast_fp16, var_19636_cast_fp16))[name = tensor("op_19688_cast_fp16")]; + tensor var_19690_interleave_0 = const()[name = tensor("op_19690_interleave_0"), val = tensor(false)]; + tensor var_19690_cast_fp16 = concat(axis = var_18159, interleave = var_19690_interleave_0, values = (var_19638_cast_fp16, var_19640_cast_fp16, var_19642_cast_fp16, var_19644_cast_fp16))[name = tensor("op_19690_cast_fp16")]; + tensor var_19692_interleave_0 = const()[name = tensor("op_19692_interleave_0"), val = tensor(false)]; + tensor var_19692_cast_fp16 = concat(axis = var_18159, interleave = var_19692_interleave_0, values = (var_19646_cast_fp16, var_19648_cast_fp16, var_19650_cast_fp16, var_19652_cast_fp16))[name = tensor("op_19692_cast_fp16")]; + tensor input_227_interleave_0 = const()[name = tensor("input_227_interleave_0"), val = tensor(false)]; + tensor input_227_cast_fp16 = concat(axis = var_18184, interleave = input_227_interleave_0, values = (var_19654_cast_fp16, var_19656_cast_fp16, var_19658_cast_fp16, var_19660_cast_fp16, var_19662_cast_fp16, var_19664_cast_fp16, var_19666_cast_fp16, var_19668_cast_fp16, var_19670_cast_fp16, var_19672_cast_fp16, var_19674_cast_fp16, var_19676_cast_fp16, var_19678_cast_fp16, var_19680_cast_fp16, var_19682_cast_fp16, var_19684_cast_fp16, var_19686_cast_fp16, var_19688_cast_fp16, var_19690_cast_fp16, var_19692_cast_fp16))[name = tensor("input_227_cast_fp16")]; + tensor var_19700 = const()[name = tensor("op_19700"), val = tensor([1, 1])]; + tensor var_19702 = const()[name = tensor("op_19702"), val = tensor([1, 1])]; + tensor pretrained_out_139_pad_type_0 = const()[name = tensor("pretrained_out_139_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_139_pad_0 = const()[name = tensor("pretrained_out_139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133630528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134449792))), name = tensor("layers_11_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_11_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134449920)))]; + tensor pretrained_out_139_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_19702, groups = var_18184, pad = pretrained_out_139_pad_0, pad_type = pretrained_out_139_pad_type_0, strides = var_19700, weight = layers_11_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = tensor("pretrained_out_139_cast_fp16")]; + tensor var_19706 = const()[name = tensor("op_19706"), val = tensor([1, 1])]; + tensor var_19708 = const()[name = tensor("op_19708"), val = tensor([1, 1])]; + tensor input_229_pad_type_0 = const()[name = tensor("input_229_pad_type_0"), val = tensor("custom")]; + tensor input_229_pad_0 = const()[name = tensor("input_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134452544)))]; + tensor input_229_cast_fp16 = conv(dilations = var_19708, groups = var_18184, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = var_19706, weight = layers_11_self_attn_o_proj_loraA_weight_to_fp16, x = input_227_cast_fp16)[name = tensor("input_229_cast_fp16")]; + tensor var_19712 = const()[name = tensor("op_19712"), val = tensor([1, 1])]; + tensor var_19714 = const()[name = tensor("op_19714"), val = tensor([1, 1])]; + tensor lora_out_277_pad_type_0 = const()[name = tensor("lora_out_277_pad_type_0"), val = tensor("custom")]; + tensor lora_out_277_pad_0 = const()[name = tensor("lora_out_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_279_weight_0_to_fp16 = const()[name = tensor("lora_out_279_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134493568)))]; + tensor lora_out_279_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_19714, groups = var_18184, pad = lora_out_277_pad_0, pad_type = lora_out_277_pad_type_0, strides = var_19712, weight = lora_out_279_weight_0_to_fp16, x = input_229_cast_fp16)[name = tensor("lora_out_279_cast_fp16")]; + tensor obj_47_cast_fp16 = add(x = pretrained_out_139_cast_fp16, y = lora_out_279_cast_fp16)[name = tensor("obj_47_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_47_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; + tensor var_19723 = const()[name = tensor("op_19723"), val = tensor([1])]; + tensor channels_mean_47_cast_fp16 = reduce_mean(axes = var_19723, keep_dims = var_18185, x = inputs_47_cast_fp16)[name = tensor("channels_mean_47_cast_fp16")]; + tensor zero_mean_47_cast_fp16 = sub(x = inputs_47_cast_fp16, y = channels_mean_47_cast_fp16)[name = tensor("zero_mean_47_cast_fp16")]; + tensor zero_mean_sq_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = zero_mean_47_cast_fp16)[name = tensor("zero_mean_sq_47_cast_fp16")]; + tensor var_19727 = const()[name = tensor("op_19727"), val = tensor([1])]; + tensor var_19728_cast_fp16 = reduce_mean(axes = var_19727, keep_dims = var_18185, x = zero_mean_sq_47_cast_fp16)[name = tensor("op_19728_cast_fp16")]; + tensor var_19729_to_fp16 = const()[name = tensor("op_19729_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_19730_cast_fp16 = add(x = var_19728_cast_fp16, y = var_19729_to_fp16)[name = tensor("op_19730_cast_fp16")]; + tensor denom_47_epsilon_0 = const()[name = tensor("denom_47_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_47_cast_fp16 = rsqrt(epsilon = denom_47_epsilon_0, x = var_19730_cast_fp16)[name = tensor("denom_47_cast_fp16")]; + tensor out_47_cast_fp16 = mul(x = zero_mean_47_cast_fp16, y = denom_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; + tensor input_231_gamma_0_to_fp16 = const()[name = tensor("input_231_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134534592)))]; + tensor input_231_beta_0_to_fp16 = const()[name = tensor("input_231_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134537216)))]; + tensor input_231_epsilon_0_to_fp16 = const()[name = tensor("input_231_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_231_cast_fp16 = batch_norm(beta = input_231_beta_0_to_fp16, epsilon = input_231_epsilon_0_to_fp16, gamma = input_231_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor("input_231_cast_fp16")]; + tensor var_19744 = const()[name = tensor("op_19744"), val = tensor([1, 1])]; + tensor var_19746 = const()[name = tensor("op_19746"), val = tensor([1, 1])]; + tensor pretrained_out_141_pad_type_0 = const()[name = tensor("pretrained_out_141_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_141_pad_0 = const()[name = tensor("pretrained_out_141_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134539840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137816704))), name = tensor("layers_11_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_11_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_11_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137816832)))]; + tensor pretrained_out_141_cast_fp16 = conv(bias = layers_11_fc1_pretrained_bias_to_fp16, dilations = var_19746, groups = var_18184, pad = pretrained_out_141_pad_0, pad_type = pretrained_out_141_pad_type_0, strides = var_19744, weight = layers_11_fc1_pretrained_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = tensor("pretrained_out_141_cast_fp16")]; + tensor var_19750 = const()[name = tensor("op_19750"), val = tensor([1, 1])]; + tensor var_19752 = const()[name = tensor("op_19752"), val = tensor([1, 1])]; + tensor input_233_pad_type_0 = const()[name = tensor("input_233_pad_type_0"), val = tensor("custom")]; + tensor input_233_pad_0 = const()[name = tensor("input_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_11_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137827136)))]; + tensor input_233_cast_fp16 = conv(dilations = var_19752, groups = var_18184, pad = input_233_pad_0, pad_type = input_233_pad_type_0, strides = var_19750, weight = layers_11_fc1_loraA_weight_to_fp16, x = input_231_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor var_19756 = const()[name = tensor("op_19756"), val = tensor([1, 1])]; + tensor var_19758 = const()[name = tensor("op_19758"), val = tensor([1, 1])]; + tensor lora_out_281_pad_type_0 = const()[name = tensor("lora_out_281_pad_type_0"), val = tensor("custom")]; + tensor lora_out_281_pad_0 = const()[name = tensor("lora_out_281_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_283_weight_0_to_fp16 = const()[name = tensor("lora_out_283_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137868160)))]; + tensor lora_out_283_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_19758, groups = var_18184, pad = lora_out_281_pad_0, pad_type = lora_out_281_pad_type_0, strides = var_19756, weight = lora_out_283_weight_0_to_fp16, x = input_233_cast_fp16)[name = tensor("lora_out_283_cast_fp16")]; + tensor input_235_cast_fp16 = add(x = pretrained_out_141_cast_fp16, y = lora_out_283_cast_fp16)[name = tensor("input_235_cast_fp16")]; + tensor input_237_mode_0 = const()[name = tensor("input_237_mode_0"), val = tensor("EXACT")]; + tensor input_237_cast_fp16 = gelu(mode = input_237_mode_0, x = input_235_cast_fp16)[name = tensor("input_237_cast_fp16")]; + tensor var_19770 = const()[name = tensor("op_19770"), val = tensor([1, 1])]; + tensor var_19772 = const()[name = tensor("op_19772"), val = tensor([1, 1])]; + tensor pretrained_out_143_pad_type_0 = const()[name = tensor("pretrained_out_143_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_143_pad_0 = const()[name = tensor("pretrained_out_143_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138032064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141308928))), name = tensor("layers_11_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_11_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_11_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141309056)))]; + tensor pretrained_out_143_cast_fp16 = conv(bias = layers_11_fc2_pretrained_bias_to_fp16, dilations = var_19772, groups = var_18184, pad = pretrained_out_143_pad_0, pad_type = pretrained_out_143_pad_type_0, strides = var_19770, weight = layers_11_fc2_pretrained_weight_to_fp16_palettized, x = input_237_cast_fp16)[name = tensor("pretrained_out_143_cast_fp16")]; + tensor var_19776 = const()[name = tensor("op_19776"), val = tensor([1, 1])]; + tensor var_19778 = const()[name = tensor("op_19778"), val = tensor([1, 1])]; + tensor input_239_pad_type_0 = const()[name = tensor("input_239_pad_type_0"), val = tensor("custom")]; + tensor input_239_pad_0 = const()[name = tensor("input_239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_11_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_11_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141311680)))]; + tensor input_239_cast_fp16 = conv(dilations = var_19778, groups = var_18184, pad = input_239_pad_0, pad_type = input_239_pad_type_0, strides = var_19776, weight = layers_11_fc2_loraA_weight_to_fp16, x = input_237_cast_fp16)[name = tensor("input_239_cast_fp16")]; + tensor var_19782 = const()[name = tensor("op_19782"), val = tensor([1, 1])]; + tensor var_19784 = const()[name = tensor("op_19784"), val = tensor([1, 1])]; + tensor lora_out_285_pad_type_0 = const()[name = tensor("lora_out_285_pad_type_0"), val = tensor("custom")]; + tensor lora_out_285_pad_0 = const()[name = tensor("lora_out_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_287_weight_0_to_fp16 = const()[name = tensor("lora_out_287_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141475584)))]; + tensor lora_out_287_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_19784, groups = var_18184, pad = lora_out_285_pad_0, pad_type = lora_out_285_pad_type_0, strides = var_19782, weight = lora_out_287_weight_0_to_fp16, x = input_239_cast_fp16)[name = tensor("lora_out_287_cast_fp16")]; + tensor hidden_states_27_cast_fp16 = add(x = pretrained_out_143_cast_fp16, y = lora_out_287_cast_fp16)[name = tensor("hidden_states_27_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_27_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; + tensor var_19794 = const()[name = tensor("op_19794"), val = tensor(3)]; + tensor var_19819 = const()[name = tensor("op_19819"), val = tensor(1)]; + tensor var_19820 = const()[name = tensor("op_19820"), val = tensor(true)]; + tensor var_19830 = const()[name = tensor("op_19830"), val = tensor([1])]; + tensor channels_mean_49_cast_fp16 = reduce_mean(axes = var_19830, keep_dims = var_19820, x = inputs_49_cast_fp16)[name = tensor("channels_mean_49_cast_fp16")]; + tensor zero_mean_49_cast_fp16 = sub(x = inputs_49_cast_fp16, y = channels_mean_49_cast_fp16)[name = tensor("zero_mean_49_cast_fp16")]; + tensor zero_mean_sq_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = zero_mean_49_cast_fp16)[name = tensor("zero_mean_sq_49_cast_fp16")]; + tensor var_19834 = const()[name = tensor("op_19834"), val = tensor([1])]; + tensor var_19835_cast_fp16 = reduce_mean(axes = var_19834, keep_dims = var_19820, x = zero_mean_sq_49_cast_fp16)[name = tensor("op_19835_cast_fp16")]; + tensor var_19836_to_fp16 = const()[name = tensor("op_19836_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_19837_cast_fp16 = add(x = var_19835_cast_fp16, y = var_19836_to_fp16)[name = tensor("op_19837_cast_fp16")]; + tensor denom_49_epsilon_0 = const()[name = tensor("denom_49_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_49_cast_fp16 = rsqrt(epsilon = denom_49_epsilon_0, x = var_19837_cast_fp16)[name = tensor("denom_49_cast_fp16")]; + tensor out_49_cast_fp16 = mul(x = zero_mean_49_cast_fp16, y = denom_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; + tensor obj_49_gamma_0_to_fp16 = const()[name = tensor("obj_49_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141516608)))]; + tensor obj_49_beta_0_to_fp16 = const()[name = tensor("obj_49_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141519232)))]; + tensor obj_49_epsilon_0_to_fp16 = const()[name = tensor("obj_49_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_49_cast_fp16 = batch_norm(beta = obj_49_beta_0_to_fp16, epsilon = obj_49_epsilon_0_to_fp16, gamma = obj_49_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor("obj_49_cast_fp16")]; + tensor var_19855 = const()[name = tensor("op_19855"), val = tensor([1, 1])]; + tensor var_19857 = const()[name = tensor("op_19857"), val = tensor([1, 1])]; + tensor pretrained_out_145_pad_type_0 = const()[name = tensor("pretrained_out_145_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_145_pad_0 = const()[name = tensor("pretrained_out_145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(141521856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142341120))), name = tensor("layers_12_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_12_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142341248)))]; + tensor pretrained_out_145_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_19857, groups = var_19819, pad = pretrained_out_145_pad_0, pad_type = pretrained_out_145_pad_type_0, strides = var_19855, weight = layers_12_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_49_cast_fp16)[name = tensor("pretrained_out_145_cast_fp16")]; + tensor var_19861 = const()[name = tensor("op_19861"), val = tensor([1, 1])]; + tensor var_19863 = const()[name = tensor("op_19863"), val = tensor([1, 1])]; + tensor input_241_pad_type_0 = const()[name = tensor("input_241_pad_type_0"), val = tensor("custom")]; + tensor input_241_pad_0 = const()[name = tensor("input_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142343872)))]; + tensor input_241_cast_fp16 = conv(dilations = var_19863, groups = var_19819, pad = input_241_pad_0, pad_type = input_241_pad_type_0, strides = var_19861, weight = layers_12_self_attn_q_proj_loraA_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor("input_241_cast_fp16")]; + tensor var_19867 = const()[name = tensor("op_19867"), val = tensor([1, 1])]; + tensor var_19869 = const()[name = tensor("op_19869"), val = tensor([1, 1])]; + tensor lora_out_289_pad_type_0 = const()[name = tensor("lora_out_289_pad_type_0"), val = tensor("custom")]; + tensor lora_out_289_pad_0 = const()[name = tensor("lora_out_289_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_291_weight_0_to_fp16 = const()[name = tensor("lora_out_291_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142384896)))]; + tensor lora_out_291_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_19869, groups = var_19819, pad = lora_out_289_pad_0, pad_type = lora_out_289_pad_type_0, strides = var_19867, weight = lora_out_291_weight_0_to_fp16, x = input_241_cast_fp16)[name = tensor("lora_out_291_cast_fp16")]; + tensor query_25_cast_fp16 = add(x = pretrained_out_145_cast_fp16, y = lora_out_291_cast_fp16)[name = tensor("query_25_cast_fp16")]; + tensor var_19879 = const()[name = tensor("op_19879"), val = tensor([1, 1])]; + tensor var_19881 = const()[name = tensor("op_19881"), val = tensor([1, 1])]; + tensor pretrained_out_147_pad_type_0 = const()[name = tensor("pretrained_out_147_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_147_pad_0 = const()[name = tensor("pretrained_out_147_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142425920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143245184))), name = tensor("layers_12_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_147_cast_fp16 = conv(dilations = var_19881, groups = var_19819, pad = pretrained_out_147_pad_0, pad_type = pretrained_out_147_pad_type_0, strides = var_19879, weight = layers_12_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_49_cast_fp16)[name = tensor("pretrained_out_147_cast_fp16")]; + tensor var_19885 = const()[name = tensor("op_19885"), val = tensor([1, 1])]; + tensor var_19887 = const()[name = tensor("op_19887"), val = tensor([1, 1])]; + tensor input_243_pad_type_0 = const()[name = tensor("input_243_pad_type_0"), val = tensor("custom")]; + tensor input_243_pad_0 = const()[name = tensor("input_243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143245312)))]; + tensor input_243_cast_fp16 = conv(dilations = var_19887, groups = var_19819, pad = input_243_pad_0, pad_type = input_243_pad_type_0, strides = var_19885, weight = layers_12_self_attn_k_proj_loraA_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor var_19891 = const()[name = tensor("op_19891"), val = tensor([1, 1])]; + tensor var_19893 = const()[name = tensor("op_19893"), val = tensor([1, 1])]; + tensor lora_out_293_pad_type_0 = const()[name = tensor("lora_out_293_pad_type_0"), val = tensor("custom")]; + tensor lora_out_293_pad_0 = const()[name = tensor("lora_out_293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_295_weight_0_to_fp16 = const()[name = tensor("lora_out_295_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143286336)))]; + tensor lora_out_295_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_19893, groups = var_19819, pad = lora_out_293_pad_0, pad_type = lora_out_293_pad_type_0, strides = var_19891, weight = lora_out_295_weight_0_to_fp16, x = input_243_cast_fp16)[name = tensor("lora_out_295_cast_fp16")]; + tensor key_25_cast_fp16 = add(x = pretrained_out_147_cast_fp16, y = lora_out_295_cast_fp16)[name = tensor("key_25_cast_fp16")]; + tensor var_19904 = const()[name = tensor("op_19904"), val = tensor([1, 1])]; + tensor var_19906 = const()[name = tensor("op_19906"), val = tensor([1, 1])]; + tensor pretrained_out_149_pad_type_0 = const()[name = tensor("pretrained_out_149_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_149_pad_0 = const()[name = tensor("pretrained_out_149_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143327360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144146624))), name = tensor("layers_12_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_12_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144146752)))]; + tensor pretrained_out_149_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_19906, groups = var_19819, pad = pretrained_out_149_pad_0, pad_type = pretrained_out_149_pad_type_0, strides = var_19904, weight = layers_12_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_49_cast_fp16)[name = tensor("pretrained_out_149_cast_fp16")]; + tensor var_19910 = const()[name = tensor("op_19910"), val = tensor([1, 1])]; + tensor var_19912 = const()[name = tensor("op_19912"), val = tensor([1, 1])]; + tensor input_245_pad_type_0 = const()[name = tensor("input_245_pad_type_0"), val = tensor("custom")]; + tensor input_245_pad_0 = const()[name = tensor("input_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144149376)))]; + tensor input_245_cast_fp16 = conv(dilations = var_19912, groups = var_19819, pad = input_245_pad_0, pad_type = input_245_pad_type_0, strides = var_19910, weight = layers_12_self_attn_v_proj_loraA_weight_to_fp16, x = obj_49_cast_fp16)[name = tensor("input_245_cast_fp16")]; + tensor var_19916 = const()[name = tensor("op_19916"), val = tensor([1, 1])]; + tensor var_19918 = const()[name = tensor("op_19918"), val = tensor([1, 1])]; + tensor lora_out_297_pad_type_0 = const()[name = tensor("lora_out_297_pad_type_0"), val = tensor("custom")]; + tensor lora_out_297_pad_0 = const()[name = tensor("lora_out_297_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_299_weight_0_to_fp16 = const()[name = tensor("lora_out_299_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144190400)))]; + tensor lora_out_299_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_19918, groups = var_19819, pad = lora_out_297_pad_0, pad_type = lora_out_297_pad_type_0, strides = var_19916, weight = lora_out_299_weight_0_to_fp16, x = input_245_cast_fp16)[name = tensor("lora_out_299_cast_fp16")]; + tensor value_25_cast_fp16 = add(x = pretrained_out_149_cast_fp16, y = lora_out_299_cast_fp16)[name = tensor("value_25_cast_fp16")]; + tensor var_19928_begin_0 = const()[name = tensor("op_19928_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_19928_end_0 = const()[name = tensor("op_19928_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_19928_end_mask_0 = const()[name = tensor("op_19928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19928_cast_fp16 = slice_by_index(begin = var_19928_begin_0, end = var_19928_end_0, end_mask = var_19928_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19928_cast_fp16")]; + tensor var_19932_begin_0 = const()[name = tensor("op_19932_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_19932_end_0 = const()[name = tensor("op_19932_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_19932_end_mask_0 = const()[name = tensor("op_19932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19932_cast_fp16 = slice_by_index(begin = var_19932_begin_0, end = var_19932_end_0, end_mask = var_19932_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19932_cast_fp16")]; + tensor var_19936_begin_0 = const()[name = tensor("op_19936_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_19936_end_0 = const()[name = tensor("op_19936_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_19936_end_mask_0 = const()[name = tensor("op_19936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19936_cast_fp16 = slice_by_index(begin = var_19936_begin_0, end = var_19936_end_0, end_mask = var_19936_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19936_cast_fp16")]; + tensor var_19940_begin_0 = const()[name = tensor("op_19940_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_19940_end_0 = const()[name = tensor("op_19940_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_19940_end_mask_0 = const()[name = tensor("op_19940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19940_cast_fp16 = slice_by_index(begin = var_19940_begin_0, end = var_19940_end_0, end_mask = var_19940_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19940_cast_fp16")]; + tensor var_19944_begin_0 = const()[name = tensor("op_19944_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_19944_end_0 = const()[name = tensor("op_19944_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_19944_end_mask_0 = const()[name = tensor("op_19944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19944_cast_fp16 = slice_by_index(begin = var_19944_begin_0, end = var_19944_end_0, end_mask = var_19944_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19944_cast_fp16")]; + tensor var_19948_begin_0 = const()[name = tensor("op_19948_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_19948_end_0 = const()[name = tensor("op_19948_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_19948_end_mask_0 = const()[name = tensor("op_19948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19948_cast_fp16 = slice_by_index(begin = var_19948_begin_0, end = var_19948_end_0, end_mask = var_19948_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19948_cast_fp16")]; + tensor var_19952_begin_0 = const()[name = tensor("op_19952_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_19952_end_0 = const()[name = tensor("op_19952_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_19952_end_mask_0 = const()[name = tensor("op_19952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19952_cast_fp16 = slice_by_index(begin = var_19952_begin_0, end = var_19952_end_0, end_mask = var_19952_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19952_cast_fp16")]; + tensor var_19956_begin_0 = const()[name = tensor("op_19956_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_19956_end_0 = const()[name = tensor("op_19956_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_19956_end_mask_0 = const()[name = tensor("op_19956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19956_cast_fp16 = slice_by_index(begin = var_19956_begin_0, end = var_19956_end_0, end_mask = var_19956_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19956_cast_fp16")]; + tensor var_19960_begin_0 = const()[name = tensor("op_19960_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_19960_end_0 = const()[name = tensor("op_19960_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_19960_end_mask_0 = const()[name = tensor("op_19960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19960_cast_fp16 = slice_by_index(begin = var_19960_begin_0, end = var_19960_end_0, end_mask = var_19960_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19960_cast_fp16")]; + tensor var_19964_begin_0 = const()[name = tensor("op_19964_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_19964_end_0 = const()[name = tensor("op_19964_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_19964_end_mask_0 = const()[name = tensor("op_19964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19964_cast_fp16 = slice_by_index(begin = var_19964_begin_0, end = var_19964_end_0, end_mask = var_19964_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19964_cast_fp16")]; + tensor var_19968_begin_0 = const()[name = tensor("op_19968_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_19968_end_0 = const()[name = tensor("op_19968_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_19968_end_mask_0 = const()[name = tensor("op_19968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19968_cast_fp16 = slice_by_index(begin = var_19968_begin_0, end = var_19968_end_0, end_mask = var_19968_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19968_cast_fp16")]; + tensor var_19972_begin_0 = const()[name = tensor("op_19972_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_19972_end_0 = const()[name = tensor("op_19972_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_19972_end_mask_0 = const()[name = tensor("op_19972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19972_cast_fp16 = slice_by_index(begin = var_19972_begin_0, end = var_19972_end_0, end_mask = var_19972_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19972_cast_fp16")]; + tensor var_19976_begin_0 = const()[name = tensor("op_19976_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_19976_end_0 = const()[name = tensor("op_19976_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_19976_end_mask_0 = const()[name = tensor("op_19976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19976_cast_fp16 = slice_by_index(begin = var_19976_begin_0, end = var_19976_end_0, end_mask = var_19976_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19976_cast_fp16")]; + tensor var_19980_begin_0 = const()[name = tensor("op_19980_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_19980_end_0 = const()[name = tensor("op_19980_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_19980_end_mask_0 = const()[name = tensor("op_19980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19980_cast_fp16 = slice_by_index(begin = var_19980_begin_0, end = var_19980_end_0, end_mask = var_19980_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19980_cast_fp16")]; + tensor var_19984_begin_0 = const()[name = tensor("op_19984_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_19984_end_0 = const()[name = tensor("op_19984_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_19984_end_mask_0 = const()[name = tensor("op_19984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19984_cast_fp16 = slice_by_index(begin = var_19984_begin_0, end = var_19984_end_0, end_mask = var_19984_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19984_cast_fp16")]; + tensor var_19988_begin_0 = const()[name = tensor("op_19988_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_19988_end_0 = const()[name = tensor("op_19988_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_19988_end_mask_0 = const()[name = tensor("op_19988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19988_cast_fp16 = slice_by_index(begin = var_19988_begin_0, end = var_19988_end_0, end_mask = var_19988_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19988_cast_fp16")]; + tensor var_19992_begin_0 = const()[name = tensor("op_19992_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_19992_end_0 = const()[name = tensor("op_19992_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_19992_end_mask_0 = const()[name = tensor("op_19992_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19992_cast_fp16 = slice_by_index(begin = var_19992_begin_0, end = var_19992_end_0, end_mask = var_19992_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19992_cast_fp16")]; + tensor var_19996_begin_0 = const()[name = tensor("op_19996_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_19996_end_0 = const()[name = tensor("op_19996_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_19996_end_mask_0 = const()[name = tensor("op_19996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_19996_cast_fp16 = slice_by_index(begin = var_19996_begin_0, end = var_19996_end_0, end_mask = var_19996_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_19996_cast_fp16")]; + tensor var_20000_begin_0 = const()[name = tensor("op_20000_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20000_end_0 = const()[name = tensor("op_20000_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20000_end_mask_0 = const()[name = tensor("op_20000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20000_cast_fp16 = slice_by_index(begin = var_20000_begin_0, end = var_20000_end_0, end_mask = var_20000_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_20000_cast_fp16")]; + tensor var_20004_begin_0 = const()[name = tensor("op_20004_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20004_end_0 = const()[name = tensor("op_20004_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20004_end_mask_0 = const()[name = tensor("op_20004_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20004_cast_fp16 = slice_by_index(begin = var_20004_begin_0, end = var_20004_end_0, end_mask = var_20004_end_mask_0, x = query_25_cast_fp16)[name = tensor("op_20004_cast_fp16")]; + tensor var_20013_begin_0 = const()[name = tensor("op_20013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20013_end_0 = const()[name = tensor("op_20013_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20013_end_mask_0 = const()[name = tensor("op_20013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20013_cast_fp16 = slice_by_index(begin = var_20013_begin_0, end = var_20013_end_0, end_mask = var_20013_end_mask_0, x = var_19928_cast_fp16)[name = tensor("op_20013_cast_fp16")]; + tensor var_20020_begin_0 = const()[name = tensor("op_20020_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20020_end_0 = const()[name = tensor("op_20020_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20020_end_mask_0 = const()[name = tensor("op_20020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20020_cast_fp16 = slice_by_index(begin = var_20020_begin_0, end = var_20020_end_0, end_mask = var_20020_end_mask_0, x = var_19928_cast_fp16)[name = tensor("op_20020_cast_fp16")]; + tensor var_20027_begin_0 = const()[name = tensor("op_20027_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20027_end_0 = const()[name = tensor("op_20027_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20027_end_mask_0 = const()[name = tensor("op_20027_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20027_cast_fp16 = slice_by_index(begin = var_20027_begin_0, end = var_20027_end_0, end_mask = var_20027_end_mask_0, x = var_19928_cast_fp16)[name = tensor("op_20027_cast_fp16")]; + tensor var_20034_begin_0 = const()[name = tensor("op_20034_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20034_end_0 = const()[name = tensor("op_20034_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20034_end_mask_0 = const()[name = tensor("op_20034_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20034_cast_fp16 = slice_by_index(begin = var_20034_begin_0, end = var_20034_end_0, end_mask = var_20034_end_mask_0, x = var_19928_cast_fp16)[name = tensor("op_20034_cast_fp16")]; + tensor var_20041_begin_0 = const()[name = tensor("op_20041_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20041_end_0 = const()[name = tensor("op_20041_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20041_end_mask_0 = const()[name = tensor("op_20041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20041_cast_fp16 = slice_by_index(begin = var_20041_begin_0, end = var_20041_end_0, end_mask = var_20041_end_mask_0, x = var_19932_cast_fp16)[name = tensor("op_20041_cast_fp16")]; + tensor var_20048_begin_0 = const()[name = tensor("op_20048_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20048_end_0 = const()[name = tensor("op_20048_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20048_end_mask_0 = const()[name = tensor("op_20048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20048_cast_fp16 = slice_by_index(begin = var_20048_begin_0, end = var_20048_end_0, end_mask = var_20048_end_mask_0, x = var_19932_cast_fp16)[name = tensor("op_20048_cast_fp16")]; + tensor var_20055_begin_0 = const()[name = tensor("op_20055_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20055_end_0 = const()[name = tensor("op_20055_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20055_end_mask_0 = const()[name = tensor("op_20055_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20055_cast_fp16 = slice_by_index(begin = var_20055_begin_0, end = var_20055_end_0, end_mask = var_20055_end_mask_0, x = var_19932_cast_fp16)[name = tensor("op_20055_cast_fp16")]; + tensor var_20062_begin_0 = const()[name = tensor("op_20062_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20062_end_0 = const()[name = tensor("op_20062_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20062_end_mask_0 = const()[name = tensor("op_20062_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20062_cast_fp16 = slice_by_index(begin = var_20062_begin_0, end = var_20062_end_0, end_mask = var_20062_end_mask_0, x = var_19932_cast_fp16)[name = tensor("op_20062_cast_fp16")]; + tensor var_20069_begin_0 = const()[name = tensor("op_20069_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20069_end_0 = const()[name = tensor("op_20069_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20069_end_mask_0 = const()[name = tensor("op_20069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20069_cast_fp16 = slice_by_index(begin = var_20069_begin_0, end = var_20069_end_0, end_mask = var_20069_end_mask_0, x = var_19936_cast_fp16)[name = tensor("op_20069_cast_fp16")]; + tensor var_20076_begin_0 = const()[name = tensor("op_20076_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20076_end_0 = const()[name = tensor("op_20076_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20076_end_mask_0 = const()[name = tensor("op_20076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20076_cast_fp16 = slice_by_index(begin = var_20076_begin_0, end = var_20076_end_0, end_mask = var_20076_end_mask_0, x = var_19936_cast_fp16)[name = tensor("op_20076_cast_fp16")]; + tensor var_20083_begin_0 = const()[name = tensor("op_20083_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20083_end_0 = const()[name = tensor("op_20083_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20083_end_mask_0 = const()[name = tensor("op_20083_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20083_cast_fp16 = slice_by_index(begin = var_20083_begin_0, end = var_20083_end_0, end_mask = var_20083_end_mask_0, x = var_19936_cast_fp16)[name = tensor("op_20083_cast_fp16")]; + tensor var_20090_begin_0 = const()[name = tensor("op_20090_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20090_end_0 = const()[name = tensor("op_20090_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20090_end_mask_0 = const()[name = tensor("op_20090_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20090_cast_fp16 = slice_by_index(begin = var_20090_begin_0, end = var_20090_end_0, end_mask = var_20090_end_mask_0, x = var_19936_cast_fp16)[name = tensor("op_20090_cast_fp16")]; + tensor var_20097_begin_0 = const()[name = tensor("op_20097_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20097_end_0 = const()[name = tensor("op_20097_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20097_end_mask_0 = const()[name = tensor("op_20097_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20097_cast_fp16 = slice_by_index(begin = var_20097_begin_0, end = var_20097_end_0, end_mask = var_20097_end_mask_0, x = var_19940_cast_fp16)[name = tensor("op_20097_cast_fp16")]; + tensor var_20104_begin_0 = const()[name = tensor("op_20104_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20104_end_0 = const()[name = tensor("op_20104_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20104_end_mask_0 = const()[name = tensor("op_20104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20104_cast_fp16 = slice_by_index(begin = var_20104_begin_0, end = var_20104_end_0, end_mask = var_20104_end_mask_0, x = var_19940_cast_fp16)[name = tensor("op_20104_cast_fp16")]; + tensor var_20111_begin_0 = const()[name = tensor("op_20111_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20111_end_0 = const()[name = tensor("op_20111_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20111_end_mask_0 = const()[name = tensor("op_20111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20111_cast_fp16 = slice_by_index(begin = var_20111_begin_0, end = var_20111_end_0, end_mask = var_20111_end_mask_0, x = var_19940_cast_fp16)[name = tensor("op_20111_cast_fp16")]; + tensor var_20118_begin_0 = const()[name = tensor("op_20118_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20118_end_0 = const()[name = tensor("op_20118_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20118_end_mask_0 = const()[name = tensor("op_20118_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20118_cast_fp16 = slice_by_index(begin = var_20118_begin_0, end = var_20118_end_0, end_mask = var_20118_end_mask_0, x = var_19940_cast_fp16)[name = tensor("op_20118_cast_fp16")]; + tensor var_20125_begin_0 = const()[name = tensor("op_20125_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20125_end_0 = const()[name = tensor("op_20125_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20125_end_mask_0 = const()[name = tensor("op_20125_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20125_cast_fp16 = slice_by_index(begin = var_20125_begin_0, end = var_20125_end_0, end_mask = var_20125_end_mask_0, x = var_19944_cast_fp16)[name = tensor("op_20125_cast_fp16")]; + tensor var_20132_begin_0 = const()[name = tensor("op_20132_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20132_end_0 = const()[name = tensor("op_20132_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20132_end_mask_0 = const()[name = tensor("op_20132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20132_cast_fp16 = slice_by_index(begin = var_20132_begin_0, end = var_20132_end_0, end_mask = var_20132_end_mask_0, x = var_19944_cast_fp16)[name = tensor("op_20132_cast_fp16")]; + tensor var_20139_begin_0 = const()[name = tensor("op_20139_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20139_end_0 = const()[name = tensor("op_20139_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20139_end_mask_0 = const()[name = tensor("op_20139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20139_cast_fp16 = slice_by_index(begin = var_20139_begin_0, end = var_20139_end_0, end_mask = var_20139_end_mask_0, x = var_19944_cast_fp16)[name = tensor("op_20139_cast_fp16")]; + tensor var_20146_begin_0 = const()[name = tensor("op_20146_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20146_end_0 = const()[name = tensor("op_20146_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20146_end_mask_0 = const()[name = tensor("op_20146_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20146_cast_fp16 = slice_by_index(begin = var_20146_begin_0, end = var_20146_end_0, end_mask = var_20146_end_mask_0, x = var_19944_cast_fp16)[name = tensor("op_20146_cast_fp16")]; + tensor var_20153_begin_0 = const()[name = tensor("op_20153_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20153_end_0 = const()[name = tensor("op_20153_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20153_end_mask_0 = const()[name = tensor("op_20153_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20153_cast_fp16 = slice_by_index(begin = var_20153_begin_0, end = var_20153_end_0, end_mask = var_20153_end_mask_0, x = var_19948_cast_fp16)[name = tensor("op_20153_cast_fp16")]; + tensor var_20160_begin_0 = const()[name = tensor("op_20160_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20160_end_0 = const()[name = tensor("op_20160_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20160_end_mask_0 = const()[name = tensor("op_20160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20160_cast_fp16 = slice_by_index(begin = var_20160_begin_0, end = var_20160_end_0, end_mask = var_20160_end_mask_0, x = var_19948_cast_fp16)[name = tensor("op_20160_cast_fp16")]; + tensor var_20167_begin_0 = const()[name = tensor("op_20167_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20167_end_0 = const()[name = tensor("op_20167_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20167_end_mask_0 = const()[name = tensor("op_20167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20167_cast_fp16 = slice_by_index(begin = var_20167_begin_0, end = var_20167_end_0, end_mask = var_20167_end_mask_0, x = var_19948_cast_fp16)[name = tensor("op_20167_cast_fp16")]; + tensor var_20174_begin_0 = const()[name = tensor("op_20174_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20174_end_0 = const()[name = tensor("op_20174_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20174_end_mask_0 = const()[name = tensor("op_20174_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20174_cast_fp16 = slice_by_index(begin = var_20174_begin_0, end = var_20174_end_0, end_mask = var_20174_end_mask_0, x = var_19948_cast_fp16)[name = tensor("op_20174_cast_fp16")]; + tensor var_20181_begin_0 = const()[name = tensor("op_20181_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20181_end_0 = const()[name = tensor("op_20181_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20181_end_mask_0 = const()[name = tensor("op_20181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20181_cast_fp16 = slice_by_index(begin = var_20181_begin_0, end = var_20181_end_0, end_mask = var_20181_end_mask_0, x = var_19952_cast_fp16)[name = tensor("op_20181_cast_fp16")]; + tensor var_20188_begin_0 = const()[name = tensor("op_20188_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20188_end_0 = const()[name = tensor("op_20188_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20188_end_mask_0 = const()[name = tensor("op_20188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20188_cast_fp16 = slice_by_index(begin = var_20188_begin_0, end = var_20188_end_0, end_mask = var_20188_end_mask_0, x = var_19952_cast_fp16)[name = tensor("op_20188_cast_fp16")]; + tensor var_20195_begin_0 = const()[name = tensor("op_20195_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20195_end_0 = const()[name = tensor("op_20195_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20195_end_mask_0 = const()[name = tensor("op_20195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20195_cast_fp16 = slice_by_index(begin = var_20195_begin_0, end = var_20195_end_0, end_mask = var_20195_end_mask_0, x = var_19952_cast_fp16)[name = tensor("op_20195_cast_fp16")]; + tensor var_20202_begin_0 = const()[name = tensor("op_20202_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20202_end_0 = const()[name = tensor("op_20202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20202_end_mask_0 = const()[name = tensor("op_20202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20202_cast_fp16 = slice_by_index(begin = var_20202_begin_0, end = var_20202_end_0, end_mask = var_20202_end_mask_0, x = var_19952_cast_fp16)[name = tensor("op_20202_cast_fp16")]; + tensor var_20209_begin_0 = const()[name = tensor("op_20209_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20209_end_0 = const()[name = tensor("op_20209_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20209_end_mask_0 = const()[name = tensor("op_20209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20209_cast_fp16 = slice_by_index(begin = var_20209_begin_0, end = var_20209_end_0, end_mask = var_20209_end_mask_0, x = var_19956_cast_fp16)[name = tensor("op_20209_cast_fp16")]; + tensor var_20216_begin_0 = const()[name = tensor("op_20216_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20216_end_0 = const()[name = tensor("op_20216_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20216_end_mask_0 = const()[name = tensor("op_20216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20216_cast_fp16 = slice_by_index(begin = var_20216_begin_0, end = var_20216_end_0, end_mask = var_20216_end_mask_0, x = var_19956_cast_fp16)[name = tensor("op_20216_cast_fp16")]; + tensor var_20223_begin_0 = const()[name = tensor("op_20223_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20223_end_0 = const()[name = tensor("op_20223_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20223_end_mask_0 = const()[name = tensor("op_20223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20223_cast_fp16 = slice_by_index(begin = var_20223_begin_0, end = var_20223_end_0, end_mask = var_20223_end_mask_0, x = var_19956_cast_fp16)[name = tensor("op_20223_cast_fp16")]; + tensor var_20230_begin_0 = const()[name = tensor("op_20230_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20230_end_0 = const()[name = tensor("op_20230_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20230_end_mask_0 = const()[name = tensor("op_20230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20230_cast_fp16 = slice_by_index(begin = var_20230_begin_0, end = var_20230_end_0, end_mask = var_20230_end_mask_0, x = var_19956_cast_fp16)[name = tensor("op_20230_cast_fp16")]; + tensor var_20237_begin_0 = const()[name = tensor("op_20237_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20237_end_0 = const()[name = tensor("op_20237_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20237_end_mask_0 = const()[name = tensor("op_20237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20237_cast_fp16 = slice_by_index(begin = var_20237_begin_0, end = var_20237_end_0, end_mask = var_20237_end_mask_0, x = var_19960_cast_fp16)[name = tensor("op_20237_cast_fp16")]; + tensor var_20244_begin_0 = const()[name = tensor("op_20244_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20244_end_0 = const()[name = tensor("op_20244_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20244_end_mask_0 = const()[name = tensor("op_20244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20244_cast_fp16 = slice_by_index(begin = var_20244_begin_0, end = var_20244_end_0, end_mask = var_20244_end_mask_0, x = var_19960_cast_fp16)[name = tensor("op_20244_cast_fp16")]; + tensor var_20251_begin_0 = const()[name = tensor("op_20251_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20251_end_0 = const()[name = tensor("op_20251_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20251_end_mask_0 = const()[name = tensor("op_20251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20251_cast_fp16 = slice_by_index(begin = var_20251_begin_0, end = var_20251_end_0, end_mask = var_20251_end_mask_0, x = var_19960_cast_fp16)[name = tensor("op_20251_cast_fp16")]; + tensor var_20258_begin_0 = const()[name = tensor("op_20258_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20258_end_0 = const()[name = tensor("op_20258_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20258_end_mask_0 = const()[name = tensor("op_20258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20258_cast_fp16 = slice_by_index(begin = var_20258_begin_0, end = var_20258_end_0, end_mask = var_20258_end_mask_0, x = var_19960_cast_fp16)[name = tensor("op_20258_cast_fp16")]; + tensor var_20265_begin_0 = const()[name = tensor("op_20265_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20265_end_0 = const()[name = tensor("op_20265_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20265_end_mask_0 = const()[name = tensor("op_20265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20265_cast_fp16 = slice_by_index(begin = var_20265_begin_0, end = var_20265_end_0, end_mask = var_20265_end_mask_0, x = var_19964_cast_fp16)[name = tensor("op_20265_cast_fp16")]; + tensor var_20272_begin_0 = const()[name = tensor("op_20272_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20272_end_0 = const()[name = tensor("op_20272_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20272_end_mask_0 = const()[name = tensor("op_20272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20272_cast_fp16 = slice_by_index(begin = var_20272_begin_0, end = var_20272_end_0, end_mask = var_20272_end_mask_0, x = var_19964_cast_fp16)[name = tensor("op_20272_cast_fp16")]; + tensor var_20279_begin_0 = const()[name = tensor("op_20279_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20279_end_0 = const()[name = tensor("op_20279_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20279_end_mask_0 = const()[name = tensor("op_20279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20279_cast_fp16 = slice_by_index(begin = var_20279_begin_0, end = var_20279_end_0, end_mask = var_20279_end_mask_0, x = var_19964_cast_fp16)[name = tensor("op_20279_cast_fp16")]; + tensor var_20286_begin_0 = const()[name = tensor("op_20286_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20286_end_0 = const()[name = tensor("op_20286_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20286_end_mask_0 = const()[name = tensor("op_20286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20286_cast_fp16 = slice_by_index(begin = var_20286_begin_0, end = var_20286_end_0, end_mask = var_20286_end_mask_0, x = var_19964_cast_fp16)[name = tensor("op_20286_cast_fp16")]; + tensor var_20293_begin_0 = const()[name = tensor("op_20293_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20293_end_0 = const()[name = tensor("op_20293_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20293_end_mask_0 = const()[name = tensor("op_20293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20293_cast_fp16 = slice_by_index(begin = var_20293_begin_0, end = var_20293_end_0, end_mask = var_20293_end_mask_0, x = var_19968_cast_fp16)[name = tensor("op_20293_cast_fp16")]; + tensor var_20300_begin_0 = const()[name = tensor("op_20300_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20300_end_0 = const()[name = tensor("op_20300_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20300_end_mask_0 = const()[name = tensor("op_20300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20300_cast_fp16 = slice_by_index(begin = var_20300_begin_0, end = var_20300_end_0, end_mask = var_20300_end_mask_0, x = var_19968_cast_fp16)[name = tensor("op_20300_cast_fp16")]; + tensor var_20307_begin_0 = const()[name = tensor("op_20307_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20307_end_0 = const()[name = tensor("op_20307_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20307_end_mask_0 = const()[name = tensor("op_20307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20307_cast_fp16 = slice_by_index(begin = var_20307_begin_0, end = var_20307_end_0, end_mask = var_20307_end_mask_0, x = var_19968_cast_fp16)[name = tensor("op_20307_cast_fp16")]; + tensor var_20314_begin_0 = const()[name = tensor("op_20314_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20314_end_0 = const()[name = tensor("op_20314_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20314_end_mask_0 = const()[name = tensor("op_20314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20314_cast_fp16 = slice_by_index(begin = var_20314_begin_0, end = var_20314_end_0, end_mask = var_20314_end_mask_0, x = var_19968_cast_fp16)[name = tensor("op_20314_cast_fp16")]; + tensor var_20321_begin_0 = const()[name = tensor("op_20321_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20321_end_0 = const()[name = tensor("op_20321_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20321_end_mask_0 = const()[name = tensor("op_20321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20321_cast_fp16 = slice_by_index(begin = var_20321_begin_0, end = var_20321_end_0, end_mask = var_20321_end_mask_0, x = var_19972_cast_fp16)[name = tensor("op_20321_cast_fp16")]; + tensor var_20328_begin_0 = const()[name = tensor("op_20328_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20328_end_0 = const()[name = tensor("op_20328_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20328_end_mask_0 = const()[name = tensor("op_20328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20328_cast_fp16 = slice_by_index(begin = var_20328_begin_0, end = var_20328_end_0, end_mask = var_20328_end_mask_0, x = var_19972_cast_fp16)[name = tensor("op_20328_cast_fp16")]; + tensor var_20335_begin_0 = const()[name = tensor("op_20335_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20335_end_0 = const()[name = tensor("op_20335_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20335_end_mask_0 = const()[name = tensor("op_20335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20335_cast_fp16 = slice_by_index(begin = var_20335_begin_0, end = var_20335_end_0, end_mask = var_20335_end_mask_0, x = var_19972_cast_fp16)[name = tensor("op_20335_cast_fp16")]; + tensor var_20342_begin_0 = const()[name = tensor("op_20342_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20342_end_0 = const()[name = tensor("op_20342_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20342_end_mask_0 = const()[name = tensor("op_20342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20342_cast_fp16 = slice_by_index(begin = var_20342_begin_0, end = var_20342_end_0, end_mask = var_20342_end_mask_0, x = var_19972_cast_fp16)[name = tensor("op_20342_cast_fp16")]; + tensor var_20349_begin_0 = const()[name = tensor("op_20349_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20349_end_0 = const()[name = tensor("op_20349_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20349_end_mask_0 = const()[name = tensor("op_20349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20349_cast_fp16 = slice_by_index(begin = var_20349_begin_0, end = var_20349_end_0, end_mask = var_20349_end_mask_0, x = var_19976_cast_fp16)[name = tensor("op_20349_cast_fp16")]; + tensor var_20356_begin_0 = const()[name = tensor("op_20356_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20356_end_0 = const()[name = tensor("op_20356_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20356_end_mask_0 = const()[name = tensor("op_20356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20356_cast_fp16 = slice_by_index(begin = var_20356_begin_0, end = var_20356_end_0, end_mask = var_20356_end_mask_0, x = var_19976_cast_fp16)[name = tensor("op_20356_cast_fp16")]; + tensor var_20363_begin_0 = const()[name = tensor("op_20363_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20363_end_0 = const()[name = tensor("op_20363_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20363_end_mask_0 = const()[name = tensor("op_20363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20363_cast_fp16 = slice_by_index(begin = var_20363_begin_0, end = var_20363_end_0, end_mask = var_20363_end_mask_0, x = var_19976_cast_fp16)[name = tensor("op_20363_cast_fp16")]; + tensor var_20370_begin_0 = const()[name = tensor("op_20370_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20370_end_0 = const()[name = tensor("op_20370_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20370_end_mask_0 = const()[name = tensor("op_20370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20370_cast_fp16 = slice_by_index(begin = var_20370_begin_0, end = var_20370_end_0, end_mask = var_20370_end_mask_0, x = var_19976_cast_fp16)[name = tensor("op_20370_cast_fp16")]; + tensor var_20377_begin_0 = const()[name = tensor("op_20377_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20377_end_0 = const()[name = tensor("op_20377_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20377_end_mask_0 = const()[name = tensor("op_20377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20377_cast_fp16 = slice_by_index(begin = var_20377_begin_0, end = var_20377_end_0, end_mask = var_20377_end_mask_0, x = var_19980_cast_fp16)[name = tensor("op_20377_cast_fp16")]; + tensor var_20384_begin_0 = const()[name = tensor("op_20384_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20384_end_0 = const()[name = tensor("op_20384_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20384_end_mask_0 = const()[name = tensor("op_20384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20384_cast_fp16 = slice_by_index(begin = var_20384_begin_0, end = var_20384_end_0, end_mask = var_20384_end_mask_0, x = var_19980_cast_fp16)[name = tensor("op_20384_cast_fp16")]; + tensor var_20391_begin_0 = const()[name = tensor("op_20391_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20391_end_0 = const()[name = tensor("op_20391_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20391_end_mask_0 = const()[name = tensor("op_20391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20391_cast_fp16 = slice_by_index(begin = var_20391_begin_0, end = var_20391_end_0, end_mask = var_20391_end_mask_0, x = var_19980_cast_fp16)[name = tensor("op_20391_cast_fp16")]; + tensor var_20398_begin_0 = const()[name = tensor("op_20398_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20398_end_0 = const()[name = tensor("op_20398_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20398_end_mask_0 = const()[name = tensor("op_20398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20398_cast_fp16 = slice_by_index(begin = var_20398_begin_0, end = var_20398_end_0, end_mask = var_20398_end_mask_0, x = var_19980_cast_fp16)[name = tensor("op_20398_cast_fp16")]; + tensor var_20405_begin_0 = const()[name = tensor("op_20405_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20405_end_0 = const()[name = tensor("op_20405_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20405_end_mask_0 = const()[name = tensor("op_20405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20405_cast_fp16 = slice_by_index(begin = var_20405_begin_0, end = var_20405_end_0, end_mask = var_20405_end_mask_0, x = var_19984_cast_fp16)[name = tensor("op_20405_cast_fp16")]; + tensor var_20412_begin_0 = const()[name = tensor("op_20412_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20412_end_0 = const()[name = tensor("op_20412_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20412_end_mask_0 = const()[name = tensor("op_20412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20412_cast_fp16 = slice_by_index(begin = var_20412_begin_0, end = var_20412_end_0, end_mask = var_20412_end_mask_0, x = var_19984_cast_fp16)[name = tensor("op_20412_cast_fp16")]; + tensor var_20419_begin_0 = const()[name = tensor("op_20419_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20419_end_0 = const()[name = tensor("op_20419_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20419_end_mask_0 = const()[name = tensor("op_20419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20419_cast_fp16 = slice_by_index(begin = var_20419_begin_0, end = var_20419_end_0, end_mask = var_20419_end_mask_0, x = var_19984_cast_fp16)[name = tensor("op_20419_cast_fp16")]; + tensor var_20426_begin_0 = const()[name = tensor("op_20426_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20426_end_0 = const()[name = tensor("op_20426_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20426_end_mask_0 = const()[name = tensor("op_20426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20426_cast_fp16 = slice_by_index(begin = var_20426_begin_0, end = var_20426_end_0, end_mask = var_20426_end_mask_0, x = var_19984_cast_fp16)[name = tensor("op_20426_cast_fp16")]; + tensor var_20433_begin_0 = const()[name = tensor("op_20433_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20433_end_0 = const()[name = tensor("op_20433_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20433_end_mask_0 = const()[name = tensor("op_20433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20433_cast_fp16 = slice_by_index(begin = var_20433_begin_0, end = var_20433_end_0, end_mask = var_20433_end_mask_0, x = var_19988_cast_fp16)[name = tensor("op_20433_cast_fp16")]; + tensor var_20440_begin_0 = const()[name = tensor("op_20440_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20440_end_0 = const()[name = tensor("op_20440_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20440_end_mask_0 = const()[name = tensor("op_20440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20440_cast_fp16 = slice_by_index(begin = var_20440_begin_0, end = var_20440_end_0, end_mask = var_20440_end_mask_0, x = var_19988_cast_fp16)[name = tensor("op_20440_cast_fp16")]; + tensor var_20447_begin_0 = const()[name = tensor("op_20447_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20447_end_0 = const()[name = tensor("op_20447_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20447_end_mask_0 = const()[name = tensor("op_20447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20447_cast_fp16 = slice_by_index(begin = var_20447_begin_0, end = var_20447_end_0, end_mask = var_20447_end_mask_0, x = var_19988_cast_fp16)[name = tensor("op_20447_cast_fp16")]; + tensor var_20454_begin_0 = const()[name = tensor("op_20454_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20454_end_0 = const()[name = tensor("op_20454_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20454_end_mask_0 = const()[name = tensor("op_20454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20454_cast_fp16 = slice_by_index(begin = var_20454_begin_0, end = var_20454_end_0, end_mask = var_20454_end_mask_0, x = var_19988_cast_fp16)[name = tensor("op_20454_cast_fp16")]; + tensor var_20461_begin_0 = const()[name = tensor("op_20461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20461_end_0 = const()[name = tensor("op_20461_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20461_end_mask_0 = const()[name = tensor("op_20461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20461_cast_fp16 = slice_by_index(begin = var_20461_begin_0, end = var_20461_end_0, end_mask = var_20461_end_mask_0, x = var_19992_cast_fp16)[name = tensor("op_20461_cast_fp16")]; + tensor var_20468_begin_0 = const()[name = tensor("op_20468_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20468_end_0 = const()[name = tensor("op_20468_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20468_end_mask_0 = const()[name = tensor("op_20468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20468_cast_fp16 = slice_by_index(begin = var_20468_begin_0, end = var_20468_end_0, end_mask = var_20468_end_mask_0, x = var_19992_cast_fp16)[name = tensor("op_20468_cast_fp16")]; + tensor var_20475_begin_0 = const()[name = tensor("op_20475_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20475_end_0 = const()[name = tensor("op_20475_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20475_end_mask_0 = const()[name = tensor("op_20475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20475_cast_fp16 = slice_by_index(begin = var_20475_begin_0, end = var_20475_end_0, end_mask = var_20475_end_mask_0, x = var_19992_cast_fp16)[name = tensor("op_20475_cast_fp16")]; + tensor var_20482_begin_0 = const()[name = tensor("op_20482_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20482_end_0 = const()[name = tensor("op_20482_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20482_end_mask_0 = const()[name = tensor("op_20482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20482_cast_fp16 = slice_by_index(begin = var_20482_begin_0, end = var_20482_end_0, end_mask = var_20482_end_mask_0, x = var_19992_cast_fp16)[name = tensor("op_20482_cast_fp16")]; + tensor var_20489_begin_0 = const()[name = tensor("op_20489_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20489_end_0 = const()[name = tensor("op_20489_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20489_end_mask_0 = const()[name = tensor("op_20489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20489_cast_fp16 = slice_by_index(begin = var_20489_begin_0, end = var_20489_end_0, end_mask = var_20489_end_mask_0, x = var_19996_cast_fp16)[name = tensor("op_20489_cast_fp16")]; + tensor var_20496_begin_0 = const()[name = tensor("op_20496_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20496_end_0 = const()[name = tensor("op_20496_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20496_end_mask_0 = const()[name = tensor("op_20496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20496_cast_fp16 = slice_by_index(begin = var_20496_begin_0, end = var_20496_end_0, end_mask = var_20496_end_mask_0, x = var_19996_cast_fp16)[name = tensor("op_20496_cast_fp16")]; + tensor var_20503_begin_0 = const()[name = tensor("op_20503_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20503_end_0 = const()[name = tensor("op_20503_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20503_end_mask_0 = const()[name = tensor("op_20503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20503_cast_fp16 = slice_by_index(begin = var_20503_begin_0, end = var_20503_end_0, end_mask = var_20503_end_mask_0, x = var_19996_cast_fp16)[name = tensor("op_20503_cast_fp16")]; + tensor var_20510_begin_0 = const()[name = tensor("op_20510_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20510_end_0 = const()[name = tensor("op_20510_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20510_end_mask_0 = const()[name = tensor("op_20510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20510_cast_fp16 = slice_by_index(begin = var_20510_begin_0, end = var_20510_end_0, end_mask = var_20510_end_mask_0, x = var_19996_cast_fp16)[name = tensor("op_20510_cast_fp16")]; + tensor var_20517_begin_0 = const()[name = tensor("op_20517_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20517_end_0 = const()[name = tensor("op_20517_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20517_end_mask_0 = const()[name = tensor("op_20517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20517_cast_fp16 = slice_by_index(begin = var_20517_begin_0, end = var_20517_end_0, end_mask = var_20517_end_mask_0, x = var_20000_cast_fp16)[name = tensor("op_20517_cast_fp16")]; + tensor var_20524_begin_0 = const()[name = tensor("op_20524_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20524_end_0 = const()[name = tensor("op_20524_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20524_end_mask_0 = const()[name = tensor("op_20524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20524_cast_fp16 = slice_by_index(begin = var_20524_begin_0, end = var_20524_end_0, end_mask = var_20524_end_mask_0, x = var_20000_cast_fp16)[name = tensor("op_20524_cast_fp16")]; + tensor var_20531_begin_0 = const()[name = tensor("op_20531_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20531_end_0 = const()[name = tensor("op_20531_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20531_end_mask_0 = const()[name = tensor("op_20531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20531_cast_fp16 = slice_by_index(begin = var_20531_begin_0, end = var_20531_end_0, end_mask = var_20531_end_mask_0, x = var_20000_cast_fp16)[name = tensor("op_20531_cast_fp16")]; + tensor var_20538_begin_0 = const()[name = tensor("op_20538_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20538_end_0 = const()[name = tensor("op_20538_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20538_end_mask_0 = const()[name = tensor("op_20538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20538_cast_fp16 = slice_by_index(begin = var_20538_begin_0, end = var_20538_end_0, end_mask = var_20538_end_mask_0, x = var_20000_cast_fp16)[name = tensor("op_20538_cast_fp16")]; + tensor var_20545_begin_0 = const()[name = tensor("op_20545_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20545_end_0 = const()[name = tensor("op_20545_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_20545_end_mask_0 = const()[name = tensor("op_20545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20545_cast_fp16 = slice_by_index(begin = var_20545_begin_0, end = var_20545_end_0, end_mask = var_20545_end_mask_0, x = var_20004_cast_fp16)[name = tensor("op_20545_cast_fp16")]; + tensor var_20552_begin_0 = const()[name = tensor("op_20552_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_20552_end_0 = const()[name = tensor("op_20552_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_20552_end_mask_0 = const()[name = tensor("op_20552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20552_cast_fp16 = slice_by_index(begin = var_20552_begin_0, end = var_20552_end_0, end_mask = var_20552_end_mask_0, x = var_20004_cast_fp16)[name = tensor("op_20552_cast_fp16")]; + tensor var_20559_begin_0 = const()[name = tensor("op_20559_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_20559_end_0 = const()[name = tensor("op_20559_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_20559_end_mask_0 = const()[name = tensor("op_20559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20559_cast_fp16 = slice_by_index(begin = var_20559_begin_0, end = var_20559_end_0, end_mask = var_20559_end_mask_0, x = var_20004_cast_fp16)[name = tensor("op_20559_cast_fp16")]; + tensor var_20566_begin_0 = const()[name = tensor("op_20566_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_20566_end_0 = const()[name = tensor("op_20566_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20566_end_mask_0 = const()[name = tensor("op_20566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20566_cast_fp16 = slice_by_index(begin = var_20566_begin_0, end = var_20566_end_0, end_mask = var_20566_end_mask_0, x = var_20004_cast_fp16)[name = tensor("op_20566_cast_fp16")]; + tensor k_25_perm_0 = const()[name = tensor("k_25_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_20571_begin_0 = const()[name = tensor("op_20571_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20571_end_0 = const()[name = tensor("op_20571_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_20571_end_mask_0 = const()[name = tensor("op_20571_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_19 = transpose(perm = k_25_perm_0, x = key_25_cast_fp16)[name = tensor("transpose_19")]; + tensor var_20571_cast_fp16 = slice_by_index(begin = var_20571_begin_0, end = var_20571_end_0, end_mask = var_20571_end_mask_0, x = transpose_19)[name = tensor("op_20571_cast_fp16")]; + tensor var_20575_begin_0 = const()[name = tensor("op_20575_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_20575_end_0 = const()[name = tensor("op_20575_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_20575_end_mask_0 = const()[name = tensor("op_20575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20575_cast_fp16 = slice_by_index(begin = var_20575_begin_0, end = var_20575_end_0, end_mask = var_20575_end_mask_0, x = transpose_19)[name = tensor("op_20575_cast_fp16")]; + tensor var_20579_begin_0 = const()[name = tensor("op_20579_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_20579_end_0 = const()[name = tensor("op_20579_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_20579_end_mask_0 = const()[name = tensor("op_20579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20579_cast_fp16 = slice_by_index(begin = var_20579_begin_0, end = var_20579_end_0, end_mask = var_20579_end_mask_0, x = transpose_19)[name = tensor("op_20579_cast_fp16")]; + tensor var_20583_begin_0 = const()[name = tensor("op_20583_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_20583_end_0 = const()[name = tensor("op_20583_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_20583_end_mask_0 = const()[name = tensor("op_20583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20583_cast_fp16 = slice_by_index(begin = var_20583_begin_0, end = var_20583_end_0, end_mask = var_20583_end_mask_0, x = transpose_19)[name = tensor("op_20583_cast_fp16")]; + tensor var_20587_begin_0 = const()[name = tensor("op_20587_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_20587_end_0 = const()[name = tensor("op_20587_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_20587_end_mask_0 = const()[name = tensor("op_20587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20587_cast_fp16 = slice_by_index(begin = var_20587_begin_0, end = var_20587_end_0, end_mask = var_20587_end_mask_0, x = transpose_19)[name = tensor("op_20587_cast_fp16")]; + tensor var_20591_begin_0 = const()[name = tensor("op_20591_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_20591_end_0 = const()[name = tensor("op_20591_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_20591_end_mask_0 = const()[name = tensor("op_20591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20591_cast_fp16 = slice_by_index(begin = var_20591_begin_0, end = var_20591_end_0, end_mask = var_20591_end_mask_0, x = transpose_19)[name = tensor("op_20591_cast_fp16")]; + tensor var_20595_begin_0 = const()[name = tensor("op_20595_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_20595_end_0 = const()[name = tensor("op_20595_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_20595_end_mask_0 = const()[name = tensor("op_20595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20595_cast_fp16 = slice_by_index(begin = var_20595_begin_0, end = var_20595_end_0, end_mask = var_20595_end_mask_0, x = transpose_19)[name = tensor("op_20595_cast_fp16")]; + tensor var_20599_begin_0 = const()[name = tensor("op_20599_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_20599_end_0 = const()[name = tensor("op_20599_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_20599_end_mask_0 = const()[name = tensor("op_20599_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20599_cast_fp16 = slice_by_index(begin = var_20599_begin_0, end = var_20599_end_0, end_mask = var_20599_end_mask_0, x = transpose_19)[name = tensor("op_20599_cast_fp16")]; + tensor var_20603_begin_0 = const()[name = tensor("op_20603_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_20603_end_0 = const()[name = tensor("op_20603_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_20603_end_mask_0 = const()[name = tensor("op_20603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20603_cast_fp16 = slice_by_index(begin = var_20603_begin_0, end = var_20603_end_0, end_mask = var_20603_end_mask_0, x = transpose_19)[name = tensor("op_20603_cast_fp16")]; + tensor var_20607_begin_0 = const()[name = tensor("op_20607_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_20607_end_0 = const()[name = tensor("op_20607_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_20607_end_mask_0 = const()[name = tensor("op_20607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20607_cast_fp16 = slice_by_index(begin = var_20607_begin_0, end = var_20607_end_0, end_mask = var_20607_end_mask_0, x = transpose_19)[name = tensor("op_20607_cast_fp16")]; + tensor var_20611_begin_0 = const()[name = tensor("op_20611_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_20611_end_0 = const()[name = tensor("op_20611_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_20611_end_mask_0 = const()[name = tensor("op_20611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20611_cast_fp16 = slice_by_index(begin = var_20611_begin_0, end = var_20611_end_0, end_mask = var_20611_end_mask_0, x = transpose_19)[name = tensor("op_20611_cast_fp16")]; + tensor var_20615_begin_0 = const()[name = tensor("op_20615_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_20615_end_0 = const()[name = tensor("op_20615_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_20615_end_mask_0 = const()[name = tensor("op_20615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20615_cast_fp16 = slice_by_index(begin = var_20615_begin_0, end = var_20615_end_0, end_mask = var_20615_end_mask_0, x = transpose_19)[name = tensor("op_20615_cast_fp16")]; + tensor var_20619_begin_0 = const()[name = tensor("op_20619_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_20619_end_0 = const()[name = tensor("op_20619_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_20619_end_mask_0 = const()[name = tensor("op_20619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20619_cast_fp16 = slice_by_index(begin = var_20619_begin_0, end = var_20619_end_0, end_mask = var_20619_end_mask_0, x = transpose_19)[name = tensor("op_20619_cast_fp16")]; + tensor var_20623_begin_0 = const()[name = tensor("op_20623_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_20623_end_0 = const()[name = tensor("op_20623_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_20623_end_mask_0 = const()[name = tensor("op_20623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20623_cast_fp16 = slice_by_index(begin = var_20623_begin_0, end = var_20623_end_0, end_mask = var_20623_end_mask_0, x = transpose_19)[name = tensor("op_20623_cast_fp16")]; + tensor var_20627_begin_0 = const()[name = tensor("op_20627_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_20627_end_0 = const()[name = tensor("op_20627_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_20627_end_mask_0 = const()[name = tensor("op_20627_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20627_cast_fp16 = slice_by_index(begin = var_20627_begin_0, end = var_20627_end_0, end_mask = var_20627_end_mask_0, x = transpose_19)[name = tensor("op_20627_cast_fp16")]; + tensor var_20631_begin_0 = const()[name = tensor("op_20631_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_20631_end_0 = const()[name = tensor("op_20631_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_20631_end_mask_0 = const()[name = tensor("op_20631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20631_cast_fp16 = slice_by_index(begin = var_20631_begin_0, end = var_20631_end_0, end_mask = var_20631_end_mask_0, x = transpose_19)[name = tensor("op_20631_cast_fp16")]; + tensor var_20635_begin_0 = const()[name = tensor("op_20635_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_20635_end_0 = const()[name = tensor("op_20635_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_20635_end_mask_0 = const()[name = tensor("op_20635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20635_cast_fp16 = slice_by_index(begin = var_20635_begin_0, end = var_20635_end_0, end_mask = var_20635_end_mask_0, x = transpose_19)[name = tensor("op_20635_cast_fp16")]; + tensor var_20639_begin_0 = const()[name = tensor("op_20639_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_20639_end_0 = const()[name = tensor("op_20639_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_20639_end_mask_0 = const()[name = tensor("op_20639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20639_cast_fp16 = slice_by_index(begin = var_20639_begin_0, end = var_20639_end_0, end_mask = var_20639_end_mask_0, x = transpose_19)[name = tensor("op_20639_cast_fp16")]; + tensor var_20643_begin_0 = const()[name = tensor("op_20643_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_20643_end_0 = const()[name = tensor("op_20643_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_20643_end_mask_0 = const()[name = tensor("op_20643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20643_cast_fp16 = slice_by_index(begin = var_20643_begin_0, end = var_20643_end_0, end_mask = var_20643_end_mask_0, x = transpose_19)[name = tensor("op_20643_cast_fp16")]; + tensor var_20647_begin_0 = const()[name = tensor("op_20647_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_20647_end_0 = const()[name = tensor("op_20647_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_20647_end_mask_0 = const()[name = tensor("op_20647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_20647_cast_fp16 = slice_by_index(begin = var_20647_begin_0, end = var_20647_end_0, end_mask = var_20647_end_mask_0, x = transpose_19)[name = tensor("op_20647_cast_fp16")]; + tensor var_20649_begin_0 = const()[name = tensor("op_20649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_20649_end_0 = const()[name = tensor("op_20649_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_20649_end_mask_0 = const()[name = tensor("op_20649_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20649_cast_fp16 = slice_by_index(begin = var_20649_begin_0, end = var_20649_end_0, end_mask = var_20649_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20649_cast_fp16")]; + tensor var_20653_begin_0 = const()[name = tensor("op_20653_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_20653_end_0 = const()[name = tensor("op_20653_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_20653_end_mask_0 = const()[name = tensor("op_20653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20653_cast_fp16 = slice_by_index(begin = var_20653_begin_0, end = var_20653_end_0, end_mask = var_20653_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20653_cast_fp16")]; + tensor var_20657_begin_0 = const()[name = tensor("op_20657_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_20657_end_0 = const()[name = tensor("op_20657_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_20657_end_mask_0 = const()[name = tensor("op_20657_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20657_cast_fp16 = slice_by_index(begin = var_20657_begin_0, end = var_20657_end_0, end_mask = var_20657_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20657_cast_fp16")]; + tensor var_20661_begin_0 = const()[name = tensor("op_20661_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_20661_end_0 = const()[name = tensor("op_20661_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_20661_end_mask_0 = const()[name = tensor("op_20661_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20661_cast_fp16 = slice_by_index(begin = var_20661_begin_0, end = var_20661_end_0, end_mask = var_20661_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20661_cast_fp16")]; + tensor var_20665_begin_0 = const()[name = tensor("op_20665_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_20665_end_0 = const()[name = tensor("op_20665_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_20665_end_mask_0 = const()[name = tensor("op_20665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20665_cast_fp16 = slice_by_index(begin = var_20665_begin_0, end = var_20665_end_0, end_mask = var_20665_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20665_cast_fp16")]; + tensor var_20669_begin_0 = const()[name = tensor("op_20669_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_20669_end_0 = const()[name = tensor("op_20669_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_20669_end_mask_0 = const()[name = tensor("op_20669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20669_cast_fp16 = slice_by_index(begin = var_20669_begin_0, end = var_20669_end_0, end_mask = var_20669_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20669_cast_fp16")]; + tensor var_20673_begin_0 = const()[name = tensor("op_20673_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_20673_end_0 = const()[name = tensor("op_20673_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_20673_end_mask_0 = const()[name = tensor("op_20673_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20673_cast_fp16 = slice_by_index(begin = var_20673_begin_0, end = var_20673_end_0, end_mask = var_20673_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20673_cast_fp16")]; + tensor var_20677_begin_0 = const()[name = tensor("op_20677_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_20677_end_0 = const()[name = tensor("op_20677_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_20677_end_mask_0 = const()[name = tensor("op_20677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20677_cast_fp16 = slice_by_index(begin = var_20677_begin_0, end = var_20677_end_0, end_mask = var_20677_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20677_cast_fp16")]; + tensor var_20681_begin_0 = const()[name = tensor("op_20681_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_20681_end_0 = const()[name = tensor("op_20681_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_20681_end_mask_0 = const()[name = tensor("op_20681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20681_cast_fp16 = slice_by_index(begin = var_20681_begin_0, end = var_20681_end_0, end_mask = var_20681_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20681_cast_fp16")]; + tensor var_20685_begin_0 = const()[name = tensor("op_20685_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_20685_end_0 = const()[name = tensor("op_20685_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_20685_end_mask_0 = const()[name = tensor("op_20685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20685_cast_fp16 = slice_by_index(begin = var_20685_begin_0, end = var_20685_end_0, end_mask = var_20685_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20685_cast_fp16")]; + tensor var_20689_begin_0 = const()[name = tensor("op_20689_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_20689_end_0 = const()[name = tensor("op_20689_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_20689_end_mask_0 = const()[name = tensor("op_20689_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20689_cast_fp16 = slice_by_index(begin = var_20689_begin_0, end = var_20689_end_0, end_mask = var_20689_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20689_cast_fp16")]; + tensor var_20693_begin_0 = const()[name = tensor("op_20693_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_20693_end_0 = const()[name = tensor("op_20693_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_20693_end_mask_0 = const()[name = tensor("op_20693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20693_cast_fp16 = slice_by_index(begin = var_20693_begin_0, end = var_20693_end_0, end_mask = var_20693_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20693_cast_fp16")]; + tensor var_20697_begin_0 = const()[name = tensor("op_20697_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_20697_end_0 = const()[name = tensor("op_20697_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_20697_end_mask_0 = const()[name = tensor("op_20697_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20697_cast_fp16 = slice_by_index(begin = var_20697_begin_0, end = var_20697_end_0, end_mask = var_20697_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20697_cast_fp16")]; + tensor var_20701_begin_0 = const()[name = tensor("op_20701_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_20701_end_0 = const()[name = tensor("op_20701_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_20701_end_mask_0 = const()[name = tensor("op_20701_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20701_cast_fp16 = slice_by_index(begin = var_20701_begin_0, end = var_20701_end_0, end_mask = var_20701_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20701_cast_fp16")]; + tensor var_20705_begin_0 = const()[name = tensor("op_20705_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_20705_end_0 = const()[name = tensor("op_20705_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_20705_end_mask_0 = const()[name = tensor("op_20705_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20705_cast_fp16 = slice_by_index(begin = var_20705_begin_0, end = var_20705_end_0, end_mask = var_20705_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20705_cast_fp16")]; + tensor var_20709_begin_0 = const()[name = tensor("op_20709_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_20709_end_0 = const()[name = tensor("op_20709_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_20709_end_mask_0 = const()[name = tensor("op_20709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20709_cast_fp16 = slice_by_index(begin = var_20709_begin_0, end = var_20709_end_0, end_mask = var_20709_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20709_cast_fp16")]; + tensor var_20713_begin_0 = const()[name = tensor("op_20713_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_20713_end_0 = const()[name = tensor("op_20713_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_20713_end_mask_0 = const()[name = tensor("op_20713_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20713_cast_fp16 = slice_by_index(begin = var_20713_begin_0, end = var_20713_end_0, end_mask = var_20713_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20713_cast_fp16")]; + tensor var_20717_begin_0 = const()[name = tensor("op_20717_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_20717_end_0 = const()[name = tensor("op_20717_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_20717_end_mask_0 = const()[name = tensor("op_20717_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20717_cast_fp16 = slice_by_index(begin = var_20717_begin_0, end = var_20717_end_0, end_mask = var_20717_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20717_cast_fp16")]; + tensor var_20721_begin_0 = const()[name = tensor("op_20721_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_20721_end_0 = const()[name = tensor("op_20721_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_20721_end_mask_0 = const()[name = tensor("op_20721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20721_cast_fp16 = slice_by_index(begin = var_20721_begin_0, end = var_20721_end_0, end_mask = var_20721_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20721_cast_fp16")]; + tensor var_20725_begin_0 = const()[name = tensor("op_20725_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_20725_end_0 = const()[name = tensor("op_20725_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_20725_end_mask_0 = const()[name = tensor("op_20725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_20725_cast_fp16 = slice_by_index(begin = var_20725_begin_0, end = var_20725_end_0, end_mask = var_20725_end_mask_0, x = value_25_cast_fp16)[name = tensor("op_20725_cast_fp16")]; + tensor var_20729_equation_0 = const()[name = tensor("op_20729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20729_cast_fp16 = einsum(equation = var_20729_equation_0, values = (var_20571_cast_fp16, var_20013_cast_fp16))[name = tensor("op_20729_cast_fp16")]; + tensor var_20730_to_fp16 = const()[name = tensor("op_20730_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1921_cast_fp16 = mul(x = var_20729_cast_fp16, y = var_20730_to_fp16)[name = tensor("aw_chunk_1921_cast_fp16")]; + tensor var_20733_equation_0 = const()[name = tensor("op_20733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20733_cast_fp16 = einsum(equation = var_20733_equation_0, values = (var_20571_cast_fp16, var_20020_cast_fp16))[name = tensor("op_20733_cast_fp16")]; + tensor var_20734_to_fp16 = const()[name = tensor("op_20734_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1923_cast_fp16 = mul(x = var_20733_cast_fp16, y = var_20734_to_fp16)[name = tensor("aw_chunk_1923_cast_fp16")]; + tensor var_20737_equation_0 = const()[name = tensor("op_20737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20737_cast_fp16 = einsum(equation = var_20737_equation_0, values = (var_20571_cast_fp16, var_20027_cast_fp16))[name = tensor("op_20737_cast_fp16")]; + tensor var_20738_to_fp16 = const()[name = tensor("op_20738_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1925_cast_fp16 = mul(x = var_20737_cast_fp16, y = var_20738_to_fp16)[name = tensor("aw_chunk_1925_cast_fp16")]; + tensor var_20741_equation_0 = const()[name = tensor("op_20741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20741_cast_fp16 = einsum(equation = var_20741_equation_0, values = (var_20571_cast_fp16, var_20034_cast_fp16))[name = tensor("op_20741_cast_fp16")]; + tensor var_20742_to_fp16 = const()[name = tensor("op_20742_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1927_cast_fp16 = mul(x = var_20741_cast_fp16, y = var_20742_to_fp16)[name = tensor("aw_chunk_1927_cast_fp16")]; + tensor var_20745_equation_0 = const()[name = tensor("op_20745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20745_cast_fp16 = einsum(equation = var_20745_equation_0, values = (var_20575_cast_fp16, var_20041_cast_fp16))[name = tensor("op_20745_cast_fp16")]; + tensor var_20746_to_fp16 = const()[name = tensor("op_20746_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1929_cast_fp16 = mul(x = var_20745_cast_fp16, y = var_20746_to_fp16)[name = tensor("aw_chunk_1929_cast_fp16")]; + tensor var_20749_equation_0 = const()[name = tensor("op_20749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20749_cast_fp16 = einsum(equation = var_20749_equation_0, values = (var_20575_cast_fp16, var_20048_cast_fp16))[name = tensor("op_20749_cast_fp16")]; + tensor var_20750_to_fp16 = const()[name = tensor("op_20750_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1931_cast_fp16 = mul(x = var_20749_cast_fp16, y = var_20750_to_fp16)[name = tensor("aw_chunk_1931_cast_fp16")]; + tensor var_20753_equation_0 = const()[name = tensor("op_20753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20753_cast_fp16 = einsum(equation = var_20753_equation_0, values = (var_20575_cast_fp16, var_20055_cast_fp16))[name = tensor("op_20753_cast_fp16")]; + tensor var_20754_to_fp16 = const()[name = tensor("op_20754_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1933_cast_fp16 = mul(x = var_20753_cast_fp16, y = var_20754_to_fp16)[name = tensor("aw_chunk_1933_cast_fp16")]; + tensor var_20757_equation_0 = const()[name = tensor("op_20757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20757_cast_fp16 = einsum(equation = var_20757_equation_0, values = (var_20575_cast_fp16, var_20062_cast_fp16))[name = tensor("op_20757_cast_fp16")]; + tensor var_20758_to_fp16 = const()[name = tensor("op_20758_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1935_cast_fp16 = mul(x = var_20757_cast_fp16, y = var_20758_to_fp16)[name = tensor("aw_chunk_1935_cast_fp16")]; + tensor var_20761_equation_0 = const()[name = tensor("op_20761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20761_cast_fp16 = einsum(equation = var_20761_equation_0, values = (var_20579_cast_fp16, var_20069_cast_fp16))[name = tensor("op_20761_cast_fp16")]; + tensor var_20762_to_fp16 = const()[name = tensor("op_20762_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1937_cast_fp16 = mul(x = var_20761_cast_fp16, y = var_20762_to_fp16)[name = tensor("aw_chunk_1937_cast_fp16")]; + tensor var_20765_equation_0 = const()[name = tensor("op_20765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20765_cast_fp16 = einsum(equation = var_20765_equation_0, values = (var_20579_cast_fp16, var_20076_cast_fp16))[name = tensor("op_20765_cast_fp16")]; + tensor var_20766_to_fp16 = const()[name = tensor("op_20766_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1939_cast_fp16 = mul(x = var_20765_cast_fp16, y = var_20766_to_fp16)[name = tensor("aw_chunk_1939_cast_fp16")]; + tensor var_20769_equation_0 = const()[name = tensor("op_20769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20769_cast_fp16 = einsum(equation = var_20769_equation_0, values = (var_20579_cast_fp16, var_20083_cast_fp16))[name = tensor("op_20769_cast_fp16")]; + tensor var_20770_to_fp16 = const()[name = tensor("op_20770_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1941_cast_fp16 = mul(x = var_20769_cast_fp16, y = var_20770_to_fp16)[name = tensor("aw_chunk_1941_cast_fp16")]; + tensor var_20773_equation_0 = const()[name = tensor("op_20773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20773_cast_fp16 = einsum(equation = var_20773_equation_0, values = (var_20579_cast_fp16, var_20090_cast_fp16))[name = tensor("op_20773_cast_fp16")]; + tensor var_20774_to_fp16 = const()[name = tensor("op_20774_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1943_cast_fp16 = mul(x = var_20773_cast_fp16, y = var_20774_to_fp16)[name = tensor("aw_chunk_1943_cast_fp16")]; + tensor var_20777_equation_0 = const()[name = tensor("op_20777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20777_cast_fp16 = einsum(equation = var_20777_equation_0, values = (var_20583_cast_fp16, var_20097_cast_fp16))[name = tensor("op_20777_cast_fp16")]; + tensor var_20778_to_fp16 = const()[name = tensor("op_20778_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1945_cast_fp16 = mul(x = var_20777_cast_fp16, y = var_20778_to_fp16)[name = tensor("aw_chunk_1945_cast_fp16")]; + tensor var_20781_equation_0 = const()[name = tensor("op_20781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20781_cast_fp16 = einsum(equation = var_20781_equation_0, values = (var_20583_cast_fp16, var_20104_cast_fp16))[name = tensor("op_20781_cast_fp16")]; + tensor var_20782_to_fp16 = const()[name = tensor("op_20782_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1947_cast_fp16 = mul(x = var_20781_cast_fp16, y = var_20782_to_fp16)[name = tensor("aw_chunk_1947_cast_fp16")]; + tensor var_20785_equation_0 = const()[name = tensor("op_20785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20785_cast_fp16 = einsum(equation = var_20785_equation_0, values = (var_20583_cast_fp16, var_20111_cast_fp16))[name = tensor("op_20785_cast_fp16")]; + tensor var_20786_to_fp16 = const()[name = tensor("op_20786_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1949_cast_fp16 = mul(x = var_20785_cast_fp16, y = var_20786_to_fp16)[name = tensor("aw_chunk_1949_cast_fp16")]; + tensor var_20789_equation_0 = const()[name = tensor("op_20789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20789_cast_fp16 = einsum(equation = var_20789_equation_0, values = (var_20583_cast_fp16, var_20118_cast_fp16))[name = tensor("op_20789_cast_fp16")]; + tensor var_20790_to_fp16 = const()[name = tensor("op_20790_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1951_cast_fp16 = mul(x = var_20789_cast_fp16, y = var_20790_to_fp16)[name = tensor("aw_chunk_1951_cast_fp16")]; + tensor var_20793_equation_0 = const()[name = tensor("op_20793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20793_cast_fp16 = einsum(equation = var_20793_equation_0, values = (var_20587_cast_fp16, var_20125_cast_fp16))[name = tensor("op_20793_cast_fp16")]; + tensor var_20794_to_fp16 = const()[name = tensor("op_20794_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1953_cast_fp16 = mul(x = var_20793_cast_fp16, y = var_20794_to_fp16)[name = tensor("aw_chunk_1953_cast_fp16")]; + tensor var_20797_equation_0 = const()[name = tensor("op_20797_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20797_cast_fp16 = einsum(equation = var_20797_equation_0, values = (var_20587_cast_fp16, var_20132_cast_fp16))[name = tensor("op_20797_cast_fp16")]; + tensor var_20798_to_fp16 = const()[name = tensor("op_20798_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1955_cast_fp16 = mul(x = var_20797_cast_fp16, y = var_20798_to_fp16)[name = tensor("aw_chunk_1955_cast_fp16")]; + tensor var_20801_equation_0 = const()[name = tensor("op_20801_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20801_cast_fp16 = einsum(equation = var_20801_equation_0, values = (var_20587_cast_fp16, var_20139_cast_fp16))[name = tensor("op_20801_cast_fp16")]; + tensor var_20802_to_fp16 = const()[name = tensor("op_20802_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1957_cast_fp16 = mul(x = var_20801_cast_fp16, y = var_20802_to_fp16)[name = tensor("aw_chunk_1957_cast_fp16")]; + tensor var_20805_equation_0 = const()[name = tensor("op_20805_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20805_cast_fp16 = einsum(equation = var_20805_equation_0, values = (var_20587_cast_fp16, var_20146_cast_fp16))[name = tensor("op_20805_cast_fp16")]; + tensor var_20806_to_fp16 = const()[name = tensor("op_20806_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1959_cast_fp16 = mul(x = var_20805_cast_fp16, y = var_20806_to_fp16)[name = tensor("aw_chunk_1959_cast_fp16")]; + tensor var_20809_equation_0 = const()[name = tensor("op_20809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20809_cast_fp16 = einsum(equation = var_20809_equation_0, values = (var_20591_cast_fp16, var_20153_cast_fp16))[name = tensor("op_20809_cast_fp16")]; + tensor var_20810_to_fp16 = const()[name = tensor("op_20810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1961_cast_fp16 = mul(x = var_20809_cast_fp16, y = var_20810_to_fp16)[name = tensor("aw_chunk_1961_cast_fp16")]; + tensor var_20813_equation_0 = const()[name = tensor("op_20813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20813_cast_fp16 = einsum(equation = var_20813_equation_0, values = (var_20591_cast_fp16, var_20160_cast_fp16))[name = tensor("op_20813_cast_fp16")]; + tensor var_20814_to_fp16 = const()[name = tensor("op_20814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1963_cast_fp16 = mul(x = var_20813_cast_fp16, y = var_20814_to_fp16)[name = tensor("aw_chunk_1963_cast_fp16")]; + tensor var_20817_equation_0 = const()[name = tensor("op_20817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20817_cast_fp16 = einsum(equation = var_20817_equation_0, values = (var_20591_cast_fp16, var_20167_cast_fp16))[name = tensor("op_20817_cast_fp16")]; + tensor var_20818_to_fp16 = const()[name = tensor("op_20818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1965_cast_fp16 = mul(x = var_20817_cast_fp16, y = var_20818_to_fp16)[name = tensor("aw_chunk_1965_cast_fp16")]; + tensor var_20821_equation_0 = const()[name = tensor("op_20821_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20821_cast_fp16 = einsum(equation = var_20821_equation_0, values = (var_20591_cast_fp16, var_20174_cast_fp16))[name = tensor("op_20821_cast_fp16")]; + tensor var_20822_to_fp16 = const()[name = tensor("op_20822_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1967_cast_fp16 = mul(x = var_20821_cast_fp16, y = var_20822_to_fp16)[name = tensor("aw_chunk_1967_cast_fp16")]; + tensor var_20825_equation_0 = const()[name = tensor("op_20825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20825_cast_fp16 = einsum(equation = var_20825_equation_0, values = (var_20595_cast_fp16, var_20181_cast_fp16))[name = tensor("op_20825_cast_fp16")]; + tensor var_20826_to_fp16 = const()[name = tensor("op_20826_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1969_cast_fp16 = mul(x = var_20825_cast_fp16, y = var_20826_to_fp16)[name = tensor("aw_chunk_1969_cast_fp16")]; + tensor var_20829_equation_0 = const()[name = tensor("op_20829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20829_cast_fp16 = einsum(equation = var_20829_equation_0, values = (var_20595_cast_fp16, var_20188_cast_fp16))[name = tensor("op_20829_cast_fp16")]; + tensor var_20830_to_fp16 = const()[name = tensor("op_20830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1971_cast_fp16 = mul(x = var_20829_cast_fp16, y = var_20830_to_fp16)[name = tensor("aw_chunk_1971_cast_fp16")]; + tensor var_20833_equation_0 = const()[name = tensor("op_20833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20833_cast_fp16 = einsum(equation = var_20833_equation_0, values = (var_20595_cast_fp16, var_20195_cast_fp16))[name = tensor("op_20833_cast_fp16")]; + tensor var_20834_to_fp16 = const()[name = tensor("op_20834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1973_cast_fp16 = mul(x = var_20833_cast_fp16, y = var_20834_to_fp16)[name = tensor("aw_chunk_1973_cast_fp16")]; + tensor var_20837_equation_0 = const()[name = tensor("op_20837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20837_cast_fp16 = einsum(equation = var_20837_equation_0, values = (var_20595_cast_fp16, var_20202_cast_fp16))[name = tensor("op_20837_cast_fp16")]; + tensor var_20838_to_fp16 = const()[name = tensor("op_20838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1975_cast_fp16 = mul(x = var_20837_cast_fp16, y = var_20838_to_fp16)[name = tensor("aw_chunk_1975_cast_fp16")]; + tensor var_20841_equation_0 = const()[name = tensor("op_20841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20841_cast_fp16 = einsum(equation = var_20841_equation_0, values = (var_20599_cast_fp16, var_20209_cast_fp16))[name = tensor("op_20841_cast_fp16")]; + tensor var_20842_to_fp16 = const()[name = tensor("op_20842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1977_cast_fp16 = mul(x = var_20841_cast_fp16, y = var_20842_to_fp16)[name = tensor("aw_chunk_1977_cast_fp16")]; + tensor var_20845_equation_0 = const()[name = tensor("op_20845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20845_cast_fp16 = einsum(equation = var_20845_equation_0, values = (var_20599_cast_fp16, var_20216_cast_fp16))[name = tensor("op_20845_cast_fp16")]; + tensor var_20846_to_fp16 = const()[name = tensor("op_20846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1979_cast_fp16 = mul(x = var_20845_cast_fp16, y = var_20846_to_fp16)[name = tensor("aw_chunk_1979_cast_fp16")]; + tensor var_20849_equation_0 = const()[name = tensor("op_20849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20849_cast_fp16 = einsum(equation = var_20849_equation_0, values = (var_20599_cast_fp16, var_20223_cast_fp16))[name = tensor("op_20849_cast_fp16")]; + tensor var_20850_to_fp16 = const()[name = tensor("op_20850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1981_cast_fp16 = mul(x = var_20849_cast_fp16, y = var_20850_to_fp16)[name = tensor("aw_chunk_1981_cast_fp16")]; + tensor var_20853_equation_0 = const()[name = tensor("op_20853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20853_cast_fp16 = einsum(equation = var_20853_equation_0, values = (var_20599_cast_fp16, var_20230_cast_fp16))[name = tensor("op_20853_cast_fp16")]; + tensor var_20854_to_fp16 = const()[name = tensor("op_20854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1983_cast_fp16 = mul(x = var_20853_cast_fp16, y = var_20854_to_fp16)[name = tensor("aw_chunk_1983_cast_fp16")]; + tensor var_20857_equation_0 = const()[name = tensor("op_20857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20857_cast_fp16 = einsum(equation = var_20857_equation_0, values = (var_20603_cast_fp16, var_20237_cast_fp16))[name = tensor("op_20857_cast_fp16")]; + tensor var_20858_to_fp16 = const()[name = tensor("op_20858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1985_cast_fp16 = mul(x = var_20857_cast_fp16, y = var_20858_to_fp16)[name = tensor("aw_chunk_1985_cast_fp16")]; + tensor var_20861_equation_0 = const()[name = tensor("op_20861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20861_cast_fp16 = einsum(equation = var_20861_equation_0, values = (var_20603_cast_fp16, var_20244_cast_fp16))[name = tensor("op_20861_cast_fp16")]; + tensor var_20862_to_fp16 = const()[name = tensor("op_20862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1987_cast_fp16 = mul(x = var_20861_cast_fp16, y = var_20862_to_fp16)[name = tensor("aw_chunk_1987_cast_fp16")]; + tensor var_20865_equation_0 = const()[name = tensor("op_20865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20865_cast_fp16 = einsum(equation = var_20865_equation_0, values = (var_20603_cast_fp16, var_20251_cast_fp16))[name = tensor("op_20865_cast_fp16")]; + tensor var_20866_to_fp16 = const()[name = tensor("op_20866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1989_cast_fp16 = mul(x = var_20865_cast_fp16, y = var_20866_to_fp16)[name = tensor("aw_chunk_1989_cast_fp16")]; + tensor var_20869_equation_0 = const()[name = tensor("op_20869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20869_cast_fp16 = einsum(equation = var_20869_equation_0, values = (var_20603_cast_fp16, var_20258_cast_fp16))[name = tensor("op_20869_cast_fp16")]; + tensor var_20870_to_fp16 = const()[name = tensor("op_20870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1991_cast_fp16 = mul(x = var_20869_cast_fp16, y = var_20870_to_fp16)[name = tensor("aw_chunk_1991_cast_fp16")]; + tensor var_20873_equation_0 = const()[name = tensor("op_20873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20873_cast_fp16 = einsum(equation = var_20873_equation_0, values = (var_20607_cast_fp16, var_20265_cast_fp16))[name = tensor("op_20873_cast_fp16")]; + tensor var_20874_to_fp16 = const()[name = tensor("op_20874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1993_cast_fp16 = mul(x = var_20873_cast_fp16, y = var_20874_to_fp16)[name = tensor("aw_chunk_1993_cast_fp16")]; + tensor var_20877_equation_0 = const()[name = tensor("op_20877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20877_cast_fp16 = einsum(equation = var_20877_equation_0, values = (var_20607_cast_fp16, var_20272_cast_fp16))[name = tensor("op_20877_cast_fp16")]; + tensor var_20878_to_fp16 = const()[name = tensor("op_20878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1995_cast_fp16 = mul(x = var_20877_cast_fp16, y = var_20878_to_fp16)[name = tensor("aw_chunk_1995_cast_fp16")]; + tensor var_20881_equation_0 = const()[name = tensor("op_20881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20881_cast_fp16 = einsum(equation = var_20881_equation_0, values = (var_20607_cast_fp16, var_20279_cast_fp16))[name = tensor("op_20881_cast_fp16")]; + tensor var_20882_to_fp16 = const()[name = tensor("op_20882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1997_cast_fp16 = mul(x = var_20881_cast_fp16, y = var_20882_to_fp16)[name = tensor("aw_chunk_1997_cast_fp16")]; + tensor var_20885_equation_0 = const()[name = tensor("op_20885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20885_cast_fp16 = einsum(equation = var_20885_equation_0, values = (var_20607_cast_fp16, var_20286_cast_fp16))[name = tensor("op_20885_cast_fp16")]; + tensor var_20886_to_fp16 = const()[name = tensor("op_20886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_1999_cast_fp16 = mul(x = var_20885_cast_fp16, y = var_20886_to_fp16)[name = tensor("aw_chunk_1999_cast_fp16")]; + tensor var_20889_equation_0 = const()[name = tensor("op_20889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20889_cast_fp16 = einsum(equation = var_20889_equation_0, values = (var_20611_cast_fp16, var_20293_cast_fp16))[name = tensor("op_20889_cast_fp16")]; + tensor var_20890_to_fp16 = const()[name = tensor("op_20890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2001_cast_fp16 = mul(x = var_20889_cast_fp16, y = var_20890_to_fp16)[name = tensor("aw_chunk_2001_cast_fp16")]; + tensor var_20893_equation_0 = const()[name = tensor("op_20893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20893_cast_fp16 = einsum(equation = var_20893_equation_0, values = (var_20611_cast_fp16, var_20300_cast_fp16))[name = tensor("op_20893_cast_fp16")]; + tensor var_20894_to_fp16 = const()[name = tensor("op_20894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2003_cast_fp16 = mul(x = var_20893_cast_fp16, y = var_20894_to_fp16)[name = tensor("aw_chunk_2003_cast_fp16")]; + tensor var_20897_equation_0 = const()[name = tensor("op_20897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20897_cast_fp16 = einsum(equation = var_20897_equation_0, values = (var_20611_cast_fp16, var_20307_cast_fp16))[name = tensor("op_20897_cast_fp16")]; + tensor var_20898_to_fp16 = const()[name = tensor("op_20898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2005_cast_fp16 = mul(x = var_20897_cast_fp16, y = var_20898_to_fp16)[name = tensor("aw_chunk_2005_cast_fp16")]; + tensor var_20901_equation_0 = const()[name = tensor("op_20901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20901_cast_fp16 = einsum(equation = var_20901_equation_0, values = (var_20611_cast_fp16, var_20314_cast_fp16))[name = tensor("op_20901_cast_fp16")]; + tensor var_20902_to_fp16 = const()[name = tensor("op_20902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2007_cast_fp16 = mul(x = var_20901_cast_fp16, y = var_20902_to_fp16)[name = tensor("aw_chunk_2007_cast_fp16")]; + tensor var_20905_equation_0 = const()[name = tensor("op_20905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20905_cast_fp16 = einsum(equation = var_20905_equation_0, values = (var_20615_cast_fp16, var_20321_cast_fp16))[name = tensor("op_20905_cast_fp16")]; + tensor var_20906_to_fp16 = const()[name = tensor("op_20906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2009_cast_fp16 = mul(x = var_20905_cast_fp16, y = var_20906_to_fp16)[name = tensor("aw_chunk_2009_cast_fp16")]; + tensor var_20909_equation_0 = const()[name = tensor("op_20909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20909_cast_fp16 = einsum(equation = var_20909_equation_0, values = (var_20615_cast_fp16, var_20328_cast_fp16))[name = tensor("op_20909_cast_fp16")]; + tensor var_20910_to_fp16 = const()[name = tensor("op_20910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2011_cast_fp16 = mul(x = var_20909_cast_fp16, y = var_20910_to_fp16)[name = tensor("aw_chunk_2011_cast_fp16")]; + tensor var_20913_equation_0 = const()[name = tensor("op_20913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20913_cast_fp16 = einsum(equation = var_20913_equation_0, values = (var_20615_cast_fp16, var_20335_cast_fp16))[name = tensor("op_20913_cast_fp16")]; + tensor var_20914_to_fp16 = const()[name = tensor("op_20914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2013_cast_fp16 = mul(x = var_20913_cast_fp16, y = var_20914_to_fp16)[name = tensor("aw_chunk_2013_cast_fp16")]; + tensor var_20917_equation_0 = const()[name = tensor("op_20917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20917_cast_fp16 = einsum(equation = var_20917_equation_0, values = (var_20615_cast_fp16, var_20342_cast_fp16))[name = tensor("op_20917_cast_fp16")]; + tensor var_20918_to_fp16 = const()[name = tensor("op_20918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2015_cast_fp16 = mul(x = var_20917_cast_fp16, y = var_20918_to_fp16)[name = tensor("aw_chunk_2015_cast_fp16")]; + tensor var_20921_equation_0 = const()[name = tensor("op_20921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20921_cast_fp16 = einsum(equation = var_20921_equation_0, values = (var_20619_cast_fp16, var_20349_cast_fp16))[name = tensor("op_20921_cast_fp16")]; + tensor var_20922_to_fp16 = const()[name = tensor("op_20922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2017_cast_fp16 = mul(x = var_20921_cast_fp16, y = var_20922_to_fp16)[name = tensor("aw_chunk_2017_cast_fp16")]; + tensor var_20925_equation_0 = const()[name = tensor("op_20925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20925_cast_fp16 = einsum(equation = var_20925_equation_0, values = (var_20619_cast_fp16, var_20356_cast_fp16))[name = tensor("op_20925_cast_fp16")]; + tensor var_20926_to_fp16 = const()[name = tensor("op_20926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2019_cast_fp16 = mul(x = var_20925_cast_fp16, y = var_20926_to_fp16)[name = tensor("aw_chunk_2019_cast_fp16")]; + tensor var_20929_equation_0 = const()[name = tensor("op_20929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20929_cast_fp16 = einsum(equation = var_20929_equation_0, values = (var_20619_cast_fp16, var_20363_cast_fp16))[name = tensor("op_20929_cast_fp16")]; + tensor var_20930_to_fp16 = const()[name = tensor("op_20930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2021_cast_fp16 = mul(x = var_20929_cast_fp16, y = var_20930_to_fp16)[name = tensor("aw_chunk_2021_cast_fp16")]; + tensor var_20933_equation_0 = const()[name = tensor("op_20933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20933_cast_fp16 = einsum(equation = var_20933_equation_0, values = (var_20619_cast_fp16, var_20370_cast_fp16))[name = tensor("op_20933_cast_fp16")]; + tensor var_20934_to_fp16 = const()[name = tensor("op_20934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2023_cast_fp16 = mul(x = var_20933_cast_fp16, y = var_20934_to_fp16)[name = tensor("aw_chunk_2023_cast_fp16")]; + tensor var_20937_equation_0 = const()[name = tensor("op_20937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20937_cast_fp16 = einsum(equation = var_20937_equation_0, values = (var_20623_cast_fp16, var_20377_cast_fp16))[name = tensor("op_20937_cast_fp16")]; + tensor var_20938_to_fp16 = const()[name = tensor("op_20938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2025_cast_fp16 = mul(x = var_20937_cast_fp16, y = var_20938_to_fp16)[name = tensor("aw_chunk_2025_cast_fp16")]; + tensor var_20941_equation_0 = const()[name = tensor("op_20941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20941_cast_fp16 = einsum(equation = var_20941_equation_0, values = (var_20623_cast_fp16, var_20384_cast_fp16))[name = tensor("op_20941_cast_fp16")]; + tensor var_20942_to_fp16 = const()[name = tensor("op_20942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2027_cast_fp16 = mul(x = var_20941_cast_fp16, y = var_20942_to_fp16)[name = tensor("aw_chunk_2027_cast_fp16")]; + tensor var_20945_equation_0 = const()[name = tensor("op_20945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20945_cast_fp16 = einsum(equation = var_20945_equation_0, values = (var_20623_cast_fp16, var_20391_cast_fp16))[name = tensor("op_20945_cast_fp16")]; + tensor var_20946_to_fp16 = const()[name = tensor("op_20946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2029_cast_fp16 = mul(x = var_20945_cast_fp16, y = var_20946_to_fp16)[name = tensor("aw_chunk_2029_cast_fp16")]; + tensor var_20949_equation_0 = const()[name = tensor("op_20949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20949_cast_fp16 = einsum(equation = var_20949_equation_0, values = (var_20623_cast_fp16, var_20398_cast_fp16))[name = tensor("op_20949_cast_fp16")]; + tensor var_20950_to_fp16 = const()[name = tensor("op_20950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2031_cast_fp16 = mul(x = var_20949_cast_fp16, y = var_20950_to_fp16)[name = tensor("aw_chunk_2031_cast_fp16")]; + tensor var_20953_equation_0 = const()[name = tensor("op_20953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20953_cast_fp16 = einsum(equation = var_20953_equation_0, values = (var_20627_cast_fp16, var_20405_cast_fp16))[name = tensor("op_20953_cast_fp16")]; + tensor var_20954_to_fp16 = const()[name = tensor("op_20954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2033_cast_fp16 = mul(x = var_20953_cast_fp16, y = var_20954_to_fp16)[name = tensor("aw_chunk_2033_cast_fp16")]; + tensor var_20957_equation_0 = const()[name = tensor("op_20957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20957_cast_fp16 = einsum(equation = var_20957_equation_0, values = (var_20627_cast_fp16, var_20412_cast_fp16))[name = tensor("op_20957_cast_fp16")]; + tensor var_20958_to_fp16 = const()[name = tensor("op_20958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2035_cast_fp16 = mul(x = var_20957_cast_fp16, y = var_20958_to_fp16)[name = tensor("aw_chunk_2035_cast_fp16")]; + tensor var_20961_equation_0 = const()[name = tensor("op_20961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20961_cast_fp16 = einsum(equation = var_20961_equation_0, values = (var_20627_cast_fp16, var_20419_cast_fp16))[name = tensor("op_20961_cast_fp16")]; + tensor var_20962_to_fp16 = const()[name = tensor("op_20962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2037_cast_fp16 = mul(x = var_20961_cast_fp16, y = var_20962_to_fp16)[name = tensor("aw_chunk_2037_cast_fp16")]; + tensor var_20965_equation_0 = const()[name = tensor("op_20965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20965_cast_fp16 = einsum(equation = var_20965_equation_0, values = (var_20627_cast_fp16, var_20426_cast_fp16))[name = tensor("op_20965_cast_fp16")]; + tensor var_20966_to_fp16 = const()[name = tensor("op_20966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2039_cast_fp16 = mul(x = var_20965_cast_fp16, y = var_20966_to_fp16)[name = tensor("aw_chunk_2039_cast_fp16")]; + tensor var_20969_equation_0 = const()[name = tensor("op_20969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20969_cast_fp16 = einsum(equation = var_20969_equation_0, values = (var_20631_cast_fp16, var_20433_cast_fp16))[name = tensor("op_20969_cast_fp16")]; + tensor var_20970_to_fp16 = const()[name = tensor("op_20970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2041_cast_fp16 = mul(x = var_20969_cast_fp16, y = var_20970_to_fp16)[name = tensor("aw_chunk_2041_cast_fp16")]; + tensor var_20973_equation_0 = const()[name = tensor("op_20973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20973_cast_fp16 = einsum(equation = var_20973_equation_0, values = (var_20631_cast_fp16, var_20440_cast_fp16))[name = tensor("op_20973_cast_fp16")]; + tensor var_20974_to_fp16 = const()[name = tensor("op_20974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2043_cast_fp16 = mul(x = var_20973_cast_fp16, y = var_20974_to_fp16)[name = tensor("aw_chunk_2043_cast_fp16")]; + tensor var_20977_equation_0 = const()[name = tensor("op_20977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20977_cast_fp16 = einsum(equation = var_20977_equation_0, values = (var_20631_cast_fp16, var_20447_cast_fp16))[name = tensor("op_20977_cast_fp16")]; + tensor var_20978_to_fp16 = const()[name = tensor("op_20978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2045_cast_fp16 = mul(x = var_20977_cast_fp16, y = var_20978_to_fp16)[name = tensor("aw_chunk_2045_cast_fp16")]; + tensor var_20981_equation_0 = const()[name = tensor("op_20981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20981_cast_fp16 = einsum(equation = var_20981_equation_0, values = (var_20631_cast_fp16, var_20454_cast_fp16))[name = tensor("op_20981_cast_fp16")]; + tensor var_20982_to_fp16 = const()[name = tensor("op_20982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2047_cast_fp16 = mul(x = var_20981_cast_fp16, y = var_20982_to_fp16)[name = tensor("aw_chunk_2047_cast_fp16")]; + tensor var_20985_equation_0 = const()[name = tensor("op_20985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20985_cast_fp16 = einsum(equation = var_20985_equation_0, values = (var_20635_cast_fp16, var_20461_cast_fp16))[name = tensor("op_20985_cast_fp16")]; + tensor var_20986_to_fp16 = const()[name = tensor("op_20986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2049_cast_fp16 = mul(x = var_20985_cast_fp16, y = var_20986_to_fp16)[name = tensor("aw_chunk_2049_cast_fp16")]; + tensor var_20989_equation_0 = const()[name = tensor("op_20989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20989_cast_fp16 = einsum(equation = var_20989_equation_0, values = (var_20635_cast_fp16, var_20468_cast_fp16))[name = tensor("op_20989_cast_fp16")]; + tensor var_20990_to_fp16 = const()[name = tensor("op_20990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2051_cast_fp16 = mul(x = var_20989_cast_fp16, y = var_20990_to_fp16)[name = tensor("aw_chunk_2051_cast_fp16")]; + tensor var_20993_equation_0 = const()[name = tensor("op_20993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20993_cast_fp16 = einsum(equation = var_20993_equation_0, values = (var_20635_cast_fp16, var_20475_cast_fp16))[name = tensor("op_20993_cast_fp16")]; + tensor var_20994_to_fp16 = const()[name = tensor("op_20994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2053_cast_fp16 = mul(x = var_20993_cast_fp16, y = var_20994_to_fp16)[name = tensor("aw_chunk_2053_cast_fp16")]; + tensor var_20997_equation_0 = const()[name = tensor("op_20997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_20997_cast_fp16 = einsum(equation = var_20997_equation_0, values = (var_20635_cast_fp16, var_20482_cast_fp16))[name = tensor("op_20997_cast_fp16")]; + tensor var_20998_to_fp16 = const()[name = tensor("op_20998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2055_cast_fp16 = mul(x = var_20997_cast_fp16, y = var_20998_to_fp16)[name = tensor("aw_chunk_2055_cast_fp16")]; + tensor var_21001_equation_0 = const()[name = tensor("op_21001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21001_cast_fp16 = einsum(equation = var_21001_equation_0, values = (var_20639_cast_fp16, var_20489_cast_fp16))[name = tensor("op_21001_cast_fp16")]; + tensor var_21002_to_fp16 = const()[name = tensor("op_21002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2057_cast_fp16 = mul(x = var_21001_cast_fp16, y = var_21002_to_fp16)[name = tensor("aw_chunk_2057_cast_fp16")]; + tensor var_21005_equation_0 = const()[name = tensor("op_21005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21005_cast_fp16 = einsum(equation = var_21005_equation_0, values = (var_20639_cast_fp16, var_20496_cast_fp16))[name = tensor("op_21005_cast_fp16")]; + tensor var_21006_to_fp16 = const()[name = tensor("op_21006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2059_cast_fp16 = mul(x = var_21005_cast_fp16, y = var_21006_to_fp16)[name = tensor("aw_chunk_2059_cast_fp16")]; + tensor var_21009_equation_0 = const()[name = tensor("op_21009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21009_cast_fp16 = einsum(equation = var_21009_equation_0, values = (var_20639_cast_fp16, var_20503_cast_fp16))[name = tensor("op_21009_cast_fp16")]; + tensor var_21010_to_fp16 = const()[name = tensor("op_21010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2061_cast_fp16 = mul(x = var_21009_cast_fp16, y = var_21010_to_fp16)[name = tensor("aw_chunk_2061_cast_fp16")]; + tensor var_21013_equation_0 = const()[name = tensor("op_21013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21013_cast_fp16 = einsum(equation = var_21013_equation_0, values = (var_20639_cast_fp16, var_20510_cast_fp16))[name = tensor("op_21013_cast_fp16")]; + tensor var_21014_to_fp16 = const()[name = tensor("op_21014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2063_cast_fp16 = mul(x = var_21013_cast_fp16, y = var_21014_to_fp16)[name = tensor("aw_chunk_2063_cast_fp16")]; + tensor var_21017_equation_0 = const()[name = tensor("op_21017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21017_cast_fp16 = einsum(equation = var_21017_equation_0, values = (var_20643_cast_fp16, var_20517_cast_fp16))[name = tensor("op_21017_cast_fp16")]; + tensor var_21018_to_fp16 = const()[name = tensor("op_21018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2065_cast_fp16 = mul(x = var_21017_cast_fp16, y = var_21018_to_fp16)[name = tensor("aw_chunk_2065_cast_fp16")]; + tensor var_21021_equation_0 = const()[name = tensor("op_21021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21021_cast_fp16 = einsum(equation = var_21021_equation_0, values = (var_20643_cast_fp16, var_20524_cast_fp16))[name = tensor("op_21021_cast_fp16")]; + tensor var_21022_to_fp16 = const()[name = tensor("op_21022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2067_cast_fp16 = mul(x = var_21021_cast_fp16, y = var_21022_to_fp16)[name = tensor("aw_chunk_2067_cast_fp16")]; + tensor var_21025_equation_0 = const()[name = tensor("op_21025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21025_cast_fp16 = einsum(equation = var_21025_equation_0, values = (var_20643_cast_fp16, var_20531_cast_fp16))[name = tensor("op_21025_cast_fp16")]; + tensor var_21026_to_fp16 = const()[name = tensor("op_21026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2069_cast_fp16 = mul(x = var_21025_cast_fp16, y = var_21026_to_fp16)[name = tensor("aw_chunk_2069_cast_fp16")]; + tensor var_21029_equation_0 = const()[name = tensor("op_21029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21029_cast_fp16 = einsum(equation = var_21029_equation_0, values = (var_20643_cast_fp16, var_20538_cast_fp16))[name = tensor("op_21029_cast_fp16")]; + tensor var_21030_to_fp16 = const()[name = tensor("op_21030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2071_cast_fp16 = mul(x = var_21029_cast_fp16, y = var_21030_to_fp16)[name = tensor("aw_chunk_2071_cast_fp16")]; + tensor var_21033_equation_0 = const()[name = tensor("op_21033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21033_cast_fp16 = einsum(equation = var_21033_equation_0, values = (var_20647_cast_fp16, var_20545_cast_fp16))[name = tensor("op_21033_cast_fp16")]; + tensor var_21034_to_fp16 = const()[name = tensor("op_21034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2073_cast_fp16 = mul(x = var_21033_cast_fp16, y = var_21034_to_fp16)[name = tensor("aw_chunk_2073_cast_fp16")]; + tensor var_21037_equation_0 = const()[name = tensor("op_21037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21037_cast_fp16 = einsum(equation = var_21037_equation_0, values = (var_20647_cast_fp16, var_20552_cast_fp16))[name = tensor("op_21037_cast_fp16")]; + tensor var_21038_to_fp16 = const()[name = tensor("op_21038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2075_cast_fp16 = mul(x = var_21037_cast_fp16, y = var_21038_to_fp16)[name = tensor("aw_chunk_2075_cast_fp16")]; + tensor var_21041_equation_0 = const()[name = tensor("op_21041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21041_cast_fp16 = einsum(equation = var_21041_equation_0, values = (var_20647_cast_fp16, var_20559_cast_fp16))[name = tensor("op_21041_cast_fp16")]; + tensor var_21042_to_fp16 = const()[name = tensor("op_21042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2077_cast_fp16 = mul(x = var_21041_cast_fp16, y = var_21042_to_fp16)[name = tensor("aw_chunk_2077_cast_fp16")]; + tensor var_21045_equation_0 = const()[name = tensor("op_21045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_21045_cast_fp16 = einsum(equation = var_21045_equation_0, values = (var_20647_cast_fp16, var_20566_cast_fp16))[name = tensor("op_21045_cast_fp16")]; + tensor var_21046_to_fp16 = const()[name = tensor("op_21046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2079_cast_fp16 = mul(x = var_21045_cast_fp16, y = var_21046_to_fp16)[name = tensor("aw_chunk_2079_cast_fp16")]; + tensor var_21048_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1921_cast_fp16)[name = tensor("op_21048_cast_fp16")]; + tensor var_21049_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1923_cast_fp16)[name = tensor("op_21049_cast_fp16")]; + tensor var_21050_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1925_cast_fp16)[name = tensor("op_21050_cast_fp16")]; + tensor var_21051_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1927_cast_fp16)[name = tensor("op_21051_cast_fp16")]; + tensor var_21052_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1929_cast_fp16)[name = tensor("op_21052_cast_fp16")]; + tensor var_21053_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1931_cast_fp16)[name = tensor("op_21053_cast_fp16")]; + tensor var_21054_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1933_cast_fp16)[name = tensor("op_21054_cast_fp16")]; + tensor var_21055_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1935_cast_fp16)[name = tensor("op_21055_cast_fp16")]; + tensor var_21056_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1937_cast_fp16)[name = tensor("op_21056_cast_fp16")]; + tensor var_21057_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1939_cast_fp16)[name = tensor("op_21057_cast_fp16")]; + tensor var_21058_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1941_cast_fp16)[name = tensor("op_21058_cast_fp16")]; + tensor var_21059_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1943_cast_fp16)[name = tensor("op_21059_cast_fp16")]; + tensor var_21060_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1945_cast_fp16)[name = tensor("op_21060_cast_fp16")]; + tensor var_21061_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1947_cast_fp16)[name = tensor("op_21061_cast_fp16")]; + tensor var_21062_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1949_cast_fp16)[name = tensor("op_21062_cast_fp16")]; + tensor var_21063_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1951_cast_fp16)[name = tensor("op_21063_cast_fp16")]; + tensor var_21064_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1953_cast_fp16)[name = tensor("op_21064_cast_fp16")]; + tensor var_21065_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1955_cast_fp16)[name = tensor("op_21065_cast_fp16")]; + tensor var_21066_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1957_cast_fp16)[name = tensor("op_21066_cast_fp16")]; + tensor var_21067_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1959_cast_fp16)[name = tensor("op_21067_cast_fp16")]; + tensor var_21068_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1961_cast_fp16)[name = tensor("op_21068_cast_fp16")]; + tensor var_21069_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1963_cast_fp16)[name = tensor("op_21069_cast_fp16")]; + tensor var_21070_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1965_cast_fp16)[name = tensor("op_21070_cast_fp16")]; + tensor var_21071_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1967_cast_fp16)[name = tensor("op_21071_cast_fp16")]; + tensor var_21072_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1969_cast_fp16)[name = tensor("op_21072_cast_fp16")]; + tensor var_21073_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1971_cast_fp16)[name = tensor("op_21073_cast_fp16")]; + tensor var_21074_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1973_cast_fp16)[name = tensor("op_21074_cast_fp16")]; + tensor var_21075_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1975_cast_fp16)[name = tensor("op_21075_cast_fp16")]; + tensor var_21076_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1977_cast_fp16)[name = tensor("op_21076_cast_fp16")]; + tensor var_21077_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1979_cast_fp16)[name = tensor("op_21077_cast_fp16")]; + tensor var_21078_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1981_cast_fp16)[name = tensor("op_21078_cast_fp16")]; + tensor var_21079_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1983_cast_fp16)[name = tensor("op_21079_cast_fp16")]; + tensor var_21080_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1985_cast_fp16)[name = tensor("op_21080_cast_fp16")]; + tensor var_21081_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1987_cast_fp16)[name = tensor("op_21081_cast_fp16")]; + tensor var_21082_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1989_cast_fp16)[name = tensor("op_21082_cast_fp16")]; + tensor var_21083_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1991_cast_fp16)[name = tensor("op_21083_cast_fp16")]; + tensor var_21084_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1993_cast_fp16)[name = tensor("op_21084_cast_fp16")]; + tensor var_21085_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1995_cast_fp16)[name = tensor("op_21085_cast_fp16")]; + tensor var_21086_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1997_cast_fp16)[name = tensor("op_21086_cast_fp16")]; + tensor var_21087_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_1999_cast_fp16)[name = tensor("op_21087_cast_fp16")]; + tensor var_21088_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2001_cast_fp16)[name = tensor("op_21088_cast_fp16")]; + tensor var_21089_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2003_cast_fp16)[name = tensor("op_21089_cast_fp16")]; + tensor var_21090_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2005_cast_fp16)[name = tensor("op_21090_cast_fp16")]; + tensor var_21091_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2007_cast_fp16)[name = tensor("op_21091_cast_fp16")]; + tensor var_21092_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2009_cast_fp16)[name = tensor("op_21092_cast_fp16")]; + tensor var_21093_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2011_cast_fp16)[name = tensor("op_21093_cast_fp16")]; + tensor var_21094_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2013_cast_fp16)[name = tensor("op_21094_cast_fp16")]; + tensor var_21095_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2015_cast_fp16)[name = tensor("op_21095_cast_fp16")]; + tensor var_21096_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2017_cast_fp16)[name = tensor("op_21096_cast_fp16")]; + tensor var_21097_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2019_cast_fp16)[name = tensor("op_21097_cast_fp16")]; + tensor var_21098_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2021_cast_fp16)[name = tensor("op_21098_cast_fp16")]; + tensor var_21099_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2023_cast_fp16)[name = tensor("op_21099_cast_fp16")]; + tensor var_21100_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2025_cast_fp16)[name = tensor("op_21100_cast_fp16")]; + tensor var_21101_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2027_cast_fp16)[name = tensor("op_21101_cast_fp16")]; + tensor var_21102_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2029_cast_fp16)[name = tensor("op_21102_cast_fp16")]; + tensor var_21103_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2031_cast_fp16)[name = tensor("op_21103_cast_fp16")]; + tensor var_21104_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2033_cast_fp16)[name = tensor("op_21104_cast_fp16")]; + tensor var_21105_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2035_cast_fp16)[name = tensor("op_21105_cast_fp16")]; + tensor var_21106_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2037_cast_fp16)[name = tensor("op_21106_cast_fp16")]; + tensor var_21107_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2039_cast_fp16)[name = tensor("op_21107_cast_fp16")]; + tensor var_21108_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2041_cast_fp16)[name = tensor("op_21108_cast_fp16")]; + tensor var_21109_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2043_cast_fp16)[name = tensor("op_21109_cast_fp16")]; + tensor var_21110_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2045_cast_fp16)[name = tensor("op_21110_cast_fp16")]; + tensor var_21111_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2047_cast_fp16)[name = tensor("op_21111_cast_fp16")]; + tensor var_21112_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2049_cast_fp16)[name = tensor("op_21112_cast_fp16")]; + tensor var_21113_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2051_cast_fp16)[name = tensor("op_21113_cast_fp16")]; + tensor var_21114_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2053_cast_fp16)[name = tensor("op_21114_cast_fp16")]; + tensor var_21115_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2055_cast_fp16)[name = tensor("op_21115_cast_fp16")]; + tensor var_21116_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2057_cast_fp16)[name = tensor("op_21116_cast_fp16")]; + tensor var_21117_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2059_cast_fp16)[name = tensor("op_21117_cast_fp16")]; + tensor var_21118_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2061_cast_fp16)[name = tensor("op_21118_cast_fp16")]; + tensor var_21119_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2063_cast_fp16)[name = tensor("op_21119_cast_fp16")]; + tensor var_21120_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2065_cast_fp16)[name = tensor("op_21120_cast_fp16")]; + tensor var_21121_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2067_cast_fp16)[name = tensor("op_21121_cast_fp16")]; + tensor var_21122_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2069_cast_fp16)[name = tensor("op_21122_cast_fp16")]; + tensor var_21123_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2071_cast_fp16)[name = tensor("op_21123_cast_fp16")]; + tensor var_21124_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2073_cast_fp16)[name = tensor("op_21124_cast_fp16")]; + tensor var_21125_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2075_cast_fp16)[name = tensor("op_21125_cast_fp16")]; + tensor var_21126_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2077_cast_fp16)[name = tensor("op_21126_cast_fp16")]; + tensor var_21127_cast_fp16 = softmax(axis = var_19819, x = aw_chunk_2079_cast_fp16)[name = tensor("op_21127_cast_fp16")]; + tensor var_21129_equation_0 = const()[name = tensor("op_21129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21129_cast_fp16 = einsum(equation = var_21129_equation_0, values = (var_20649_cast_fp16, var_21048_cast_fp16))[name = tensor("op_21129_cast_fp16")]; + tensor var_21131_equation_0 = const()[name = tensor("op_21131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21131_cast_fp16 = einsum(equation = var_21131_equation_0, values = (var_20649_cast_fp16, var_21049_cast_fp16))[name = tensor("op_21131_cast_fp16")]; + tensor var_21133_equation_0 = const()[name = tensor("op_21133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21133_cast_fp16 = einsum(equation = var_21133_equation_0, values = (var_20649_cast_fp16, var_21050_cast_fp16))[name = tensor("op_21133_cast_fp16")]; + tensor var_21135_equation_0 = const()[name = tensor("op_21135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21135_cast_fp16 = einsum(equation = var_21135_equation_0, values = (var_20649_cast_fp16, var_21051_cast_fp16))[name = tensor("op_21135_cast_fp16")]; + tensor var_21137_equation_0 = const()[name = tensor("op_21137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21137_cast_fp16 = einsum(equation = var_21137_equation_0, values = (var_20653_cast_fp16, var_21052_cast_fp16))[name = tensor("op_21137_cast_fp16")]; + tensor var_21139_equation_0 = const()[name = tensor("op_21139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21139_cast_fp16 = einsum(equation = var_21139_equation_0, values = (var_20653_cast_fp16, var_21053_cast_fp16))[name = tensor("op_21139_cast_fp16")]; + tensor var_21141_equation_0 = const()[name = tensor("op_21141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21141_cast_fp16 = einsum(equation = var_21141_equation_0, values = (var_20653_cast_fp16, var_21054_cast_fp16))[name = tensor("op_21141_cast_fp16")]; + tensor var_21143_equation_0 = const()[name = tensor("op_21143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21143_cast_fp16 = einsum(equation = var_21143_equation_0, values = (var_20653_cast_fp16, var_21055_cast_fp16))[name = tensor("op_21143_cast_fp16")]; + tensor var_21145_equation_0 = const()[name = tensor("op_21145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21145_cast_fp16 = einsum(equation = var_21145_equation_0, values = (var_20657_cast_fp16, var_21056_cast_fp16))[name = tensor("op_21145_cast_fp16")]; + tensor var_21147_equation_0 = const()[name = tensor("op_21147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21147_cast_fp16 = einsum(equation = var_21147_equation_0, values = (var_20657_cast_fp16, var_21057_cast_fp16))[name = tensor("op_21147_cast_fp16")]; + tensor var_21149_equation_0 = const()[name = tensor("op_21149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21149_cast_fp16 = einsum(equation = var_21149_equation_0, values = (var_20657_cast_fp16, var_21058_cast_fp16))[name = tensor("op_21149_cast_fp16")]; + tensor var_21151_equation_0 = const()[name = tensor("op_21151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21151_cast_fp16 = einsum(equation = var_21151_equation_0, values = (var_20657_cast_fp16, var_21059_cast_fp16))[name = tensor("op_21151_cast_fp16")]; + tensor var_21153_equation_0 = const()[name = tensor("op_21153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21153_cast_fp16 = einsum(equation = var_21153_equation_0, values = (var_20661_cast_fp16, var_21060_cast_fp16))[name = tensor("op_21153_cast_fp16")]; + tensor var_21155_equation_0 = const()[name = tensor("op_21155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21155_cast_fp16 = einsum(equation = var_21155_equation_0, values = (var_20661_cast_fp16, var_21061_cast_fp16))[name = tensor("op_21155_cast_fp16")]; + tensor var_21157_equation_0 = const()[name = tensor("op_21157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21157_cast_fp16 = einsum(equation = var_21157_equation_0, values = (var_20661_cast_fp16, var_21062_cast_fp16))[name = tensor("op_21157_cast_fp16")]; + tensor var_21159_equation_0 = const()[name = tensor("op_21159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21159_cast_fp16 = einsum(equation = var_21159_equation_0, values = (var_20661_cast_fp16, var_21063_cast_fp16))[name = tensor("op_21159_cast_fp16")]; + tensor var_21161_equation_0 = const()[name = tensor("op_21161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21161_cast_fp16 = einsum(equation = var_21161_equation_0, values = (var_20665_cast_fp16, var_21064_cast_fp16))[name = tensor("op_21161_cast_fp16")]; + tensor var_21163_equation_0 = const()[name = tensor("op_21163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21163_cast_fp16 = einsum(equation = var_21163_equation_0, values = (var_20665_cast_fp16, var_21065_cast_fp16))[name = tensor("op_21163_cast_fp16")]; + tensor var_21165_equation_0 = const()[name = tensor("op_21165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21165_cast_fp16 = einsum(equation = var_21165_equation_0, values = (var_20665_cast_fp16, var_21066_cast_fp16))[name = tensor("op_21165_cast_fp16")]; + tensor var_21167_equation_0 = const()[name = tensor("op_21167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21167_cast_fp16 = einsum(equation = var_21167_equation_0, values = (var_20665_cast_fp16, var_21067_cast_fp16))[name = tensor("op_21167_cast_fp16")]; + tensor var_21169_equation_0 = const()[name = tensor("op_21169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21169_cast_fp16 = einsum(equation = var_21169_equation_0, values = (var_20669_cast_fp16, var_21068_cast_fp16))[name = tensor("op_21169_cast_fp16")]; + tensor var_21171_equation_0 = const()[name = tensor("op_21171_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21171_cast_fp16 = einsum(equation = var_21171_equation_0, values = (var_20669_cast_fp16, var_21069_cast_fp16))[name = tensor("op_21171_cast_fp16")]; + tensor var_21173_equation_0 = const()[name = tensor("op_21173_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21173_cast_fp16 = einsum(equation = var_21173_equation_0, values = (var_20669_cast_fp16, var_21070_cast_fp16))[name = tensor("op_21173_cast_fp16")]; + tensor var_21175_equation_0 = const()[name = tensor("op_21175_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21175_cast_fp16 = einsum(equation = var_21175_equation_0, values = (var_20669_cast_fp16, var_21071_cast_fp16))[name = tensor("op_21175_cast_fp16")]; + tensor var_21177_equation_0 = const()[name = tensor("op_21177_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21177_cast_fp16 = einsum(equation = var_21177_equation_0, values = (var_20673_cast_fp16, var_21072_cast_fp16))[name = tensor("op_21177_cast_fp16")]; + tensor var_21179_equation_0 = const()[name = tensor("op_21179_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21179_cast_fp16 = einsum(equation = var_21179_equation_0, values = (var_20673_cast_fp16, var_21073_cast_fp16))[name = tensor("op_21179_cast_fp16")]; + tensor var_21181_equation_0 = const()[name = tensor("op_21181_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21181_cast_fp16 = einsum(equation = var_21181_equation_0, values = (var_20673_cast_fp16, var_21074_cast_fp16))[name = tensor("op_21181_cast_fp16")]; + tensor var_21183_equation_0 = const()[name = tensor("op_21183_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21183_cast_fp16 = einsum(equation = var_21183_equation_0, values = (var_20673_cast_fp16, var_21075_cast_fp16))[name = tensor("op_21183_cast_fp16")]; + tensor var_21185_equation_0 = const()[name = tensor("op_21185_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21185_cast_fp16 = einsum(equation = var_21185_equation_0, values = (var_20677_cast_fp16, var_21076_cast_fp16))[name = tensor("op_21185_cast_fp16")]; + tensor var_21187_equation_0 = const()[name = tensor("op_21187_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21187_cast_fp16 = einsum(equation = var_21187_equation_0, values = (var_20677_cast_fp16, var_21077_cast_fp16))[name = tensor("op_21187_cast_fp16")]; + tensor var_21189_equation_0 = const()[name = tensor("op_21189_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21189_cast_fp16 = einsum(equation = var_21189_equation_0, values = (var_20677_cast_fp16, var_21078_cast_fp16))[name = tensor("op_21189_cast_fp16")]; + tensor var_21191_equation_0 = const()[name = tensor("op_21191_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21191_cast_fp16 = einsum(equation = var_21191_equation_0, values = (var_20677_cast_fp16, var_21079_cast_fp16))[name = tensor("op_21191_cast_fp16")]; + tensor var_21193_equation_0 = const()[name = tensor("op_21193_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21193_cast_fp16 = einsum(equation = var_21193_equation_0, values = (var_20681_cast_fp16, var_21080_cast_fp16))[name = tensor("op_21193_cast_fp16")]; + tensor var_21195_equation_0 = const()[name = tensor("op_21195_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21195_cast_fp16 = einsum(equation = var_21195_equation_0, values = (var_20681_cast_fp16, var_21081_cast_fp16))[name = tensor("op_21195_cast_fp16")]; + tensor var_21197_equation_0 = const()[name = tensor("op_21197_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21197_cast_fp16 = einsum(equation = var_21197_equation_0, values = (var_20681_cast_fp16, var_21082_cast_fp16))[name = tensor("op_21197_cast_fp16")]; + tensor var_21199_equation_0 = const()[name = tensor("op_21199_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21199_cast_fp16 = einsum(equation = var_21199_equation_0, values = (var_20681_cast_fp16, var_21083_cast_fp16))[name = tensor("op_21199_cast_fp16")]; + tensor var_21201_equation_0 = const()[name = tensor("op_21201_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21201_cast_fp16 = einsum(equation = var_21201_equation_0, values = (var_20685_cast_fp16, var_21084_cast_fp16))[name = tensor("op_21201_cast_fp16")]; + tensor var_21203_equation_0 = const()[name = tensor("op_21203_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21203_cast_fp16 = einsum(equation = var_21203_equation_0, values = (var_20685_cast_fp16, var_21085_cast_fp16))[name = tensor("op_21203_cast_fp16")]; + tensor var_21205_equation_0 = const()[name = tensor("op_21205_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21205_cast_fp16 = einsum(equation = var_21205_equation_0, values = (var_20685_cast_fp16, var_21086_cast_fp16))[name = tensor("op_21205_cast_fp16")]; + tensor var_21207_equation_0 = const()[name = tensor("op_21207_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21207_cast_fp16 = einsum(equation = var_21207_equation_0, values = (var_20685_cast_fp16, var_21087_cast_fp16))[name = tensor("op_21207_cast_fp16")]; + tensor var_21209_equation_0 = const()[name = tensor("op_21209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21209_cast_fp16 = einsum(equation = var_21209_equation_0, values = (var_20689_cast_fp16, var_21088_cast_fp16))[name = tensor("op_21209_cast_fp16")]; + tensor var_21211_equation_0 = const()[name = tensor("op_21211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21211_cast_fp16 = einsum(equation = var_21211_equation_0, values = (var_20689_cast_fp16, var_21089_cast_fp16))[name = tensor("op_21211_cast_fp16")]; + tensor var_21213_equation_0 = const()[name = tensor("op_21213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21213_cast_fp16 = einsum(equation = var_21213_equation_0, values = (var_20689_cast_fp16, var_21090_cast_fp16))[name = tensor("op_21213_cast_fp16")]; + tensor var_21215_equation_0 = const()[name = tensor("op_21215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21215_cast_fp16 = einsum(equation = var_21215_equation_0, values = (var_20689_cast_fp16, var_21091_cast_fp16))[name = tensor("op_21215_cast_fp16")]; + tensor var_21217_equation_0 = const()[name = tensor("op_21217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21217_cast_fp16 = einsum(equation = var_21217_equation_0, values = (var_20693_cast_fp16, var_21092_cast_fp16))[name = tensor("op_21217_cast_fp16")]; + tensor var_21219_equation_0 = const()[name = tensor("op_21219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21219_cast_fp16 = einsum(equation = var_21219_equation_0, values = (var_20693_cast_fp16, var_21093_cast_fp16))[name = tensor("op_21219_cast_fp16")]; + tensor var_21221_equation_0 = const()[name = tensor("op_21221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21221_cast_fp16 = einsum(equation = var_21221_equation_0, values = (var_20693_cast_fp16, var_21094_cast_fp16))[name = tensor("op_21221_cast_fp16")]; + tensor var_21223_equation_0 = const()[name = tensor("op_21223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21223_cast_fp16 = einsum(equation = var_21223_equation_0, values = (var_20693_cast_fp16, var_21095_cast_fp16))[name = tensor("op_21223_cast_fp16")]; + tensor var_21225_equation_0 = const()[name = tensor("op_21225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21225_cast_fp16 = einsum(equation = var_21225_equation_0, values = (var_20697_cast_fp16, var_21096_cast_fp16))[name = tensor("op_21225_cast_fp16")]; + tensor var_21227_equation_0 = const()[name = tensor("op_21227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21227_cast_fp16 = einsum(equation = var_21227_equation_0, values = (var_20697_cast_fp16, var_21097_cast_fp16))[name = tensor("op_21227_cast_fp16")]; + tensor var_21229_equation_0 = const()[name = tensor("op_21229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21229_cast_fp16 = einsum(equation = var_21229_equation_0, values = (var_20697_cast_fp16, var_21098_cast_fp16))[name = tensor("op_21229_cast_fp16")]; + tensor var_21231_equation_0 = const()[name = tensor("op_21231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21231_cast_fp16 = einsum(equation = var_21231_equation_0, values = (var_20697_cast_fp16, var_21099_cast_fp16))[name = tensor("op_21231_cast_fp16")]; + tensor var_21233_equation_0 = const()[name = tensor("op_21233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21233_cast_fp16 = einsum(equation = var_21233_equation_0, values = (var_20701_cast_fp16, var_21100_cast_fp16))[name = tensor("op_21233_cast_fp16")]; + tensor var_21235_equation_0 = const()[name = tensor("op_21235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21235_cast_fp16 = einsum(equation = var_21235_equation_0, values = (var_20701_cast_fp16, var_21101_cast_fp16))[name = tensor("op_21235_cast_fp16")]; + tensor var_21237_equation_0 = const()[name = tensor("op_21237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21237_cast_fp16 = einsum(equation = var_21237_equation_0, values = (var_20701_cast_fp16, var_21102_cast_fp16))[name = tensor("op_21237_cast_fp16")]; + tensor var_21239_equation_0 = const()[name = tensor("op_21239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21239_cast_fp16 = einsum(equation = var_21239_equation_0, values = (var_20701_cast_fp16, var_21103_cast_fp16))[name = tensor("op_21239_cast_fp16")]; + tensor var_21241_equation_0 = const()[name = tensor("op_21241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21241_cast_fp16 = einsum(equation = var_21241_equation_0, values = (var_20705_cast_fp16, var_21104_cast_fp16))[name = tensor("op_21241_cast_fp16")]; + tensor var_21243_equation_0 = const()[name = tensor("op_21243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21243_cast_fp16 = einsum(equation = var_21243_equation_0, values = (var_20705_cast_fp16, var_21105_cast_fp16))[name = tensor("op_21243_cast_fp16")]; + tensor var_21245_equation_0 = const()[name = tensor("op_21245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21245_cast_fp16 = einsum(equation = var_21245_equation_0, values = (var_20705_cast_fp16, var_21106_cast_fp16))[name = tensor("op_21245_cast_fp16")]; + tensor var_21247_equation_0 = const()[name = tensor("op_21247_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21247_cast_fp16 = einsum(equation = var_21247_equation_0, values = (var_20705_cast_fp16, var_21107_cast_fp16))[name = tensor("op_21247_cast_fp16")]; + tensor var_21249_equation_0 = const()[name = tensor("op_21249_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21249_cast_fp16 = einsum(equation = var_21249_equation_0, values = (var_20709_cast_fp16, var_21108_cast_fp16))[name = tensor("op_21249_cast_fp16")]; + tensor var_21251_equation_0 = const()[name = tensor("op_21251_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21251_cast_fp16 = einsum(equation = var_21251_equation_0, values = (var_20709_cast_fp16, var_21109_cast_fp16))[name = tensor("op_21251_cast_fp16")]; + tensor var_21253_equation_0 = const()[name = tensor("op_21253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21253_cast_fp16 = einsum(equation = var_21253_equation_0, values = (var_20709_cast_fp16, var_21110_cast_fp16))[name = tensor("op_21253_cast_fp16")]; + tensor var_21255_equation_0 = const()[name = tensor("op_21255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21255_cast_fp16 = einsum(equation = var_21255_equation_0, values = (var_20709_cast_fp16, var_21111_cast_fp16))[name = tensor("op_21255_cast_fp16")]; + tensor var_21257_equation_0 = const()[name = tensor("op_21257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21257_cast_fp16 = einsum(equation = var_21257_equation_0, values = (var_20713_cast_fp16, var_21112_cast_fp16))[name = tensor("op_21257_cast_fp16")]; + tensor var_21259_equation_0 = const()[name = tensor("op_21259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21259_cast_fp16 = einsum(equation = var_21259_equation_0, values = (var_20713_cast_fp16, var_21113_cast_fp16))[name = tensor("op_21259_cast_fp16")]; + tensor var_21261_equation_0 = const()[name = tensor("op_21261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21261_cast_fp16 = einsum(equation = var_21261_equation_0, values = (var_20713_cast_fp16, var_21114_cast_fp16))[name = tensor("op_21261_cast_fp16")]; + tensor var_21263_equation_0 = const()[name = tensor("op_21263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21263_cast_fp16 = einsum(equation = var_21263_equation_0, values = (var_20713_cast_fp16, var_21115_cast_fp16))[name = tensor("op_21263_cast_fp16")]; + tensor var_21265_equation_0 = const()[name = tensor("op_21265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21265_cast_fp16 = einsum(equation = var_21265_equation_0, values = (var_20717_cast_fp16, var_21116_cast_fp16))[name = tensor("op_21265_cast_fp16")]; + tensor var_21267_equation_0 = const()[name = tensor("op_21267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21267_cast_fp16 = einsum(equation = var_21267_equation_0, values = (var_20717_cast_fp16, var_21117_cast_fp16))[name = tensor("op_21267_cast_fp16")]; + tensor var_21269_equation_0 = const()[name = tensor("op_21269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21269_cast_fp16 = einsum(equation = var_21269_equation_0, values = (var_20717_cast_fp16, var_21118_cast_fp16))[name = tensor("op_21269_cast_fp16")]; + tensor var_21271_equation_0 = const()[name = tensor("op_21271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21271_cast_fp16 = einsum(equation = var_21271_equation_0, values = (var_20717_cast_fp16, var_21119_cast_fp16))[name = tensor("op_21271_cast_fp16")]; + tensor var_21273_equation_0 = const()[name = tensor("op_21273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21273_cast_fp16 = einsum(equation = var_21273_equation_0, values = (var_20721_cast_fp16, var_21120_cast_fp16))[name = tensor("op_21273_cast_fp16")]; + tensor var_21275_equation_0 = const()[name = tensor("op_21275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21275_cast_fp16 = einsum(equation = var_21275_equation_0, values = (var_20721_cast_fp16, var_21121_cast_fp16))[name = tensor("op_21275_cast_fp16")]; + tensor var_21277_equation_0 = const()[name = tensor("op_21277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21277_cast_fp16 = einsum(equation = var_21277_equation_0, values = (var_20721_cast_fp16, var_21122_cast_fp16))[name = tensor("op_21277_cast_fp16")]; + tensor var_21279_equation_0 = const()[name = tensor("op_21279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21279_cast_fp16 = einsum(equation = var_21279_equation_0, values = (var_20721_cast_fp16, var_21123_cast_fp16))[name = tensor("op_21279_cast_fp16")]; + tensor var_21281_equation_0 = const()[name = tensor("op_21281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21281_cast_fp16 = einsum(equation = var_21281_equation_0, values = (var_20725_cast_fp16, var_21124_cast_fp16))[name = tensor("op_21281_cast_fp16")]; + tensor var_21283_equation_0 = const()[name = tensor("op_21283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21283_cast_fp16 = einsum(equation = var_21283_equation_0, values = (var_20725_cast_fp16, var_21125_cast_fp16))[name = tensor("op_21283_cast_fp16")]; + tensor var_21285_equation_0 = const()[name = tensor("op_21285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21285_cast_fp16 = einsum(equation = var_21285_equation_0, values = (var_20725_cast_fp16, var_21126_cast_fp16))[name = tensor("op_21285_cast_fp16")]; + tensor var_21287_equation_0 = const()[name = tensor("op_21287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_21287_cast_fp16 = einsum(equation = var_21287_equation_0, values = (var_20725_cast_fp16, var_21127_cast_fp16))[name = tensor("op_21287_cast_fp16")]; + tensor var_21289_interleave_0 = const()[name = tensor("op_21289_interleave_0"), val = tensor(false)]; + tensor var_21289_cast_fp16 = concat(axis = var_19794, interleave = var_21289_interleave_0, values = (var_21129_cast_fp16, var_21131_cast_fp16, var_21133_cast_fp16, var_21135_cast_fp16))[name = tensor("op_21289_cast_fp16")]; + tensor var_21291_interleave_0 = const()[name = tensor("op_21291_interleave_0"), val = tensor(false)]; + tensor var_21291_cast_fp16 = concat(axis = var_19794, interleave = var_21291_interleave_0, values = (var_21137_cast_fp16, var_21139_cast_fp16, var_21141_cast_fp16, var_21143_cast_fp16))[name = tensor("op_21291_cast_fp16")]; + tensor var_21293_interleave_0 = const()[name = tensor("op_21293_interleave_0"), val = tensor(false)]; + tensor var_21293_cast_fp16 = concat(axis = var_19794, interleave = var_21293_interleave_0, values = (var_21145_cast_fp16, var_21147_cast_fp16, var_21149_cast_fp16, var_21151_cast_fp16))[name = tensor("op_21293_cast_fp16")]; + tensor var_21295_interleave_0 = const()[name = tensor("op_21295_interleave_0"), val = tensor(false)]; + tensor var_21295_cast_fp16 = concat(axis = var_19794, interleave = var_21295_interleave_0, values = (var_21153_cast_fp16, var_21155_cast_fp16, var_21157_cast_fp16, var_21159_cast_fp16))[name = tensor("op_21295_cast_fp16")]; + tensor var_21297_interleave_0 = const()[name = tensor("op_21297_interleave_0"), val = tensor(false)]; + tensor var_21297_cast_fp16 = concat(axis = var_19794, interleave = var_21297_interleave_0, values = (var_21161_cast_fp16, var_21163_cast_fp16, var_21165_cast_fp16, var_21167_cast_fp16))[name = tensor("op_21297_cast_fp16")]; + tensor var_21299_interleave_0 = const()[name = tensor("op_21299_interleave_0"), val = tensor(false)]; + tensor var_21299_cast_fp16 = concat(axis = var_19794, interleave = var_21299_interleave_0, values = (var_21169_cast_fp16, var_21171_cast_fp16, var_21173_cast_fp16, var_21175_cast_fp16))[name = tensor("op_21299_cast_fp16")]; + tensor var_21301_interleave_0 = const()[name = tensor("op_21301_interleave_0"), val = tensor(false)]; + tensor var_21301_cast_fp16 = concat(axis = var_19794, interleave = var_21301_interleave_0, values = (var_21177_cast_fp16, var_21179_cast_fp16, var_21181_cast_fp16, var_21183_cast_fp16))[name = tensor("op_21301_cast_fp16")]; + tensor var_21303_interleave_0 = const()[name = tensor("op_21303_interleave_0"), val = tensor(false)]; + tensor var_21303_cast_fp16 = concat(axis = var_19794, interleave = var_21303_interleave_0, values = (var_21185_cast_fp16, var_21187_cast_fp16, var_21189_cast_fp16, var_21191_cast_fp16))[name = tensor("op_21303_cast_fp16")]; + tensor var_21305_interleave_0 = const()[name = tensor("op_21305_interleave_0"), val = tensor(false)]; + tensor var_21305_cast_fp16 = concat(axis = var_19794, interleave = var_21305_interleave_0, values = (var_21193_cast_fp16, var_21195_cast_fp16, var_21197_cast_fp16, var_21199_cast_fp16))[name = tensor("op_21305_cast_fp16")]; + tensor var_21307_interleave_0 = const()[name = tensor("op_21307_interleave_0"), val = tensor(false)]; + tensor var_21307_cast_fp16 = concat(axis = var_19794, interleave = var_21307_interleave_0, values = (var_21201_cast_fp16, var_21203_cast_fp16, var_21205_cast_fp16, var_21207_cast_fp16))[name = tensor("op_21307_cast_fp16")]; + tensor var_21309_interleave_0 = const()[name = tensor("op_21309_interleave_0"), val = tensor(false)]; + tensor var_21309_cast_fp16 = concat(axis = var_19794, interleave = var_21309_interleave_0, values = (var_21209_cast_fp16, var_21211_cast_fp16, var_21213_cast_fp16, var_21215_cast_fp16))[name = tensor("op_21309_cast_fp16")]; + tensor var_21311_interleave_0 = const()[name = tensor("op_21311_interleave_0"), val = tensor(false)]; + tensor var_21311_cast_fp16 = concat(axis = var_19794, interleave = var_21311_interleave_0, values = (var_21217_cast_fp16, var_21219_cast_fp16, var_21221_cast_fp16, var_21223_cast_fp16))[name = tensor("op_21311_cast_fp16")]; + tensor var_21313_interleave_0 = const()[name = tensor("op_21313_interleave_0"), val = tensor(false)]; + tensor var_21313_cast_fp16 = concat(axis = var_19794, interleave = var_21313_interleave_0, values = (var_21225_cast_fp16, var_21227_cast_fp16, var_21229_cast_fp16, var_21231_cast_fp16))[name = tensor("op_21313_cast_fp16")]; + tensor var_21315_interleave_0 = const()[name = tensor("op_21315_interleave_0"), val = tensor(false)]; + tensor var_21315_cast_fp16 = concat(axis = var_19794, interleave = var_21315_interleave_0, values = (var_21233_cast_fp16, var_21235_cast_fp16, var_21237_cast_fp16, var_21239_cast_fp16))[name = tensor("op_21315_cast_fp16")]; + tensor var_21317_interleave_0 = const()[name = tensor("op_21317_interleave_0"), val = tensor(false)]; + tensor var_21317_cast_fp16 = concat(axis = var_19794, interleave = var_21317_interleave_0, values = (var_21241_cast_fp16, var_21243_cast_fp16, var_21245_cast_fp16, var_21247_cast_fp16))[name = tensor("op_21317_cast_fp16")]; + tensor var_21319_interleave_0 = const()[name = tensor("op_21319_interleave_0"), val = tensor(false)]; + tensor var_21319_cast_fp16 = concat(axis = var_19794, interleave = var_21319_interleave_0, values = (var_21249_cast_fp16, var_21251_cast_fp16, var_21253_cast_fp16, var_21255_cast_fp16))[name = tensor("op_21319_cast_fp16")]; + tensor var_21321_interleave_0 = const()[name = tensor("op_21321_interleave_0"), val = tensor(false)]; + tensor var_21321_cast_fp16 = concat(axis = var_19794, interleave = var_21321_interleave_0, values = (var_21257_cast_fp16, var_21259_cast_fp16, var_21261_cast_fp16, var_21263_cast_fp16))[name = tensor("op_21321_cast_fp16")]; + tensor var_21323_interleave_0 = const()[name = tensor("op_21323_interleave_0"), val = tensor(false)]; + tensor var_21323_cast_fp16 = concat(axis = var_19794, interleave = var_21323_interleave_0, values = (var_21265_cast_fp16, var_21267_cast_fp16, var_21269_cast_fp16, var_21271_cast_fp16))[name = tensor("op_21323_cast_fp16")]; + tensor var_21325_interleave_0 = const()[name = tensor("op_21325_interleave_0"), val = tensor(false)]; + tensor var_21325_cast_fp16 = concat(axis = var_19794, interleave = var_21325_interleave_0, values = (var_21273_cast_fp16, var_21275_cast_fp16, var_21277_cast_fp16, var_21279_cast_fp16))[name = tensor("op_21325_cast_fp16")]; + tensor var_21327_interleave_0 = const()[name = tensor("op_21327_interleave_0"), val = tensor(false)]; + tensor var_21327_cast_fp16 = concat(axis = var_19794, interleave = var_21327_interleave_0, values = (var_21281_cast_fp16, var_21283_cast_fp16, var_21285_cast_fp16, var_21287_cast_fp16))[name = tensor("op_21327_cast_fp16")]; + tensor input_247_interleave_0 = const()[name = tensor("input_247_interleave_0"), val = tensor(false)]; + tensor input_247_cast_fp16 = concat(axis = var_19819, interleave = input_247_interleave_0, values = (var_21289_cast_fp16, var_21291_cast_fp16, var_21293_cast_fp16, var_21295_cast_fp16, var_21297_cast_fp16, var_21299_cast_fp16, var_21301_cast_fp16, var_21303_cast_fp16, var_21305_cast_fp16, var_21307_cast_fp16, var_21309_cast_fp16, var_21311_cast_fp16, var_21313_cast_fp16, var_21315_cast_fp16, var_21317_cast_fp16, var_21319_cast_fp16, var_21321_cast_fp16, var_21323_cast_fp16, var_21325_cast_fp16, var_21327_cast_fp16))[name = tensor("input_247_cast_fp16")]; + tensor var_21335 = const()[name = tensor("op_21335"), val = tensor([1, 1])]; + tensor var_21337 = const()[name = tensor("op_21337"), val = tensor([1, 1])]; + tensor pretrained_out_151_pad_type_0 = const()[name = tensor("pretrained_out_151_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_151_pad_0 = const()[name = tensor("pretrained_out_151_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144231424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145050688))), name = tensor("layers_12_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_12_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145050816)))]; + tensor pretrained_out_151_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_21337, groups = var_19819, pad = pretrained_out_151_pad_0, pad_type = pretrained_out_151_pad_type_0, strides = var_21335, weight = layers_12_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_247_cast_fp16)[name = tensor("pretrained_out_151_cast_fp16")]; + tensor var_21341 = const()[name = tensor("op_21341"), val = tensor([1, 1])]; + tensor var_21343 = const()[name = tensor("op_21343"), val = tensor([1, 1])]; + tensor input_249_pad_type_0 = const()[name = tensor("input_249_pad_type_0"), val = tensor("custom")]; + tensor input_249_pad_0 = const()[name = tensor("input_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145053440)))]; + tensor input_249_cast_fp16 = conv(dilations = var_21343, groups = var_19819, pad = input_249_pad_0, pad_type = input_249_pad_type_0, strides = var_21341, weight = layers_12_self_attn_o_proj_loraA_weight_to_fp16, x = input_247_cast_fp16)[name = tensor("input_249_cast_fp16")]; + tensor var_21347 = const()[name = tensor("op_21347"), val = tensor([1, 1])]; + tensor var_21349 = const()[name = tensor("op_21349"), val = tensor([1, 1])]; + tensor lora_out_301_pad_type_0 = const()[name = tensor("lora_out_301_pad_type_0"), val = tensor("custom")]; + tensor lora_out_301_pad_0 = const()[name = tensor("lora_out_301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_303_weight_0_to_fp16 = const()[name = tensor("lora_out_303_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145094464)))]; + tensor lora_out_303_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_21349, groups = var_19819, pad = lora_out_301_pad_0, pad_type = lora_out_301_pad_type_0, strides = var_21347, weight = lora_out_303_weight_0_to_fp16, x = input_249_cast_fp16)[name = tensor("lora_out_303_cast_fp16")]; + tensor obj_51_cast_fp16 = add(x = pretrained_out_151_cast_fp16, y = lora_out_303_cast_fp16)[name = tensor("obj_51_cast_fp16")]; + tensor inputs_51_cast_fp16 = add(x = inputs_49_cast_fp16, y = obj_51_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; + tensor var_21358 = const()[name = tensor("op_21358"), val = tensor([1])]; + tensor channels_mean_51_cast_fp16 = reduce_mean(axes = var_21358, keep_dims = var_19820, x = inputs_51_cast_fp16)[name = tensor("channels_mean_51_cast_fp16")]; + tensor zero_mean_51_cast_fp16 = sub(x = inputs_51_cast_fp16, y = channels_mean_51_cast_fp16)[name = tensor("zero_mean_51_cast_fp16")]; + tensor zero_mean_sq_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = zero_mean_51_cast_fp16)[name = tensor("zero_mean_sq_51_cast_fp16")]; + tensor var_21362 = const()[name = tensor("op_21362"), val = tensor([1])]; + tensor var_21363_cast_fp16 = reduce_mean(axes = var_21362, keep_dims = var_19820, x = zero_mean_sq_51_cast_fp16)[name = tensor("op_21363_cast_fp16")]; + tensor var_21364_to_fp16 = const()[name = tensor("op_21364_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_21365_cast_fp16 = add(x = var_21363_cast_fp16, y = var_21364_to_fp16)[name = tensor("op_21365_cast_fp16")]; + tensor denom_51_epsilon_0 = const()[name = tensor("denom_51_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_51_cast_fp16 = rsqrt(epsilon = denom_51_epsilon_0, x = var_21365_cast_fp16)[name = tensor("denom_51_cast_fp16")]; + tensor out_51_cast_fp16 = mul(x = zero_mean_51_cast_fp16, y = denom_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = tensor("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145135488)))]; + tensor input_251_beta_0_to_fp16 = const()[name = tensor("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145138112)))]; + tensor input_251_epsilon_0_to_fp16 = const()[name = tensor("input_251_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor var_21379 = const()[name = tensor("op_21379"), val = tensor([1, 1])]; + tensor var_21381 = const()[name = tensor("op_21381"), val = tensor([1, 1])]; + tensor pretrained_out_153_pad_type_0 = const()[name = tensor("pretrained_out_153_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_153_pad_0 = const()[name = tensor("pretrained_out_153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145140736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148417600))), name = tensor("layers_12_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_12_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_12_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148417728)))]; + tensor pretrained_out_153_cast_fp16 = conv(bias = layers_12_fc1_pretrained_bias_to_fp16, dilations = var_21381, groups = var_19819, pad = pretrained_out_153_pad_0, pad_type = pretrained_out_153_pad_type_0, strides = var_21379, weight = layers_12_fc1_pretrained_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = tensor("pretrained_out_153_cast_fp16")]; + tensor var_21385 = const()[name = tensor("op_21385"), val = tensor([1, 1])]; + tensor var_21387 = const()[name = tensor("op_21387"), val = tensor([1, 1])]; + tensor input_253_pad_type_0 = const()[name = tensor("input_253_pad_type_0"), val = tensor("custom")]; + tensor input_253_pad_0 = const()[name = tensor("input_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_12_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148428032)))]; + tensor input_253_cast_fp16 = conv(dilations = var_21387, groups = var_19819, pad = input_253_pad_0, pad_type = input_253_pad_type_0, strides = var_21385, weight = layers_12_fc1_loraA_weight_to_fp16, x = input_251_cast_fp16)[name = tensor("input_253_cast_fp16")]; + tensor var_21391 = const()[name = tensor("op_21391"), val = tensor([1, 1])]; + tensor var_21393 = const()[name = tensor("op_21393"), val = tensor([1, 1])]; + tensor lora_out_305_pad_type_0 = const()[name = tensor("lora_out_305_pad_type_0"), val = tensor("custom")]; + tensor lora_out_305_pad_0 = const()[name = tensor("lora_out_305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_307_weight_0_to_fp16 = const()[name = tensor("lora_out_307_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148469056)))]; + tensor lora_out_307_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_21393, groups = var_19819, pad = lora_out_305_pad_0, pad_type = lora_out_305_pad_type_0, strides = var_21391, weight = lora_out_307_weight_0_to_fp16, x = input_253_cast_fp16)[name = tensor("lora_out_307_cast_fp16")]; + tensor input_255_cast_fp16 = add(x = pretrained_out_153_cast_fp16, y = lora_out_307_cast_fp16)[name = tensor("input_255_cast_fp16")]; + tensor input_257_mode_0 = const()[name = tensor("input_257_mode_0"), val = tensor("EXACT")]; + tensor input_257_cast_fp16 = gelu(mode = input_257_mode_0, x = input_255_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor var_21405 = const()[name = tensor("op_21405"), val = tensor([1, 1])]; + tensor var_21407 = const()[name = tensor("op_21407"), val = tensor([1, 1])]; + tensor pretrained_out_155_pad_type_0 = const()[name = tensor("pretrained_out_155_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_155_pad_0 = const()[name = tensor("pretrained_out_155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(148632960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151909824))), name = tensor("layers_12_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_12_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_12_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151909952)))]; + tensor pretrained_out_155_cast_fp16 = conv(bias = layers_12_fc2_pretrained_bias_to_fp16, dilations = var_21407, groups = var_19819, pad = pretrained_out_155_pad_0, pad_type = pretrained_out_155_pad_type_0, strides = var_21405, weight = layers_12_fc2_pretrained_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = tensor("pretrained_out_155_cast_fp16")]; + tensor var_21411 = const()[name = tensor("op_21411"), val = tensor([1, 1])]; + tensor var_21413 = const()[name = tensor("op_21413"), val = tensor([1, 1])]; + tensor input_259_pad_type_0 = const()[name = tensor("input_259_pad_type_0"), val = tensor("custom")]; + tensor input_259_pad_0 = const()[name = tensor("input_259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_12_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_12_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151912576)))]; + tensor input_259_cast_fp16 = conv(dilations = var_21413, groups = var_19819, pad = input_259_pad_0, pad_type = input_259_pad_type_0, strides = var_21411, weight = layers_12_fc2_loraA_weight_to_fp16, x = input_257_cast_fp16)[name = tensor("input_259_cast_fp16")]; + tensor var_21417 = const()[name = tensor("op_21417"), val = tensor([1, 1])]; + tensor var_21419 = const()[name = tensor("op_21419"), val = tensor([1, 1])]; + tensor lora_out_309_pad_type_0 = const()[name = tensor("lora_out_309_pad_type_0"), val = tensor("custom")]; + tensor lora_out_309_pad_0 = const()[name = tensor("lora_out_309_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_311_weight_0_to_fp16 = const()[name = tensor("lora_out_311_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152076480)))]; + tensor lora_out_311_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_21419, groups = var_19819, pad = lora_out_309_pad_0, pad_type = lora_out_309_pad_type_0, strides = var_21417, weight = lora_out_311_weight_0_to_fp16, x = input_259_cast_fp16)[name = tensor("lora_out_311_cast_fp16")]; + tensor hidden_states_29_cast_fp16 = add(x = pretrained_out_155_cast_fp16, y = lora_out_311_cast_fp16)[name = tensor("hidden_states_29_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = hidden_states_29_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; + tensor var_21429 = const()[name = tensor("op_21429"), val = tensor(3)]; + tensor var_21454 = const()[name = tensor("op_21454"), val = tensor(1)]; + tensor var_21455 = const()[name = tensor("op_21455"), val = tensor(true)]; + tensor var_21465 = const()[name = tensor("op_21465"), val = tensor([1])]; + tensor channels_mean_53_cast_fp16 = reduce_mean(axes = var_21465, keep_dims = var_21455, x = inputs_53_cast_fp16)[name = tensor("channels_mean_53_cast_fp16")]; + tensor zero_mean_53_cast_fp16 = sub(x = inputs_53_cast_fp16, y = channels_mean_53_cast_fp16)[name = tensor("zero_mean_53_cast_fp16")]; + tensor zero_mean_sq_53_cast_fp16 = mul(x = zero_mean_53_cast_fp16, y = zero_mean_53_cast_fp16)[name = tensor("zero_mean_sq_53_cast_fp16")]; + tensor var_21469 = const()[name = tensor("op_21469"), val = tensor([1])]; + tensor var_21470_cast_fp16 = reduce_mean(axes = var_21469, keep_dims = var_21455, x = zero_mean_sq_53_cast_fp16)[name = tensor("op_21470_cast_fp16")]; + tensor var_21471_to_fp16 = const()[name = tensor("op_21471_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_21472_cast_fp16 = add(x = var_21470_cast_fp16, y = var_21471_to_fp16)[name = tensor("op_21472_cast_fp16")]; + tensor denom_53_epsilon_0 = const()[name = tensor("denom_53_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_53_cast_fp16 = rsqrt(epsilon = denom_53_epsilon_0, x = var_21472_cast_fp16)[name = tensor("denom_53_cast_fp16")]; + tensor out_53_cast_fp16 = mul(x = zero_mean_53_cast_fp16, y = denom_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; + tensor obj_53_gamma_0_to_fp16 = const()[name = tensor("obj_53_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152117504)))]; + tensor obj_53_beta_0_to_fp16 = const()[name = tensor("obj_53_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152120128)))]; + tensor obj_53_epsilon_0_to_fp16 = const()[name = tensor("obj_53_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_53_cast_fp16 = batch_norm(beta = obj_53_beta_0_to_fp16, epsilon = obj_53_epsilon_0_to_fp16, gamma = obj_53_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor("obj_53_cast_fp16")]; + tensor var_21490 = const()[name = tensor("op_21490"), val = tensor([1, 1])]; + tensor var_21492 = const()[name = tensor("op_21492"), val = tensor([1, 1])]; + tensor pretrained_out_157_pad_type_0 = const()[name = tensor("pretrained_out_157_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_157_pad_0 = const()[name = tensor("pretrained_out_157_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152122752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152942016))), name = tensor("layers_13_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_13_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152942144)))]; + tensor pretrained_out_157_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_21492, groups = var_21454, pad = pretrained_out_157_pad_0, pad_type = pretrained_out_157_pad_type_0, strides = var_21490, weight = layers_13_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_53_cast_fp16)[name = tensor("pretrained_out_157_cast_fp16")]; + tensor var_21496 = const()[name = tensor("op_21496"), val = tensor([1, 1])]; + tensor var_21498 = const()[name = tensor("op_21498"), val = tensor([1, 1])]; + tensor input_261_pad_type_0 = const()[name = tensor("input_261_pad_type_0"), val = tensor("custom")]; + tensor input_261_pad_0 = const()[name = tensor("input_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152944768)))]; + tensor input_261_cast_fp16 = conv(dilations = var_21498, groups = var_21454, pad = input_261_pad_0, pad_type = input_261_pad_type_0, strides = var_21496, weight = layers_13_self_attn_q_proj_loraA_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor("input_261_cast_fp16")]; + tensor var_21502 = const()[name = tensor("op_21502"), val = tensor([1, 1])]; + tensor var_21504 = const()[name = tensor("op_21504"), val = tensor([1, 1])]; + tensor lora_out_313_pad_type_0 = const()[name = tensor("lora_out_313_pad_type_0"), val = tensor("custom")]; + tensor lora_out_313_pad_0 = const()[name = tensor("lora_out_313_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_315_weight_0_to_fp16 = const()[name = tensor("lora_out_315_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152985792)))]; + tensor lora_out_315_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_21504, groups = var_21454, pad = lora_out_313_pad_0, pad_type = lora_out_313_pad_type_0, strides = var_21502, weight = lora_out_315_weight_0_to_fp16, x = input_261_cast_fp16)[name = tensor("lora_out_315_cast_fp16")]; + tensor query_27_cast_fp16 = add(x = pretrained_out_157_cast_fp16, y = lora_out_315_cast_fp16)[name = tensor("query_27_cast_fp16")]; + tensor var_21514 = const()[name = tensor("op_21514"), val = tensor([1, 1])]; + tensor var_21516 = const()[name = tensor("op_21516"), val = tensor([1, 1])]; + tensor pretrained_out_159_pad_type_0 = const()[name = tensor("pretrained_out_159_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_159_pad_0 = const()[name = tensor("pretrained_out_159_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153026816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153846080))), name = tensor("layers_13_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_159_cast_fp16 = conv(dilations = var_21516, groups = var_21454, pad = pretrained_out_159_pad_0, pad_type = pretrained_out_159_pad_type_0, strides = var_21514, weight = layers_13_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_53_cast_fp16)[name = tensor("pretrained_out_159_cast_fp16")]; + tensor var_21520 = const()[name = tensor("op_21520"), val = tensor([1, 1])]; + tensor var_21522 = const()[name = tensor("op_21522"), val = tensor([1, 1])]; + tensor input_263_pad_type_0 = const()[name = tensor("input_263_pad_type_0"), val = tensor("custom")]; + tensor input_263_pad_0 = const()[name = tensor("input_263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153846208)))]; + tensor input_263_cast_fp16 = conv(dilations = var_21522, groups = var_21454, pad = input_263_pad_0, pad_type = input_263_pad_type_0, strides = var_21520, weight = layers_13_self_attn_k_proj_loraA_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor("input_263_cast_fp16")]; + tensor var_21526 = const()[name = tensor("op_21526"), val = tensor([1, 1])]; + tensor var_21528 = const()[name = tensor("op_21528"), val = tensor([1, 1])]; + tensor lora_out_317_pad_type_0 = const()[name = tensor("lora_out_317_pad_type_0"), val = tensor("custom")]; + tensor lora_out_317_pad_0 = const()[name = tensor("lora_out_317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_319_weight_0_to_fp16 = const()[name = tensor("lora_out_319_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153887232)))]; + tensor lora_out_319_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_21528, groups = var_21454, pad = lora_out_317_pad_0, pad_type = lora_out_317_pad_type_0, strides = var_21526, weight = lora_out_319_weight_0_to_fp16, x = input_263_cast_fp16)[name = tensor("lora_out_319_cast_fp16")]; + tensor key_27_cast_fp16 = add(x = pretrained_out_159_cast_fp16, y = lora_out_319_cast_fp16)[name = tensor("key_27_cast_fp16")]; + tensor var_21539 = const()[name = tensor("op_21539"), val = tensor([1, 1])]; + tensor var_21541 = const()[name = tensor("op_21541"), val = tensor([1, 1])]; + tensor pretrained_out_161_pad_type_0 = const()[name = tensor("pretrained_out_161_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_161_pad_0 = const()[name = tensor("pretrained_out_161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153928256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154747520))), name = tensor("layers_13_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_13_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154747648)))]; + tensor pretrained_out_161_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_21541, groups = var_21454, pad = pretrained_out_161_pad_0, pad_type = pretrained_out_161_pad_type_0, strides = var_21539, weight = layers_13_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_53_cast_fp16)[name = tensor("pretrained_out_161_cast_fp16")]; + tensor var_21545 = const()[name = tensor("op_21545"), val = tensor([1, 1])]; + tensor var_21547 = const()[name = tensor("op_21547"), val = tensor([1, 1])]; + tensor input_265_pad_type_0 = const()[name = tensor("input_265_pad_type_0"), val = tensor("custom")]; + tensor input_265_pad_0 = const()[name = tensor("input_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154750272)))]; + tensor input_265_cast_fp16 = conv(dilations = var_21547, groups = var_21454, pad = input_265_pad_0, pad_type = input_265_pad_type_0, strides = var_21545, weight = layers_13_self_attn_v_proj_loraA_weight_to_fp16, x = obj_53_cast_fp16)[name = tensor("input_265_cast_fp16")]; + tensor var_21551 = const()[name = tensor("op_21551"), val = tensor([1, 1])]; + tensor var_21553 = const()[name = tensor("op_21553"), val = tensor([1, 1])]; + tensor lora_out_321_pad_type_0 = const()[name = tensor("lora_out_321_pad_type_0"), val = tensor("custom")]; + tensor lora_out_321_pad_0 = const()[name = tensor("lora_out_321_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_323_weight_0_to_fp16 = const()[name = tensor("lora_out_323_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154791296)))]; + tensor lora_out_323_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_21553, groups = var_21454, pad = lora_out_321_pad_0, pad_type = lora_out_321_pad_type_0, strides = var_21551, weight = lora_out_323_weight_0_to_fp16, x = input_265_cast_fp16)[name = tensor("lora_out_323_cast_fp16")]; + tensor value_27_cast_fp16 = add(x = pretrained_out_161_cast_fp16, y = lora_out_323_cast_fp16)[name = tensor("value_27_cast_fp16")]; + tensor var_21563_begin_0 = const()[name = tensor("op_21563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21563_end_0 = const()[name = tensor("op_21563_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21563_end_mask_0 = const()[name = tensor("op_21563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21563_cast_fp16 = slice_by_index(begin = var_21563_begin_0, end = var_21563_end_0, end_mask = var_21563_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21563_cast_fp16")]; + tensor var_21567_begin_0 = const()[name = tensor("op_21567_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_21567_end_0 = const()[name = tensor("op_21567_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_21567_end_mask_0 = const()[name = tensor("op_21567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21567_cast_fp16 = slice_by_index(begin = var_21567_begin_0, end = var_21567_end_0, end_mask = var_21567_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21567_cast_fp16")]; + tensor var_21571_begin_0 = const()[name = tensor("op_21571_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_21571_end_0 = const()[name = tensor("op_21571_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_21571_end_mask_0 = const()[name = tensor("op_21571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21571_cast_fp16 = slice_by_index(begin = var_21571_begin_0, end = var_21571_end_0, end_mask = var_21571_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21571_cast_fp16")]; + tensor var_21575_begin_0 = const()[name = tensor("op_21575_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_21575_end_0 = const()[name = tensor("op_21575_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_21575_end_mask_0 = const()[name = tensor("op_21575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21575_cast_fp16 = slice_by_index(begin = var_21575_begin_0, end = var_21575_end_0, end_mask = var_21575_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21575_cast_fp16")]; + tensor var_21579_begin_0 = const()[name = tensor("op_21579_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_21579_end_0 = const()[name = tensor("op_21579_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_21579_end_mask_0 = const()[name = tensor("op_21579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21579_cast_fp16 = slice_by_index(begin = var_21579_begin_0, end = var_21579_end_0, end_mask = var_21579_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21579_cast_fp16")]; + tensor var_21583_begin_0 = const()[name = tensor("op_21583_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_21583_end_0 = const()[name = tensor("op_21583_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_21583_end_mask_0 = const()[name = tensor("op_21583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21583_cast_fp16 = slice_by_index(begin = var_21583_begin_0, end = var_21583_end_0, end_mask = var_21583_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21583_cast_fp16")]; + tensor var_21587_begin_0 = const()[name = tensor("op_21587_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_21587_end_0 = const()[name = tensor("op_21587_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_21587_end_mask_0 = const()[name = tensor("op_21587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21587_cast_fp16 = slice_by_index(begin = var_21587_begin_0, end = var_21587_end_0, end_mask = var_21587_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21587_cast_fp16")]; + tensor var_21591_begin_0 = const()[name = tensor("op_21591_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_21591_end_0 = const()[name = tensor("op_21591_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_21591_end_mask_0 = const()[name = tensor("op_21591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21591_cast_fp16 = slice_by_index(begin = var_21591_begin_0, end = var_21591_end_0, end_mask = var_21591_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21591_cast_fp16")]; + tensor var_21595_begin_0 = const()[name = tensor("op_21595_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_21595_end_0 = const()[name = tensor("op_21595_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_21595_end_mask_0 = const()[name = tensor("op_21595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21595_cast_fp16 = slice_by_index(begin = var_21595_begin_0, end = var_21595_end_0, end_mask = var_21595_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21595_cast_fp16")]; + tensor var_21599_begin_0 = const()[name = tensor("op_21599_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_21599_end_0 = const()[name = tensor("op_21599_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_21599_end_mask_0 = const()[name = tensor("op_21599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21599_cast_fp16 = slice_by_index(begin = var_21599_begin_0, end = var_21599_end_0, end_mask = var_21599_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21599_cast_fp16")]; + tensor var_21603_begin_0 = const()[name = tensor("op_21603_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_21603_end_0 = const()[name = tensor("op_21603_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_21603_end_mask_0 = const()[name = tensor("op_21603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21603_cast_fp16 = slice_by_index(begin = var_21603_begin_0, end = var_21603_end_0, end_mask = var_21603_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21603_cast_fp16")]; + tensor var_21607_begin_0 = const()[name = tensor("op_21607_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_21607_end_0 = const()[name = tensor("op_21607_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_21607_end_mask_0 = const()[name = tensor("op_21607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21607_cast_fp16 = slice_by_index(begin = var_21607_begin_0, end = var_21607_end_0, end_mask = var_21607_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21607_cast_fp16")]; + tensor var_21611_begin_0 = const()[name = tensor("op_21611_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_21611_end_0 = const()[name = tensor("op_21611_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_21611_end_mask_0 = const()[name = tensor("op_21611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21611_cast_fp16 = slice_by_index(begin = var_21611_begin_0, end = var_21611_end_0, end_mask = var_21611_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21611_cast_fp16")]; + tensor var_21615_begin_0 = const()[name = tensor("op_21615_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_21615_end_0 = const()[name = tensor("op_21615_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_21615_end_mask_0 = const()[name = tensor("op_21615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21615_cast_fp16 = slice_by_index(begin = var_21615_begin_0, end = var_21615_end_0, end_mask = var_21615_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21615_cast_fp16")]; + tensor var_21619_begin_0 = const()[name = tensor("op_21619_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_21619_end_0 = const()[name = tensor("op_21619_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_21619_end_mask_0 = const()[name = tensor("op_21619_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21619_cast_fp16 = slice_by_index(begin = var_21619_begin_0, end = var_21619_end_0, end_mask = var_21619_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21619_cast_fp16")]; + tensor var_21623_begin_0 = const()[name = tensor("op_21623_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_21623_end_0 = const()[name = tensor("op_21623_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_21623_end_mask_0 = const()[name = tensor("op_21623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21623_cast_fp16 = slice_by_index(begin = var_21623_begin_0, end = var_21623_end_0, end_mask = var_21623_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21623_cast_fp16")]; + tensor var_21627_begin_0 = const()[name = tensor("op_21627_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_21627_end_0 = const()[name = tensor("op_21627_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_21627_end_mask_0 = const()[name = tensor("op_21627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21627_cast_fp16 = slice_by_index(begin = var_21627_begin_0, end = var_21627_end_0, end_mask = var_21627_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21627_cast_fp16")]; + tensor var_21631_begin_0 = const()[name = tensor("op_21631_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_21631_end_0 = const()[name = tensor("op_21631_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_21631_end_mask_0 = const()[name = tensor("op_21631_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21631_cast_fp16 = slice_by_index(begin = var_21631_begin_0, end = var_21631_end_0, end_mask = var_21631_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21631_cast_fp16")]; + tensor var_21635_begin_0 = const()[name = tensor("op_21635_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_21635_end_0 = const()[name = tensor("op_21635_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_21635_end_mask_0 = const()[name = tensor("op_21635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21635_cast_fp16 = slice_by_index(begin = var_21635_begin_0, end = var_21635_end_0, end_mask = var_21635_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21635_cast_fp16")]; + tensor var_21639_begin_0 = const()[name = tensor("op_21639_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_21639_end_0 = const()[name = tensor("op_21639_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_21639_end_mask_0 = const()[name = tensor("op_21639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_21639_cast_fp16 = slice_by_index(begin = var_21639_begin_0, end = var_21639_end_0, end_mask = var_21639_end_mask_0, x = query_27_cast_fp16)[name = tensor("op_21639_cast_fp16")]; + tensor var_21648_begin_0 = const()[name = tensor("op_21648_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21648_end_0 = const()[name = tensor("op_21648_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21648_end_mask_0 = const()[name = tensor("op_21648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21648_cast_fp16 = slice_by_index(begin = var_21648_begin_0, end = var_21648_end_0, end_mask = var_21648_end_mask_0, x = var_21563_cast_fp16)[name = tensor("op_21648_cast_fp16")]; + tensor var_21655_begin_0 = const()[name = tensor("op_21655_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21655_end_0 = const()[name = tensor("op_21655_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21655_end_mask_0 = const()[name = tensor("op_21655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21655_cast_fp16 = slice_by_index(begin = var_21655_begin_0, end = var_21655_end_0, end_mask = var_21655_end_mask_0, x = var_21563_cast_fp16)[name = tensor("op_21655_cast_fp16")]; + tensor var_21662_begin_0 = const()[name = tensor("op_21662_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21662_end_0 = const()[name = tensor("op_21662_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21662_end_mask_0 = const()[name = tensor("op_21662_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21662_cast_fp16 = slice_by_index(begin = var_21662_begin_0, end = var_21662_end_0, end_mask = var_21662_end_mask_0, x = var_21563_cast_fp16)[name = tensor("op_21662_cast_fp16")]; + tensor var_21669_begin_0 = const()[name = tensor("op_21669_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21669_end_0 = const()[name = tensor("op_21669_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21669_end_mask_0 = const()[name = tensor("op_21669_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21669_cast_fp16 = slice_by_index(begin = var_21669_begin_0, end = var_21669_end_0, end_mask = var_21669_end_mask_0, x = var_21563_cast_fp16)[name = tensor("op_21669_cast_fp16")]; + tensor var_21676_begin_0 = const()[name = tensor("op_21676_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21676_end_0 = const()[name = tensor("op_21676_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21676_end_mask_0 = const()[name = tensor("op_21676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21676_cast_fp16 = slice_by_index(begin = var_21676_begin_0, end = var_21676_end_0, end_mask = var_21676_end_mask_0, x = var_21567_cast_fp16)[name = tensor("op_21676_cast_fp16")]; + tensor var_21683_begin_0 = const()[name = tensor("op_21683_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21683_end_0 = const()[name = tensor("op_21683_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21683_end_mask_0 = const()[name = tensor("op_21683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21683_cast_fp16 = slice_by_index(begin = var_21683_begin_0, end = var_21683_end_0, end_mask = var_21683_end_mask_0, x = var_21567_cast_fp16)[name = tensor("op_21683_cast_fp16")]; + tensor var_21690_begin_0 = const()[name = tensor("op_21690_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21690_end_0 = const()[name = tensor("op_21690_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21690_end_mask_0 = const()[name = tensor("op_21690_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21690_cast_fp16 = slice_by_index(begin = var_21690_begin_0, end = var_21690_end_0, end_mask = var_21690_end_mask_0, x = var_21567_cast_fp16)[name = tensor("op_21690_cast_fp16")]; + tensor var_21697_begin_0 = const()[name = tensor("op_21697_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21697_end_0 = const()[name = tensor("op_21697_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21697_end_mask_0 = const()[name = tensor("op_21697_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21697_cast_fp16 = slice_by_index(begin = var_21697_begin_0, end = var_21697_end_0, end_mask = var_21697_end_mask_0, x = var_21567_cast_fp16)[name = tensor("op_21697_cast_fp16")]; + tensor var_21704_begin_0 = const()[name = tensor("op_21704_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21704_end_0 = const()[name = tensor("op_21704_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21704_end_mask_0 = const()[name = tensor("op_21704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21704_cast_fp16 = slice_by_index(begin = var_21704_begin_0, end = var_21704_end_0, end_mask = var_21704_end_mask_0, x = var_21571_cast_fp16)[name = tensor("op_21704_cast_fp16")]; + tensor var_21711_begin_0 = const()[name = tensor("op_21711_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21711_end_0 = const()[name = tensor("op_21711_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21711_end_mask_0 = const()[name = tensor("op_21711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21711_cast_fp16 = slice_by_index(begin = var_21711_begin_0, end = var_21711_end_0, end_mask = var_21711_end_mask_0, x = var_21571_cast_fp16)[name = tensor("op_21711_cast_fp16")]; + tensor var_21718_begin_0 = const()[name = tensor("op_21718_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21718_end_0 = const()[name = tensor("op_21718_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21718_end_mask_0 = const()[name = tensor("op_21718_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21718_cast_fp16 = slice_by_index(begin = var_21718_begin_0, end = var_21718_end_0, end_mask = var_21718_end_mask_0, x = var_21571_cast_fp16)[name = tensor("op_21718_cast_fp16")]; + tensor var_21725_begin_0 = const()[name = tensor("op_21725_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21725_end_0 = const()[name = tensor("op_21725_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21725_end_mask_0 = const()[name = tensor("op_21725_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21725_cast_fp16 = slice_by_index(begin = var_21725_begin_0, end = var_21725_end_0, end_mask = var_21725_end_mask_0, x = var_21571_cast_fp16)[name = tensor("op_21725_cast_fp16")]; + tensor var_21732_begin_0 = const()[name = tensor("op_21732_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21732_end_0 = const()[name = tensor("op_21732_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21732_end_mask_0 = const()[name = tensor("op_21732_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21732_cast_fp16 = slice_by_index(begin = var_21732_begin_0, end = var_21732_end_0, end_mask = var_21732_end_mask_0, x = var_21575_cast_fp16)[name = tensor("op_21732_cast_fp16")]; + tensor var_21739_begin_0 = const()[name = tensor("op_21739_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21739_end_0 = const()[name = tensor("op_21739_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21739_end_mask_0 = const()[name = tensor("op_21739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21739_cast_fp16 = slice_by_index(begin = var_21739_begin_0, end = var_21739_end_0, end_mask = var_21739_end_mask_0, x = var_21575_cast_fp16)[name = tensor("op_21739_cast_fp16")]; + tensor var_21746_begin_0 = const()[name = tensor("op_21746_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21746_end_0 = const()[name = tensor("op_21746_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21746_end_mask_0 = const()[name = tensor("op_21746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21746_cast_fp16 = slice_by_index(begin = var_21746_begin_0, end = var_21746_end_0, end_mask = var_21746_end_mask_0, x = var_21575_cast_fp16)[name = tensor("op_21746_cast_fp16")]; + tensor var_21753_begin_0 = const()[name = tensor("op_21753_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21753_end_0 = const()[name = tensor("op_21753_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21753_end_mask_0 = const()[name = tensor("op_21753_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21753_cast_fp16 = slice_by_index(begin = var_21753_begin_0, end = var_21753_end_0, end_mask = var_21753_end_mask_0, x = var_21575_cast_fp16)[name = tensor("op_21753_cast_fp16")]; + tensor var_21760_begin_0 = const()[name = tensor("op_21760_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21760_end_0 = const()[name = tensor("op_21760_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21760_end_mask_0 = const()[name = tensor("op_21760_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21760_cast_fp16 = slice_by_index(begin = var_21760_begin_0, end = var_21760_end_0, end_mask = var_21760_end_mask_0, x = var_21579_cast_fp16)[name = tensor("op_21760_cast_fp16")]; + tensor var_21767_begin_0 = const()[name = tensor("op_21767_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21767_end_0 = const()[name = tensor("op_21767_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21767_end_mask_0 = const()[name = tensor("op_21767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21767_cast_fp16 = slice_by_index(begin = var_21767_begin_0, end = var_21767_end_0, end_mask = var_21767_end_mask_0, x = var_21579_cast_fp16)[name = tensor("op_21767_cast_fp16")]; + tensor var_21774_begin_0 = const()[name = tensor("op_21774_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21774_end_0 = const()[name = tensor("op_21774_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21774_end_mask_0 = const()[name = tensor("op_21774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21774_cast_fp16 = slice_by_index(begin = var_21774_begin_0, end = var_21774_end_0, end_mask = var_21774_end_mask_0, x = var_21579_cast_fp16)[name = tensor("op_21774_cast_fp16")]; + tensor var_21781_begin_0 = const()[name = tensor("op_21781_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21781_end_0 = const()[name = tensor("op_21781_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21781_end_mask_0 = const()[name = tensor("op_21781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21781_cast_fp16 = slice_by_index(begin = var_21781_begin_0, end = var_21781_end_0, end_mask = var_21781_end_mask_0, x = var_21579_cast_fp16)[name = tensor("op_21781_cast_fp16")]; + tensor var_21788_begin_0 = const()[name = tensor("op_21788_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21788_end_0 = const()[name = tensor("op_21788_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21788_end_mask_0 = const()[name = tensor("op_21788_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21788_cast_fp16 = slice_by_index(begin = var_21788_begin_0, end = var_21788_end_0, end_mask = var_21788_end_mask_0, x = var_21583_cast_fp16)[name = tensor("op_21788_cast_fp16")]; + tensor var_21795_begin_0 = const()[name = tensor("op_21795_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21795_end_0 = const()[name = tensor("op_21795_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21795_end_mask_0 = const()[name = tensor("op_21795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21795_cast_fp16 = slice_by_index(begin = var_21795_begin_0, end = var_21795_end_0, end_mask = var_21795_end_mask_0, x = var_21583_cast_fp16)[name = tensor("op_21795_cast_fp16")]; + tensor var_21802_begin_0 = const()[name = tensor("op_21802_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21802_end_0 = const()[name = tensor("op_21802_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21802_end_mask_0 = const()[name = tensor("op_21802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21802_cast_fp16 = slice_by_index(begin = var_21802_begin_0, end = var_21802_end_0, end_mask = var_21802_end_mask_0, x = var_21583_cast_fp16)[name = tensor("op_21802_cast_fp16")]; + tensor var_21809_begin_0 = const()[name = tensor("op_21809_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21809_end_0 = const()[name = tensor("op_21809_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21809_end_mask_0 = const()[name = tensor("op_21809_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21809_cast_fp16 = slice_by_index(begin = var_21809_begin_0, end = var_21809_end_0, end_mask = var_21809_end_mask_0, x = var_21583_cast_fp16)[name = tensor("op_21809_cast_fp16")]; + tensor var_21816_begin_0 = const()[name = tensor("op_21816_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21816_end_0 = const()[name = tensor("op_21816_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21816_end_mask_0 = const()[name = tensor("op_21816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21816_cast_fp16 = slice_by_index(begin = var_21816_begin_0, end = var_21816_end_0, end_mask = var_21816_end_mask_0, x = var_21587_cast_fp16)[name = tensor("op_21816_cast_fp16")]; + tensor var_21823_begin_0 = const()[name = tensor("op_21823_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21823_end_0 = const()[name = tensor("op_21823_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21823_end_mask_0 = const()[name = tensor("op_21823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21823_cast_fp16 = slice_by_index(begin = var_21823_begin_0, end = var_21823_end_0, end_mask = var_21823_end_mask_0, x = var_21587_cast_fp16)[name = tensor("op_21823_cast_fp16")]; + tensor var_21830_begin_0 = const()[name = tensor("op_21830_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21830_end_0 = const()[name = tensor("op_21830_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21830_end_mask_0 = const()[name = tensor("op_21830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21830_cast_fp16 = slice_by_index(begin = var_21830_begin_0, end = var_21830_end_0, end_mask = var_21830_end_mask_0, x = var_21587_cast_fp16)[name = tensor("op_21830_cast_fp16")]; + tensor var_21837_begin_0 = const()[name = tensor("op_21837_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21837_end_0 = const()[name = tensor("op_21837_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21837_end_mask_0 = const()[name = tensor("op_21837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21837_cast_fp16 = slice_by_index(begin = var_21837_begin_0, end = var_21837_end_0, end_mask = var_21837_end_mask_0, x = var_21587_cast_fp16)[name = tensor("op_21837_cast_fp16")]; + tensor var_21844_begin_0 = const()[name = tensor("op_21844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21844_end_0 = const()[name = tensor("op_21844_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21844_end_mask_0 = const()[name = tensor("op_21844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21844_cast_fp16 = slice_by_index(begin = var_21844_begin_0, end = var_21844_end_0, end_mask = var_21844_end_mask_0, x = var_21591_cast_fp16)[name = tensor("op_21844_cast_fp16")]; + tensor var_21851_begin_0 = const()[name = tensor("op_21851_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21851_end_0 = const()[name = tensor("op_21851_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21851_end_mask_0 = const()[name = tensor("op_21851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21851_cast_fp16 = slice_by_index(begin = var_21851_begin_0, end = var_21851_end_0, end_mask = var_21851_end_mask_0, x = var_21591_cast_fp16)[name = tensor("op_21851_cast_fp16")]; + tensor var_21858_begin_0 = const()[name = tensor("op_21858_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21858_end_0 = const()[name = tensor("op_21858_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21858_end_mask_0 = const()[name = tensor("op_21858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21858_cast_fp16 = slice_by_index(begin = var_21858_begin_0, end = var_21858_end_0, end_mask = var_21858_end_mask_0, x = var_21591_cast_fp16)[name = tensor("op_21858_cast_fp16")]; + tensor var_21865_begin_0 = const()[name = tensor("op_21865_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21865_end_0 = const()[name = tensor("op_21865_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21865_end_mask_0 = const()[name = tensor("op_21865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21865_cast_fp16 = slice_by_index(begin = var_21865_begin_0, end = var_21865_end_0, end_mask = var_21865_end_mask_0, x = var_21591_cast_fp16)[name = tensor("op_21865_cast_fp16")]; + tensor var_21872_begin_0 = const()[name = tensor("op_21872_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21872_end_0 = const()[name = tensor("op_21872_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21872_end_mask_0 = const()[name = tensor("op_21872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21872_cast_fp16 = slice_by_index(begin = var_21872_begin_0, end = var_21872_end_0, end_mask = var_21872_end_mask_0, x = var_21595_cast_fp16)[name = tensor("op_21872_cast_fp16")]; + tensor var_21879_begin_0 = const()[name = tensor("op_21879_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21879_end_0 = const()[name = tensor("op_21879_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21879_end_mask_0 = const()[name = tensor("op_21879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21879_cast_fp16 = slice_by_index(begin = var_21879_begin_0, end = var_21879_end_0, end_mask = var_21879_end_mask_0, x = var_21595_cast_fp16)[name = tensor("op_21879_cast_fp16")]; + tensor var_21886_begin_0 = const()[name = tensor("op_21886_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21886_end_0 = const()[name = tensor("op_21886_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21886_end_mask_0 = const()[name = tensor("op_21886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21886_cast_fp16 = slice_by_index(begin = var_21886_begin_0, end = var_21886_end_0, end_mask = var_21886_end_mask_0, x = var_21595_cast_fp16)[name = tensor("op_21886_cast_fp16")]; + tensor var_21893_begin_0 = const()[name = tensor("op_21893_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21893_end_0 = const()[name = tensor("op_21893_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21893_end_mask_0 = const()[name = tensor("op_21893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21893_cast_fp16 = slice_by_index(begin = var_21893_begin_0, end = var_21893_end_0, end_mask = var_21893_end_mask_0, x = var_21595_cast_fp16)[name = tensor("op_21893_cast_fp16")]; + tensor var_21900_begin_0 = const()[name = tensor("op_21900_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21900_end_0 = const()[name = tensor("op_21900_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21900_end_mask_0 = const()[name = tensor("op_21900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21900_cast_fp16 = slice_by_index(begin = var_21900_begin_0, end = var_21900_end_0, end_mask = var_21900_end_mask_0, x = var_21599_cast_fp16)[name = tensor("op_21900_cast_fp16")]; + tensor var_21907_begin_0 = const()[name = tensor("op_21907_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21907_end_0 = const()[name = tensor("op_21907_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21907_end_mask_0 = const()[name = tensor("op_21907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21907_cast_fp16 = slice_by_index(begin = var_21907_begin_0, end = var_21907_end_0, end_mask = var_21907_end_mask_0, x = var_21599_cast_fp16)[name = tensor("op_21907_cast_fp16")]; + tensor var_21914_begin_0 = const()[name = tensor("op_21914_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21914_end_0 = const()[name = tensor("op_21914_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21914_end_mask_0 = const()[name = tensor("op_21914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21914_cast_fp16 = slice_by_index(begin = var_21914_begin_0, end = var_21914_end_0, end_mask = var_21914_end_mask_0, x = var_21599_cast_fp16)[name = tensor("op_21914_cast_fp16")]; + tensor var_21921_begin_0 = const()[name = tensor("op_21921_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21921_end_0 = const()[name = tensor("op_21921_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21921_end_mask_0 = const()[name = tensor("op_21921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21921_cast_fp16 = slice_by_index(begin = var_21921_begin_0, end = var_21921_end_0, end_mask = var_21921_end_mask_0, x = var_21599_cast_fp16)[name = tensor("op_21921_cast_fp16")]; + tensor var_21928_begin_0 = const()[name = tensor("op_21928_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21928_end_0 = const()[name = tensor("op_21928_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21928_end_mask_0 = const()[name = tensor("op_21928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21928_cast_fp16 = slice_by_index(begin = var_21928_begin_0, end = var_21928_end_0, end_mask = var_21928_end_mask_0, x = var_21603_cast_fp16)[name = tensor("op_21928_cast_fp16")]; + tensor var_21935_begin_0 = const()[name = tensor("op_21935_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21935_end_0 = const()[name = tensor("op_21935_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21935_end_mask_0 = const()[name = tensor("op_21935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21935_cast_fp16 = slice_by_index(begin = var_21935_begin_0, end = var_21935_end_0, end_mask = var_21935_end_mask_0, x = var_21603_cast_fp16)[name = tensor("op_21935_cast_fp16")]; + tensor var_21942_begin_0 = const()[name = tensor("op_21942_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21942_end_0 = const()[name = tensor("op_21942_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21942_end_mask_0 = const()[name = tensor("op_21942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21942_cast_fp16 = slice_by_index(begin = var_21942_begin_0, end = var_21942_end_0, end_mask = var_21942_end_mask_0, x = var_21603_cast_fp16)[name = tensor("op_21942_cast_fp16")]; + tensor var_21949_begin_0 = const()[name = tensor("op_21949_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21949_end_0 = const()[name = tensor("op_21949_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21949_end_mask_0 = const()[name = tensor("op_21949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21949_cast_fp16 = slice_by_index(begin = var_21949_begin_0, end = var_21949_end_0, end_mask = var_21949_end_mask_0, x = var_21603_cast_fp16)[name = tensor("op_21949_cast_fp16")]; + tensor var_21956_begin_0 = const()[name = tensor("op_21956_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21956_end_0 = const()[name = tensor("op_21956_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21956_end_mask_0 = const()[name = tensor("op_21956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21956_cast_fp16 = slice_by_index(begin = var_21956_begin_0, end = var_21956_end_0, end_mask = var_21956_end_mask_0, x = var_21607_cast_fp16)[name = tensor("op_21956_cast_fp16")]; + tensor var_21963_begin_0 = const()[name = tensor("op_21963_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21963_end_0 = const()[name = tensor("op_21963_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21963_end_mask_0 = const()[name = tensor("op_21963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21963_cast_fp16 = slice_by_index(begin = var_21963_begin_0, end = var_21963_end_0, end_mask = var_21963_end_mask_0, x = var_21607_cast_fp16)[name = tensor("op_21963_cast_fp16")]; + tensor var_21970_begin_0 = const()[name = tensor("op_21970_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21970_end_0 = const()[name = tensor("op_21970_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21970_end_mask_0 = const()[name = tensor("op_21970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21970_cast_fp16 = slice_by_index(begin = var_21970_begin_0, end = var_21970_end_0, end_mask = var_21970_end_mask_0, x = var_21607_cast_fp16)[name = tensor("op_21970_cast_fp16")]; + tensor var_21977_begin_0 = const()[name = tensor("op_21977_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_21977_end_0 = const()[name = tensor("op_21977_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_21977_end_mask_0 = const()[name = tensor("op_21977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21977_cast_fp16 = slice_by_index(begin = var_21977_begin_0, end = var_21977_end_0, end_mask = var_21977_end_mask_0, x = var_21607_cast_fp16)[name = tensor("op_21977_cast_fp16")]; + tensor var_21984_begin_0 = const()[name = tensor("op_21984_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_21984_end_0 = const()[name = tensor("op_21984_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_21984_end_mask_0 = const()[name = tensor("op_21984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21984_cast_fp16 = slice_by_index(begin = var_21984_begin_0, end = var_21984_end_0, end_mask = var_21984_end_mask_0, x = var_21611_cast_fp16)[name = tensor("op_21984_cast_fp16")]; + tensor var_21991_begin_0 = const()[name = tensor("op_21991_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_21991_end_0 = const()[name = tensor("op_21991_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_21991_end_mask_0 = const()[name = tensor("op_21991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21991_cast_fp16 = slice_by_index(begin = var_21991_begin_0, end = var_21991_end_0, end_mask = var_21991_end_mask_0, x = var_21611_cast_fp16)[name = tensor("op_21991_cast_fp16")]; + tensor var_21998_begin_0 = const()[name = tensor("op_21998_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_21998_end_0 = const()[name = tensor("op_21998_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_21998_end_mask_0 = const()[name = tensor("op_21998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_21998_cast_fp16 = slice_by_index(begin = var_21998_begin_0, end = var_21998_end_0, end_mask = var_21998_end_mask_0, x = var_21611_cast_fp16)[name = tensor("op_21998_cast_fp16")]; + tensor var_22005_begin_0 = const()[name = tensor("op_22005_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22005_end_0 = const()[name = tensor("op_22005_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22005_end_mask_0 = const()[name = tensor("op_22005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22005_cast_fp16 = slice_by_index(begin = var_22005_begin_0, end = var_22005_end_0, end_mask = var_22005_end_mask_0, x = var_21611_cast_fp16)[name = tensor("op_22005_cast_fp16")]; + tensor var_22012_begin_0 = const()[name = tensor("op_22012_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22012_end_0 = const()[name = tensor("op_22012_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22012_end_mask_0 = const()[name = tensor("op_22012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22012_cast_fp16 = slice_by_index(begin = var_22012_begin_0, end = var_22012_end_0, end_mask = var_22012_end_mask_0, x = var_21615_cast_fp16)[name = tensor("op_22012_cast_fp16")]; + tensor var_22019_begin_0 = const()[name = tensor("op_22019_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22019_end_0 = const()[name = tensor("op_22019_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22019_end_mask_0 = const()[name = tensor("op_22019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22019_cast_fp16 = slice_by_index(begin = var_22019_begin_0, end = var_22019_end_0, end_mask = var_22019_end_mask_0, x = var_21615_cast_fp16)[name = tensor("op_22019_cast_fp16")]; + tensor var_22026_begin_0 = const()[name = tensor("op_22026_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22026_end_0 = const()[name = tensor("op_22026_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22026_end_mask_0 = const()[name = tensor("op_22026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22026_cast_fp16 = slice_by_index(begin = var_22026_begin_0, end = var_22026_end_0, end_mask = var_22026_end_mask_0, x = var_21615_cast_fp16)[name = tensor("op_22026_cast_fp16")]; + tensor var_22033_begin_0 = const()[name = tensor("op_22033_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22033_end_0 = const()[name = tensor("op_22033_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22033_end_mask_0 = const()[name = tensor("op_22033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22033_cast_fp16 = slice_by_index(begin = var_22033_begin_0, end = var_22033_end_0, end_mask = var_22033_end_mask_0, x = var_21615_cast_fp16)[name = tensor("op_22033_cast_fp16")]; + tensor var_22040_begin_0 = const()[name = tensor("op_22040_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22040_end_0 = const()[name = tensor("op_22040_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22040_end_mask_0 = const()[name = tensor("op_22040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22040_cast_fp16 = slice_by_index(begin = var_22040_begin_0, end = var_22040_end_0, end_mask = var_22040_end_mask_0, x = var_21619_cast_fp16)[name = tensor("op_22040_cast_fp16")]; + tensor var_22047_begin_0 = const()[name = tensor("op_22047_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22047_end_0 = const()[name = tensor("op_22047_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22047_end_mask_0 = const()[name = tensor("op_22047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22047_cast_fp16 = slice_by_index(begin = var_22047_begin_0, end = var_22047_end_0, end_mask = var_22047_end_mask_0, x = var_21619_cast_fp16)[name = tensor("op_22047_cast_fp16")]; + tensor var_22054_begin_0 = const()[name = tensor("op_22054_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22054_end_0 = const()[name = tensor("op_22054_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22054_end_mask_0 = const()[name = tensor("op_22054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22054_cast_fp16 = slice_by_index(begin = var_22054_begin_0, end = var_22054_end_0, end_mask = var_22054_end_mask_0, x = var_21619_cast_fp16)[name = tensor("op_22054_cast_fp16")]; + tensor var_22061_begin_0 = const()[name = tensor("op_22061_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22061_end_0 = const()[name = tensor("op_22061_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22061_end_mask_0 = const()[name = tensor("op_22061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22061_cast_fp16 = slice_by_index(begin = var_22061_begin_0, end = var_22061_end_0, end_mask = var_22061_end_mask_0, x = var_21619_cast_fp16)[name = tensor("op_22061_cast_fp16")]; + tensor var_22068_begin_0 = const()[name = tensor("op_22068_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22068_end_0 = const()[name = tensor("op_22068_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22068_end_mask_0 = const()[name = tensor("op_22068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22068_cast_fp16 = slice_by_index(begin = var_22068_begin_0, end = var_22068_end_0, end_mask = var_22068_end_mask_0, x = var_21623_cast_fp16)[name = tensor("op_22068_cast_fp16")]; + tensor var_22075_begin_0 = const()[name = tensor("op_22075_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22075_end_0 = const()[name = tensor("op_22075_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22075_end_mask_0 = const()[name = tensor("op_22075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22075_cast_fp16 = slice_by_index(begin = var_22075_begin_0, end = var_22075_end_0, end_mask = var_22075_end_mask_0, x = var_21623_cast_fp16)[name = tensor("op_22075_cast_fp16")]; + tensor var_22082_begin_0 = const()[name = tensor("op_22082_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22082_end_0 = const()[name = tensor("op_22082_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22082_end_mask_0 = const()[name = tensor("op_22082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22082_cast_fp16 = slice_by_index(begin = var_22082_begin_0, end = var_22082_end_0, end_mask = var_22082_end_mask_0, x = var_21623_cast_fp16)[name = tensor("op_22082_cast_fp16")]; + tensor var_22089_begin_0 = const()[name = tensor("op_22089_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22089_end_0 = const()[name = tensor("op_22089_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22089_end_mask_0 = const()[name = tensor("op_22089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22089_cast_fp16 = slice_by_index(begin = var_22089_begin_0, end = var_22089_end_0, end_mask = var_22089_end_mask_0, x = var_21623_cast_fp16)[name = tensor("op_22089_cast_fp16")]; + tensor var_22096_begin_0 = const()[name = tensor("op_22096_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22096_end_0 = const()[name = tensor("op_22096_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22096_end_mask_0 = const()[name = tensor("op_22096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22096_cast_fp16 = slice_by_index(begin = var_22096_begin_0, end = var_22096_end_0, end_mask = var_22096_end_mask_0, x = var_21627_cast_fp16)[name = tensor("op_22096_cast_fp16")]; + tensor var_22103_begin_0 = const()[name = tensor("op_22103_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22103_end_0 = const()[name = tensor("op_22103_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22103_end_mask_0 = const()[name = tensor("op_22103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22103_cast_fp16 = slice_by_index(begin = var_22103_begin_0, end = var_22103_end_0, end_mask = var_22103_end_mask_0, x = var_21627_cast_fp16)[name = tensor("op_22103_cast_fp16")]; + tensor var_22110_begin_0 = const()[name = tensor("op_22110_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22110_end_0 = const()[name = tensor("op_22110_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22110_end_mask_0 = const()[name = tensor("op_22110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22110_cast_fp16 = slice_by_index(begin = var_22110_begin_0, end = var_22110_end_0, end_mask = var_22110_end_mask_0, x = var_21627_cast_fp16)[name = tensor("op_22110_cast_fp16")]; + tensor var_22117_begin_0 = const()[name = tensor("op_22117_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22117_end_0 = const()[name = tensor("op_22117_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22117_end_mask_0 = const()[name = tensor("op_22117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22117_cast_fp16 = slice_by_index(begin = var_22117_begin_0, end = var_22117_end_0, end_mask = var_22117_end_mask_0, x = var_21627_cast_fp16)[name = tensor("op_22117_cast_fp16")]; + tensor var_22124_begin_0 = const()[name = tensor("op_22124_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22124_end_0 = const()[name = tensor("op_22124_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22124_end_mask_0 = const()[name = tensor("op_22124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22124_cast_fp16 = slice_by_index(begin = var_22124_begin_0, end = var_22124_end_0, end_mask = var_22124_end_mask_0, x = var_21631_cast_fp16)[name = tensor("op_22124_cast_fp16")]; + tensor var_22131_begin_0 = const()[name = tensor("op_22131_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22131_end_0 = const()[name = tensor("op_22131_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22131_end_mask_0 = const()[name = tensor("op_22131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22131_cast_fp16 = slice_by_index(begin = var_22131_begin_0, end = var_22131_end_0, end_mask = var_22131_end_mask_0, x = var_21631_cast_fp16)[name = tensor("op_22131_cast_fp16")]; + tensor var_22138_begin_0 = const()[name = tensor("op_22138_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22138_end_0 = const()[name = tensor("op_22138_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22138_end_mask_0 = const()[name = tensor("op_22138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22138_cast_fp16 = slice_by_index(begin = var_22138_begin_0, end = var_22138_end_0, end_mask = var_22138_end_mask_0, x = var_21631_cast_fp16)[name = tensor("op_22138_cast_fp16")]; + tensor var_22145_begin_0 = const()[name = tensor("op_22145_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22145_end_0 = const()[name = tensor("op_22145_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22145_end_mask_0 = const()[name = tensor("op_22145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22145_cast_fp16 = slice_by_index(begin = var_22145_begin_0, end = var_22145_end_0, end_mask = var_22145_end_mask_0, x = var_21631_cast_fp16)[name = tensor("op_22145_cast_fp16")]; + tensor var_22152_begin_0 = const()[name = tensor("op_22152_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22152_end_0 = const()[name = tensor("op_22152_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22152_end_mask_0 = const()[name = tensor("op_22152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22152_cast_fp16 = slice_by_index(begin = var_22152_begin_0, end = var_22152_end_0, end_mask = var_22152_end_mask_0, x = var_21635_cast_fp16)[name = tensor("op_22152_cast_fp16")]; + tensor var_22159_begin_0 = const()[name = tensor("op_22159_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22159_end_0 = const()[name = tensor("op_22159_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22159_end_mask_0 = const()[name = tensor("op_22159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22159_cast_fp16 = slice_by_index(begin = var_22159_begin_0, end = var_22159_end_0, end_mask = var_22159_end_mask_0, x = var_21635_cast_fp16)[name = tensor("op_22159_cast_fp16")]; + tensor var_22166_begin_0 = const()[name = tensor("op_22166_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22166_end_0 = const()[name = tensor("op_22166_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22166_end_mask_0 = const()[name = tensor("op_22166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22166_cast_fp16 = slice_by_index(begin = var_22166_begin_0, end = var_22166_end_0, end_mask = var_22166_end_mask_0, x = var_21635_cast_fp16)[name = tensor("op_22166_cast_fp16")]; + tensor var_22173_begin_0 = const()[name = tensor("op_22173_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22173_end_0 = const()[name = tensor("op_22173_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22173_end_mask_0 = const()[name = tensor("op_22173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22173_cast_fp16 = slice_by_index(begin = var_22173_begin_0, end = var_22173_end_0, end_mask = var_22173_end_mask_0, x = var_21635_cast_fp16)[name = tensor("op_22173_cast_fp16")]; + tensor var_22180_begin_0 = const()[name = tensor("op_22180_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22180_end_0 = const()[name = tensor("op_22180_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_22180_end_mask_0 = const()[name = tensor("op_22180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22180_cast_fp16 = slice_by_index(begin = var_22180_begin_0, end = var_22180_end_0, end_mask = var_22180_end_mask_0, x = var_21639_cast_fp16)[name = tensor("op_22180_cast_fp16")]; + tensor var_22187_begin_0 = const()[name = tensor("op_22187_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_22187_end_0 = const()[name = tensor("op_22187_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_22187_end_mask_0 = const()[name = tensor("op_22187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22187_cast_fp16 = slice_by_index(begin = var_22187_begin_0, end = var_22187_end_0, end_mask = var_22187_end_mask_0, x = var_21639_cast_fp16)[name = tensor("op_22187_cast_fp16")]; + tensor var_22194_begin_0 = const()[name = tensor("op_22194_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_22194_end_0 = const()[name = tensor("op_22194_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_22194_end_mask_0 = const()[name = tensor("op_22194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22194_cast_fp16 = slice_by_index(begin = var_22194_begin_0, end = var_22194_end_0, end_mask = var_22194_end_mask_0, x = var_21639_cast_fp16)[name = tensor("op_22194_cast_fp16")]; + tensor var_22201_begin_0 = const()[name = tensor("op_22201_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_22201_end_0 = const()[name = tensor("op_22201_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22201_end_mask_0 = const()[name = tensor("op_22201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22201_cast_fp16 = slice_by_index(begin = var_22201_begin_0, end = var_22201_end_0, end_mask = var_22201_end_mask_0, x = var_21639_cast_fp16)[name = tensor("op_22201_cast_fp16")]; + tensor k_27_perm_0 = const()[name = tensor("k_27_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_22206_begin_0 = const()[name = tensor("op_22206_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22206_end_0 = const()[name = tensor("op_22206_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_22206_end_mask_0 = const()[name = tensor("op_22206_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_18 = transpose(perm = k_27_perm_0, x = key_27_cast_fp16)[name = tensor("transpose_18")]; + tensor var_22206_cast_fp16 = slice_by_index(begin = var_22206_begin_0, end = var_22206_end_0, end_mask = var_22206_end_mask_0, x = transpose_18)[name = tensor("op_22206_cast_fp16")]; + tensor var_22210_begin_0 = const()[name = tensor("op_22210_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_22210_end_0 = const()[name = tensor("op_22210_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_22210_end_mask_0 = const()[name = tensor("op_22210_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22210_cast_fp16 = slice_by_index(begin = var_22210_begin_0, end = var_22210_end_0, end_mask = var_22210_end_mask_0, x = transpose_18)[name = tensor("op_22210_cast_fp16")]; + tensor var_22214_begin_0 = const()[name = tensor("op_22214_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_22214_end_0 = const()[name = tensor("op_22214_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_22214_end_mask_0 = const()[name = tensor("op_22214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22214_cast_fp16 = slice_by_index(begin = var_22214_begin_0, end = var_22214_end_0, end_mask = var_22214_end_mask_0, x = transpose_18)[name = tensor("op_22214_cast_fp16")]; + tensor var_22218_begin_0 = const()[name = tensor("op_22218_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_22218_end_0 = const()[name = tensor("op_22218_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_22218_end_mask_0 = const()[name = tensor("op_22218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22218_cast_fp16 = slice_by_index(begin = var_22218_begin_0, end = var_22218_end_0, end_mask = var_22218_end_mask_0, x = transpose_18)[name = tensor("op_22218_cast_fp16")]; + tensor var_22222_begin_0 = const()[name = tensor("op_22222_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_22222_end_0 = const()[name = tensor("op_22222_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_22222_end_mask_0 = const()[name = tensor("op_22222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22222_cast_fp16 = slice_by_index(begin = var_22222_begin_0, end = var_22222_end_0, end_mask = var_22222_end_mask_0, x = transpose_18)[name = tensor("op_22222_cast_fp16")]; + tensor var_22226_begin_0 = const()[name = tensor("op_22226_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_22226_end_0 = const()[name = tensor("op_22226_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_22226_end_mask_0 = const()[name = tensor("op_22226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22226_cast_fp16 = slice_by_index(begin = var_22226_begin_0, end = var_22226_end_0, end_mask = var_22226_end_mask_0, x = transpose_18)[name = tensor("op_22226_cast_fp16")]; + tensor var_22230_begin_0 = const()[name = tensor("op_22230_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_22230_end_0 = const()[name = tensor("op_22230_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_22230_end_mask_0 = const()[name = tensor("op_22230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22230_cast_fp16 = slice_by_index(begin = var_22230_begin_0, end = var_22230_end_0, end_mask = var_22230_end_mask_0, x = transpose_18)[name = tensor("op_22230_cast_fp16")]; + tensor var_22234_begin_0 = const()[name = tensor("op_22234_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_22234_end_0 = const()[name = tensor("op_22234_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_22234_end_mask_0 = const()[name = tensor("op_22234_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22234_cast_fp16 = slice_by_index(begin = var_22234_begin_0, end = var_22234_end_0, end_mask = var_22234_end_mask_0, x = transpose_18)[name = tensor("op_22234_cast_fp16")]; + tensor var_22238_begin_0 = const()[name = tensor("op_22238_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_22238_end_0 = const()[name = tensor("op_22238_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_22238_end_mask_0 = const()[name = tensor("op_22238_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22238_cast_fp16 = slice_by_index(begin = var_22238_begin_0, end = var_22238_end_0, end_mask = var_22238_end_mask_0, x = transpose_18)[name = tensor("op_22238_cast_fp16")]; + tensor var_22242_begin_0 = const()[name = tensor("op_22242_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_22242_end_0 = const()[name = tensor("op_22242_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_22242_end_mask_0 = const()[name = tensor("op_22242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22242_cast_fp16 = slice_by_index(begin = var_22242_begin_0, end = var_22242_end_0, end_mask = var_22242_end_mask_0, x = transpose_18)[name = tensor("op_22242_cast_fp16")]; + tensor var_22246_begin_0 = const()[name = tensor("op_22246_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_22246_end_0 = const()[name = tensor("op_22246_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_22246_end_mask_0 = const()[name = tensor("op_22246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22246_cast_fp16 = slice_by_index(begin = var_22246_begin_0, end = var_22246_end_0, end_mask = var_22246_end_mask_0, x = transpose_18)[name = tensor("op_22246_cast_fp16")]; + tensor var_22250_begin_0 = const()[name = tensor("op_22250_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_22250_end_0 = const()[name = tensor("op_22250_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_22250_end_mask_0 = const()[name = tensor("op_22250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22250_cast_fp16 = slice_by_index(begin = var_22250_begin_0, end = var_22250_end_0, end_mask = var_22250_end_mask_0, x = transpose_18)[name = tensor("op_22250_cast_fp16")]; + tensor var_22254_begin_0 = const()[name = tensor("op_22254_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_22254_end_0 = const()[name = tensor("op_22254_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_22254_end_mask_0 = const()[name = tensor("op_22254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22254_cast_fp16 = slice_by_index(begin = var_22254_begin_0, end = var_22254_end_0, end_mask = var_22254_end_mask_0, x = transpose_18)[name = tensor("op_22254_cast_fp16")]; + tensor var_22258_begin_0 = const()[name = tensor("op_22258_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_22258_end_0 = const()[name = tensor("op_22258_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_22258_end_mask_0 = const()[name = tensor("op_22258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22258_cast_fp16 = slice_by_index(begin = var_22258_begin_0, end = var_22258_end_0, end_mask = var_22258_end_mask_0, x = transpose_18)[name = tensor("op_22258_cast_fp16")]; + tensor var_22262_begin_0 = const()[name = tensor("op_22262_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_22262_end_0 = const()[name = tensor("op_22262_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_22262_end_mask_0 = const()[name = tensor("op_22262_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22262_cast_fp16 = slice_by_index(begin = var_22262_begin_0, end = var_22262_end_0, end_mask = var_22262_end_mask_0, x = transpose_18)[name = tensor("op_22262_cast_fp16")]; + tensor var_22266_begin_0 = const()[name = tensor("op_22266_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_22266_end_0 = const()[name = tensor("op_22266_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_22266_end_mask_0 = const()[name = tensor("op_22266_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22266_cast_fp16 = slice_by_index(begin = var_22266_begin_0, end = var_22266_end_0, end_mask = var_22266_end_mask_0, x = transpose_18)[name = tensor("op_22266_cast_fp16")]; + tensor var_22270_begin_0 = const()[name = tensor("op_22270_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_22270_end_0 = const()[name = tensor("op_22270_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_22270_end_mask_0 = const()[name = tensor("op_22270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22270_cast_fp16 = slice_by_index(begin = var_22270_begin_0, end = var_22270_end_0, end_mask = var_22270_end_mask_0, x = transpose_18)[name = tensor("op_22270_cast_fp16")]; + tensor var_22274_begin_0 = const()[name = tensor("op_22274_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_22274_end_0 = const()[name = tensor("op_22274_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_22274_end_mask_0 = const()[name = tensor("op_22274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22274_cast_fp16 = slice_by_index(begin = var_22274_begin_0, end = var_22274_end_0, end_mask = var_22274_end_mask_0, x = transpose_18)[name = tensor("op_22274_cast_fp16")]; + tensor var_22278_begin_0 = const()[name = tensor("op_22278_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_22278_end_0 = const()[name = tensor("op_22278_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_22278_end_mask_0 = const()[name = tensor("op_22278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22278_cast_fp16 = slice_by_index(begin = var_22278_begin_0, end = var_22278_end_0, end_mask = var_22278_end_mask_0, x = transpose_18)[name = tensor("op_22278_cast_fp16")]; + tensor var_22282_begin_0 = const()[name = tensor("op_22282_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_22282_end_0 = const()[name = tensor("op_22282_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_22282_end_mask_0 = const()[name = tensor("op_22282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_22282_cast_fp16 = slice_by_index(begin = var_22282_begin_0, end = var_22282_end_0, end_mask = var_22282_end_mask_0, x = transpose_18)[name = tensor("op_22282_cast_fp16")]; + tensor var_22284_begin_0 = const()[name = tensor("op_22284_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_22284_end_0 = const()[name = tensor("op_22284_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_22284_end_mask_0 = const()[name = tensor("op_22284_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22284_cast_fp16 = slice_by_index(begin = var_22284_begin_0, end = var_22284_end_0, end_mask = var_22284_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22284_cast_fp16")]; + tensor var_22288_begin_0 = const()[name = tensor("op_22288_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_22288_end_0 = const()[name = tensor("op_22288_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_22288_end_mask_0 = const()[name = tensor("op_22288_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22288_cast_fp16 = slice_by_index(begin = var_22288_begin_0, end = var_22288_end_0, end_mask = var_22288_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22288_cast_fp16")]; + tensor var_22292_begin_0 = const()[name = tensor("op_22292_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_22292_end_0 = const()[name = tensor("op_22292_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_22292_end_mask_0 = const()[name = tensor("op_22292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22292_cast_fp16 = slice_by_index(begin = var_22292_begin_0, end = var_22292_end_0, end_mask = var_22292_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22292_cast_fp16")]; + tensor var_22296_begin_0 = const()[name = tensor("op_22296_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_22296_end_0 = const()[name = tensor("op_22296_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_22296_end_mask_0 = const()[name = tensor("op_22296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22296_cast_fp16 = slice_by_index(begin = var_22296_begin_0, end = var_22296_end_0, end_mask = var_22296_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22296_cast_fp16")]; + tensor var_22300_begin_0 = const()[name = tensor("op_22300_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_22300_end_0 = const()[name = tensor("op_22300_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_22300_end_mask_0 = const()[name = tensor("op_22300_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22300_cast_fp16 = slice_by_index(begin = var_22300_begin_0, end = var_22300_end_0, end_mask = var_22300_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22300_cast_fp16")]; + tensor var_22304_begin_0 = const()[name = tensor("op_22304_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_22304_end_0 = const()[name = tensor("op_22304_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_22304_end_mask_0 = const()[name = tensor("op_22304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22304_cast_fp16 = slice_by_index(begin = var_22304_begin_0, end = var_22304_end_0, end_mask = var_22304_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22304_cast_fp16")]; + tensor var_22308_begin_0 = const()[name = tensor("op_22308_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_22308_end_0 = const()[name = tensor("op_22308_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_22308_end_mask_0 = const()[name = tensor("op_22308_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22308_cast_fp16 = slice_by_index(begin = var_22308_begin_0, end = var_22308_end_0, end_mask = var_22308_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22308_cast_fp16")]; + tensor var_22312_begin_0 = const()[name = tensor("op_22312_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_22312_end_0 = const()[name = tensor("op_22312_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_22312_end_mask_0 = const()[name = tensor("op_22312_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22312_cast_fp16 = slice_by_index(begin = var_22312_begin_0, end = var_22312_end_0, end_mask = var_22312_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22312_cast_fp16")]; + tensor var_22316_begin_0 = const()[name = tensor("op_22316_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_22316_end_0 = const()[name = tensor("op_22316_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_22316_end_mask_0 = const()[name = tensor("op_22316_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22316_cast_fp16 = slice_by_index(begin = var_22316_begin_0, end = var_22316_end_0, end_mask = var_22316_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22316_cast_fp16")]; + tensor var_22320_begin_0 = const()[name = tensor("op_22320_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_22320_end_0 = const()[name = tensor("op_22320_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_22320_end_mask_0 = const()[name = tensor("op_22320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22320_cast_fp16 = slice_by_index(begin = var_22320_begin_0, end = var_22320_end_0, end_mask = var_22320_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22320_cast_fp16")]; + tensor var_22324_begin_0 = const()[name = tensor("op_22324_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_22324_end_0 = const()[name = tensor("op_22324_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_22324_end_mask_0 = const()[name = tensor("op_22324_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22324_cast_fp16 = slice_by_index(begin = var_22324_begin_0, end = var_22324_end_0, end_mask = var_22324_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22324_cast_fp16")]; + tensor var_22328_begin_0 = const()[name = tensor("op_22328_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_22328_end_0 = const()[name = tensor("op_22328_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_22328_end_mask_0 = const()[name = tensor("op_22328_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22328_cast_fp16 = slice_by_index(begin = var_22328_begin_0, end = var_22328_end_0, end_mask = var_22328_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22328_cast_fp16")]; + tensor var_22332_begin_0 = const()[name = tensor("op_22332_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_22332_end_0 = const()[name = tensor("op_22332_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_22332_end_mask_0 = const()[name = tensor("op_22332_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22332_cast_fp16 = slice_by_index(begin = var_22332_begin_0, end = var_22332_end_0, end_mask = var_22332_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22332_cast_fp16")]; + tensor var_22336_begin_0 = const()[name = tensor("op_22336_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_22336_end_0 = const()[name = tensor("op_22336_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_22336_end_mask_0 = const()[name = tensor("op_22336_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22336_cast_fp16 = slice_by_index(begin = var_22336_begin_0, end = var_22336_end_0, end_mask = var_22336_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22336_cast_fp16")]; + tensor var_22340_begin_0 = const()[name = tensor("op_22340_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_22340_end_0 = const()[name = tensor("op_22340_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_22340_end_mask_0 = const()[name = tensor("op_22340_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22340_cast_fp16 = slice_by_index(begin = var_22340_begin_0, end = var_22340_end_0, end_mask = var_22340_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22340_cast_fp16")]; + tensor var_22344_begin_0 = const()[name = tensor("op_22344_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_22344_end_0 = const()[name = tensor("op_22344_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_22344_end_mask_0 = const()[name = tensor("op_22344_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22344_cast_fp16 = slice_by_index(begin = var_22344_begin_0, end = var_22344_end_0, end_mask = var_22344_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22344_cast_fp16")]; + tensor var_22348_begin_0 = const()[name = tensor("op_22348_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_22348_end_0 = const()[name = tensor("op_22348_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_22348_end_mask_0 = const()[name = tensor("op_22348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22348_cast_fp16 = slice_by_index(begin = var_22348_begin_0, end = var_22348_end_0, end_mask = var_22348_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22348_cast_fp16")]; + tensor var_22352_begin_0 = const()[name = tensor("op_22352_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_22352_end_0 = const()[name = tensor("op_22352_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_22352_end_mask_0 = const()[name = tensor("op_22352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22352_cast_fp16 = slice_by_index(begin = var_22352_begin_0, end = var_22352_end_0, end_mask = var_22352_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22352_cast_fp16")]; + tensor var_22356_begin_0 = const()[name = tensor("op_22356_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_22356_end_0 = const()[name = tensor("op_22356_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_22356_end_mask_0 = const()[name = tensor("op_22356_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22356_cast_fp16 = slice_by_index(begin = var_22356_begin_0, end = var_22356_end_0, end_mask = var_22356_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22356_cast_fp16")]; + tensor var_22360_begin_0 = const()[name = tensor("op_22360_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_22360_end_0 = const()[name = tensor("op_22360_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_22360_end_mask_0 = const()[name = tensor("op_22360_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_22360_cast_fp16 = slice_by_index(begin = var_22360_begin_0, end = var_22360_end_0, end_mask = var_22360_end_mask_0, x = value_27_cast_fp16)[name = tensor("op_22360_cast_fp16")]; + tensor var_22364_equation_0 = const()[name = tensor("op_22364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22364_cast_fp16 = einsum(equation = var_22364_equation_0, values = (var_22206_cast_fp16, var_21648_cast_fp16))[name = tensor("op_22364_cast_fp16")]; + tensor var_22365_to_fp16 = const()[name = tensor("op_22365_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2081_cast_fp16 = mul(x = var_22364_cast_fp16, y = var_22365_to_fp16)[name = tensor("aw_chunk_2081_cast_fp16")]; + tensor var_22368_equation_0 = const()[name = tensor("op_22368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22368_cast_fp16 = einsum(equation = var_22368_equation_0, values = (var_22206_cast_fp16, var_21655_cast_fp16))[name = tensor("op_22368_cast_fp16")]; + tensor var_22369_to_fp16 = const()[name = tensor("op_22369_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2083_cast_fp16 = mul(x = var_22368_cast_fp16, y = var_22369_to_fp16)[name = tensor("aw_chunk_2083_cast_fp16")]; + tensor var_22372_equation_0 = const()[name = tensor("op_22372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22372_cast_fp16 = einsum(equation = var_22372_equation_0, values = (var_22206_cast_fp16, var_21662_cast_fp16))[name = tensor("op_22372_cast_fp16")]; + tensor var_22373_to_fp16 = const()[name = tensor("op_22373_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2085_cast_fp16 = mul(x = var_22372_cast_fp16, y = var_22373_to_fp16)[name = tensor("aw_chunk_2085_cast_fp16")]; + tensor var_22376_equation_0 = const()[name = tensor("op_22376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22376_cast_fp16 = einsum(equation = var_22376_equation_0, values = (var_22206_cast_fp16, var_21669_cast_fp16))[name = tensor("op_22376_cast_fp16")]; + tensor var_22377_to_fp16 = const()[name = tensor("op_22377_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2087_cast_fp16 = mul(x = var_22376_cast_fp16, y = var_22377_to_fp16)[name = tensor("aw_chunk_2087_cast_fp16")]; + tensor var_22380_equation_0 = const()[name = tensor("op_22380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22380_cast_fp16 = einsum(equation = var_22380_equation_0, values = (var_22210_cast_fp16, var_21676_cast_fp16))[name = tensor("op_22380_cast_fp16")]; + tensor var_22381_to_fp16 = const()[name = tensor("op_22381_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2089_cast_fp16 = mul(x = var_22380_cast_fp16, y = var_22381_to_fp16)[name = tensor("aw_chunk_2089_cast_fp16")]; + tensor var_22384_equation_0 = const()[name = tensor("op_22384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22384_cast_fp16 = einsum(equation = var_22384_equation_0, values = (var_22210_cast_fp16, var_21683_cast_fp16))[name = tensor("op_22384_cast_fp16")]; + tensor var_22385_to_fp16 = const()[name = tensor("op_22385_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2091_cast_fp16 = mul(x = var_22384_cast_fp16, y = var_22385_to_fp16)[name = tensor("aw_chunk_2091_cast_fp16")]; + tensor var_22388_equation_0 = const()[name = tensor("op_22388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22388_cast_fp16 = einsum(equation = var_22388_equation_0, values = (var_22210_cast_fp16, var_21690_cast_fp16))[name = tensor("op_22388_cast_fp16")]; + tensor var_22389_to_fp16 = const()[name = tensor("op_22389_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2093_cast_fp16 = mul(x = var_22388_cast_fp16, y = var_22389_to_fp16)[name = tensor("aw_chunk_2093_cast_fp16")]; + tensor var_22392_equation_0 = const()[name = tensor("op_22392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22392_cast_fp16 = einsum(equation = var_22392_equation_0, values = (var_22210_cast_fp16, var_21697_cast_fp16))[name = tensor("op_22392_cast_fp16")]; + tensor var_22393_to_fp16 = const()[name = tensor("op_22393_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2095_cast_fp16 = mul(x = var_22392_cast_fp16, y = var_22393_to_fp16)[name = tensor("aw_chunk_2095_cast_fp16")]; + tensor var_22396_equation_0 = const()[name = tensor("op_22396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22396_cast_fp16 = einsum(equation = var_22396_equation_0, values = (var_22214_cast_fp16, var_21704_cast_fp16))[name = tensor("op_22396_cast_fp16")]; + tensor var_22397_to_fp16 = const()[name = tensor("op_22397_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2097_cast_fp16 = mul(x = var_22396_cast_fp16, y = var_22397_to_fp16)[name = tensor("aw_chunk_2097_cast_fp16")]; + tensor var_22400_equation_0 = const()[name = tensor("op_22400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22400_cast_fp16 = einsum(equation = var_22400_equation_0, values = (var_22214_cast_fp16, var_21711_cast_fp16))[name = tensor("op_22400_cast_fp16")]; + tensor var_22401_to_fp16 = const()[name = tensor("op_22401_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2099_cast_fp16 = mul(x = var_22400_cast_fp16, y = var_22401_to_fp16)[name = tensor("aw_chunk_2099_cast_fp16")]; + tensor var_22404_equation_0 = const()[name = tensor("op_22404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22404_cast_fp16 = einsum(equation = var_22404_equation_0, values = (var_22214_cast_fp16, var_21718_cast_fp16))[name = tensor("op_22404_cast_fp16")]; + tensor var_22405_to_fp16 = const()[name = tensor("op_22405_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2101_cast_fp16 = mul(x = var_22404_cast_fp16, y = var_22405_to_fp16)[name = tensor("aw_chunk_2101_cast_fp16")]; + tensor var_22408_equation_0 = const()[name = tensor("op_22408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22408_cast_fp16 = einsum(equation = var_22408_equation_0, values = (var_22214_cast_fp16, var_21725_cast_fp16))[name = tensor("op_22408_cast_fp16")]; + tensor var_22409_to_fp16 = const()[name = tensor("op_22409_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2103_cast_fp16 = mul(x = var_22408_cast_fp16, y = var_22409_to_fp16)[name = tensor("aw_chunk_2103_cast_fp16")]; + tensor var_22412_equation_0 = const()[name = tensor("op_22412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22412_cast_fp16 = einsum(equation = var_22412_equation_0, values = (var_22218_cast_fp16, var_21732_cast_fp16))[name = tensor("op_22412_cast_fp16")]; + tensor var_22413_to_fp16 = const()[name = tensor("op_22413_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2105_cast_fp16 = mul(x = var_22412_cast_fp16, y = var_22413_to_fp16)[name = tensor("aw_chunk_2105_cast_fp16")]; + tensor var_22416_equation_0 = const()[name = tensor("op_22416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22416_cast_fp16 = einsum(equation = var_22416_equation_0, values = (var_22218_cast_fp16, var_21739_cast_fp16))[name = tensor("op_22416_cast_fp16")]; + tensor var_22417_to_fp16 = const()[name = tensor("op_22417_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2107_cast_fp16 = mul(x = var_22416_cast_fp16, y = var_22417_to_fp16)[name = tensor("aw_chunk_2107_cast_fp16")]; + tensor var_22420_equation_0 = const()[name = tensor("op_22420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22420_cast_fp16 = einsum(equation = var_22420_equation_0, values = (var_22218_cast_fp16, var_21746_cast_fp16))[name = tensor("op_22420_cast_fp16")]; + tensor var_22421_to_fp16 = const()[name = tensor("op_22421_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2109_cast_fp16 = mul(x = var_22420_cast_fp16, y = var_22421_to_fp16)[name = tensor("aw_chunk_2109_cast_fp16")]; + tensor var_22424_equation_0 = const()[name = tensor("op_22424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22424_cast_fp16 = einsum(equation = var_22424_equation_0, values = (var_22218_cast_fp16, var_21753_cast_fp16))[name = tensor("op_22424_cast_fp16")]; + tensor var_22425_to_fp16 = const()[name = tensor("op_22425_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2111_cast_fp16 = mul(x = var_22424_cast_fp16, y = var_22425_to_fp16)[name = tensor("aw_chunk_2111_cast_fp16")]; + tensor var_22428_equation_0 = const()[name = tensor("op_22428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22428_cast_fp16 = einsum(equation = var_22428_equation_0, values = (var_22222_cast_fp16, var_21760_cast_fp16))[name = tensor("op_22428_cast_fp16")]; + tensor var_22429_to_fp16 = const()[name = tensor("op_22429_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2113_cast_fp16 = mul(x = var_22428_cast_fp16, y = var_22429_to_fp16)[name = tensor("aw_chunk_2113_cast_fp16")]; + tensor var_22432_equation_0 = const()[name = tensor("op_22432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22432_cast_fp16 = einsum(equation = var_22432_equation_0, values = (var_22222_cast_fp16, var_21767_cast_fp16))[name = tensor("op_22432_cast_fp16")]; + tensor var_22433_to_fp16 = const()[name = tensor("op_22433_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2115_cast_fp16 = mul(x = var_22432_cast_fp16, y = var_22433_to_fp16)[name = tensor("aw_chunk_2115_cast_fp16")]; + tensor var_22436_equation_0 = const()[name = tensor("op_22436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22436_cast_fp16 = einsum(equation = var_22436_equation_0, values = (var_22222_cast_fp16, var_21774_cast_fp16))[name = tensor("op_22436_cast_fp16")]; + tensor var_22437_to_fp16 = const()[name = tensor("op_22437_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2117_cast_fp16 = mul(x = var_22436_cast_fp16, y = var_22437_to_fp16)[name = tensor("aw_chunk_2117_cast_fp16")]; + tensor var_22440_equation_0 = const()[name = tensor("op_22440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22440_cast_fp16 = einsum(equation = var_22440_equation_0, values = (var_22222_cast_fp16, var_21781_cast_fp16))[name = tensor("op_22440_cast_fp16")]; + tensor var_22441_to_fp16 = const()[name = tensor("op_22441_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2119_cast_fp16 = mul(x = var_22440_cast_fp16, y = var_22441_to_fp16)[name = tensor("aw_chunk_2119_cast_fp16")]; + tensor var_22444_equation_0 = const()[name = tensor("op_22444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22444_cast_fp16 = einsum(equation = var_22444_equation_0, values = (var_22226_cast_fp16, var_21788_cast_fp16))[name = tensor("op_22444_cast_fp16")]; + tensor var_22445_to_fp16 = const()[name = tensor("op_22445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2121_cast_fp16 = mul(x = var_22444_cast_fp16, y = var_22445_to_fp16)[name = tensor("aw_chunk_2121_cast_fp16")]; + tensor var_22448_equation_0 = const()[name = tensor("op_22448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22448_cast_fp16 = einsum(equation = var_22448_equation_0, values = (var_22226_cast_fp16, var_21795_cast_fp16))[name = tensor("op_22448_cast_fp16")]; + tensor var_22449_to_fp16 = const()[name = tensor("op_22449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2123_cast_fp16 = mul(x = var_22448_cast_fp16, y = var_22449_to_fp16)[name = tensor("aw_chunk_2123_cast_fp16")]; + tensor var_22452_equation_0 = const()[name = tensor("op_22452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22452_cast_fp16 = einsum(equation = var_22452_equation_0, values = (var_22226_cast_fp16, var_21802_cast_fp16))[name = tensor("op_22452_cast_fp16")]; + tensor var_22453_to_fp16 = const()[name = tensor("op_22453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2125_cast_fp16 = mul(x = var_22452_cast_fp16, y = var_22453_to_fp16)[name = tensor("aw_chunk_2125_cast_fp16")]; + tensor var_22456_equation_0 = const()[name = tensor("op_22456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22456_cast_fp16 = einsum(equation = var_22456_equation_0, values = (var_22226_cast_fp16, var_21809_cast_fp16))[name = tensor("op_22456_cast_fp16")]; + tensor var_22457_to_fp16 = const()[name = tensor("op_22457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2127_cast_fp16 = mul(x = var_22456_cast_fp16, y = var_22457_to_fp16)[name = tensor("aw_chunk_2127_cast_fp16")]; + tensor var_22460_equation_0 = const()[name = tensor("op_22460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22460_cast_fp16 = einsum(equation = var_22460_equation_0, values = (var_22230_cast_fp16, var_21816_cast_fp16))[name = tensor("op_22460_cast_fp16")]; + tensor var_22461_to_fp16 = const()[name = tensor("op_22461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2129_cast_fp16 = mul(x = var_22460_cast_fp16, y = var_22461_to_fp16)[name = tensor("aw_chunk_2129_cast_fp16")]; + tensor var_22464_equation_0 = const()[name = tensor("op_22464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22464_cast_fp16 = einsum(equation = var_22464_equation_0, values = (var_22230_cast_fp16, var_21823_cast_fp16))[name = tensor("op_22464_cast_fp16")]; + tensor var_22465_to_fp16 = const()[name = tensor("op_22465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2131_cast_fp16 = mul(x = var_22464_cast_fp16, y = var_22465_to_fp16)[name = tensor("aw_chunk_2131_cast_fp16")]; + tensor var_22468_equation_0 = const()[name = tensor("op_22468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22468_cast_fp16 = einsum(equation = var_22468_equation_0, values = (var_22230_cast_fp16, var_21830_cast_fp16))[name = tensor("op_22468_cast_fp16")]; + tensor var_22469_to_fp16 = const()[name = tensor("op_22469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2133_cast_fp16 = mul(x = var_22468_cast_fp16, y = var_22469_to_fp16)[name = tensor("aw_chunk_2133_cast_fp16")]; + tensor var_22472_equation_0 = const()[name = tensor("op_22472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22472_cast_fp16 = einsum(equation = var_22472_equation_0, values = (var_22230_cast_fp16, var_21837_cast_fp16))[name = tensor("op_22472_cast_fp16")]; + tensor var_22473_to_fp16 = const()[name = tensor("op_22473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2135_cast_fp16 = mul(x = var_22472_cast_fp16, y = var_22473_to_fp16)[name = tensor("aw_chunk_2135_cast_fp16")]; + tensor var_22476_equation_0 = const()[name = tensor("op_22476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22476_cast_fp16 = einsum(equation = var_22476_equation_0, values = (var_22234_cast_fp16, var_21844_cast_fp16))[name = tensor("op_22476_cast_fp16")]; + tensor var_22477_to_fp16 = const()[name = tensor("op_22477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2137_cast_fp16 = mul(x = var_22476_cast_fp16, y = var_22477_to_fp16)[name = tensor("aw_chunk_2137_cast_fp16")]; + tensor var_22480_equation_0 = const()[name = tensor("op_22480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22480_cast_fp16 = einsum(equation = var_22480_equation_0, values = (var_22234_cast_fp16, var_21851_cast_fp16))[name = tensor("op_22480_cast_fp16")]; + tensor var_22481_to_fp16 = const()[name = tensor("op_22481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2139_cast_fp16 = mul(x = var_22480_cast_fp16, y = var_22481_to_fp16)[name = tensor("aw_chunk_2139_cast_fp16")]; + tensor var_22484_equation_0 = const()[name = tensor("op_22484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22484_cast_fp16 = einsum(equation = var_22484_equation_0, values = (var_22234_cast_fp16, var_21858_cast_fp16))[name = tensor("op_22484_cast_fp16")]; + tensor var_22485_to_fp16 = const()[name = tensor("op_22485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2141_cast_fp16 = mul(x = var_22484_cast_fp16, y = var_22485_to_fp16)[name = tensor("aw_chunk_2141_cast_fp16")]; + tensor var_22488_equation_0 = const()[name = tensor("op_22488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22488_cast_fp16 = einsum(equation = var_22488_equation_0, values = (var_22234_cast_fp16, var_21865_cast_fp16))[name = tensor("op_22488_cast_fp16")]; + tensor var_22489_to_fp16 = const()[name = tensor("op_22489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2143_cast_fp16 = mul(x = var_22488_cast_fp16, y = var_22489_to_fp16)[name = tensor("aw_chunk_2143_cast_fp16")]; + tensor var_22492_equation_0 = const()[name = tensor("op_22492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22492_cast_fp16 = einsum(equation = var_22492_equation_0, values = (var_22238_cast_fp16, var_21872_cast_fp16))[name = tensor("op_22492_cast_fp16")]; + tensor var_22493_to_fp16 = const()[name = tensor("op_22493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2145_cast_fp16 = mul(x = var_22492_cast_fp16, y = var_22493_to_fp16)[name = tensor("aw_chunk_2145_cast_fp16")]; + tensor var_22496_equation_0 = const()[name = tensor("op_22496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22496_cast_fp16 = einsum(equation = var_22496_equation_0, values = (var_22238_cast_fp16, var_21879_cast_fp16))[name = tensor("op_22496_cast_fp16")]; + tensor var_22497_to_fp16 = const()[name = tensor("op_22497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2147_cast_fp16 = mul(x = var_22496_cast_fp16, y = var_22497_to_fp16)[name = tensor("aw_chunk_2147_cast_fp16")]; + tensor var_22500_equation_0 = const()[name = tensor("op_22500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22500_cast_fp16 = einsum(equation = var_22500_equation_0, values = (var_22238_cast_fp16, var_21886_cast_fp16))[name = tensor("op_22500_cast_fp16")]; + tensor var_22501_to_fp16 = const()[name = tensor("op_22501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2149_cast_fp16 = mul(x = var_22500_cast_fp16, y = var_22501_to_fp16)[name = tensor("aw_chunk_2149_cast_fp16")]; + tensor var_22504_equation_0 = const()[name = tensor("op_22504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22504_cast_fp16 = einsum(equation = var_22504_equation_0, values = (var_22238_cast_fp16, var_21893_cast_fp16))[name = tensor("op_22504_cast_fp16")]; + tensor var_22505_to_fp16 = const()[name = tensor("op_22505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2151_cast_fp16 = mul(x = var_22504_cast_fp16, y = var_22505_to_fp16)[name = tensor("aw_chunk_2151_cast_fp16")]; + tensor var_22508_equation_0 = const()[name = tensor("op_22508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22508_cast_fp16 = einsum(equation = var_22508_equation_0, values = (var_22242_cast_fp16, var_21900_cast_fp16))[name = tensor("op_22508_cast_fp16")]; + tensor var_22509_to_fp16 = const()[name = tensor("op_22509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2153_cast_fp16 = mul(x = var_22508_cast_fp16, y = var_22509_to_fp16)[name = tensor("aw_chunk_2153_cast_fp16")]; + tensor var_22512_equation_0 = const()[name = tensor("op_22512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22512_cast_fp16 = einsum(equation = var_22512_equation_0, values = (var_22242_cast_fp16, var_21907_cast_fp16))[name = tensor("op_22512_cast_fp16")]; + tensor var_22513_to_fp16 = const()[name = tensor("op_22513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2155_cast_fp16 = mul(x = var_22512_cast_fp16, y = var_22513_to_fp16)[name = tensor("aw_chunk_2155_cast_fp16")]; + tensor var_22516_equation_0 = const()[name = tensor("op_22516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22516_cast_fp16 = einsum(equation = var_22516_equation_0, values = (var_22242_cast_fp16, var_21914_cast_fp16))[name = tensor("op_22516_cast_fp16")]; + tensor var_22517_to_fp16 = const()[name = tensor("op_22517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2157_cast_fp16 = mul(x = var_22516_cast_fp16, y = var_22517_to_fp16)[name = tensor("aw_chunk_2157_cast_fp16")]; + tensor var_22520_equation_0 = const()[name = tensor("op_22520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22520_cast_fp16 = einsum(equation = var_22520_equation_0, values = (var_22242_cast_fp16, var_21921_cast_fp16))[name = tensor("op_22520_cast_fp16")]; + tensor var_22521_to_fp16 = const()[name = tensor("op_22521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2159_cast_fp16 = mul(x = var_22520_cast_fp16, y = var_22521_to_fp16)[name = tensor("aw_chunk_2159_cast_fp16")]; + tensor var_22524_equation_0 = const()[name = tensor("op_22524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22524_cast_fp16 = einsum(equation = var_22524_equation_0, values = (var_22246_cast_fp16, var_21928_cast_fp16))[name = tensor("op_22524_cast_fp16")]; + tensor var_22525_to_fp16 = const()[name = tensor("op_22525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2161_cast_fp16 = mul(x = var_22524_cast_fp16, y = var_22525_to_fp16)[name = tensor("aw_chunk_2161_cast_fp16")]; + tensor var_22528_equation_0 = const()[name = tensor("op_22528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22528_cast_fp16 = einsum(equation = var_22528_equation_0, values = (var_22246_cast_fp16, var_21935_cast_fp16))[name = tensor("op_22528_cast_fp16")]; + tensor var_22529_to_fp16 = const()[name = tensor("op_22529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2163_cast_fp16 = mul(x = var_22528_cast_fp16, y = var_22529_to_fp16)[name = tensor("aw_chunk_2163_cast_fp16")]; + tensor var_22532_equation_0 = const()[name = tensor("op_22532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22532_cast_fp16 = einsum(equation = var_22532_equation_0, values = (var_22246_cast_fp16, var_21942_cast_fp16))[name = tensor("op_22532_cast_fp16")]; + tensor var_22533_to_fp16 = const()[name = tensor("op_22533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2165_cast_fp16 = mul(x = var_22532_cast_fp16, y = var_22533_to_fp16)[name = tensor("aw_chunk_2165_cast_fp16")]; + tensor var_22536_equation_0 = const()[name = tensor("op_22536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22536_cast_fp16 = einsum(equation = var_22536_equation_0, values = (var_22246_cast_fp16, var_21949_cast_fp16))[name = tensor("op_22536_cast_fp16")]; + tensor var_22537_to_fp16 = const()[name = tensor("op_22537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2167_cast_fp16 = mul(x = var_22536_cast_fp16, y = var_22537_to_fp16)[name = tensor("aw_chunk_2167_cast_fp16")]; + tensor var_22540_equation_0 = const()[name = tensor("op_22540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22540_cast_fp16 = einsum(equation = var_22540_equation_0, values = (var_22250_cast_fp16, var_21956_cast_fp16))[name = tensor("op_22540_cast_fp16")]; + tensor var_22541_to_fp16 = const()[name = tensor("op_22541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2169_cast_fp16 = mul(x = var_22540_cast_fp16, y = var_22541_to_fp16)[name = tensor("aw_chunk_2169_cast_fp16")]; + tensor var_22544_equation_0 = const()[name = tensor("op_22544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22544_cast_fp16 = einsum(equation = var_22544_equation_0, values = (var_22250_cast_fp16, var_21963_cast_fp16))[name = tensor("op_22544_cast_fp16")]; + tensor var_22545_to_fp16 = const()[name = tensor("op_22545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2171_cast_fp16 = mul(x = var_22544_cast_fp16, y = var_22545_to_fp16)[name = tensor("aw_chunk_2171_cast_fp16")]; + tensor var_22548_equation_0 = const()[name = tensor("op_22548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22548_cast_fp16 = einsum(equation = var_22548_equation_0, values = (var_22250_cast_fp16, var_21970_cast_fp16))[name = tensor("op_22548_cast_fp16")]; + tensor var_22549_to_fp16 = const()[name = tensor("op_22549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2173_cast_fp16 = mul(x = var_22548_cast_fp16, y = var_22549_to_fp16)[name = tensor("aw_chunk_2173_cast_fp16")]; + tensor var_22552_equation_0 = const()[name = tensor("op_22552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22552_cast_fp16 = einsum(equation = var_22552_equation_0, values = (var_22250_cast_fp16, var_21977_cast_fp16))[name = tensor("op_22552_cast_fp16")]; + tensor var_22553_to_fp16 = const()[name = tensor("op_22553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2175_cast_fp16 = mul(x = var_22552_cast_fp16, y = var_22553_to_fp16)[name = tensor("aw_chunk_2175_cast_fp16")]; + tensor var_22556_equation_0 = const()[name = tensor("op_22556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22556_cast_fp16 = einsum(equation = var_22556_equation_0, values = (var_22254_cast_fp16, var_21984_cast_fp16))[name = tensor("op_22556_cast_fp16")]; + tensor var_22557_to_fp16 = const()[name = tensor("op_22557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2177_cast_fp16 = mul(x = var_22556_cast_fp16, y = var_22557_to_fp16)[name = tensor("aw_chunk_2177_cast_fp16")]; + tensor var_22560_equation_0 = const()[name = tensor("op_22560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22560_cast_fp16 = einsum(equation = var_22560_equation_0, values = (var_22254_cast_fp16, var_21991_cast_fp16))[name = tensor("op_22560_cast_fp16")]; + tensor var_22561_to_fp16 = const()[name = tensor("op_22561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2179_cast_fp16 = mul(x = var_22560_cast_fp16, y = var_22561_to_fp16)[name = tensor("aw_chunk_2179_cast_fp16")]; + tensor var_22564_equation_0 = const()[name = tensor("op_22564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22564_cast_fp16 = einsum(equation = var_22564_equation_0, values = (var_22254_cast_fp16, var_21998_cast_fp16))[name = tensor("op_22564_cast_fp16")]; + tensor var_22565_to_fp16 = const()[name = tensor("op_22565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2181_cast_fp16 = mul(x = var_22564_cast_fp16, y = var_22565_to_fp16)[name = tensor("aw_chunk_2181_cast_fp16")]; + tensor var_22568_equation_0 = const()[name = tensor("op_22568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22568_cast_fp16 = einsum(equation = var_22568_equation_0, values = (var_22254_cast_fp16, var_22005_cast_fp16))[name = tensor("op_22568_cast_fp16")]; + tensor var_22569_to_fp16 = const()[name = tensor("op_22569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2183_cast_fp16 = mul(x = var_22568_cast_fp16, y = var_22569_to_fp16)[name = tensor("aw_chunk_2183_cast_fp16")]; + tensor var_22572_equation_0 = const()[name = tensor("op_22572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22572_cast_fp16 = einsum(equation = var_22572_equation_0, values = (var_22258_cast_fp16, var_22012_cast_fp16))[name = tensor("op_22572_cast_fp16")]; + tensor var_22573_to_fp16 = const()[name = tensor("op_22573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2185_cast_fp16 = mul(x = var_22572_cast_fp16, y = var_22573_to_fp16)[name = tensor("aw_chunk_2185_cast_fp16")]; + tensor var_22576_equation_0 = const()[name = tensor("op_22576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22576_cast_fp16 = einsum(equation = var_22576_equation_0, values = (var_22258_cast_fp16, var_22019_cast_fp16))[name = tensor("op_22576_cast_fp16")]; + tensor var_22577_to_fp16 = const()[name = tensor("op_22577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2187_cast_fp16 = mul(x = var_22576_cast_fp16, y = var_22577_to_fp16)[name = tensor("aw_chunk_2187_cast_fp16")]; + tensor var_22580_equation_0 = const()[name = tensor("op_22580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22580_cast_fp16 = einsum(equation = var_22580_equation_0, values = (var_22258_cast_fp16, var_22026_cast_fp16))[name = tensor("op_22580_cast_fp16")]; + tensor var_22581_to_fp16 = const()[name = tensor("op_22581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2189_cast_fp16 = mul(x = var_22580_cast_fp16, y = var_22581_to_fp16)[name = tensor("aw_chunk_2189_cast_fp16")]; + tensor var_22584_equation_0 = const()[name = tensor("op_22584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22584_cast_fp16 = einsum(equation = var_22584_equation_0, values = (var_22258_cast_fp16, var_22033_cast_fp16))[name = tensor("op_22584_cast_fp16")]; + tensor var_22585_to_fp16 = const()[name = tensor("op_22585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2191_cast_fp16 = mul(x = var_22584_cast_fp16, y = var_22585_to_fp16)[name = tensor("aw_chunk_2191_cast_fp16")]; + tensor var_22588_equation_0 = const()[name = tensor("op_22588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22588_cast_fp16 = einsum(equation = var_22588_equation_0, values = (var_22262_cast_fp16, var_22040_cast_fp16))[name = tensor("op_22588_cast_fp16")]; + tensor var_22589_to_fp16 = const()[name = tensor("op_22589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2193_cast_fp16 = mul(x = var_22588_cast_fp16, y = var_22589_to_fp16)[name = tensor("aw_chunk_2193_cast_fp16")]; + tensor var_22592_equation_0 = const()[name = tensor("op_22592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22592_cast_fp16 = einsum(equation = var_22592_equation_0, values = (var_22262_cast_fp16, var_22047_cast_fp16))[name = tensor("op_22592_cast_fp16")]; + tensor var_22593_to_fp16 = const()[name = tensor("op_22593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2195_cast_fp16 = mul(x = var_22592_cast_fp16, y = var_22593_to_fp16)[name = tensor("aw_chunk_2195_cast_fp16")]; + tensor var_22596_equation_0 = const()[name = tensor("op_22596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22596_cast_fp16 = einsum(equation = var_22596_equation_0, values = (var_22262_cast_fp16, var_22054_cast_fp16))[name = tensor("op_22596_cast_fp16")]; + tensor var_22597_to_fp16 = const()[name = tensor("op_22597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2197_cast_fp16 = mul(x = var_22596_cast_fp16, y = var_22597_to_fp16)[name = tensor("aw_chunk_2197_cast_fp16")]; + tensor var_22600_equation_0 = const()[name = tensor("op_22600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22600_cast_fp16 = einsum(equation = var_22600_equation_0, values = (var_22262_cast_fp16, var_22061_cast_fp16))[name = tensor("op_22600_cast_fp16")]; + tensor var_22601_to_fp16 = const()[name = tensor("op_22601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2199_cast_fp16 = mul(x = var_22600_cast_fp16, y = var_22601_to_fp16)[name = tensor("aw_chunk_2199_cast_fp16")]; + tensor var_22604_equation_0 = const()[name = tensor("op_22604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22604_cast_fp16 = einsum(equation = var_22604_equation_0, values = (var_22266_cast_fp16, var_22068_cast_fp16))[name = tensor("op_22604_cast_fp16")]; + tensor var_22605_to_fp16 = const()[name = tensor("op_22605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2201_cast_fp16 = mul(x = var_22604_cast_fp16, y = var_22605_to_fp16)[name = tensor("aw_chunk_2201_cast_fp16")]; + tensor var_22608_equation_0 = const()[name = tensor("op_22608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22608_cast_fp16 = einsum(equation = var_22608_equation_0, values = (var_22266_cast_fp16, var_22075_cast_fp16))[name = tensor("op_22608_cast_fp16")]; + tensor var_22609_to_fp16 = const()[name = tensor("op_22609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2203_cast_fp16 = mul(x = var_22608_cast_fp16, y = var_22609_to_fp16)[name = tensor("aw_chunk_2203_cast_fp16")]; + tensor var_22612_equation_0 = const()[name = tensor("op_22612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22612_cast_fp16 = einsum(equation = var_22612_equation_0, values = (var_22266_cast_fp16, var_22082_cast_fp16))[name = tensor("op_22612_cast_fp16")]; + tensor var_22613_to_fp16 = const()[name = tensor("op_22613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2205_cast_fp16 = mul(x = var_22612_cast_fp16, y = var_22613_to_fp16)[name = tensor("aw_chunk_2205_cast_fp16")]; + tensor var_22616_equation_0 = const()[name = tensor("op_22616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22616_cast_fp16 = einsum(equation = var_22616_equation_0, values = (var_22266_cast_fp16, var_22089_cast_fp16))[name = tensor("op_22616_cast_fp16")]; + tensor var_22617_to_fp16 = const()[name = tensor("op_22617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2207_cast_fp16 = mul(x = var_22616_cast_fp16, y = var_22617_to_fp16)[name = tensor("aw_chunk_2207_cast_fp16")]; + tensor var_22620_equation_0 = const()[name = tensor("op_22620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22620_cast_fp16 = einsum(equation = var_22620_equation_0, values = (var_22270_cast_fp16, var_22096_cast_fp16))[name = tensor("op_22620_cast_fp16")]; + tensor var_22621_to_fp16 = const()[name = tensor("op_22621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2209_cast_fp16 = mul(x = var_22620_cast_fp16, y = var_22621_to_fp16)[name = tensor("aw_chunk_2209_cast_fp16")]; + tensor var_22624_equation_0 = const()[name = tensor("op_22624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22624_cast_fp16 = einsum(equation = var_22624_equation_0, values = (var_22270_cast_fp16, var_22103_cast_fp16))[name = tensor("op_22624_cast_fp16")]; + tensor var_22625_to_fp16 = const()[name = tensor("op_22625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2211_cast_fp16 = mul(x = var_22624_cast_fp16, y = var_22625_to_fp16)[name = tensor("aw_chunk_2211_cast_fp16")]; + tensor var_22628_equation_0 = const()[name = tensor("op_22628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22628_cast_fp16 = einsum(equation = var_22628_equation_0, values = (var_22270_cast_fp16, var_22110_cast_fp16))[name = tensor("op_22628_cast_fp16")]; + tensor var_22629_to_fp16 = const()[name = tensor("op_22629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2213_cast_fp16 = mul(x = var_22628_cast_fp16, y = var_22629_to_fp16)[name = tensor("aw_chunk_2213_cast_fp16")]; + tensor var_22632_equation_0 = const()[name = tensor("op_22632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22632_cast_fp16 = einsum(equation = var_22632_equation_0, values = (var_22270_cast_fp16, var_22117_cast_fp16))[name = tensor("op_22632_cast_fp16")]; + tensor var_22633_to_fp16 = const()[name = tensor("op_22633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2215_cast_fp16 = mul(x = var_22632_cast_fp16, y = var_22633_to_fp16)[name = tensor("aw_chunk_2215_cast_fp16")]; + tensor var_22636_equation_0 = const()[name = tensor("op_22636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22636_cast_fp16 = einsum(equation = var_22636_equation_0, values = (var_22274_cast_fp16, var_22124_cast_fp16))[name = tensor("op_22636_cast_fp16")]; + tensor var_22637_to_fp16 = const()[name = tensor("op_22637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2217_cast_fp16 = mul(x = var_22636_cast_fp16, y = var_22637_to_fp16)[name = tensor("aw_chunk_2217_cast_fp16")]; + tensor var_22640_equation_0 = const()[name = tensor("op_22640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22640_cast_fp16 = einsum(equation = var_22640_equation_0, values = (var_22274_cast_fp16, var_22131_cast_fp16))[name = tensor("op_22640_cast_fp16")]; + tensor var_22641_to_fp16 = const()[name = tensor("op_22641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2219_cast_fp16 = mul(x = var_22640_cast_fp16, y = var_22641_to_fp16)[name = tensor("aw_chunk_2219_cast_fp16")]; + tensor var_22644_equation_0 = const()[name = tensor("op_22644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22644_cast_fp16 = einsum(equation = var_22644_equation_0, values = (var_22274_cast_fp16, var_22138_cast_fp16))[name = tensor("op_22644_cast_fp16")]; + tensor var_22645_to_fp16 = const()[name = tensor("op_22645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2221_cast_fp16 = mul(x = var_22644_cast_fp16, y = var_22645_to_fp16)[name = tensor("aw_chunk_2221_cast_fp16")]; + tensor var_22648_equation_0 = const()[name = tensor("op_22648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22648_cast_fp16 = einsum(equation = var_22648_equation_0, values = (var_22274_cast_fp16, var_22145_cast_fp16))[name = tensor("op_22648_cast_fp16")]; + tensor var_22649_to_fp16 = const()[name = tensor("op_22649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2223_cast_fp16 = mul(x = var_22648_cast_fp16, y = var_22649_to_fp16)[name = tensor("aw_chunk_2223_cast_fp16")]; + tensor var_22652_equation_0 = const()[name = tensor("op_22652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22652_cast_fp16 = einsum(equation = var_22652_equation_0, values = (var_22278_cast_fp16, var_22152_cast_fp16))[name = tensor("op_22652_cast_fp16")]; + tensor var_22653_to_fp16 = const()[name = tensor("op_22653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2225_cast_fp16 = mul(x = var_22652_cast_fp16, y = var_22653_to_fp16)[name = tensor("aw_chunk_2225_cast_fp16")]; + tensor var_22656_equation_0 = const()[name = tensor("op_22656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22656_cast_fp16 = einsum(equation = var_22656_equation_0, values = (var_22278_cast_fp16, var_22159_cast_fp16))[name = tensor("op_22656_cast_fp16")]; + tensor var_22657_to_fp16 = const()[name = tensor("op_22657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2227_cast_fp16 = mul(x = var_22656_cast_fp16, y = var_22657_to_fp16)[name = tensor("aw_chunk_2227_cast_fp16")]; + tensor var_22660_equation_0 = const()[name = tensor("op_22660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22660_cast_fp16 = einsum(equation = var_22660_equation_0, values = (var_22278_cast_fp16, var_22166_cast_fp16))[name = tensor("op_22660_cast_fp16")]; + tensor var_22661_to_fp16 = const()[name = tensor("op_22661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2229_cast_fp16 = mul(x = var_22660_cast_fp16, y = var_22661_to_fp16)[name = tensor("aw_chunk_2229_cast_fp16")]; + tensor var_22664_equation_0 = const()[name = tensor("op_22664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22664_cast_fp16 = einsum(equation = var_22664_equation_0, values = (var_22278_cast_fp16, var_22173_cast_fp16))[name = tensor("op_22664_cast_fp16")]; + tensor var_22665_to_fp16 = const()[name = tensor("op_22665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2231_cast_fp16 = mul(x = var_22664_cast_fp16, y = var_22665_to_fp16)[name = tensor("aw_chunk_2231_cast_fp16")]; + tensor var_22668_equation_0 = const()[name = tensor("op_22668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22668_cast_fp16 = einsum(equation = var_22668_equation_0, values = (var_22282_cast_fp16, var_22180_cast_fp16))[name = tensor("op_22668_cast_fp16")]; + tensor var_22669_to_fp16 = const()[name = tensor("op_22669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2233_cast_fp16 = mul(x = var_22668_cast_fp16, y = var_22669_to_fp16)[name = tensor("aw_chunk_2233_cast_fp16")]; + tensor var_22672_equation_0 = const()[name = tensor("op_22672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22672_cast_fp16 = einsum(equation = var_22672_equation_0, values = (var_22282_cast_fp16, var_22187_cast_fp16))[name = tensor("op_22672_cast_fp16")]; + tensor var_22673_to_fp16 = const()[name = tensor("op_22673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2235_cast_fp16 = mul(x = var_22672_cast_fp16, y = var_22673_to_fp16)[name = tensor("aw_chunk_2235_cast_fp16")]; + tensor var_22676_equation_0 = const()[name = tensor("op_22676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22676_cast_fp16 = einsum(equation = var_22676_equation_0, values = (var_22282_cast_fp16, var_22194_cast_fp16))[name = tensor("op_22676_cast_fp16")]; + tensor var_22677_to_fp16 = const()[name = tensor("op_22677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2237_cast_fp16 = mul(x = var_22676_cast_fp16, y = var_22677_to_fp16)[name = tensor("aw_chunk_2237_cast_fp16")]; + tensor var_22680_equation_0 = const()[name = tensor("op_22680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_22680_cast_fp16 = einsum(equation = var_22680_equation_0, values = (var_22282_cast_fp16, var_22201_cast_fp16))[name = tensor("op_22680_cast_fp16")]; + tensor var_22681_to_fp16 = const()[name = tensor("op_22681_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2239_cast_fp16 = mul(x = var_22680_cast_fp16, y = var_22681_to_fp16)[name = tensor("aw_chunk_2239_cast_fp16")]; + tensor var_22683_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2081_cast_fp16)[name = tensor("op_22683_cast_fp16")]; + tensor var_22684_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2083_cast_fp16)[name = tensor("op_22684_cast_fp16")]; + tensor var_22685_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2085_cast_fp16)[name = tensor("op_22685_cast_fp16")]; + tensor var_22686_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2087_cast_fp16)[name = tensor("op_22686_cast_fp16")]; + tensor var_22687_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2089_cast_fp16)[name = tensor("op_22687_cast_fp16")]; + tensor var_22688_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2091_cast_fp16)[name = tensor("op_22688_cast_fp16")]; + tensor var_22689_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2093_cast_fp16)[name = tensor("op_22689_cast_fp16")]; + tensor var_22690_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2095_cast_fp16)[name = tensor("op_22690_cast_fp16")]; + tensor var_22691_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2097_cast_fp16)[name = tensor("op_22691_cast_fp16")]; + tensor var_22692_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2099_cast_fp16)[name = tensor("op_22692_cast_fp16")]; + tensor var_22693_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2101_cast_fp16)[name = tensor("op_22693_cast_fp16")]; + tensor var_22694_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2103_cast_fp16)[name = tensor("op_22694_cast_fp16")]; + tensor var_22695_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2105_cast_fp16)[name = tensor("op_22695_cast_fp16")]; + tensor var_22696_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2107_cast_fp16)[name = tensor("op_22696_cast_fp16")]; + tensor var_22697_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2109_cast_fp16)[name = tensor("op_22697_cast_fp16")]; + tensor var_22698_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2111_cast_fp16)[name = tensor("op_22698_cast_fp16")]; + tensor var_22699_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2113_cast_fp16)[name = tensor("op_22699_cast_fp16")]; + tensor var_22700_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2115_cast_fp16)[name = tensor("op_22700_cast_fp16")]; + tensor var_22701_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2117_cast_fp16)[name = tensor("op_22701_cast_fp16")]; + tensor var_22702_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2119_cast_fp16)[name = tensor("op_22702_cast_fp16")]; + tensor var_22703_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2121_cast_fp16)[name = tensor("op_22703_cast_fp16")]; + tensor var_22704_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2123_cast_fp16)[name = tensor("op_22704_cast_fp16")]; + tensor var_22705_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2125_cast_fp16)[name = tensor("op_22705_cast_fp16")]; + tensor var_22706_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2127_cast_fp16)[name = tensor("op_22706_cast_fp16")]; + tensor var_22707_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2129_cast_fp16)[name = tensor("op_22707_cast_fp16")]; + tensor var_22708_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2131_cast_fp16)[name = tensor("op_22708_cast_fp16")]; + tensor var_22709_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2133_cast_fp16)[name = tensor("op_22709_cast_fp16")]; + tensor var_22710_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2135_cast_fp16)[name = tensor("op_22710_cast_fp16")]; + tensor var_22711_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2137_cast_fp16)[name = tensor("op_22711_cast_fp16")]; + tensor var_22712_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2139_cast_fp16)[name = tensor("op_22712_cast_fp16")]; + tensor var_22713_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2141_cast_fp16)[name = tensor("op_22713_cast_fp16")]; + tensor var_22714_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2143_cast_fp16)[name = tensor("op_22714_cast_fp16")]; + tensor var_22715_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2145_cast_fp16)[name = tensor("op_22715_cast_fp16")]; + tensor var_22716_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2147_cast_fp16)[name = tensor("op_22716_cast_fp16")]; + tensor var_22717_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2149_cast_fp16)[name = tensor("op_22717_cast_fp16")]; + tensor var_22718_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2151_cast_fp16)[name = tensor("op_22718_cast_fp16")]; + tensor var_22719_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2153_cast_fp16)[name = tensor("op_22719_cast_fp16")]; + tensor var_22720_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2155_cast_fp16)[name = tensor("op_22720_cast_fp16")]; + tensor var_22721_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2157_cast_fp16)[name = tensor("op_22721_cast_fp16")]; + tensor var_22722_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2159_cast_fp16)[name = tensor("op_22722_cast_fp16")]; + tensor var_22723_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2161_cast_fp16)[name = tensor("op_22723_cast_fp16")]; + tensor var_22724_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2163_cast_fp16)[name = tensor("op_22724_cast_fp16")]; + tensor var_22725_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2165_cast_fp16)[name = tensor("op_22725_cast_fp16")]; + tensor var_22726_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2167_cast_fp16)[name = tensor("op_22726_cast_fp16")]; + tensor var_22727_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2169_cast_fp16)[name = tensor("op_22727_cast_fp16")]; + tensor var_22728_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2171_cast_fp16)[name = tensor("op_22728_cast_fp16")]; + tensor var_22729_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2173_cast_fp16)[name = tensor("op_22729_cast_fp16")]; + tensor var_22730_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2175_cast_fp16)[name = tensor("op_22730_cast_fp16")]; + tensor var_22731_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2177_cast_fp16)[name = tensor("op_22731_cast_fp16")]; + tensor var_22732_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2179_cast_fp16)[name = tensor("op_22732_cast_fp16")]; + tensor var_22733_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2181_cast_fp16)[name = tensor("op_22733_cast_fp16")]; + tensor var_22734_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2183_cast_fp16)[name = tensor("op_22734_cast_fp16")]; + tensor var_22735_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2185_cast_fp16)[name = tensor("op_22735_cast_fp16")]; + tensor var_22736_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2187_cast_fp16)[name = tensor("op_22736_cast_fp16")]; + tensor var_22737_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2189_cast_fp16)[name = tensor("op_22737_cast_fp16")]; + tensor var_22738_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2191_cast_fp16)[name = tensor("op_22738_cast_fp16")]; + tensor var_22739_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2193_cast_fp16)[name = tensor("op_22739_cast_fp16")]; + tensor var_22740_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2195_cast_fp16)[name = tensor("op_22740_cast_fp16")]; + tensor var_22741_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2197_cast_fp16)[name = tensor("op_22741_cast_fp16")]; + tensor var_22742_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2199_cast_fp16)[name = tensor("op_22742_cast_fp16")]; + tensor var_22743_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2201_cast_fp16)[name = tensor("op_22743_cast_fp16")]; + tensor var_22744_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2203_cast_fp16)[name = tensor("op_22744_cast_fp16")]; + tensor var_22745_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2205_cast_fp16)[name = tensor("op_22745_cast_fp16")]; + tensor var_22746_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2207_cast_fp16)[name = tensor("op_22746_cast_fp16")]; + tensor var_22747_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2209_cast_fp16)[name = tensor("op_22747_cast_fp16")]; + tensor var_22748_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2211_cast_fp16)[name = tensor("op_22748_cast_fp16")]; + tensor var_22749_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2213_cast_fp16)[name = tensor("op_22749_cast_fp16")]; + tensor var_22750_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2215_cast_fp16)[name = tensor("op_22750_cast_fp16")]; + tensor var_22751_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2217_cast_fp16)[name = tensor("op_22751_cast_fp16")]; + tensor var_22752_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2219_cast_fp16)[name = tensor("op_22752_cast_fp16")]; + tensor var_22753_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2221_cast_fp16)[name = tensor("op_22753_cast_fp16")]; + tensor var_22754_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2223_cast_fp16)[name = tensor("op_22754_cast_fp16")]; + tensor var_22755_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2225_cast_fp16)[name = tensor("op_22755_cast_fp16")]; + tensor var_22756_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2227_cast_fp16)[name = tensor("op_22756_cast_fp16")]; + tensor var_22757_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2229_cast_fp16)[name = tensor("op_22757_cast_fp16")]; + tensor var_22758_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2231_cast_fp16)[name = tensor("op_22758_cast_fp16")]; + tensor var_22759_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2233_cast_fp16)[name = tensor("op_22759_cast_fp16")]; + tensor var_22760_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2235_cast_fp16)[name = tensor("op_22760_cast_fp16")]; + tensor var_22761_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2237_cast_fp16)[name = tensor("op_22761_cast_fp16")]; + tensor var_22762_cast_fp16 = softmax(axis = var_21454, x = aw_chunk_2239_cast_fp16)[name = tensor("op_22762_cast_fp16")]; + tensor var_22764_equation_0 = const()[name = tensor("op_22764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22764_cast_fp16 = einsum(equation = var_22764_equation_0, values = (var_22284_cast_fp16, var_22683_cast_fp16))[name = tensor("op_22764_cast_fp16")]; + tensor var_22766_equation_0 = const()[name = tensor("op_22766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22766_cast_fp16 = einsum(equation = var_22766_equation_0, values = (var_22284_cast_fp16, var_22684_cast_fp16))[name = tensor("op_22766_cast_fp16")]; + tensor var_22768_equation_0 = const()[name = tensor("op_22768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22768_cast_fp16 = einsum(equation = var_22768_equation_0, values = (var_22284_cast_fp16, var_22685_cast_fp16))[name = tensor("op_22768_cast_fp16")]; + tensor var_22770_equation_0 = const()[name = tensor("op_22770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22770_cast_fp16 = einsum(equation = var_22770_equation_0, values = (var_22284_cast_fp16, var_22686_cast_fp16))[name = tensor("op_22770_cast_fp16")]; + tensor var_22772_equation_0 = const()[name = tensor("op_22772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22772_cast_fp16 = einsum(equation = var_22772_equation_0, values = (var_22288_cast_fp16, var_22687_cast_fp16))[name = tensor("op_22772_cast_fp16")]; + tensor var_22774_equation_0 = const()[name = tensor("op_22774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22774_cast_fp16 = einsum(equation = var_22774_equation_0, values = (var_22288_cast_fp16, var_22688_cast_fp16))[name = tensor("op_22774_cast_fp16")]; + tensor var_22776_equation_0 = const()[name = tensor("op_22776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22776_cast_fp16 = einsum(equation = var_22776_equation_0, values = (var_22288_cast_fp16, var_22689_cast_fp16))[name = tensor("op_22776_cast_fp16")]; + tensor var_22778_equation_0 = const()[name = tensor("op_22778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22778_cast_fp16 = einsum(equation = var_22778_equation_0, values = (var_22288_cast_fp16, var_22690_cast_fp16))[name = tensor("op_22778_cast_fp16")]; + tensor var_22780_equation_0 = const()[name = tensor("op_22780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22780_cast_fp16 = einsum(equation = var_22780_equation_0, values = (var_22292_cast_fp16, var_22691_cast_fp16))[name = tensor("op_22780_cast_fp16")]; + tensor var_22782_equation_0 = const()[name = tensor("op_22782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22782_cast_fp16 = einsum(equation = var_22782_equation_0, values = (var_22292_cast_fp16, var_22692_cast_fp16))[name = tensor("op_22782_cast_fp16")]; + tensor var_22784_equation_0 = const()[name = tensor("op_22784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22784_cast_fp16 = einsum(equation = var_22784_equation_0, values = (var_22292_cast_fp16, var_22693_cast_fp16))[name = tensor("op_22784_cast_fp16")]; + tensor var_22786_equation_0 = const()[name = tensor("op_22786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22786_cast_fp16 = einsum(equation = var_22786_equation_0, values = (var_22292_cast_fp16, var_22694_cast_fp16))[name = tensor("op_22786_cast_fp16")]; + tensor var_22788_equation_0 = const()[name = tensor("op_22788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22788_cast_fp16 = einsum(equation = var_22788_equation_0, values = (var_22296_cast_fp16, var_22695_cast_fp16))[name = tensor("op_22788_cast_fp16")]; + tensor var_22790_equation_0 = const()[name = tensor("op_22790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22790_cast_fp16 = einsum(equation = var_22790_equation_0, values = (var_22296_cast_fp16, var_22696_cast_fp16))[name = tensor("op_22790_cast_fp16")]; + tensor var_22792_equation_0 = const()[name = tensor("op_22792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22792_cast_fp16 = einsum(equation = var_22792_equation_0, values = (var_22296_cast_fp16, var_22697_cast_fp16))[name = tensor("op_22792_cast_fp16")]; + tensor var_22794_equation_0 = const()[name = tensor("op_22794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22794_cast_fp16 = einsum(equation = var_22794_equation_0, values = (var_22296_cast_fp16, var_22698_cast_fp16))[name = tensor("op_22794_cast_fp16")]; + tensor var_22796_equation_0 = const()[name = tensor("op_22796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22796_cast_fp16 = einsum(equation = var_22796_equation_0, values = (var_22300_cast_fp16, var_22699_cast_fp16))[name = tensor("op_22796_cast_fp16")]; + tensor var_22798_equation_0 = const()[name = tensor("op_22798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22798_cast_fp16 = einsum(equation = var_22798_equation_0, values = (var_22300_cast_fp16, var_22700_cast_fp16))[name = tensor("op_22798_cast_fp16")]; + tensor var_22800_equation_0 = const()[name = tensor("op_22800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22800_cast_fp16 = einsum(equation = var_22800_equation_0, values = (var_22300_cast_fp16, var_22701_cast_fp16))[name = tensor("op_22800_cast_fp16")]; + tensor var_22802_equation_0 = const()[name = tensor("op_22802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22802_cast_fp16 = einsum(equation = var_22802_equation_0, values = (var_22300_cast_fp16, var_22702_cast_fp16))[name = tensor("op_22802_cast_fp16")]; + tensor var_22804_equation_0 = const()[name = tensor("op_22804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22804_cast_fp16 = einsum(equation = var_22804_equation_0, values = (var_22304_cast_fp16, var_22703_cast_fp16))[name = tensor("op_22804_cast_fp16")]; + tensor var_22806_equation_0 = const()[name = tensor("op_22806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22806_cast_fp16 = einsum(equation = var_22806_equation_0, values = (var_22304_cast_fp16, var_22704_cast_fp16))[name = tensor("op_22806_cast_fp16")]; + tensor var_22808_equation_0 = const()[name = tensor("op_22808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22808_cast_fp16 = einsum(equation = var_22808_equation_0, values = (var_22304_cast_fp16, var_22705_cast_fp16))[name = tensor("op_22808_cast_fp16")]; + tensor var_22810_equation_0 = const()[name = tensor("op_22810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22810_cast_fp16 = einsum(equation = var_22810_equation_0, values = (var_22304_cast_fp16, var_22706_cast_fp16))[name = tensor("op_22810_cast_fp16")]; + tensor var_22812_equation_0 = const()[name = tensor("op_22812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22812_cast_fp16 = einsum(equation = var_22812_equation_0, values = (var_22308_cast_fp16, var_22707_cast_fp16))[name = tensor("op_22812_cast_fp16")]; + tensor var_22814_equation_0 = const()[name = tensor("op_22814_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22814_cast_fp16 = einsum(equation = var_22814_equation_0, values = (var_22308_cast_fp16, var_22708_cast_fp16))[name = tensor("op_22814_cast_fp16")]; + tensor var_22816_equation_0 = const()[name = tensor("op_22816_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22816_cast_fp16 = einsum(equation = var_22816_equation_0, values = (var_22308_cast_fp16, var_22709_cast_fp16))[name = tensor("op_22816_cast_fp16")]; + tensor var_22818_equation_0 = const()[name = tensor("op_22818_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22818_cast_fp16 = einsum(equation = var_22818_equation_0, values = (var_22308_cast_fp16, var_22710_cast_fp16))[name = tensor("op_22818_cast_fp16")]; + tensor var_22820_equation_0 = const()[name = tensor("op_22820_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22820_cast_fp16 = einsum(equation = var_22820_equation_0, values = (var_22312_cast_fp16, var_22711_cast_fp16))[name = tensor("op_22820_cast_fp16")]; + tensor var_22822_equation_0 = const()[name = tensor("op_22822_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22822_cast_fp16 = einsum(equation = var_22822_equation_0, values = (var_22312_cast_fp16, var_22712_cast_fp16))[name = tensor("op_22822_cast_fp16")]; + tensor var_22824_equation_0 = const()[name = tensor("op_22824_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22824_cast_fp16 = einsum(equation = var_22824_equation_0, values = (var_22312_cast_fp16, var_22713_cast_fp16))[name = tensor("op_22824_cast_fp16")]; + tensor var_22826_equation_0 = const()[name = tensor("op_22826_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22826_cast_fp16 = einsum(equation = var_22826_equation_0, values = (var_22312_cast_fp16, var_22714_cast_fp16))[name = tensor("op_22826_cast_fp16")]; + tensor var_22828_equation_0 = const()[name = tensor("op_22828_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22828_cast_fp16 = einsum(equation = var_22828_equation_0, values = (var_22316_cast_fp16, var_22715_cast_fp16))[name = tensor("op_22828_cast_fp16")]; + tensor var_22830_equation_0 = const()[name = tensor("op_22830_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22830_cast_fp16 = einsum(equation = var_22830_equation_0, values = (var_22316_cast_fp16, var_22716_cast_fp16))[name = tensor("op_22830_cast_fp16")]; + tensor var_22832_equation_0 = const()[name = tensor("op_22832_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22832_cast_fp16 = einsum(equation = var_22832_equation_0, values = (var_22316_cast_fp16, var_22717_cast_fp16))[name = tensor("op_22832_cast_fp16")]; + tensor var_22834_equation_0 = const()[name = tensor("op_22834_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22834_cast_fp16 = einsum(equation = var_22834_equation_0, values = (var_22316_cast_fp16, var_22718_cast_fp16))[name = tensor("op_22834_cast_fp16")]; + tensor var_22836_equation_0 = const()[name = tensor("op_22836_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22836_cast_fp16 = einsum(equation = var_22836_equation_0, values = (var_22320_cast_fp16, var_22719_cast_fp16))[name = tensor("op_22836_cast_fp16")]; + tensor var_22838_equation_0 = const()[name = tensor("op_22838_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22838_cast_fp16 = einsum(equation = var_22838_equation_0, values = (var_22320_cast_fp16, var_22720_cast_fp16))[name = tensor("op_22838_cast_fp16")]; + tensor var_22840_equation_0 = const()[name = tensor("op_22840_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22840_cast_fp16 = einsum(equation = var_22840_equation_0, values = (var_22320_cast_fp16, var_22721_cast_fp16))[name = tensor("op_22840_cast_fp16")]; + tensor var_22842_equation_0 = const()[name = tensor("op_22842_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22842_cast_fp16 = einsum(equation = var_22842_equation_0, values = (var_22320_cast_fp16, var_22722_cast_fp16))[name = tensor("op_22842_cast_fp16")]; + tensor var_22844_equation_0 = const()[name = tensor("op_22844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22844_cast_fp16 = einsum(equation = var_22844_equation_0, values = (var_22324_cast_fp16, var_22723_cast_fp16))[name = tensor("op_22844_cast_fp16")]; + tensor var_22846_equation_0 = const()[name = tensor("op_22846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22846_cast_fp16 = einsum(equation = var_22846_equation_0, values = (var_22324_cast_fp16, var_22724_cast_fp16))[name = tensor("op_22846_cast_fp16")]; + tensor var_22848_equation_0 = const()[name = tensor("op_22848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22848_cast_fp16 = einsum(equation = var_22848_equation_0, values = (var_22324_cast_fp16, var_22725_cast_fp16))[name = tensor("op_22848_cast_fp16")]; + tensor var_22850_equation_0 = const()[name = tensor("op_22850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22850_cast_fp16 = einsum(equation = var_22850_equation_0, values = (var_22324_cast_fp16, var_22726_cast_fp16))[name = tensor("op_22850_cast_fp16")]; + tensor var_22852_equation_0 = const()[name = tensor("op_22852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22852_cast_fp16 = einsum(equation = var_22852_equation_0, values = (var_22328_cast_fp16, var_22727_cast_fp16))[name = tensor("op_22852_cast_fp16")]; + tensor var_22854_equation_0 = const()[name = tensor("op_22854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22854_cast_fp16 = einsum(equation = var_22854_equation_0, values = (var_22328_cast_fp16, var_22728_cast_fp16))[name = tensor("op_22854_cast_fp16")]; + tensor var_22856_equation_0 = const()[name = tensor("op_22856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22856_cast_fp16 = einsum(equation = var_22856_equation_0, values = (var_22328_cast_fp16, var_22729_cast_fp16))[name = tensor("op_22856_cast_fp16")]; + tensor var_22858_equation_0 = const()[name = tensor("op_22858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22858_cast_fp16 = einsum(equation = var_22858_equation_0, values = (var_22328_cast_fp16, var_22730_cast_fp16))[name = tensor("op_22858_cast_fp16")]; + tensor var_22860_equation_0 = const()[name = tensor("op_22860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22860_cast_fp16 = einsum(equation = var_22860_equation_0, values = (var_22332_cast_fp16, var_22731_cast_fp16))[name = tensor("op_22860_cast_fp16")]; + tensor var_22862_equation_0 = const()[name = tensor("op_22862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22862_cast_fp16 = einsum(equation = var_22862_equation_0, values = (var_22332_cast_fp16, var_22732_cast_fp16))[name = tensor("op_22862_cast_fp16")]; + tensor var_22864_equation_0 = const()[name = tensor("op_22864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22864_cast_fp16 = einsum(equation = var_22864_equation_0, values = (var_22332_cast_fp16, var_22733_cast_fp16))[name = tensor("op_22864_cast_fp16")]; + tensor var_22866_equation_0 = const()[name = tensor("op_22866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22866_cast_fp16 = einsum(equation = var_22866_equation_0, values = (var_22332_cast_fp16, var_22734_cast_fp16))[name = tensor("op_22866_cast_fp16")]; + tensor var_22868_equation_0 = const()[name = tensor("op_22868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22868_cast_fp16 = einsum(equation = var_22868_equation_0, values = (var_22336_cast_fp16, var_22735_cast_fp16))[name = tensor("op_22868_cast_fp16")]; + tensor var_22870_equation_0 = const()[name = tensor("op_22870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22870_cast_fp16 = einsum(equation = var_22870_equation_0, values = (var_22336_cast_fp16, var_22736_cast_fp16))[name = tensor("op_22870_cast_fp16")]; + tensor var_22872_equation_0 = const()[name = tensor("op_22872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22872_cast_fp16 = einsum(equation = var_22872_equation_0, values = (var_22336_cast_fp16, var_22737_cast_fp16))[name = tensor("op_22872_cast_fp16")]; + tensor var_22874_equation_0 = const()[name = tensor("op_22874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22874_cast_fp16 = einsum(equation = var_22874_equation_0, values = (var_22336_cast_fp16, var_22738_cast_fp16))[name = tensor("op_22874_cast_fp16")]; + tensor var_22876_equation_0 = const()[name = tensor("op_22876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22876_cast_fp16 = einsum(equation = var_22876_equation_0, values = (var_22340_cast_fp16, var_22739_cast_fp16))[name = tensor("op_22876_cast_fp16")]; + tensor var_22878_equation_0 = const()[name = tensor("op_22878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22878_cast_fp16 = einsum(equation = var_22878_equation_0, values = (var_22340_cast_fp16, var_22740_cast_fp16))[name = tensor("op_22878_cast_fp16")]; + tensor var_22880_equation_0 = const()[name = tensor("op_22880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22880_cast_fp16 = einsum(equation = var_22880_equation_0, values = (var_22340_cast_fp16, var_22741_cast_fp16))[name = tensor("op_22880_cast_fp16")]; + tensor var_22882_equation_0 = const()[name = tensor("op_22882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22882_cast_fp16 = einsum(equation = var_22882_equation_0, values = (var_22340_cast_fp16, var_22742_cast_fp16))[name = tensor("op_22882_cast_fp16")]; + tensor var_22884_equation_0 = const()[name = tensor("op_22884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22884_cast_fp16 = einsum(equation = var_22884_equation_0, values = (var_22344_cast_fp16, var_22743_cast_fp16))[name = tensor("op_22884_cast_fp16")]; + tensor var_22886_equation_0 = const()[name = tensor("op_22886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22886_cast_fp16 = einsum(equation = var_22886_equation_0, values = (var_22344_cast_fp16, var_22744_cast_fp16))[name = tensor("op_22886_cast_fp16")]; + tensor var_22888_equation_0 = const()[name = tensor("op_22888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22888_cast_fp16 = einsum(equation = var_22888_equation_0, values = (var_22344_cast_fp16, var_22745_cast_fp16))[name = tensor("op_22888_cast_fp16")]; + tensor var_22890_equation_0 = const()[name = tensor("op_22890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22890_cast_fp16 = einsum(equation = var_22890_equation_0, values = (var_22344_cast_fp16, var_22746_cast_fp16))[name = tensor("op_22890_cast_fp16")]; + tensor var_22892_equation_0 = const()[name = tensor("op_22892_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22892_cast_fp16 = einsum(equation = var_22892_equation_0, values = (var_22348_cast_fp16, var_22747_cast_fp16))[name = tensor("op_22892_cast_fp16")]; + tensor var_22894_equation_0 = const()[name = tensor("op_22894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22894_cast_fp16 = einsum(equation = var_22894_equation_0, values = (var_22348_cast_fp16, var_22748_cast_fp16))[name = tensor("op_22894_cast_fp16")]; + tensor var_22896_equation_0 = const()[name = tensor("op_22896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22896_cast_fp16 = einsum(equation = var_22896_equation_0, values = (var_22348_cast_fp16, var_22749_cast_fp16))[name = tensor("op_22896_cast_fp16")]; + tensor var_22898_equation_0 = const()[name = tensor("op_22898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22898_cast_fp16 = einsum(equation = var_22898_equation_0, values = (var_22348_cast_fp16, var_22750_cast_fp16))[name = tensor("op_22898_cast_fp16")]; + tensor var_22900_equation_0 = const()[name = tensor("op_22900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22900_cast_fp16 = einsum(equation = var_22900_equation_0, values = (var_22352_cast_fp16, var_22751_cast_fp16))[name = tensor("op_22900_cast_fp16")]; + tensor var_22902_equation_0 = const()[name = tensor("op_22902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22902_cast_fp16 = einsum(equation = var_22902_equation_0, values = (var_22352_cast_fp16, var_22752_cast_fp16))[name = tensor("op_22902_cast_fp16")]; + tensor var_22904_equation_0 = const()[name = tensor("op_22904_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22904_cast_fp16 = einsum(equation = var_22904_equation_0, values = (var_22352_cast_fp16, var_22753_cast_fp16))[name = tensor("op_22904_cast_fp16")]; + tensor var_22906_equation_0 = const()[name = tensor("op_22906_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22906_cast_fp16 = einsum(equation = var_22906_equation_0, values = (var_22352_cast_fp16, var_22754_cast_fp16))[name = tensor("op_22906_cast_fp16")]; + tensor var_22908_equation_0 = const()[name = tensor("op_22908_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22908_cast_fp16 = einsum(equation = var_22908_equation_0, values = (var_22356_cast_fp16, var_22755_cast_fp16))[name = tensor("op_22908_cast_fp16")]; + tensor var_22910_equation_0 = const()[name = tensor("op_22910_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22910_cast_fp16 = einsum(equation = var_22910_equation_0, values = (var_22356_cast_fp16, var_22756_cast_fp16))[name = tensor("op_22910_cast_fp16")]; + tensor var_22912_equation_0 = const()[name = tensor("op_22912_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22912_cast_fp16 = einsum(equation = var_22912_equation_0, values = (var_22356_cast_fp16, var_22757_cast_fp16))[name = tensor("op_22912_cast_fp16")]; + tensor var_22914_equation_0 = const()[name = tensor("op_22914_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22914_cast_fp16 = einsum(equation = var_22914_equation_0, values = (var_22356_cast_fp16, var_22758_cast_fp16))[name = tensor("op_22914_cast_fp16")]; + tensor var_22916_equation_0 = const()[name = tensor("op_22916_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22916_cast_fp16 = einsum(equation = var_22916_equation_0, values = (var_22360_cast_fp16, var_22759_cast_fp16))[name = tensor("op_22916_cast_fp16")]; + tensor var_22918_equation_0 = const()[name = tensor("op_22918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22918_cast_fp16 = einsum(equation = var_22918_equation_0, values = (var_22360_cast_fp16, var_22760_cast_fp16))[name = tensor("op_22918_cast_fp16")]; + tensor var_22920_equation_0 = const()[name = tensor("op_22920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22920_cast_fp16 = einsum(equation = var_22920_equation_0, values = (var_22360_cast_fp16, var_22761_cast_fp16))[name = tensor("op_22920_cast_fp16")]; + tensor var_22922_equation_0 = const()[name = tensor("op_22922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_22922_cast_fp16 = einsum(equation = var_22922_equation_0, values = (var_22360_cast_fp16, var_22762_cast_fp16))[name = tensor("op_22922_cast_fp16")]; + tensor var_22924_interleave_0 = const()[name = tensor("op_22924_interleave_0"), val = tensor(false)]; + tensor var_22924_cast_fp16 = concat(axis = var_21429, interleave = var_22924_interleave_0, values = (var_22764_cast_fp16, var_22766_cast_fp16, var_22768_cast_fp16, var_22770_cast_fp16))[name = tensor("op_22924_cast_fp16")]; + tensor var_22926_interleave_0 = const()[name = tensor("op_22926_interleave_0"), val = tensor(false)]; + tensor var_22926_cast_fp16 = concat(axis = var_21429, interleave = var_22926_interleave_0, values = (var_22772_cast_fp16, var_22774_cast_fp16, var_22776_cast_fp16, var_22778_cast_fp16))[name = tensor("op_22926_cast_fp16")]; + tensor var_22928_interleave_0 = const()[name = tensor("op_22928_interleave_0"), val = tensor(false)]; + tensor var_22928_cast_fp16 = concat(axis = var_21429, interleave = var_22928_interleave_0, values = (var_22780_cast_fp16, var_22782_cast_fp16, var_22784_cast_fp16, var_22786_cast_fp16))[name = tensor("op_22928_cast_fp16")]; + tensor var_22930_interleave_0 = const()[name = tensor("op_22930_interleave_0"), val = tensor(false)]; + tensor var_22930_cast_fp16 = concat(axis = var_21429, interleave = var_22930_interleave_0, values = (var_22788_cast_fp16, var_22790_cast_fp16, var_22792_cast_fp16, var_22794_cast_fp16))[name = tensor("op_22930_cast_fp16")]; + tensor var_22932_interleave_0 = const()[name = tensor("op_22932_interleave_0"), val = tensor(false)]; + tensor var_22932_cast_fp16 = concat(axis = var_21429, interleave = var_22932_interleave_0, values = (var_22796_cast_fp16, var_22798_cast_fp16, var_22800_cast_fp16, var_22802_cast_fp16))[name = tensor("op_22932_cast_fp16")]; + tensor var_22934_interleave_0 = const()[name = tensor("op_22934_interleave_0"), val = tensor(false)]; + tensor var_22934_cast_fp16 = concat(axis = var_21429, interleave = var_22934_interleave_0, values = (var_22804_cast_fp16, var_22806_cast_fp16, var_22808_cast_fp16, var_22810_cast_fp16))[name = tensor("op_22934_cast_fp16")]; + tensor var_22936_interleave_0 = const()[name = tensor("op_22936_interleave_0"), val = tensor(false)]; + tensor var_22936_cast_fp16 = concat(axis = var_21429, interleave = var_22936_interleave_0, values = (var_22812_cast_fp16, var_22814_cast_fp16, var_22816_cast_fp16, var_22818_cast_fp16))[name = tensor("op_22936_cast_fp16")]; + tensor var_22938_interleave_0 = const()[name = tensor("op_22938_interleave_0"), val = tensor(false)]; + tensor var_22938_cast_fp16 = concat(axis = var_21429, interleave = var_22938_interleave_0, values = (var_22820_cast_fp16, var_22822_cast_fp16, var_22824_cast_fp16, var_22826_cast_fp16))[name = tensor("op_22938_cast_fp16")]; + tensor var_22940_interleave_0 = const()[name = tensor("op_22940_interleave_0"), val = tensor(false)]; + tensor var_22940_cast_fp16 = concat(axis = var_21429, interleave = var_22940_interleave_0, values = (var_22828_cast_fp16, var_22830_cast_fp16, var_22832_cast_fp16, var_22834_cast_fp16))[name = tensor("op_22940_cast_fp16")]; + tensor var_22942_interleave_0 = const()[name = tensor("op_22942_interleave_0"), val = tensor(false)]; + tensor var_22942_cast_fp16 = concat(axis = var_21429, interleave = var_22942_interleave_0, values = (var_22836_cast_fp16, var_22838_cast_fp16, var_22840_cast_fp16, var_22842_cast_fp16))[name = tensor("op_22942_cast_fp16")]; + tensor var_22944_interleave_0 = const()[name = tensor("op_22944_interleave_0"), val = tensor(false)]; + tensor var_22944_cast_fp16 = concat(axis = var_21429, interleave = var_22944_interleave_0, values = (var_22844_cast_fp16, var_22846_cast_fp16, var_22848_cast_fp16, var_22850_cast_fp16))[name = tensor("op_22944_cast_fp16")]; + tensor var_22946_interleave_0 = const()[name = tensor("op_22946_interleave_0"), val = tensor(false)]; + tensor var_22946_cast_fp16 = concat(axis = var_21429, interleave = var_22946_interleave_0, values = (var_22852_cast_fp16, var_22854_cast_fp16, var_22856_cast_fp16, var_22858_cast_fp16))[name = tensor("op_22946_cast_fp16")]; + tensor var_22948_interleave_0 = const()[name = tensor("op_22948_interleave_0"), val = tensor(false)]; + tensor var_22948_cast_fp16 = concat(axis = var_21429, interleave = var_22948_interleave_0, values = (var_22860_cast_fp16, var_22862_cast_fp16, var_22864_cast_fp16, var_22866_cast_fp16))[name = tensor("op_22948_cast_fp16")]; + tensor var_22950_interleave_0 = const()[name = tensor("op_22950_interleave_0"), val = tensor(false)]; + tensor var_22950_cast_fp16 = concat(axis = var_21429, interleave = var_22950_interleave_0, values = (var_22868_cast_fp16, var_22870_cast_fp16, var_22872_cast_fp16, var_22874_cast_fp16))[name = tensor("op_22950_cast_fp16")]; + tensor var_22952_interleave_0 = const()[name = tensor("op_22952_interleave_0"), val = tensor(false)]; + tensor var_22952_cast_fp16 = concat(axis = var_21429, interleave = var_22952_interleave_0, values = (var_22876_cast_fp16, var_22878_cast_fp16, var_22880_cast_fp16, var_22882_cast_fp16))[name = tensor("op_22952_cast_fp16")]; + tensor var_22954_interleave_0 = const()[name = tensor("op_22954_interleave_0"), val = tensor(false)]; + tensor var_22954_cast_fp16 = concat(axis = var_21429, interleave = var_22954_interleave_0, values = (var_22884_cast_fp16, var_22886_cast_fp16, var_22888_cast_fp16, var_22890_cast_fp16))[name = tensor("op_22954_cast_fp16")]; + tensor var_22956_interleave_0 = const()[name = tensor("op_22956_interleave_0"), val = tensor(false)]; + tensor var_22956_cast_fp16 = concat(axis = var_21429, interleave = var_22956_interleave_0, values = (var_22892_cast_fp16, var_22894_cast_fp16, var_22896_cast_fp16, var_22898_cast_fp16))[name = tensor("op_22956_cast_fp16")]; + tensor var_22958_interleave_0 = const()[name = tensor("op_22958_interleave_0"), val = tensor(false)]; + tensor var_22958_cast_fp16 = concat(axis = var_21429, interleave = var_22958_interleave_0, values = (var_22900_cast_fp16, var_22902_cast_fp16, var_22904_cast_fp16, var_22906_cast_fp16))[name = tensor("op_22958_cast_fp16")]; + tensor var_22960_interleave_0 = const()[name = tensor("op_22960_interleave_0"), val = tensor(false)]; + tensor var_22960_cast_fp16 = concat(axis = var_21429, interleave = var_22960_interleave_0, values = (var_22908_cast_fp16, var_22910_cast_fp16, var_22912_cast_fp16, var_22914_cast_fp16))[name = tensor("op_22960_cast_fp16")]; + tensor var_22962_interleave_0 = const()[name = tensor("op_22962_interleave_0"), val = tensor(false)]; + tensor var_22962_cast_fp16 = concat(axis = var_21429, interleave = var_22962_interleave_0, values = (var_22916_cast_fp16, var_22918_cast_fp16, var_22920_cast_fp16, var_22922_cast_fp16))[name = tensor("op_22962_cast_fp16")]; + tensor input_267_interleave_0 = const()[name = tensor("input_267_interleave_0"), val = tensor(false)]; + tensor input_267_cast_fp16 = concat(axis = var_21454, interleave = input_267_interleave_0, values = (var_22924_cast_fp16, var_22926_cast_fp16, var_22928_cast_fp16, var_22930_cast_fp16, var_22932_cast_fp16, var_22934_cast_fp16, var_22936_cast_fp16, var_22938_cast_fp16, var_22940_cast_fp16, var_22942_cast_fp16, var_22944_cast_fp16, var_22946_cast_fp16, var_22948_cast_fp16, var_22950_cast_fp16, var_22952_cast_fp16, var_22954_cast_fp16, var_22956_cast_fp16, var_22958_cast_fp16, var_22960_cast_fp16, var_22962_cast_fp16))[name = tensor("input_267_cast_fp16")]; + tensor var_22970 = const()[name = tensor("op_22970"), val = tensor([1, 1])]; + tensor var_22972 = const()[name = tensor("op_22972"), val = tensor([1, 1])]; + tensor pretrained_out_163_pad_type_0 = const()[name = tensor("pretrained_out_163_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_163_pad_0 = const()[name = tensor("pretrained_out_163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154832320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155651584))), name = tensor("layers_13_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_13_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155651712)))]; + tensor pretrained_out_163_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_22972, groups = var_21454, pad = pretrained_out_163_pad_0, pad_type = pretrained_out_163_pad_type_0, strides = var_22970, weight = layers_13_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_267_cast_fp16)[name = tensor("pretrained_out_163_cast_fp16")]; + tensor var_22976 = const()[name = tensor("op_22976"), val = tensor([1, 1])]; + tensor var_22978 = const()[name = tensor("op_22978"), val = tensor([1, 1])]; + tensor input_269_pad_type_0 = const()[name = tensor("input_269_pad_type_0"), val = tensor("custom")]; + tensor input_269_pad_0 = const()[name = tensor("input_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155654336)))]; + tensor input_269_cast_fp16 = conv(dilations = var_22978, groups = var_21454, pad = input_269_pad_0, pad_type = input_269_pad_type_0, strides = var_22976, weight = layers_13_self_attn_o_proj_loraA_weight_to_fp16, x = input_267_cast_fp16)[name = tensor("input_269_cast_fp16")]; + tensor var_22982 = const()[name = tensor("op_22982"), val = tensor([1, 1])]; + tensor var_22984 = const()[name = tensor("op_22984"), val = tensor([1, 1])]; + tensor lora_out_325_pad_type_0 = const()[name = tensor("lora_out_325_pad_type_0"), val = tensor("custom")]; + tensor lora_out_325_pad_0 = const()[name = tensor("lora_out_325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_327_weight_0_to_fp16 = const()[name = tensor("lora_out_327_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155695360)))]; + tensor lora_out_327_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_22984, groups = var_21454, pad = lora_out_325_pad_0, pad_type = lora_out_325_pad_type_0, strides = var_22982, weight = lora_out_327_weight_0_to_fp16, x = input_269_cast_fp16)[name = tensor("lora_out_327_cast_fp16")]; + tensor obj_55_cast_fp16 = add(x = pretrained_out_163_cast_fp16, y = lora_out_327_cast_fp16)[name = tensor("obj_55_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_55_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; + tensor var_22993 = const()[name = tensor("op_22993"), val = tensor([1])]; + tensor channels_mean_55_cast_fp16 = reduce_mean(axes = var_22993, keep_dims = var_21455, x = inputs_55_cast_fp16)[name = tensor("channels_mean_55_cast_fp16")]; + tensor zero_mean_55_cast_fp16 = sub(x = inputs_55_cast_fp16, y = channels_mean_55_cast_fp16)[name = tensor("zero_mean_55_cast_fp16")]; + tensor zero_mean_sq_55_cast_fp16 = mul(x = zero_mean_55_cast_fp16, y = zero_mean_55_cast_fp16)[name = tensor("zero_mean_sq_55_cast_fp16")]; + tensor var_22997 = const()[name = tensor("op_22997"), val = tensor([1])]; + tensor var_22998_cast_fp16 = reduce_mean(axes = var_22997, keep_dims = var_21455, x = zero_mean_sq_55_cast_fp16)[name = tensor("op_22998_cast_fp16")]; + tensor var_22999_to_fp16 = const()[name = tensor("op_22999_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_23000_cast_fp16 = add(x = var_22998_cast_fp16, y = var_22999_to_fp16)[name = tensor("op_23000_cast_fp16")]; + tensor denom_55_epsilon_0 = const()[name = tensor("denom_55_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_55_cast_fp16 = rsqrt(epsilon = denom_55_epsilon_0, x = var_23000_cast_fp16)[name = tensor("denom_55_cast_fp16")]; + tensor out_55_cast_fp16 = mul(x = zero_mean_55_cast_fp16, y = denom_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; + tensor input_271_gamma_0_to_fp16 = const()[name = tensor("input_271_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155736384)))]; + tensor input_271_beta_0_to_fp16 = const()[name = tensor("input_271_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155739008)))]; + tensor input_271_epsilon_0_to_fp16 = const()[name = tensor("input_271_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_271_cast_fp16 = batch_norm(beta = input_271_beta_0_to_fp16, epsilon = input_271_epsilon_0_to_fp16, gamma = input_271_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor("input_271_cast_fp16")]; + tensor var_23014 = const()[name = tensor("op_23014"), val = tensor([1, 1])]; + tensor var_23016 = const()[name = tensor("op_23016"), val = tensor([1, 1])]; + tensor pretrained_out_165_pad_type_0 = const()[name = tensor("pretrained_out_165_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_165_pad_0 = const()[name = tensor("pretrained_out_165_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155741632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159018496))), name = tensor("layers_13_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_13_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_13_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159018624)))]; + tensor pretrained_out_165_cast_fp16 = conv(bias = layers_13_fc1_pretrained_bias_to_fp16, dilations = var_23016, groups = var_21454, pad = pretrained_out_165_pad_0, pad_type = pretrained_out_165_pad_type_0, strides = var_23014, weight = layers_13_fc1_pretrained_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = tensor("pretrained_out_165_cast_fp16")]; + tensor var_23020 = const()[name = tensor("op_23020"), val = tensor([1, 1])]; + tensor var_23022 = const()[name = tensor("op_23022"), val = tensor([1, 1])]; + tensor input_273_pad_type_0 = const()[name = tensor("input_273_pad_type_0"), val = tensor("custom")]; + tensor input_273_pad_0 = const()[name = tensor("input_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_13_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159028928)))]; + tensor input_273_cast_fp16 = conv(dilations = var_23022, groups = var_21454, pad = input_273_pad_0, pad_type = input_273_pad_type_0, strides = var_23020, weight = layers_13_fc1_loraA_weight_to_fp16, x = input_271_cast_fp16)[name = tensor("input_273_cast_fp16")]; + tensor var_23026 = const()[name = tensor("op_23026"), val = tensor([1, 1])]; + tensor var_23028 = const()[name = tensor("op_23028"), val = tensor([1, 1])]; + tensor lora_out_329_pad_type_0 = const()[name = tensor("lora_out_329_pad_type_0"), val = tensor("custom")]; + tensor lora_out_329_pad_0 = const()[name = tensor("lora_out_329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_331_weight_0_to_fp16 = const()[name = tensor("lora_out_331_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159069952)))]; + tensor lora_out_331_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_23028, groups = var_21454, pad = lora_out_329_pad_0, pad_type = lora_out_329_pad_type_0, strides = var_23026, weight = lora_out_331_weight_0_to_fp16, x = input_273_cast_fp16)[name = tensor("lora_out_331_cast_fp16")]; + tensor input_275_cast_fp16 = add(x = pretrained_out_165_cast_fp16, y = lora_out_331_cast_fp16)[name = tensor("input_275_cast_fp16")]; + tensor input_277_mode_0 = const()[name = tensor("input_277_mode_0"), val = tensor("EXACT")]; + tensor input_277_cast_fp16 = gelu(mode = input_277_mode_0, x = input_275_cast_fp16)[name = tensor("input_277_cast_fp16")]; + tensor var_23040 = const()[name = tensor("op_23040"), val = tensor([1, 1])]; + tensor var_23042 = const()[name = tensor("op_23042"), val = tensor([1, 1])]; + tensor pretrained_out_167_pad_type_0 = const()[name = tensor("pretrained_out_167_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_167_pad_0 = const()[name = tensor("pretrained_out_167_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(159233856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162510720))), name = tensor("layers_13_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_13_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_13_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162510848)))]; + tensor pretrained_out_167_cast_fp16 = conv(bias = layers_13_fc2_pretrained_bias_to_fp16, dilations = var_23042, groups = var_21454, pad = pretrained_out_167_pad_0, pad_type = pretrained_out_167_pad_type_0, strides = var_23040, weight = layers_13_fc2_pretrained_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = tensor("pretrained_out_167_cast_fp16")]; + tensor var_23046 = const()[name = tensor("op_23046"), val = tensor([1, 1])]; + tensor var_23048 = const()[name = tensor("op_23048"), val = tensor([1, 1])]; + tensor input_279_pad_type_0 = const()[name = tensor("input_279_pad_type_0"), val = tensor("custom")]; + tensor input_279_pad_0 = const()[name = tensor("input_279_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_13_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_13_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162513472)))]; + tensor input_279_cast_fp16 = conv(dilations = var_23048, groups = var_21454, pad = input_279_pad_0, pad_type = input_279_pad_type_0, strides = var_23046, weight = layers_13_fc2_loraA_weight_to_fp16, x = input_277_cast_fp16)[name = tensor("input_279_cast_fp16")]; + tensor var_23052 = const()[name = tensor("op_23052"), val = tensor([1, 1])]; + tensor var_23054 = const()[name = tensor("op_23054"), val = tensor([1, 1])]; + tensor lora_out_333_pad_type_0 = const()[name = tensor("lora_out_333_pad_type_0"), val = tensor("custom")]; + tensor lora_out_333_pad_0 = const()[name = tensor("lora_out_333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_335_weight_0_to_fp16 = const()[name = tensor("lora_out_335_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162677376)))]; + tensor lora_out_335_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_23054, groups = var_21454, pad = lora_out_333_pad_0, pad_type = lora_out_333_pad_type_0, strides = var_23052, weight = lora_out_335_weight_0_to_fp16, x = input_279_cast_fp16)[name = tensor("lora_out_335_cast_fp16")]; + tensor hidden_states_31_cast_fp16 = add(x = pretrained_out_167_cast_fp16, y = lora_out_335_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = hidden_states_31_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; + tensor var_23064 = const()[name = tensor("op_23064"), val = tensor(3)]; + tensor var_23089 = const()[name = tensor("op_23089"), val = tensor(1)]; + tensor var_23090 = const()[name = tensor("op_23090"), val = tensor(true)]; + tensor var_23100 = const()[name = tensor("op_23100"), val = tensor([1])]; + tensor channels_mean_57_cast_fp16 = reduce_mean(axes = var_23100, keep_dims = var_23090, x = inputs_57_cast_fp16)[name = tensor("channels_mean_57_cast_fp16")]; + tensor zero_mean_57_cast_fp16 = sub(x = inputs_57_cast_fp16, y = channels_mean_57_cast_fp16)[name = tensor("zero_mean_57_cast_fp16")]; + tensor zero_mean_sq_57_cast_fp16 = mul(x = zero_mean_57_cast_fp16, y = zero_mean_57_cast_fp16)[name = tensor("zero_mean_sq_57_cast_fp16")]; + tensor var_23104 = const()[name = tensor("op_23104"), val = tensor([1])]; + tensor var_23105_cast_fp16 = reduce_mean(axes = var_23104, keep_dims = var_23090, x = zero_mean_sq_57_cast_fp16)[name = tensor("op_23105_cast_fp16")]; + tensor var_23106_to_fp16 = const()[name = tensor("op_23106_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_23107_cast_fp16 = add(x = var_23105_cast_fp16, y = var_23106_to_fp16)[name = tensor("op_23107_cast_fp16")]; + tensor denom_57_epsilon_0 = const()[name = tensor("denom_57_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_57_cast_fp16 = rsqrt(epsilon = denom_57_epsilon_0, x = var_23107_cast_fp16)[name = tensor("denom_57_cast_fp16")]; + tensor out_57_cast_fp16 = mul(x = zero_mean_57_cast_fp16, y = denom_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; + tensor obj_57_gamma_0_to_fp16 = const()[name = tensor("obj_57_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162718400)))]; + tensor obj_57_beta_0_to_fp16 = const()[name = tensor("obj_57_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162721024)))]; + tensor obj_57_epsilon_0_to_fp16 = const()[name = tensor("obj_57_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_57_cast_fp16 = batch_norm(beta = obj_57_beta_0_to_fp16, epsilon = obj_57_epsilon_0_to_fp16, gamma = obj_57_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor("obj_57_cast_fp16")]; + tensor var_23125 = const()[name = tensor("op_23125"), val = tensor([1, 1])]; + tensor var_23127 = const()[name = tensor("op_23127"), val = tensor([1, 1])]; + tensor pretrained_out_169_pad_type_0 = const()[name = tensor("pretrained_out_169_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_169_pad_0 = const()[name = tensor("pretrained_out_169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162723648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163542912))), name = tensor("layers_14_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_14_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163543040)))]; + tensor pretrained_out_169_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_23127, groups = var_23089, pad = pretrained_out_169_pad_0, pad_type = pretrained_out_169_pad_type_0, strides = var_23125, weight = layers_14_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("pretrained_out_169_cast_fp16")]; + tensor var_23131 = const()[name = tensor("op_23131"), val = tensor([1, 1])]; + tensor var_23133 = const()[name = tensor("op_23133"), val = tensor([1, 1])]; + tensor input_281_pad_type_0 = const()[name = tensor("input_281_pad_type_0"), val = tensor("custom")]; + tensor input_281_pad_0 = const()[name = tensor("input_281_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163545664)))]; + tensor input_281_cast_fp16 = conv(dilations = var_23133, groups = var_23089, pad = input_281_pad_0, pad_type = input_281_pad_type_0, strides = var_23131, weight = layers_14_self_attn_q_proj_loraA_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("input_281_cast_fp16")]; + tensor var_23137 = const()[name = tensor("op_23137"), val = tensor([1, 1])]; + tensor var_23139 = const()[name = tensor("op_23139"), val = tensor([1, 1])]; + tensor lora_out_337_pad_type_0 = const()[name = tensor("lora_out_337_pad_type_0"), val = tensor("custom")]; + tensor lora_out_337_pad_0 = const()[name = tensor("lora_out_337_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_339_weight_0_to_fp16 = const()[name = tensor("lora_out_339_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163586688)))]; + tensor lora_out_339_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_23139, groups = var_23089, pad = lora_out_337_pad_0, pad_type = lora_out_337_pad_type_0, strides = var_23137, weight = lora_out_339_weight_0_to_fp16, x = input_281_cast_fp16)[name = tensor("lora_out_339_cast_fp16")]; + tensor query_29_cast_fp16 = add(x = pretrained_out_169_cast_fp16, y = lora_out_339_cast_fp16)[name = tensor("query_29_cast_fp16")]; + tensor var_23149 = const()[name = tensor("op_23149"), val = tensor([1, 1])]; + tensor var_23151 = const()[name = tensor("op_23151"), val = tensor([1, 1])]; + tensor pretrained_out_171_pad_type_0 = const()[name = tensor("pretrained_out_171_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_171_pad_0 = const()[name = tensor("pretrained_out_171_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163627712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164446976))), name = tensor("layers_14_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_171_cast_fp16 = conv(dilations = var_23151, groups = var_23089, pad = pretrained_out_171_pad_0, pad_type = pretrained_out_171_pad_type_0, strides = var_23149, weight = layers_14_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("pretrained_out_171_cast_fp16")]; + tensor var_23155 = const()[name = tensor("op_23155"), val = tensor([1, 1])]; + tensor var_23157 = const()[name = tensor("op_23157"), val = tensor([1, 1])]; + tensor input_283_pad_type_0 = const()[name = tensor("input_283_pad_type_0"), val = tensor("custom")]; + tensor input_283_pad_0 = const()[name = tensor("input_283_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164447104)))]; + tensor input_283_cast_fp16 = conv(dilations = var_23157, groups = var_23089, pad = input_283_pad_0, pad_type = input_283_pad_type_0, strides = var_23155, weight = layers_14_self_attn_k_proj_loraA_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("input_283_cast_fp16")]; + tensor var_23161 = const()[name = tensor("op_23161"), val = tensor([1, 1])]; + tensor var_23163 = const()[name = tensor("op_23163"), val = tensor([1, 1])]; + tensor lora_out_341_pad_type_0 = const()[name = tensor("lora_out_341_pad_type_0"), val = tensor("custom")]; + tensor lora_out_341_pad_0 = const()[name = tensor("lora_out_341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_343_weight_0_to_fp16 = const()[name = tensor("lora_out_343_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164488128)))]; + tensor lora_out_343_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_23163, groups = var_23089, pad = lora_out_341_pad_0, pad_type = lora_out_341_pad_type_0, strides = var_23161, weight = lora_out_343_weight_0_to_fp16, x = input_283_cast_fp16)[name = tensor("lora_out_343_cast_fp16")]; + tensor key_29_cast_fp16 = add(x = pretrained_out_171_cast_fp16, y = lora_out_343_cast_fp16)[name = tensor("key_29_cast_fp16")]; + tensor var_23174 = const()[name = tensor("op_23174"), val = tensor([1, 1])]; + tensor var_23176 = const()[name = tensor("op_23176"), val = tensor([1, 1])]; + tensor pretrained_out_173_pad_type_0 = const()[name = tensor("pretrained_out_173_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_173_pad_0 = const()[name = tensor("pretrained_out_173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164529152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165348416))), name = tensor("layers_14_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_14_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165348544)))]; + tensor pretrained_out_173_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_23176, groups = var_23089, pad = pretrained_out_173_pad_0, pad_type = pretrained_out_173_pad_type_0, strides = var_23174, weight = layers_14_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_57_cast_fp16)[name = tensor("pretrained_out_173_cast_fp16")]; + tensor var_23180 = const()[name = tensor("op_23180"), val = tensor([1, 1])]; + tensor var_23182 = const()[name = tensor("op_23182"), val = tensor([1, 1])]; + tensor input_285_pad_type_0 = const()[name = tensor("input_285_pad_type_0"), val = tensor("custom")]; + tensor input_285_pad_0 = const()[name = tensor("input_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165351168)))]; + tensor input_285_cast_fp16 = conv(dilations = var_23182, groups = var_23089, pad = input_285_pad_0, pad_type = input_285_pad_type_0, strides = var_23180, weight = layers_14_self_attn_v_proj_loraA_weight_to_fp16, x = obj_57_cast_fp16)[name = tensor("input_285_cast_fp16")]; + tensor var_23186 = const()[name = tensor("op_23186"), val = tensor([1, 1])]; + tensor var_23188 = const()[name = tensor("op_23188"), val = tensor([1, 1])]; + tensor lora_out_345_pad_type_0 = const()[name = tensor("lora_out_345_pad_type_0"), val = tensor("custom")]; + tensor lora_out_345_pad_0 = const()[name = tensor("lora_out_345_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_347_weight_0_to_fp16 = const()[name = tensor("lora_out_347_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165392192)))]; + tensor lora_out_347_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_23188, groups = var_23089, pad = lora_out_345_pad_0, pad_type = lora_out_345_pad_type_0, strides = var_23186, weight = lora_out_347_weight_0_to_fp16, x = input_285_cast_fp16)[name = tensor("lora_out_347_cast_fp16")]; + tensor value_29_cast_fp16 = add(x = pretrained_out_173_cast_fp16, y = lora_out_347_cast_fp16)[name = tensor("value_29_cast_fp16")]; + tensor var_23198_begin_0 = const()[name = tensor("op_23198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23198_end_0 = const()[name = tensor("op_23198_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23198_end_mask_0 = const()[name = tensor("op_23198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23198_cast_fp16 = slice_by_index(begin = var_23198_begin_0, end = var_23198_end_0, end_mask = var_23198_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23198_cast_fp16")]; + tensor var_23202_begin_0 = const()[name = tensor("op_23202_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23202_end_0 = const()[name = tensor("op_23202_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23202_end_mask_0 = const()[name = tensor("op_23202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23202_cast_fp16 = slice_by_index(begin = var_23202_begin_0, end = var_23202_end_0, end_mask = var_23202_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23202_cast_fp16")]; + tensor var_23206_begin_0 = const()[name = tensor("op_23206_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23206_end_0 = const()[name = tensor("op_23206_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23206_end_mask_0 = const()[name = tensor("op_23206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23206_cast_fp16 = slice_by_index(begin = var_23206_begin_0, end = var_23206_end_0, end_mask = var_23206_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23206_cast_fp16")]; + tensor var_23210_begin_0 = const()[name = tensor("op_23210_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23210_end_0 = const()[name = tensor("op_23210_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23210_end_mask_0 = const()[name = tensor("op_23210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23210_cast_fp16 = slice_by_index(begin = var_23210_begin_0, end = var_23210_end_0, end_mask = var_23210_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23210_cast_fp16")]; + tensor var_23214_begin_0 = const()[name = tensor("op_23214_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23214_end_0 = const()[name = tensor("op_23214_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23214_end_mask_0 = const()[name = tensor("op_23214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23214_cast_fp16 = slice_by_index(begin = var_23214_begin_0, end = var_23214_end_0, end_mask = var_23214_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23214_cast_fp16")]; + tensor var_23218_begin_0 = const()[name = tensor("op_23218_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23218_end_0 = const()[name = tensor("op_23218_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23218_end_mask_0 = const()[name = tensor("op_23218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23218_cast_fp16 = slice_by_index(begin = var_23218_begin_0, end = var_23218_end_0, end_mask = var_23218_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23218_cast_fp16")]; + tensor var_23222_begin_0 = const()[name = tensor("op_23222_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23222_end_0 = const()[name = tensor("op_23222_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23222_end_mask_0 = const()[name = tensor("op_23222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23222_cast_fp16 = slice_by_index(begin = var_23222_begin_0, end = var_23222_end_0, end_mask = var_23222_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23222_cast_fp16")]; + tensor var_23226_begin_0 = const()[name = tensor("op_23226_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23226_end_0 = const()[name = tensor("op_23226_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23226_end_mask_0 = const()[name = tensor("op_23226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23226_cast_fp16 = slice_by_index(begin = var_23226_begin_0, end = var_23226_end_0, end_mask = var_23226_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23226_cast_fp16")]; + tensor var_23230_begin_0 = const()[name = tensor("op_23230_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23230_end_0 = const()[name = tensor("op_23230_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23230_end_mask_0 = const()[name = tensor("op_23230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23230_cast_fp16 = slice_by_index(begin = var_23230_begin_0, end = var_23230_end_0, end_mask = var_23230_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23230_cast_fp16")]; + tensor var_23234_begin_0 = const()[name = tensor("op_23234_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23234_end_0 = const()[name = tensor("op_23234_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23234_end_mask_0 = const()[name = tensor("op_23234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23234_cast_fp16 = slice_by_index(begin = var_23234_begin_0, end = var_23234_end_0, end_mask = var_23234_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23234_cast_fp16")]; + tensor var_23238_begin_0 = const()[name = tensor("op_23238_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23238_end_0 = const()[name = tensor("op_23238_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23238_end_mask_0 = const()[name = tensor("op_23238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23238_cast_fp16 = slice_by_index(begin = var_23238_begin_0, end = var_23238_end_0, end_mask = var_23238_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23238_cast_fp16")]; + tensor var_23242_begin_0 = const()[name = tensor("op_23242_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23242_end_0 = const()[name = tensor("op_23242_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23242_end_mask_0 = const()[name = tensor("op_23242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23242_cast_fp16 = slice_by_index(begin = var_23242_begin_0, end = var_23242_end_0, end_mask = var_23242_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23242_cast_fp16")]; + tensor var_23246_begin_0 = const()[name = tensor("op_23246_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23246_end_0 = const()[name = tensor("op_23246_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23246_end_mask_0 = const()[name = tensor("op_23246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23246_cast_fp16 = slice_by_index(begin = var_23246_begin_0, end = var_23246_end_0, end_mask = var_23246_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23246_cast_fp16")]; + tensor var_23250_begin_0 = const()[name = tensor("op_23250_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23250_end_0 = const()[name = tensor("op_23250_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23250_end_mask_0 = const()[name = tensor("op_23250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23250_cast_fp16 = slice_by_index(begin = var_23250_begin_0, end = var_23250_end_0, end_mask = var_23250_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23250_cast_fp16")]; + tensor var_23254_begin_0 = const()[name = tensor("op_23254_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23254_end_0 = const()[name = tensor("op_23254_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23254_end_mask_0 = const()[name = tensor("op_23254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23254_cast_fp16 = slice_by_index(begin = var_23254_begin_0, end = var_23254_end_0, end_mask = var_23254_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23254_cast_fp16")]; + tensor var_23258_begin_0 = const()[name = tensor("op_23258_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23258_end_0 = const()[name = tensor("op_23258_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23258_end_mask_0 = const()[name = tensor("op_23258_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23258_cast_fp16 = slice_by_index(begin = var_23258_begin_0, end = var_23258_end_0, end_mask = var_23258_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23258_cast_fp16")]; + tensor var_23262_begin_0 = const()[name = tensor("op_23262_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23262_end_0 = const()[name = tensor("op_23262_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23262_end_mask_0 = const()[name = tensor("op_23262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23262_cast_fp16 = slice_by_index(begin = var_23262_begin_0, end = var_23262_end_0, end_mask = var_23262_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23262_cast_fp16")]; + tensor var_23266_begin_0 = const()[name = tensor("op_23266_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23266_end_0 = const()[name = tensor("op_23266_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23266_end_mask_0 = const()[name = tensor("op_23266_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23266_cast_fp16 = slice_by_index(begin = var_23266_begin_0, end = var_23266_end_0, end_mask = var_23266_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23266_cast_fp16")]; + tensor var_23270_begin_0 = const()[name = tensor("op_23270_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23270_end_0 = const()[name = tensor("op_23270_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23270_end_mask_0 = const()[name = tensor("op_23270_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23270_cast_fp16 = slice_by_index(begin = var_23270_begin_0, end = var_23270_end_0, end_mask = var_23270_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23270_cast_fp16")]; + tensor var_23274_begin_0 = const()[name = tensor("op_23274_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23274_end_0 = const()[name = tensor("op_23274_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23274_end_mask_0 = const()[name = tensor("op_23274_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23274_cast_fp16 = slice_by_index(begin = var_23274_begin_0, end = var_23274_end_0, end_mask = var_23274_end_mask_0, x = query_29_cast_fp16)[name = tensor("op_23274_cast_fp16")]; + tensor var_23283_begin_0 = const()[name = tensor("op_23283_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23283_end_0 = const()[name = tensor("op_23283_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23283_end_mask_0 = const()[name = tensor("op_23283_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23283_cast_fp16 = slice_by_index(begin = var_23283_begin_0, end = var_23283_end_0, end_mask = var_23283_end_mask_0, x = var_23198_cast_fp16)[name = tensor("op_23283_cast_fp16")]; + tensor var_23290_begin_0 = const()[name = tensor("op_23290_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23290_end_0 = const()[name = tensor("op_23290_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23290_end_mask_0 = const()[name = tensor("op_23290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23290_cast_fp16 = slice_by_index(begin = var_23290_begin_0, end = var_23290_end_0, end_mask = var_23290_end_mask_0, x = var_23198_cast_fp16)[name = tensor("op_23290_cast_fp16")]; + tensor var_23297_begin_0 = const()[name = tensor("op_23297_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23297_end_0 = const()[name = tensor("op_23297_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23297_end_mask_0 = const()[name = tensor("op_23297_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23297_cast_fp16 = slice_by_index(begin = var_23297_begin_0, end = var_23297_end_0, end_mask = var_23297_end_mask_0, x = var_23198_cast_fp16)[name = tensor("op_23297_cast_fp16")]; + tensor var_23304_begin_0 = const()[name = tensor("op_23304_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23304_end_0 = const()[name = tensor("op_23304_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23304_end_mask_0 = const()[name = tensor("op_23304_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23304_cast_fp16 = slice_by_index(begin = var_23304_begin_0, end = var_23304_end_0, end_mask = var_23304_end_mask_0, x = var_23198_cast_fp16)[name = tensor("op_23304_cast_fp16")]; + tensor var_23311_begin_0 = const()[name = tensor("op_23311_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23311_end_0 = const()[name = tensor("op_23311_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23311_end_mask_0 = const()[name = tensor("op_23311_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23311_cast_fp16 = slice_by_index(begin = var_23311_begin_0, end = var_23311_end_0, end_mask = var_23311_end_mask_0, x = var_23202_cast_fp16)[name = tensor("op_23311_cast_fp16")]; + tensor var_23318_begin_0 = const()[name = tensor("op_23318_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23318_end_0 = const()[name = tensor("op_23318_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23318_end_mask_0 = const()[name = tensor("op_23318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23318_cast_fp16 = slice_by_index(begin = var_23318_begin_0, end = var_23318_end_0, end_mask = var_23318_end_mask_0, x = var_23202_cast_fp16)[name = tensor("op_23318_cast_fp16")]; + tensor var_23325_begin_0 = const()[name = tensor("op_23325_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23325_end_0 = const()[name = tensor("op_23325_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23325_end_mask_0 = const()[name = tensor("op_23325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23325_cast_fp16 = slice_by_index(begin = var_23325_begin_0, end = var_23325_end_0, end_mask = var_23325_end_mask_0, x = var_23202_cast_fp16)[name = tensor("op_23325_cast_fp16")]; + tensor var_23332_begin_0 = const()[name = tensor("op_23332_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23332_end_0 = const()[name = tensor("op_23332_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23332_end_mask_0 = const()[name = tensor("op_23332_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23332_cast_fp16 = slice_by_index(begin = var_23332_begin_0, end = var_23332_end_0, end_mask = var_23332_end_mask_0, x = var_23202_cast_fp16)[name = tensor("op_23332_cast_fp16")]; + tensor var_23339_begin_0 = const()[name = tensor("op_23339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23339_end_0 = const()[name = tensor("op_23339_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23339_end_mask_0 = const()[name = tensor("op_23339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23339_cast_fp16 = slice_by_index(begin = var_23339_begin_0, end = var_23339_end_0, end_mask = var_23339_end_mask_0, x = var_23206_cast_fp16)[name = tensor("op_23339_cast_fp16")]; + tensor var_23346_begin_0 = const()[name = tensor("op_23346_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23346_end_0 = const()[name = tensor("op_23346_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23346_end_mask_0 = const()[name = tensor("op_23346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23346_cast_fp16 = slice_by_index(begin = var_23346_begin_0, end = var_23346_end_0, end_mask = var_23346_end_mask_0, x = var_23206_cast_fp16)[name = tensor("op_23346_cast_fp16")]; + tensor var_23353_begin_0 = const()[name = tensor("op_23353_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23353_end_0 = const()[name = tensor("op_23353_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23353_end_mask_0 = const()[name = tensor("op_23353_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23353_cast_fp16 = slice_by_index(begin = var_23353_begin_0, end = var_23353_end_0, end_mask = var_23353_end_mask_0, x = var_23206_cast_fp16)[name = tensor("op_23353_cast_fp16")]; + tensor var_23360_begin_0 = const()[name = tensor("op_23360_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23360_end_0 = const()[name = tensor("op_23360_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23360_end_mask_0 = const()[name = tensor("op_23360_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23360_cast_fp16 = slice_by_index(begin = var_23360_begin_0, end = var_23360_end_0, end_mask = var_23360_end_mask_0, x = var_23206_cast_fp16)[name = tensor("op_23360_cast_fp16")]; + tensor var_23367_begin_0 = const()[name = tensor("op_23367_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23367_end_0 = const()[name = tensor("op_23367_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23367_end_mask_0 = const()[name = tensor("op_23367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23367_cast_fp16 = slice_by_index(begin = var_23367_begin_0, end = var_23367_end_0, end_mask = var_23367_end_mask_0, x = var_23210_cast_fp16)[name = tensor("op_23367_cast_fp16")]; + tensor var_23374_begin_0 = const()[name = tensor("op_23374_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23374_end_0 = const()[name = tensor("op_23374_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23374_end_mask_0 = const()[name = tensor("op_23374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23374_cast_fp16 = slice_by_index(begin = var_23374_begin_0, end = var_23374_end_0, end_mask = var_23374_end_mask_0, x = var_23210_cast_fp16)[name = tensor("op_23374_cast_fp16")]; + tensor var_23381_begin_0 = const()[name = tensor("op_23381_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23381_end_0 = const()[name = tensor("op_23381_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23381_end_mask_0 = const()[name = tensor("op_23381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23381_cast_fp16 = slice_by_index(begin = var_23381_begin_0, end = var_23381_end_0, end_mask = var_23381_end_mask_0, x = var_23210_cast_fp16)[name = tensor("op_23381_cast_fp16")]; + tensor var_23388_begin_0 = const()[name = tensor("op_23388_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23388_end_0 = const()[name = tensor("op_23388_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23388_end_mask_0 = const()[name = tensor("op_23388_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23388_cast_fp16 = slice_by_index(begin = var_23388_begin_0, end = var_23388_end_0, end_mask = var_23388_end_mask_0, x = var_23210_cast_fp16)[name = tensor("op_23388_cast_fp16")]; + tensor var_23395_begin_0 = const()[name = tensor("op_23395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23395_end_0 = const()[name = tensor("op_23395_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23395_end_mask_0 = const()[name = tensor("op_23395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23395_cast_fp16 = slice_by_index(begin = var_23395_begin_0, end = var_23395_end_0, end_mask = var_23395_end_mask_0, x = var_23214_cast_fp16)[name = tensor("op_23395_cast_fp16")]; + tensor var_23402_begin_0 = const()[name = tensor("op_23402_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23402_end_0 = const()[name = tensor("op_23402_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23402_end_mask_0 = const()[name = tensor("op_23402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23402_cast_fp16 = slice_by_index(begin = var_23402_begin_0, end = var_23402_end_0, end_mask = var_23402_end_mask_0, x = var_23214_cast_fp16)[name = tensor("op_23402_cast_fp16")]; + tensor var_23409_begin_0 = const()[name = tensor("op_23409_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23409_end_0 = const()[name = tensor("op_23409_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23409_end_mask_0 = const()[name = tensor("op_23409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23409_cast_fp16 = slice_by_index(begin = var_23409_begin_0, end = var_23409_end_0, end_mask = var_23409_end_mask_0, x = var_23214_cast_fp16)[name = tensor("op_23409_cast_fp16")]; + tensor var_23416_begin_0 = const()[name = tensor("op_23416_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23416_end_0 = const()[name = tensor("op_23416_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23416_end_mask_0 = const()[name = tensor("op_23416_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23416_cast_fp16 = slice_by_index(begin = var_23416_begin_0, end = var_23416_end_0, end_mask = var_23416_end_mask_0, x = var_23214_cast_fp16)[name = tensor("op_23416_cast_fp16")]; + tensor var_23423_begin_0 = const()[name = tensor("op_23423_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23423_end_0 = const()[name = tensor("op_23423_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23423_end_mask_0 = const()[name = tensor("op_23423_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23423_cast_fp16 = slice_by_index(begin = var_23423_begin_0, end = var_23423_end_0, end_mask = var_23423_end_mask_0, x = var_23218_cast_fp16)[name = tensor("op_23423_cast_fp16")]; + tensor var_23430_begin_0 = const()[name = tensor("op_23430_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23430_end_0 = const()[name = tensor("op_23430_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23430_end_mask_0 = const()[name = tensor("op_23430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23430_cast_fp16 = slice_by_index(begin = var_23430_begin_0, end = var_23430_end_0, end_mask = var_23430_end_mask_0, x = var_23218_cast_fp16)[name = tensor("op_23430_cast_fp16")]; + tensor var_23437_begin_0 = const()[name = tensor("op_23437_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23437_end_0 = const()[name = tensor("op_23437_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23437_end_mask_0 = const()[name = tensor("op_23437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23437_cast_fp16 = slice_by_index(begin = var_23437_begin_0, end = var_23437_end_0, end_mask = var_23437_end_mask_0, x = var_23218_cast_fp16)[name = tensor("op_23437_cast_fp16")]; + tensor var_23444_begin_0 = const()[name = tensor("op_23444_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23444_end_0 = const()[name = tensor("op_23444_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23444_end_mask_0 = const()[name = tensor("op_23444_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23444_cast_fp16 = slice_by_index(begin = var_23444_begin_0, end = var_23444_end_0, end_mask = var_23444_end_mask_0, x = var_23218_cast_fp16)[name = tensor("op_23444_cast_fp16")]; + tensor var_23451_begin_0 = const()[name = tensor("op_23451_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23451_end_0 = const()[name = tensor("op_23451_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23451_end_mask_0 = const()[name = tensor("op_23451_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23451_cast_fp16 = slice_by_index(begin = var_23451_begin_0, end = var_23451_end_0, end_mask = var_23451_end_mask_0, x = var_23222_cast_fp16)[name = tensor("op_23451_cast_fp16")]; + tensor var_23458_begin_0 = const()[name = tensor("op_23458_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23458_end_0 = const()[name = tensor("op_23458_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23458_end_mask_0 = const()[name = tensor("op_23458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23458_cast_fp16 = slice_by_index(begin = var_23458_begin_0, end = var_23458_end_0, end_mask = var_23458_end_mask_0, x = var_23222_cast_fp16)[name = tensor("op_23458_cast_fp16")]; + tensor var_23465_begin_0 = const()[name = tensor("op_23465_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23465_end_0 = const()[name = tensor("op_23465_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23465_end_mask_0 = const()[name = tensor("op_23465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23465_cast_fp16 = slice_by_index(begin = var_23465_begin_0, end = var_23465_end_0, end_mask = var_23465_end_mask_0, x = var_23222_cast_fp16)[name = tensor("op_23465_cast_fp16")]; + tensor var_23472_begin_0 = const()[name = tensor("op_23472_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23472_end_0 = const()[name = tensor("op_23472_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23472_end_mask_0 = const()[name = tensor("op_23472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23472_cast_fp16 = slice_by_index(begin = var_23472_begin_0, end = var_23472_end_0, end_mask = var_23472_end_mask_0, x = var_23222_cast_fp16)[name = tensor("op_23472_cast_fp16")]; + tensor var_23479_begin_0 = const()[name = tensor("op_23479_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23479_end_0 = const()[name = tensor("op_23479_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23479_end_mask_0 = const()[name = tensor("op_23479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23479_cast_fp16 = slice_by_index(begin = var_23479_begin_0, end = var_23479_end_0, end_mask = var_23479_end_mask_0, x = var_23226_cast_fp16)[name = tensor("op_23479_cast_fp16")]; + tensor var_23486_begin_0 = const()[name = tensor("op_23486_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23486_end_0 = const()[name = tensor("op_23486_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23486_end_mask_0 = const()[name = tensor("op_23486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23486_cast_fp16 = slice_by_index(begin = var_23486_begin_0, end = var_23486_end_0, end_mask = var_23486_end_mask_0, x = var_23226_cast_fp16)[name = tensor("op_23486_cast_fp16")]; + tensor var_23493_begin_0 = const()[name = tensor("op_23493_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23493_end_0 = const()[name = tensor("op_23493_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23493_end_mask_0 = const()[name = tensor("op_23493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23493_cast_fp16 = slice_by_index(begin = var_23493_begin_0, end = var_23493_end_0, end_mask = var_23493_end_mask_0, x = var_23226_cast_fp16)[name = tensor("op_23493_cast_fp16")]; + tensor var_23500_begin_0 = const()[name = tensor("op_23500_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23500_end_0 = const()[name = tensor("op_23500_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23500_end_mask_0 = const()[name = tensor("op_23500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23500_cast_fp16 = slice_by_index(begin = var_23500_begin_0, end = var_23500_end_0, end_mask = var_23500_end_mask_0, x = var_23226_cast_fp16)[name = tensor("op_23500_cast_fp16")]; + tensor var_23507_begin_0 = const()[name = tensor("op_23507_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23507_end_0 = const()[name = tensor("op_23507_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23507_end_mask_0 = const()[name = tensor("op_23507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23507_cast_fp16 = slice_by_index(begin = var_23507_begin_0, end = var_23507_end_0, end_mask = var_23507_end_mask_0, x = var_23230_cast_fp16)[name = tensor("op_23507_cast_fp16")]; + tensor var_23514_begin_0 = const()[name = tensor("op_23514_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23514_end_0 = const()[name = tensor("op_23514_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23514_end_mask_0 = const()[name = tensor("op_23514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23514_cast_fp16 = slice_by_index(begin = var_23514_begin_0, end = var_23514_end_0, end_mask = var_23514_end_mask_0, x = var_23230_cast_fp16)[name = tensor("op_23514_cast_fp16")]; + tensor var_23521_begin_0 = const()[name = tensor("op_23521_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23521_end_0 = const()[name = tensor("op_23521_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23521_end_mask_0 = const()[name = tensor("op_23521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23521_cast_fp16 = slice_by_index(begin = var_23521_begin_0, end = var_23521_end_0, end_mask = var_23521_end_mask_0, x = var_23230_cast_fp16)[name = tensor("op_23521_cast_fp16")]; + tensor var_23528_begin_0 = const()[name = tensor("op_23528_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23528_end_0 = const()[name = tensor("op_23528_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23528_end_mask_0 = const()[name = tensor("op_23528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23528_cast_fp16 = slice_by_index(begin = var_23528_begin_0, end = var_23528_end_0, end_mask = var_23528_end_mask_0, x = var_23230_cast_fp16)[name = tensor("op_23528_cast_fp16")]; + tensor var_23535_begin_0 = const()[name = tensor("op_23535_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23535_end_0 = const()[name = tensor("op_23535_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23535_end_mask_0 = const()[name = tensor("op_23535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23535_cast_fp16 = slice_by_index(begin = var_23535_begin_0, end = var_23535_end_0, end_mask = var_23535_end_mask_0, x = var_23234_cast_fp16)[name = tensor("op_23535_cast_fp16")]; + tensor var_23542_begin_0 = const()[name = tensor("op_23542_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23542_end_0 = const()[name = tensor("op_23542_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23542_end_mask_0 = const()[name = tensor("op_23542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23542_cast_fp16 = slice_by_index(begin = var_23542_begin_0, end = var_23542_end_0, end_mask = var_23542_end_mask_0, x = var_23234_cast_fp16)[name = tensor("op_23542_cast_fp16")]; + tensor var_23549_begin_0 = const()[name = tensor("op_23549_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23549_end_0 = const()[name = tensor("op_23549_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23549_end_mask_0 = const()[name = tensor("op_23549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23549_cast_fp16 = slice_by_index(begin = var_23549_begin_0, end = var_23549_end_0, end_mask = var_23549_end_mask_0, x = var_23234_cast_fp16)[name = tensor("op_23549_cast_fp16")]; + tensor var_23556_begin_0 = const()[name = tensor("op_23556_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23556_end_0 = const()[name = tensor("op_23556_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23556_end_mask_0 = const()[name = tensor("op_23556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23556_cast_fp16 = slice_by_index(begin = var_23556_begin_0, end = var_23556_end_0, end_mask = var_23556_end_mask_0, x = var_23234_cast_fp16)[name = tensor("op_23556_cast_fp16")]; + tensor var_23563_begin_0 = const()[name = tensor("op_23563_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23563_end_0 = const()[name = tensor("op_23563_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23563_end_mask_0 = const()[name = tensor("op_23563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23563_cast_fp16 = slice_by_index(begin = var_23563_begin_0, end = var_23563_end_0, end_mask = var_23563_end_mask_0, x = var_23238_cast_fp16)[name = tensor("op_23563_cast_fp16")]; + tensor var_23570_begin_0 = const()[name = tensor("op_23570_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23570_end_0 = const()[name = tensor("op_23570_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23570_end_mask_0 = const()[name = tensor("op_23570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23570_cast_fp16 = slice_by_index(begin = var_23570_begin_0, end = var_23570_end_0, end_mask = var_23570_end_mask_0, x = var_23238_cast_fp16)[name = tensor("op_23570_cast_fp16")]; + tensor var_23577_begin_0 = const()[name = tensor("op_23577_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23577_end_0 = const()[name = tensor("op_23577_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23577_end_mask_0 = const()[name = tensor("op_23577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23577_cast_fp16 = slice_by_index(begin = var_23577_begin_0, end = var_23577_end_0, end_mask = var_23577_end_mask_0, x = var_23238_cast_fp16)[name = tensor("op_23577_cast_fp16")]; + tensor var_23584_begin_0 = const()[name = tensor("op_23584_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23584_end_0 = const()[name = tensor("op_23584_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23584_end_mask_0 = const()[name = tensor("op_23584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23584_cast_fp16 = slice_by_index(begin = var_23584_begin_0, end = var_23584_end_0, end_mask = var_23584_end_mask_0, x = var_23238_cast_fp16)[name = tensor("op_23584_cast_fp16")]; + tensor var_23591_begin_0 = const()[name = tensor("op_23591_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23591_end_0 = const()[name = tensor("op_23591_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23591_end_mask_0 = const()[name = tensor("op_23591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23591_cast_fp16 = slice_by_index(begin = var_23591_begin_0, end = var_23591_end_0, end_mask = var_23591_end_mask_0, x = var_23242_cast_fp16)[name = tensor("op_23591_cast_fp16")]; + tensor var_23598_begin_0 = const()[name = tensor("op_23598_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23598_end_0 = const()[name = tensor("op_23598_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23598_end_mask_0 = const()[name = tensor("op_23598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23598_cast_fp16 = slice_by_index(begin = var_23598_begin_0, end = var_23598_end_0, end_mask = var_23598_end_mask_0, x = var_23242_cast_fp16)[name = tensor("op_23598_cast_fp16")]; + tensor var_23605_begin_0 = const()[name = tensor("op_23605_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23605_end_0 = const()[name = tensor("op_23605_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23605_end_mask_0 = const()[name = tensor("op_23605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23605_cast_fp16 = slice_by_index(begin = var_23605_begin_0, end = var_23605_end_0, end_mask = var_23605_end_mask_0, x = var_23242_cast_fp16)[name = tensor("op_23605_cast_fp16")]; + tensor var_23612_begin_0 = const()[name = tensor("op_23612_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23612_end_0 = const()[name = tensor("op_23612_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23612_end_mask_0 = const()[name = tensor("op_23612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23612_cast_fp16 = slice_by_index(begin = var_23612_begin_0, end = var_23612_end_0, end_mask = var_23612_end_mask_0, x = var_23242_cast_fp16)[name = tensor("op_23612_cast_fp16")]; + tensor var_23619_begin_0 = const()[name = tensor("op_23619_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23619_end_0 = const()[name = tensor("op_23619_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23619_end_mask_0 = const()[name = tensor("op_23619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23619_cast_fp16 = slice_by_index(begin = var_23619_begin_0, end = var_23619_end_0, end_mask = var_23619_end_mask_0, x = var_23246_cast_fp16)[name = tensor("op_23619_cast_fp16")]; + tensor var_23626_begin_0 = const()[name = tensor("op_23626_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23626_end_0 = const()[name = tensor("op_23626_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23626_end_mask_0 = const()[name = tensor("op_23626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23626_cast_fp16 = slice_by_index(begin = var_23626_begin_0, end = var_23626_end_0, end_mask = var_23626_end_mask_0, x = var_23246_cast_fp16)[name = tensor("op_23626_cast_fp16")]; + tensor var_23633_begin_0 = const()[name = tensor("op_23633_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23633_end_0 = const()[name = tensor("op_23633_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23633_end_mask_0 = const()[name = tensor("op_23633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23633_cast_fp16 = slice_by_index(begin = var_23633_begin_0, end = var_23633_end_0, end_mask = var_23633_end_mask_0, x = var_23246_cast_fp16)[name = tensor("op_23633_cast_fp16")]; + tensor var_23640_begin_0 = const()[name = tensor("op_23640_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23640_end_0 = const()[name = tensor("op_23640_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23640_end_mask_0 = const()[name = tensor("op_23640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23640_cast_fp16 = slice_by_index(begin = var_23640_begin_0, end = var_23640_end_0, end_mask = var_23640_end_mask_0, x = var_23246_cast_fp16)[name = tensor("op_23640_cast_fp16")]; + tensor var_23647_begin_0 = const()[name = tensor("op_23647_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23647_end_0 = const()[name = tensor("op_23647_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23647_end_mask_0 = const()[name = tensor("op_23647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23647_cast_fp16 = slice_by_index(begin = var_23647_begin_0, end = var_23647_end_0, end_mask = var_23647_end_mask_0, x = var_23250_cast_fp16)[name = tensor("op_23647_cast_fp16")]; + tensor var_23654_begin_0 = const()[name = tensor("op_23654_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23654_end_0 = const()[name = tensor("op_23654_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23654_end_mask_0 = const()[name = tensor("op_23654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23654_cast_fp16 = slice_by_index(begin = var_23654_begin_0, end = var_23654_end_0, end_mask = var_23654_end_mask_0, x = var_23250_cast_fp16)[name = tensor("op_23654_cast_fp16")]; + tensor var_23661_begin_0 = const()[name = tensor("op_23661_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23661_end_0 = const()[name = tensor("op_23661_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23661_end_mask_0 = const()[name = tensor("op_23661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23661_cast_fp16 = slice_by_index(begin = var_23661_begin_0, end = var_23661_end_0, end_mask = var_23661_end_mask_0, x = var_23250_cast_fp16)[name = tensor("op_23661_cast_fp16")]; + tensor var_23668_begin_0 = const()[name = tensor("op_23668_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23668_end_0 = const()[name = tensor("op_23668_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23668_end_mask_0 = const()[name = tensor("op_23668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23668_cast_fp16 = slice_by_index(begin = var_23668_begin_0, end = var_23668_end_0, end_mask = var_23668_end_mask_0, x = var_23250_cast_fp16)[name = tensor("op_23668_cast_fp16")]; + tensor var_23675_begin_0 = const()[name = tensor("op_23675_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23675_end_0 = const()[name = tensor("op_23675_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23675_end_mask_0 = const()[name = tensor("op_23675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23675_cast_fp16 = slice_by_index(begin = var_23675_begin_0, end = var_23675_end_0, end_mask = var_23675_end_mask_0, x = var_23254_cast_fp16)[name = tensor("op_23675_cast_fp16")]; + tensor var_23682_begin_0 = const()[name = tensor("op_23682_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23682_end_0 = const()[name = tensor("op_23682_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23682_end_mask_0 = const()[name = tensor("op_23682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23682_cast_fp16 = slice_by_index(begin = var_23682_begin_0, end = var_23682_end_0, end_mask = var_23682_end_mask_0, x = var_23254_cast_fp16)[name = tensor("op_23682_cast_fp16")]; + tensor var_23689_begin_0 = const()[name = tensor("op_23689_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23689_end_0 = const()[name = tensor("op_23689_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23689_end_mask_0 = const()[name = tensor("op_23689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23689_cast_fp16 = slice_by_index(begin = var_23689_begin_0, end = var_23689_end_0, end_mask = var_23689_end_mask_0, x = var_23254_cast_fp16)[name = tensor("op_23689_cast_fp16")]; + tensor var_23696_begin_0 = const()[name = tensor("op_23696_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23696_end_0 = const()[name = tensor("op_23696_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23696_end_mask_0 = const()[name = tensor("op_23696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23696_cast_fp16 = slice_by_index(begin = var_23696_begin_0, end = var_23696_end_0, end_mask = var_23696_end_mask_0, x = var_23254_cast_fp16)[name = tensor("op_23696_cast_fp16")]; + tensor var_23703_begin_0 = const()[name = tensor("op_23703_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23703_end_0 = const()[name = tensor("op_23703_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23703_end_mask_0 = const()[name = tensor("op_23703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23703_cast_fp16 = slice_by_index(begin = var_23703_begin_0, end = var_23703_end_0, end_mask = var_23703_end_mask_0, x = var_23258_cast_fp16)[name = tensor("op_23703_cast_fp16")]; + tensor var_23710_begin_0 = const()[name = tensor("op_23710_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23710_end_0 = const()[name = tensor("op_23710_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23710_end_mask_0 = const()[name = tensor("op_23710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23710_cast_fp16 = slice_by_index(begin = var_23710_begin_0, end = var_23710_end_0, end_mask = var_23710_end_mask_0, x = var_23258_cast_fp16)[name = tensor("op_23710_cast_fp16")]; + tensor var_23717_begin_0 = const()[name = tensor("op_23717_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23717_end_0 = const()[name = tensor("op_23717_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23717_end_mask_0 = const()[name = tensor("op_23717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23717_cast_fp16 = slice_by_index(begin = var_23717_begin_0, end = var_23717_end_0, end_mask = var_23717_end_mask_0, x = var_23258_cast_fp16)[name = tensor("op_23717_cast_fp16")]; + tensor var_23724_begin_0 = const()[name = tensor("op_23724_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23724_end_0 = const()[name = tensor("op_23724_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23724_end_mask_0 = const()[name = tensor("op_23724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23724_cast_fp16 = slice_by_index(begin = var_23724_begin_0, end = var_23724_end_0, end_mask = var_23724_end_mask_0, x = var_23258_cast_fp16)[name = tensor("op_23724_cast_fp16")]; + tensor var_23731_begin_0 = const()[name = tensor("op_23731_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23731_end_0 = const()[name = tensor("op_23731_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23731_end_mask_0 = const()[name = tensor("op_23731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23731_cast_fp16 = slice_by_index(begin = var_23731_begin_0, end = var_23731_end_0, end_mask = var_23731_end_mask_0, x = var_23262_cast_fp16)[name = tensor("op_23731_cast_fp16")]; + tensor var_23738_begin_0 = const()[name = tensor("op_23738_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23738_end_0 = const()[name = tensor("op_23738_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23738_end_mask_0 = const()[name = tensor("op_23738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23738_cast_fp16 = slice_by_index(begin = var_23738_begin_0, end = var_23738_end_0, end_mask = var_23738_end_mask_0, x = var_23262_cast_fp16)[name = tensor("op_23738_cast_fp16")]; + tensor var_23745_begin_0 = const()[name = tensor("op_23745_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23745_end_0 = const()[name = tensor("op_23745_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23745_end_mask_0 = const()[name = tensor("op_23745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23745_cast_fp16 = slice_by_index(begin = var_23745_begin_0, end = var_23745_end_0, end_mask = var_23745_end_mask_0, x = var_23262_cast_fp16)[name = tensor("op_23745_cast_fp16")]; + tensor var_23752_begin_0 = const()[name = tensor("op_23752_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23752_end_0 = const()[name = tensor("op_23752_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23752_end_mask_0 = const()[name = tensor("op_23752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23752_cast_fp16 = slice_by_index(begin = var_23752_begin_0, end = var_23752_end_0, end_mask = var_23752_end_mask_0, x = var_23262_cast_fp16)[name = tensor("op_23752_cast_fp16")]; + tensor var_23759_begin_0 = const()[name = tensor("op_23759_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23759_end_0 = const()[name = tensor("op_23759_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23759_end_mask_0 = const()[name = tensor("op_23759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23759_cast_fp16 = slice_by_index(begin = var_23759_begin_0, end = var_23759_end_0, end_mask = var_23759_end_mask_0, x = var_23266_cast_fp16)[name = tensor("op_23759_cast_fp16")]; + tensor var_23766_begin_0 = const()[name = tensor("op_23766_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23766_end_0 = const()[name = tensor("op_23766_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23766_end_mask_0 = const()[name = tensor("op_23766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23766_cast_fp16 = slice_by_index(begin = var_23766_begin_0, end = var_23766_end_0, end_mask = var_23766_end_mask_0, x = var_23266_cast_fp16)[name = tensor("op_23766_cast_fp16")]; + tensor var_23773_begin_0 = const()[name = tensor("op_23773_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23773_end_0 = const()[name = tensor("op_23773_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23773_end_mask_0 = const()[name = tensor("op_23773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23773_cast_fp16 = slice_by_index(begin = var_23773_begin_0, end = var_23773_end_0, end_mask = var_23773_end_mask_0, x = var_23266_cast_fp16)[name = tensor("op_23773_cast_fp16")]; + tensor var_23780_begin_0 = const()[name = tensor("op_23780_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23780_end_0 = const()[name = tensor("op_23780_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23780_end_mask_0 = const()[name = tensor("op_23780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23780_cast_fp16 = slice_by_index(begin = var_23780_begin_0, end = var_23780_end_0, end_mask = var_23780_end_mask_0, x = var_23266_cast_fp16)[name = tensor("op_23780_cast_fp16")]; + tensor var_23787_begin_0 = const()[name = tensor("op_23787_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23787_end_0 = const()[name = tensor("op_23787_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23787_end_mask_0 = const()[name = tensor("op_23787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23787_cast_fp16 = slice_by_index(begin = var_23787_begin_0, end = var_23787_end_0, end_mask = var_23787_end_mask_0, x = var_23270_cast_fp16)[name = tensor("op_23787_cast_fp16")]; + tensor var_23794_begin_0 = const()[name = tensor("op_23794_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23794_end_0 = const()[name = tensor("op_23794_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23794_end_mask_0 = const()[name = tensor("op_23794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23794_cast_fp16 = slice_by_index(begin = var_23794_begin_0, end = var_23794_end_0, end_mask = var_23794_end_mask_0, x = var_23270_cast_fp16)[name = tensor("op_23794_cast_fp16")]; + tensor var_23801_begin_0 = const()[name = tensor("op_23801_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23801_end_0 = const()[name = tensor("op_23801_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23801_end_mask_0 = const()[name = tensor("op_23801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23801_cast_fp16 = slice_by_index(begin = var_23801_begin_0, end = var_23801_end_0, end_mask = var_23801_end_mask_0, x = var_23270_cast_fp16)[name = tensor("op_23801_cast_fp16")]; + tensor var_23808_begin_0 = const()[name = tensor("op_23808_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23808_end_0 = const()[name = tensor("op_23808_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23808_end_mask_0 = const()[name = tensor("op_23808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23808_cast_fp16 = slice_by_index(begin = var_23808_begin_0, end = var_23808_end_0, end_mask = var_23808_end_mask_0, x = var_23270_cast_fp16)[name = tensor("op_23808_cast_fp16")]; + tensor var_23815_begin_0 = const()[name = tensor("op_23815_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23815_end_0 = const()[name = tensor("op_23815_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_23815_end_mask_0 = const()[name = tensor("op_23815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23815_cast_fp16 = slice_by_index(begin = var_23815_begin_0, end = var_23815_end_0, end_mask = var_23815_end_mask_0, x = var_23274_cast_fp16)[name = tensor("op_23815_cast_fp16")]; + tensor var_23822_begin_0 = const()[name = tensor("op_23822_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_23822_end_0 = const()[name = tensor("op_23822_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_23822_end_mask_0 = const()[name = tensor("op_23822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23822_cast_fp16 = slice_by_index(begin = var_23822_begin_0, end = var_23822_end_0, end_mask = var_23822_end_mask_0, x = var_23274_cast_fp16)[name = tensor("op_23822_cast_fp16")]; + tensor var_23829_begin_0 = const()[name = tensor("op_23829_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_23829_end_0 = const()[name = tensor("op_23829_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_23829_end_mask_0 = const()[name = tensor("op_23829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23829_cast_fp16 = slice_by_index(begin = var_23829_begin_0, end = var_23829_end_0, end_mask = var_23829_end_mask_0, x = var_23274_cast_fp16)[name = tensor("op_23829_cast_fp16")]; + tensor var_23836_begin_0 = const()[name = tensor("op_23836_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_23836_end_0 = const()[name = tensor("op_23836_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23836_end_mask_0 = const()[name = tensor("op_23836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23836_cast_fp16 = slice_by_index(begin = var_23836_begin_0, end = var_23836_end_0, end_mask = var_23836_end_mask_0, x = var_23274_cast_fp16)[name = tensor("op_23836_cast_fp16")]; + tensor k_29_perm_0 = const()[name = tensor("k_29_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_23841_begin_0 = const()[name = tensor("op_23841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23841_end_0 = const()[name = tensor("op_23841_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_23841_end_mask_0 = const()[name = tensor("op_23841_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_17 = transpose(perm = k_29_perm_0, x = key_29_cast_fp16)[name = tensor("transpose_17")]; + tensor var_23841_cast_fp16 = slice_by_index(begin = var_23841_begin_0, end = var_23841_end_0, end_mask = var_23841_end_mask_0, x = transpose_17)[name = tensor("op_23841_cast_fp16")]; + tensor var_23845_begin_0 = const()[name = tensor("op_23845_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_23845_end_0 = const()[name = tensor("op_23845_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_23845_end_mask_0 = const()[name = tensor("op_23845_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23845_cast_fp16 = slice_by_index(begin = var_23845_begin_0, end = var_23845_end_0, end_mask = var_23845_end_mask_0, x = transpose_17)[name = tensor("op_23845_cast_fp16")]; + tensor var_23849_begin_0 = const()[name = tensor("op_23849_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_23849_end_0 = const()[name = tensor("op_23849_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_23849_end_mask_0 = const()[name = tensor("op_23849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23849_cast_fp16 = slice_by_index(begin = var_23849_begin_0, end = var_23849_end_0, end_mask = var_23849_end_mask_0, x = transpose_17)[name = tensor("op_23849_cast_fp16")]; + tensor var_23853_begin_0 = const()[name = tensor("op_23853_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_23853_end_0 = const()[name = tensor("op_23853_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_23853_end_mask_0 = const()[name = tensor("op_23853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23853_cast_fp16 = slice_by_index(begin = var_23853_begin_0, end = var_23853_end_0, end_mask = var_23853_end_mask_0, x = transpose_17)[name = tensor("op_23853_cast_fp16")]; + tensor var_23857_begin_0 = const()[name = tensor("op_23857_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_23857_end_0 = const()[name = tensor("op_23857_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_23857_end_mask_0 = const()[name = tensor("op_23857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23857_cast_fp16 = slice_by_index(begin = var_23857_begin_0, end = var_23857_end_0, end_mask = var_23857_end_mask_0, x = transpose_17)[name = tensor("op_23857_cast_fp16")]; + tensor var_23861_begin_0 = const()[name = tensor("op_23861_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_23861_end_0 = const()[name = tensor("op_23861_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_23861_end_mask_0 = const()[name = tensor("op_23861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23861_cast_fp16 = slice_by_index(begin = var_23861_begin_0, end = var_23861_end_0, end_mask = var_23861_end_mask_0, x = transpose_17)[name = tensor("op_23861_cast_fp16")]; + tensor var_23865_begin_0 = const()[name = tensor("op_23865_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_23865_end_0 = const()[name = tensor("op_23865_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_23865_end_mask_0 = const()[name = tensor("op_23865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23865_cast_fp16 = slice_by_index(begin = var_23865_begin_0, end = var_23865_end_0, end_mask = var_23865_end_mask_0, x = transpose_17)[name = tensor("op_23865_cast_fp16")]; + tensor var_23869_begin_0 = const()[name = tensor("op_23869_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_23869_end_0 = const()[name = tensor("op_23869_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_23869_end_mask_0 = const()[name = tensor("op_23869_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23869_cast_fp16 = slice_by_index(begin = var_23869_begin_0, end = var_23869_end_0, end_mask = var_23869_end_mask_0, x = transpose_17)[name = tensor("op_23869_cast_fp16")]; + tensor var_23873_begin_0 = const()[name = tensor("op_23873_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_23873_end_0 = const()[name = tensor("op_23873_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_23873_end_mask_0 = const()[name = tensor("op_23873_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23873_cast_fp16 = slice_by_index(begin = var_23873_begin_0, end = var_23873_end_0, end_mask = var_23873_end_mask_0, x = transpose_17)[name = tensor("op_23873_cast_fp16")]; + tensor var_23877_begin_0 = const()[name = tensor("op_23877_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_23877_end_0 = const()[name = tensor("op_23877_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_23877_end_mask_0 = const()[name = tensor("op_23877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23877_cast_fp16 = slice_by_index(begin = var_23877_begin_0, end = var_23877_end_0, end_mask = var_23877_end_mask_0, x = transpose_17)[name = tensor("op_23877_cast_fp16")]; + tensor var_23881_begin_0 = const()[name = tensor("op_23881_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_23881_end_0 = const()[name = tensor("op_23881_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_23881_end_mask_0 = const()[name = tensor("op_23881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23881_cast_fp16 = slice_by_index(begin = var_23881_begin_0, end = var_23881_end_0, end_mask = var_23881_end_mask_0, x = transpose_17)[name = tensor("op_23881_cast_fp16")]; + tensor var_23885_begin_0 = const()[name = tensor("op_23885_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_23885_end_0 = const()[name = tensor("op_23885_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_23885_end_mask_0 = const()[name = tensor("op_23885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23885_cast_fp16 = slice_by_index(begin = var_23885_begin_0, end = var_23885_end_0, end_mask = var_23885_end_mask_0, x = transpose_17)[name = tensor("op_23885_cast_fp16")]; + tensor var_23889_begin_0 = const()[name = tensor("op_23889_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_23889_end_0 = const()[name = tensor("op_23889_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_23889_end_mask_0 = const()[name = tensor("op_23889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23889_cast_fp16 = slice_by_index(begin = var_23889_begin_0, end = var_23889_end_0, end_mask = var_23889_end_mask_0, x = transpose_17)[name = tensor("op_23889_cast_fp16")]; + tensor var_23893_begin_0 = const()[name = tensor("op_23893_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_23893_end_0 = const()[name = tensor("op_23893_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_23893_end_mask_0 = const()[name = tensor("op_23893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23893_cast_fp16 = slice_by_index(begin = var_23893_begin_0, end = var_23893_end_0, end_mask = var_23893_end_mask_0, x = transpose_17)[name = tensor("op_23893_cast_fp16")]; + tensor var_23897_begin_0 = const()[name = tensor("op_23897_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_23897_end_0 = const()[name = tensor("op_23897_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_23897_end_mask_0 = const()[name = tensor("op_23897_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23897_cast_fp16 = slice_by_index(begin = var_23897_begin_0, end = var_23897_end_0, end_mask = var_23897_end_mask_0, x = transpose_17)[name = tensor("op_23897_cast_fp16")]; + tensor var_23901_begin_0 = const()[name = tensor("op_23901_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_23901_end_0 = const()[name = tensor("op_23901_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_23901_end_mask_0 = const()[name = tensor("op_23901_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23901_cast_fp16 = slice_by_index(begin = var_23901_begin_0, end = var_23901_end_0, end_mask = var_23901_end_mask_0, x = transpose_17)[name = tensor("op_23901_cast_fp16")]; + tensor var_23905_begin_0 = const()[name = tensor("op_23905_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_23905_end_0 = const()[name = tensor("op_23905_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_23905_end_mask_0 = const()[name = tensor("op_23905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23905_cast_fp16 = slice_by_index(begin = var_23905_begin_0, end = var_23905_end_0, end_mask = var_23905_end_mask_0, x = transpose_17)[name = tensor("op_23905_cast_fp16")]; + tensor var_23909_begin_0 = const()[name = tensor("op_23909_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_23909_end_0 = const()[name = tensor("op_23909_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_23909_end_mask_0 = const()[name = tensor("op_23909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23909_cast_fp16 = slice_by_index(begin = var_23909_begin_0, end = var_23909_end_0, end_mask = var_23909_end_mask_0, x = transpose_17)[name = tensor("op_23909_cast_fp16")]; + tensor var_23913_begin_0 = const()[name = tensor("op_23913_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_23913_end_0 = const()[name = tensor("op_23913_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_23913_end_mask_0 = const()[name = tensor("op_23913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23913_cast_fp16 = slice_by_index(begin = var_23913_begin_0, end = var_23913_end_0, end_mask = var_23913_end_mask_0, x = transpose_17)[name = tensor("op_23913_cast_fp16")]; + tensor var_23917_begin_0 = const()[name = tensor("op_23917_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_23917_end_0 = const()[name = tensor("op_23917_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_23917_end_mask_0 = const()[name = tensor("op_23917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_23917_cast_fp16 = slice_by_index(begin = var_23917_begin_0, end = var_23917_end_0, end_mask = var_23917_end_mask_0, x = transpose_17)[name = tensor("op_23917_cast_fp16")]; + tensor var_23919_begin_0 = const()[name = tensor("op_23919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_23919_end_0 = const()[name = tensor("op_23919_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_23919_end_mask_0 = const()[name = tensor("op_23919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23919_cast_fp16 = slice_by_index(begin = var_23919_begin_0, end = var_23919_end_0, end_mask = var_23919_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23919_cast_fp16")]; + tensor var_23923_begin_0 = const()[name = tensor("op_23923_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_23923_end_0 = const()[name = tensor("op_23923_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_23923_end_mask_0 = const()[name = tensor("op_23923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23923_cast_fp16 = slice_by_index(begin = var_23923_begin_0, end = var_23923_end_0, end_mask = var_23923_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23923_cast_fp16")]; + tensor var_23927_begin_0 = const()[name = tensor("op_23927_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_23927_end_0 = const()[name = tensor("op_23927_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_23927_end_mask_0 = const()[name = tensor("op_23927_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23927_cast_fp16 = slice_by_index(begin = var_23927_begin_0, end = var_23927_end_0, end_mask = var_23927_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23927_cast_fp16")]; + tensor var_23931_begin_0 = const()[name = tensor("op_23931_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_23931_end_0 = const()[name = tensor("op_23931_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_23931_end_mask_0 = const()[name = tensor("op_23931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23931_cast_fp16 = slice_by_index(begin = var_23931_begin_0, end = var_23931_end_0, end_mask = var_23931_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23931_cast_fp16")]; + tensor var_23935_begin_0 = const()[name = tensor("op_23935_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_23935_end_0 = const()[name = tensor("op_23935_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_23935_end_mask_0 = const()[name = tensor("op_23935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23935_cast_fp16 = slice_by_index(begin = var_23935_begin_0, end = var_23935_end_0, end_mask = var_23935_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23935_cast_fp16")]; + tensor var_23939_begin_0 = const()[name = tensor("op_23939_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_23939_end_0 = const()[name = tensor("op_23939_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_23939_end_mask_0 = const()[name = tensor("op_23939_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23939_cast_fp16 = slice_by_index(begin = var_23939_begin_0, end = var_23939_end_0, end_mask = var_23939_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23939_cast_fp16")]; + tensor var_23943_begin_0 = const()[name = tensor("op_23943_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_23943_end_0 = const()[name = tensor("op_23943_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_23943_end_mask_0 = const()[name = tensor("op_23943_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23943_cast_fp16 = slice_by_index(begin = var_23943_begin_0, end = var_23943_end_0, end_mask = var_23943_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23943_cast_fp16")]; + tensor var_23947_begin_0 = const()[name = tensor("op_23947_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_23947_end_0 = const()[name = tensor("op_23947_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_23947_end_mask_0 = const()[name = tensor("op_23947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23947_cast_fp16 = slice_by_index(begin = var_23947_begin_0, end = var_23947_end_0, end_mask = var_23947_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23947_cast_fp16")]; + tensor var_23951_begin_0 = const()[name = tensor("op_23951_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_23951_end_0 = const()[name = tensor("op_23951_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_23951_end_mask_0 = const()[name = tensor("op_23951_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23951_cast_fp16 = slice_by_index(begin = var_23951_begin_0, end = var_23951_end_0, end_mask = var_23951_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23951_cast_fp16")]; + tensor var_23955_begin_0 = const()[name = tensor("op_23955_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_23955_end_0 = const()[name = tensor("op_23955_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_23955_end_mask_0 = const()[name = tensor("op_23955_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23955_cast_fp16 = slice_by_index(begin = var_23955_begin_0, end = var_23955_end_0, end_mask = var_23955_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23955_cast_fp16")]; + tensor var_23959_begin_0 = const()[name = tensor("op_23959_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_23959_end_0 = const()[name = tensor("op_23959_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_23959_end_mask_0 = const()[name = tensor("op_23959_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23959_cast_fp16 = slice_by_index(begin = var_23959_begin_0, end = var_23959_end_0, end_mask = var_23959_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23959_cast_fp16")]; + tensor var_23963_begin_0 = const()[name = tensor("op_23963_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_23963_end_0 = const()[name = tensor("op_23963_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_23963_end_mask_0 = const()[name = tensor("op_23963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23963_cast_fp16 = slice_by_index(begin = var_23963_begin_0, end = var_23963_end_0, end_mask = var_23963_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23963_cast_fp16")]; + tensor var_23967_begin_0 = const()[name = tensor("op_23967_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_23967_end_0 = const()[name = tensor("op_23967_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_23967_end_mask_0 = const()[name = tensor("op_23967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23967_cast_fp16 = slice_by_index(begin = var_23967_begin_0, end = var_23967_end_0, end_mask = var_23967_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23967_cast_fp16")]; + tensor var_23971_begin_0 = const()[name = tensor("op_23971_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_23971_end_0 = const()[name = tensor("op_23971_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_23971_end_mask_0 = const()[name = tensor("op_23971_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23971_cast_fp16 = slice_by_index(begin = var_23971_begin_0, end = var_23971_end_0, end_mask = var_23971_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23971_cast_fp16")]; + tensor var_23975_begin_0 = const()[name = tensor("op_23975_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_23975_end_0 = const()[name = tensor("op_23975_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_23975_end_mask_0 = const()[name = tensor("op_23975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23975_cast_fp16 = slice_by_index(begin = var_23975_begin_0, end = var_23975_end_0, end_mask = var_23975_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23975_cast_fp16")]; + tensor var_23979_begin_0 = const()[name = tensor("op_23979_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_23979_end_0 = const()[name = tensor("op_23979_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_23979_end_mask_0 = const()[name = tensor("op_23979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23979_cast_fp16 = slice_by_index(begin = var_23979_begin_0, end = var_23979_end_0, end_mask = var_23979_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23979_cast_fp16")]; + tensor var_23983_begin_0 = const()[name = tensor("op_23983_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_23983_end_0 = const()[name = tensor("op_23983_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_23983_end_mask_0 = const()[name = tensor("op_23983_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23983_cast_fp16 = slice_by_index(begin = var_23983_begin_0, end = var_23983_end_0, end_mask = var_23983_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23983_cast_fp16")]; + tensor var_23987_begin_0 = const()[name = tensor("op_23987_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_23987_end_0 = const()[name = tensor("op_23987_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_23987_end_mask_0 = const()[name = tensor("op_23987_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23987_cast_fp16 = slice_by_index(begin = var_23987_begin_0, end = var_23987_end_0, end_mask = var_23987_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23987_cast_fp16")]; + tensor var_23991_begin_0 = const()[name = tensor("op_23991_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_23991_end_0 = const()[name = tensor("op_23991_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_23991_end_mask_0 = const()[name = tensor("op_23991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23991_cast_fp16 = slice_by_index(begin = var_23991_begin_0, end = var_23991_end_0, end_mask = var_23991_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23991_cast_fp16")]; + tensor var_23995_begin_0 = const()[name = tensor("op_23995_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_23995_end_0 = const()[name = tensor("op_23995_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_23995_end_mask_0 = const()[name = tensor("op_23995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_23995_cast_fp16 = slice_by_index(begin = var_23995_begin_0, end = var_23995_end_0, end_mask = var_23995_end_mask_0, x = value_29_cast_fp16)[name = tensor("op_23995_cast_fp16")]; + tensor var_23999_equation_0 = const()[name = tensor("op_23999_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_23999_cast_fp16 = einsum(equation = var_23999_equation_0, values = (var_23841_cast_fp16, var_23283_cast_fp16))[name = tensor("op_23999_cast_fp16")]; + tensor var_24000_to_fp16 = const()[name = tensor("op_24000_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2241_cast_fp16 = mul(x = var_23999_cast_fp16, y = var_24000_to_fp16)[name = tensor("aw_chunk_2241_cast_fp16")]; + tensor var_24003_equation_0 = const()[name = tensor("op_24003_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24003_cast_fp16 = einsum(equation = var_24003_equation_0, values = (var_23841_cast_fp16, var_23290_cast_fp16))[name = tensor("op_24003_cast_fp16")]; + tensor var_24004_to_fp16 = const()[name = tensor("op_24004_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2243_cast_fp16 = mul(x = var_24003_cast_fp16, y = var_24004_to_fp16)[name = tensor("aw_chunk_2243_cast_fp16")]; + tensor var_24007_equation_0 = const()[name = tensor("op_24007_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24007_cast_fp16 = einsum(equation = var_24007_equation_0, values = (var_23841_cast_fp16, var_23297_cast_fp16))[name = tensor("op_24007_cast_fp16")]; + tensor var_24008_to_fp16 = const()[name = tensor("op_24008_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2245_cast_fp16 = mul(x = var_24007_cast_fp16, y = var_24008_to_fp16)[name = tensor("aw_chunk_2245_cast_fp16")]; + tensor var_24011_equation_0 = const()[name = tensor("op_24011_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24011_cast_fp16 = einsum(equation = var_24011_equation_0, values = (var_23841_cast_fp16, var_23304_cast_fp16))[name = tensor("op_24011_cast_fp16")]; + tensor var_24012_to_fp16 = const()[name = tensor("op_24012_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2247_cast_fp16 = mul(x = var_24011_cast_fp16, y = var_24012_to_fp16)[name = tensor("aw_chunk_2247_cast_fp16")]; + tensor var_24015_equation_0 = const()[name = tensor("op_24015_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24015_cast_fp16 = einsum(equation = var_24015_equation_0, values = (var_23845_cast_fp16, var_23311_cast_fp16))[name = tensor("op_24015_cast_fp16")]; + tensor var_24016_to_fp16 = const()[name = tensor("op_24016_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2249_cast_fp16 = mul(x = var_24015_cast_fp16, y = var_24016_to_fp16)[name = tensor("aw_chunk_2249_cast_fp16")]; + tensor var_24019_equation_0 = const()[name = tensor("op_24019_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24019_cast_fp16 = einsum(equation = var_24019_equation_0, values = (var_23845_cast_fp16, var_23318_cast_fp16))[name = tensor("op_24019_cast_fp16")]; + tensor var_24020_to_fp16 = const()[name = tensor("op_24020_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2251_cast_fp16 = mul(x = var_24019_cast_fp16, y = var_24020_to_fp16)[name = tensor("aw_chunk_2251_cast_fp16")]; + tensor var_24023_equation_0 = const()[name = tensor("op_24023_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24023_cast_fp16 = einsum(equation = var_24023_equation_0, values = (var_23845_cast_fp16, var_23325_cast_fp16))[name = tensor("op_24023_cast_fp16")]; + tensor var_24024_to_fp16 = const()[name = tensor("op_24024_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2253_cast_fp16 = mul(x = var_24023_cast_fp16, y = var_24024_to_fp16)[name = tensor("aw_chunk_2253_cast_fp16")]; + tensor var_24027_equation_0 = const()[name = tensor("op_24027_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24027_cast_fp16 = einsum(equation = var_24027_equation_0, values = (var_23845_cast_fp16, var_23332_cast_fp16))[name = tensor("op_24027_cast_fp16")]; + tensor var_24028_to_fp16 = const()[name = tensor("op_24028_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2255_cast_fp16 = mul(x = var_24027_cast_fp16, y = var_24028_to_fp16)[name = tensor("aw_chunk_2255_cast_fp16")]; + tensor var_24031_equation_0 = const()[name = tensor("op_24031_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24031_cast_fp16 = einsum(equation = var_24031_equation_0, values = (var_23849_cast_fp16, var_23339_cast_fp16))[name = tensor("op_24031_cast_fp16")]; + tensor var_24032_to_fp16 = const()[name = tensor("op_24032_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2257_cast_fp16 = mul(x = var_24031_cast_fp16, y = var_24032_to_fp16)[name = tensor("aw_chunk_2257_cast_fp16")]; + tensor var_24035_equation_0 = const()[name = tensor("op_24035_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24035_cast_fp16 = einsum(equation = var_24035_equation_0, values = (var_23849_cast_fp16, var_23346_cast_fp16))[name = tensor("op_24035_cast_fp16")]; + tensor var_24036_to_fp16 = const()[name = tensor("op_24036_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2259_cast_fp16 = mul(x = var_24035_cast_fp16, y = var_24036_to_fp16)[name = tensor("aw_chunk_2259_cast_fp16")]; + tensor var_24039_equation_0 = const()[name = tensor("op_24039_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24039_cast_fp16 = einsum(equation = var_24039_equation_0, values = (var_23849_cast_fp16, var_23353_cast_fp16))[name = tensor("op_24039_cast_fp16")]; + tensor var_24040_to_fp16 = const()[name = tensor("op_24040_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2261_cast_fp16 = mul(x = var_24039_cast_fp16, y = var_24040_to_fp16)[name = tensor("aw_chunk_2261_cast_fp16")]; + tensor var_24043_equation_0 = const()[name = tensor("op_24043_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24043_cast_fp16 = einsum(equation = var_24043_equation_0, values = (var_23849_cast_fp16, var_23360_cast_fp16))[name = tensor("op_24043_cast_fp16")]; + tensor var_24044_to_fp16 = const()[name = tensor("op_24044_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2263_cast_fp16 = mul(x = var_24043_cast_fp16, y = var_24044_to_fp16)[name = tensor("aw_chunk_2263_cast_fp16")]; + tensor var_24047_equation_0 = const()[name = tensor("op_24047_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24047_cast_fp16 = einsum(equation = var_24047_equation_0, values = (var_23853_cast_fp16, var_23367_cast_fp16))[name = tensor("op_24047_cast_fp16")]; + tensor var_24048_to_fp16 = const()[name = tensor("op_24048_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2265_cast_fp16 = mul(x = var_24047_cast_fp16, y = var_24048_to_fp16)[name = tensor("aw_chunk_2265_cast_fp16")]; + tensor var_24051_equation_0 = const()[name = tensor("op_24051_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24051_cast_fp16 = einsum(equation = var_24051_equation_0, values = (var_23853_cast_fp16, var_23374_cast_fp16))[name = tensor("op_24051_cast_fp16")]; + tensor var_24052_to_fp16 = const()[name = tensor("op_24052_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2267_cast_fp16 = mul(x = var_24051_cast_fp16, y = var_24052_to_fp16)[name = tensor("aw_chunk_2267_cast_fp16")]; + tensor var_24055_equation_0 = const()[name = tensor("op_24055_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24055_cast_fp16 = einsum(equation = var_24055_equation_0, values = (var_23853_cast_fp16, var_23381_cast_fp16))[name = tensor("op_24055_cast_fp16")]; + tensor var_24056_to_fp16 = const()[name = tensor("op_24056_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2269_cast_fp16 = mul(x = var_24055_cast_fp16, y = var_24056_to_fp16)[name = tensor("aw_chunk_2269_cast_fp16")]; + tensor var_24059_equation_0 = const()[name = tensor("op_24059_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24059_cast_fp16 = einsum(equation = var_24059_equation_0, values = (var_23853_cast_fp16, var_23388_cast_fp16))[name = tensor("op_24059_cast_fp16")]; + tensor var_24060_to_fp16 = const()[name = tensor("op_24060_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2271_cast_fp16 = mul(x = var_24059_cast_fp16, y = var_24060_to_fp16)[name = tensor("aw_chunk_2271_cast_fp16")]; + tensor var_24063_equation_0 = const()[name = tensor("op_24063_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24063_cast_fp16 = einsum(equation = var_24063_equation_0, values = (var_23857_cast_fp16, var_23395_cast_fp16))[name = tensor("op_24063_cast_fp16")]; + tensor var_24064_to_fp16 = const()[name = tensor("op_24064_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2273_cast_fp16 = mul(x = var_24063_cast_fp16, y = var_24064_to_fp16)[name = tensor("aw_chunk_2273_cast_fp16")]; + tensor var_24067_equation_0 = const()[name = tensor("op_24067_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24067_cast_fp16 = einsum(equation = var_24067_equation_0, values = (var_23857_cast_fp16, var_23402_cast_fp16))[name = tensor("op_24067_cast_fp16")]; + tensor var_24068_to_fp16 = const()[name = tensor("op_24068_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2275_cast_fp16 = mul(x = var_24067_cast_fp16, y = var_24068_to_fp16)[name = tensor("aw_chunk_2275_cast_fp16")]; + tensor var_24071_equation_0 = const()[name = tensor("op_24071_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24071_cast_fp16 = einsum(equation = var_24071_equation_0, values = (var_23857_cast_fp16, var_23409_cast_fp16))[name = tensor("op_24071_cast_fp16")]; + tensor var_24072_to_fp16 = const()[name = tensor("op_24072_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2277_cast_fp16 = mul(x = var_24071_cast_fp16, y = var_24072_to_fp16)[name = tensor("aw_chunk_2277_cast_fp16")]; + tensor var_24075_equation_0 = const()[name = tensor("op_24075_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24075_cast_fp16 = einsum(equation = var_24075_equation_0, values = (var_23857_cast_fp16, var_23416_cast_fp16))[name = tensor("op_24075_cast_fp16")]; + tensor var_24076_to_fp16 = const()[name = tensor("op_24076_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2279_cast_fp16 = mul(x = var_24075_cast_fp16, y = var_24076_to_fp16)[name = tensor("aw_chunk_2279_cast_fp16")]; + tensor var_24079_equation_0 = const()[name = tensor("op_24079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24079_cast_fp16 = einsum(equation = var_24079_equation_0, values = (var_23861_cast_fp16, var_23423_cast_fp16))[name = tensor("op_24079_cast_fp16")]; + tensor var_24080_to_fp16 = const()[name = tensor("op_24080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2281_cast_fp16 = mul(x = var_24079_cast_fp16, y = var_24080_to_fp16)[name = tensor("aw_chunk_2281_cast_fp16")]; + tensor var_24083_equation_0 = const()[name = tensor("op_24083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24083_cast_fp16 = einsum(equation = var_24083_equation_0, values = (var_23861_cast_fp16, var_23430_cast_fp16))[name = tensor("op_24083_cast_fp16")]; + tensor var_24084_to_fp16 = const()[name = tensor("op_24084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2283_cast_fp16 = mul(x = var_24083_cast_fp16, y = var_24084_to_fp16)[name = tensor("aw_chunk_2283_cast_fp16")]; + tensor var_24087_equation_0 = const()[name = tensor("op_24087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24087_cast_fp16 = einsum(equation = var_24087_equation_0, values = (var_23861_cast_fp16, var_23437_cast_fp16))[name = tensor("op_24087_cast_fp16")]; + tensor var_24088_to_fp16 = const()[name = tensor("op_24088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2285_cast_fp16 = mul(x = var_24087_cast_fp16, y = var_24088_to_fp16)[name = tensor("aw_chunk_2285_cast_fp16")]; + tensor var_24091_equation_0 = const()[name = tensor("op_24091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24091_cast_fp16 = einsum(equation = var_24091_equation_0, values = (var_23861_cast_fp16, var_23444_cast_fp16))[name = tensor("op_24091_cast_fp16")]; + tensor var_24092_to_fp16 = const()[name = tensor("op_24092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2287_cast_fp16 = mul(x = var_24091_cast_fp16, y = var_24092_to_fp16)[name = tensor("aw_chunk_2287_cast_fp16")]; + tensor var_24095_equation_0 = const()[name = tensor("op_24095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24095_cast_fp16 = einsum(equation = var_24095_equation_0, values = (var_23865_cast_fp16, var_23451_cast_fp16))[name = tensor("op_24095_cast_fp16")]; + tensor var_24096_to_fp16 = const()[name = tensor("op_24096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2289_cast_fp16 = mul(x = var_24095_cast_fp16, y = var_24096_to_fp16)[name = tensor("aw_chunk_2289_cast_fp16")]; + tensor var_24099_equation_0 = const()[name = tensor("op_24099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24099_cast_fp16 = einsum(equation = var_24099_equation_0, values = (var_23865_cast_fp16, var_23458_cast_fp16))[name = tensor("op_24099_cast_fp16")]; + tensor var_24100_to_fp16 = const()[name = tensor("op_24100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2291_cast_fp16 = mul(x = var_24099_cast_fp16, y = var_24100_to_fp16)[name = tensor("aw_chunk_2291_cast_fp16")]; + tensor var_24103_equation_0 = const()[name = tensor("op_24103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24103_cast_fp16 = einsum(equation = var_24103_equation_0, values = (var_23865_cast_fp16, var_23465_cast_fp16))[name = tensor("op_24103_cast_fp16")]; + tensor var_24104_to_fp16 = const()[name = tensor("op_24104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2293_cast_fp16 = mul(x = var_24103_cast_fp16, y = var_24104_to_fp16)[name = tensor("aw_chunk_2293_cast_fp16")]; + tensor var_24107_equation_0 = const()[name = tensor("op_24107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24107_cast_fp16 = einsum(equation = var_24107_equation_0, values = (var_23865_cast_fp16, var_23472_cast_fp16))[name = tensor("op_24107_cast_fp16")]; + tensor var_24108_to_fp16 = const()[name = tensor("op_24108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2295_cast_fp16 = mul(x = var_24107_cast_fp16, y = var_24108_to_fp16)[name = tensor("aw_chunk_2295_cast_fp16")]; + tensor var_24111_equation_0 = const()[name = tensor("op_24111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24111_cast_fp16 = einsum(equation = var_24111_equation_0, values = (var_23869_cast_fp16, var_23479_cast_fp16))[name = tensor("op_24111_cast_fp16")]; + tensor var_24112_to_fp16 = const()[name = tensor("op_24112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2297_cast_fp16 = mul(x = var_24111_cast_fp16, y = var_24112_to_fp16)[name = tensor("aw_chunk_2297_cast_fp16")]; + tensor var_24115_equation_0 = const()[name = tensor("op_24115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24115_cast_fp16 = einsum(equation = var_24115_equation_0, values = (var_23869_cast_fp16, var_23486_cast_fp16))[name = tensor("op_24115_cast_fp16")]; + tensor var_24116_to_fp16 = const()[name = tensor("op_24116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2299_cast_fp16 = mul(x = var_24115_cast_fp16, y = var_24116_to_fp16)[name = tensor("aw_chunk_2299_cast_fp16")]; + tensor var_24119_equation_0 = const()[name = tensor("op_24119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24119_cast_fp16 = einsum(equation = var_24119_equation_0, values = (var_23869_cast_fp16, var_23493_cast_fp16))[name = tensor("op_24119_cast_fp16")]; + tensor var_24120_to_fp16 = const()[name = tensor("op_24120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2301_cast_fp16 = mul(x = var_24119_cast_fp16, y = var_24120_to_fp16)[name = tensor("aw_chunk_2301_cast_fp16")]; + tensor var_24123_equation_0 = const()[name = tensor("op_24123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24123_cast_fp16 = einsum(equation = var_24123_equation_0, values = (var_23869_cast_fp16, var_23500_cast_fp16))[name = tensor("op_24123_cast_fp16")]; + tensor var_24124_to_fp16 = const()[name = tensor("op_24124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2303_cast_fp16 = mul(x = var_24123_cast_fp16, y = var_24124_to_fp16)[name = tensor("aw_chunk_2303_cast_fp16")]; + tensor var_24127_equation_0 = const()[name = tensor("op_24127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24127_cast_fp16 = einsum(equation = var_24127_equation_0, values = (var_23873_cast_fp16, var_23507_cast_fp16))[name = tensor("op_24127_cast_fp16")]; + tensor var_24128_to_fp16 = const()[name = tensor("op_24128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2305_cast_fp16 = mul(x = var_24127_cast_fp16, y = var_24128_to_fp16)[name = tensor("aw_chunk_2305_cast_fp16")]; + tensor var_24131_equation_0 = const()[name = tensor("op_24131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24131_cast_fp16 = einsum(equation = var_24131_equation_0, values = (var_23873_cast_fp16, var_23514_cast_fp16))[name = tensor("op_24131_cast_fp16")]; + tensor var_24132_to_fp16 = const()[name = tensor("op_24132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2307_cast_fp16 = mul(x = var_24131_cast_fp16, y = var_24132_to_fp16)[name = tensor("aw_chunk_2307_cast_fp16")]; + tensor var_24135_equation_0 = const()[name = tensor("op_24135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24135_cast_fp16 = einsum(equation = var_24135_equation_0, values = (var_23873_cast_fp16, var_23521_cast_fp16))[name = tensor("op_24135_cast_fp16")]; + tensor var_24136_to_fp16 = const()[name = tensor("op_24136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2309_cast_fp16 = mul(x = var_24135_cast_fp16, y = var_24136_to_fp16)[name = tensor("aw_chunk_2309_cast_fp16")]; + tensor var_24139_equation_0 = const()[name = tensor("op_24139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24139_cast_fp16 = einsum(equation = var_24139_equation_0, values = (var_23873_cast_fp16, var_23528_cast_fp16))[name = tensor("op_24139_cast_fp16")]; + tensor var_24140_to_fp16 = const()[name = tensor("op_24140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2311_cast_fp16 = mul(x = var_24139_cast_fp16, y = var_24140_to_fp16)[name = tensor("aw_chunk_2311_cast_fp16")]; + tensor var_24143_equation_0 = const()[name = tensor("op_24143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24143_cast_fp16 = einsum(equation = var_24143_equation_0, values = (var_23877_cast_fp16, var_23535_cast_fp16))[name = tensor("op_24143_cast_fp16")]; + tensor var_24144_to_fp16 = const()[name = tensor("op_24144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2313_cast_fp16 = mul(x = var_24143_cast_fp16, y = var_24144_to_fp16)[name = tensor("aw_chunk_2313_cast_fp16")]; + tensor var_24147_equation_0 = const()[name = tensor("op_24147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24147_cast_fp16 = einsum(equation = var_24147_equation_0, values = (var_23877_cast_fp16, var_23542_cast_fp16))[name = tensor("op_24147_cast_fp16")]; + tensor var_24148_to_fp16 = const()[name = tensor("op_24148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2315_cast_fp16 = mul(x = var_24147_cast_fp16, y = var_24148_to_fp16)[name = tensor("aw_chunk_2315_cast_fp16")]; + tensor var_24151_equation_0 = const()[name = tensor("op_24151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24151_cast_fp16 = einsum(equation = var_24151_equation_0, values = (var_23877_cast_fp16, var_23549_cast_fp16))[name = tensor("op_24151_cast_fp16")]; + tensor var_24152_to_fp16 = const()[name = tensor("op_24152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2317_cast_fp16 = mul(x = var_24151_cast_fp16, y = var_24152_to_fp16)[name = tensor("aw_chunk_2317_cast_fp16")]; + tensor var_24155_equation_0 = const()[name = tensor("op_24155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24155_cast_fp16 = einsum(equation = var_24155_equation_0, values = (var_23877_cast_fp16, var_23556_cast_fp16))[name = tensor("op_24155_cast_fp16")]; + tensor var_24156_to_fp16 = const()[name = tensor("op_24156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2319_cast_fp16 = mul(x = var_24155_cast_fp16, y = var_24156_to_fp16)[name = tensor("aw_chunk_2319_cast_fp16")]; + tensor var_24159_equation_0 = const()[name = tensor("op_24159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24159_cast_fp16 = einsum(equation = var_24159_equation_0, values = (var_23881_cast_fp16, var_23563_cast_fp16))[name = tensor("op_24159_cast_fp16")]; + tensor var_24160_to_fp16 = const()[name = tensor("op_24160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2321_cast_fp16 = mul(x = var_24159_cast_fp16, y = var_24160_to_fp16)[name = tensor("aw_chunk_2321_cast_fp16")]; + tensor var_24163_equation_0 = const()[name = tensor("op_24163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24163_cast_fp16 = einsum(equation = var_24163_equation_0, values = (var_23881_cast_fp16, var_23570_cast_fp16))[name = tensor("op_24163_cast_fp16")]; + tensor var_24164_to_fp16 = const()[name = tensor("op_24164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2323_cast_fp16 = mul(x = var_24163_cast_fp16, y = var_24164_to_fp16)[name = tensor("aw_chunk_2323_cast_fp16")]; + tensor var_24167_equation_0 = const()[name = tensor("op_24167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24167_cast_fp16 = einsum(equation = var_24167_equation_0, values = (var_23881_cast_fp16, var_23577_cast_fp16))[name = tensor("op_24167_cast_fp16")]; + tensor var_24168_to_fp16 = const()[name = tensor("op_24168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2325_cast_fp16 = mul(x = var_24167_cast_fp16, y = var_24168_to_fp16)[name = tensor("aw_chunk_2325_cast_fp16")]; + tensor var_24171_equation_0 = const()[name = tensor("op_24171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24171_cast_fp16 = einsum(equation = var_24171_equation_0, values = (var_23881_cast_fp16, var_23584_cast_fp16))[name = tensor("op_24171_cast_fp16")]; + tensor var_24172_to_fp16 = const()[name = tensor("op_24172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2327_cast_fp16 = mul(x = var_24171_cast_fp16, y = var_24172_to_fp16)[name = tensor("aw_chunk_2327_cast_fp16")]; + tensor var_24175_equation_0 = const()[name = tensor("op_24175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24175_cast_fp16 = einsum(equation = var_24175_equation_0, values = (var_23885_cast_fp16, var_23591_cast_fp16))[name = tensor("op_24175_cast_fp16")]; + tensor var_24176_to_fp16 = const()[name = tensor("op_24176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2329_cast_fp16 = mul(x = var_24175_cast_fp16, y = var_24176_to_fp16)[name = tensor("aw_chunk_2329_cast_fp16")]; + tensor var_24179_equation_0 = const()[name = tensor("op_24179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24179_cast_fp16 = einsum(equation = var_24179_equation_0, values = (var_23885_cast_fp16, var_23598_cast_fp16))[name = tensor("op_24179_cast_fp16")]; + tensor var_24180_to_fp16 = const()[name = tensor("op_24180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2331_cast_fp16 = mul(x = var_24179_cast_fp16, y = var_24180_to_fp16)[name = tensor("aw_chunk_2331_cast_fp16")]; + tensor var_24183_equation_0 = const()[name = tensor("op_24183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24183_cast_fp16 = einsum(equation = var_24183_equation_0, values = (var_23885_cast_fp16, var_23605_cast_fp16))[name = tensor("op_24183_cast_fp16")]; + tensor var_24184_to_fp16 = const()[name = tensor("op_24184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2333_cast_fp16 = mul(x = var_24183_cast_fp16, y = var_24184_to_fp16)[name = tensor("aw_chunk_2333_cast_fp16")]; + tensor var_24187_equation_0 = const()[name = tensor("op_24187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24187_cast_fp16 = einsum(equation = var_24187_equation_0, values = (var_23885_cast_fp16, var_23612_cast_fp16))[name = tensor("op_24187_cast_fp16")]; + tensor var_24188_to_fp16 = const()[name = tensor("op_24188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2335_cast_fp16 = mul(x = var_24187_cast_fp16, y = var_24188_to_fp16)[name = tensor("aw_chunk_2335_cast_fp16")]; + tensor var_24191_equation_0 = const()[name = tensor("op_24191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24191_cast_fp16 = einsum(equation = var_24191_equation_0, values = (var_23889_cast_fp16, var_23619_cast_fp16))[name = tensor("op_24191_cast_fp16")]; + tensor var_24192_to_fp16 = const()[name = tensor("op_24192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2337_cast_fp16 = mul(x = var_24191_cast_fp16, y = var_24192_to_fp16)[name = tensor("aw_chunk_2337_cast_fp16")]; + tensor var_24195_equation_0 = const()[name = tensor("op_24195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24195_cast_fp16 = einsum(equation = var_24195_equation_0, values = (var_23889_cast_fp16, var_23626_cast_fp16))[name = tensor("op_24195_cast_fp16")]; + tensor var_24196_to_fp16 = const()[name = tensor("op_24196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2339_cast_fp16 = mul(x = var_24195_cast_fp16, y = var_24196_to_fp16)[name = tensor("aw_chunk_2339_cast_fp16")]; + tensor var_24199_equation_0 = const()[name = tensor("op_24199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24199_cast_fp16 = einsum(equation = var_24199_equation_0, values = (var_23889_cast_fp16, var_23633_cast_fp16))[name = tensor("op_24199_cast_fp16")]; + tensor var_24200_to_fp16 = const()[name = tensor("op_24200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2341_cast_fp16 = mul(x = var_24199_cast_fp16, y = var_24200_to_fp16)[name = tensor("aw_chunk_2341_cast_fp16")]; + tensor var_24203_equation_0 = const()[name = tensor("op_24203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24203_cast_fp16 = einsum(equation = var_24203_equation_0, values = (var_23889_cast_fp16, var_23640_cast_fp16))[name = tensor("op_24203_cast_fp16")]; + tensor var_24204_to_fp16 = const()[name = tensor("op_24204_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2343_cast_fp16 = mul(x = var_24203_cast_fp16, y = var_24204_to_fp16)[name = tensor("aw_chunk_2343_cast_fp16")]; + tensor var_24207_equation_0 = const()[name = tensor("op_24207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24207_cast_fp16 = einsum(equation = var_24207_equation_0, values = (var_23893_cast_fp16, var_23647_cast_fp16))[name = tensor("op_24207_cast_fp16")]; + tensor var_24208_to_fp16 = const()[name = tensor("op_24208_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2345_cast_fp16 = mul(x = var_24207_cast_fp16, y = var_24208_to_fp16)[name = tensor("aw_chunk_2345_cast_fp16")]; + tensor var_24211_equation_0 = const()[name = tensor("op_24211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24211_cast_fp16 = einsum(equation = var_24211_equation_0, values = (var_23893_cast_fp16, var_23654_cast_fp16))[name = tensor("op_24211_cast_fp16")]; + tensor var_24212_to_fp16 = const()[name = tensor("op_24212_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2347_cast_fp16 = mul(x = var_24211_cast_fp16, y = var_24212_to_fp16)[name = tensor("aw_chunk_2347_cast_fp16")]; + tensor var_24215_equation_0 = const()[name = tensor("op_24215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24215_cast_fp16 = einsum(equation = var_24215_equation_0, values = (var_23893_cast_fp16, var_23661_cast_fp16))[name = tensor("op_24215_cast_fp16")]; + tensor var_24216_to_fp16 = const()[name = tensor("op_24216_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2349_cast_fp16 = mul(x = var_24215_cast_fp16, y = var_24216_to_fp16)[name = tensor("aw_chunk_2349_cast_fp16")]; + tensor var_24219_equation_0 = const()[name = tensor("op_24219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24219_cast_fp16 = einsum(equation = var_24219_equation_0, values = (var_23893_cast_fp16, var_23668_cast_fp16))[name = tensor("op_24219_cast_fp16")]; + tensor var_24220_to_fp16 = const()[name = tensor("op_24220_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2351_cast_fp16 = mul(x = var_24219_cast_fp16, y = var_24220_to_fp16)[name = tensor("aw_chunk_2351_cast_fp16")]; + tensor var_24223_equation_0 = const()[name = tensor("op_24223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24223_cast_fp16 = einsum(equation = var_24223_equation_0, values = (var_23897_cast_fp16, var_23675_cast_fp16))[name = tensor("op_24223_cast_fp16")]; + tensor var_24224_to_fp16 = const()[name = tensor("op_24224_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2353_cast_fp16 = mul(x = var_24223_cast_fp16, y = var_24224_to_fp16)[name = tensor("aw_chunk_2353_cast_fp16")]; + tensor var_24227_equation_0 = const()[name = tensor("op_24227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24227_cast_fp16 = einsum(equation = var_24227_equation_0, values = (var_23897_cast_fp16, var_23682_cast_fp16))[name = tensor("op_24227_cast_fp16")]; + tensor var_24228_to_fp16 = const()[name = tensor("op_24228_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2355_cast_fp16 = mul(x = var_24227_cast_fp16, y = var_24228_to_fp16)[name = tensor("aw_chunk_2355_cast_fp16")]; + tensor var_24231_equation_0 = const()[name = tensor("op_24231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24231_cast_fp16 = einsum(equation = var_24231_equation_0, values = (var_23897_cast_fp16, var_23689_cast_fp16))[name = tensor("op_24231_cast_fp16")]; + tensor var_24232_to_fp16 = const()[name = tensor("op_24232_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2357_cast_fp16 = mul(x = var_24231_cast_fp16, y = var_24232_to_fp16)[name = tensor("aw_chunk_2357_cast_fp16")]; + tensor var_24235_equation_0 = const()[name = tensor("op_24235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24235_cast_fp16 = einsum(equation = var_24235_equation_0, values = (var_23897_cast_fp16, var_23696_cast_fp16))[name = tensor("op_24235_cast_fp16")]; + tensor var_24236_to_fp16 = const()[name = tensor("op_24236_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2359_cast_fp16 = mul(x = var_24235_cast_fp16, y = var_24236_to_fp16)[name = tensor("aw_chunk_2359_cast_fp16")]; + tensor var_24239_equation_0 = const()[name = tensor("op_24239_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24239_cast_fp16 = einsum(equation = var_24239_equation_0, values = (var_23901_cast_fp16, var_23703_cast_fp16))[name = tensor("op_24239_cast_fp16")]; + tensor var_24240_to_fp16 = const()[name = tensor("op_24240_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2361_cast_fp16 = mul(x = var_24239_cast_fp16, y = var_24240_to_fp16)[name = tensor("aw_chunk_2361_cast_fp16")]; + tensor var_24243_equation_0 = const()[name = tensor("op_24243_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24243_cast_fp16 = einsum(equation = var_24243_equation_0, values = (var_23901_cast_fp16, var_23710_cast_fp16))[name = tensor("op_24243_cast_fp16")]; + tensor var_24244_to_fp16 = const()[name = tensor("op_24244_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2363_cast_fp16 = mul(x = var_24243_cast_fp16, y = var_24244_to_fp16)[name = tensor("aw_chunk_2363_cast_fp16")]; + tensor var_24247_equation_0 = const()[name = tensor("op_24247_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24247_cast_fp16 = einsum(equation = var_24247_equation_0, values = (var_23901_cast_fp16, var_23717_cast_fp16))[name = tensor("op_24247_cast_fp16")]; + tensor var_24248_to_fp16 = const()[name = tensor("op_24248_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2365_cast_fp16 = mul(x = var_24247_cast_fp16, y = var_24248_to_fp16)[name = tensor("aw_chunk_2365_cast_fp16")]; + tensor var_24251_equation_0 = const()[name = tensor("op_24251_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24251_cast_fp16 = einsum(equation = var_24251_equation_0, values = (var_23901_cast_fp16, var_23724_cast_fp16))[name = tensor("op_24251_cast_fp16")]; + tensor var_24252_to_fp16 = const()[name = tensor("op_24252_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2367_cast_fp16 = mul(x = var_24251_cast_fp16, y = var_24252_to_fp16)[name = tensor("aw_chunk_2367_cast_fp16")]; + tensor var_24255_equation_0 = const()[name = tensor("op_24255_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24255_cast_fp16 = einsum(equation = var_24255_equation_0, values = (var_23905_cast_fp16, var_23731_cast_fp16))[name = tensor("op_24255_cast_fp16")]; + tensor var_24256_to_fp16 = const()[name = tensor("op_24256_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2369_cast_fp16 = mul(x = var_24255_cast_fp16, y = var_24256_to_fp16)[name = tensor("aw_chunk_2369_cast_fp16")]; + tensor var_24259_equation_0 = const()[name = tensor("op_24259_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24259_cast_fp16 = einsum(equation = var_24259_equation_0, values = (var_23905_cast_fp16, var_23738_cast_fp16))[name = tensor("op_24259_cast_fp16")]; + tensor var_24260_to_fp16 = const()[name = tensor("op_24260_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2371_cast_fp16 = mul(x = var_24259_cast_fp16, y = var_24260_to_fp16)[name = tensor("aw_chunk_2371_cast_fp16")]; + tensor var_24263_equation_0 = const()[name = tensor("op_24263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24263_cast_fp16 = einsum(equation = var_24263_equation_0, values = (var_23905_cast_fp16, var_23745_cast_fp16))[name = tensor("op_24263_cast_fp16")]; + tensor var_24264_to_fp16 = const()[name = tensor("op_24264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2373_cast_fp16 = mul(x = var_24263_cast_fp16, y = var_24264_to_fp16)[name = tensor("aw_chunk_2373_cast_fp16")]; + tensor var_24267_equation_0 = const()[name = tensor("op_24267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24267_cast_fp16 = einsum(equation = var_24267_equation_0, values = (var_23905_cast_fp16, var_23752_cast_fp16))[name = tensor("op_24267_cast_fp16")]; + tensor var_24268_to_fp16 = const()[name = tensor("op_24268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2375_cast_fp16 = mul(x = var_24267_cast_fp16, y = var_24268_to_fp16)[name = tensor("aw_chunk_2375_cast_fp16")]; + tensor var_24271_equation_0 = const()[name = tensor("op_24271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24271_cast_fp16 = einsum(equation = var_24271_equation_0, values = (var_23909_cast_fp16, var_23759_cast_fp16))[name = tensor("op_24271_cast_fp16")]; + tensor var_24272_to_fp16 = const()[name = tensor("op_24272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2377_cast_fp16 = mul(x = var_24271_cast_fp16, y = var_24272_to_fp16)[name = tensor("aw_chunk_2377_cast_fp16")]; + tensor var_24275_equation_0 = const()[name = tensor("op_24275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24275_cast_fp16 = einsum(equation = var_24275_equation_0, values = (var_23909_cast_fp16, var_23766_cast_fp16))[name = tensor("op_24275_cast_fp16")]; + tensor var_24276_to_fp16 = const()[name = tensor("op_24276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2379_cast_fp16 = mul(x = var_24275_cast_fp16, y = var_24276_to_fp16)[name = tensor("aw_chunk_2379_cast_fp16")]; + tensor var_24279_equation_0 = const()[name = tensor("op_24279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24279_cast_fp16 = einsum(equation = var_24279_equation_0, values = (var_23909_cast_fp16, var_23773_cast_fp16))[name = tensor("op_24279_cast_fp16")]; + tensor var_24280_to_fp16 = const()[name = tensor("op_24280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2381_cast_fp16 = mul(x = var_24279_cast_fp16, y = var_24280_to_fp16)[name = tensor("aw_chunk_2381_cast_fp16")]; + tensor var_24283_equation_0 = const()[name = tensor("op_24283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24283_cast_fp16 = einsum(equation = var_24283_equation_0, values = (var_23909_cast_fp16, var_23780_cast_fp16))[name = tensor("op_24283_cast_fp16")]; + tensor var_24284_to_fp16 = const()[name = tensor("op_24284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2383_cast_fp16 = mul(x = var_24283_cast_fp16, y = var_24284_to_fp16)[name = tensor("aw_chunk_2383_cast_fp16")]; + tensor var_24287_equation_0 = const()[name = tensor("op_24287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24287_cast_fp16 = einsum(equation = var_24287_equation_0, values = (var_23913_cast_fp16, var_23787_cast_fp16))[name = tensor("op_24287_cast_fp16")]; + tensor var_24288_to_fp16 = const()[name = tensor("op_24288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2385_cast_fp16 = mul(x = var_24287_cast_fp16, y = var_24288_to_fp16)[name = tensor("aw_chunk_2385_cast_fp16")]; + tensor var_24291_equation_0 = const()[name = tensor("op_24291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24291_cast_fp16 = einsum(equation = var_24291_equation_0, values = (var_23913_cast_fp16, var_23794_cast_fp16))[name = tensor("op_24291_cast_fp16")]; + tensor var_24292_to_fp16 = const()[name = tensor("op_24292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2387_cast_fp16 = mul(x = var_24291_cast_fp16, y = var_24292_to_fp16)[name = tensor("aw_chunk_2387_cast_fp16")]; + tensor var_24295_equation_0 = const()[name = tensor("op_24295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24295_cast_fp16 = einsum(equation = var_24295_equation_0, values = (var_23913_cast_fp16, var_23801_cast_fp16))[name = tensor("op_24295_cast_fp16")]; + tensor var_24296_to_fp16 = const()[name = tensor("op_24296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2389_cast_fp16 = mul(x = var_24295_cast_fp16, y = var_24296_to_fp16)[name = tensor("aw_chunk_2389_cast_fp16")]; + tensor var_24299_equation_0 = const()[name = tensor("op_24299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24299_cast_fp16 = einsum(equation = var_24299_equation_0, values = (var_23913_cast_fp16, var_23808_cast_fp16))[name = tensor("op_24299_cast_fp16")]; + tensor var_24300_to_fp16 = const()[name = tensor("op_24300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2391_cast_fp16 = mul(x = var_24299_cast_fp16, y = var_24300_to_fp16)[name = tensor("aw_chunk_2391_cast_fp16")]; + tensor var_24303_equation_0 = const()[name = tensor("op_24303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24303_cast_fp16 = einsum(equation = var_24303_equation_0, values = (var_23917_cast_fp16, var_23815_cast_fp16))[name = tensor("op_24303_cast_fp16")]; + tensor var_24304_to_fp16 = const()[name = tensor("op_24304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2393_cast_fp16 = mul(x = var_24303_cast_fp16, y = var_24304_to_fp16)[name = tensor("aw_chunk_2393_cast_fp16")]; + tensor var_24307_equation_0 = const()[name = tensor("op_24307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24307_cast_fp16 = einsum(equation = var_24307_equation_0, values = (var_23917_cast_fp16, var_23822_cast_fp16))[name = tensor("op_24307_cast_fp16")]; + tensor var_24308_to_fp16 = const()[name = tensor("op_24308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2395_cast_fp16 = mul(x = var_24307_cast_fp16, y = var_24308_to_fp16)[name = tensor("aw_chunk_2395_cast_fp16")]; + tensor var_24311_equation_0 = const()[name = tensor("op_24311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24311_cast_fp16 = einsum(equation = var_24311_equation_0, values = (var_23917_cast_fp16, var_23829_cast_fp16))[name = tensor("op_24311_cast_fp16")]; + tensor var_24312_to_fp16 = const()[name = tensor("op_24312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2397_cast_fp16 = mul(x = var_24311_cast_fp16, y = var_24312_to_fp16)[name = tensor("aw_chunk_2397_cast_fp16")]; + tensor var_24315_equation_0 = const()[name = tensor("op_24315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_24315_cast_fp16 = einsum(equation = var_24315_equation_0, values = (var_23917_cast_fp16, var_23836_cast_fp16))[name = tensor("op_24315_cast_fp16")]; + tensor var_24316_to_fp16 = const()[name = tensor("op_24316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2399_cast_fp16 = mul(x = var_24315_cast_fp16, y = var_24316_to_fp16)[name = tensor("aw_chunk_2399_cast_fp16")]; + tensor var_24318_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2241_cast_fp16)[name = tensor("op_24318_cast_fp16")]; + tensor var_24319_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2243_cast_fp16)[name = tensor("op_24319_cast_fp16")]; + tensor var_24320_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2245_cast_fp16)[name = tensor("op_24320_cast_fp16")]; + tensor var_24321_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2247_cast_fp16)[name = tensor("op_24321_cast_fp16")]; + tensor var_24322_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2249_cast_fp16)[name = tensor("op_24322_cast_fp16")]; + tensor var_24323_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2251_cast_fp16)[name = tensor("op_24323_cast_fp16")]; + tensor var_24324_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2253_cast_fp16)[name = tensor("op_24324_cast_fp16")]; + tensor var_24325_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2255_cast_fp16)[name = tensor("op_24325_cast_fp16")]; + tensor var_24326_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2257_cast_fp16)[name = tensor("op_24326_cast_fp16")]; + tensor var_24327_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2259_cast_fp16)[name = tensor("op_24327_cast_fp16")]; + tensor var_24328_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2261_cast_fp16)[name = tensor("op_24328_cast_fp16")]; + tensor var_24329_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2263_cast_fp16)[name = tensor("op_24329_cast_fp16")]; + tensor var_24330_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2265_cast_fp16)[name = tensor("op_24330_cast_fp16")]; + tensor var_24331_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2267_cast_fp16)[name = tensor("op_24331_cast_fp16")]; + tensor var_24332_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2269_cast_fp16)[name = tensor("op_24332_cast_fp16")]; + tensor var_24333_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2271_cast_fp16)[name = tensor("op_24333_cast_fp16")]; + tensor var_24334_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2273_cast_fp16)[name = tensor("op_24334_cast_fp16")]; + tensor var_24335_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2275_cast_fp16)[name = tensor("op_24335_cast_fp16")]; + tensor var_24336_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2277_cast_fp16)[name = tensor("op_24336_cast_fp16")]; + tensor var_24337_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2279_cast_fp16)[name = tensor("op_24337_cast_fp16")]; + tensor var_24338_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2281_cast_fp16)[name = tensor("op_24338_cast_fp16")]; + tensor var_24339_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2283_cast_fp16)[name = tensor("op_24339_cast_fp16")]; + tensor var_24340_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2285_cast_fp16)[name = tensor("op_24340_cast_fp16")]; + tensor var_24341_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2287_cast_fp16)[name = tensor("op_24341_cast_fp16")]; + tensor var_24342_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2289_cast_fp16)[name = tensor("op_24342_cast_fp16")]; + tensor var_24343_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2291_cast_fp16)[name = tensor("op_24343_cast_fp16")]; + tensor var_24344_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2293_cast_fp16)[name = tensor("op_24344_cast_fp16")]; + tensor var_24345_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2295_cast_fp16)[name = tensor("op_24345_cast_fp16")]; + tensor var_24346_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2297_cast_fp16)[name = tensor("op_24346_cast_fp16")]; + tensor var_24347_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2299_cast_fp16)[name = tensor("op_24347_cast_fp16")]; + tensor var_24348_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2301_cast_fp16)[name = tensor("op_24348_cast_fp16")]; + tensor var_24349_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2303_cast_fp16)[name = tensor("op_24349_cast_fp16")]; + tensor var_24350_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2305_cast_fp16)[name = tensor("op_24350_cast_fp16")]; + tensor var_24351_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2307_cast_fp16)[name = tensor("op_24351_cast_fp16")]; + tensor var_24352_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2309_cast_fp16)[name = tensor("op_24352_cast_fp16")]; + tensor var_24353_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2311_cast_fp16)[name = tensor("op_24353_cast_fp16")]; + tensor var_24354_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2313_cast_fp16)[name = tensor("op_24354_cast_fp16")]; + tensor var_24355_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2315_cast_fp16)[name = tensor("op_24355_cast_fp16")]; + tensor var_24356_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2317_cast_fp16)[name = tensor("op_24356_cast_fp16")]; + tensor var_24357_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2319_cast_fp16)[name = tensor("op_24357_cast_fp16")]; + tensor var_24358_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2321_cast_fp16)[name = tensor("op_24358_cast_fp16")]; + tensor var_24359_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2323_cast_fp16)[name = tensor("op_24359_cast_fp16")]; + tensor var_24360_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2325_cast_fp16)[name = tensor("op_24360_cast_fp16")]; + tensor var_24361_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2327_cast_fp16)[name = tensor("op_24361_cast_fp16")]; + tensor var_24362_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2329_cast_fp16)[name = tensor("op_24362_cast_fp16")]; + tensor var_24363_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2331_cast_fp16)[name = tensor("op_24363_cast_fp16")]; + tensor var_24364_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2333_cast_fp16)[name = tensor("op_24364_cast_fp16")]; + tensor var_24365_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2335_cast_fp16)[name = tensor("op_24365_cast_fp16")]; + tensor var_24366_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2337_cast_fp16)[name = tensor("op_24366_cast_fp16")]; + tensor var_24367_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2339_cast_fp16)[name = tensor("op_24367_cast_fp16")]; + tensor var_24368_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2341_cast_fp16)[name = tensor("op_24368_cast_fp16")]; + tensor var_24369_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2343_cast_fp16)[name = tensor("op_24369_cast_fp16")]; + tensor var_24370_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2345_cast_fp16)[name = tensor("op_24370_cast_fp16")]; + tensor var_24371_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2347_cast_fp16)[name = tensor("op_24371_cast_fp16")]; + tensor var_24372_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2349_cast_fp16)[name = tensor("op_24372_cast_fp16")]; + tensor var_24373_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2351_cast_fp16)[name = tensor("op_24373_cast_fp16")]; + tensor var_24374_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2353_cast_fp16)[name = tensor("op_24374_cast_fp16")]; + tensor var_24375_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2355_cast_fp16)[name = tensor("op_24375_cast_fp16")]; + tensor var_24376_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2357_cast_fp16)[name = tensor("op_24376_cast_fp16")]; + tensor var_24377_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2359_cast_fp16)[name = tensor("op_24377_cast_fp16")]; + tensor var_24378_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2361_cast_fp16)[name = tensor("op_24378_cast_fp16")]; + tensor var_24379_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2363_cast_fp16)[name = tensor("op_24379_cast_fp16")]; + tensor var_24380_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2365_cast_fp16)[name = tensor("op_24380_cast_fp16")]; + tensor var_24381_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2367_cast_fp16)[name = tensor("op_24381_cast_fp16")]; + tensor var_24382_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2369_cast_fp16)[name = tensor("op_24382_cast_fp16")]; + tensor var_24383_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2371_cast_fp16)[name = tensor("op_24383_cast_fp16")]; + tensor var_24384_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2373_cast_fp16)[name = tensor("op_24384_cast_fp16")]; + tensor var_24385_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2375_cast_fp16)[name = tensor("op_24385_cast_fp16")]; + tensor var_24386_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2377_cast_fp16)[name = tensor("op_24386_cast_fp16")]; + tensor var_24387_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2379_cast_fp16)[name = tensor("op_24387_cast_fp16")]; + tensor var_24388_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2381_cast_fp16)[name = tensor("op_24388_cast_fp16")]; + tensor var_24389_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2383_cast_fp16)[name = tensor("op_24389_cast_fp16")]; + tensor var_24390_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2385_cast_fp16)[name = tensor("op_24390_cast_fp16")]; + tensor var_24391_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2387_cast_fp16)[name = tensor("op_24391_cast_fp16")]; + tensor var_24392_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2389_cast_fp16)[name = tensor("op_24392_cast_fp16")]; + tensor var_24393_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2391_cast_fp16)[name = tensor("op_24393_cast_fp16")]; + tensor var_24394_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2393_cast_fp16)[name = tensor("op_24394_cast_fp16")]; + tensor var_24395_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2395_cast_fp16)[name = tensor("op_24395_cast_fp16")]; + tensor var_24396_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2397_cast_fp16)[name = tensor("op_24396_cast_fp16")]; + tensor var_24397_cast_fp16 = softmax(axis = var_23089, x = aw_chunk_2399_cast_fp16)[name = tensor("op_24397_cast_fp16")]; + tensor var_24399_equation_0 = const()[name = tensor("op_24399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24399_cast_fp16 = einsum(equation = var_24399_equation_0, values = (var_23919_cast_fp16, var_24318_cast_fp16))[name = tensor("op_24399_cast_fp16")]; + tensor var_24401_equation_0 = const()[name = tensor("op_24401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24401_cast_fp16 = einsum(equation = var_24401_equation_0, values = (var_23919_cast_fp16, var_24319_cast_fp16))[name = tensor("op_24401_cast_fp16")]; + tensor var_24403_equation_0 = const()[name = tensor("op_24403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24403_cast_fp16 = einsum(equation = var_24403_equation_0, values = (var_23919_cast_fp16, var_24320_cast_fp16))[name = tensor("op_24403_cast_fp16")]; + tensor var_24405_equation_0 = const()[name = tensor("op_24405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24405_cast_fp16 = einsum(equation = var_24405_equation_0, values = (var_23919_cast_fp16, var_24321_cast_fp16))[name = tensor("op_24405_cast_fp16")]; + tensor var_24407_equation_0 = const()[name = tensor("op_24407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24407_cast_fp16 = einsum(equation = var_24407_equation_0, values = (var_23923_cast_fp16, var_24322_cast_fp16))[name = tensor("op_24407_cast_fp16")]; + tensor var_24409_equation_0 = const()[name = tensor("op_24409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24409_cast_fp16 = einsum(equation = var_24409_equation_0, values = (var_23923_cast_fp16, var_24323_cast_fp16))[name = tensor("op_24409_cast_fp16")]; + tensor var_24411_equation_0 = const()[name = tensor("op_24411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24411_cast_fp16 = einsum(equation = var_24411_equation_0, values = (var_23923_cast_fp16, var_24324_cast_fp16))[name = tensor("op_24411_cast_fp16")]; + tensor var_24413_equation_0 = const()[name = tensor("op_24413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24413_cast_fp16 = einsum(equation = var_24413_equation_0, values = (var_23923_cast_fp16, var_24325_cast_fp16))[name = tensor("op_24413_cast_fp16")]; + tensor var_24415_equation_0 = const()[name = tensor("op_24415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24415_cast_fp16 = einsum(equation = var_24415_equation_0, values = (var_23927_cast_fp16, var_24326_cast_fp16))[name = tensor("op_24415_cast_fp16")]; + tensor var_24417_equation_0 = const()[name = tensor("op_24417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24417_cast_fp16 = einsum(equation = var_24417_equation_0, values = (var_23927_cast_fp16, var_24327_cast_fp16))[name = tensor("op_24417_cast_fp16")]; + tensor var_24419_equation_0 = const()[name = tensor("op_24419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24419_cast_fp16 = einsum(equation = var_24419_equation_0, values = (var_23927_cast_fp16, var_24328_cast_fp16))[name = tensor("op_24419_cast_fp16")]; + tensor var_24421_equation_0 = const()[name = tensor("op_24421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24421_cast_fp16 = einsum(equation = var_24421_equation_0, values = (var_23927_cast_fp16, var_24329_cast_fp16))[name = tensor("op_24421_cast_fp16")]; + tensor var_24423_equation_0 = const()[name = tensor("op_24423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24423_cast_fp16 = einsum(equation = var_24423_equation_0, values = (var_23931_cast_fp16, var_24330_cast_fp16))[name = tensor("op_24423_cast_fp16")]; + tensor var_24425_equation_0 = const()[name = tensor("op_24425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24425_cast_fp16 = einsum(equation = var_24425_equation_0, values = (var_23931_cast_fp16, var_24331_cast_fp16))[name = tensor("op_24425_cast_fp16")]; + tensor var_24427_equation_0 = const()[name = tensor("op_24427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24427_cast_fp16 = einsum(equation = var_24427_equation_0, values = (var_23931_cast_fp16, var_24332_cast_fp16))[name = tensor("op_24427_cast_fp16")]; + tensor var_24429_equation_0 = const()[name = tensor("op_24429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24429_cast_fp16 = einsum(equation = var_24429_equation_0, values = (var_23931_cast_fp16, var_24333_cast_fp16))[name = tensor("op_24429_cast_fp16")]; + tensor var_24431_equation_0 = const()[name = tensor("op_24431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24431_cast_fp16 = einsum(equation = var_24431_equation_0, values = (var_23935_cast_fp16, var_24334_cast_fp16))[name = tensor("op_24431_cast_fp16")]; + tensor var_24433_equation_0 = const()[name = tensor("op_24433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24433_cast_fp16 = einsum(equation = var_24433_equation_0, values = (var_23935_cast_fp16, var_24335_cast_fp16))[name = tensor("op_24433_cast_fp16")]; + tensor var_24435_equation_0 = const()[name = tensor("op_24435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24435_cast_fp16 = einsum(equation = var_24435_equation_0, values = (var_23935_cast_fp16, var_24336_cast_fp16))[name = tensor("op_24435_cast_fp16")]; + tensor var_24437_equation_0 = const()[name = tensor("op_24437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24437_cast_fp16 = einsum(equation = var_24437_equation_0, values = (var_23935_cast_fp16, var_24337_cast_fp16))[name = tensor("op_24437_cast_fp16")]; + tensor var_24439_equation_0 = const()[name = tensor("op_24439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24439_cast_fp16 = einsum(equation = var_24439_equation_0, values = (var_23939_cast_fp16, var_24338_cast_fp16))[name = tensor("op_24439_cast_fp16")]; + tensor var_24441_equation_0 = const()[name = tensor("op_24441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24441_cast_fp16 = einsum(equation = var_24441_equation_0, values = (var_23939_cast_fp16, var_24339_cast_fp16))[name = tensor("op_24441_cast_fp16")]; + tensor var_24443_equation_0 = const()[name = tensor("op_24443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24443_cast_fp16 = einsum(equation = var_24443_equation_0, values = (var_23939_cast_fp16, var_24340_cast_fp16))[name = tensor("op_24443_cast_fp16")]; + tensor var_24445_equation_0 = const()[name = tensor("op_24445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24445_cast_fp16 = einsum(equation = var_24445_equation_0, values = (var_23939_cast_fp16, var_24341_cast_fp16))[name = tensor("op_24445_cast_fp16")]; + tensor var_24447_equation_0 = const()[name = tensor("op_24447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24447_cast_fp16 = einsum(equation = var_24447_equation_0, values = (var_23943_cast_fp16, var_24342_cast_fp16))[name = tensor("op_24447_cast_fp16")]; + tensor var_24449_equation_0 = const()[name = tensor("op_24449_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24449_cast_fp16 = einsum(equation = var_24449_equation_0, values = (var_23943_cast_fp16, var_24343_cast_fp16))[name = tensor("op_24449_cast_fp16")]; + tensor var_24451_equation_0 = const()[name = tensor("op_24451_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24451_cast_fp16 = einsum(equation = var_24451_equation_0, values = (var_23943_cast_fp16, var_24344_cast_fp16))[name = tensor("op_24451_cast_fp16")]; + tensor var_24453_equation_0 = const()[name = tensor("op_24453_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24453_cast_fp16 = einsum(equation = var_24453_equation_0, values = (var_23943_cast_fp16, var_24345_cast_fp16))[name = tensor("op_24453_cast_fp16")]; + tensor var_24455_equation_0 = const()[name = tensor("op_24455_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24455_cast_fp16 = einsum(equation = var_24455_equation_0, values = (var_23947_cast_fp16, var_24346_cast_fp16))[name = tensor("op_24455_cast_fp16")]; + tensor var_24457_equation_0 = const()[name = tensor("op_24457_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24457_cast_fp16 = einsum(equation = var_24457_equation_0, values = (var_23947_cast_fp16, var_24347_cast_fp16))[name = tensor("op_24457_cast_fp16")]; + tensor var_24459_equation_0 = const()[name = tensor("op_24459_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24459_cast_fp16 = einsum(equation = var_24459_equation_0, values = (var_23947_cast_fp16, var_24348_cast_fp16))[name = tensor("op_24459_cast_fp16")]; + tensor var_24461_equation_0 = const()[name = tensor("op_24461_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24461_cast_fp16 = einsum(equation = var_24461_equation_0, values = (var_23947_cast_fp16, var_24349_cast_fp16))[name = tensor("op_24461_cast_fp16")]; + tensor var_24463_equation_0 = const()[name = tensor("op_24463_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24463_cast_fp16 = einsum(equation = var_24463_equation_0, values = (var_23951_cast_fp16, var_24350_cast_fp16))[name = tensor("op_24463_cast_fp16")]; + tensor var_24465_equation_0 = const()[name = tensor("op_24465_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24465_cast_fp16 = einsum(equation = var_24465_equation_0, values = (var_23951_cast_fp16, var_24351_cast_fp16))[name = tensor("op_24465_cast_fp16")]; + tensor var_24467_equation_0 = const()[name = tensor("op_24467_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24467_cast_fp16 = einsum(equation = var_24467_equation_0, values = (var_23951_cast_fp16, var_24352_cast_fp16))[name = tensor("op_24467_cast_fp16")]; + tensor var_24469_equation_0 = const()[name = tensor("op_24469_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24469_cast_fp16 = einsum(equation = var_24469_equation_0, values = (var_23951_cast_fp16, var_24353_cast_fp16))[name = tensor("op_24469_cast_fp16")]; + tensor var_24471_equation_0 = const()[name = tensor("op_24471_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24471_cast_fp16 = einsum(equation = var_24471_equation_0, values = (var_23955_cast_fp16, var_24354_cast_fp16))[name = tensor("op_24471_cast_fp16")]; + tensor var_24473_equation_0 = const()[name = tensor("op_24473_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24473_cast_fp16 = einsum(equation = var_24473_equation_0, values = (var_23955_cast_fp16, var_24355_cast_fp16))[name = tensor("op_24473_cast_fp16")]; + tensor var_24475_equation_0 = const()[name = tensor("op_24475_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24475_cast_fp16 = einsum(equation = var_24475_equation_0, values = (var_23955_cast_fp16, var_24356_cast_fp16))[name = tensor("op_24475_cast_fp16")]; + tensor var_24477_equation_0 = const()[name = tensor("op_24477_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24477_cast_fp16 = einsum(equation = var_24477_equation_0, values = (var_23955_cast_fp16, var_24357_cast_fp16))[name = tensor("op_24477_cast_fp16")]; + tensor var_24479_equation_0 = const()[name = tensor("op_24479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24479_cast_fp16 = einsum(equation = var_24479_equation_0, values = (var_23959_cast_fp16, var_24358_cast_fp16))[name = tensor("op_24479_cast_fp16")]; + tensor var_24481_equation_0 = const()[name = tensor("op_24481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24481_cast_fp16 = einsum(equation = var_24481_equation_0, values = (var_23959_cast_fp16, var_24359_cast_fp16))[name = tensor("op_24481_cast_fp16")]; + tensor var_24483_equation_0 = const()[name = tensor("op_24483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24483_cast_fp16 = einsum(equation = var_24483_equation_0, values = (var_23959_cast_fp16, var_24360_cast_fp16))[name = tensor("op_24483_cast_fp16")]; + tensor var_24485_equation_0 = const()[name = tensor("op_24485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24485_cast_fp16 = einsum(equation = var_24485_equation_0, values = (var_23959_cast_fp16, var_24361_cast_fp16))[name = tensor("op_24485_cast_fp16")]; + tensor var_24487_equation_0 = const()[name = tensor("op_24487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24487_cast_fp16 = einsum(equation = var_24487_equation_0, values = (var_23963_cast_fp16, var_24362_cast_fp16))[name = tensor("op_24487_cast_fp16")]; + tensor var_24489_equation_0 = const()[name = tensor("op_24489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24489_cast_fp16 = einsum(equation = var_24489_equation_0, values = (var_23963_cast_fp16, var_24363_cast_fp16))[name = tensor("op_24489_cast_fp16")]; + tensor var_24491_equation_0 = const()[name = tensor("op_24491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24491_cast_fp16 = einsum(equation = var_24491_equation_0, values = (var_23963_cast_fp16, var_24364_cast_fp16))[name = tensor("op_24491_cast_fp16")]; + tensor var_24493_equation_0 = const()[name = tensor("op_24493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24493_cast_fp16 = einsum(equation = var_24493_equation_0, values = (var_23963_cast_fp16, var_24365_cast_fp16))[name = tensor("op_24493_cast_fp16")]; + tensor var_24495_equation_0 = const()[name = tensor("op_24495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24495_cast_fp16 = einsum(equation = var_24495_equation_0, values = (var_23967_cast_fp16, var_24366_cast_fp16))[name = tensor("op_24495_cast_fp16")]; + tensor var_24497_equation_0 = const()[name = tensor("op_24497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24497_cast_fp16 = einsum(equation = var_24497_equation_0, values = (var_23967_cast_fp16, var_24367_cast_fp16))[name = tensor("op_24497_cast_fp16")]; + tensor var_24499_equation_0 = const()[name = tensor("op_24499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24499_cast_fp16 = einsum(equation = var_24499_equation_0, values = (var_23967_cast_fp16, var_24368_cast_fp16))[name = tensor("op_24499_cast_fp16")]; + tensor var_24501_equation_0 = const()[name = tensor("op_24501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24501_cast_fp16 = einsum(equation = var_24501_equation_0, values = (var_23967_cast_fp16, var_24369_cast_fp16))[name = tensor("op_24501_cast_fp16")]; + tensor var_24503_equation_0 = const()[name = tensor("op_24503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24503_cast_fp16 = einsum(equation = var_24503_equation_0, values = (var_23971_cast_fp16, var_24370_cast_fp16))[name = tensor("op_24503_cast_fp16")]; + tensor var_24505_equation_0 = const()[name = tensor("op_24505_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24505_cast_fp16 = einsum(equation = var_24505_equation_0, values = (var_23971_cast_fp16, var_24371_cast_fp16))[name = tensor("op_24505_cast_fp16")]; + tensor var_24507_equation_0 = const()[name = tensor("op_24507_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24507_cast_fp16 = einsum(equation = var_24507_equation_0, values = (var_23971_cast_fp16, var_24372_cast_fp16))[name = tensor("op_24507_cast_fp16")]; + tensor var_24509_equation_0 = const()[name = tensor("op_24509_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24509_cast_fp16 = einsum(equation = var_24509_equation_0, values = (var_23971_cast_fp16, var_24373_cast_fp16))[name = tensor("op_24509_cast_fp16")]; + tensor var_24511_equation_0 = const()[name = tensor("op_24511_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24511_cast_fp16 = einsum(equation = var_24511_equation_0, values = (var_23975_cast_fp16, var_24374_cast_fp16))[name = tensor("op_24511_cast_fp16")]; + tensor var_24513_equation_0 = const()[name = tensor("op_24513_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24513_cast_fp16 = einsum(equation = var_24513_equation_0, values = (var_23975_cast_fp16, var_24375_cast_fp16))[name = tensor("op_24513_cast_fp16")]; + tensor var_24515_equation_0 = const()[name = tensor("op_24515_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24515_cast_fp16 = einsum(equation = var_24515_equation_0, values = (var_23975_cast_fp16, var_24376_cast_fp16))[name = tensor("op_24515_cast_fp16")]; + tensor var_24517_equation_0 = const()[name = tensor("op_24517_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24517_cast_fp16 = einsum(equation = var_24517_equation_0, values = (var_23975_cast_fp16, var_24377_cast_fp16))[name = tensor("op_24517_cast_fp16")]; + tensor var_24519_equation_0 = const()[name = tensor("op_24519_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24519_cast_fp16 = einsum(equation = var_24519_equation_0, values = (var_23979_cast_fp16, var_24378_cast_fp16))[name = tensor("op_24519_cast_fp16")]; + tensor var_24521_equation_0 = const()[name = tensor("op_24521_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24521_cast_fp16 = einsum(equation = var_24521_equation_0, values = (var_23979_cast_fp16, var_24379_cast_fp16))[name = tensor("op_24521_cast_fp16")]; + tensor var_24523_equation_0 = const()[name = tensor("op_24523_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24523_cast_fp16 = einsum(equation = var_24523_equation_0, values = (var_23979_cast_fp16, var_24380_cast_fp16))[name = tensor("op_24523_cast_fp16")]; + tensor var_24525_equation_0 = const()[name = tensor("op_24525_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24525_cast_fp16 = einsum(equation = var_24525_equation_0, values = (var_23979_cast_fp16, var_24381_cast_fp16))[name = tensor("op_24525_cast_fp16")]; + tensor var_24527_equation_0 = const()[name = tensor("op_24527_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24527_cast_fp16 = einsum(equation = var_24527_equation_0, values = (var_23983_cast_fp16, var_24382_cast_fp16))[name = tensor("op_24527_cast_fp16")]; + tensor var_24529_equation_0 = const()[name = tensor("op_24529_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24529_cast_fp16 = einsum(equation = var_24529_equation_0, values = (var_23983_cast_fp16, var_24383_cast_fp16))[name = tensor("op_24529_cast_fp16")]; + tensor var_24531_equation_0 = const()[name = tensor("op_24531_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24531_cast_fp16 = einsum(equation = var_24531_equation_0, values = (var_23983_cast_fp16, var_24384_cast_fp16))[name = tensor("op_24531_cast_fp16")]; + tensor var_24533_equation_0 = const()[name = tensor("op_24533_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24533_cast_fp16 = einsum(equation = var_24533_equation_0, values = (var_23983_cast_fp16, var_24385_cast_fp16))[name = tensor("op_24533_cast_fp16")]; + tensor var_24535_equation_0 = const()[name = tensor("op_24535_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24535_cast_fp16 = einsum(equation = var_24535_equation_0, values = (var_23987_cast_fp16, var_24386_cast_fp16))[name = tensor("op_24535_cast_fp16")]; + tensor var_24537_equation_0 = const()[name = tensor("op_24537_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24537_cast_fp16 = einsum(equation = var_24537_equation_0, values = (var_23987_cast_fp16, var_24387_cast_fp16))[name = tensor("op_24537_cast_fp16")]; + tensor var_24539_equation_0 = const()[name = tensor("op_24539_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24539_cast_fp16 = einsum(equation = var_24539_equation_0, values = (var_23987_cast_fp16, var_24388_cast_fp16))[name = tensor("op_24539_cast_fp16")]; + tensor var_24541_equation_0 = const()[name = tensor("op_24541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24541_cast_fp16 = einsum(equation = var_24541_equation_0, values = (var_23987_cast_fp16, var_24389_cast_fp16))[name = tensor("op_24541_cast_fp16")]; + tensor var_24543_equation_0 = const()[name = tensor("op_24543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24543_cast_fp16 = einsum(equation = var_24543_equation_0, values = (var_23991_cast_fp16, var_24390_cast_fp16))[name = tensor("op_24543_cast_fp16")]; + tensor var_24545_equation_0 = const()[name = tensor("op_24545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24545_cast_fp16 = einsum(equation = var_24545_equation_0, values = (var_23991_cast_fp16, var_24391_cast_fp16))[name = tensor("op_24545_cast_fp16")]; + tensor var_24547_equation_0 = const()[name = tensor("op_24547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24547_cast_fp16 = einsum(equation = var_24547_equation_0, values = (var_23991_cast_fp16, var_24392_cast_fp16))[name = tensor("op_24547_cast_fp16")]; + tensor var_24549_equation_0 = const()[name = tensor("op_24549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24549_cast_fp16 = einsum(equation = var_24549_equation_0, values = (var_23991_cast_fp16, var_24393_cast_fp16))[name = tensor("op_24549_cast_fp16")]; + tensor var_24551_equation_0 = const()[name = tensor("op_24551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24551_cast_fp16 = einsum(equation = var_24551_equation_0, values = (var_23995_cast_fp16, var_24394_cast_fp16))[name = tensor("op_24551_cast_fp16")]; + tensor var_24553_equation_0 = const()[name = tensor("op_24553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24553_cast_fp16 = einsum(equation = var_24553_equation_0, values = (var_23995_cast_fp16, var_24395_cast_fp16))[name = tensor("op_24553_cast_fp16")]; + tensor var_24555_equation_0 = const()[name = tensor("op_24555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24555_cast_fp16 = einsum(equation = var_24555_equation_0, values = (var_23995_cast_fp16, var_24396_cast_fp16))[name = tensor("op_24555_cast_fp16")]; + tensor var_24557_equation_0 = const()[name = tensor("op_24557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_24557_cast_fp16 = einsum(equation = var_24557_equation_0, values = (var_23995_cast_fp16, var_24397_cast_fp16))[name = tensor("op_24557_cast_fp16")]; + tensor var_24559_interleave_0 = const()[name = tensor("op_24559_interleave_0"), val = tensor(false)]; + tensor var_24559_cast_fp16 = concat(axis = var_23064, interleave = var_24559_interleave_0, values = (var_24399_cast_fp16, var_24401_cast_fp16, var_24403_cast_fp16, var_24405_cast_fp16))[name = tensor("op_24559_cast_fp16")]; + tensor var_24561_interleave_0 = const()[name = tensor("op_24561_interleave_0"), val = tensor(false)]; + tensor var_24561_cast_fp16 = concat(axis = var_23064, interleave = var_24561_interleave_0, values = (var_24407_cast_fp16, var_24409_cast_fp16, var_24411_cast_fp16, var_24413_cast_fp16))[name = tensor("op_24561_cast_fp16")]; + tensor var_24563_interleave_0 = const()[name = tensor("op_24563_interleave_0"), val = tensor(false)]; + tensor var_24563_cast_fp16 = concat(axis = var_23064, interleave = var_24563_interleave_0, values = (var_24415_cast_fp16, var_24417_cast_fp16, var_24419_cast_fp16, var_24421_cast_fp16))[name = tensor("op_24563_cast_fp16")]; + tensor var_24565_interleave_0 = const()[name = tensor("op_24565_interleave_0"), val = tensor(false)]; + tensor var_24565_cast_fp16 = concat(axis = var_23064, interleave = var_24565_interleave_0, values = (var_24423_cast_fp16, var_24425_cast_fp16, var_24427_cast_fp16, var_24429_cast_fp16))[name = tensor("op_24565_cast_fp16")]; + tensor var_24567_interleave_0 = const()[name = tensor("op_24567_interleave_0"), val = tensor(false)]; + tensor var_24567_cast_fp16 = concat(axis = var_23064, interleave = var_24567_interleave_0, values = (var_24431_cast_fp16, var_24433_cast_fp16, var_24435_cast_fp16, var_24437_cast_fp16))[name = tensor("op_24567_cast_fp16")]; + tensor var_24569_interleave_0 = const()[name = tensor("op_24569_interleave_0"), val = tensor(false)]; + tensor var_24569_cast_fp16 = concat(axis = var_23064, interleave = var_24569_interleave_0, values = (var_24439_cast_fp16, var_24441_cast_fp16, var_24443_cast_fp16, var_24445_cast_fp16))[name = tensor("op_24569_cast_fp16")]; + tensor var_24571_interleave_0 = const()[name = tensor("op_24571_interleave_0"), val = tensor(false)]; + tensor var_24571_cast_fp16 = concat(axis = var_23064, interleave = var_24571_interleave_0, values = (var_24447_cast_fp16, var_24449_cast_fp16, var_24451_cast_fp16, var_24453_cast_fp16))[name = tensor("op_24571_cast_fp16")]; + tensor var_24573_interleave_0 = const()[name = tensor("op_24573_interleave_0"), val = tensor(false)]; + tensor var_24573_cast_fp16 = concat(axis = var_23064, interleave = var_24573_interleave_0, values = (var_24455_cast_fp16, var_24457_cast_fp16, var_24459_cast_fp16, var_24461_cast_fp16))[name = tensor("op_24573_cast_fp16")]; + tensor var_24575_interleave_0 = const()[name = tensor("op_24575_interleave_0"), val = tensor(false)]; + tensor var_24575_cast_fp16 = concat(axis = var_23064, interleave = var_24575_interleave_0, values = (var_24463_cast_fp16, var_24465_cast_fp16, var_24467_cast_fp16, var_24469_cast_fp16))[name = tensor("op_24575_cast_fp16")]; + tensor var_24577_interleave_0 = const()[name = tensor("op_24577_interleave_0"), val = tensor(false)]; + tensor var_24577_cast_fp16 = concat(axis = var_23064, interleave = var_24577_interleave_0, values = (var_24471_cast_fp16, var_24473_cast_fp16, var_24475_cast_fp16, var_24477_cast_fp16))[name = tensor("op_24577_cast_fp16")]; + tensor var_24579_interleave_0 = const()[name = tensor("op_24579_interleave_0"), val = tensor(false)]; + tensor var_24579_cast_fp16 = concat(axis = var_23064, interleave = var_24579_interleave_0, values = (var_24479_cast_fp16, var_24481_cast_fp16, var_24483_cast_fp16, var_24485_cast_fp16))[name = tensor("op_24579_cast_fp16")]; + tensor var_24581_interleave_0 = const()[name = tensor("op_24581_interleave_0"), val = tensor(false)]; + tensor var_24581_cast_fp16 = concat(axis = var_23064, interleave = var_24581_interleave_0, values = (var_24487_cast_fp16, var_24489_cast_fp16, var_24491_cast_fp16, var_24493_cast_fp16))[name = tensor("op_24581_cast_fp16")]; + tensor var_24583_interleave_0 = const()[name = tensor("op_24583_interleave_0"), val = tensor(false)]; + tensor var_24583_cast_fp16 = concat(axis = var_23064, interleave = var_24583_interleave_0, values = (var_24495_cast_fp16, var_24497_cast_fp16, var_24499_cast_fp16, var_24501_cast_fp16))[name = tensor("op_24583_cast_fp16")]; + tensor var_24585_interleave_0 = const()[name = tensor("op_24585_interleave_0"), val = tensor(false)]; + tensor var_24585_cast_fp16 = concat(axis = var_23064, interleave = var_24585_interleave_0, values = (var_24503_cast_fp16, var_24505_cast_fp16, var_24507_cast_fp16, var_24509_cast_fp16))[name = tensor("op_24585_cast_fp16")]; + tensor var_24587_interleave_0 = const()[name = tensor("op_24587_interleave_0"), val = tensor(false)]; + tensor var_24587_cast_fp16 = concat(axis = var_23064, interleave = var_24587_interleave_0, values = (var_24511_cast_fp16, var_24513_cast_fp16, var_24515_cast_fp16, var_24517_cast_fp16))[name = tensor("op_24587_cast_fp16")]; + tensor var_24589_interleave_0 = const()[name = tensor("op_24589_interleave_0"), val = tensor(false)]; + tensor var_24589_cast_fp16 = concat(axis = var_23064, interleave = var_24589_interleave_0, values = (var_24519_cast_fp16, var_24521_cast_fp16, var_24523_cast_fp16, var_24525_cast_fp16))[name = tensor("op_24589_cast_fp16")]; + tensor var_24591_interleave_0 = const()[name = tensor("op_24591_interleave_0"), val = tensor(false)]; + tensor var_24591_cast_fp16 = concat(axis = var_23064, interleave = var_24591_interleave_0, values = (var_24527_cast_fp16, var_24529_cast_fp16, var_24531_cast_fp16, var_24533_cast_fp16))[name = tensor("op_24591_cast_fp16")]; + tensor var_24593_interleave_0 = const()[name = tensor("op_24593_interleave_0"), val = tensor(false)]; + tensor var_24593_cast_fp16 = concat(axis = var_23064, interleave = var_24593_interleave_0, values = (var_24535_cast_fp16, var_24537_cast_fp16, var_24539_cast_fp16, var_24541_cast_fp16))[name = tensor("op_24593_cast_fp16")]; + tensor var_24595_interleave_0 = const()[name = tensor("op_24595_interleave_0"), val = tensor(false)]; + tensor var_24595_cast_fp16 = concat(axis = var_23064, interleave = var_24595_interleave_0, values = (var_24543_cast_fp16, var_24545_cast_fp16, var_24547_cast_fp16, var_24549_cast_fp16))[name = tensor("op_24595_cast_fp16")]; + tensor var_24597_interleave_0 = const()[name = tensor("op_24597_interleave_0"), val = tensor(false)]; + tensor var_24597_cast_fp16 = concat(axis = var_23064, interleave = var_24597_interleave_0, values = (var_24551_cast_fp16, var_24553_cast_fp16, var_24555_cast_fp16, var_24557_cast_fp16))[name = tensor("op_24597_cast_fp16")]; + tensor input_287_interleave_0 = const()[name = tensor("input_287_interleave_0"), val = tensor(false)]; + tensor input_287_cast_fp16 = concat(axis = var_23089, interleave = input_287_interleave_0, values = (var_24559_cast_fp16, var_24561_cast_fp16, var_24563_cast_fp16, var_24565_cast_fp16, var_24567_cast_fp16, var_24569_cast_fp16, var_24571_cast_fp16, var_24573_cast_fp16, var_24575_cast_fp16, var_24577_cast_fp16, var_24579_cast_fp16, var_24581_cast_fp16, var_24583_cast_fp16, var_24585_cast_fp16, var_24587_cast_fp16, var_24589_cast_fp16, var_24591_cast_fp16, var_24593_cast_fp16, var_24595_cast_fp16, var_24597_cast_fp16))[name = tensor("input_287_cast_fp16")]; + tensor var_24605 = const()[name = tensor("op_24605"), val = tensor([1, 1])]; + tensor var_24607 = const()[name = tensor("op_24607"), val = tensor([1, 1])]; + tensor pretrained_out_175_pad_type_0 = const()[name = tensor("pretrained_out_175_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_175_pad_0 = const()[name = tensor("pretrained_out_175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165433216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166252480))), name = tensor("layers_14_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_14_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166252608)))]; + tensor pretrained_out_175_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_24607, groups = var_23089, pad = pretrained_out_175_pad_0, pad_type = pretrained_out_175_pad_type_0, strides = var_24605, weight = layers_14_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_287_cast_fp16)[name = tensor("pretrained_out_175_cast_fp16")]; + tensor var_24611 = const()[name = tensor("op_24611"), val = tensor([1, 1])]; + tensor var_24613 = const()[name = tensor("op_24613"), val = tensor([1, 1])]; + tensor input_289_pad_type_0 = const()[name = tensor("input_289_pad_type_0"), val = tensor("custom")]; + tensor input_289_pad_0 = const()[name = tensor("input_289_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166255232)))]; + tensor input_289_cast_fp16 = conv(dilations = var_24613, groups = var_23089, pad = input_289_pad_0, pad_type = input_289_pad_type_0, strides = var_24611, weight = layers_14_self_attn_o_proj_loraA_weight_to_fp16, x = input_287_cast_fp16)[name = tensor("input_289_cast_fp16")]; + tensor var_24617 = const()[name = tensor("op_24617"), val = tensor([1, 1])]; + tensor var_24619 = const()[name = tensor("op_24619"), val = tensor([1, 1])]; + tensor lora_out_349_pad_type_0 = const()[name = tensor("lora_out_349_pad_type_0"), val = tensor("custom")]; + tensor lora_out_349_pad_0 = const()[name = tensor("lora_out_349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_351_weight_0_to_fp16 = const()[name = tensor("lora_out_351_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166296256)))]; + tensor lora_out_351_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_24619, groups = var_23089, pad = lora_out_349_pad_0, pad_type = lora_out_349_pad_type_0, strides = var_24617, weight = lora_out_351_weight_0_to_fp16, x = input_289_cast_fp16)[name = tensor("lora_out_351_cast_fp16")]; + tensor obj_59_cast_fp16 = add(x = pretrained_out_175_cast_fp16, y = lora_out_351_cast_fp16)[name = tensor("obj_59_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = obj_59_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; + tensor var_24628 = const()[name = tensor("op_24628"), val = tensor([1])]; + tensor channels_mean_59_cast_fp16 = reduce_mean(axes = var_24628, keep_dims = var_23090, x = inputs_59_cast_fp16)[name = tensor("channels_mean_59_cast_fp16")]; + tensor zero_mean_59_cast_fp16 = sub(x = inputs_59_cast_fp16, y = channels_mean_59_cast_fp16)[name = tensor("zero_mean_59_cast_fp16")]; + tensor zero_mean_sq_59_cast_fp16 = mul(x = zero_mean_59_cast_fp16, y = zero_mean_59_cast_fp16)[name = tensor("zero_mean_sq_59_cast_fp16")]; + tensor var_24632 = const()[name = tensor("op_24632"), val = tensor([1])]; + tensor var_24633_cast_fp16 = reduce_mean(axes = var_24632, keep_dims = var_23090, x = zero_mean_sq_59_cast_fp16)[name = tensor("op_24633_cast_fp16")]; + tensor var_24634_to_fp16 = const()[name = tensor("op_24634_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_24635_cast_fp16 = add(x = var_24633_cast_fp16, y = var_24634_to_fp16)[name = tensor("op_24635_cast_fp16")]; + tensor denom_59_epsilon_0 = const()[name = tensor("denom_59_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_59_cast_fp16 = rsqrt(epsilon = denom_59_epsilon_0, x = var_24635_cast_fp16)[name = tensor("denom_59_cast_fp16")]; + tensor out_59_cast_fp16 = mul(x = zero_mean_59_cast_fp16, y = denom_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; + tensor input_291_gamma_0_to_fp16 = const()[name = tensor("input_291_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166337280)))]; + tensor input_291_beta_0_to_fp16 = const()[name = tensor("input_291_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166339904)))]; + tensor input_291_epsilon_0_to_fp16 = const()[name = tensor("input_291_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_291_cast_fp16 = batch_norm(beta = input_291_beta_0_to_fp16, epsilon = input_291_epsilon_0_to_fp16, gamma = input_291_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor("input_291_cast_fp16")]; + tensor var_24649 = const()[name = tensor("op_24649"), val = tensor([1, 1])]; + tensor var_24651 = const()[name = tensor("op_24651"), val = tensor([1, 1])]; + tensor pretrained_out_177_pad_type_0 = const()[name = tensor("pretrained_out_177_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_177_pad_0 = const()[name = tensor("pretrained_out_177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166342528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169619392))), name = tensor("layers_14_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_14_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_14_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169619520)))]; + tensor pretrained_out_177_cast_fp16 = conv(bias = layers_14_fc1_pretrained_bias_to_fp16, dilations = var_24651, groups = var_23089, pad = pretrained_out_177_pad_0, pad_type = pretrained_out_177_pad_type_0, strides = var_24649, weight = layers_14_fc1_pretrained_weight_to_fp16_palettized, x = input_291_cast_fp16)[name = tensor("pretrained_out_177_cast_fp16")]; + tensor var_24655 = const()[name = tensor("op_24655"), val = tensor([1, 1])]; + tensor var_24657 = const()[name = tensor("op_24657"), val = tensor([1, 1])]; + tensor input_293_pad_type_0 = const()[name = tensor("input_293_pad_type_0"), val = tensor("custom")]; + tensor input_293_pad_0 = const()[name = tensor("input_293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_14_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169629824)))]; + tensor input_293_cast_fp16 = conv(dilations = var_24657, groups = var_23089, pad = input_293_pad_0, pad_type = input_293_pad_type_0, strides = var_24655, weight = layers_14_fc1_loraA_weight_to_fp16, x = input_291_cast_fp16)[name = tensor("input_293_cast_fp16")]; + tensor var_24661 = const()[name = tensor("op_24661"), val = tensor([1, 1])]; + tensor var_24663 = const()[name = tensor("op_24663"), val = tensor([1, 1])]; + tensor lora_out_353_pad_type_0 = const()[name = tensor("lora_out_353_pad_type_0"), val = tensor("custom")]; + tensor lora_out_353_pad_0 = const()[name = tensor("lora_out_353_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_355_weight_0_to_fp16 = const()[name = tensor("lora_out_355_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169670848)))]; + tensor lora_out_355_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_24663, groups = var_23089, pad = lora_out_353_pad_0, pad_type = lora_out_353_pad_type_0, strides = var_24661, weight = lora_out_355_weight_0_to_fp16, x = input_293_cast_fp16)[name = tensor("lora_out_355_cast_fp16")]; + tensor input_295_cast_fp16 = add(x = pretrained_out_177_cast_fp16, y = lora_out_355_cast_fp16)[name = tensor("input_295_cast_fp16")]; + tensor input_297_mode_0 = const()[name = tensor("input_297_mode_0"), val = tensor("EXACT")]; + tensor input_297_cast_fp16 = gelu(mode = input_297_mode_0, x = input_295_cast_fp16)[name = tensor("input_297_cast_fp16")]; + tensor var_24675 = const()[name = tensor("op_24675"), val = tensor([1, 1])]; + tensor var_24677 = const()[name = tensor("op_24677"), val = tensor([1, 1])]; + tensor pretrained_out_179_pad_type_0 = const()[name = tensor("pretrained_out_179_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_179_pad_0 = const()[name = tensor("pretrained_out_179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169834752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173111616))), name = tensor("layers_14_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_14_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_14_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173111744)))]; + tensor pretrained_out_179_cast_fp16 = conv(bias = layers_14_fc2_pretrained_bias_to_fp16, dilations = var_24677, groups = var_23089, pad = pretrained_out_179_pad_0, pad_type = pretrained_out_179_pad_type_0, strides = var_24675, weight = layers_14_fc2_pretrained_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = tensor("pretrained_out_179_cast_fp16")]; + tensor var_24681 = const()[name = tensor("op_24681"), val = tensor([1, 1])]; + tensor var_24683 = const()[name = tensor("op_24683"), val = tensor([1, 1])]; + tensor input_299_pad_type_0 = const()[name = tensor("input_299_pad_type_0"), val = tensor("custom")]; + tensor input_299_pad_0 = const()[name = tensor("input_299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_14_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_14_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173114368)))]; + tensor input_299_cast_fp16 = conv(dilations = var_24683, groups = var_23089, pad = input_299_pad_0, pad_type = input_299_pad_type_0, strides = var_24681, weight = layers_14_fc2_loraA_weight_to_fp16, x = input_297_cast_fp16)[name = tensor("input_299_cast_fp16")]; + tensor var_24687 = const()[name = tensor("op_24687"), val = tensor([1, 1])]; + tensor var_24689 = const()[name = tensor("op_24689"), val = tensor([1, 1])]; + tensor lora_out_357_pad_type_0 = const()[name = tensor("lora_out_357_pad_type_0"), val = tensor("custom")]; + tensor lora_out_357_pad_0 = const()[name = tensor("lora_out_357_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_359_weight_0_to_fp16 = const()[name = tensor("lora_out_359_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173278272)))]; + tensor lora_out_359_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_24689, groups = var_23089, pad = lora_out_357_pad_0, pad_type = lora_out_357_pad_type_0, strides = var_24687, weight = lora_out_359_weight_0_to_fp16, x = input_299_cast_fp16)[name = tensor("lora_out_359_cast_fp16")]; + tensor hidden_states_33_cast_fp16 = add(x = pretrained_out_179_cast_fp16, y = lora_out_359_cast_fp16)[name = tensor("hidden_states_33_cast_fp16")]; + tensor inputs_61_cast_fp16 = add(x = inputs_59_cast_fp16, y = hidden_states_33_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; + tensor var_24699 = const()[name = tensor("op_24699"), val = tensor(3)]; + tensor var_24724 = const()[name = tensor("op_24724"), val = tensor(1)]; + tensor var_24725 = const()[name = tensor("op_24725"), val = tensor(true)]; + tensor var_24735 = const()[name = tensor("op_24735"), val = tensor([1])]; + tensor channels_mean_61_cast_fp16 = reduce_mean(axes = var_24735, keep_dims = var_24725, x = inputs_61_cast_fp16)[name = tensor("channels_mean_61_cast_fp16")]; + tensor zero_mean_61_cast_fp16 = sub(x = inputs_61_cast_fp16, y = channels_mean_61_cast_fp16)[name = tensor("zero_mean_61_cast_fp16")]; + tensor zero_mean_sq_61_cast_fp16 = mul(x = zero_mean_61_cast_fp16, y = zero_mean_61_cast_fp16)[name = tensor("zero_mean_sq_61_cast_fp16")]; + tensor var_24739 = const()[name = tensor("op_24739"), val = tensor([1])]; + tensor var_24740_cast_fp16 = reduce_mean(axes = var_24739, keep_dims = var_24725, x = zero_mean_sq_61_cast_fp16)[name = tensor("op_24740_cast_fp16")]; + tensor var_24741_to_fp16 = const()[name = tensor("op_24741_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_24742_cast_fp16 = add(x = var_24740_cast_fp16, y = var_24741_to_fp16)[name = tensor("op_24742_cast_fp16")]; + tensor denom_61_epsilon_0 = const()[name = tensor("denom_61_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_61_cast_fp16 = rsqrt(epsilon = denom_61_epsilon_0, x = var_24742_cast_fp16)[name = tensor("denom_61_cast_fp16")]; + tensor out_61_cast_fp16 = mul(x = zero_mean_61_cast_fp16, y = denom_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; + tensor obj_61_gamma_0_to_fp16 = const()[name = tensor("obj_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173319296)))]; + tensor obj_61_beta_0_to_fp16 = const()[name = tensor("obj_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173321920)))]; + tensor obj_61_epsilon_0_to_fp16 = const()[name = tensor("obj_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_61_cast_fp16 = batch_norm(beta = obj_61_beta_0_to_fp16, epsilon = obj_61_epsilon_0_to_fp16, gamma = obj_61_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor("obj_61_cast_fp16")]; + tensor var_24760 = const()[name = tensor("op_24760"), val = tensor([1, 1])]; + tensor var_24762 = const()[name = tensor("op_24762"), val = tensor([1, 1])]; + tensor pretrained_out_181_pad_type_0 = const()[name = tensor("pretrained_out_181_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_181_pad_0 = const()[name = tensor("pretrained_out_181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173324544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174143808))), name = tensor("layers_15_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_15_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174143936)))]; + tensor pretrained_out_181_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_24762, groups = var_24724, pad = pretrained_out_181_pad_0, pad_type = pretrained_out_181_pad_type_0, strides = var_24760, weight = layers_15_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_61_cast_fp16)[name = tensor("pretrained_out_181_cast_fp16")]; + tensor var_24766 = const()[name = tensor("op_24766"), val = tensor([1, 1])]; + tensor var_24768 = const()[name = tensor("op_24768"), val = tensor([1, 1])]; + tensor input_301_pad_type_0 = const()[name = tensor("input_301_pad_type_0"), val = tensor("custom")]; + tensor input_301_pad_0 = const()[name = tensor("input_301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174146560)))]; + tensor input_301_cast_fp16 = conv(dilations = var_24768, groups = var_24724, pad = input_301_pad_0, pad_type = input_301_pad_type_0, strides = var_24766, weight = layers_15_self_attn_q_proj_loraA_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor("input_301_cast_fp16")]; + tensor var_24772 = const()[name = tensor("op_24772"), val = tensor([1, 1])]; + tensor var_24774 = const()[name = tensor("op_24774"), val = tensor([1, 1])]; + tensor lora_out_361_pad_type_0 = const()[name = tensor("lora_out_361_pad_type_0"), val = tensor("custom")]; + tensor lora_out_361_pad_0 = const()[name = tensor("lora_out_361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_363_weight_0_to_fp16 = const()[name = tensor("lora_out_363_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174187584)))]; + tensor lora_out_363_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_24774, groups = var_24724, pad = lora_out_361_pad_0, pad_type = lora_out_361_pad_type_0, strides = var_24772, weight = lora_out_363_weight_0_to_fp16, x = input_301_cast_fp16)[name = tensor("lora_out_363_cast_fp16")]; + tensor query_31_cast_fp16 = add(x = pretrained_out_181_cast_fp16, y = lora_out_363_cast_fp16)[name = tensor("query_31_cast_fp16")]; + tensor var_24784 = const()[name = tensor("op_24784"), val = tensor([1, 1])]; + tensor var_24786 = const()[name = tensor("op_24786"), val = tensor([1, 1])]; + tensor pretrained_out_183_pad_type_0 = const()[name = tensor("pretrained_out_183_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_183_pad_0 = const()[name = tensor("pretrained_out_183_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174228608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175047872))), name = tensor("layers_15_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_183_cast_fp16 = conv(dilations = var_24786, groups = var_24724, pad = pretrained_out_183_pad_0, pad_type = pretrained_out_183_pad_type_0, strides = var_24784, weight = layers_15_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_61_cast_fp16)[name = tensor("pretrained_out_183_cast_fp16")]; + tensor var_24790 = const()[name = tensor("op_24790"), val = tensor([1, 1])]; + tensor var_24792 = const()[name = tensor("op_24792"), val = tensor([1, 1])]; + tensor input_303_pad_type_0 = const()[name = tensor("input_303_pad_type_0"), val = tensor("custom")]; + tensor input_303_pad_0 = const()[name = tensor("input_303_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175048000)))]; + tensor input_303_cast_fp16 = conv(dilations = var_24792, groups = var_24724, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = var_24790, weight = layers_15_self_attn_k_proj_loraA_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor("input_303_cast_fp16")]; + tensor var_24796 = const()[name = tensor("op_24796"), val = tensor([1, 1])]; + tensor var_24798 = const()[name = tensor("op_24798"), val = tensor([1, 1])]; + tensor lora_out_365_pad_type_0 = const()[name = tensor("lora_out_365_pad_type_0"), val = tensor("custom")]; + tensor lora_out_365_pad_0 = const()[name = tensor("lora_out_365_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_367_weight_0_to_fp16 = const()[name = tensor("lora_out_367_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175089024)))]; + tensor lora_out_367_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_24798, groups = var_24724, pad = lora_out_365_pad_0, pad_type = lora_out_365_pad_type_0, strides = var_24796, weight = lora_out_367_weight_0_to_fp16, x = input_303_cast_fp16)[name = tensor("lora_out_367_cast_fp16")]; + tensor key_31_cast_fp16 = add(x = pretrained_out_183_cast_fp16, y = lora_out_367_cast_fp16)[name = tensor("key_31_cast_fp16")]; + tensor var_24809 = const()[name = tensor("op_24809"), val = tensor([1, 1])]; + tensor var_24811 = const()[name = tensor("op_24811"), val = tensor([1, 1])]; + tensor pretrained_out_185_pad_type_0 = const()[name = tensor("pretrained_out_185_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_185_pad_0 = const()[name = tensor("pretrained_out_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175130048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175949312))), name = tensor("layers_15_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_15_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175949440)))]; + tensor pretrained_out_185_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_24811, groups = var_24724, pad = pretrained_out_185_pad_0, pad_type = pretrained_out_185_pad_type_0, strides = var_24809, weight = layers_15_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_61_cast_fp16)[name = tensor("pretrained_out_185_cast_fp16")]; + tensor var_24815 = const()[name = tensor("op_24815"), val = tensor([1, 1])]; + tensor var_24817 = const()[name = tensor("op_24817"), val = tensor([1, 1])]; + tensor input_305_pad_type_0 = const()[name = tensor("input_305_pad_type_0"), val = tensor("custom")]; + tensor input_305_pad_0 = const()[name = tensor("input_305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175952064)))]; + tensor input_305_cast_fp16 = conv(dilations = var_24817, groups = var_24724, pad = input_305_pad_0, pad_type = input_305_pad_type_0, strides = var_24815, weight = layers_15_self_attn_v_proj_loraA_weight_to_fp16, x = obj_61_cast_fp16)[name = tensor("input_305_cast_fp16")]; + tensor var_24821 = const()[name = tensor("op_24821"), val = tensor([1, 1])]; + tensor var_24823 = const()[name = tensor("op_24823"), val = tensor([1, 1])]; + tensor lora_out_369_pad_type_0 = const()[name = tensor("lora_out_369_pad_type_0"), val = tensor("custom")]; + tensor lora_out_369_pad_0 = const()[name = tensor("lora_out_369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_371_weight_0_to_fp16 = const()[name = tensor("lora_out_371_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175993088)))]; + tensor lora_out_371_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_24823, groups = var_24724, pad = lora_out_369_pad_0, pad_type = lora_out_369_pad_type_0, strides = var_24821, weight = lora_out_371_weight_0_to_fp16, x = input_305_cast_fp16)[name = tensor("lora_out_371_cast_fp16")]; + tensor value_31_cast_fp16 = add(x = pretrained_out_185_cast_fp16, y = lora_out_371_cast_fp16)[name = tensor("value_31_cast_fp16")]; + tensor var_24833_begin_0 = const()[name = tensor("op_24833_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24833_end_0 = const()[name = tensor("op_24833_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24833_end_mask_0 = const()[name = tensor("op_24833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24833_cast_fp16 = slice_by_index(begin = var_24833_begin_0, end = var_24833_end_0, end_mask = var_24833_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24833_cast_fp16")]; + tensor var_24837_begin_0 = const()[name = tensor("op_24837_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_24837_end_0 = const()[name = tensor("op_24837_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_24837_end_mask_0 = const()[name = tensor("op_24837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24837_cast_fp16 = slice_by_index(begin = var_24837_begin_0, end = var_24837_end_0, end_mask = var_24837_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24837_cast_fp16")]; + tensor var_24841_begin_0 = const()[name = tensor("op_24841_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_24841_end_0 = const()[name = tensor("op_24841_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_24841_end_mask_0 = const()[name = tensor("op_24841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24841_cast_fp16 = slice_by_index(begin = var_24841_begin_0, end = var_24841_end_0, end_mask = var_24841_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24841_cast_fp16")]; + tensor var_24845_begin_0 = const()[name = tensor("op_24845_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_24845_end_0 = const()[name = tensor("op_24845_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_24845_end_mask_0 = const()[name = tensor("op_24845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24845_cast_fp16 = slice_by_index(begin = var_24845_begin_0, end = var_24845_end_0, end_mask = var_24845_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24845_cast_fp16")]; + tensor var_24849_begin_0 = const()[name = tensor("op_24849_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_24849_end_0 = const()[name = tensor("op_24849_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_24849_end_mask_0 = const()[name = tensor("op_24849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24849_cast_fp16 = slice_by_index(begin = var_24849_begin_0, end = var_24849_end_0, end_mask = var_24849_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24849_cast_fp16")]; + tensor var_24853_begin_0 = const()[name = tensor("op_24853_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_24853_end_0 = const()[name = tensor("op_24853_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_24853_end_mask_0 = const()[name = tensor("op_24853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24853_cast_fp16 = slice_by_index(begin = var_24853_begin_0, end = var_24853_end_0, end_mask = var_24853_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24853_cast_fp16")]; + tensor var_24857_begin_0 = const()[name = tensor("op_24857_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_24857_end_0 = const()[name = tensor("op_24857_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_24857_end_mask_0 = const()[name = tensor("op_24857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24857_cast_fp16 = slice_by_index(begin = var_24857_begin_0, end = var_24857_end_0, end_mask = var_24857_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24857_cast_fp16")]; + tensor var_24861_begin_0 = const()[name = tensor("op_24861_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_24861_end_0 = const()[name = tensor("op_24861_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_24861_end_mask_0 = const()[name = tensor("op_24861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24861_cast_fp16 = slice_by_index(begin = var_24861_begin_0, end = var_24861_end_0, end_mask = var_24861_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24861_cast_fp16")]; + tensor var_24865_begin_0 = const()[name = tensor("op_24865_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_24865_end_0 = const()[name = tensor("op_24865_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_24865_end_mask_0 = const()[name = tensor("op_24865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24865_cast_fp16 = slice_by_index(begin = var_24865_begin_0, end = var_24865_end_0, end_mask = var_24865_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24865_cast_fp16")]; + tensor var_24869_begin_0 = const()[name = tensor("op_24869_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_24869_end_0 = const()[name = tensor("op_24869_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_24869_end_mask_0 = const()[name = tensor("op_24869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24869_cast_fp16 = slice_by_index(begin = var_24869_begin_0, end = var_24869_end_0, end_mask = var_24869_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24869_cast_fp16")]; + tensor var_24873_begin_0 = const()[name = tensor("op_24873_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_24873_end_0 = const()[name = tensor("op_24873_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_24873_end_mask_0 = const()[name = tensor("op_24873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24873_cast_fp16 = slice_by_index(begin = var_24873_begin_0, end = var_24873_end_0, end_mask = var_24873_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24873_cast_fp16")]; + tensor var_24877_begin_0 = const()[name = tensor("op_24877_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_24877_end_0 = const()[name = tensor("op_24877_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_24877_end_mask_0 = const()[name = tensor("op_24877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24877_cast_fp16 = slice_by_index(begin = var_24877_begin_0, end = var_24877_end_0, end_mask = var_24877_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24877_cast_fp16")]; + tensor var_24881_begin_0 = const()[name = tensor("op_24881_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_24881_end_0 = const()[name = tensor("op_24881_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_24881_end_mask_0 = const()[name = tensor("op_24881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24881_cast_fp16 = slice_by_index(begin = var_24881_begin_0, end = var_24881_end_0, end_mask = var_24881_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24881_cast_fp16")]; + tensor var_24885_begin_0 = const()[name = tensor("op_24885_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_24885_end_0 = const()[name = tensor("op_24885_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_24885_end_mask_0 = const()[name = tensor("op_24885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24885_cast_fp16 = slice_by_index(begin = var_24885_begin_0, end = var_24885_end_0, end_mask = var_24885_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24885_cast_fp16")]; + tensor var_24889_begin_0 = const()[name = tensor("op_24889_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_24889_end_0 = const()[name = tensor("op_24889_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_24889_end_mask_0 = const()[name = tensor("op_24889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24889_cast_fp16 = slice_by_index(begin = var_24889_begin_0, end = var_24889_end_0, end_mask = var_24889_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24889_cast_fp16")]; + tensor var_24893_begin_0 = const()[name = tensor("op_24893_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_24893_end_0 = const()[name = tensor("op_24893_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_24893_end_mask_0 = const()[name = tensor("op_24893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24893_cast_fp16 = slice_by_index(begin = var_24893_begin_0, end = var_24893_end_0, end_mask = var_24893_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24893_cast_fp16")]; + tensor var_24897_begin_0 = const()[name = tensor("op_24897_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_24897_end_0 = const()[name = tensor("op_24897_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_24897_end_mask_0 = const()[name = tensor("op_24897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24897_cast_fp16 = slice_by_index(begin = var_24897_begin_0, end = var_24897_end_0, end_mask = var_24897_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24897_cast_fp16")]; + tensor var_24901_begin_0 = const()[name = tensor("op_24901_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_24901_end_0 = const()[name = tensor("op_24901_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_24901_end_mask_0 = const()[name = tensor("op_24901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24901_cast_fp16 = slice_by_index(begin = var_24901_begin_0, end = var_24901_end_0, end_mask = var_24901_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24901_cast_fp16")]; + tensor var_24905_begin_0 = const()[name = tensor("op_24905_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_24905_end_0 = const()[name = tensor("op_24905_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_24905_end_mask_0 = const()[name = tensor("op_24905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24905_cast_fp16 = slice_by_index(begin = var_24905_begin_0, end = var_24905_end_0, end_mask = var_24905_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24905_cast_fp16")]; + tensor var_24909_begin_0 = const()[name = tensor("op_24909_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_24909_end_0 = const()[name = tensor("op_24909_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_24909_end_mask_0 = const()[name = tensor("op_24909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_24909_cast_fp16 = slice_by_index(begin = var_24909_begin_0, end = var_24909_end_0, end_mask = var_24909_end_mask_0, x = query_31_cast_fp16)[name = tensor("op_24909_cast_fp16")]; + tensor var_24918_begin_0 = const()[name = tensor("op_24918_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24918_end_0 = const()[name = tensor("op_24918_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24918_end_mask_0 = const()[name = tensor("op_24918_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24918_cast_fp16 = slice_by_index(begin = var_24918_begin_0, end = var_24918_end_0, end_mask = var_24918_end_mask_0, x = var_24833_cast_fp16)[name = tensor("op_24918_cast_fp16")]; + tensor var_24925_begin_0 = const()[name = tensor("op_24925_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24925_end_0 = const()[name = tensor("op_24925_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24925_end_mask_0 = const()[name = tensor("op_24925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24925_cast_fp16 = slice_by_index(begin = var_24925_begin_0, end = var_24925_end_0, end_mask = var_24925_end_mask_0, x = var_24833_cast_fp16)[name = tensor("op_24925_cast_fp16")]; + tensor var_24932_begin_0 = const()[name = tensor("op_24932_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24932_end_0 = const()[name = tensor("op_24932_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24932_end_mask_0 = const()[name = tensor("op_24932_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24932_cast_fp16 = slice_by_index(begin = var_24932_begin_0, end = var_24932_end_0, end_mask = var_24932_end_mask_0, x = var_24833_cast_fp16)[name = tensor("op_24932_cast_fp16")]; + tensor var_24939_begin_0 = const()[name = tensor("op_24939_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24939_end_0 = const()[name = tensor("op_24939_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24939_end_mask_0 = const()[name = tensor("op_24939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24939_cast_fp16 = slice_by_index(begin = var_24939_begin_0, end = var_24939_end_0, end_mask = var_24939_end_mask_0, x = var_24833_cast_fp16)[name = tensor("op_24939_cast_fp16")]; + tensor var_24946_begin_0 = const()[name = tensor("op_24946_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24946_end_0 = const()[name = tensor("op_24946_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24946_end_mask_0 = const()[name = tensor("op_24946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24946_cast_fp16 = slice_by_index(begin = var_24946_begin_0, end = var_24946_end_0, end_mask = var_24946_end_mask_0, x = var_24837_cast_fp16)[name = tensor("op_24946_cast_fp16")]; + tensor var_24953_begin_0 = const()[name = tensor("op_24953_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24953_end_0 = const()[name = tensor("op_24953_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24953_end_mask_0 = const()[name = tensor("op_24953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24953_cast_fp16 = slice_by_index(begin = var_24953_begin_0, end = var_24953_end_0, end_mask = var_24953_end_mask_0, x = var_24837_cast_fp16)[name = tensor("op_24953_cast_fp16")]; + tensor var_24960_begin_0 = const()[name = tensor("op_24960_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24960_end_0 = const()[name = tensor("op_24960_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24960_end_mask_0 = const()[name = tensor("op_24960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24960_cast_fp16 = slice_by_index(begin = var_24960_begin_0, end = var_24960_end_0, end_mask = var_24960_end_mask_0, x = var_24837_cast_fp16)[name = tensor("op_24960_cast_fp16")]; + tensor var_24967_begin_0 = const()[name = tensor("op_24967_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24967_end_0 = const()[name = tensor("op_24967_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24967_end_mask_0 = const()[name = tensor("op_24967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24967_cast_fp16 = slice_by_index(begin = var_24967_begin_0, end = var_24967_end_0, end_mask = var_24967_end_mask_0, x = var_24837_cast_fp16)[name = tensor("op_24967_cast_fp16")]; + tensor var_24974_begin_0 = const()[name = tensor("op_24974_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_24974_end_0 = const()[name = tensor("op_24974_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_24974_end_mask_0 = const()[name = tensor("op_24974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24974_cast_fp16 = slice_by_index(begin = var_24974_begin_0, end = var_24974_end_0, end_mask = var_24974_end_mask_0, x = var_24841_cast_fp16)[name = tensor("op_24974_cast_fp16")]; + tensor var_24981_begin_0 = const()[name = tensor("op_24981_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_24981_end_0 = const()[name = tensor("op_24981_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_24981_end_mask_0 = const()[name = tensor("op_24981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24981_cast_fp16 = slice_by_index(begin = var_24981_begin_0, end = var_24981_end_0, end_mask = var_24981_end_mask_0, x = var_24841_cast_fp16)[name = tensor("op_24981_cast_fp16")]; + tensor var_24988_begin_0 = const()[name = tensor("op_24988_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_24988_end_0 = const()[name = tensor("op_24988_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_24988_end_mask_0 = const()[name = tensor("op_24988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24988_cast_fp16 = slice_by_index(begin = var_24988_begin_0, end = var_24988_end_0, end_mask = var_24988_end_mask_0, x = var_24841_cast_fp16)[name = tensor("op_24988_cast_fp16")]; + tensor var_24995_begin_0 = const()[name = tensor("op_24995_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_24995_end_0 = const()[name = tensor("op_24995_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_24995_end_mask_0 = const()[name = tensor("op_24995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_24995_cast_fp16 = slice_by_index(begin = var_24995_begin_0, end = var_24995_end_0, end_mask = var_24995_end_mask_0, x = var_24841_cast_fp16)[name = tensor("op_24995_cast_fp16")]; + tensor var_25002_begin_0 = const()[name = tensor("op_25002_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25002_end_0 = const()[name = tensor("op_25002_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25002_end_mask_0 = const()[name = tensor("op_25002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25002_cast_fp16 = slice_by_index(begin = var_25002_begin_0, end = var_25002_end_0, end_mask = var_25002_end_mask_0, x = var_24845_cast_fp16)[name = tensor("op_25002_cast_fp16")]; + tensor var_25009_begin_0 = const()[name = tensor("op_25009_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25009_end_0 = const()[name = tensor("op_25009_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25009_end_mask_0 = const()[name = tensor("op_25009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25009_cast_fp16 = slice_by_index(begin = var_25009_begin_0, end = var_25009_end_0, end_mask = var_25009_end_mask_0, x = var_24845_cast_fp16)[name = tensor("op_25009_cast_fp16")]; + tensor var_25016_begin_0 = const()[name = tensor("op_25016_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25016_end_0 = const()[name = tensor("op_25016_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25016_end_mask_0 = const()[name = tensor("op_25016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25016_cast_fp16 = slice_by_index(begin = var_25016_begin_0, end = var_25016_end_0, end_mask = var_25016_end_mask_0, x = var_24845_cast_fp16)[name = tensor("op_25016_cast_fp16")]; + tensor var_25023_begin_0 = const()[name = tensor("op_25023_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25023_end_0 = const()[name = tensor("op_25023_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25023_end_mask_0 = const()[name = tensor("op_25023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25023_cast_fp16 = slice_by_index(begin = var_25023_begin_0, end = var_25023_end_0, end_mask = var_25023_end_mask_0, x = var_24845_cast_fp16)[name = tensor("op_25023_cast_fp16")]; + tensor var_25030_begin_0 = const()[name = tensor("op_25030_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25030_end_0 = const()[name = tensor("op_25030_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25030_end_mask_0 = const()[name = tensor("op_25030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25030_cast_fp16 = slice_by_index(begin = var_25030_begin_0, end = var_25030_end_0, end_mask = var_25030_end_mask_0, x = var_24849_cast_fp16)[name = tensor("op_25030_cast_fp16")]; + tensor var_25037_begin_0 = const()[name = tensor("op_25037_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25037_end_0 = const()[name = tensor("op_25037_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25037_end_mask_0 = const()[name = tensor("op_25037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25037_cast_fp16 = slice_by_index(begin = var_25037_begin_0, end = var_25037_end_0, end_mask = var_25037_end_mask_0, x = var_24849_cast_fp16)[name = tensor("op_25037_cast_fp16")]; + tensor var_25044_begin_0 = const()[name = tensor("op_25044_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25044_end_0 = const()[name = tensor("op_25044_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25044_end_mask_0 = const()[name = tensor("op_25044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25044_cast_fp16 = slice_by_index(begin = var_25044_begin_0, end = var_25044_end_0, end_mask = var_25044_end_mask_0, x = var_24849_cast_fp16)[name = tensor("op_25044_cast_fp16")]; + tensor var_25051_begin_0 = const()[name = tensor("op_25051_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25051_end_0 = const()[name = tensor("op_25051_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25051_end_mask_0 = const()[name = tensor("op_25051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25051_cast_fp16 = slice_by_index(begin = var_25051_begin_0, end = var_25051_end_0, end_mask = var_25051_end_mask_0, x = var_24849_cast_fp16)[name = tensor("op_25051_cast_fp16")]; + tensor var_25058_begin_0 = const()[name = tensor("op_25058_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25058_end_0 = const()[name = tensor("op_25058_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25058_end_mask_0 = const()[name = tensor("op_25058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25058_cast_fp16 = slice_by_index(begin = var_25058_begin_0, end = var_25058_end_0, end_mask = var_25058_end_mask_0, x = var_24853_cast_fp16)[name = tensor("op_25058_cast_fp16")]; + tensor var_25065_begin_0 = const()[name = tensor("op_25065_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25065_end_0 = const()[name = tensor("op_25065_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25065_end_mask_0 = const()[name = tensor("op_25065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25065_cast_fp16 = slice_by_index(begin = var_25065_begin_0, end = var_25065_end_0, end_mask = var_25065_end_mask_0, x = var_24853_cast_fp16)[name = tensor("op_25065_cast_fp16")]; + tensor var_25072_begin_0 = const()[name = tensor("op_25072_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25072_end_0 = const()[name = tensor("op_25072_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25072_end_mask_0 = const()[name = tensor("op_25072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25072_cast_fp16 = slice_by_index(begin = var_25072_begin_0, end = var_25072_end_0, end_mask = var_25072_end_mask_0, x = var_24853_cast_fp16)[name = tensor("op_25072_cast_fp16")]; + tensor var_25079_begin_0 = const()[name = tensor("op_25079_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25079_end_0 = const()[name = tensor("op_25079_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25079_end_mask_0 = const()[name = tensor("op_25079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25079_cast_fp16 = slice_by_index(begin = var_25079_begin_0, end = var_25079_end_0, end_mask = var_25079_end_mask_0, x = var_24853_cast_fp16)[name = tensor("op_25079_cast_fp16")]; + tensor var_25086_begin_0 = const()[name = tensor("op_25086_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25086_end_0 = const()[name = tensor("op_25086_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25086_end_mask_0 = const()[name = tensor("op_25086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25086_cast_fp16 = slice_by_index(begin = var_25086_begin_0, end = var_25086_end_0, end_mask = var_25086_end_mask_0, x = var_24857_cast_fp16)[name = tensor("op_25086_cast_fp16")]; + tensor var_25093_begin_0 = const()[name = tensor("op_25093_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25093_end_0 = const()[name = tensor("op_25093_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25093_end_mask_0 = const()[name = tensor("op_25093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25093_cast_fp16 = slice_by_index(begin = var_25093_begin_0, end = var_25093_end_0, end_mask = var_25093_end_mask_0, x = var_24857_cast_fp16)[name = tensor("op_25093_cast_fp16")]; + tensor var_25100_begin_0 = const()[name = tensor("op_25100_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25100_end_0 = const()[name = tensor("op_25100_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25100_end_mask_0 = const()[name = tensor("op_25100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25100_cast_fp16 = slice_by_index(begin = var_25100_begin_0, end = var_25100_end_0, end_mask = var_25100_end_mask_0, x = var_24857_cast_fp16)[name = tensor("op_25100_cast_fp16")]; + tensor var_25107_begin_0 = const()[name = tensor("op_25107_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25107_end_0 = const()[name = tensor("op_25107_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25107_end_mask_0 = const()[name = tensor("op_25107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25107_cast_fp16 = slice_by_index(begin = var_25107_begin_0, end = var_25107_end_0, end_mask = var_25107_end_mask_0, x = var_24857_cast_fp16)[name = tensor("op_25107_cast_fp16")]; + tensor var_25114_begin_0 = const()[name = tensor("op_25114_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25114_end_0 = const()[name = tensor("op_25114_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25114_end_mask_0 = const()[name = tensor("op_25114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25114_cast_fp16 = slice_by_index(begin = var_25114_begin_0, end = var_25114_end_0, end_mask = var_25114_end_mask_0, x = var_24861_cast_fp16)[name = tensor("op_25114_cast_fp16")]; + tensor var_25121_begin_0 = const()[name = tensor("op_25121_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25121_end_0 = const()[name = tensor("op_25121_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25121_end_mask_0 = const()[name = tensor("op_25121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25121_cast_fp16 = slice_by_index(begin = var_25121_begin_0, end = var_25121_end_0, end_mask = var_25121_end_mask_0, x = var_24861_cast_fp16)[name = tensor("op_25121_cast_fp16")]; + tensor var_25128_begin_0 = const()[name = tensor("op_25128_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25128_end_0 = const()[name = tensor("op_25128_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25128_end_mask_0 = const()[name = tensor("op_25128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25128_cast_fp16 = slice_by_index(begin = var_25128_begin_0, end = var_25128_end_0, end_mask = var_25128_end_mask_0, x = var_24861_cast_fp16)[name = tensor("op_25128_cast_fp16")]; + tensor var_25135_begin_0 = const()[name = tensor("op_25135_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25135_end_0 = const()[name = tensor("op_25135_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25135_end_mask_0 = const()[name = tensor("op_25135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25135_cast_fp16 = slice_by_index(begin = var_25135_begin_0, end = var_25135_end_0, end_mask = var_25135_end_mask_0, x = var_24861_cast_fp16)[name = tensor("op_25135_cast_fp16")]; + tensor var_25142_begin_0 = const()[name = tensor("op_25142_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25142_end_0 = const()[name = tensor("op_25142_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25142_end_mask_0 = const()[name = tensor("op_25142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25142_cast_fp16 = slice_by_index(begin = var_25142_begin_0, end = var_25142_end_0, end_mask = var_25142_end_mask_0, x = var_24865_cast_fp16)[name = tensor("op_25142_cast_fp16")]; + tensor var_25149_begin_0 = const()[name = tensor("op_25149_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25149_end_0 = const()[name = tensor("op_25149_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25149_end_mask_0 = const()[name = tensor("op_25149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25149_cast_fp16 = slice_by_index(begin = var_25149_begin_0, end = var_25149_end_0, end_mask = var_25149_end_mask_0, x = var_24865_cast_fp16)[name = tensor("op_25149_cast_fp16")]; + tensor var_25156_begin_0 = const()[name = tensor("op_25156_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25156_end_0 = const()[name = tensor("op_25156_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25156_end_mask_0 = const()[name = tensor("op_25156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25156_cast_fp16 = slice_by_index(begin = var_25156_begin_0, end = var_25156_end_0, end_mask = var_25156_end_mask_0, x = var_24865_cast_fp16)[name = tensor("op_25156_cast_fp16")]; + tensor var_25163_begin_0 = const()[name = tensor("op_25163_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25163_end_0 = const()[name = tensor("op_25163_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25163_end_mask_0 = const()[name = tensor("op_25163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25163_cast_fp16 = slice_by_index(begin = var_25163_begin_0, end = var_25163_end_0, end_mask = var_25163_end_mask_0, x = var_24865_cast_fp16)[name = tensor("op_25163_cast_fp16")]; + tensor var_25170_begin_0 = const()[name = tensor("op_25170_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25170_end_0 = const()[name = tensor("op_25170_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25170_end_mask_0 = const()[name = tensor("op_25170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25170_cast_fp16 = slice_by_index(begin = var_25170_begin_0, end = var_25170_end_0, end_mask = var_25170_end_mask_0, x = var_24869_cast_fp16)[name = tensor("op_25170_cast_fp16")]; + tensor var_25177_begin_0 = const()[name = tensor("op_25177_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25177_end_0 = const()[name = tensor("op_25177_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25177_end_mask_0 = const()[name = tensor("op_25177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25177_cast_fp16 = slice_by_index(begin = var_25177_begin_0, end = var_25177_end_0, end_mask = var_25177_end_mask_0, x = var_24869_cast_fp16)[name = tensor("op_25177_cast_fp16")]; + tensor var_25184_begin_0 = const()[name = tensor("op_25184_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25184_end_0 = const()[name = tensor("op_25184_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25184_end_mask_0 = const()[name = tensor("op_25184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25184_cast_fp16 = slice_by_index(begin = var_25184_begin_0, end = var_25184_end_0, end_mask = var_25184_end_mask_0, x = var_24869_cast_fp16)[name = tensor("op_25184_cast_fp16")]; + tensor var_25191_begin_0 = const()[name = tensor("op_25191_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25191_end_0 = const()[name = tensor("op_25191_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25191_end_mask_0 = const()[name = tensor("op_25191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25191_cast_fp16 = slice_by_index(begin = var_25191_begin_0, end = var_25191_end_0, end_mask = var_25191_end_mask_0, x = var_24869_cast_fp16)[name = tensor("op_25191_cast_fp16")]; + tensor var_25198_begin_0 = const()[name = tensor("op_25198_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25198_end_0 = const()[name = tensor("op_25198_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25198_end_mask_0 = const()[name = tensor("op_25198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25198_cast_fp16 = slice_by_index(begin = var_25198_begin_0, end = var_25198_end_0, end_mask = var_25198_end_mask_0, x = var_24873_cast_fp16)[name = tensor("op_25198_cast_fp16")]; + tensor var_25205_begin_0 = const()[name = tensor("op_25205_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25205_end_0 = const()[name = tensor("op_25205_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25205_end_mask_0 = const()[name = tensor("op_25205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25205_cast_fp16 = slice_by_index(begin = var_25205_begin_0, end = var_25205_end_0, end_mask = var_25205_end_mask_0, x = var_24873_cast_fp16)[name = tensor("op_25205_cast_fp16")]; + tensor var_25212_begin_0 = const()[name = tensor("op_25212_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25212_end_0 = const()[name = tensor("op_25212_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25212_end_mask_0 = const()[name = tensor("op_25212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25212_cast_fp16 = slice_by_index(begin = var_25212_begin_0, end = var_25212_end_0, end_mask = var_25212_end_mask_0, x = var_24873_cast_fp16)[name = tensor("op_25212_cast_fp16")]; + tensor var_25219_begin_0 = const()[name = tensor("op_25219_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25219_end_0 = const()[name = tensor("op_25219_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25219_end_mask_0 = const()[name = tensor("op_25219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25219_cast_fp16 = slice_by_index(begin = var_25219_begin_0, end = var_25219_end_0, end_mask = var_25219_end_mask_0, x = var_24873_cast_fp16)[name = tensor("op_25219_cast_fp16")]; + tensor var_25226_begin_0 = const()[name = tensor("op_25226_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25226_end_0 = const()[name = tensor("op_25226_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25226_end_mask_0 = const()[name = tensor("op_25226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25226_cast_fp16 = slice_by_index(begin = var_25226_begin_0, end = var_25226_end_0, end_mask = var_25226_end_mask_0, x = var_24877_cast_fp16)[name = tensor("op_25226_cast_fp16")]; + tensor var_25233_begin_0 = const()[name = tensor("op_25233_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25233_end_0 = const()[name = tensor("op_25233_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25233_end_mask_0 = const()[name = tensor("op_25233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25233_cast_fp16 = slice_by_index(begin = var_25233_begin_0, end = var_25233_end_0, end_mask = var_25233_end_mask_0, x = var_24877_cast_fp16)[name = tensor("op_25233_cast_fp16")]; + tensor var_25240_begin_0 = const()[name = tensor("op_25240_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25240_end_0 = const()[name = tensor("op_25240_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25240_end_mask_0 = const()[name = tensor("op_25240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25240_cast_fp16 = slice_by_index(begin = var_25240_begin_0, end = var_25240_end_0, end_mask = var_25240_end_mask_0, x = var_24877_cast_fp16)[name = tensor("op_25240_cast_fp16")]; + tensor var_25247_begin_0 = const()[name = tensor("op_25247_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25247_end_0 = const()[name = tensor("op_25247_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25247_end_mask_0 = const()[name = tensor("op_25247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25247_cast_fp16 = slice_by_index(begin = var_25247_begin_0, end = var_25247_end_0, end_mask = var_25247_end_mask_0, x = var_24877_cast_fp16)[name = tensor("op_25247_cast_fp16")]; + tensor var_25254_begin_0 = const()[name = tensor("op_25254_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25254_end_0 = const()[name = tensor("op_25254_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25254_end_mask_0 = const()[name = tensor("op_25254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25254_cast_fp16 = slice_by_index(begin = var_25254_begin_0, end = var_25254_end_0, end_mask = var_25254_end_mask_0, x = var_24881_cast_fp16)[name = tensor("op_25254_cast_fp16")]; + tensor var_25261_begin_0 = const()[name = tensor("op_25261_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25261_end_0 = const()[name = tensor("op_25261_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25261_end_mask_0 = const()[name = tensor("op_25261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25261_cast_fp16 = slice_by_index(begin = var_25261_begin_0, end = var_25261_end_0, end_mask = var_25261_end_mask_0, x = var_24881_cast_fp16)[name = tensor("op_25261_cast_fp16")]; + tensor var_25268_begin_0 = const()[name = tensor("op_25268_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25268_end_0 = const()[name = tensor("op_25268_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25268_end_mask_0 = const()[name = tensor("op_25268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25268_cast_fp16 = slice_by_index(begin = var_25268_begin_0, end = var_25268_end_0, end_mask = var_25268_end_mask_0, x = var_24881_cast_fp16)[name = tensor("op_25268_cast_fp16")]; + tensor var_25275_begin_0 = const()[name = tensor("op_25275_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25275_end_0 = const()[name = tensor("op_25275_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25275_end_mask_0 = const()[name = tensor("op_25275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25275_cast_fp16 = slice_by_index(begin = var_25275_begin_0, end = var_25275_end_0, end_mask = var_25275_end_mask_0, x = var_24881_cast_fp16)[name = tensor("op_25275_cast_fp16")]; + tensor var_25282_begin_0 = const()[name = tensor("op_25282_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25282_end_0 = const()[name = tensor("op_25282_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25282_end_mask_0 = const()[name = tensor("op_25282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25282_cast_fp16 = slice_by_index(begin = var_25282_begin_0, end = var_25282_end_0, end_mask = var_25282_end_mask_0, x = var_24885_cast_fp16)[name = tensor("op_25282_cast_fp16")]; + tensor var_25289_begin_0 = const()[name = tensor("op_25289_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25289_end_0 = const()[name = tensor("op_25289_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25289_end_mask_0 = const()[name = tensor("op_25289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25289_cast_fp16 = slice_by_index(begin = var_25289_begin_0, end = var_25289_end_0, end_mask = var_25289_end_mask_0, x = var_24885_cast_fp16)[name = tensor("op_25289_cast_fp16")]; + tensor var_25296_begin_0 = const()[name = tensor("op_25296_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25296_end_0 = const()[name = tensor("op_25296_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25296_end_mask_0 = const()[name = tensor("op_25296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25296_cast_fp16 = slice_by_index(begin = var_25296_begin_0, end = var_25296_end_0, end_mask = var_25296_end_mask_0, x = var_24885_cast_fp16)[name = tensor("op_25296_cast_fp16")]; + tensor var_25303_begin_0 = const()[name = tensor("op_25303_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25303_end_0 = const()[name = tensor("op_25303_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25303_end_mask_0 = const()[name = tensor("op_25303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25303_cast_fp16 = slice_by_index(begin = var_25303_begin_0, end = var_25303_end_0, end_mask = var_25303_end_mask_0, x = var_24885_cast_fp16)[name = tensor("op_25303_cast_fp16")]; + tensor var_25310_begin_0 = const()[name = tensor("op_25310_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25310_end_0 = const()[name = tensor("op_25310_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25310_end_mask_0 = const()[name = tensor("op_25310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25310_cast_fp16 = slice_by_index(begin = var_25310_begin_0, end = var_25310_end_0, end_mask = var_25310_end_mask_0, x = var_24889_cast_fp16)[name = tensor("op_25310_cast_fp16")]; + tensor var_25317_begin_0 = const()[name = tensor("op_25317_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25317_end_0 = const()[name = tensor("op_25317_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25317_end_mask_0 = const()[name = tensor("op_25317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25317_cast_fp16 = slice_by_index(begin = var_25317_begin_0, end = var_25317_end_0, end_mask = var_25317_end_mask_0, x = var_24889_cast_fp16)[name = tensor("op_25317_cast_fp16")]; + tensor var_25324_begin_0 = const()[name = tensor("op_25324_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25324_end_0 = const()[name = tensor("op_25324_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25324_end_mask_0 = const()[name = tensor("op_25324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25324_cast_fp16 = slice_by_index(begin = var_25324_begin_0, end = var_25324_end_0, end_mask = var_25324_end_mask_0, x = var_24889_cast_fp16)[name = tensor("op_25324_cast_fp16")]; + tensor var_25331_begin_0 = const()[name = tensor("op_25331_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25331_end_0 = const()[name = tensor("op_25331_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25331_end_mask_0 = const()[name = tensor("op_25331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25331_cast_fp16 = slice_by_index(begin = var_25331_begin_0, end = var_25331_end_0, end_mask = var_25331_end_mask_0, x = var_24889_cast_fp16)[name = tensor("op_25331_cast_fp16")]; + tensor var_25338_begin_0 = const()[name = tensor("op_25338_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25338_end_0 = const()[name = tensor("op_25338_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25338_end_mask_0 = const()[name = tensor("op_25338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25338_cast_fp16 = slice_by_index(begin = var_25338_begin_0, end = var_25338_end_0, end_mask = var_25338_end_mask_0, x = var_24893_cast_fp16)[name = tensor("op_25338_cast_fp16")]; + tensor var_25345_begin_0 = const()[name = tensor("op_25345_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25345_end_0 = const()[name = tensor("op_25345_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25345_end_mask_0 = const()[name = tensor("op_25345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25345_cast_fp16 = slice_by_index(begin = var_25345_begin_0, end = var_25345_end_0, end_mask = var_25345_end_mask_0, x = var_24893_cast_fp16)[name = tensor("op_25345_cast_fp16")]; + tensor var_25352_begin_0 = const()[name = tensor("op_25352_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25352_end_0 = const()[name = tensor("op_25352_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25352_end_mask_0 = const()[name = tensor("op_25352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25352_cast_fp16 = slice_by_index(begin = var_25352_begin_0, end = var_25352_end_0, end_mask = var_25352_end_mask_0, x = var_24893_cast_fp16)[name = tensor("op_25352_cast_fp16")]; + tensor var_25359_begin_0 = const()[name = tensor("op_25359_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25359_end_0 = const()[name = tensor("op_25359_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25359_end_mask_0 = const()[name = tensor("op_25359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25359_cast_fp16 = slice_by_index(begin = var_25359_begin_0, end = var_25359_end_0, end_mask = var_25359_end_mask_0, x = var_24893_cast_fp16)[name = tensor("op_25359_cast_fp16")]; + tensor var_25366_begin_0 = const()[name = tensor("op_25366_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25366_end_0 = const()[name = tensor("op_25366_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25366_end_mask_0 = const()[name = tensor("op_25366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25366_cast_fp16 = slice_by_index(begin = var_25366_begin_0, end = var_25366_end_0, end_mask = var_25366_end_mask_0, x = var_24897_cast_fp16)[name = tensor("op_25366_cast_fp16")]; + tensor var_25373_begin_0 = const()[name = tensor("op_25373_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25373_end_0 = const()[name = tensor("op_25373_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25373_end_mask_0 = const()[name = tensor("op_25373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25373_cast_fp16 = slice_by_index(begin = var_25373_begin_0, end = var_25373_end_0, end_mask = var_25373_end_mask_0, x = var_24897_cast_fp16)[name = tensor("op_25373_cast_fp16")]; + tensor var_25380_begin_0 = const()[name = tensor("op_25380_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25380_end_0 = const()[name = tensor("op_25380_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25380_end_mask_0 = const()[name = tensor("op_25380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25380_cast_fp16 = slice_by_index(begin = var_25380_begin_0, end = var_25380_end_0, end_mask = var_25380_end_mask_0, x = var_24897_cast_fp16)[name = tensor("op_25380_cast_fp16")]; + tensor var_25387_begin_0 = const()[name = tensor("op_25387_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25387_end_0 = const()[name = tensor("op_25387_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25387_end_mask_0 = const()[name = tensor("op_25387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25387_cast_fp16 = slice_by_index(begin = var_25387_begin_0, end = var_25387_end_0, end_mask = var_25387_end_mask_0, x = var_24897_cast_fp16)[name = tensor("op_25387_cast_fp16")]; + tensor var_25394_begin_0 = const()[name = tensor("op_25394_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25394_end_0 = const()[name = tensor("op_25394_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25394_end_mask_0 = const()[name = tensor("op_25394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25394_cast_fp16 = slice_by_index(begin = var_25394_begin_0, end = var_25394_end_0, end_mask = var_25394_end_mask_0, x = var_24901_cast_fp16)[name = tensor("op_25394_cast_fp16")]; + tensor var_25401_begin_0 = const()[name = tensor("op_25401_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25401_end_0 = const()[name = tensor("op_25401_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25401_end_mask_0 = const()[name = tensor("op_25401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25401_cast_fp16 = slice_by_index(begin = var_25401_begin_0, end = var_25401_end_0, end_mask = var_25401_end_mask_0, x = var_24901_cast_fp16)[name = tensor("op_25401_cast_fp16")]; + tensor var_25408_begin_0 = const()[name = tensor("op_25408_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25408_end_0 = const()[name = tensor("op_25408_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25408_end_mask_0 = const()[name = tensor("op_25408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25408_cast_fp16 = slice_by_index(begin = var_25408_begin_0, end = var_25408_end_0, end_mask = var_25408_end_mask_0, x = var_24901_cast_fp16)[name = tensor("op_25408_cast_fp16")]; + tensor var_25415_begin_0 = const()[name = tensor("op_25415_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25415_end_0 = const()[name = tensor("op_25415_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25415_end_mask_0 = const()[name = tensor("op_25415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25415_cast_fp16 = slice_by_index(begin = var_25415_begin_0, end = var_25415_end_0, end_mask = var_25415_end_mask_0, x = var_24901_cast_fp16)[name = tensor("op_25415_cast_fp16")]; + tensor var_25422_begin_0 = const()[name = tensor("op_25422_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25422_end_0 = const()[name = tensor("op_25422_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25422_end_mask_0 = const()[name = tensor("op_25422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25422_cast_fp16 = slice_by_index(begin = var_25422_begin_0, end = var_25422_end_0, end_mask = var_25422_end_mask_0, x = var_24905_cast_fp16)[name = tensor("op_25422_cast_fp16")]; + tensor var_25429_begin_0 = const()[name = tensor("op_25429_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25429_end_0 = const()[name = tensor("op_25429_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25429_end_mask_0 = const()[name = tensor("op_25429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25429_cast_fp16 = slice_by_index(begin = var_25429_begin_0, end = var_25429_end_0, end_mask = var_25429_end_mask_0, x = var_24905_cast_fp16)[name = tensor("op_25429_cast_fp16")]; + tensor var_25436_begin_0 = const()[name = tensor("op_25436_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25436_end_0 = const()[name = tensor("op_25436_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25436_end_mask_0 = const()[name = tensor("op_25436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25436_cast_fp16 = slice_by_index(begin = var_25436_begin_0, end = var_25436_end_0, end_mask = var_25436_end_mask_0, x = var_24905_cast_fp16)[name = tensor("op_25436_cast_fp16")]; + tensor var_25443_begin_0 = const()[name = tensor("op_25443_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25443_end_0 = const()[name = tensor("op_25443_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25443_end_mask_0 = const()[name = tensor("op_25443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25443_cast_fp16 = slice_by_index(begin = var_25443_begin_0, end = var_25443_end_0, end_mask = var_25443_end_mask_0, x = var_24905_cast_fp16)[name = tensor("op_25443_cast_fp16")]; + tensor var_25450_begin_0 = const()[name = tensor("op_25450_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25450_end_0 = const()[name = tensor("op_25450_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_25450_end_mask_0 = const()[name = tensor("op_25450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25450_cast_fp16 = slice_by_index(begin = var_25450_begin_0, end = var_25450_end_0, end_mask = var_25450_end_mask_0, x = var_24909_cast_fp16)[name = tensor("op_25450_cast_fp16")]; + tensor var_25457_begin_0 = const()[name = tensor("op_25457_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_25457_end_0 = const()[name = tensor("op_25457_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_25457_end_mask_0 = const()[name = tensor("op_25457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25457_cast_fp16 = slice_by_index(begin = var_25457_begin_0, end = var_25457_end_0, end_mask = var_25457_end_mask_0, x = var_24909_cast_fp16)[name = tensor("op_25457_cast_fp16")]; + tensor var_25464_begin_0 = const()[name = tensor("op_25464_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_25464_end_0 = const()[name = tensor("op_25464_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_25464_end_mask_0 = const()[name = tensor("op_25464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25464_cast_fp16 = slice_by_index(begin = var_25464_begin_0, end = var_25464_end_0, end_mask = var_25464_end_mask_0, x = var_24909_cast_fp16)[name = tensor("op_25464_cast_fp16")]; + tensor var_25471_begin_0 = const()[name = tensor("op_25471_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_25471_end_0 = const()[name = tensor("op_25471_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25471_end_mask_0 = const()[name = tensor("op_25471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25471_cast_fp16 = slice_by_index(begin = var_25471_begin_0, end = var_25471_end_0, end_mask = var_25471_end_mask_0, x = var_24909_cast_fp16)[name = tensor("op_25471_cast_fp16")]; + tensor k_31_perm_0 = const()[name = tensor("k_31_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_25476_begin_0 = const()[name = tensor("op_25476_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25476_end_0 = const()[name = tensor("op_25476_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_25476_end_mask_0 = const()[name = tensor("op_25476_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_16 = transpose(perm = k_31_perm_0, x = key_31_cast_fp16)[name = tensor("transpose_16")]; + tensor var_25476_cast_fp16 = slice_by_index(begin = var_25476_begin_0, end = var_25476_end_0, end_mask = var_25476_end_mask_0, x = transpose_16)[name = tensor("op_25476_cast_fp16")]; + tensor var_25480_begin_0 = const()[name = tensor("op_25480_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_25480_end_0 = const()[name = tensor("op_25480_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_25480_end_mask_0 = const()[name = tensor("op_25480_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25480_cast_fp16 = slice_by_index(begin = var_25480_begin_0, end = var_25480_end_0, end_mask = var_25480_end_mask_0, x = transpose_16)[name = tensor("op_25480_cast_fp16")]; + tensor var_25484_begin_0 = const()[name = tensor("op_25484_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_25484_end_0 = const()[name = tensor("op_25484_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_25484_end_mask_0 = const()[name = tensor("op_25484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25484_cast_fp16 = slice_by_index(begin = var_25484_begin_0, end = var_25484_end_0, end_mask = var_25484_end_mask_0, x = transpose_16)[name = tensor("op_25484_cast_fp16")]; + tensor var_25488_begin_0 = const()[name = tensor("op_25488_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_25488_end_0 = const()[name = tensor("op_25488_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_25488_end_mask_0 = const()[name = tensor("op_25488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25488_cast_fp16 = slice_by_index(begin = var_25488_begin_0, end = var_25488_end_0, end_mask = var_25488_end_mask_0, x = transpose_16)[name = tensor("op_25488_cast_fp16")]; + tensor var_25492_begin_0 = const()[name = tensor("op_25492_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_25492_end_0 = const()[name = tensor("op_25492_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_25492_end_mask_0 = const()[name = tensor("op_25492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25492_cast_fp16 = slice_by_index(begin = var_25492_begin_0, end = var_25492_end_0, end_mask = var_25492_end_mask_0, x = transpose_16)[name = tensor("op_25492_cast_fp16")]; + tensor var_25496_begin_0 = const()[name = tensor("op_25496_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_25496_end_0 = const()[name = tensor("op_25496_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_25496_end_mask_0 = const()[name = tensor("op_25496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25496_cast_fp16 = slice_by_index(begin = var_25496_begin_0, end = var_25496_end_0, end_mask = var_25496_end_mask_0, x = transpose_16)[name = tensor("op_25496_cast_fp16")]; + tensor var_25500_begin_0 = const()[name = tensor("op_25500_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_25500_end_0 = const()[name = tensor("op_25500_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_25500_end_mask_0 = const()[name = tensor("op_25500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25500_cast_fp16 = slice_by_index(begin = var_25500_begin_0, end = var_25500_end_0, end_mask = var_25500_end_mask_0, x = transpose_16)[name = tensor("op_25500_cast_fp16")]; + tensor var_25504_begin_0 = const()[name = tensor("op_25504_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_25504_end_0 = const()[name = tensor("op_25504_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_25504_end_mask_0 = const()[name = tensor("op_25504_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25504_cast_fp16 = slice_by_index(begin = var_25504_begin_0, end = var_25504_end_0, end_mask = var_25504_end_mask_0, x = transpose_16)[name = tensor("op_25504_cast_fp16")]; + tensor var_25508_begin_0 = const()[name = tensor("op_25508_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_25508_end_0 = const()[name = tensor("op_25508_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_25508_end_mask_0 = const()[name = tensor("op_25508_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25508_cast_fp16 = slice_by_index(begin = var_25508_begin_0, end = var_25508_end_0, end_mask = var_25508_end_mask_0, x = transpose_16)[name = tensor("op_25508_cast_fp16")]; + tensor var_25512_begin_0 = const()[name = tensor("op_25512_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_25512_end_0 = const()[name = tensor("op_25512_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_25512_end_mask_0 = const()[name = tensor("op_25512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25512_cast_fp16 = slice_by_index(begin = var_25512_begin_0, end = var_25512_end_0, end_mask = var_25512_end_mask_0, x = transpose_16)[name = tensor("op_25512_cast_fp16")]; + tensor var_25516_begin_0 = const()[name = tensor("op_25516_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_25516_end_0 = const()[name = tensor("op_25516_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_25516_end_mask_0 = const()[name = tensor("op_25516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25516_cast_fp16 = slice_by_index(begin = var_25516_begin_0, end = var_25516_end_0, end_mask = var_25516_end_mask_0, x = transpose_16)[name = tensor("op_25516_cast_fp16")]; + tensor var_25520_begin_0 = const()[name = tensor("op_25520_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_25520_end_0 = const()[name = tensor("op_25520_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_25520_end_mask_0 = const()[name = tensor("op_25520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25520_cast_fp16 = slice_by_index(begin = var_25520_begin_0, end = var_25520_end_0, end_mask = var_25520_end_mask_0, x = transpose_16)[name = tensor("op_25520_cast_fp16")]; + tensor var_25524_begin_0 = const()[name = tensor("op_25524_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_25524_end_0 = const()[name = tensor("op_25524_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_25524_end_mask_0 = const()[name = tensor("op_25524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25524_cast_fp16 = slice_by_index(begin = var_25524_begin_0, end = var_25524_end_0, end_mask = var_25524_end_mask_0, x = transpose_16)[name = tensor("op_25524_cast_fp16")]; + tensor var_25528_begin_0 = const()[name = tensor("op_25528_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_25528_end_0 = const()[name = tensor("op_25528_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_25528_end_mask_0 = const()[name = tensor("op_25528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25528_cast_fp16 = slice_by_index(begin = var_25528_begin_0, end = var_25528_end_0, end_mask = var_25528_end_mask_0, x = transpose_16)[name = tensor("op_25528_cast_fp16")]; + tensor var_25532_begin_0 = const()[name = tensor("op_25532_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_25532_end_0 = const()[name = tensor("op_25532_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_25532_end_mask_0 = const()[name = tensor("op_25532_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25532_cast_fp16 = slice_by_index(begin = var_25532_begin_0, end = var_25532_end_0, end_mask = var_25532_end_mask_0, x = transpose_16)[name = tensor("op_25532_cast_fp16")]; + tensor var_25536_begin_0 = const()[name = tensor("op_25536_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_25536_end_0 = const()[name = tensor("op_25536_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_25536_end_mask_0 = const()[name = tensor("op_25536_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25536_cast_fp16 = slice_by_index(begin = var_25536_begin_0, end = var_25536_end_0, end_mask = var_25536_end_mask_0, x = transpose_16)[name = tensor("op_25536_cast_fp16")]; + tensor var_25540_begin_0 = const()[name = tensor("op_25540_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_25540_end_0 = const()[name = tensor("op_25540_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_25540_end_mask_0 = const()[name = tensor("op_25540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25540_cast_fp16 = slice_by_index(begin = var_25540_begin_0, end = var_25540_end_0, end_mask = var_25540_end_mask_0, x = transpose_16)[name = tensor("op_25540_cast_fp16")]; + tensor var_25544_begin_0 = const()[name = tensor("op_25544_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_25544_end_0 = const()[name = tensor("op_25544_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_25544_end_mask_0 = const()[name = tensor("op_25544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25544_cast_fp16 = slice_by_index(begin = var_25544_begin_0, end = var_25544_end_0, end_mask = var_25544_end_mask_0, x = transpose_16)[name = tensor("op_25544_cast_fp16")]; + tensor var_25548_begin_0 = const()[name = tensor("op_25548_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_25548_end_0 = const()[name = tensor("op_25548_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_25548_end_mask_0 = const()[name = tensor("op_25548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25548_cast_fp16 = slice_by_index(begin = var_25548_begin_0, end = var_25548_end_0, end_mask = var_25548_end_mask_0, x = transpose_16)[name = tensor("op_25548_cast_fp16")]; + tensor var_25552_begin_0 = const()[name = tensor("op_25552_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_25552_end_0 = const()[name = tensor("op_25552_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_25552_end_mask_0 = const()[name = tensor("op_25552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_25552_cast_fp16 = slice_by_index(begin = var_25552_begin_0, end = var_25552_end_0, end_mask = var_25552_end_mask_0, x = transpose_16)[name = tensor("op_25552_cast_fp16")]; + tensor var_25554_begin_0 = const()[name = tensor("op_25554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_25554_end_0 = const()[name = tensor("op_25554_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_25554_end_mask_0 = const()[name = tensor("op_25554_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25554_cast_fp16 = slice_by_index(begin = var_25554_begin_0, end = var_25554_end_0, end_mask = var_25554_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25554_cast_fp16")]; + tensor var_25558_begin_0 = const()[name = tensor("op_25558_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_25558_end_0 = const()[name = tensor("op_25558_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_25558_end_mask_0 = const()[name = tensor("op_25558_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25558_cast_fp16 = slice_by_index(begin = var_25558_begin_0, end = var_25558_end_0, end_mask = var_25558_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25558_cast_fp16")]; + tensor var_25562_begin_0 = const()[name = tensor("op_25562_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_25562_end_0 = const()[name = tensor("op_25562_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_25562_end_mask_0 = const()[name = tensor("op_25562_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25562_cast_fp16 = slice_by_index(begin = var_25562_begin_0, end = var_25562_end_0, end_mask = var_25562_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25562_cast_fp16")]; + tensor var_25566_begin_0 = const()[name = tensor("op_25566_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_25566_end_0 = const()[name = tensor("op_25566_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_25566_end_mask_0 = const()[name = tensor("op_25566_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25566_cast_fp16 = slice_by_index(begin = var_25566_begin_0, end = var_25566_end_0, end_mask = var_25566_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25566_cast_fp16")]; + tensor var_25570_begin_0 = const()[name = tensor("op_25570_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_25570_end_0 = const()[name = tensor("op_25570_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_25570_end_mask_0 = const()[name = tensor("op_25570_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25570_cast_fp16 = slice_by_index(begin = var_25570_begin_0, end = var_25570_end_0, end_mask = var_25570_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25570_cast_fp16")]; + tensor var_25574_begin_0 = const()[name = tensor("op_25574_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_25574_end_0 = const()[name = tensor("op_25574_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_25574_end_mask_0 = const()[name = tensor("op_25574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25574_cast_fp16 = slice_by_index(begin = var_25574_begin_0, end = var_25574_end_0, end_mask = var_25574_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25574_cast_fp16")]; + tensor var_25578_begin_0 = const()[name = tensor("op_25578_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_25578_end_0 = const()[name = tensor("op_25578_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_25578_end_mask_0 = const()[name = tensor("op_25578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25578_cast_fp16 = slice_by_index(begin = var_25578_begin_0, end = var_25578_end_0, end_mask = var_25578_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25578_cast_fp16")]; + tensor var_25582_begin_0 = const()[name = tensor("op_25582_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_25582_end_0 = const()[name = tensor("op_25582_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_25582_end_mask_0 = const()[name = tensor("op_25582_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25582_cast_fp16 = slice_by_index(begin = var_25582_begin_0, end = var_25582_end_0, end_mask = var_25582_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25582_cast_fp16")]; + tensor var_25586_begin_0 = const()[name = tensor("op_25586_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_25586_end_0 = const()[name = tensor("op_25586_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_25586_end_mask_0 = const()[name = tensor("op_25586_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25586_cast_fp16 = slice_by_index(begin = var_25586_begin_0, end = var_25586_end_0, end_mask = var_25586_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25586_cast_fp16")]; + tensor var_25590_begin_0 = const()[name = tensor("op_25590_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_25590_end_0 = const()[name = tensor("op_25590_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_25590_end_mask_0 = const()[name = tensor("op_25590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25590_cast_fp16 = slice_by_index(begin = var_25590_begin_0, end = var_25590_end_0, end_mask = var_25590_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25590_cast_fp16")]; + tensor var_25594_begin_0 = const()[name = tensor("op_25594_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_25594_end_0 = const()[name = tensor("op_25594_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_25594_end_mask_0 = const()[name = tensor("op_25594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25594_cast_fp16 = slice_by_index(begin = var_25594_begin_0, end = var_25594_end_0, end_mask = var_25594_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25594_cast_fp16")]; + tensor var_25598_begin_0 = const()[name = tensor("op_25598_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_25598_end_0 = const()[name = tensor("op_25598_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_25598_end_mask_0 = const()[name = tensor("op_25598_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25598_cast_fp16 = slice_by_index(begin = var_25598_begin_0, end = var_25598_end_0, end_mask = var_25598_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25598_cast_fp16")]; + tensor var_25602_begin_0 = const()[name = tensor("op_25602_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_25602_end_0 = const()[name = tensor("op_25602_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_25602_end_mask_0 = const()[name = tensor("op_25602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25602_cast_fp16 = slice_by_index(begin = var_25602_begin_0, end = var_25602_end_0, end_mask = var_25602_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25602_cast_fp16")]; + tensor var_25606_begin_0 = const()[name = tensor("op_25606_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_25606_end_0 = const()[name = tensor("op_25606_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_25606_end_mask_0 = const()[name = tensor("op_25606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25606_cast_fp16 = slice_by_index(begin = var_25606_begin_0, end = var_25606_end_0, end_mask = var_25606_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25606_cast_fp16")]; + tensor var_25610_begin_0 = const()[name = tensor("op_25610_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_25610_end_0 = const()[name = tensor("op_25610_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_25610_end_mask_0 = const()[name = tensor("op_25610_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25610_cast_fp16 = slice_by_index(begin = var_25610_begin_0, end = var_25610_end_0, end_mask = var_25610_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25610_cast_fp16")]; + tensor var_25614_begin_0 = const()[name = tensor("op_25614_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_25614_end_0 = const()[name = tensor("op_25614_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_25614_end_mask_0 = const()[name = tensor("op_25614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25614_cast_fp16 = slice_by_index(begin = var_25614_begin_0, end = var_25614_end_0, end_mask = var_25614_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25614_cast_fp16")]; + tensor var_25618_begin_0 = const()[name = tensor("op_25618_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_25618_end_0 = const()[name = tensor("op_25618_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_25618_end_mask_0 = const()[name = tensor("op_25618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25618_cast_fp16 = slice_by_index(begin = var_25618_begin_0, end = var_25618_end_0, end_mask = var_25618_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25618_cast_fp16")]; + tensor var_25622_begin_0 = const()[name = tensor("op_25622_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_25622_end_0 = const()[name = tensor("op_25622_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_25622_end_mask_0 = const()[name = tensor("op_25622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25622_cast_fp16 = slice_by_index(begin = var_25622_begin_0, end = var_25622_end_0, end_mask = var_25622_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25622_cast_fp16")]; + tensor var_25626_begin_0 = const()[name = tensor("op_25626_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_25626_end_0 = const()[name = tensor("op_25626_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_25626_end_mask_0 = const()[name = tensor("op_25626_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25626_cast_fp16 = slice_by_index(begin = var_25626_begin_0, end = var_25626_end_0, end_mask = var_25626_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25626_cast_fp16")]; + tensor var_25630_begin_0 = const()[name = tensor("op_25630_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_25630_end_0 = const()[name = tensor("op_25630_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_25630_end_mask_0 = const()[name = tensor("op_25630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_25630_cast_fp16 = slice_by_index(begin = var_25630_begin_0, end = var_25630_end_0, end_mask = var_25630_end_mask_0, x = value_31_cast_fp16)[name = tensor("op_25630_cast_fp16")]; + tensor var_25634_equation_0 = const()[name = tensor("op_25634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25634_cast_fp16 = einsum(equation = var_25634_equation_0, values = (var_25476_cast_fp16, var_24918_cast_fp16))[name = tensor("op_25634_cast_fp16")]; + tensor var_25635_to_fp16 = const()[name = tensor("op_25635_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2401_cast_fp16 = mul(x = var_25634_cast_fp16, y = var_25635_to_fp16)[name = tensor("aw_chunk_2401_cast_fp16")]; + tensor var_25638_equation_0 = const()[name = tensor("op_25638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25638_cast_fp16 = einsum(equation = var_25638_equation_0, values = (var_25476_cast_fp16, var_24925_cast_fp16))[name = tensor("op_25638_cast_fp16")]; + tensor var_25639_to_fp16 = const()[name = tensor("op_25639_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2403_cast_fp16 = mul(x = var_25638_cast_fp16, y = var_25639_to_fp16)[name = tensor("aw_chunk_2403_cast_fp16")]; + tensor var_25642_equation_0 = const()[name = tensor("op_25642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25642_cast_fp16 = einsum(equation = var_25642_equation_0, values = (var_25476_cast_fp16, var_24932_cast_fp16))[name = tensor("op_25642_cast_fp16")]; + tensor var_25643_to_fp16 = const()[name = tensor("op_25643_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2405_cast_fp16 = mul(x = var_25642_cast_fp16, y = var_25643_to_fp16)[name = tensor("aw_chunk_2405_cast_fp16")]; + tensor var_25646_equation_0 = const()[name = tensor("op_25646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25646_cast_fp16 = einsum(equation = var_25646_equation_0, values = (var_25476_cast_fp16, var_24939_cast_fp16))[name = tensor("op_25646_cast_fp16")]; + tensor var_25647_to_fp16 = const()[name = tensor("op_25647_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2407_cast_fp16 = mul(x = var_25646_cast_fp16, y = var_25647_to_fp16)[name = tensor("aw_chunk_2407_cast_fp16")]; + tensor var_25650_equation_0 = const()[name = tensor("op_25650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25650_cast_fp16 = einsum(equation = var_25650_equation_0, values = (var_25480_cast_fp16, var_24946_cast_fp16))[name = tensor("op_25650_cast_fp16")]; + tensor var_25651_to_fp16 = const()[name = tensor("op_25651_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2409_cast_fp16 = mul(x = var_25650_cast_fp16, y = var_25651_to_fp16)[name = tensor("aw_chunk_2409_cast_fp16")]; + tensor var_25654_equation_0 = const()[name = tensor("op_25654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25654_cast_fp16 = einsum(equation = var_25654_equation_0, values = (var_25480_cast_fp16, var_24953_cast_fp16))[name = tensor("op_25654_cast_fp16")]; + tensor var_25655_to_fp16 = const()[name = tensor("op_25655_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2411_cast_fp16 = mul(x = var_25654_cast_fp16, y = var_25655_to_fp16)[name = tensor("aw_chunk_2411_cast_fp16")]; + tensor var_25658_equation_0 = const()[name = tensor("op_25658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25658_cast_fp16 = einsum(equation = var_25658_equation_0, values = (var_25480_cast_fp16, var_24960_cast_fp16))[name = tensor("op_25658_cast_fp16")]; + tensor var_25659_to_fp16 = const()[name = tensor("op_25659_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2413_cast_fp16 = mul(x = var_25658_cast_fp16, y = var_25659_to_fp16)[name = tensor("aw_chunk_2413_cast_fp16")]; + tensor var_25662_equation_0 = const()[name = tensor("op_25662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25662_cast_fp16 = einsum(equation = var_25662_equation_0, values = (var_25480_cast_fp16, var_24967_cast_fp16))[name = tensor("op_25662_cast_fp16")]; + tensor var_25663_to_fp16 = const()[name = tensor("op_25663_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2415_cast_fp16 = mul(x = var_25662_cast_fp16, y = var_25663_to_fp16)[name = tensor("aw_chunk_2415_cast_fp16")]; + tensor var_25666_equation_0 = const()[name = tensor("op_25666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25666_cast_fp16 = einsum(equation = var_25666_equation_0, values = (var_25484_cast_fp16, var_24974_cast_fp16))[name = tensor("op_25666_cast_fp16")]; + tensor var_25667_to_fp16 = const()[name = tensor("op_25667_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2417_cast_fp16 = mul(x = var_25666_cast_fp16, y = var_25667_to_fp16)[name = tensor("aw_chunk_2417_cast_fp16")]; + tensor var_25670_equation_0 = const()[name = tensor("op_25670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25670_cast_fp16 = einsum(equation = var_25670_equation_0, values = (var_25484_cast_fp16, var_24981_cast_fp16))[name = tensor("op_25670_cast_fp16")]; + tensor var_25671_to_fp16 = const()[name = tensor("op_25671_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2419_cast_fp16 = mul(x = var_25670_cast_fp16, y = var_25671_to_fp16)[name = tensor("aw_chunk_2419_cast_fp16")]; + tensor var_25674_equation_0 = const()[name = tensor("op_25674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25674_cast_fp16 = einsum(equation = var_25674_equation_0, values = (var_25484_cast_fp16, var_24988_cast_fp16))[name = tensor("op_25674_cast_fp16")]; + tensor var_25675_to_fp16 = const()[name = tensor("op_25675_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2421_cast_fp16 = mul(x = var_25674_cast_fp16, y = var_25675_to_fp16)[name = tensor("aw_chunk_2421_cast_fp16")]; + tensor var_25678_equation_0 = const()[name = tensor("op_25678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25678_cast_fp16 = einsum(equation = var_25678_equation_0, values = (var_25484_cast_fp16, var_24995_cast_fp16))[name = tensor("op_25678_cast_fp16")]; + tensor var_25679_to_fp16 = const()[name = tensor("op_25679_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2423_cast_fp16 = mul(x = var_25678_cast_fp16, y = var_25679_to_fp16)[name = tensor("aw_chunk_2423_cast_fp16")]; + tensor var_25682_equation_0 = const()[name = tensor("op_25682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25682_cast_fp16 = einsum(equation = var_25682_equation_0, values = (var_25488_cast_fp16, var_25002_cast_fp16))[name = tensor("op_25682_cast_fp16")]; + tensor var_25683_to_fp16 = const()[name = tensor("op_25683_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2425_cast_fp16 = mul(x = var_25682_cast_fp16, y = var_25683_to_fp16)[name = tensor("aw_chunk_2425_cast_fp16")]; + tensor var_25686_equation_0 = const()[name = tensor("op_25686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25686_cast_fp16 = einsum(equation = var_25686_equation_0, values = (var_25488_cast_fp16, var_25009_cast_fp16))[name = tensor("op_25686_cast_fp16")]; + tensor var_25687_to_fp16 = const()[name = tensor("op_25687_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2427_cast_fp16 = mul(x = var_25686_cast_fp16, y = var_25687_to_fp16)[name = tensor("aw_chunk_2427_cast_fp16")]; + tensor var_25690_equation_0 = const()[name = tensor("op_25690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25690_cast_fp16 = einsum(equation = var_25690_equation_0, values = (var_25488_cast_fp16, var_25016_cast_fp16))[name = tensor("op_25690_cast_fp16")]; + tensor var_25691_to_fp16 = const()[name = tensor("op_25691_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2429_cast_fp16 = mul(x = var_25690_cast_fp16, y = var_25691_to_fp16)[name = tensor("aw_chunk_2429_cast_fp16")]; + tensor var_25694_equation_0 = const()[name = tensor("op_25694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25694_cast_fp16 = einsum(equation = var_25694_equation_0, values = (var_25488_cast_fp16, var_25023_cast_fp16))[name = tensor("op_25694_cast_fp16")]; + tensor var_25695_to_fp16 = const()[name = tensor("op_25695_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2431_cast_fp16 = mul(x = var_25694_cast_fp16, y = var_25695_to_fp16)[name = tensor("aw_chunk_2431_cast_fp16")]; + tensor var_25698_equation_0 = const()[name = tensor("op_25698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25698_cast_fp16 = einsum(equation = var_25698_equation_0, values = (var_25492_cast_fp16, var_25030_cast_fp16))[name = tensor("op_25698_cast_fp16")]; + tensor var_25699_to_fp16 = const()[name = tensor("op_25699_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2433_cast_fp16 = mul(x = var_25698_cast_fp16, y = var_25699_to_fp16)[name = tensor("aw_chunk_2433_cast_fp16")]; + tensor var_25702_equation_0 = const()[name = tensor("op_25702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25702_cast_fp16 = einsum(equation = var_25702_equation_0, values = (var_25492_cast_fp16, var_25037_cast_fp16))[name = tensor("op_25702_cast_fp16")]; + tensor var_25703_to_fp16 = const()[name = tensor("op_25703_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2435_cast_fp16 = mul(x = var_25702_cast_fp16, y = var_25703_to_fp16)[name = tensor("aw_chunk_2435_cast_fp16")]; + tensor var_25706_equation_0 = const()[name = tensor("op_25706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25706_cast_fp16 = einsum(equation = var_25706_equation_0, values = (var_25492_cast_fp16, var_25044_cast_fp16))[name = tensor("op_25706_cast_fp16")]; + tensor var_25707_to_fp16 = const()[name = tensor("op_25707_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2437_cast_fp16 = mul(x = var_25706_cast_fp16, y = var_25707_to_fp16)[name = tensor("aw_chunk_2437_cast_fp16")]; + tensor var_25710_equation_0 = const()[name = tensor("op_25710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25710_cast_fp16 = einsum(equation = var_25710_equation_0, values = (var_25492_cast_fp16, var_25051_cast_fp16))[name = tensor("op_25710_cast_fp16")]; + tensor var_25711_to_fp16 = const()[name = tensor("op_25711_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2439_cast_fp16 = mul(x = var_25710_cast_fp16, y = var_25711_to_fp16)[name = tensor("aw_chunk_2439_cast_fp16")]; + tensor var_25714_equation_0 = const()[name = tensor("op_25714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25714_cast_fp16 = einsum(equation = var_25714_equation_0, values = (var_25496_cast_fp16, var_25058_cast_fp16))[name = tensor("op_25714_cast_fp16")]; + tensor var_25715_to_fp16 = const()[name = tensor("op_25715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2441_cast_fp16 = mul(x = var_25714_cast_fp16, y = var_25715_to_fp16)[name = tensor("aw_chunk_2441_cast_fp16")]; + tensor var_25718_equation_0 = const()[name = tensor("op_25718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25718_cast_fp16 = einsum(equation = var_25718_equation_0, values = (var_25496_cast_fp16, var_25065_cast_fp16))[name = tensor("op_25718_cast_fp16")]; + tensor var_25719_to_fp16 = const()[name = tensor("op_25719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2443_cast_fp16 = mul(x = var_25718_cast_fp16, y = var_25719_to_fp16)[name = tensor("aw_chunk_2443_cast_fp16")]; + tensor var_25722_equation_0 = const()[name = tensor("op_25722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25722_cast_fp16 = einsum(equation = var_25722_equation_0, values = (var_25496_cast_fp16, var_25072_cast_fp16))[name = tensor("op_25722_cast_fp16")]; + tensor var_25723_to_fp16 = const()[name = tensor("op_25723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2445_cast_fp16 = mul(x = var_25722_cast_fp16, y = var_25723_to_fp16)[name = tensor("aw_chunk_2445_cast_fp16")]; + tensor var_25726_equation_0 = const()[name = tensor("op_25726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25726_cast_fp16 = einsum(equation = var_25726_equation_0, values = (var_25496_cast_fp16, var_25079_cast_fp16))[name = tensor("op_25726_cast_fp16")]; + tensor var_25727_to_fp16 = const()[name = tensor("op_25727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2447_cast_fp16 = mul(x = var_25726_cast_fp16, y = var_25727_to_fp16)[name = tensor("aw_chunk_2447_cast_fp16")]; + tensor var_25730_equation_0 = const()[name = tensor("op_25730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25730_cast_fp16 = einsum(equation = var_25730_equation_0, values = (var_25500_cast_fp16, var_25086_cast_fp16))[name = tensor("op_25730_cast_fp16")]; + tensor var_25731_to_fp16 = const()[name = tensor("op_25731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2449_cast_fp16 = mul(x = var_25730_cast_fp16, y = var_25731_to_fp16)[name = tensor("aw_chunk_2449_cast_fp16")]; + tensor var_25734_equation_0 = const()[name = tensor("op_25734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25734_cast_fp16 = einsum(equation = var_25734_equation_0, values = (var_25500_cast_fp16, var_25093_cast_fp16))[name = tensor("op_25734_cast_fp16")]; + tensor var_25735_to_fp16 = const()[name = tensor("op_25735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2451_cast_fp16 = mul(x = var_25734_cast_fp16, y = var_25735_to_fp16)[name = tensor("aw_chunk_2451_cast_fp16")]; + tensor var_25738_equation_0 = const()[name = tensor("op_25738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25738_cast_fp16 = einsum(equation = var_25738_equation_0, values = (var_25500_cast_fp16, var_25100_cast_fp16))[name = tensor("op_25738_cast_fp16")]; + tensor var_25739_to_fp16 = const()[name = tensor("op_25739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2453_cast_fp16 = mul(x = var_25738_cast_fp16, y = var_25739_to_fp16)[name = tensor("aw_chunk_2453_cast_fp16")]; + tensor var_25742_equation_0 = const()[name = tensor("op_25742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25742_cast_fp16 = einsum(equation = var_25742_equation_0, values = (var_25500_cast_fp16, var_25107_cast_fp16))[name = tensor("op_25742_cast_fp16")]; + tensor var_25743_to_fp16 = const()[name = tensor("op_25743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2455_cast_fp16 = mul(x = var_25742_cast_fp16, y = var_25743_to_fp16)[name = tensor("aw_chunk_2455_cast_fp16")]; + tensor var_25746_equation_0 = const()[name = tensor("op_25746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25746_cast_fp16 = einsum(equation = var_25746_equation_0, values = (var_25504_cast_fp16, var_25114_cast_fp16))[name = tensor("op_25746_cast_fp16")]; + tensor var_25747_to_fp16 = const()[name = tensor("op_25747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2457_cast_fp16 = mul(x = var_25746_cast_fp16, y = var_25747_to_fp16)[name = tensor("aw_chunk_2457_cast_fp16")]; + tensor var_25750_equation_0 = const()[name = tensor("op_25750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25750_cast_fp16 = einsum(equation = var_25750_equation_0, values = (var_25504_cast_fp16, var_25121_cast_fp16))[name = tensor("op_25750_cast_fp16")]; + tensor var_25751_to_fp16 = const()[name = tensor("op_25751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2459_cast_fp16 = mul(x = var_25750_cast_fp16, y = var_25751_to_fp16)[name = tensor("aw_chunk_2459_cast_fp16")]; + tensor var_25754_equation_0 = const()[name = tensor("op_25754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25754_cast_fp16 = einsum(equation = var_25754_equation_0, values = (var_25504_cast_fp16, var_25128_cast_fp16))[name = tensor("op_25754_cast_fp16")]; + tensor var_25755_to_fp16 = const()[name = tensor("op_25755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2461_cast_fp16 = mul(x = var_25754_cast_fp16, y = var_25755_to_fp16)[name = tensor("aw_chunk_2461_cast_fp16")]; + tensor var_25758_equation_0 = const()[name = tensor("op_25758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25758_cast_fp16 = einsum(equation = var_25758_equation_0, values = (var_25504_cast_fp16, var_25135_cast_fp16))[name = tensor("op_25758_cast_fp16")]; + tensor var_25759_to_fp16 = const()[name = tensor("op_25759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2463_cast_fp16 = mul(x = var_25758_cast_fp16, y = var_25759_to_fp16)[name = tensor("aw_chunk_2463_cast_fp16")]; + tensor var_25762_equation_0 = const()[name = tensor("op_25762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25762_cast_fp16 = einsum(equation = var_25762_equation_0, values = (var_25508_cast_fp16, var_25142_cast_fp16))[name = tensor("op_25762_cast_fp16")]; + tensor var_25763_to_fp16 = const()[name = tensor("op_25763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2465_cast_fp16 = mul(x = var_25762_cast_fp16, y = var_25763_to_fp16)[name = tensor("aw_chunk_2465_cast_fp16")]; + tensor var_25766_equation_0 = const()[name = tensor("op_25766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25766_cast_fp16 = einsum(equation = var_25766_equation_0, values = (var_25508_cast_fp16, var_25149_cast_fp16))[name = tensor("op_25766_cast_fp16")]; + tensor var_25767_to_fp16 = const()[name = tensor("op_25767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2467_cast_fp16 = mul(x = var_25766_cast_fp16, y = var_25767_to_fp16)[name = tensor("aw_chunk_2467_cast_fp16")]; + tensor var_25770_equation_0 = const()[name = tensor("op_25770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25770_cast_fp16 = einsum(equation = var_25770_equation_0, values = (var_25508_cast_fp16, var_25156_cast_fp16))[name = tensor("op_25770_cast_fp16")]; + tensor var_25771_to_fp16 = const()[name = tensor("op_25771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2469_cast_fp16 = mul(x = var_25770_cast_fp16, y = var_25771_to_fp16)[name = tensor("aw_chunk_2469_cast_fp16")]; + tensor var_25774_equation_0 = const()[name = tensor("op_25774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25774_cast_fp16 = einsum(equation = var_25774_equation_0, values = (var_25508_cast_fp16, var_25163_cast_fp16))[name = tensor("op_25774_cast_fp16")]; + tensor var_25775_to_fp16 = const()[name = tensor("op_25775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2471_cast_fp16 = mul(x = var_25774_cast_fp16, y = var_25775_to_fp16)[name = tensor("aw_chunk_2471_cast_fp16")]; + tensor var_25778_equation_0 = const()[name = tensor("op_25778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25778_cast_fp16 = einsum(equation = var_25778_equation_0, values = (var_25512_cast_fp16, var_25170_cast_fp16))[name = tensor("op_25778_cast_fp16")]; + tensor var_25779_to_fp16 = const()[name = tensor("op_25779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2473_cast_fp16 = mul(x = var_25778_cast_fp16, y = var_25779_to_fp16)[name = tensor("aw_chunk_2473_cast_fp16")]; + tensor var_25782_equation_0 = const()[name = tensor("op_25782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25782_cast_fp16 = einsum(equation = var_25782_equation_0, values = (var_25512_cast_fp16, var_25177_cast_fp16))[name = tensor("op_25782_cast_fp16")]; + tensor var_25783_to_fp16 = const()[name = tensor("op_25783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2475_cast_fp16 = mul(x = var_25782_cast_fp16, y = var_25783_to_fp16)[name = tensor("aw_chunk_2475_cast_fp16")]; + tensor var_25786_equation_0 = const()[name = tensor("op_25786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25786_cast_fp16 = einsum(equation = var_25786_equation_0, values = (var_25512_cast_fp16, var_25184_cast_fp16))[name = tensor("op_25786_cast_fp16")]; + tensor var_25787_to_fp16 = const()[name = tensor("op_25787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2477_cast_fp16 = mul(x = var_25786_cast_fp16, y = var_25787_to_fp16)[name = tensor("aw_chunk_2477_cast_fp16")]; + tensor var_25790_equation_0 = const()[name = tensor("op_25790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25790_cast_fp16 = einsum(equation = var_25790_equation_0, values = (var_25512_cast_fp16, var_25191_cast_fp16))[name = tensor("op_25790_cast_fp16")]; + tensor var_25791_to_fp16 = const()[name = tensor("op_25791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2479_cast_fp16 = mul(x = var_25790_cast_fp16, y = var_25791_to_fp16)[name = tensor("aw_chunk_2479_cast_fp16")]; + tensor var_25794_equation_0 = const()[name = tensor("op_25794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25794_cast_fp16 = einsum(equation = var_25794_equation_0, values = (var_25516_cast_fp16, var_25198_cast_fp16))[name = tensor("op_25794_cast_fp16")]; + tensor var_25795_to_fp16 = const()[name = tensor("op_25795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2481_cast_fp16 = mul(x = var_25794_cast_fp16, y = var_25795_to_fp16)[name = tensor("aw_chunk_2481_cast_fp16")]; + tensor var_25798_equation_0 = const()[name = tensor("op_25798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25798_cast_fp16 = einsum(equation = var_25798_equation_0, values = (var_25516_cast_fp16, var_25205_cast_fp16))[name = tensor("op_25798_cast_fp16")]; + tensor var_25799_to_fp16 = const()[name = tensor("op_25799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2483_cast_fp16 = mul(x = var_25798_cast_fp16, y = var_25799_to_fp16)[name = tensor("aw_chunk_2483_cast_fp16")]; + tensor var_25802_equation_0 = const()[name = tensor("op_25802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25802_cast_fp16 = einsum(equation = var_25802_equation_0, values = (var_25516_cast_fp16, var_25212_cast_fp16))[name = tensor("op_25802_cast_fp16")]; + tensor var_25803_to_fp16 = const()[name = tensor("op_25803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2485_cast_fp16 = mul(x = var_25802_cast_fp16, y = var_25803_to_fp16)[name = tensor("aw_chunk_2485_cast_fp16")]; + tensor var_25806_equation_0 = const()[name = tensor("op_25806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25806_cast_fp16 = einsum(equation = var_25806_equation_0, values = (var_25516_cast_fp16, var_25219_cast_fp16))[name = tensor("op_25806_cast_fp16")]; + tensor var_25807_to_fp16 = const()[name = tensor("op_25807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2487_cast_fp16 = mul(x = var_25806_cast_fp16, y = var_25807_to_fp16)[name = tensor("aw_chunk_2487_cast_fp16")]; + tensor var_25810_equation_0 = const()[name = tensor("op_25810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25810_cast_fp16 = einsum(equation = var_25810_equation_0, values = (var_25520_cast_fp16, var_25226_cast_fp16))[name = tensor("op_25810_cast_fp16")]; + tensor var_25811_to_fp16 = const()[name = tensor("op_25811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2489_cast_fp16 = mul(x = var_25810_cast_fp16, y = var_25811_to_fp16)[name = tensor("aw_chunk_2489_cast_fp16")]; + tensor var_25814_equation_0 = const()[name = tensor("op_25814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25814_cast_fp16 = einsum(equation = var_25814_equation_0, values = (var_25520_cast_fp16, var_25233_cast_fp16))[name = tensor("op_25814_cast_fp16")]; + tensor var_25815_to_fp16 = const()[name = tensor("op_25815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2491_cast_fp16 = mul(x = var_25814_cast_fp16, y = var_25815_to_fp16)[name = tensor("aw_chunk_2491_cast_fp16")]; + tensor var_25818_equation_0 = const()[name = tensor("op_25818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25818_cast_fp16 = einsum(equation = var_25818_equation_0, values = (var_25520_cast_fp16, var_25240_cast_fp16))[name = tensor("op_25818_cast_fp16")]; + tensor var_25819_to_fp16 = const()[name = tensor("op_25819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2493_cast_fp16 = mul(x = var_25818_cast_fp16, y = var_25819_to_fp16)[name = tensor("aw_chunk_2493_cast_fp16")]; + tensor var_25822_equation_0 = const()[name = tensor("op_25822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25822_cast_fp16 = einsum(equation = var_25822_equation_0, values = (var_25520_cast_fp16, var_25247_cast_fp16))[name = tensor("op_25822_cast_fp16")]; + tensor var_25823_to_fp16 = const()[name = tensor("op_25823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2495_cast_fp16 = mul(x = var_25822_cast_fp16, y = var_25823_to_fp16)[name = tensor("aw_chunk_2495_cast_fp16")]; + tensor var_25826_equation_0 = const()[name = tensor("op_25826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25826_cast_fp16 = einsum(equation = var_25826_equation_0, values = (var_25524_cast_fp16, var_25254_cast_fp16))[name = tensor("op_25826_cast_fp16")]; + tensor var_25827_to_fp16 = const()[name = tensor("op_25827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2497_cast_fp16 = mul(x = var_25826_cast_fp16, y = var_25827_to_fp16)[name = tensor("aw_chunk_2497_cast_fp16")]; + tensor var_25830_equation_0 = const()[name = tensor("op_25830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25830_cast_fp16 = einsum(equation = var_25830_equation_0, values = (var_25524_cast_fp16, var_25261_cast_fp16))[name = tensor("op_25830_cast_fp16")]; + tensor var_25831_to_fp16 = const()[name = tensor("op_25831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2499_cast_fp16 = mul(x = var_25830_cast_fp16, y = var_25831_to_fp16)[name = tensor("aw_chunk_2499_cast_fp16")]; + tensor var_25834_equation_0 = const()[name = tensor("op_25834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25834_cast_fp16 = einsum(equation = var_25834_equation_0, values = (var_25524_cast_fp16, var_25268_cast_fp16))[name = tensor("op_25834_cast_fp16")]; + tensor var_25835_to_fp16 = const()[name = tensor("op_25835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2501_cast_fp16 = mul(x = var_25834_cast_fp16, y = var_25835_to_fp16)[name = tensor("aw_chunk_2501_cast_fp16")]; + tensor var_25838_equation_0 = const()[name = tensor("op_25838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25838_cast_fp16 = einsum(equation = var_25838_equation_0, values = (var_25524_cast_fp16, var_25275_cast_fp16))[name = tensor("op_25838_cast_fp16")]; + tensor var_25839_to_fp16 = const()[name = tensor("op_25839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2503_cast_fp16 = mul(x = var_25838_cast_fp16, y = var_25839_to_fp16)[name = tensor("aw_chunk_2503_cast_fp16")]; + tensor var_25842_equation_0 = const()[name = tensor("op_25842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25842_cast_fp16 = einsum(equation = var_25842_equation_0, values = (var_25528_cast_fp16, var_25282_cast_fp16))[name = tensor("op_25842_cast_fp16")]; + tensor var_25843_to_fp16 = const()[name = tensor("op_25843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2505_cast_fp16 = mul(x = var_25842_cast_fp16, y = var_25843_to_fp16)[name = tensor("aw_chunk_2505_cast_fp16")]; + tensor var_25846_equation_0 = const()[name = tensor("op_25846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25846_cast_fp16 = einsum(equation = var_25846_equation_0, values = (var_25528_cast_fp16, var_25289_cast_fp16))[name = tensor("op_25846_cast_fp16")]; + tensor var_25847_to_fp16 = const()[name = tensor("op_25847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2507_cast_fp16 = mul(x = var_25846_cast_fp16, y = var_25847_to_fp16)[name = tensor("aw_chunk_2507_cast_fp16")]; + tensor var_25850_equation_0 = const()[name = tensor("op_25850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25850_cast_fp16 = einsum(equation = var_25850_equation_0, values = (var_25528_cast_fp16, var_25296_cast_fp16))[name = tensor("op_25850_cast_fp16")]; + tensor var_25851_to_fp16 = const()[name = tensor("op_25851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2509_cast_fp16 = mul(x = var_25850_cast_fp16, y = var_25851_to_fp16)[name = tensor("aw_chunk_2509_cast_fp16")]; + tensor var_25854_equation_0 = const()[name = tensor("op_25854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25854_cast_fp16 = einsum(equation = var_25854_equation_0, values = (var_25528_cast_fp16, var_25303_cast_fp16))[name = tensor("op_25854_cast_fp16")]; + tensor var_25855_to_fp16 = const()[name = tensor("op_25855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2511_cast_fp16 = mul(x = var_25854_cast_fp16, y = var_25855_to_fp16)[name = tensor("aw_chunk_2511_cast_fp16")]; + tensor var_25858_equation_0 = const()[name = tensor("op_25858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25858_cast_fp16 = einsum(equation = var_25858_equation_0, values = (var_25532_cast_fp16, var_25310_cast_fp16))[name = tensor("op_25858_cast_fp16")]; + tensor var_25859_to_fp16 = const()[name = tensor("op_25859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2513_cast_fp16 = mul(x = var_25858_cast_fp16, y = var_25859_to_fp16)[name = tensor("aw_chunk_2513_cast_fp16")]; + tensor var_25862_equation_0 = const()[name = tensor("op_25862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25862_cast_fp16 = einsum(equation = var_25862_equation_0, values = (var_25532_cast_fp16, var_25317_cast_fp16))[name = tensor("op_25862_cast_fp16")]; + tensor var_25863_to_fp16 = const()[name = tensor("op_25863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2515_cast_fp16 = mul(x = var_25862_cast_fp16, y = var_25863_to_fp16)[name = tensor("aw_chunk_2515_cast_fp16")]; + tensor var_25866_equation_0 = const()[name = tensor("op_25866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25866_cast_fp16 = einsum(equation = var_25866_equation_0, values = (var_25532_cast_fp16, var_25324_cast_fp16))[name = tensor("op_25866_cast_fp16")]; + tensor var_25867_to_fp16 = const()[name = tensor("op_25867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2517_cast_fp16 = mul(x = var_25866_cast_fp16, y = var_25867_to_fp16)[name = tensor("aw_chunk_2517_cast_fp16")]; + tensor var_25870_equation_0 = const()[name = tensor("op_25870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25870_cast_fp16 = einsum(equation = var_25870_equation_0, values = (var_25532_cast_fp16, var_25331_cast_fp16))[name = tensor("op_25870_cast_fp16")]; + tensor var_25871_to_fp16 = const()[name = tensor("op_25871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2519_cast_fp16 = mul(x = var_25870_cast_fp16, y = var_25871_to_fp16)[name = tensor("aw_chunk_2519_cast_fp16")]; + tensor var_25874_equation_0 = const()[name = tensor("op_25874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25874_cast_fp16 = einsum(equation = var_25874_equation_0, values = (var_25536_cast_fp16, var_25338_cast_fp16))[name = tensor("op_25874_cast_fp16")]; + tensor var_25875_to_fp16 = const()[name = tensor("op_25875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2521_cast_fp16 = mul(x = var_25874_cast_fp16, y = var_25875_to_fp16)[name = tensor("aw_chunk_2521_cast_fp16")]; + tensor var_25878_equation_0 = const()[name = tensor("op_25878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25878_cast_fp16 = einsum(equation = var_25878_equation_0, values = (var_25536_cast_fp16, var_25345_cast_fp16))[name = tensor("op_25878_cast_fp16")]; + tensor var_25879_to_fp16 = const()[name = tensor("op_25879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2523_cast_fp16 = mul(x = var_25878_cast_fp16, y = var_25879_to_fp16)[name = tensor("aw_chunk_2523_cast_fp16")]; + tensor var_25882_equation_0 = const()[name = tensor("op_25882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25882_cast_fp16 = einsum(equation = var_25882_equation_0, values = (var_25536_cast_fp16, var_25352_cast_fp16))[name = tensor("op_25882_cast_fp16")]; + tensor var_25883_to_fp16 = const()[name = tensor("op_25883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2525_cast_fp16 = mul(x = var_25882_cast_fp16, y = var_25883_to_fp16)[name = tensor("aw_chunk_2525_cast_fp16")]; + tensor var_25886_equation_0 = const()[name = tensor("op_25886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25886_cast_fp16 = einsum(equation = var_25886_equation_0, values = (var_25536_cast_fp16, var_25359_cast_fp16))[name = tensor("op_25886_cast_fp16")]; + tensor var_25887_to_fp16 = const()[name = tensor("op_25887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2527_cast_fp16 = mul(x = var_25886_cast_fp16, y = var_25887_to_fp16)[name = tensor("aw_chunk_2527_cast_fp16")]; + tensor var_25890_equation_0 = const()[name = tensor("op_25890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25890_cast_fp16 = einsum(equation = var_25890_equation_0, values = (var_25540_cast_fp16, var_25366_cast_fp16))[name = tensor("op_25890_cast_fp16")]; + tensor var_25891_to_fp16 = const()[name = tensor("op_25891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2529_cast_fp16 = mul(x = var_25890_cast_fp16, y = var_25891_to_fp16)[name = tensor("aw_chunk_2529_cast_fp16")]; + tensor var_25894_equation_0 = const()[name = tensor("op_25894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25894_cast_fp16 = einsum(equation = var_25894_equation_0, values = (var_25540_cast_fp16, var_25373_cast_fp16))[name = tensor("op_25894_cast_fp16")]; + tensor var_25895_to_fp16 = const()[name = tensor("op_25895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2531_cast_fp16 = mul(x = var_25894_cast_fp16, y = var_25895_to_fp16)[name = tensor("aw_chunk_2531_cast_fp16")]; + tensor var_25898_equation_0 = const()[name = tensor("op_25898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25898_cast_fp16 = einsum(equation = var_25898_equation_0, values = (var_25540_cast_fp16, var_25380_cast_fp16))[name = tensor("op_25898_cast_fp16")]; + tensor var_25899_to_fp16 = const()[name = tensor("op_25899_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2533_cast_fp16 = mul(x = var_25898_cast_fp16, y = var_25899_to_fp16)[name = tensor("aw_chunk_2533_cast_fp16")]; + tensor var_25902_equation_0 = const()[name = tensor("op_25902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25902_cast_fp16 = einsum(equation = var_25902_equation_0, values = (var_25540_cast_fp16, var_25387_cast_fp16))[name = tensor("op_25902_cast_fp16")]; + tensor var_25903_to_fp16 = const()[name = tensor("op_25903_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2535_cast_fp16 = mul(x = var_25902_cast_fp16, y = var_25903_to_fp16)[name = tensor("aw_chunk_2535_cast_fp16")]; + tensor var_25906_equation_0 = const()[name = tensor("op_25906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25906_cast_fp16 = einsum(equation = var_25906_equation_0, values = (var_25544_cast_fp16, var_25394_cast_fp16))[name = tensor("op_25906_cast_fp16")]; + tensor var_25907_to_fp16 = const()[name = tensor("op_25907_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2537_cast_fp16 = mul(x = var_25906_cast_fp16, y = var_25907_to_fp16)[name = tensor("aw_chunk_2537_cast_fp16")]; + tensor var_25910_equation_0 = const()[name = tensor("op_25910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25910_cast_fp16 = einsum(equation = var_25910_equation_0, values = (var_25544_cast_fp16, var_25401_cast_fp16))[name = tensor("op_25910_cast_fp16")]; + tensor var_25911_to_fp16 = const()[name = tensor("op_25911_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2539_cast_fp16 = mul(x = var_25910_cast_fp16, y = var_25911_to_fp16)[name = tensor("aw_chunk_2539_cast_fp16")]; + tensor var_25914_equation_0 = const()[name = tensor("op_25914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25914_cast_fp16 = einsum(equation = var_25914_equation_0, values = (var_25544_cast_fp16, var_25408_cast_fp16))[name = tensor("op_25914_cast_fp16")]; + tensor var_25915_to_fp16 = const()[name = tensor("op_25915_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2541_cast_fp16 = mul(x = var_25914_cast_fp16, y = var_25915_to_fp16)[name = tensor("aw_chunk_2541_cast_fp16")]; + tensor var_25918_equation_0 = const()[name = tensor("op_25918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25918_cast_fp16 = einsum(equation = var_25918_equation_0, values = (var_25544_cast_fp16, var_25415_cast_fp16))[name = tensor("op_25918_cast_fp16")]; + tensor var_25919_to_fp16 = const()[name = tensor("op_25919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2543_cast_fp16 = mul(x = var_25918_cast_fp16, y = var_25919_to_fp16)[name = tensor("aw_chunk_2543_cast_fp16")]; + tensor var_25922_equation_0 = const()[name = tensor("op_25922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25922_cast_fp16 = einsum(equation = var_25922_equation_0, values = (var_25548_cast_fp16, var_25422_cast_fp16))[name = tensor("op_25922_cast_fp16")]; + tensor var_25923_to_fp16 = const()[name = tensor("op_25923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2545_cast_fp16 = mul(x = var_25922_cast_fp16, y = var_25923_to_fp16)[name = tensor("aw_chunk_2545_cast_fp16")]; + tensor var_25926_equation_0 = const()[name = tensor("op_25926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25926_cast_fp16 = einsum(equation = var_25926_equation_0, values = (var_25548_cast_fp16, var_25429_cast_fp16))[name = tensor("op_25926_cast_fp16")]; + tensor var_25927_to_fp16 = const()[name = tensor("op_25927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2547_cast_fp16 = mul(x = var_25926_cast_fp16, y = var_25927_to_fp16)[name = tensor("aw_chunk_2547_cast_fp16")]; + tensor var_25930_equation_0 = const()[name = tensor("op_25930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25930_cast_fp16 = einsum(equation = var_25930_equation_0, values = (var_25548_cast_fp16, var_25436_cast_fp16))[name = tensor("op_25930_cast_fp16")]; + tensor var_25931_to_fp16 = const()[name = tensor("op_25931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2549_cast_fp16 = mul(x = var_25930_cast_fp16, y = var_25931_to_fp16)[name = tensor("aw_chunk_2549_cast_fp16")]; + tensor var_25934_equation_0 = const()[name = tensor("op_25934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25934_cast_fp16 = einsum(equation = var_25934_equation_0, values = (var_25548_cast_fp16, var_25443_cast_fp16))[name = tensor("op_25934_cast_fp16")]; + tensor var_25935_to_fp16 = const()[name = tensor("op_25935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2551_cast_fp16 = mul(x = var_25934_cast_fp16, y = var_25935_to_fp16)[name = tensor("aw_chunk_2551_cast_fp16")]; + tensor var_25938_equation_0 = const()[name = tensor("op_25938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25938_cast_fp16 = einsum(equation = var_25938_equation_0, values = (var_25552_cast_fp16, var_25450_cast_fp16))[name = tensor("op_25938_cast_fp16")]; + tensor var_25939_to_fp16 = const()[name = tensor("op_25939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2553_cast_fp16 = mul(x = var_25938_cast_fp16, y = var_25939_to_fp16)[name = tensor("aw_chunk_2553_cast_fp16")]; + tensor var_25942_equation_0 = const()[name = tensor("op_25942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25942_cast_fp16 = einsum(equation = var_25942_equation_0, values = (var_25552_cast_fp16, var_25457_cast_fp16))[name = tensor("op_25942_cast_fp16")]; + tensor var_25943_to_fp16 = const()[name = tensor("op_25943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2555_cast_fp16 = mul(x = var_25942_cast_fp16, y = var_25943_to_fp16)[name = tensor("aw_chunk_2555_cast_fp16")]; + tensor var_25946_equation_0 = const()[name = tensor("op_25946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25946_cast_fp16 = einsum(equation = var_25946_equation_0, values = (var_25552_cast_fp16, var_25464_cast_fp16))[name = tensor("op_25946_cast_fp16")]; + tensor var_25947_to_fp16 = const()[name = tensor("op_25947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2557_cast_fp16 = mul(x = var_25946_cast_fp16, y = var_25947_to_fp16)[name = tensor("aw_chunk_2557_cast_fp16")]; + tensor var_25950_equation_0 = const()[name = tensor("op_25950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_25950_cast_fp16 = einsum(equation = var_25950_equation_0, values = (var_25552_cast_fp16, var_25471_cast_fp16))[name = tensor("op_25950_cast_fp16")]; + tensor var_25951_to_fp16 = const()[name = tensor("op_25951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2559_cast_fp16 = mul(x = var_25950_cast_fp16, y = var_25951_to_fp16)[name = tensor("aw_chunk_2559_cast_fp16")]; + tensor var_25953_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2401_cast_fp16)[name = tensor("op_25953_cast_fp16")]; + tensor var_25954_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2403_cast_fp16)[name = tensor("op_25954_cast_fp16")]; + tensor var_25955_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2405_cast_fp16)[name = tensor("op_25955_cast_fp16")]; + tensor var_25956_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2407_cast_fp16)[name = tensor("op_25956_cast_fp16")]; + tensor var_25957_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2409_cast_fp16)[name = tensor("op_25957_cast_fp16")]; + tensor var_25958_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2411_cast_fp16)[name = tensor("op_25958_cast_fp16")]; + tensor var_25959_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2413_cast_fp16)[name = tensor("op_25959_cast_fp16")]; + tensor var_25960_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2415_cast_fp16)[name = tensor("op_25960_cast_fp16")]; + tensor var_25961_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2417_cast_fp16)[name = tensor("op_25961_cast_fp16")]; + tensor var_25962_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2419_cast_fp16)[name = tensor("op_25962_cast_fp16")]; + tensor var_25963_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2421_cast_fp16)[name = tensor("op_25963_cast_fp16")]; + tensor var_25964_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2423_cast_fp16)[name = tensor("op_25964_cast_fp16")]; + tensor var_25965_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2425_cast_fp16)[name = tensor("op_25965_cast_fp16")]; + tensor var_25966_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2427_cast_fp16)[name = tensor("op_25966_cast_fp16")]; + tensor var_25967_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2429_cast_fp16)[name = tensor("op_25967_cast_fp16")]; + tensor var_25968_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2431_cast_fp16)[name = tensor("op_25968_cast_fp16")]; + tensor var_25969_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2433_cast_fp16)[name = tensor("op_25969_cast_fp16")]; + tensor var_25970_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2435_cast_fp16)[name = tensor("op_25970_cast_fp16")]; + tensor var_25971_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2437_cast_fp16)[name = tensor("op_25971_cast_fp16")]; + tensor var_25972_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2439_cast_fp16)[name = tensor("op_25972_cast_fp16")]; + tensor var_25973_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2441_cast_fp16)[name = tensor("op_25973_cast_fp16")]; + tensor var_25974_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2443_cast_fp16)[name = tensor("op_25974_cast_fp16")]; + tensor var_25975_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2445_cast_fp16)[name = tensor("op_25975_cast_fp16")]; + tensor var_25976_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2447_cast_fp16)[name = tensor("op_25976_cast_fp16")]; + tensor var_25977_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2449_cast_fp16)[name = tensor("op_25977_cast_fp16")]; + tensor var_25978_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2451_cast_fp16)[name = tensor("op_25978_cast_fp16")]; + tensor var_25979_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2453_cast_fp16)[name = tensor("op_25979_cast_fp16")]; + tensor var_25980_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2455_cast_fp16)[name = tensor("op_25980_cast_fp16")]; + tensor var_25981_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2457_cast_fp16)[name = tensor("op_25981_cast_fp16")]; + tensor var_25982_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2459_cast_fp16)[name = tensor("op_25982_cast_fp16")]; + tensor var_25983_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2461_cast_fp16)[name = tensor("op_25983_cast_fp16")]; + tensor var_25984_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2463_cast_fp16)[name = tensor("op_25984_cast_fp16")]; + tensor var_25985_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2465_cast_fp16)[name = tensor("op_25985_cast_fp16")]; + tensor var_25986_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2467_cast_fp16)[name = tensor("op_25986_cast_fp16")]; + tensor var_25987_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2469_cast_fp16)[name = tensor("op_25987_cast_fp16")]; + tensor var_25988_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2471_cast_fp16)[name = tensor("op_25988_cast_fp16")]; + tensor var_25989_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2473_cast_fp16)[name = tensor("op_25989_cast_fp16")]; + tensor var_25990_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2475_cast_fp16)[name = tensor("op_25990_cast_fp16")]; + tensor var_25991_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2477_cast_fp16)[name = tensor("op_25991_cast_fp16")]; + tensor var_25992_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2479_cast_fp16)[name = tensor("op_25992_cast_fp16")]; + tensor var_25993_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2481_cast_fp16)[name = tensor("op_25993_cast_fp16")]; + tensor var_25994_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2483_cast_fp16)[name = tensor("op_25994_cast_fp16")]; + tensor var_25995_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2485_cast_fp16)[name = tensor("op_25995_cast_fp16")]; + tensor var_25996_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2487_cast_fp16)[name = tensor("op_25996_cast_fp16")]; + tensor var_25997_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2489_cast_fp16)[name = tensor("op_25997_cast_fp16")]; + tensor var_25998_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2491_cast_fp16)[name = tensor("op_25998_cast_fp16")]; + tensor var_25999_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2493_cast_fp16)[name = tensor("op_25999_cast_fp16")]; + tensor var_26000_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2495_cast_fp16)[name = tensor("op_26000_cast_fp16")]; + tensor var_26001_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2497_cast_fp16)[name = tensor("op_26001_cast_fp16")]; + tensor var_26002_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2499_cast_fp16)[name = tensor("op_26002_cast_fp16")]; + tensor var_26003_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2501_cast_fp16)[name = tensor("op_26003_cast_fp16")]; + tensor var_26004_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2503_cast_fp16)[name = tensor("op_26004_cast_fp16")]; + tensor var_26005_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2505_cast_fp16)[name = tensor("op_26005_cast_fp16")]; + tensor var_26006_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2507_cast_fp16)[name = tensor("op_26006_cast_fp16")]; + tensor var_26007_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2509_cast_fp16)[name = tensor("op_26007_cast_fp16")]; + tensor var_26008_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2511_cast_fp16)[name = tensor("op_26008_cast_fp16")]; + tensor var_26009_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2513_cast_fp16)[name = tensor("op_26009_cast_fp16")]; + tensor var_26010_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2515_cast_fp16)[name = tensor("op_26010_cast_fp16")]; + tensor var_26011_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2517_cast_fp16)[name = tensor("op_26011_cast_fp16")]; + tensor var_26012_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2519_cast_fp16)[name = tensor("op_26012_cast_fp16")]; + tensor var_26013_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2521_cast_fp16)[name = tensor("op_26013_cast_fp16")]; + tensor var_26014_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2523_cast_fp16)[name = tensor("op_26014_cast_fp16")]; + tensor var_26015_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2525_cast_fp16)[name = tensor("op_26015_cast_fp16")]; + tensor var_26016_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2527_cast_fp16)[name = tensor("op_26016_cast_fp16")]; + tensor var_26017_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2529_cast_fp16)[name = tensor("op_26017_cast_fp16")]; + tensor var_26018_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2531_cast_fp16)[name = tensor("op_26018_cast_fp16")]; + tensor var_26019_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2533_cast_fp16)[name = tensor("op_26019_cast_fp16")]; + tensor var_26020_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2535_cast_fp16)[name = tensor("op_26020_cast_fp16")]; + tensor var_26021_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2537_cast_fp16)[name = tensor("op_26021_cast_fp16")]; + tensor var_26022_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2539_cast_fp16)[name = tensor("op_26022_cast_fp16")]; + tensor var_26023_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2541_cast_fp16)[name = tensor("op_26023_cast_fp16")]; + tensor var_26024_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2543_cast_fp16)[name = tensor("op_26024_cast_fp16")]; + tensor var_26025_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2545_cast_fp16)[name = tensor("op_26025_cast_fp16")]; + tensor var_26026_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2547_cast_fp16)[name = tensor("op_26026_cast_fp16")]; + tensor var_26027_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2549_cast_fp16)[name = tensor("op_26027_cast_fp16")]; + tensor var_26028_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2551_cast_fp16)[name = tensor("op_26028_cast_fp16")]; + tensor var_26029_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2553_cast_fp16)[name = tensor("op_26029_cast_fp16")]; + tensor var_26030_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2555_cast_fp16)[name = tensor("op_26030_cast_fp16")]; + tensor var_26031_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2557_cast_fp16)[name = tensor("op_26031_cast_fp16")]; + tensor var_26032_cast_fp16 = softmax(axis = var_24724, x = aw_chunk_2559_cast_fp16)[name = tensor("op_26032_cast_fp16")]; + tensor var_26034_equation_0 = const()[name = tensor("op_26034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26034_cast_fp16 = einsum(equation = var_26034_equation_0, values = (var_25554_cast_fp16, var_25953_cast_fp16))[name = tensor("op_26034_cast_fp16")]; + tensor var_26036_equation_0 = const()[name = tensor("op_26036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26036_cast_fp16 = einsum(equation = var_26036_equation_0, values = (var_25554_cast_fp16, var_25954_cast_fp16))[name = tensor("op_26036_cast_fp16")]; + tensor var_26038_equation_0 = const()[name = tensor("op_26038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26038_cast_fp16 = einsum(equation = var_26038_equation_0, values = (var_25554_cast_fp16, var_25955_cast_fp16))[name = tensor("op_26038_cast_fp16")]; + tensor var_26040_equation_0 = const()[name = tensor("op_26040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26040_cast_fp16 = einsum(equation = var_26040_equation_0, values = (var_25554_cast_fp16, var_25956_cast_fp16))[name = tensor("op_26040_cast_fp16")]; + tensor var_26042_equation_0 = const()[name = tensor("op_26042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26042_cast_fp16 = einsum(equation = var_26042_equation_0, values = (var_25558_cast_fp16, var_25957_cast_fp16))[name = tensor("op_26042_cast_fp16")]; + tensor var_26044_equation_0 = const()[name = tensor("op_26044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26044_cast_fp16 = einsum(equation = var_26044_equation_0, values = (var_25558_cast_fp16, var_25958_cast_fp16))[name = tensor("op_26044_cast_fp16")]; + tensor var_26046_equation_0 = const()[name = tensor("op_26046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26046_cast_fp16 = einsum(equation = var_26046_equation_0, values = (var_25558_cast_fp16, var_25959_cast_fp16))[name = tensor("op_26046_cast_fp16")]; + tensor var_26048_equation_0 = const()[name = tensor("op_26048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26048_cast_fp16 = einsum(equation = var_26048_equation_0, values = (var_25558_cast_fp16, var_25960_cast_fp16))[name = tensor("op_26048_cast_fp16")]; + tensor var_26050_equation_0 = const()[name = tensor("op_26050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26050_cast_fp16 = einsum(equation = var_26050_equation_0, values = (var_25562_cast_fp16, var_25961_cast_fp16))[name = tensor("op_26050_cast_fp16")]; + tensor var_26052_equation_0 = const()[name = tensor("op_26052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26052_cast_fp16 = einsum(equation = var_26052_equation_0, values = (var_25562_cast_fp16, var_25962_cast_fp16))[name = tensor("op_26052_cast_fp16")]; + tensor var_26054_equation_0 = const()[name = tensor("op_26054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26054_cast_fp16 = einsum(equation = var_26054_equation_0, values = (var_25562_cast_fp16, var_25963_cast_fp16))[name = tensor("op_26054_cast_fp16")]; + tensor var_26056_equation_0 = const()[name = tensor("op_26056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26056_cast_fp16 = einsum(equation = var_26056_equation_0, values = (var_25562_cast_fp16, var_25964_cast_fp16))[name = tensor("op_26056_cast_fp16")]; + tensor var_26058_equation_0 = const()[name = tensor("op_26058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26058_cast_fp16 = einsum(equation = var_26058_equation_0, values = (var_25566_cast_fp16, var_25965_cast_fp16))[name = tensor("op_26058_cast_fp16")]; + tensor var_26060_equation_0 = const()[name = tensor("op_26060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26060_cast_fp16 = einsum(equation = var_26060_equation_0, values = (var_25566_cast_fp16, var_25966_cast_fp16))[name = tensor("op_26060_cast_fp16")]; + tensor var_26062_equation_0 = const()[name = tensor("op_26062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26062_cast_fp16 = einsum(equation = var_26062_equation_0, values = (var_25566_cast_fp16, var_25967_cast_fp16))[name = tensor("op_26062_cast_fp16")]; + tensor var_26064_equation_0 = const()[name = tensor("op_26064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26064_cast_fp16 = einsum(equation = var_26064_equation_0, values = (var_25566_cast_fp16, var_25968_cast_fp16))[name = tensor("op_26064_cast_fp16")]; + tensor var_26066_equation_0 = const()[name = tensor("op_26066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26066_cast_fp16 = einsum(equation = var_26066_equation_0, values = (var_25570_cast_fp16, var_25969_cast_fp16))[name = tensor("op_26066_cast_fp16")]; + tensor var_26068_equation_0 = const()[name = tensor("op_26068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26068_cast_fp16 = einsum(equation = var_26068_equation_0, values = (var_25570_cast_fp16, var_25970_cast_fp16))[name = tensor("op_26068_cast_fp16")]; + tensor var_26070_equation_0 = const()[name = tensor("op_26070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26070_cast_fp16 = einsum(equation = var_26070_equation_0, values = (var_25570_cast_fp16, var_25971_cast_fp16))[name = tensor("op_26070_cast_fp16")]; + tensor var_26072_equation_0 = const()[name = tensor("op_26072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26072_cast_fp16 = einsum(equation = var_26072_equation_0, values = (var_25570_cast_fp16, var_25972_cast_fp16))[name = tensor("op_26072_cast_fp16")]; + tensor var_26074_equation_0 = const()[name = tensor("op_26074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26074_cast_fp16 = einsum(equation = var_26074_equation_0, values = (var_25574_cast_fp16, var_25973_cast_fp16))[name = tensor("op_26074_cast_fp16")]; + tensor var_26076_equation_0 = const()[name = tensor("op_26076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26076_cast_fp16 = einsum(equation = var_26076_equation_0, values = (var_25574_cast_fp16, var_25974_cast_fp16))[name = tensor("op_26076_cast_fp16")]; + tensor var_26078_equation_0 = const()[name = tensor("op_26078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26078_cast_fp16 = einsum(equation = var_26078_equation_0, values = (var_25574_cast_fp16, var_25975_cast_fp16))[name = tensor("op_26078_cast_fp16")]; + tensor var_26080_equation_0 = const()[name = tensor("op_26080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26080_cast_fp16 = einsum(equation = var_26080_equation_0, values = (var_25574_cast_fp16, var_25976_cast_fp16))[name = tensor("op_26080_cast_fp16")]; + tensor var_26082_equation_0 = const()[name = tensor("op_26082_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26082_cast_fp16 = einsum(equation = var_26082_equation_0, values = (var_25578_cast_fp16, var_25977_cast_fp16))[name = tensor("op_26082_cast_fp16")]; + tensor var_26084_equation_0 = const()[name = tensor("op_26084_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26084_cast_fp16 = einsum(equation = var_26084_equation_0, values = (var_25578_cast_fp16, var_25978_cast_fp16))[name = tensor("op_26084_cast_fp16")]; + tensor var_26086_equation_0 = const()[name = tensor("op_26086_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26086_cast_fp16 = einsum(equation = var_26086_equation_0, values = (var_25578_cast_fp16, var_25979_cast_fp16))[name = tensor("op_26086_cast_fp16")]; + tensor var_26088_equation_0 = const()[name = tensor("op_26088_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26088_cast_fp16 = einsum(equation = var_26088_equation_0, values = (var_25578_cast_fp16, var_25980_cast_fp16))[name = tensor("op_26088_cast_fp16")]; + tensor var_26090_equation_0 = const()[name = tensor("op_26090_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26090_cast_fp16 = einsum(equation = var_26090_equation_0, values = (var_25582_cast_fp16, var_25981_cast_fp16))[name = tensor("op_26090_cast_fp16")]; + tensor var_26092_equation_0 = const()[name = tensor("op_26092_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26092_cast_fp16 = einsum(equation = var_26092_equation_0, values = (var_25582_cast_fp16, var_25982_cast_fp16))[name = tensor("op_26092_cast_fp16")]; + tensor var_26094_equation_0 = const()[name = tensor("op_26094_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26094_cast_fp16 = einsum(equation = var_26094_equation_0, values = (var_25582_cast_fp16, var_25983_cast_fp16))[name = tensor("op_26094_cast_fp16")]; + tensor var_26096_equation_0 = const()[name = tensor("op_26096_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26096_cast_fp16 = einsum(equation = var_26096_equation_0, values = (var_25582_cast_fp16, var_25984_cast_fp16))[name = tensor("op_26096_cast_fp16")]; + tensor var_26098_equation_0 = const()[name = tensor("op_26098_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26098_cast_fp16 = einsum(equation = var_26098_equation_0, values = (var_25586_cast_fp16, var_25985_cast_fp16))[name = tensor("op_26098_cast_fp16")]; + tensor var_26100_equation_0 = const()[name = tensor("op_26100_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26100_cast_fp16 = einsum(equation = var_26100_equation_0, values = (var_25586_cast_fp16, var_25986_cast_fp16))[name = tensor("op_26100_cast_fp16")]; + tensor var_26102_equation_0 = const()[name = tensor("op_26102_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26102_cast_fp16 = einsum(equation = var_26102_equation_0, values = (var_25586_cast_fp16, var_25987_cast_fp16))[name = tensor("op_26102_cast_fp16")]; + tensor var_26104_equation_0 = const()[name = tensor("op_26104_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26104_cast_fp16 = einsum(equation = var_26104_equation_0, values = (var_25586_cast_fp16, var_25988_cast_fp16))[name = tensor("op_26104_cast_fp16")]; + tensor var_26106_equation_0 = const()[name = tensor("op_26106_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26106_cast_fp16 = einsum(equation = var_26106_equation_0, values = (var_25590_cast_fp16, var_25989_cast_fp16))[name = tensor("op_26106_cast_fp16")]; + tensor var_26108_equation_0 = const()[name = tensor("op_26108_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26108_cast_fp16 = einsum(equation = var_26108_equation_0, values = (var_25590_cast_fp16, var_25990_cast_fp16))[name = tensor("op_26108_cast_fp16")]; + tensor var_26110_equation_0 = const()[name = tensor("op_26110_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26110_cast_fp16 = einsum(equation = var_26110_equation_0, values = (var_25590_cast_fp16, var_25991_cast_fp16))[name = tensor("op_26110_cast_fp16")]; + tensor var_26112_equation_0 = const()[name = tensor("op_26112_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26112_cast_fp16 = einsum(equation = var_26112_equation_0, values = (var_25590_cast_fp16, var_25992_cast_fp16))[name = tensor("op_26112_cast_fp16")]; + tensor var_26114_equation_0 = const()[name = tensor("op_26114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26114_cast_fp16 = einsum(equation = var_26114_equation_0, values = (var_25594_cast_fp16, var_25993_cast_fp16))[name = tensor("op_26114_cast_fp16")]; + tensor var_26116_equation_0 = const()[name = tensor("op_26116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26116_cast_fp16 = einsum(equation = var_26116_equation_0, values = (var_25594_cast_fp16, var_25994_cast_fp16))[name = tensor("op_26116_cast_fp16")]; + tensor var_26118_equation_0 = const()[name = tensor("op_26118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26118_cast_fp16 = einsum(equation = var_26118_equation_0, values = (var_25594_cast_fp16, var_25995_cast_fp16))[name = tensor("op_26118_cast_fp16")]; + tensor var_26120_equation_0 = const()[name = tensor("op_26120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26120_cast_fp16 = einsum(equation = var_26120_equation_0, values = (var_25594_cast_fp16, var_25996_cast_fp16))[name = tensor("op_26120_cast_fp16")]; + tensor var_26122_equation_0 = const()[name = tensor("op_26122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26122_cast_fp16 = einsum(equation = var_26122_equation_0, values = (var_25598_cast_fp16, var_25997_cast_fp16))[name = tensor("op_26122_cast_fp16")]; + tensor var_26124_equation_0 = const()[name = tensor("op_26124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26124_cast_fp16 = einsum(equation = var_26124_equation_0, values = (var_25598_cast_fp16, var_25998_cast_fp16))[name = tensor("op_26124_cast_fp16")]; + tensor var_26126_equation_0 = const()[name = tensor("op_26126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26126_cast_fp16 = einsum(equation = var_26126_equation_0, values = (var_25598_cast_fp16, var_25999_cast_fp16))[name = tensor("op_26126_cast_fp16")]; + tensor var_26128_equation_0 = const()[name = tensor("op_26128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26128_cast_fp16 = einsum(equation = var_26128_equation_0, values = (var_25598_cast_fp16, var_26000_cast_fp16))[name = tensor("op_26128_cast_fp16")]; + tensor var_26130_equation_0 = const()[name = tensor("op_26130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26130_cast_fp16 = einsum(equation = var_26130_equation_0, values = (var_25602_cast_fp16, var_26001_cast_fp16))[name = tensor("op_26130_cast_fp16")]; + tensor var_26132_equation_0 = const()[name = tensor("op_26132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26132_cast_fp16 = einsum(equation = var_26132_equation_0, values = (var_25602_cast_fp16, var_26002_cast_fp16))[name = tensor("op_26132_cast_fp16")]; + tensor var_26134_equation_0 = const()[name = tensor("op_26134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26134_cast_fp16 = einsum(equation = var_26134_equation_0, values = (var_25602_cast_fp16, var_26003_cast_fp16))[name = tensor("op_26134_cast_fp16")]; + tensor var_26136_equation_0 = const()[name = tensor("op_26136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26136_cast_fp16 = einsum(equation = var_26136_equation_0, values = (var_25602_cast_fp16, var_26004_cast_fp16))[name = tensor("op_26136_cast_fp16")]; + tensor var_26138_equation_0 = const()[name = tensor("op_26138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26138_cast_fp16 = einsum(equation = var_26138_equation_0, values = (var_25606_cast_fp16, var_26005_cast_fp16))[name = tensor("op_26138_cast_fp16")]; + tensor var_26140_equation_0 = const()[name = tensor("op_26140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26140_cast_fp16 = einsum(equation = var_26140_equation_0, values = (var_25606_cast_fp16, var_26006_cast_fp16))[name = tensor("op_26140_cast_fp16")]; + tensor var_26142_equation_0 = const()[name = tensor("op_26142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26142_cast_fp16 = einsum(equation = var_26142_equation_0, values = (var_25606_cast_fp16, var_26007_cast_fp16))[name = tensor("op_26142_cast_fp16")]; + tensor var_26144_equation_0 = const()[name = tensor("op_26144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26144_cast_fp16 = einsum(equation = var_26144_equation_0, values = (var_25606_cast_fp16, var_26008_cast_fp16))[name = tensor("op_26144_cast_fp16")]; + tensor var_26146_equation_0 = const()[name = tensor("op_26146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26146_cast_fp16 = einsum(equation = var_26146_equation_0, values = (var_25610_cast_fp16, var_26009_cast_fp16))[name = tensor("op_26146_cast_fp16")]; + tensor var_26148_equation_0 = const()[name = tensor("op_26148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26148_cast_fp16 = einsum(equation = var_26148_equation_0, values = (var_25610_cast_fp16, var_26010_cast_fp16))[name = tensor("op_26148_cast_fp16")]; + tensor var_26150_equation_0 = const()[name = tensor("op_26150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26150_cast_fp16 = einsum(equation = var_26150_equation_0, values = (var_25610_cast_fp16, var_26011_cast_fp16))[name = tensor("op_26150_cast_fp16")]; + tensor var_26152_equation_0 = const()[name = tensor("op_26152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26152_cast_fp16 = einsum(equation = var_26152_equation_0, values = (var_25610_cast_fp16, var_26012_cast_fp16))[name = tensor("op_26152_cast_fp16")]; + tensor var_26154_equation_0 = const()[name = tensor("op_26154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26154_cast_fp16 = einsum(equation = var_26154_equation_0, values = (var_25614_cast_fp16, var_26013_cast_fp16))[name = tensor("op_26154_cast_fp16")]; + tensor var_26156_equation_0 = const()[name = tensor("op_26156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26156_cast_fp16 = einsum(equation = var_26156_equation_0, values = (var_25614_cast_fp16, var_26014_cast_fp16))[name = tensor("op_26156_cast_fp16")]; + tensor var_26158_equation_0 = const()[name = tensor("op_26158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26158_cast_fp16 = einsum(equation = var_26158_equation_0, values = (var_25614_cast_fp16, var_26015_cast_fp16))[name = tensor("op_26158_cast_fp16")]; + tensor var_26160_equation_0 = const()[name = tensor("op_26160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26160_cast_fp16 = einsum(equation = var_26160_equation_0, values = (var_25614_cast_fp16, var_26016_cast_fp16))[name = tensor("op_26160_cast_fp16")]; + tensor var_26162_equation_0 = const()[name = tensor("op_26162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26162_cast_fp16 = einsum(equation = var_26162_equation_0, values = (var_25618_cast_fp16, var_26017_cast_fp16))[name = tensor("op_26162_cast_fp16")]; + tensor var_26164_equation_0 = const()[name = tensor("op_26164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26164_cast_fp16 = einsum(equation = var_26164_equation_0, values = (var_25618_cast_fp16, var_26018_cast_fp16))[name = tensor("op_26164_cast_fp16")]; + tensor var_26166_equation_0 = const()[name = tensor("op_26166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26166_cast_fp16 = einsum(equation = var_26166_equation_0, values = (var_25618_cast_fp16, var_26019_cast_fp16))[name = tensor("op_26166_cast_fp16")]; + tensor var_26168_equation_0 = const()[name = tensor("op_26168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26168_cast_fp16 = einsum(equation = var_26168_equation_0, values = (var_25618_cast_fp16, var_26020_cast_fp16))[name = tensor("op_26168_cast_fp16")]; + tensor var_26170_equation_0 = const()[name = tensor("op_26170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26170_cast_fp16 = einsum(equation = var_26170_equation_0, values = (var_25622_cast_fp16, var_26021_cast_fp16))[name = tensor("op_26170_cast_fp16")]; + tensor var_26172_equation_0 = const()[name = tensor("op_26172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26172_cast_fp16 = einsum(equation = var_26172_equation_0, values = (var_25622_cast_fp16, var_26022_cast_fp16))[name = tensor("op_26172_cast_fp16")]; + tensor var_26174_equation_0 = const()[name = tensor("op_26174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26174_cast_fp16 = einsum(equation = var_26174_equation_0, values = (var_25622_cast_fp16, var_26023_cast_fp16))[name = tensor("op_26174_cast_fp16")]; + tensor var_26176_equation_0 = const()[name = tensor("op_26176_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26176_cast_fp16 = einsum(equation = var_26176_equation_0, values = (var_25622_cast_fp16, var_26024_cast_fp16))[name = tensor("op_26176_cast_fp16")]; + tensor var_26178_equation_0 = const()[name = tensor("op_26178_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26178_cast_fp16 = einsum(equation = var_26178_equation_0, values = (var_25626_cast_fp16, var_26025_cast_fp16))[name = tensor("op_26178_cast_fp16")]; + tensor var_26180_equation_0 = const()[name = tensor("op_26180_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26180_cast_fp16 = einsum(equation = var_26180_equation_0, values = (var_25626_cast_fp16, var_26026_cast_fp16))[name = tensor("op_26180_cast_fp16")]; + tensor var_26182_equation_0 = const()[name = tensor("op_26182_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26182_cast_fp16 = einsum(equation = var_26182_equation_0, values = (var_25626_cast_fp16, var_26027_cast_fp16))[name = tensor("op_26182_cast_fp16")]; + tensor var_26184_equation_0 = const()[name = tensor("op_26184_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26184_cast_fp16 = einsum(equation = var_26184_equation_0, values = (var_25626_cast_fp16, var_26028_cast_fp16))[name = tensor("op_26184_cast_fp16")]; + tensor var_26186_equation_0 = const()[name = tensor("op_26186_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26186_cast_fp16 = einsum(equation = var_26186_equation_0, values = (var_25630_cast_fp16, var_26029_cast_fp16))[name = tensor("op_26186_cast_fp16")]; + tensor var_26188_equation_0 = const()[name = tensor("op_26188_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26188_cast_fp16 = einsum(equation = var_26188_equation_0, values = (var_25630_cast_fp16, var_26030_cast_fp16))[name = tensor("op_26188_cast_fp16")]; + tensor var_26190_equation_0 = const()[name = tensor("op_26190_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26190_cast_fp16 = einsum(equation = var_26190_equation_0, values = (var_25630_cast_fp16, var_26031_cast_fp16))[name = tensor("op_26190_cast_fp16")]; + tensor var_26192_equation_0 = const()[name = tensor("op_26192_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_26192_cast_fp16 = einsum(equation = var_26192_equation_0, values = (var_25630_cast_fp16, var_26032_cast_fp16))[name = tensor("op_26192_cast_fp16")]; + tensor var_26194_interleave_0 = const()[name = tensor("op_26194_interleave_0"), val = tensor(false)]; + tensor var_26194_cast_fp16 = concat(axis = var_24699, interleave = var_26194_interleave_0, values = (var_26034_cast_fp16, var_26036_cast_fp16, var_26038_cast_fp16, var_26040_cast_fp16))[name = tensor("op_26194_cast_fp16")]; + tensor var_26196_interleave_0 = const()[name = tensor("op_26196_interleave_0"), val = tensor(false)]; + tensor var_26196_cast_fp16 = concat(axis = var_24699, interleave = var_26196_interleave_0, values = (var_26042_cast_fp16, var_26044_cast_fp16, var_26046_cast_fp16, var_26048_cast_fp16))[name = tensor("op_26196_cast_fp16")]; + tensor var_26198_interleave_0 = const()[name = tensor("op_26198_interleave_0"), val = tensor(false)]; + tensor var_26198_cast_fp16 = concat(axis = var_24699, interleave = var_26198_interleave_0, values = (var_26050_cast_fp16, var_26052_cast_fp16, var_26054_cast_fp16, var_26056_cast_fp16))[name = tensor("op_26198_cast_fp16")]; + tensor var_26200_interleave_0 = const()[name = tensor("op_26200_interleave_0"), val = tensor(false)]; + tensor var_26200_cast_fp16 = concat(axis = var_24699, interleave = var_26200_interleave_0, values = (var_26058_cast_fp16, var_26060_cast_fp16, var_26062_cast_fp16, var_26064_cast_fp16))[name = tensor("op_26200_cast_fp16")]; + tensor var_26202_interleave_0 = const()[name = tensor("op_26202_interleave_0"), val = tensor(false)]; + tensor var_26202_cast_fp16 = concat(axis = var_24699, interleave = var_26202_interleave_0, values = (var_26066_cast_fp16, var_26068_cast_fp16, var_26070_cast_fp16, var_26072_cast_fp16))[name = tensor("op_26202_cast_fp16")]; + tensor var_26204_interleave_0 = const()[name = tensor("op_26204_interleave_0"), val = tensor(false)]; + tensor var_26204_cast_fp16 = concat(axis = var_24699, interleave = var_26204_interleave_0, values = (var_26074_cast_fp16, var_26076_cast_fp16, var_26078_cast_fp16, var_26080_cast_fp16))[name = tensor("op_26204_cast_fp16")]; + tensor var_26206_interleave_0 = const()[name = tensor("op_26206_interleave_0"), val = tensor(false)]; + tensor var_26206_cast_fp16 = concat(axis = var_24699, interleave = var_26206_interleave_0, values = (var_26082_cast_fp16, var_26084_cast_fp16, var_26086_cast_fp16, var_26088_cast_fp16))[name = tensor("op_26206_cast_fp16")]; + tensor var_26208_interleave_0 = const()[name = tensor("op_26208_interleave_0"), val = tensor(false)]; + tensor var_26208_cast_fp16 = concat(axis = var_24699, interleave = var_26208_interleave_0, values = (var_26090_cast_fp16, var_26092_cast_fp16, var_26094_cast_fp16, var_26096_cast_fp16))[name = tensor("op_26208_cast_fp16")]; + tensor var_26210_interleave_0 = const()[name = tensor("op_26210_interleave_0"), val = tensor(false)]; + tensor var_26210_cast_fp16 = concat(axis = var_24699, interleave = var_26210_interleave_0, values = (var_26098_cast_fp16, var_26100_cast_fp16, var_26102_cast_fp16, var_26104_cast_fp16))[name = tensor("op_26210_cast_fp16")]; + tensor var_26212_interleave_0 = const()[name = tensor("op_26212_interleave_0"), val = tensor(false)]; + tensor var_26212_cast_fp16 = concat(axis = var_24699, interleave = var_26212_interleave_0, values = (var_26106_cast_fp16, var_26108_cast_fp16, var_26110_cast_fp16, var_26112_cast_fp16))[name = tensor("op_26212_cast_fp16")]; + tensor var_26214_interleave_0 = const()[name = tensor("op_26214_interleave_0"), val = tensor(false)]; + tensor var_26214_cast_fp16 = concat(axis = var_24699, interleave = var_26214_interleave_0, values = (var_26114_cast_fp16, var_26116_cast_fp16, var_26118_cast_fp16, var_26120_cast_fp16))[name = tensor("op_26214_cast_fp16")]; + tensor var_26216_interleave_0 = const()[name = tensor("op_26216_interleave_0"), val = tensor(false)]; + tensor var_26216_cast_fp16 = concat(axis = var_24699, interleave = var_26216_interleave_0, values = (var_26122_cast_fp16, var_26124_cast_fp16, var_26126_cast_fp16, var_26128_cast_fp16))[name = tensor("op_26216_cast_fp16")]; + tensor var_26218_interleave_0 = const()[name = tensor("op_26218_interleave_0"), val = tensor(false)]; + tensor var_26218_cast_fp16 = concat(axis = var_24699, interleave = var_26218_interleave_0, values = (var_26130_cast_fp16, var_26132_cast_fp16, var_26134_cast_fp16, var_26136_cast_fp16))[name = tensor("op_26218_cast_fp16")]; + tensor var_26220_interleave_0 = const()[name = tensor("op_26220_interleave_0"), val = tensor(false)]; + tensor var_26220_cast_fp16 = concat(axis = var_24699, interleave = var_26220_interleave_0, values = (var_26138_cast_fp16, var_26140_cast_fp16, var_26142_cast_fp16, var_26144_cast_fp16))[name = tensor("op_26220_cast_fp16")]; + tensor var_26222_interleave_0 = const()[name = tensor("op_26222_interleave_0"), val = tensor(false)]; + tensor var_26222_cast_fp16 = concat(axis = var_24699, interleave = var_26222_interleave_0, values = (var_26146_cast_fp16, var_26148_cast_fp16, var_26150_cast_fp16, var_26152_cast_fp16))[name = tensor("op_26222_cast_fp16")]; + tensor var_26224_interleave_0 = const()[name = tensor("op_26224_interleave_0"), val = tensor(false)]; + tensor var_26224_cast_fp16 = concat(axis = var_24699, interleave = var_26224_interleave_0, values = (var_26154_cast_fp16, var_26156_cast_fp16, var_26158_cast_fp16, var_26160_cast_fp16))[name = tensor("op_26224_cast_fp16")]; + tensor var_26226_interleave_0 = const()[name = tensor("op_26226_interleave_0"), val = tensor(false)]; + tensor var_26226_cast_fp16 = concat(axis = var_24699, interleave = var_26226_interleave_0, values = (var_26162_cast_fp16, var_26164_cast_fp16, var_26166_cast_fp16, var_26168_cast_fp16))[name = tensor("op_26226_cast_fp16")]; + tensor var_26228_interleave_0 = const()[name = tensor("op_26228_interleave_0"), val = tensor(false)]; + tensor var_26228_cast_fp16 = concat(axis = var_24699, interleave = var_26228_interleave_0, values = (var_26170_cast_fp16, var_26172_cast_fp16, var_26174_cast_fp16, var_26176_cast_fp16))[name = tensor("op_26228_cast_fp16")]; + tensor var_26230_interleave_0 = const()[name = tensor("op_26230_interleave_0"), val = tensor(false)]; + tensor var_26230_cast_fp16 = concat(axis = var_24699, interleave = var_26230_interleave_0, values = (var_26178_cast_fp16, var_26180_cast_fp16, var_26182_cast_fp16, var_26184_cast_fp16))[name = tensor("op_26230_cast_fp16")]; + tensor var_26232_interleave_0 = const()[name = tensor("op_26232_interleave_0"), val = tensor(false)]; + tensor var_26232_cast_fp16 = concat(axis = var_24699, interleave = var_26232_interleave_0, values = (var_26186_cast_fp16, var_26188_cast_fp16, var_26190_cast_fp16, var_26192_cast_fp16))[name = tensor("op_26232_cast_fp16")]; + tensor input_307_interleave_0 = const()[name = tensor("input_307_interleave_0"), val = tensor(false)]; + tensor input_307_cast_fp16 = concat(axis = var_24724, interleave = input_307_interleave_0, values = (var_26194_cast_fp16, var_26196_cast_fp16, var_26198_cast_fp16, var_26200_cast_fp16, var_26202_cast_fp16, var_26204_cast_fp16, var_26206_cast_fp16, var_26208_cast_fp16, var_26210_cast_fp16, var_26212_cast_fp16, var_26214_cast_fp16, var_26216_cast_fp16, var_26218_cast_fp16, var_26220_cast_fp16, var_26222_cast_fp16, var_26224_cast_fp16, var_26226_cast_fp16, var_26228_cast_fp16, var_26230_cast_fp16, var_26232_cast_fp16))[name = tensor("input_307_cast_fp16")]; + tensor var_26240 = const()[name = tensor("op_26240"), val = tensor([1, 1])]; + tensor var_26242 = const()[name = tensor("op_26242"), val = tensor([1, 1])]; + tensor pretrained_out_187_pad_type_0 = const()[name = tensor("pretrained_out_187_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_187_pad_0 = const()[name = tensor("pretrained_out_187_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176034112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176853376))), name = tensor("layers_15_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_15_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176853504)))]; + tensor pretrained_out_187_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_26242, groups = var_24724, pad = pretrained_out_187_pad_0, pad_type = pretrained_out_187_pad_type_0, strides = var_26240, weight = layers_15_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_307_cast_fp16)[name = tensor("pretrained_out_187_cast_fp16")]; + tensor var_26246 = const()[name = tensor("op_26246"), val = tensor([1, 1])]; + tensor var_26248 = const()[name = tensor("op_26248"), val = tensor([1, 1])]; + tensor input_309_pad_type_0 = const()[name = tensor("input_309_pad_type_0"), val = tensor("custom")]; + tensor input_309_pad_0 = const()[name = tensor("input_309_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176856128)))]; + tensor input_309_cast_fp16 = conv(dilations = var_26248, groups = var_24724, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = var_26246, weight = layers_15_self_attn_o_proj_loraA_weight_to_fp16, x = input_307_cast_fp16)[name = tensor("input_309_cast_fp16")]; + tensor var_26252 = const()[name = tensor("op_26252"), val = tensor([1, 1])]; + tensor var_26254 = const()[name = tensor("op_26254"), val = tensor([1, 1])]; + tensor lora_out_373_pad_type_0 = const()[name = tensor("lora_out_373_pad_type_0"), val = tensor("custom")]; + tensor lora_out_373_pad_0 = const()[name = tensor("lora_out_373_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_375_weight_0_to_fp16 = const()[name = tensor("lora_out_375_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176897152)))]; + tensor lora_out_375_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_26254, groups = var_24724, pad = lora_out_373_pad_0, pad_type = lora_out_373_pad_type_0, strides = var_26252, weight = lora_out_375_weight_0_to_fp16, x = input_309_cast_fp16)[name = tensor("lora_out_375_cast_fp16")]; + tensor obj_63_cast_fp16 = add(x = pretrained_out_187_cast_fp16, y = lora_out_375_cast_fp16)[name = tensor("obj_63_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = obj_63_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; + tensor var_26263 = const()[name = tensor("op_26263"), val = tensor([1])]; + tensor channels_mean_63_cast_fp16 = reduce_mean(axes = var_26263, keep_dims = var_24725, x = inputs_63_cast_fp16)[name = tensor("channels_mean_63_cast_fp16")]; + tensor zero_mean_63_cast_fp16 = sub(x = inputs_63_cast_fp16, y = channels_mean_63_cast_fp16)[name = tensor("zero_mean_63_cast_fp16")]; + tensor zero_mean_sq_63_cast_fp16 = mul(x = zero_mean_63_cast_fp16, y = zero_mean_63_cast_fp16)[name = tensor("zero_mean_sq_63_cast_fp16")]; + tensor var_26267 = const()[name = tensor("op_26267"), val = tensor([1])]; + tensor var_26268_cast_fp16 = reduce_mean(axes = var_26267, keep_dims = var_24725, x = zero_mean_sq_63_cast_fp16)[name = tensor("op_26268_cast_fp16")]; + tensor var_26269_to_fp16 = const()[name = tensor("op_26269_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_26270_cast_fp16 = add(x = var_26268_cast_fp16, y = var_26269_to_fp16)[name = tensor("op_26270_cast_fp16")]; + tensor denom_63_epsilon_0 = const()[name = tensor("denom_63_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_63_cast_fp16 = rsqrt(epsilon = denom_63_epsilon_0, x = var_26270_cast_fp16)[name = tensor("denom_63_cast_fp16")]; + tensor out_63_cast_fp16 = mul(x = zero_mean_63_cast_fp16, y = denom_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; + tensor input_311_gamma_0_to_fp16 = const()[name = tensor("input_311_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176938176)))]; + tensor input_311_beta_0_to_fp16 = const()[name = tensor("input_311_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176940800)))]; + tensor input_311_epsilon_0_to_fp16 = const()[name = tensor("input_311_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_311_cast_fp16 = batch_norm(beta = input_311_beta_0_to_fp16, epsilon = input_311_epsilon_0_to_fp16, gamma = input_311_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor("input_311_cast_fp16")]; + tensor var_26284 = const()[name = tensor("op_26284"), val = tensor([1, 1])]; + tensor var_26286 = const()[name = tensor("op_26286"), val = tensor([1, 1])]; + tensor pretrained_out_189_pad_type_0 = const()[name = tensor("pretrained_out_189_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_189_pad_0 = const()[name = tensor("pretrained_out_189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176943424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180220288))), name = tensor("layers_15_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_15_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_15_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180220416)))]; + tensor pretrained_out_189_cast_fp16 = conv(bias = layers_15_fc1_pretrained_bias_to_fp16, dilations = var_26286, groups = var_24724, pad = pretrained_out_189_pad_0, pad_type = pretrained_out_189_pad_type_0, strides = var_26284, weight = layers_15_fc1_pretrained_weight_to_fp16_palettized, x = input_311_cast_fp16)[name = tensor("pretrained_out_189_cast_fp16")]; + tensor var_26290 = const()[name = tensor("op_26290"), val = tensor([1, 1])]; + tensor var_26292 = const()[name = tensor("op_26292"), val = tensor([1, 1])]; + tensor input_313_pad_type_0 = const()[name = tensor("input_313_pad_type_0"), val = tensor("custom")]; + tensor input_313_pad_0 = const()[name = tensor("input_313_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_15_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180230720)))]; + tensor input_313_cast_fp16 = conv(dilations = var_26292, groups = var_24724, pad = input_313_pad_0, pad_type = input_313_pad_type_0, strides = var_26290, weight = layers_15_fc1_loraA_weight_to_fp16, x = input_311_cast_fp16)[name = tensor("input_313_cast_fp16")]; + tensor var_26296 = const()[name = tensor("op_26296"), val = tensor([1, 1])]; + tensor var_26298 = const()[name = tensor("op_26298"), val = tensor([1, 1])]; + tensor lora_out_377_pad_type_0 = const()[name = tensor("lora_out_377_pad_type_0"), val = tensor("custom")]; + tensor lora_out_377_pad_0 = const()[name = tensor("lora_out_377_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_379_weight_0_to_fp16 = const()[name = tensor("lora_out_379_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180271744)))]; + tensor lora_out_379_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_26298, groups = var_24724, pad = lora_out_377_pad_0, pad_type = lora_out_377_pad_type_0, strides = var_26296, weight = lora_out_379_weight_0_to_fp16, x = input_313_cast_fp16)[name = tensor("lora_out_379_cast_fp16")]; + tensor input_315_cast_fp16 = add(x = pretrained_out_189_cast_fp16, y = lora_out_379_cast_fp16)[name = tensor("input_315_cast_fp16")]; + tensor input_317_mode_0 = const()[name = tensor("input_317_mode_0"), val = tensor("EXACT")]; + tensor input_317_cast_fp16 = gelu(mode = input_317_mode_0, x = input_315_cast_fp16)[name = tensor("input_317_cast_fp16")]; + tensor var_26310 = const()[name = tensor("op_26310"), val = tensor([1, 1])]; + tensor var_26312 = const()[name = tensor("op_26312"), val = tensor([1, 1])]; + tensor pretrained_out_191_pad_type_0 = const()[name = tensor("pretrained_out_191_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_191_pad_0 = const()[name = tensor("pretrained_out_191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180435648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183712512))), name = tensor("layers_15_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_15_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_15_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183712640)))]; + tensor pretrained_out_191_cast_fp16 = conv(bias = layers_15_fc2_pretrained_bias_to_fp16, dilations = var_26312, groups = var_24724, pad = pretrained_out_191_pad_0, pad_type = pretrained_out_191_pad_type_0, strides = var_26310, weight = layers_15_fc2_pretrained_weight_to_fp16_palettized, x = input_317_cast_fp16)[name = tensor("pretrained_out_191_cast_fp16")]; + tensor var_26316 = const()[name = tensor("op_26316"), val = tensor([1, 1])]; + tensor var_26318 = const()[name = tensor("op_26318"), val = tensor([1, 1])]; + tensor input_319_pad_type_0 = const()[name = tensor("input_319_pad_type_0"), val = tensor("custom")]; + tensor input_319_pad_0 = const()[name = tensor("input_319_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_15_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_15_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183715264)))]; + tensor input_319_cast_fp16 = conv(dilations = var_26318, groups = var_24724, pad = input_319_pad_0, pad_type = input_319_pad_type_0, strides = var_26316, weight = layers_15_fc2_loraA_weight_to_fp16, x = input_317_cast_fp16)[name = tensor("input_319_cast_fp16")]; + tensor var_26322 = const()[name = tensor("op_26322"), val = tensor([1, 1])]; + tensor var_26324 = const()[name = tensor("op_26324"), val = tensor([1, 1])]; + tensor lora_out_381_pad_type_0 = const()[name = tensor("lora_out_381_pad_type_0"), val = tensor("custom")]; + tensor lora_out_381_pad_0 = const()[name = tensor("lora_out_381_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_383_weight_0_to_fp16 = const()[name = tensor("lora_out_383_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183879168)))]; + tensor lora_out_383_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_26324, groups = var_24724, pad = lora_out_381_pad_0, pad_type = lora_out_381_pad_type_0, strides = var_26322, weight = lora_out_383_weight_0_to_fp16, x = input_319_cast_fp16)[name = tensor("lora_out_383_cast_fp16")]; + tensor hidden_states_35_cast_fp16 = add(x = pretrained_out_191_cast_fp16, y = lora_out_383_cast_fp16)[name = tensor("hidden_states_35_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = hidden_states_35_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; + tensor var_26334 = const()[name = tensor("op_26334"), val = tensor(3)]; + tensor var_26359 = const()[name = tensor("op_26359"), val = tensor(1)]; + tensor var_26360 = const()[name = tensor("op_26360"), val = tensor(true)]; + tensor var_26370 = const()[name = tensor("op_26370"), val = tensor([1])]; + tensor channels_mean_65_cast_fp16 = reduce_mean(axes = var_26370, keep_dims = var_26360, x = inputs_65_cast_fp16)[name = tensor("channels_mean_65_cast_fp16")]; + tensor zero_mean_65_cast_fp16 = sub(x = inputs_65_cast_fp16, y = channels_mean_65_cast_fp16)[name = tensor("zero_mean_65_cast_fp16")]; + tensor zero_mean_sq_65_cast_fp16 = mul(x = zero_mean_65_cast_fp16, y = zero_mean_65_cast_fp16)[name = tensor("zero_mean_sq_65_cast_fp16")]; + tensor var_26374 = const()[name = tensor("op_26374"), val = tensor([1])]; + tensor var_26375_cast_fp16 = reduce_mean(axes = var_26374, keep_dims = var_26360, x = zero_mean_sq_65_cast_fp16)[name = tensor("op_26375_cast_fp16")]; + tensor var_26376_to_fp16 = const()[name = tensor("op_26376_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_26377_cast_fp16 = add(x = var_26375_cast_fp16, y = var_26376_to_fp16)[name = tensor("op_26377_cast_fp16")]; + tensor denom_65_epsilon_0 = const()[name = tensor("denom_65_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_65_cast_fp16 = rsqrt(epsilon = denom_65_epsilon_0, x = var_26377_cast_fp16)[name = tensor("denom_65_cast_fp16")]; + tensor out_65_cast_fp16 = mul(x = zero_mean_65_cast_fp16, y = denom_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; + tensor obj_65_gamma_0_to_fp16 = const()[name = tensor("obj_65_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183920192)))]; + tensor obj_65_beta_0_to_fp16 = const()[name = tensor("obj_65_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183922816)))]; + tensor obj_65_epsilon_0_to_fp16 = const()[name = tensor("obj_65_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_65_cast_fp16 = batch_norm(beta = obj_65_beta_0_to_fp16, epsilon = obj_65_epsilon_0_to_fp16, gamma = obj_65_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor("obj_65_cast_fp16")]; + tensor var_26395 = const()[name = tensor("op_26395"), val = tensor([1, 1])]; + tensor var_26397 = const()[name = tensor("op_26397"), val = tensor([1, 1])]; + tensor pretrained_out_193_pad_type_0 = const()[name = tensor("pretrained_out_193_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_193_pad_0 = const()[name = tensor("pretrained_out_193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(183925440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184744704))), name = tensor("layers_16_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_16_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184744832)))]; + tensor pretrained_out_193_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_26397, groups = var_26359, pad = pretrained_out_193_pad_0, pad_type = pretrained_out_193_pad_type_0, strides = var_26395, weight = layers_16_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_65_cast_fp16)[name = tensor("pretrained_out_193_cast_fp16")]; + tensor var_26401 = const()[name = tensor("op_26401"), val = tensor([1, 1])]; + tensor var_26403 = const()[name = tensor("op_26403"), val = tensor([1, 1])]; + tensor input_321_pad_type_0 = const()[name = tensor("input_321_pad_type_0"), val = tensor("custom")]; + tensor input_321_pad_0 = const()[name = tensor("input_321_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184747456)))]; + tensor input_321_cast_fp16 = conv(dilations = var_26403, groups = var_26359, pad = input_321_pad_0, pad_type = input_321_pad_type_0, strides = var_26401, weight = layers_16_self_attn_q_proj_loraA_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor("input_321_cast_fp16")]; + tensor var_26407 = const()[name = tensor("op_26407"), val = tensor([1, 1])]; + tensor var_26409 = const()[name = tensor("op_26409"), val = tensor([1, 1])]; + tensor lora_out_385_pad_type_0 = const()[name = tensor("lora_out_385_pad_type_0"), val = tensor("custom")]; + tensor lora_out_385_pad_0 = const()[name = tensor("lora_out_385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_387_weight_0_to_fp16 = const()[name = tensor("lora_out_387_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184788480)))]; + tensor lora_out_387_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_26409, groups = var_26359, pad = lora_out_385_pad_0, pad_type = lora_out_385_pad_type_0, strides = var_26407, weight = lora_out_387_weight_0_to_fp16, x = input_321_cast_fp16)[name = tensor("lora_out_387_cast_fp16")]; + tensor query_33_cast_fp16 = add(x = pretrained_out_193_cast_fp16, y = lora_out_387_cast_fp16)[name = tensor("query_33_cast_fp16")]; + tensor var_26419 = const()[name = tensor("op_26419"), val = tensor([1, 1])]; + tensor var_26421 = const()[name = tensor("op_26421"), val = tensor([1, 1])]; + tensor pretrained_out_195_pad_type_0 = const()[name = tensor("pretrained_out_195_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_195_pad_0 = const()[name = tensor("pretrained_out_195_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184829504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185648768))), name = tensor("layers_16_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_195_cast_fp16 = conv(dilations = var_26421, groups = var_26359, pad = pretrained_out_195_pad_0, pad_type = pretrained_out_195_pad_type_0, strides = var_26419, weight = layers_16_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_65_cast_fp16)[name = tensor("pretrained_out_195_cast_fp16")]; + tensor var_26425 = const()[name = tensor("op_26425"), val = tensor([1, 1])]; + tensor var_26427 = const()[name = tensor("op_26427"), val = tensor([1, 1])]; + tensor input_323_pad_type_0 = const()[name = tensor("input_323_pad_type_0"), val = tensor("custom")]; + tensor input_323_pad_0 = const()[name = tensor("input_323_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185648896)))]; + tensor input_323_cast_fp16 = conv(dilations = var_26427, groups = var_26359, pad = input_323_pad_0, pad_type = input_323_pad_type_0, strides = var_26425, weight = layers_16_self_attn_k_proj_loraA_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor("input_323_cast_fp16")]; + tensor var_26431 = const()[name = tensor("op_26431"), val = tensor([1, 1])]; + tensor var_26433 = const()[name = tensor("op_26433"), val = tensor([1, 1])]; + tensor lora_out_389_pad_type_0 = const()[name = tensor("lora_out_389_pad_type_0"), val = tensor("custom")]; + tensor lora_out_389_pad_0 = const()[name = tensor("lora_out_389_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_391_weight_0_to_fp16 = const()[name = tensor("lora_out_391_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185689920)))]; + tensor lora_out_391_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_26433, groups = var_26359, pad = lora_out_389_pad_0, pad_type = lora_out_389_pad_type_0, strides = var_26431, weight = lora_out_391_weight_0_to_fp16, x = input_323_cast_fp16)[name = tensor("lora_out_391_cast_fp16")]; + tensor key_33_cast_fp16 = add(x = pretrained_out_195_cast_fp16, y = lora_out_391_cast_fp16)[name = tensor("key_33_cast_fp16")]; + tensor var_26444 = const()[name = tensor("op_26444"), val = tensor([1, 1])]; + tensor var_26446 = const()[name = tensor("op_26446"), val = tensor([1, 1])]; + tensor pretrained_out_197_pad_type_0 = const()[name = tensor("pretrained_out_197_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_197_pad_0 = const()[name = tensor("pretrained_out_197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185730944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186550208))), name = tensor("layers_16_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_16_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186550336)))]; + tensor pretrained_out_197_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_26446, groups = var_26359, pad = pretrained_out_197_pad_0, pad_type = pretrained_out_197_pad_type_0, strides = var_26444, weight = layers_16_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_65_cast_fp16)[name = tensor("pretrained_out_197_cast_fp16")]; + tensor var_26450 = const()[name = tensor("op_26450"), val = tensor([1, 1])]; + tensor var_26452 = const()[name = tensor("op_26452"), val = tensor([1, 1])]; + tensor input_325_pad_type_0 = const()[name = tensor("input_325_pad_type_0"), val = tensor("custom")]; + tensor input_325_pad_0 = const()[name = tensor("input_325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186552960)))]; + tensor input_325_cast_fp16 = conv(dilations = var_26452, groups = var_26359, pad = input_325_pad_0, pad_type = input_325_pad_type_0, strides = var_26450, weight = layers_16_self_attn_v_proj_loraA_weight_to_fp16, x = obj_65_cast_fp16)[name = tensor("input_325_cast_fp16")]; + tensor var_26456 = const()[name = tensor("op_26456"), val = tensor([1, 1])]; + tensor var_26458 = const()[name = tensor("op_26458"), val = tensor([1, 1])]; + tensor lora_out_393_pad_type_0 = const()[name = tensor("lora_out_393_pad_type_0"), val = tensor("custom")]; + tensor lora_out_393_pad_0 = const()[name = tensor("lora_out_393_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_395_weight_0_to_fp16 = const()[name = tensor("lora_out_395_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186593984)))]; + tensor lora_out_395_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_26458, groups = var_26359, pad = lora_out_393_pad_0, pad_type = lora_out_393_pad_type_0, strides = var_26456, weight = lora_out_395_weight_0_to_fp16, x = input_325_cast_fp16)[name = tensor("lora_out_395_cast_fp16")]; + tensor value_33_cast_fp16 = add(x = pretrained_out_197_cast_fp16, y = lora_out_395_cast_fp16)[name = tensor("value_33_cast_fp16")]; + tensor var_26468_begin_0 = const()[name = tensor("op_26468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26468_end_0 = const()[name = tensor("op_26468_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26468_end_mask_0 = const()[name = tensor("op_26468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26468_cast_fp16 = slice_by_index(begin = var_26468_begin_0, end = var_26468_end_0, end_mask = var_26468_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26468_cast_fp16")]; + tensor var_26472_begin_0 = const()[name = tensor("op_26472_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_26472_end_0 = const()[name = tensor("op_26472_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_26472_end_mask_0 = const()[name = tensor("op_26472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26472_cast_fp16 = slice_by_index(begin = var_26472_begin_0, end = var_26472_end_0, end_mask = var_26472_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26472_cast_fp16")]; + tensor var_26476_begin_0 = const()[name = tensor("op_26476_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_26476_end_0 = const()[name = tensor("op_26476_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_26476_end_mask_0 = const()[name = tensor("op_26476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26476_cast_fp16 = slice_by_index(begin = var_26476_begin_0, end = var_26476_end_0, end_mask = var_26476_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26476_cast_fp16")]; + tensor var_26480_begin_0 = const()[name = tensor("op_26480_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_26480_end_0 = const()[name = tensor("op_26480_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_26480_end_mask_0 = const()[name = tensor("op_26480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26480_cast_fp16 = slice_by_index(begin = var_26480_begin_0, end = var_26480_end_0, end_mask = var_26480_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26480_cast_fp16")]; + tensor var_26484_begin_0 = const()[name = tensor("op_26484_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_26484_end_0 = const()[name = tensor("op_26484_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_26484_end_mask_0 = const()[name = tensor("op_26484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26484_cast_fp16 = slice_by_index(begin = var_26484_begin_0, end = var_26484_end_0, end_mask = var_26484_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26484_cast_fp16")]; + tensor var_26488_begin_0 = const()[name = tensor("op_26488_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_26488_end_0 = const()[name = tensor("op_26488_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_26488_end_mask_0 = const()[name = tensor("op_26488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26488_cast_fp16 = slice_by_index(begin = var_26488_begin_0, end = var_26488_end_0, end_mask = var_26488_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26488_cast_fp16")]; + tensor var_26492_begin_0 = const()[name = tensor("op_26492_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_26492_end_0 = const()[name = tensor("op_26492_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_26492_end_mask_0 = const()[name = tensor("op_26492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26492_cast_fp16 = slice_by_index(begin = var_26492_begin_0, end = var_26492_end_0, end_mask = var_26492_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26492_cast_fp16")]; + tensor var_26496_begin_0 = const()[name = tensor("op_26496_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_26496_end_0 = const()[name = tensor("op_26496_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_26496_end_mask_0 = const()[name = tensor("op_26496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26496_cast_fp16 = slice_by_index(begin = var_26496_begin_0, end = var_26496_end_0, end_mask = var_26496_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26496_cast_fp16")]; + tensor var_26500_begin_0 = const()[name = tensor("op_26500_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_26500_end_0 = const()[name = tensor("op_26500_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_26500_end_mask_0 = const()[name = tensor("op_26500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26500_cast_fp16 = slice_by_index(begin = var_26500_begin_0, end = var_26500_end_0, end_mask = var_26500_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26500_cast_fp16")]; + tensor var_26504_begin_0 = const()[name = tensor("op_26504_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_26504_end_0 = const()[name = tensor("op_26504_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_26504_end_mask_0 = const()[name = tensor("op_26504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26504_cast_fp16 = slice_by_index(begin = var_26504_begin_0, end = var_26504_end_0, end_mask = var_26504_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26504_cast_fp16")]; + tensor var_26508_begin_0 = const()[name = tensor("op_26508_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_26508_end_0 = const()[name = tensor("op_26508_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_26508_end_mask_0 = const()[name = tensor("op_26508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26508_cast_fp16 = slice_by_index(begin = var_26508_begin_0, end = var_26508_end_0, end_mask = var_26508_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26508_cast_fp16")]; + tensor var_26512_begin_0 = const()[name = tensor("op_26512_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_26512_end_0 = const()[name = tensor("op_26512_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_26512_end_mask_0 = const()[name = tensor("op_26512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26512_cast_fp16 = slice_by_index(begin = var_26512_begin_0, end = var_26512_end_0, end_mask = var_26512_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26512_cast_fp16")]; + tensor var_26516_begin_0 = const()[name = tensor("op_26516_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_26516_end_0 = const()[name = tensor("op_26516_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_26516_end_mask_0 = const()[name = tensor("op_26516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26516_cast_fp16 = slice_by_index(begin = var_26516_begin_0, end = var_26516_end_0, end_mask = var_26516_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26516_cast_fp16")]; + tensor var_26520_begin_0 = const()[name = tensor("op_26520_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_26520_end_0 = const()[name = tensor("op_26520_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_26520_end_mask_0 = const()[name = tensor("op_26520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26520_cast_fp16 = slice_by_index(begin = var_26520_begin_0, end = var_26520_end_0, end_mask = var_26520_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26520_cast_fp16")]; + tensor var_26524_begin_0 = const()[name = tensor("op_26524_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_26524_end_0 = const()[name = tensor("op_26524_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_26524_end_mask_0 = const()[name = tensor("op_26524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26524_cast_fp16 = slice_by_index(begin = var_26524_begin_0, end = var_26524_end_0, end_mask = var_26524_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26524_cast_fp16")]; + tensor var_26528_begin_0 = const()[name = tensor("op_26528_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_26528_end_0 = const()[name = tensor("op_26528_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_26528_end_mask_0 = const()[name = tensor("op_26528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26528_cast_fp16 = slice_by_index(begin = var_26528_begin_0, end = var_26528_end_0, end_mask = var_26528_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26528_cast_fp16")]; + tensor var_26532_begin_0 = const()[name = tensor("op_26532_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_26532_end_0 = const()[name = tensor("op_26532_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_26532_end_mask_0 = const()[name = tensor("op_26532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26532_cast_fp16 = slice_by_index(begin = var_26532_begin_0, end = var_26532_end_0, end_mask = var_26532_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26532_cast_fp16")]; + tensor var_26536_begin_0 = const()[name = tensor("op_26536_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_26536_end_0 = const()[name = tensor("op_26536_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_26536_end_mask_0 = const()[name = tensor("op_26536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26536_cast_fp16 = slice_by_index(begin = var_26536_begin_0, end = var_26536_end_0, end_mask = var_26536_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26536_cast_fp16")]; + tensor var_26540_begin_0 = const()[name = tensor("op_26540_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_26540_end_0 = const()[name = tensor("op_26540_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_26540_end_mask_0 = const()[name = tensor("op_26540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26540_cast_fp16 = slice_by_index(begin = var_26540_begin_0, end = var_26540_end_0, end_mask = var_26540_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26540_cast_fp16")]; + tensor var_26544_begin_0 = const()[name = tensor("op_26544_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_26544_end_0 = const()[name = tensor("op_26544_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_26544_end_mask_0 = const()[name = tensor("op_26544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_26544_cast_fp16 = slice_by_index(begin = var_26544_begin_0, end = var_26544_end_0, end_mask = var_26544_end_mask_0, x = query_33_cast_fp16)[name = tensor("op_26544_cast_fp16")]; + tensor var_26553_begin_0 = const()[name = tensor("op_26553_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26553_end_0 = const()[name = tensor("op_26553_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26553_end_mask_0 = const()[name = tensor("op_26553_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26553_cast_fp16 = slice_by_index(begin = var_26553_begin_0, end = var_26553_end_0, end_mask = var_26553_end_mask_0, x = var_26468_cast_fp16)[name = tensor("op_26553_cast_fp16")]; + tensor var_26560_begin_0 = const()[name = tensor("op_26560_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26560_end_0 = const()[name = tensor("op_26560_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26560_end_mask_0 = const()[name = tensor("op_26560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26560_cast_fp16 = slice_by_index(begin = var_26560_begin_0, end = var_26560_end_0, end_mask = var_26560_end_mask_0, x = var_26468_cast_fp16)[name = tensor("op_26560_cast_fp16")]; + tensor var_26567_begin_0 = const()[name = tensor("op_26567_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26567_end_0 = const()[name = tensor("op_26567_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26567_end_mask_0 = const()[name = tensor("op_26567_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26567_cast_fp16 = slice_by_index(begin = var_26567_begin_0, end = var_26567_end_0, end_mask = var_26567_end_mask_0, x = var_26468_cast_fp16)[name = tensor("op_26567_cast_fp16")]; + tensor var_26574_begin_0 = const()[name = tensor("op_26574_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26574_end_0 = const()[name = tensor("op_26574_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26574_end_mask_0 = const()[name = tensor("op_26574_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26574_cast_fp16 = slice_by_index(begin = var_26574_begin_0, end = var_26574_end_0, end_mask = var_26574_end_mask_0, x = var_26468_cast_fp16)[name = tensor("op_26574_cast_fp16")]; + tensor var_26581_begin_0 = const()[name = tensor("op_26581_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26581_end_0 = const()[name = tensor("op_26581_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26581_end_mask_0 = const()[name = tensor("op_26581_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26581_cast_fp16 = slice_by_index(begin = var_26581_begin_0, end = var_26581_end_0, end_mask = var_26581_end_mask_0, x = var_26472_cast_fp16)[name = tensor("op_26581_cast_fp16")]; + tensor var_26588_begin_0 = const()[name = tensor("op_26588_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26588_end_0 = const()[name = tensor("op_26588_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26588_end_mask_0 = const()[name = tensor("op_26588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26588_cast_fp16 = slice_by_index(begin = var_26588_begin_0, end = var_26588_end_0, end_mask = var_26588_end_mask_0, x = var_26472_cast_fp16)[name = tensor("op_26588_cast_fp16")]; + tensor var_26595_begin_0 = const()[name = tensor("op_26595_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26595_end_0 = const()[name = tensor("op_26595_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26595_end_mask_0 = const()[name = tensor("op_26595_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26595_cast_fp16 = slice_by_index(begin = var_26595_begin_0, end = var_26595_end_0, end_mask = var_26595_end_mask_0, x = var_26472_cast_fp16)[name = tensor("op_26595_cast_fp16")]; + tensor var_26602_begin_0 = const()[name = tensor("op_26602_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26602_end_0 = const()[name = tensor("op_26602_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26602_end_mask_0 = const()[name = tensor("op_26602_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26602_cast_fp16 = slice_by_index(begin = var_26602_begin_0, end = var_26602_end_0, end_mask = var_26602_end_mask_0, x = var_26472_cast_fp16)[name = tensor("op_26602_cast_fp16")]; + tensor var_26609_begin_0 = const()[name = tensor("op_26609_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26609_end_0 = const()[name = tensor("op_26609_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26609_end_mask_0 = const()[name = tensor("op_26609_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26609_cast_fp16 = slice_by_index(begin = var_26609_begin_0, end = var_26609_end_0, end_mask = var_26609_end_mask_0, x = var_26476_cast_fp16)[name = tensor("op_26609_cast_fp16")]; + tensor var_26616_begin_0 = const()[name = tensor("op_26616_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26616_end_0 = const()[name = tensor("op_26616_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26616_end_mask_0 = const()[name = tensor("op_26616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26616_cast_fp16 = slice_by_index(begin = var_26616_begin_0, end = var_26616_end_0, end_mask = var_26616_end_mask_0, x = var_26476_cast_fp16)[name = tensor("op_26616_cast_fp16")]; + tensor var_26623_begin_0 = const()[name = tensor("op_26623_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26623_end_0 = const()[name = tensor("op_26623_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26623_end_mask_0 = const()[name = tensor("op_26623_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26623_cast_fp16 = slice_by_index(begin = var_26623_begin_0, end = var_26623_end_0, end_mask = var_26623_end_mask_0, x = var_26476_cast_fp16)[name = tensor("op_26623_cast_fp16")]; + tensor var_26630_begin_0 = const()[name = tensor("op_26630_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26630_end_0 = const()[name = tensor("op_26630_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26630_end_mask_0 = const()[name = tensor("op_26630_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26630_cast_fp16 = slice_by_index(begin = var_26630_begin_0, end = var_26630_end_0, end_mask = var_26630_end_mask_0, x = var_26476_cast_fp16)[name = tensor("op_26630_cast_fp16")]; + tensor var_26637_begin_0 = const()[name = tensor("op_26637_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26637_end_0 = const()[name = tensor("op_26637_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26637_end_mask_0 = const()[name = tensor("op_26637_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26637_cast_fp16 = slice_by_index(begin = var_26637_begin_0, end = var_26637_end_0, end_mask = var_26637_end_mask_0, x = var_26480_cast_fp16)[name = tensor("op_26637_cast_fp16")]; + tensor var_26644_begin_0 = const()[name = tensor("op_26644_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26644_end_0 = const()[name = tensor("op_26644_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26644_end_mask_0 = const()[name = tensor("op_26644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26644_cast_fp16 = slice_by_index(begin = var_26644_begin_0, end = var_26644_end_0, end_mask = var_26644_end_mask_0, x = var_26480_cast_fp16)[name = tensor("op_26644_cast_fp16")]; + tensor var_26651_begin_0 = const()[name = tensor("op_26651_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26651_end_0 = const()[name = tensor("op_26651_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26651_end_mask_0 = const()[name = tensor("op_26651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26651_cast_fp16 = slice_by_index(begin = var_26651_begin_0, end = var_26651_end_0, end_mask = var_26651_end_mask_0, x = var_26480_cast_fp16)[name = tensor("op_26651_cast_fp16")]; + tensor var_26658_begin_0 = const()[name = tensor("op_26658_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26658_end_0 = const()[name = tensor("op_26658_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26658_end_mask_0 = const()[name = tensor("op_26658_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26658_cast_fp16 = slice_by_index(begin = var_26658_begin_0, end = var_26658_end_0, end_mask = var_26658_end_mask_0, x = var_26480_cast_fp16)[name = tensor("op_26658_cast_fp16")]; + tensor var_26665_begin_0 = const()[name = tensor("op_26665_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26665_end_0 = const()[name = tensor("op_26665_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26665_end_mask_0 = const()[name = tensor("op_26665_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26665_cast_fp16 = slice_by_index(begin = var_26665_begin_0, end = var_26665_end_0, end_mask = var_26665_end_mask_0, x = var_26484_cast_fp16)[name = tensor("op_26665_cast_fp16")]; + tensor var_26672_begin_0 = const()[name = tensor("op_26672_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26672_end_0 = const()[name = tensor("op_26672_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26672_end_mask_0 = const()[name = tensor("op_26672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26672_cast_fp16 = slice_by_index(begin = var_26672_begin_0, end = var_26672_end_0, end_mask = var_26672_end_mask_0, x = var_26484_cast_fp16)[name = tensor("op_26672_cast_fp16")]; + tensor var_26679_begin_0 = const()[name = tensor("op_26679_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26679_end_0 = const()[name = tensor("op_26679_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26679_end_mask_0 = const()[name = tensor("op_26679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26679_cast_fp16 = slice_by_index(begin = var_26679_begin_0, end = var_26679_end_0, end_mask = var_26679_end_mask_0, x = var_26484_cast_fp16)[name = tensor("op_26679_cast_fp16")]; + tensor var_26686_begin_0 = const()[name = tensor("op_26686_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26686_end_0 = const()[name = tensor("op_26686_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26686_end_mask_0 = const()[name = tensor("op_26686_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26686_cast_fp16 = slice_by_index(begin = var_26686_begin_0, end = var_26686_end_0, end_mask = var_26686_end_mask_0, x = var_26484_cast_fp16)[name = tensor("op_26686_cast_fp16")]; + tensor var_26693_begin_0 = const()[name = tensor("op_26693_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26693_end_0 = const()[name = tensor("op_26693_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26693_end_mask_0 = const()[name = tensor("op_26693_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26693_cast_fp16 = slice_by_index(begin = var_26693_begin_0, end = var_26693_end_0, end_mask = var_26693_end_mask_0, x = var_26488_cast_fp16)[name = tensor("op_26693_cast_fp16")]; + tensor var_26700_begin_0 = const()[name = tensor("op_26700_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26700_end_0 = const()[name = tensor("op_26700_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26700_end_mask_0 = const()[name = tensor("op_26700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26700_cast_fp16 = slice_by_index(begin = var_26700_begin_0, end = var_26700_end_0, end_mask = var_26700_end_mask_0, x = var_26488_cast_fp16)[name = tensor("op_26700_cast_fp16")]; + tensor var_26707_begin_0 = const()[name = tensor("op_26707_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26707_end_0 = const()[name = tensor("op_26707_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26707_end_mask_0 = const()[name = tensor("op_26707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26707_cast_fp16 = slice_by_index(begin = var_26707_begin_0, end = var_26707_end_0, end_mask = var_26707_end_mask_0, x = var_26488_cast_fp16)[name = tensor("op_26707_cast_fp16")]; + tensor var_26714_begin_0 = const()[name = tensor("op_26714_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26714_end_0 = const()[name = tensor("op_26714_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26714_end_mask_0 = const()[name = tensor("op_26714_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26714_cast_fp16 = slice_by_index(begin = var_26714_begin_0, end = var_26714_end_0, end_mask = var_26714_end_mask_0, x = var_26488_cast_fp16)[name = tensor("op_26714_cast_fp16")]; + tensor var_26721_begin_0 = const()[name = tensor("op_26721_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26721_end_0 = const()[name = tensor("op_26721_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26721_end_mask_0 = const()[name = tensor("op_26721_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26721_cast_fp16 = slice_by_index(begin = var_26721_begin_0, end = var_26721_end_0, end_mask = var_26721_end_mask_0, x = var_26492_cast_fp16)[name = tensor("op_26721_cast_fp16")]; + tensor var_26728_begin_0 = const()[name = tensor("op_26728_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26728_end_0 = const()[name = tensor("op_26728_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26728_end_mask_0 = const()[name = tensor("op_26728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26728_cast_fp16 = slice_by_index(begin = var_26728_begin_0, end = var_26728_end_0, end_mask = var_26728_end_mask_0, x = var_26492_cast_fp16)[name = tensor("op_26728_cast_fp16")]; + tensor var_26735_begin_0 = const()[name = tensor("op_26735_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26735_end_0 = const()[name = tensor("op_26735_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26735_end_mask_0 = const()[name = tensor("op_26735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26735_cast_fp16 = slice_by_index(begin = var_26735_begin_0, end = var_26735_end_0, end_mask = var_26735_end_mask_0, x = var_26492_cast_fp16)[name = tensor("op_26735_cast_fp16")]; + tensor var_26742_begin_0 = const()[name = tensor("op_26742_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26742_end_0 = const()[name = tensor("op_26742_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26742_end_mask_0 = const()[name = tensor("op_26742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26742_cast_fp16 = slice_by_index(begin = var_26742_begin_0, end = var_26742_end_0, end_mask = var_26742_end_mask_0, x = var_26492_cast_fp16)[name = tensor("op_26742_cast_fp16")]; + tensor var_26749_begin_0 = const()[name = tensor("op_26749_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26749_end_0 = const()[name = tensor("op_26749_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26749_end_mask_0 = const()[name = tensor("op_26749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26749_cast_fp16 = slice_by_index(begin = var_26749_begin_0, end = var_26749_end_0, end_mask = var_26749_end_mask_0, x = var_26496_cast_fp16)[name = tensor("op_26749_cast_fp16")]; + tensor var_26756_begin_0 = const()[name = tensor("op_26756_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26756_end_0 = const()[name = tensor("op_26756_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26756_end_mask_0 = const()[name = tensor("op_26756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26756_cast_fp16 = slice_by_index(begin = var_26756_begin_0, end = var_26756_end_0, end_mask = var_26756_end_mask_0, x = var_26496_cast_fp16)[name = tensor("op_26756_cast_fp16")]; + tensor var_26763_begin_0 = const()[name = tensor("op_26763_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26763_end_0 = const()[name = tensor("op_26763_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26763_end_mask_0 = const()[name = tensor("op_26763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26763_cast_fp16 = slice_by_index(begin = var_26763_begin_0, end = var_26763_end_0, end_mask = var_26763_end_mask_0, x = var_26496_cast_fp16)[name = tensor("op_26763_cast_fp16")]; + tensor var_26770_begin_0 = const()[name = tensor("op_26770_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26770_end_0 = const()[name = tensor("op_26770_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26770_end_mask_0 = const()[name = tensor("op_26770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26770_cast_fp16 = slice_by_index(begin = var_26770_begin_0, end = var_26770_end_0, end_mask = var_26770_end_mask_0, x = var_26496_cast_fp16)[name = tensor("op_26770_cast_fp16")]; + tensor var_26777_begin_0 = const()[name = tensor("op_26777_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26777_end_0 = const()[name = tensor("op_26777_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26777_end_mask_0 = const()[name = tensor("op_26777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26777_cast_fp16 = slice_by_index(begin = var_26777_begin_0, end = var_26777_end_0, end_mask = var_26777_end_mask_0, x = var_26500_cast_fp16)[name = tensor("op_26777_cast_fp16")]; + tensor var_26784_begin_0 = const()[name = tensor("op_26784_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26784_end_0 = const()[name = tensor("op_26784_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26784_end_mask_0 = const()[name = tensor("op_26784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26784_cast_fp16 = slice_by_index(begin = var_26784_begin_0, end = var_26784_end_0, end_mask = var_26784_end_mask_0, x = var_26500_cast_fp16)[name = tensor("op_26784_cast_fp16")]; + tensor var_26791_begin_0 = const()[name = tensor("op_26791_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26791_end_0 = const()[name = tensor("op_26791_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26791_end_mask_0 = const()[name = tensor("op_26791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26791_cast_fp16 = slice_by_index(begin = var_26791_begin_0, end = var_26791_end_0, end_mask = var_26791_end_mask_0, x = var_26500_cast_fp16)[name = tensor("op_26791_cast_fp16")]; + tensor var_26798_begin_0 = const()[name = tensor("op_26798_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26798_end_0 = const()[name = tensor("op_26798_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26798_end_mask_0 = const()[name = tensor("op_26798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26798_cast_fp16 = slice_by_index(begin = var_26798_begin_0, end = var_26798_end_0, end_mask = var_26798_end_mask_0, x = var_26500_cast_fp16)[name = tensor("op_26798_cast_fp16")]; + tensor var_26805_begin_0 = const()[name = tensor("op_26805_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26805_end_0 = const()[name = tensor("op_26805_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26805_end_mask_0 = const()[name = tensor("op_26805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26805_cast_fp16 = slice_by_index(begin = var_26805_begin_0, end = var_26805_end_0, end_mask = var_26805_end_mask_0, x = var_26504_cast_fp16)[name = tensor("op_26805_cast_fp16")]; + tensor var_26812_begin_0 = const()[name = tensor("op_26812_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26812_end_0 = const()[name = tensor("op_26812_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26812_end_mask_0 = const()[name = tensor("op_26812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26812_cast_fp16 = slice_by_index(begin = var_26812_begin_0, end = var_26812_end_0, end_mask = var_26812_end_mask_0, x = var_26504_cast_fp16)[name = tensor("op_26812_cast_fp16")]; + tensor var_26819_begin_0 = const()[name = tensor("op_26819_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26819_end_0 = const()[name = tensor("op_26819_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26819_end_mask_0 = const()[name = tensor("op_26819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26819_cast_fp16 = slice_by_index(begin = var_26819_begin_0, end = var_26819_end_0, end_mask = var_26819_end_mask_0, x = var_26504_cast_fp16)[name = tensor("op_26819_cast_fp16")]; + tensor var_26826_begin_0 = const()[name = tensor("op_26826_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26826_end_0 = const()[name = tensor("op_26826_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26826_end_mask_0 = const()[name = tensor("op_26826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26826_cast_fp16 = slice_by_index(begin = var_26826_begin_0, end = var_26826_end_0, end_mask = var_26826_end_mask_0, x = var_26504_cast_fp16)[name = tensor("op_26826_cast_fp16")]; + tensor var_26833_begin_0 = const()[name = tensor("op_26833_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26833_end_0 = const()[name = tensor("op_26833_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26833_end_mask_0 = const()[name = tensor("op_26833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26833_cast_fp16 = slice_by_index(begin = var_26833_begin_0, end = var_26833_end_0, end_mask = var_26833_end_mask_0, x = var_26508_cast_fp16)[name = tensor("op_26833_cast_fp16")]; + tensor var_26840_begin_0 = const()[name = tensor("op_26840_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26840_end_0 = const()[name = tensor("op_26840_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26840_end_mask_0 = const()[name = tensor("op_26840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26840_cast_fp16 = slice_by_index(begin = var_26840_begin_0, end = var_26840_end_0, end_mask = var_26840_end_mask_0, x = var_26508_cast_fp16)[name = tensor("op_26840_cast_fp16")]; + tensor var_26847_begin_0 = const()[name = tensor("op_26847_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26847_end_0 = const()[name = tensor("op_26847_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26847_end_mask_0 = const()[name = tensor("op_26847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26847_cast_fp16 = slice_by_index(begin = var_26847_begin_0, end = var_26847_end_0, end_mask = var_26847_end_mask_0, x = var_26508_cast_fp16)[name = tensor("op_26847_cast_fp16")]; + tensor var_26854_begin_0 = const()[name = tensor("op_26854_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26854_end_0 = const()[name = tensor("op_26854_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26854_end_mask_0 = const()[name = tensor("op_26854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26854_cast_fp16 = slice_by_index(begin = var_26854_begin_0, end = var_26854_end_0, end_mask = var_26854_end_mask_0, x = var_26508_cast_fp16)[name = tensor("op_26854_cast_fp16")]; + tensor var_26861_begin_0 = const()[name = tensor("op_26861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26861_end_0 = const()[name = tensor("op_26861_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26861_end_mask_0 = const()[name = tensor("op_26861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26861_cast_fp16 = slice_by_index(begin = var_26861_begin_0, end = var_26861_end_0, end_mask = var_26861_end_mask_0, x = var_26512_cast_fp16)[name = tensor("op_26861_cast_fp16")]; + tensor var_26868_begin_0 = const()[name = tensor("op_26868_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26868_end_0 = const()[name = tensor("op_26868_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26868_end_mask_0 = const()[name = tensor("op_26868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26868_cast_fp16 = slice_by_index(begin = var_26868_begin_0, end = var_26868_end_0, end_mask = var_26868_end_mask_0, x = var_26512_cast_fp16)[name = tensor("op_26868_cast_fp16")]; + tensor var_26875_begin_0 = const()[name = tensor("op_26875_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26875_end_0 = const()[name = tensor("op_26875_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26875_end_mask_0 = const()[name = tensor("op_26875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26875_cast_fp16 = slice_by_index(begin = var_26875_begin_0, end = var_26875_end_0, end_mask = var_26875_end_mask_0, x = var_26512_cast_fp16)[name = tensor("op_26875_cast_fp16")]; + tensor var_26882_begin_0 = const()[name = tensor("op_26882_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26882_end_0 = const()[name = tensor("op_26882_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26882_end_mask_0 = const()[name = tensor("op_26882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26882_cast_fp16 = slice_by_index(begin = var_26882_begin_0, end = var_26882_end_0, end_mask = var_26882_end_mask_0, x = var_26512_cast_fp16)[name = tensor("op_26882_cast_fp16")]; + tensor var_26889_begin_0 = const()[name = tensor("op_26889_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26889_end_0 = const()[name = tensor("op_26889_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26889_end_mask_0 = const()[name = tensor("op_26889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26889_cast_fp16 = slice_by_index(begin = var_26889_begin_0, end = var_26889_end_0, end_mask = var_26889_end_mask_0, x = var_26516_cast_fp16)[name = tensor("op_26889_cast_fp16")]; + tensor var_26896_begin_0 = const()[name = tensor("op_26896_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26896_end_0 = const()[name = tensor("op_26896_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26896_end_mask_0 = const()[name = tensor("op_26896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26896_cast_fp16 = slice_by_index(begin = var_26896_begin_0, end = var_26896_end_0, end_mask = var_26896_end_mask_0, x = var_26516_cast_fp16)[name = tensor("op_26896_cast_fp16")]; + tensor var_26903_begin_0 = const()[name = tensor("op_26903_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26903_end_0 = const()[name = tensor("op_26903_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26903_end_mask_0 = const()[name = tensor("op_26903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26903_cast_fp16 = slice_by_index(begin = var_26903_begin_0, end = var_26903_end_0, end_mask = var_26903_end_mask_0, x = var_26516_cast_fp16)[name = tensor("op_26903_cast_fp16")]; + tensor var_26910_begin_0 = const()[name = tensor("op_26910_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26910_end_0 = const()[name = tensor("op_26910_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26910_end_mask_0 = const()[name = tensor("op_26910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26910_cast_fp16 = slice_by_index(begin = var_26910_begin_0, end = var_26910_end_0, end_mask = var_26910_end_mask_0, x = var_26516_cast_fp16)[name = tensor("op_26910_cast_fp16")]; + tensor var_26917_begin_0 = const()[name = tensor("op_26917_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26917_end_0 = const()[name = tensor("op_26917_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26917_end_mask_0 = const()[name = tensor("op_26917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26917_cast_fp16 = slice_by_index(begin = var_26917_begin_0, end = var_26917_end_0, end_mask = var_26917_end_mask_0, x = var_26520_cast_fp16)[name = tensor("op_26917_cast_fp16")]; + tensor var_26924_begin_0 = const()[name = tensor("op_26924_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26924_end_0 = const()[name = tensor("op_26924_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26924_end_mask_0 = const()[name = tensor("op_26924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26924_cast_fp16 = slice_by_index(begin = var_26924_begin_0, end = var_26924_end_0, end_mask = var_26924_end_mask_0, x = var_26520_cast_fp16)[name = tensor("op_26924_cast_fp16")]; + tensor var_26931_begin_0 = const()[name = tensor("op_26931_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26931_end_0 = const()[name = tensor("op_26931_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26931_end_mask_0 = const()[name = tensor("op_26931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26931_cast_fp16 = slice_by_index(begin = var_26931_begin_0, end = var_26931_end_0, end_mask = var_26931_end_mask_0, x = var_26520_cast_fp16)[name = tensor("op_26931_cast_fp16")]; + tensor var_26938_begin_0 = const()[name = tensor("op_26938_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26938_end_0 = const()[name = tensor("op_26938_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26938_end_mask_0 = const()[name = tensor("op_26938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26938_cast_fp16 = slice_by_index(begin = var_26938_begin_0, end = var_26938_end_0, end_mask = var_26938_end_mask_0, x = var_26520_cast_fp16)[name = tensor("op_26938_cast_fp16")]; + tensor var_26945_begin_0 = const()[name = tensor("op_26945_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26945_end_0 = const()[name = tensor("op_26945_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26945_end_mask_0 = const()[name = tensor("op_26945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26945_cast_fp16 = slice_by_index(begin = var_26945_begin_0, end = var_26945_end_0, end_mask = var_26945_end_mask_0, x = var_26524_cast_fp16)[name = tensor("op_26945_cast_fp16")]; + tensor var_26952_begin_0 = const()[name = tensor("op_26952_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26952_end_0 = const()[name = tensor("op_26952_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26952_end_mask_0 = const()[name = tensor("op_26952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26952_cast_fp16 = slice_by_index(begin = var_26952_begin_0, end = var_26952_end_0, end_mask = var_26952_end_mask_0, x = var_26524_cast_fp16)[name = tensor("op_26952_cast_fp16")]; + tensor var_26959_begin_0 = const()[name = tensor("op_26959_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26959_end_0 = const()[name = tensor("op_26959_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26959_end_mask_0 = const()[name = tensor("op_26959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26959_cast_fp16 = slice_by_index(begin = var_26959_begin_0, end = var_26959_end_0, end_mask = var_26959_end_mask_0, x = var_26524_cast_fp16)[name = tensor("op_26959_cast_fp16")]; + tensor var_26966_begin_0 = const()[name = tensor("op_26966_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26966_end_0 = const()[name = tensor("op_26966_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26966_end_mask_0 = const()[name = tensor("op_26966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26966_cast_fp16 = slice_by_index(begin = var_26966_begin_0, end = var_26966_end_0, end_mask = var_26966_end_mask_0, x = var_26524_cast_fp16)[name = tensor("op_26966_cast_fp16")]; + tensor var_26973_begin_0 = const()[name = tensor("op_26973_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_26973_end_0 = const()[name = tensor("op_26973_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_26973_end_mask_0 = const()[name = tensor("op_26973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26973_cast_fp16 = slice_by_index(begin = var_26973_begin_0, end = var_26973_end_0, end_mask = var_26973_end_mask_0, x = var_26528_cast_fp16)[name = tensor("op_26973_cast_fp16")]; + tensor var_26980_begin_0 = const()[name = tensor("op_26980_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_26980_end_0 = const()[name = tensor("op_26980_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_26980_end_mask_0 = const()[name = tensor("op_26980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26980_cast_fp16 = slice_by_index(begin = var_26980_begin_0, end = var_26980_end_0, end_mask = var_26980_end_mask_0, x = var_26528_cast_fp16)[name = tensor("op_26980_cast_fp16")]; + tensor var_26987_begin_0 = const()[name = tensor("op_26987_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_26987_end_0 = const()[name = tensor("op_26987_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_26987_end_mask_0 = const()[name = tensor("op_26987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26987_cast_fp16 = slice_by_index(begin = var_26987_begin_0, end = var_26987_end_0, end_mask = var_26987_end_mask_0, x = var_26528_cast_fp16)[name = tensor("op_26987_cast_fp16")]; + tensor var_26994_begin_0 = const()[name = tensor("op_26994_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_26994_end_0 = const()[name = tensor("op_26994_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_26994_end_mask_0 = const()[name = tensor("op_26994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_26994_cast_fp16 = slice_by_index(begin = var_26994_begin_0, end = var_26994_end_0, end_mask = var_26994_end_mask_0, x = var_26528_cast_fp16)[name = tensor("op_26994_cast_fp16")]; + tensor var_27001_begin_0 = const()[name = tensor("op_27001_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27001_end_0 = const()[name = tensor("op_27001_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27001_end_mask_0 = const()[name = tensor("op_27001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27001_cast_fp16 = slice_by_index(begin = var_27001_begin_0, end = var_27001_end_0, end_mask = var_27001_end_mask_0, x = var_26532_cast_fp16)[name = tensor("op_27001_cast_fp16")]; + tensor var_27008_begin_0 = const()[name = tensor("op_27008_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27008_end_0 = const()[name = tensor("op_27008_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27008_end_mask_0 = const()[name = tensor("op_27008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27008_cast_fp16 = slice_by_index(begin = var_27008_begin_0, end = var_27008_end_0, end_mask = var_27008_end_mask_0, x = var_26532_cast_fp16)[name = tensor("op_27008_cast_fp16")]; + tensor var_27015_begin_0 = const()[name = tensor("op_27015_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27015_end_0 = const()[name = tensor("op_27015_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27015_end_mask_0 = const()[name = tensor("op_27015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27015_cast_fp16 = slice_by_index(begin = var_27015_begin_0, end = var_27015_end_0, end_mask = var_27015_end_mask_0, x = var_26532_cast_fp16)[name = tensor("op_27015_cast_fp16")]; + tensor var_27022_begin_0 = const()[name = tensor("op_27022_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27022_end_0 = const()[name = tensor("op_27022_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27022_end_mask_0 = const()[name = tensor("op_27022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27022_cast_fp16 = slice_by_index(begin = var_27022_begin_0, end = var_27022_end_0, end_mask = var_27022_end_mask_0, x = var_26532_cast_fp16)[name = tensor("op_27022_cast_fp16")]; + tensor var_27029_begin_0 = const()[name = tensor("op_27029_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27029_end_0 = const()[name = tensor("op_27029_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27029_end_mask_0 = const()[name = tensor("op_27029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27029_cast_fp16 = slice_by_index(begin = var_27029_begin_0, end = var_27029_end_0, end_mask = var_27029_end_mask_0, x = var_26536_cast_fp16)[name = tensor("op_27029_cast_fp16")]; + tensor var_27036_begin_0 = const()[name = tensor("op_27036_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27036_end_0 = const()[name = tensor("op_27036_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27036_end_mask_0 = const()[name = tensor("op_27036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27036_cast_fp16 = slice_by_index(begin = var_27036_begin_0, end = var_27036_end_0, end_mask = var_27036_end_mask_0, x = var_26536_cast_fp16)[name = tensor("op_27036_cast_fp16")]; + tensor var_27043_begin_0 = const()[name = tensor("op_27043_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27043_end_0 = const()[name = tensor("op_27043_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27043_end_mask_0 = const()[name = tensor("op_27043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27043_cast_fp16 = slice_by_index(begin = var_27043_begin_0, end = var_27043_end_0, end_mask = var_27043_end_mask_0, x = var_26536_cast_fp16)[name = tensor("op_27043_cast_fp16")]; + tensor var_27050_begin_0 = const()[name = tensor("op_27050_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27050_end_0 = const()[name = tensor("op_27050_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27050_end_mask_0 = const()[name = tensor("op_27050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27050_cast_fp16 = slice_by_index(begin = var_27050_begin_0, end = var_27050_end_0, end_mask = var_27050_end_mask_0, x = var_26536_cast_fp16)[name = tensor("op_27050_cast_fp16")]; + tensor var_27057_begin_0 = const()[name = tensor("op_27057_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27057_end_0 = const()[name = tensor("op_27057_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27057_end_mask_0 = const()[name = tensor("op_27057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27057_cast_fp16 = slice_by_index(begin = var_27057_begin_0, end = var_27057_end_0, end_mask = var_27057_end_mask_0, x = var_26540_cast_fp16)[name = tensor("op_27057_cast_fp16")]; + tensor var_27064_begin_0 = const()[name = tensor("op_27064_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27064_end_0 = const()[name = tensor("op_27064_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27064_end_mask_0 = const()[name = tensor("op_27064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27064_cast_fp16 = slice_by_index(begin = var_27064_begin_0, end = var_27064_end_0, end_mask = var_27064_end_mask_0, x = var_26540_cast_fp16)[name = tensor("op_27064_cast_fp16")]; + tensor var_27071_begin_0 = const()[name = tensor("op_27071_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27071_end_0 = const()[name = tensor("op_27071_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27071_end_mask_0 = const()[name = tensor("op_27071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27071_cast_fp16 = slice_by_index(begin = var_27071_begin_0, end = var_27071_end_0, end_mask = var_27071_end_mask_0, x = var_26540_cast_fp16)[name = tensor("op_27071_cast_fp16")]; + tensor var_27078_begin_0 = const()[name = tensor("op_27078_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27078_end_0 = const()[name = tensor("op_27078_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27078_end_mask_0 = const()[name = tensor("op_27078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27078_cast_fp16 = slice_by_index(begin = var_27078_begin_0, end = var_27078_end_0, end_mask = var_27078_end_mask_0, x = var_26540_cast_fp16)[name = tensor("op_27078_cast_fp16")]; + tensor var_27085_begin_0 = const()[name = tensor("op_27085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27085_end_0 = const()[name = tensor("op_27085_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_27085_end_mask_0 = const()[name = tensor("op_27085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27085_cast_fp16 = slice_by_index(begin = var_27085_begin_0, end = var_27085_end_0, end_mask = var_27085_end_mask_0, x = var_26544_cast_fp16)[name = tensor("op_27085_cast_fp16")]; + tensor var_27092_begin_0 = const()[name = tensor("op_27092_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_27092_end_0 = const()[name = tensor("op_27092_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_27092_end_mask_0 = const()[name = tensor("op_27092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27092_cast_fp16 = slice_by_index(begin = var_27092_begin_0, end = var_27092_end_0, end_mask = var_27092_end_mask_0, x = var_26544_cast_fp16)[name = tensor("op_27092_cast_fp16")]; + tensor var_27099_begin_0 = const()[name = tensor("op_27099_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_27099_end_0 = const()[name = tensor("op_27099_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_27099_end_mask_0 = const()[name = tensor("op_27099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27099_cast_fp16 = slice_by_index(begin = var_27099_begin_0, end = var_27099_end_0, end_mask = var_27099_end_mask_0, x = var_26544_cast_fp16)[name = tensor("op_27099_cast_fp16")]; + tensor var_27106_begin_0 = const()[name = tensor("op_27106_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_27106_end_0 = const()[name = tensor("op_27106_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27106_end_mask_0 = const()[name = tensor("op_27106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27106_cast_fp16 = slice_by_index(begin = var_27106_begin_0, end = var_27106_end_0, end_mask = var_27106_end_mask_0, x = var_26544_cast_fp16)[name = tensor("op_27106_cast_fp16")]; + tensor k_33_perm_0 = const()[name = tensor("k_33_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_27111_begin_0 = const()[name = tensor("op_27111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27111_end_0 = const()[name = tensor("op_27111_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_27111_end_mask_0 = const()[name = tensor("op_27111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_15 = transpose(perm = k_33_perm_0, x = key_33_cast_fp16)[name = tensor("transpose_15")]; + tensor var_27111_cast_fp16 = slice_by_index(begin = var_27111_begin_0, end = var_27111_end_0, end_mask = var_27111_end_mask_0, x = transpose_15)[name = tensor("op_27111_cast_fp16")]; + tensor var_27115_begin_0 = const()[name = tensor("op_27115_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_27115_end_0 = const()[name = tensor("op_27115_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_27115_end_mask_0 = const()[name = tensor("op_27115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27115_cast_fp16 = slice_by_index(begin = var_27115_begin_0, end = var_27115_end_0, end_mask = var_27115_end_mask_0, x = transpose_15)[name = tensor("op_27115_cast_fp16")]; + tensor var_27119_begin_0 = const()[name = tensor("op_27119_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_27119_end_0 = const()[name = tensor("op_27119_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_27119_end_mask_0 = const()[name = tensor("op_27119_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27119_cast_fp16 = slice_by_index(begin = var_27119_begin_0, end = var_27119_end_0, end_mask = var_27119_end_mask_0, x = transpose_15)[name = tensor("op_27119_cast_fp16")]; + tensor var_27123_begin_0 = const()[name = tensor("op_27123_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_27123_end_0 = const()[name = tensor("op_27123_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_27123_end_mask_0 = const()[name = tensor("op_27123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27123_cast_fp16 = slice_by_index(begin = var_27123_begin_0, end = var_27123_end_0, end_mask = var_27123_end_mask_0, x = transpose_15)[name = tensor("op_27123_cast_fp16")]; + tensor var_27127_begin_0 = const()[name = tensor("op_27127_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_27127_end_0 = const()[name = tensor("op_27127_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_27127_end_mask_0 = const()[name = tensor("op_27127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27127_cast_fp16 = slice_by_index(begin = var_27127_begin_0, end = var_27127_end_0, end_mask = var_27127_end_mask_0, x = transpose_15)[name = tensor("op_27127_cast_fp16")]; + tensor var_27131_begin_0 = const()[name = tensor("op_27131_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_27131_end_0 = const()[name = tensor("op_27131_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_27131_end_mask_0 = const()[name = tensor("op_27131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27131_cast_fp16 = slice_by_index(begin = var_27131_begin_0, end = var_27131_end_0, end_mask = var_27131_end_mask_0, x = transpose_15)[name = tensor("op_27131_cast_fp16")]; + tensor var_27135_begin_0 = const()[name = tensor("op_27135_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_27135_end_0 = const()[name = tensor("op_27135_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_27135_end_mask_0 = const()[name = tensor("op_27135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27135_cast_fp16 = slice_by_index(begin = var_27135_begin_0, end = var_27135_end_0, end_mask = var_27135_end_mask_0, x = transpose_15)[name = tensor("op_27135_cast_fp16")]; + tensor var_27139_begin_0 = const()[name = tensor("op_27139_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_27139_end_0 = const()[name = tensor("op_27139_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_27139_end_mask_0 = const()[name = tensor("op_27139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27139_cast_fp16 = slice_by_index(begin = var_27139_begin_0, end = var_27139_end_0, end_mask = var_27139_end_mask_0, x = transpose_15)[name = tensor("op_27139_cast_fp16")]; + tensor var_27143_begin_0 = const()[name = tensor("op_27143_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_27143_end_0 = const()[name = tensor("op_27143_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_27143_end_mask_0 = const()[name = tensor("op_27143_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27143_cast_fp16 = slice_by_index(begin = var_27143_begin_0, end = var_27143_end_0, end_mask = var_27143_end_mask_0, x = transpose_15)[name = tensor("op_27143_cast_fp16")]; + tensor var_27147_begin_0 = const()[name = tensor("op_27147_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_27147_end_0 = const()[name = tensor("op_27147_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_27147_end_mask_0 = const()[name = tensor("op_27147_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27147_cast_fp16 = slice_by_index(begin = var_27147_begin_0, end = var_27147_end_0, end_mask = var_27147_end_mask_0, x = transpose_15)[name = tensor("op_27147_cast_fp16")]; + tensor var_27151_begin_0 = const()[name = tensor("op_27151_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_27151_end_0 = const()[name = tensor("op_27151_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_27151_end_mask_0 = const()[name = tensor("op_27151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27151_cast_fp16 = slice_by_index(begin = var_27151_begin_0, end = var_27151_end_0, end_mask = var_27151_end_mask_0, x = transpose_15)[name = tensor("op_27151_cast_fp16")]; + tensor var_27155_begin_0 = const()[name = tensor("op_27155_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_27155_end_0 = const()[name = tensor("op_27155_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_27155_end_mask_0 = const()[name = tensor("op_27155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27155_cast_fp16 = slice_by_index(begin = var_27155_begin_0, end = var_27155_end_0, end_mask = var_27155_end_mask_0, x = transpose_15)[name = tensor("op_27155_cast_fp16")]; + tensor var_27159_begin_0 = const()[name = tensor("op_27159_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_27159_end_0 = const()[name = tensor("op_27159_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_27159_end_mask_0 = const()[name = tensor("op_27159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27159_cast_fp16 = slice_by_index(begin = var_27159_begin_0, end = var_27159_end_0, end_mask = var_27159_end_mask_0, x = transpose_15)[name = tensor("op_27159_cast_fp16")]; + tensor var_27163_begin_0 = const()[name = tensor("op_27163_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_27163_end_0 = const()[name = tensor("op_27163_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_27163_end_mask_0 = const()[name = tensor("op_27163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27163_cast_fp16 = slice_by_index(begin = var_27163_begin_0, end = var_27163_end_0, end_mask = var_27163_end_mask_0, x = transpose_15)[name = tensor("op_27163_cast_fp16")]; + tensor var_27167_begin_0 = const()[name = tensor("op_27167_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_27167_end_0 = const()[name = tensor("op_27167_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_27167_end_mask_0 = const()[name = tensor("op_27167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27167_cast_fp16 = slice_by_index(begin = var_27167_begin_0, end = var_27167_end_0, end_mask = var_27167_end_mask_0, x = transpose_15)[name = tensor("op_27167_cast_fp16")]; + tensor var_27171_begin_0 = const()[name = tensor("op_27171_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_27171_end_0 = const()[name = tensor("op_27171_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_27171_end_mask_0 = const()[name = tensor("op_27171_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27171_cast_fp16 = slice_by_index(begin = var_27171_begin_0, end = var_27171_end_0, end_mask = var_27171_end_mask_0, x = transpose_15)[name = tensor("op_27171_cast_fp16")]; + tensor var_27175_begin_0 = const()[name = tensor("op_27175_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_27175_end_0 = const()[name = tensor("op_27175_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_27175_end_mask_0 = const()[name = tensor("op_27175_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27175_cast_fp16 = slice_by_index(begin = var_27175_begin_0, end = var_27175_end_0, end_mask = var_27175_end_mask_0, x = transpose_15)[name = tensor("op_27175_cast_fp16")]; + tensor var_27179_begin_0 = const()[name = tensor("op_27179_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_27179_end_0 = const()[name = tensor("op_27179_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_27179_end_mask_0 = const()[name = tensor("op_27179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27179_cast_fp16 = slice_by_index(begin = var_27179_begin_0, end = var_27179_end_0, end_mask = var_27179_end_mask_0, x = transpose_15)[name = tensor("op_27179_cast_fp16")]; + tensor var_27183_begin_0 = const()[name = tensor("op_27183_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_27183_end_0 = const()[name = tensor("op_27183_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_27183_end_mask_0 = const()[name = tensor("op_27183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27183_cast_fp16 = slice_by_index(begin = var_27183_begin_0, end = var_27183_end_0, end_mask = var_27183_end_mask_0, x = transpose_15)[name = tensor("op_27183_cast_fp16")]; + tensor var_27187_begin_0 = const()[name = tensor("op_27187_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_27187_end_0 = const()[name = tensor("op_27187_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_27187_end_mask_0 = const()[name = tensor("op_27187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_27187_cast_fp16 = slice_by_index(begin = var_27187_begin_0, end = var_27187_end_0, end_mask = var_27187_end_mask_0, x = transpose_15)[name = tensor("op_27187_cast_fp16")]; + tensor var_27189_begin_0 = const()[name = tensor("op_27189_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_27189_end_0 = const()[name = tensor("op_27189_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_27189_end_mask_0 = const()[name = tensor("op_27189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27189_cast_fp16 = slice_by_index(begin = var_27189_begin_0, end = var_27189_end_0, end_mask = var_27189_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27189_cast_fp16")]; + tensor var_27193_begin_0 = const()[name = tensor("op_27193_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_27193_end_0 = const()[name = tensor("op_27193_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_27193_end_mask_0 = const()[name = tensor("op_27193_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27193_cast_fp16 = slice_by_index(begin = var_27193_begin_0, end = var_27193_end_0, end_mask = var_27193_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27193_cast_fp16")]; + tensor var_27197_begin_0 = const()[name = tensor("op_27197_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_27197_end_0 = const()[name = tensor("op_27197_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_27197_end_mask_0 = const()[name = tensor("op_27197_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27197_cast_fp16 = slice_by_index(begin = var_27197_begin_0, end = var_27197_end_0, end_mask = var_27197_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27197_cast_fp16")]; + tensor var_27201_begin_0 = const()[name = tensor("op_27201_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_27201_end_0 = const()[name = tensor("op_27201_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_27201_end_mask_0 = const()[name = tensor("op_27201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27201_cast_fp16 = slice_by_index(begin = var_27201_begin_0, end = var_27201_end_0, end_mask = var_27201_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27201_cast_fp16")]; + tensor var_27205_begin_0 = const()[name = tensor("op_27205_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_27205_end_0 = const()[name = tensor("op_27205_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_27205_end_mask_0 = const()[name = tensor("op_27205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27205_cast_fp16 = slice_by_index(begin = var_27205_begin_0, end = var_27205_end_0, end_mask = var_27205_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27205_cast_fp16")]; + tensor var_27209_begin_0 = const()[name = tensor("op_27209_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_27209_end_0 = const()[name = tensor("op_27209_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_27209_end_mask_0 = const()[name = tensor("op_27209_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27209_cast_fp16 = slice_by_index(begin = var_27209_begin_0, end = var_27209_end_0, end_mask = var_27209_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27209_cast_fp16")]; + tensor var_27213_begin_0 = const()[name = tensor("op_27213_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_27213_end_0 = const()[name = tensor("op_27213_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_27213_end_mask_0 = const()[name = tensor("op_27213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27213_cast_fp16 = slice_by_index(begin = var_27213_begin_0, end = var_27213_end_0, end_mask = var_27213_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27213_cast_fp16")]; + tensor var_27217_begin_0 = const()[name = tensor("op_27217_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_27217_end_0 = const()[name = tensor("op_27217_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_27217_end_mask_0 = const()[name = tensor("op_27217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27217_cast_fp16 = slice_by_index(begin = var_27217_begin_0, end = var_27217_end_0, end_mask = var_27217_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27217_cast_fp16")]; + tensor var_27221_begin_0 = const()[name = tensor("op_27221_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_27221_end_0 = const()[name = tensor("op_27221_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_27221_end_mask_0 = const()[name = tensor("op_27221_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27221_cast_fp16 = slice_by_index(begin = var_27221_begin_0, end = var_27221_end_0, end_mask = var_27221_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27221_cast_fp16")]; + tensor var_27225_begin_0 = const()[name = tensor("op_27225_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_27225_end_0 = const()[name = tensor("op_27225_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_27225_end_mask_0 = const()[name = tensor("op_27225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27225_cast_fp16 = slice_by_index(begin = var_27225_begin_0, end = var_27225_end_0, end_mask = var_27225_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27225_cast_fp16")]; + tensor var_27229_begin_0 = const()[name = tensor("op_27229_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_27229_end_0 = const()[name = tensor("op_27229_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_27229_end_mask_0 = const()[name = tensor("op_27229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27229_cast_fp16 = slice_by_index(begin = var_27229_begin_0, end = var_27229_end_0, end_mask = var_27229_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27229_cast_fp16")]; + tensor var_27233_begin_0 = const()[name = tensor("op_27233_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_27233_end_0 = const()[name = tensor("op_27233_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_27233_end_mask_0 = const()[name = tensor("op_27233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27233_cast_fp16 = slice_by_index(begin = var_27233_begin_0, end = var_27233_end_0, end_mask = var_27233_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27233_cast_fp16")]; + tensor var_27237_begin_0 = const()[name = tensor("op_27237_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_27237_end_0 = const()[name = tensor("op_27237_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_27237_end_mask_0 = const()[name = tensor("op_27237_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27237_cast_fp16 = slice_by_index(begin = var_27237_begin_0, end = var_27237_end_0, end_mask = var_27237_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27237_cast_fp16")]; + tensor var_27241_begin_0 = const()[name = tensor("op_27241_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_27241_end_0 = const()[name = tensor("op_27241_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_27241_end_mask_0 = const()[name = tensor("op_27241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27241_cast_fp16 = slice_by_index(begin = var_27241_begin_0, end = var_27241_end_0, end_mask = var_27241_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27241_cast_fp16")]; + tensor var_27245_begin_0 = const()[name = tensor("op_27245_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_27245_end_0 = const()[name = tensor("op_27245_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_27245_end_mask_0 = const()[name = tensor("op_27245_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27245_cast_fp16 = slice_by_index(begin = var_27245_begin_0, end = var_27245_end_0, end_mask = var_27245_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27245_cast_fp16")]; + tensor var_27249_begin_0 = const()[name = tensor("op_27249_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_27249_end_0 = const()[name = tensor("op_27249_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_27249_end_mask_0 = const()[name = tensor("op_27249_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27249_cast_fp16 = slice_by_index(begin = var_27249_begin_0, end = var_27249_end_0, end_mask = var_27249_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27249_cast_fp16")]; + tensor var_27253_begin_0 = const()[name = tensor("op_27253_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_27253_end_0 = const()[name = tensor("op_27253_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_27253_end_mask_0 = const()[name = tensor("op_27253_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27253_cast_fp16 = slice_by_index(begin = var_27253_begin_0, end = var_27253_end_0, end_mask = var_27253_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27253_cast_fp16")]; + tensor var_27257_begin_0 = const()[name = tensor("op_27257_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_27257_end_0 = const()[name = tensor("op_27257_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_27257_end_mask_0 = const()[name = tensor("op_27257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27257_cast_fp16 = slice_by_index(begin = var_27257_begin_0, end = var_27257_end_0, end_mask = var_27257_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27257_cast_fp16")]; + tensor var_27261_begin_0 = const()[name = tensor("op_27261_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_27261_end_0 = const()[name = tensor("op_27261_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_27261_end_mask_0 = const()[name = tensor("op_27261_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27261_cast_fp16 = slice_by_index(begin = var_27261_begin_0, end = var_27261_end_0, end_mask = var_27261_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27261_cast_fp16")]; + tensor var_27265_begin_0 = const()[name = tensor("op_27265_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_27265_end_0 = const()[name = tensor("op_27265_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_27265_end_mask_0 = const()[name = tensor("op_27265_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_27265_cast_fp16 = slice_by_index(begin = var_27265_begin_0, end = var_27265_end_0, end_mask = var_27265_end_mask_0, x = value_33_cast_fp16)[name = tensor("op_27265_cast_fp16")]; + tensor var_27269_equation_0 = const()[name = tensor("op_27269_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27269_cast_fp16 = einsum(equation = var_27269_equation_0, values = (var_27111_cast_fp16, var_26553_cast_fp16))[name = tensor("op_27269_cast_fp16")]; + tensor var_27270_to_fp16 = const()[name = tensor("op_27270_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2561_cast_fp16 = mul(x = var_27269_cast_fp16, y = var_27270_to_fp16)[name = tensor("aw_chunk_2561_cast_fp16")]; + tensor var_27273_equation_0 = const()[name = tensor("op_27273_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27273_cast_fp16 = einsum(equation = var_27273_equation_0, values = (var_27111_cast_fp16, var_26560_cast_fp16))[name = tensor("op_27273_cast_fp16")]; + tensor var_27274_to_fp16 = const()[name = tensor("op_27274_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2563_cast_fp16 = mul(x = var_27273_cast_fp16, y = var_27274_to_fp16)[name = tensor("aw_chunk_2563_cast_fp16")]; + tensor var_27277_equation_0 = const()[name = tensor("op_27277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27277_cast_fp16 = einsum(equation = var_27277_equation_0, values = (var_27111_cast_fp16, var_26567_cast_fp16))[name = tensor("op_27277_cast_fp16")]; + tensor var_27278_to_fp16 = const()[name = tensor("op_27278_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2565_cast_fp16 = mul(x = var_27277_cast_fp16, y = var_27278_to_fp16)[name = tensor("aw_chunk_2565_cast_fp16")]; + tensor var_27281_equation_0 = const()[name = tensor("op_27281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27281_cast_fp16 = einsum(equation = var_27281_equation_0, values = (var_27111_cast_fp16, var_26574_cast_fp16))[name = tensor("op_27281_cast_fp16")]; + tensor var_27282_to_fp16 = const()[name = tensor("op_27282_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2567_cast_fp16 = mul(x = var_27281_cast_fp16, y = var_27282_to_fp16)[name = tensor("aw_chunk_2567_cast_fp16")]; + tensor var_27285_equation_0 = const()[name = tensor("op_27285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27285_cast_fp16 = einsum(equation = var_27285_equation_0, values = (var_27115_cast_fp16, var_26581_cast_fp16))[name = tensor("op_27285_cast_fp16")]; + tensor var_27286_to_fp16 = const()[name = tensor("op_27286_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2569_cast_fp16 = mul(x = var_27285_cast_fp16, y = var_27286_to_fp16)[name = tensor("aw_chunk_2569_cast_fp16")]; + tensor var_27289_equation_0 = const()[name = tensor("op_27289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27289_cast_fp16 = einsum(equation = var_27289_equation_0, values = (var_27115_cast_fp16, var_26588_cast_fp16))[name = tensor("op_27289_cast_fp16")]; + tensor var_27290_to_fp16 = const()[name = tensor("op_27290_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2571_cast_fp16 = mul(x = var_27289_cast_fp16, y = var_27290_to_fp16)[name = tensor("aw_chunk_2571_cast_fp16")]; + tensor var_27293_equation_0 = const()[name = tensor("op_27293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27293_cast_fp16 = einsum(equation = var_27293_equation_0, values = (var_27115_cast_fp16, var_26595_cast_fp16))[name = tensor("op_27293_cast_fp16")]; + tensor var_27294_to_fp16 = const()[name = tensor("op_27294_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2573_cast_fp16 = mul(x = var_27293_cast_fp16, y = var_27294_to_fp16)[name = tensor("aw_chunk_2573_cast_fp16")]; + tensor var_27297_equation_0 = const()[name = tensor("op_27297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27297_cast_fp16 = einsum(equation = var_27297_equation_0, values = (var_27115_cast_fp16, var_26602_cast_fp16))[name = tensor("op_27297_cast_fp16")]; + tensor var_27298_to_fp16 = const()[name = tensor("op_27298_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2575_cast_fp16 = mul(x = var_27297_cast_fp16, y = var_27298_to_fp16)[name = tensor("aw_chunk_2575_cast_fp16")]; + tensor var_27301_equation_0 = const()[name = tensor("op_27301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27301_cast_fp16 = einsum(equation = var_27301_equation_0, values = (var_27119_cast_fp16, var_26609_cast_fp16))[name = tensor("op_27301_cast_fp16")]; + tensor var_27302_to_fp16 = const()[name = tensor("op_27302_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2577_cast_fp16 = mul(x = var_27301_cast_fp16, y = var_27302_to_fp16)[name = tensor("aw_chunk_2577_cast_fp16")]; + tensor var_27305_equation_0 = const()[name = tensor("op_27305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27305_cast_fp16 = einsum(equation = var_27305_equation_0, values = (var_27119_cast_fp16, var_26616_cast_fp16))[name = tensor("op_27305_cast_fp16")]; + tensor var_27306_to_fp16 = const()[name = tensor("op_27306_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2579_cast_fp16 = mul(x = var_27305_cast_fp16, y = var_27306_to_fp16)[name = tensor("aw_chunk_2579_cast_fp16")]; + tensor var_27309_equation_0 = const()[name = tensor("op_27309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27309_cast_fp16 = einsum(equation = var_27309_equation_0, values = (var_27119_cast_fp16, var_26623_cast_fp16))[name = tensor("op_27309_cast_fp16")]; + tensor var_27310_to_fp16 = const()[name = tensor("op_27310_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2581_cast_fp16 = mul(x = var_27309_cast_fp16, y = var_27310_to_fp16)[name = tensor("aw_chunk_2581_cast_fp16")]; + tensor var_27313_equation_0 = const()[name = tensor("op_27313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27313_cast_fp16 = einsum(equation = var_27313_equation_0, values = (var_27119_cast_fp16, var_26630_cast_fp16))[name = tensor("op_27313_cast_fp16")]; + tensor var_27314_to_fp16 = const()[name = tensor("op_27314_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2583_cast_fp16 = mul(x = var_27313_cast_fp16, y = var_27314_to_fp16)[name = tensor("aw_chunk_2583_cast_fp16")]; + tensor var_27317_equation_0 = const()[name = tensor("op_27317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27317_cast_fp16 = einsum(equation = var_27317_equation_0, values = (var_27123_cast_fp16, var_26637_cast_fp16))[name = tensor("op_27317_cast_fp16")]; + tensor var_27318_to_fp16 = const()[name = tensor("op_27318_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2585_cast_fp16 = mul(x = var_27317_cast_fp16, y = var_27318_to_fp16)[name = tensor("aw_chunk_2585_cast_fp16")]; + tensor var_27321_equation_0 = const()[name = tensor("op_27321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27321_cast_fp16 = einsum(equation = var_27321_equation_0, values = (var_27123_cast_fp16, var_26644_cast_fp16))[name = tensor("op_27321_cast_fp16")]; + tensor var_27322_to_fp16 = const()[name = tensor("op_27322_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2587_cast_fp16 = mul(x = var_27321_cast_fp16, y = var_27322_to_fp16)[name = tensor("aw_chunk_2587_cast_fp16")]; + tensor var_27325_equation_0 = const()[name = tensor("op_27325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27325_cast_fp16 = einsum(equation = var_27325_equation_0, values = (var_27123_cast_fp16, var_26651_cast_fp16))[name = tensor("op_27325_cast_fp16")]; + tensor var_27326_to_fp16 = const()[name = tensor("op_27326_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2589_cast_fp16 = mul(x = var_27325_cast_fp16, y = var_27326_to_fp16)[name = tensor("aw_chunk_2589_cast_fp16")]; + tensor var_27329_equation_0 = const()[name = tensor("op_27329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27329_cast_fp16 = einsum(equation = var_27329_equation_0, values = (var_27123_cast_fp16, var_26658_cast_fp16))[name = tensor("op_27329_cast_fp16")]; + tensor var_27330_to_fp16 = const()[name = tensor("op_27330_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2591_cast_fp16 = mul(x = var_27329_cast_fp16, y = var_27330_to_fp16)[name = tensor("aw_chunk_2591_cast_fp16")]; + tensor var_27333_equation_0 = const()[name = tensor("op_27333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27333_cast_fp16 = einsum(equation = var_27333_equation_0, values = (var_27127_cast_fp16, var_26665_cast_fp16))[name = tensor("op_27333_cast_fp16")]; + tensor var_27334_to_fp16 = const()[name = tensor("op_27334_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2593_cast_fp16 = mul(x = var_27333_cast_fp16, y = var_27334_to_fp16)[name = tensor("aw_chunk_2593_cast_fp16")]; + tensor var_27337_equation_0 = const()[name = tensor("op_27337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27337_cast_fp16 = einsum(equation = var_27337_equation_0, values = (var_27127_cast_fp16, var_26672_cast_fp16))[name = tensor("op_27337_cast_fp16")]; + tensor var_27338_to_fp16 = const()[name = tensor("op_27338_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2595_cast_fp16 = mul(x = var_27337_cast_fp16, y = var_27338_to_fp16)[name = tensor("aw_chunk_2595_cast_fp16")]; + tensor var_27341_equation_0 = const()[name = tensor("op_27341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27341_cast_fp16 = einsum(equation = var_27341_equation_0, values = (var_27127_cast_fp16, var_26679_cast_fp16))[name = tensor("op_27341_cast_fp16")]; + tensor var_27342_to_fp16 = const()[name = tensor("op_27342_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2597_cast_fp16 = mul(x = var_27341_cast_fp16, y = var_27342_to_fp16)[name = tensor("aw_chunk_2597_cast_fp16")]; + tensor var_27345_equation_0 = const()[name = tensor("op_27345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27345_cast_fp16 = einsum(equation = var_27345_equation_0, values = (var_27127_cast_fp16, var_26686_cast_fp16))[name = tensor("op_27345_cast_fp16")]; + tensor var_27346_to_fp16 = const()[name = tensor("op_27346_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2599_cast_fp16 = mul(x = var_27345_cast_fp16, y = var_27346_to_fp16)[name = tensor("aw_chunk_2599_cast_fp16")]; + tensor var_27349_equation_0 = const()[name = tensor("op_27349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27349_cast_fp16 = einsum(equation = var_27349_equation_0, values = (var_27131_cast_fp16, var_26693_cast_fp16))[name = tensor("op_27349_cast_fp16")]; + tensor var_27350_to_fp16 = const()[name = tensor("op_27350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2601_cast_fp16 = mul(x = var_27349_cast_fp16, y = var_27350_to_fp16)[name = tensor("aw_chunk_2601_cast_fp16")]; + tensor var_27353_equation_0 = const()[name = tensor("op_27353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27353_cast_fp16 = einsum(equation = var_27353_equation_0, values = (var_27131_cast_fp16, var_26700_cast_fp16))[name = tensor("op_27353_cast_fp16")]; + tensor var_27354_to_fp16 = const()[name = tensor("op_27354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2603_cast_fp16 = mul(x = var_27353_cast_fp16, y = var_27354_to_fp16)[name = tensor("aw_chunk_2603_cast_fp16")]; + tensor var_27357_equation_0 = const()[name = tensor("op_27357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27357_cast_fp16 = einsum(equation = var_27357_equation_0, values = (var_27131_cast_fp16, var_26707_cast_fp16))[name = tensor("op_27357_cast_fp16")]; + tensor var_27358_to_fp16 = const()[name = tensor("op_27358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2605_cast_fp16 = mul(x = var_27357_cast_fp16, y = var_27358_to_fp16)[name = tensor("aw_chunk_2605_cast_fp16")]; + tensor var_27361_equation_0 = const()[name = tensor("op_27361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27361_cast_fp16 = einsum(equation = var_27361_equation_0, values = (var_27131_cast_fp16, var_26714_cast_fp16))[name = tensor("op_27361_cast_fp16")]; + tensor var_27362_to_fp16 = const()[name = tensor("op_27362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2607_cast_fp16 = mul(x = var_27361_cast_fp16, y = var_27362_to_fp16)[name = tensor("aw_chunk_2607_cast_fp16")]; + tensor var_27365_equation_0 = const()[name = tensor("op_27365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27365_cast_fp16 = einsum(equation = var_27365_equation_0, values = (var_27135_cast_fp16, var_26721_cast_fp16))[name = tensor("op_27365_cast_fp16")]; + tensor var_27366_to_fp16 = const()[name = tensor("op_27366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2609_cast_fp16 = mul(x = var_27365_cast_fp16, y = var_27366_to_fp16)[name = tensor("aw_chunk_2609_cast_fp16")]; + tensor var_27369_equation_0 = const()[name = tensor("op_27369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27369_cast_fp16 = einsum(equation = var_27369_equation_0, values = (var_27135_cast_fp16, var_26728_cast_fp16))[name = tensor("op_27369_cast_fp16")]; + tensor var_27370_to_fp16 = const()[name = tensor("op_27370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2611_cast_fp16 = mul(x = var_27369_cast_fp16, y = var_27370_to_fp16)[name = tensor("aw_chunk_2611_cast_fp16")]; + tensor var_27373_equation_0 = const()[name = tensor("op_27373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27373_cast_fp16 = einsum(equation = var_27373_equation_0, values = (var_27135_cast_fp16, var_26735_cast_fp16))[name = tensor("op_27373_cast_fp16")]; + tensor var_27374_to_fp16 = const()[name = tensor("op_27374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2613_cast_fp16 = mul(x = var_27373_cast_fp16, y = var_27374_to_fp16)[name = tensor("aw_chunk_2613_cast_fp16")]; + tensor var_27377_equation_0 = const()[name = tensor("op_27377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27377_cast_fp16 = einsum(equation = var_27377_equation_0, values = (var_27135_cast_fp16, var_26742_cast_fp16))[name = tensor("op_27377_cast_fp16")]; + tensor var_27378_to_fp16 = const()[name = tensor("op_27378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2615_cast_fp16 = mul(x = var_27377_cast_fp16, y = var_27378_to_fp16)[name = tensor("aw_chunk_2615_cast_fp16")]; + tensor var_27381_equation_0 = const()[name = tensor("op_27381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27381_cast_fp16 = einsum(equation = var_27381_equation_0, values = (var_27139_cast_fp16, var_26749_cast_fp16))[name = tensor("op_27381_cast_fp16")]; + tensor var_27382_to_fp16 = const()[name = tensor("op_27382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2617_cast_fp16 = mul(x = var_27381_cast_fp16, y = var_27382_to_fp16)[name = tensor("aw_chunk_2617_cast_fp16")]; + tensor var_27385_equation_0 = const()[name = tensor("op_27385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27385_cast_fp16 = einsum(equation = var_27385_equation_0, values = (var_27139_cast_fp16, var_26756_cast_fp16))[name = tensor("op_27385_cast_fp16")]; + tensor var_27386_to_fp16 = const()[name = tensor("op_27386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2619_cast_fp16 = mul(x = var_27385_cast_fp16, y = var_27386_to_fp16)[name = tensor("aw_chunk_2619_cast_fp16")]; + tensor var_27389_equation_0 = const()[name = tensor("op_27389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27389_cast_fp16 = einsum(equation = var_27389_equation_0, values = (var_27139_cast_fp16, var_26763_cast_fp16))[name = tensor("op_27389_cast_fp16")]; + tensor var_27390_to_fp16 = const()[name = tensor("op_27390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2621_cast_fp16 = mul(x = var_27389_cast_fp16, y = var_27390_to_fp16)[name = tensor("aw_chunk_2621_cast_fp16")]; + tensor var_27393_equation_0 = const()[name = tensor("op_27393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27393_cast_fp16 = einsum(equation = var_27393_equation_0, values = (var_27139_cast_fp16, var_26770_cast_fp16))[name = tensor("op_27393_cast_fp16")]; + tensor var_27394_to_fp16 = const()[name = tensor("op_27394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2623_cast_fp16 = mul(x = var_27393_cast_fp16, y = var_27394_to_fp16)[name = tensor("aw_chunk_2623_cast_fp16")]; + tensor var_27397_equation_0 = const()[name = tensor("op_27397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27397_cast_fp16 = einsum(equation = var_27397_equation_0, values = (var_27143_cast_fp16, var_26777_cast_fp16))[name = tensor("op_27397_cast_fp16")]; + tensor var_27398_to_fp16 = const()[name = tensor("op_27398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2625_cast_fp16 = mul(x = var_27397_cast_fp16, y = var_27398_to_fp16)[name = tensor("aw_chunk_2625_cast_fp16")]; + tensor var_27401_equation_0 = const()[name = tensor("op_27401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27401_cast_fp16 = einsum(equation = var_27401_equation_0, values = (var_27143_cast_fp16, var_26784_cast_fp16))[name = tensor("op_27401_cast_fp16")]; + tensor var_27402_to_fp16 = const()[name = tensor("op_27402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2627_cast_fp16 = mul(x = var_27401_cast_fp16, y = var_27402_to_fp16)[name = tensor("aw_chunk_2627_cast_fp16")]; + tensor var_27405_equation_0 = const()[name = tensor("op_27405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27405_cast_fp16 = einsum(equation = var_27405_equation_0, values = (var_27143_cast_fp16, var_26791_cast_fp16))[name = tensor("op_27405_cast_fp16")]; + tensor var_27406_to_fp16 = const()[name = tensor("op_27406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2629_cast_fp16 = mul(x = var_27405_cast_fp16, y = var_27406_to_fp16)[name = tensor("aw_chunk_2629_cast_fp16")]; + tensor var_27409_equation_0 = const()[name = tensor("op_27409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27409_cast_fp16 = einsum(equation = var_27409_equation_0, values = (var_27143_cast_fp16, var_26798_cast_fp16))[name = tensor("op_27409_cast_fp16")]; + tensor var_27410_to_fp16 = const()[name = tensor("op_27410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2631_cast_fp16 = mul(x = var_27409_cast_fp16, y = var_27410_to_fp16)[name = tensor("aw_chunk_2631_cast_fp16")]; + tensor var_27413_equation_0 = const()[name = tensor("op_27413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27413_cast_fp16 = einsum(equation = var_27413_equation_0, values = (var_27147_cast_fp16, var_26805_cast_fp16))[name = tensor("op_27413_cast_fp16")]; + tensor var_27414_to_fp16 = const()[name = tensor("op_27414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2633_cast_fp16 = mul(x = var_27413_cast_fp16, y = var_27414_to_fp16)[name = tensor("aw_chunk_2633_cast_fp16")]; + tensor var_27417_equation_0 = const()[name = tensor("op_27417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27417_cast_fp16 = einsum(equation = var_27417_equation_0, values = (var_27147_cast_fp16, var_26812_cast_fp16))[name = tensor("op_27417_cast_fp16")]; + tensor var_27418_to_fp16 = const()[name = tensor("op_27418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2635_cast_fp16 = mul(x = var_27417_cast_fp16, y = var_27418_to_fp16)[name = tensor("aw_chunk_2635_cast_fp16")]; + tensor var_27421_equation_0 = const()[name = tensor("op_27421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27421_cast_fp16 = einsum(equation = var_27421_equation_0, values = (var_27147_cast_fp16, var_26819_cast_fp16))[name = tensor("op_27421_cast_fp16")]; + tensor var_27422_to_fp16 = const()[name = tensor("op_27422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2637_cast_fp16 = mul(x = var_27421_cast_fp16, y = var_27422_to_fp16)[name = tensor("aw_chunk_2637_cast_fp16")]; + tensor var_27425_equation_0 = const()[name = tensor("op_27425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27425_cast_fp16 = einsum(equation = var_27425_equation_0, values = (var_27147_cast_fp16, var_26826_cast_fp16))[name = tensor("op_27425_cast_fp16")]; + tensor var_27426_to_fp16 = const()[name = tensor("op_27426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2639_cast_fp16 = mul(x = var_27425_cast_fp16, y = var_27426_to_fp16)[name = tensor("aw_chunk_2639_cast_fp16")]; + tensor var_27429_equation_0 = const()[name = tensor("op_27429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27429_cast_fp16 = einsum(equation = var_27429_equation_0, values = (var_27151_cast_fp16, var_26833_cast_fp16))[name = tensor("op_27429_cast_fp16")]; + tensor var_27430_to_fp16 = const()[name = tensor("op_27430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2641_cast_fp16 = mul(x = var_27429_cast_fp16, y = var_27430_to_fp16)[name = tensor("aw_chunk_2641_cast_fp16")]; + tensor var_27433_equation_0 = const()[name = tensor("op_27433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27433_cast_fp16 = einsum(equation = var_27433_equation_0, values = (var_27151_cast_fp16, var_26840_cast_fp16))[name = tensor("op_27433_cast_fp16")]; + tensor var_27434_to_fp16 = const()[name = tensor("op_27434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2643_cast_fp16 = mul(x = var_27433_cast_fp16, y = var_27434_to_fp16)[name = tensor("aw_chunk_2643_cast_fp16")]; + tensor var_27437_equation_0 = const()[name = tensor("op_27437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27437_cast_fp16 = einsum(equation = var_27437_equation_0, values = (var_27151_cast_fp16, var_26847_cast_fp16))[name = tensor("op_27437_cast_fp16")]; + tensor var_27438_to_fp16 = const()[name = tensor("op_27438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2645_cast_fp16 = mul(x = var_27437_cast_fp16, y = var_27438_to_fp16)[name = tensor("aw_chunk_2645_cast_fp16")]; + tensor var_27441_equation_0 = const()[name = tensor("op_27441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27441_cast_fp16 = einsum(equation = var_27441_equation_0, values = (var_27151_cast_fp16, var_26854_cast_fp16))[name = tensor("op_27441_cast_fp16")]; + tensor var_27442_to_fp16 = const()[name = tensor("op_27442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2647_cast_fp16 = mul(x = var_27441_cast_fp16, y = var_27442_to_fp16)[name = tensor("aw_chunk_2647_cast_fp16")]; + tensor var_27445_equation_0 = const()[name = tensor("op_27445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27445_cast_fp16 = einsum(equation = var_27445_equation_0, values = (var_27155_cast_fp16, var_26861_cast_fp16))[name = tensor("op_27445_cast_fp16")]; + tensor var_27446_to_fp16 = const()[name = tensor("op_27446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2649_cast_fp16 = mul(x = var_27445_cast_fp16, y = var_27446_to_fp16)[name = tensor("aw_chunk_2649_cast_fp16")]; + tensor var_27449_equation_0 = const()[name = tensor("op_27449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27449_cast_fp16 = einsum(equation = var_27449_equation_0, values = (var_27155_cast_fp16, var_26868_cast_fp16))[name = tensor("op_27449_cast_fp16")]; + tensor var_27450_to_fp16 = const()[name = tensor("op_27450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2651_cast_fp16 = mul(x = var_27449_cast_fp16, y = var_27450_to_fp16)[name = tensor("aw_chunk_2651_cast_fp16")]; + tensor var_27453_equation_0 = const()[name = tensor("op_27453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27453_cast_fp16 = einsum(equation = var_27453_equation_0, values = (var_27155_cast_fp16, var_26875_cast_fp16))[name = tensor("op_27453_cast_fp16")]; + tensor var_27454_to_fp16 = const()[name = tensor("op_27454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2653_cast_fp16 = mul(x = var_27453_cast_fp16, y = var_27454_to_fp16)[name = tensor("aw_chunk_2653_cast_fp16")]; + tensor var_27457_equation_0 = const()[name = tensor("op_27457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27457_cast_fp16 = einsum(equation = var_27457_equation_0, values = (var_27155_cast_fp16, var_26882_cast_fp16))[name = tensor("op_27457_cast_fp16")]; + tensor var_27458_to_fp16 = const()[name = tensor("op_27458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2655_cast_fp16 = mul(x = var_27457_cast_fp16, y = var_27458_to_fp16)[name = tensor("aw_chunk_2655_cast_fp16")]; + tensor var_27461_equation_0 = const()[name = tensor("op_27461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27461_cast_fp16 = einsum(equation = var_27461_equation_0, values = (var_27159_cast_fp16, var_26889_cast_fp16))[name = tensor("op_27461_cast_fp16")]; + tensor var_27462_to_fp16 = const()[name = tensor("op_27462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2657_cast_fp16 = mul(x = var_27461_cast_fp16, y = var_27462_to_fp16)[name = tensor("aw_chunk_2657_cast_fp16")]; + tensor var_27465_equation_0 = const()[name = tensor("op_27465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27465_cast_fp16 = einsum(equation = var_27465_equation_0, values = (var_27159_cast_fp16, var_26896_cast_fp16))[name = tensor("op_27465_cast_fp16")]; + tensor var_27466_to_fp16 = const()[name = tensor("op_27466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2659_cast_fp16 = mul(x = var_27465_cast_fp16, y = var_27466_to_fp16)[name = tensor("aw_chunk_2659_cast_fp16")]; + tensor var_27469_equation_0 = const()[name = tensor("op_27469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27469_cast_fp16 = einsum(equation = var_27469_equation_0, values = (var_27159_cast_fp16, var_26903_cast_fp16))[name = tensor("op_27469_cast_fp16")]; + tensor var_27470_to_fp16 = const()[name = tensor("op_27470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2661_cast_fp16 = mul(x = var_27469_cast_fp16, y = var_27470_to_fp16)[name = tensor("aw_chunk_2661_cast_fp16")]; + tensor var_27473_equation_0 = const()[name = tensor("op_27473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27473_cast_fp16 = einsum(equation = var_27473_equation_0, values = (var_27159_cast_fp16, var_26910_cast_fp16))[name = tensor("op_27473_cast_fp16")]; + tensor var_27474_to_fp16 = const()[name = tensor("op_27474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2663_cast_fp16 = mul(x = var_27473_cast_fp16, y = var_27474_to_fp16)[name = tensor("aw_chunk_2663_cast_fp16")]; + tensor var_27477_equation_0 = const()[name = tensor("op_27477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27477_cast_fp16 = einsum(equation = var_27477_equation_0, values = (var_27163_cast_fp16, var_26917_cast_fp16))[name = tensor("op_27477_cast_fp16")]; + tensor var_27478_to_fp16 = const()[name = tensor("op_27478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2665_cast_fp16 = mul(x = var_27477_cast_fp16, y = var_27478_to_fp16)[name = tensor("aw_chunk_2665_cast_fp16")]; + tensor var_27481_equation_0 = const()[name = tensor("op_27481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27481_cast_fp16 = einsum(equation = var_27481_equation_0, values = (var_27163_cast_fp16, var_26924_cast_fp16))[name = tensor("op_27481_cast_fp16")]; + tensor var_27482_to_fp16 = const()[name = tensor("op_27482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2667_cast_fp16 = mul(x = var_27481_cast_fp16, y = var_27482_to_fp16)[name = tensor("aw_chunk_2667_cast_fp16")]; + tensor var_27485_equation_0 = const()[name = tensor("op_27485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27485_cast_fp16 = einsum(equation = var_27485_equation_0, values = (var_27163_cast_fp16, var_26931_cast_fp16))[name = tensor("op_27485_cast_fp16")]; + tensor var_27486_to_fp16 = const()[name = tensor("op_27486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2669_cast_fp16 = mul(x = var_27485_cast_fp16, y = var_27486_to_fp16)[name = tensor("aw_chunk_2669_cast_fp16")]; + tensor var_27489_equation_0 = const()[name = tensor("op_27489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27489_cast_fp16 = einsum(equation = var_27489_equation_0, values = (var_27163_cast_fp16, var_26938_cast_fp16))[name = tensor("op_27489_cast_fp16")]; + tensor var_27490_to_fp16 = const()[name = tensor("op_27490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2671_cast_fp16 = mul(x = var_27489_cast_fp16, y = var_27490_to_fp16)[name = tensor("aw_chunk_2671_cast_fp16")]; + tensor var_27493_equation_0 = const()[name = tensor("op_27493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27493_cast_fp16 = einsum(equation = var_27493_equation_0, values = (var_27167_cast_fp16, var_26945_cast_fp16))[name = tensor("op_27493_cast_fp16")]; + tensor var_27494_to_fp16 = const()[name = tensor("op_27494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2673_cast_fp16 = mul(x = var_27493_cast_fp16, y = var_27494_to_fp16)[name = tensor("aw_chunk_2673_cast_fp16")]; + tensor var_27497_equation_0 = const()[name = tensor("op_27497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27497_cast_fp16 = einsum(equation = var_27497_equation_0, values = (var_27167_cast_fp16, var_26952_cast_fp16))[name = tensor("op_27497_cast_fp16")]; + tensor var_27498_to_fp16 = const()[name = tensor("op_27498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2675_cast_fp16 = mul(x = var_27497_cast_fp16, y = var_27498_to_fp16)[name = tensor("aw_chunk_2675_cast_fp16")]; + tensor var_27501_equation_0 = const()[name = tensor("op_27501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27501_cast_fp16 = einsum(equation = var_27501_equation_0, values = (var_27167_cast_fp16, var_26959_cast_fp16))[name = tensor("op_27501_cast_fp16")]; + tensor var_27502_to_fp16 = const()[name = tensor("op_27502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2677_cast_fp16 = mul(x = var_27501_cast_fp16, y = var_27502_to_fp16)[name = tensor("aw_chunk_2677_cast_fp16")]; + tensor var_27505_equation_0 = const()[name = tensor("op_27505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27505_cast_fp16 = einsum(equation = var_27505_equation_0, values = (var_27167_cast_fp16, var_26966_cast_fp16))[name = tensor("op_27505_cast_fp16")]; + tensor var_27506_to_fp16 = const()[name = tensor("op_27506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2679_cast_fp16 = mul(x = var_27505_cast_fp16, y = var_27506_to_fp16)[name = tensor("aw_chunk_2679_cast_fp16")]; + tensor var_27509_equation_0 = const()[name = tensor("op_27509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27509_cast_fp16 = einsum(equation = var_27509_equation_0, values = (var_27171_cast_fp16, var_26973_cast_fp16))[name = tensor("op_27509_cast_fp16")]; + tensor var_27510_to_fp16 = const()[name = tensor("op_27510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2681_cast_fp16 = mul(x = var_27509_cast_fp16, y = var_27510_to_fp16)[name = tensor("aw_chunk_2681_cast_fp16")]; + tensor var_27513_equation_0 = const()[name = tensor("op_27513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27513_cast_fp16 = einsum(equation = var_27513_equation_0, values = (var_27171_cast_fp16, var_26980_cast_fp16))[name = tensor("op_27513_cast_fp16")]; + tensor var_27514_to_fp16 = const()[name = tensor("op_27514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2683_cast_fp16 = mul(x = var_27513_cast_fp16, y = var_27514_to_fp16)[name = tensor("aw_chunk_2683_cast_fp16")]; + tensor var_27517_equation_0 = const()[name = tensor("op_27517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27517_cast_fp16 = einsum(equation = var_27517_equation_0, values = (var_27171_cast_fp16, var_26987_cast_fp16))[name = tensor("op_27517_cast_fp16")]; + tensor var_27518_to_fp16 = const()[name = tensor("op_27518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2685_cast_fp16 = mul(x = var_27517_cast_fp16, y = var_27518_to_fp16)[name = tensor("aw_chunk_2685_cast_fp16")]; + tensor var_27521_equation_0 = const()[name = tensor("op_27521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27521_cast_fp16 = einsum(equation = var_27521_equation_0, values = (var_27171_cast_fp16, var_26994_cast_fp16))[name = tensor("op_27521_cast_fp16")]; + tensor var_27522_to_fp16 = const()[name = tensor("op_27522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2687_cast_fp16 = mul(x = var_27521_cast_fp16, y = var_27522_to_fp16)[name = tensor("aw_chunk_2687_cast_fp16")]; + tensor var_27525_equation_0 = const()[name = tensor("op_27525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27525_cast_fp16 = einsum(equation = var_27525_equation_0, values = (var_27175_cast_fp16, var_27001_cast_fp16))[name = tensor("op_27525_cast_fp16")]; + tensor var_27526_to_fp16 = const()[name = tensor("op_27526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2689_cast_fp16 = mul(x = var_27525_cast_fp16, y = var_27526_to_fp16)[name = tensor("aw_chunk_2689_cast_fp16")]; + tensor var_27529_equation_0 = const()[name = tensor("op_27529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27529_cast_fp16 = einsum(equation = var_27529_equation_0, values = (var_27175_cast_fp16, var_27008_cast_fp16))[name = tensor("op_27529_cast_fp16")]; + tensor var_27530_to_fp16 = const()[name = tensor("op_27530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2691_cast_fp16 = mul(x = var_27529_cast_fp16, y = var_27530_to_fp16)[name = tensor("aw_chunk_2691_cast_fp16")]; + tensor var_27533_equation_0 = const()[name = tensor("op_27533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27533_cast_fp16 = einsum(equation = var_27533_equation_0, values = (var_27175_cast_fp16, var_27015_cast_fp16))[name = tensor("op_27533_cast_fp16")]; + tensor var_27534_to_fp16 = const()[name = tensor("op_27534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2693_cast_fp16 = mul(x = var_27533_cast_fp16, y = var_27534_to_fp16)[name = tensor("aw_chunk_2693_cast_fp16")]; + tensor var_27537_equation_0 = const()[name = tensor("op_27537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27537_cast_fp16 = einsum(equation = var_27537_equation_0, values = (var_27175_cast_fp16, var_27022_cast_fp16))[name = tensor("op_27537_cast_fp16")]; + tensor var_27538_to_fp16 = const()[name = tensor("op_27538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2695_cast_fp16 = mul(x = var_27537_cast_fp16, y = var_27538_to_fp16)[name = tensor("aw_chunk_2695_cast_fp16")]; + tensor var_27541_equation_0 = const()[name = tensor("op_27541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27541_cast_fp16 = einsum(equation = var_27541_equation_0, values = (var_27179_cast_fp16, var_27029_cast_fp16))[name = tensor("op_27541_cast_fp16")]; + tensor var_27542_to_fp16 = const()[name = tensor("op_27542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2697_cast_fp16 = mul(x = var_27541_cast_fp16, y = var_27542_to_fp16)[name = tensor("aw_chunk_2697_cast_fp16")]; + tensor var_27545_equation_0 = const()[name = tensor("op_27545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27545_cast_fp16 = einsum(equation = var_27545_equation_0, values = (var_27179_cast_fp16, var_27036_cast_fp16))[name = tensor("op_27545_cast_fp16")]; + tensor var_27546_to_fp16 = const()[name = tensor("op_27546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2699_cast_fp16 = mul(x = var_27545_cast_fp16, y = var_27546_to_fp16)[name = tensor("aw_chunk_2699_cast_fp16")]; + tensor var_27549_equation_0 = const()[name = tensor("op_27549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27549_cast_fp16 = einsum(equation = var_27549_equation_0, values = (var_27179_cast_fp16, var_27043_cast_fp16))[name = tensor("op_27549_cast_fp16")]; + tensor var_27550_to_fp16 = const()[name = tensor("op_27550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2701_cast_fp16 = mul(x = var_27549_cast_fp16, y = var_27550_to_fp16)[name = tensor("aw_chunk_2701_cast_fp16")]; + tensor var_27553_equation_0 = const()[name = tensor("op_27553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27553_cast_fp16 = einsum(equation = var_27553_equation_0, values = (var_27179_cast_fp16, var_27050_cast_fp16))[name = tensor("op_27553_cast_fp16")]; + tensor var_27554_to_fp16 = const()[name = tensor("op_27554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2703_cast_fp16 = mul(x = var_27553_cast_fp16, y = var_27554_to_fp16)[name = tensor("aw_chunk_2703_cast_fp16")]; + tensor var_27557_equation_0 = const()[name = tensor("op_27557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27557_cast_fp16 = einsum(equation = var_27557_equation_0, values = (var_27183_cast_fp16, var_27057_cast_fp16))[name = tensor("op_27557_cast_fp16")]; + tensor var_27558_to_fp16 = const()[name = tensor("op_27558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2705_cast_fp16 = mul(x = var_27557_cast_fp16, y = var_27558_to_fp16)[name = tensor("aw_chunk_2705_cast_fp16")]; + tensor var_27561_equation_0 = const()[name = tensor("op_27561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27561_cast_fp16 = einsum(equation = var_27561_equation_0, values = (var_27183_cast_fp16, var_27064_cast_fp16))[name = tensor("op_27561_cast_fp16")]; + tensor var_27562_to_fp16 = const()[name = tensor("op_27562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2707_cast_fp16 = mul(x = var_27561_cast_fp16, y = var_27562_to_fp16)[name = tensor("aw_chunk_2707_cast_fp16")]; + tensor var_27565_equation_0 = const()[name = tensor("op_27565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27565_cast_fp16 = einsum(equation = var_27565_equation_0, values = (var_27183_cast_fp16, var_27071_cast_fp16))[name = tensor("op_27565_cast_fp16")]; + tensor var_27566_to_fp16 = const()[name = tensor("op_27566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2709_cast_fp16 = mul(x = var_27565_cast_fp16, y = var_27566_to_fp16)[name = tensor("aw_chunk_2709_cast_fp16")]; + tensor var_27569_equation_0 = const()[name = tensor("op_27569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27569_cast_fp16 = einsum(equation = var_27569_equation_0, values = (var_27183_cast_fp16, var_27078_cast_fp16))[name = tensor("op_27569_cast_fp16")]; + tensor var_27570_to_fp16 = const()[name = tensor("op_27570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2711_cast_fp16 = mul(x = var_27569_cast_fp16, y = var_27570_to_fp16)[name = tensor("aw_chunk_2711_cast_fp16")]; + tensor var_27573_equation_0 = const()[name = tensor("op_27573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27573_cast_fp16 = einsum(equation = var_27573_equation_0, values = (var_27187_cast_fp16, var_27085_cast_fp16))[name = tensor("op_27573_cast_fp16")]; + tensor var_27574_to_fp16 = const()[name = tensor("op_27574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2713_cast_fp16 = mul(x = var_27573_cast_fp16, y = var_27574_to_fp16)[name = tensor("aw_chunk_2713_cast_fp16")]; + tensor var_27577_equation_0 = const()[name = tensor("op_27577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27577_cast_fp16 = einsum(equation = var_27577_equation_0, values = (var_27187_cast_fp16, var_27092_cast_fp16))[name = tensor("op_27577_cast_fp16")]; + tensor var_27578_to_fp16 = const()[name = tensor("op_27578_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2715_cast_fp16 = mul(x = var_27577_cast_fp16, y = var_27578_to_fp16)[name = tensor("aw_chunk_2715_cast_fp16")]; + tensor var_27581_equation_0 = const()[name = tensor("op_27581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27581_cast_fp16 = einsum(equation = var_27581_equation_0, values = (var_27187_cast_fp16, var_27099_cast_fp16))[name = tensor("op_27581_cast_fp16")]; + tensor var_27582_to_fp16 = const()[name = tensor("op_27582_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2717_cast_fp16 = mul(x = var_27581_cast_fp16, y = var_27582_to_fp16)[name = tensor("aw_chunk_2717_cast_fp16")]; + tensor var_27585_equation_0 = const()[name = tensor("op_27585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_27585_cast_fp16 = einsum(equation = var_27585_equation_0, values = (var_27187_cast_fp16, var_27106_cast_fp16))[name = tensor("op_27585_cast_fp16")]; + tensor var_27586_to_fp16 = const()[name = tensor("op_27586_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2719_cast_fp16 = mul(x = var_27585_cast_fp16, y = var_27586_to_fp16)[name = tensor("aw_chunk_2719_cast_fp16")]; + tensor var_27588_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2561_cast_fp16)[name = tensor("op_27588_cast_fp16")]; + tensor var_27589_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2563_cast_fp16)[name = tensor("op_27589_cast_fp16")]; + tensor var_27590_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2565_cast_fp16)[name = tensor("op_27590_cast_fp16")]; + tensor var_27591_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2567_cast_fp16)[name = tensor("op_27591_cast_fp16")]; + tensor var_27592_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2569_cast_fp16)[name = tensor("op_27592_cast_fp16")]; + tensor var_27593_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2571_cast_fp16)[name = tensor("op_27593_cast_fp16")]; + tensor var_27594_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2573_cast_fp16)[name = tensor("op_27594_cast_fp16")]; + tensor var_27595_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2575_cast_fp16)[name = tensor("op_27595_cast_fp16")]; + tensor var_27596_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2577_cast_fp16)[name = tensor("op_27596_cast_fp16")]; + tensor var_27597_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2579_cast_fp16)[name = tensor("op_27597_cast_fp16")]; + tensor var_27598_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2581_cast_fp16)[name = tensor("op_27598_cast_fp16")]; + tensor var_27599_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2583_cast_fp16)[name = tensor("op_27599_cast_fp16")]; + tensor var_27600_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2585_cast_fp16)[name = tensor("op_27600_cast_fp16")]; + tensor var_27601_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2587_cast_fp16)[name = tensor("op_27601_cast_fp16")]; + tensor var_27602_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2589_cast_fp16)[name = tensor("op_27602_cast_fp16")]; + tensor var_27603_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2591_cast_fp16)[name = tensor("op_27603_cast_fp16")]; + tensor var_27604_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2593_cast_fp16)[name = tensor("op_27604_cast_fp16")]; + tensor var_27605_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2595_cast_fp16)[name = tensor("op_27605_cast_fp16")]; + tensor var_27606_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2597_cast_fp16)[name = tensor("op_27606_cast_fp16")]; + tensor var_27607_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2599_cast_fp16)[name = tensor("op_27607_cast_fp16")]; + tensor var_27608_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2601_cast_fp16)[name = tensor("op_27608_cast_fp16")]; + tensor var_27609_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2603_cast_fp16)[name = tensor("op_27609_cast_fp16")]; + tensor var_27610_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2605_cast_fp16)[name = tensor("op_27610_cast_fp16")]; + tensor var_27611_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2607_cast_fp16)[name = tensor("op_27611_cast_fp16")]; + tensor var_27612_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2609_cast_fp16)[name = tensor("op_27612_cast_fp16")]; + tensor var_27613_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2611_cast_fp16)[name = tensor("op_27613_cast_fp16")]; + tensor var_27614_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2613_cast_fp16)[name = tensor("op_27614_cast_fp16")]; + tensor var_27615_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2615_cast_fp16)[name = tensor("op_27615_cast_fp16")]; + tensor var_27616_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2617_cast_fp16)[name = tensor("op_27616_cast_fp16")]; + tensor var_27617_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2619_cast_fp16)[name = tensor("op_27617_cast_fp16")]; + tensor var_27618_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2621_cast_fp16)[name = tensor("op_27618_cast_fp16")]; + tensor var_27619_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2623_cast_fp16)[name = tensor("op_27619_cast_fp16")]; + tensor var_27620_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2625_cast_fp16)[name = tensor("op_27620_cast_fp16")]; + tensor var_27621_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2627_cast_fp16)[name = tensor("op_27621_cast_fp16")]; + tensor var_27622_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2629_cast_fp16)[name = tensor("op_27622_cast_fp16")]; + tensor var_27623_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2631_cast_fp16)[name = tensor("op_27623_cast_fp16")]; + tensor var_27624_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2633_cast_fp16)[name = tensor("op_27624_cast_fp16")]; + tensor var_27625_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2635_cast_fp16)[name = tensor("op_27625_cast_fp16")]; + tensor var_27626_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2637_cast_fp16)[name = tensor("op_27626_cast_fp16")]; + tensor var_27627_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2639_cast_fp16)[name = tensor("op_27627_cast_fp16")]; + tensor var_27628_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2641_cast_fp16)[name = tensor("op_27628_cast_fp16")]; + tensor var_27629_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2643_cast_fp16)[name = tensor("op_27629_cast_fp16")]; + tensor var_27630_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2645_cast_fp16)[name = tensor("op_27630_cast_fp16")]; + tensor var_27631_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2647_cast_fp16)[name = tensor("op_27631_cast_fp16")]; + tensor var_27632_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2649_cast_fp16)[name = tensor("op_27632_cast_fp16")]; + tensor var_27633_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2651_cast_fp16)[name = tensor("op_27633_cast_fp16")]; + tensor var_27634_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2653_cast_fp16)[name = tensor("op_27634_cast_fp16")]; + tensor var_27635_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2655_cast_fp16)[name = tensor("op_27635_cast_fp16")]; + tensor var_27636_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2657_cast_fp16)[name = tensor("op_27636_cast_fp16")]; + tensor var_27637_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2659_cast_fp16)[name = tensor("op_27637_cast_fp16")]; + tensor var_27638_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2661_cast_fp16)[name = tensor("op_27638_cast_fp16")]; + tensor var_27639_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2663_cast_fp16)[name = tensor("op_27639_cast_fp16")]; + tensor var_27640_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2665_cast_fp16)[name = tensor("op_27640_cast_fp16")]; + tensor var_27641_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2667_cast_fp16)[name = tensor("op_27641_cast_fp16")]; + tensor var_27642_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2669_cast_fp16)[name = tensor("op_27642_cast_fp16")]; + tensor var_27643_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2671_cast_fp16)[name = tensor("op_27643_cast_fp16")]; + tensor var_27644_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2673_cast_fp16)[name = tensor("op_27644_cast_fp16")]; + tensor var_27645_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2675_cast_fp16)[name = tensor("op_27645_cast_fp16")]; + tensor var_27646_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2677_cast_fp16)[name = tensor("op_27646_cast_fp16")]; + tensor var_27647_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2679_cast_fp16)[name = tensor("op_27647_cast_fp16")]; + tensor var_27648_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2681_cast_fp16)[name = tensor("op_27648_cast_fp16")]; + tensor var_27649_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2683_cast_fp16)[name = tensor("op_27649_cast_fp16")]; + tensor var_27650_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2685_cast_fp16)[name = tensor("op_27650_cast_fp16")]; + tensor var_27651_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2687_cast_fp16)[name = tensor("op_27651_cast_fp16")]; + tensor var_27652_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2689_cast_fp16)[name = tensor("op_27652_cast_fp16")]; + tensor var_27653_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2691_cast_fp16)[name = tensor("op_27653_cast_fp16")]; + tensor var_27654_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2693_cast_fp16)[name = tensor("op_27654_cast_fp16")]; + tensor var_27655_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2695_cast_fp16)[name = tensor("op_27655_cast_fp16")]; + tensor var_27656_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2697_cast_fp16)[name = tensor("op_27656_cast_fp16")]; + tensor var_27657_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2699_cast_fp16)[name = tensor("op_27657_cast_fp16")]; + tensor var_27658_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2701_cast_fp16)[name = tensor("op_27658_cast_fp16")]; + tensor var_27659_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2703_cast_fp16)[name = tensor("op_27659_cast_fp16")]; + tensor var_27660_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2705_cast_fp16)[name = tensor("op_27660_cast_fp16")]; + tensor var_27661_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2707_cast_fp16)[name = tensor("op_27661_cast_fp16")]; + tensor var_27662_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2709_cast_fp16)[name = tensor("op_27662_cast_fp16")]; + tensor var_27663_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2711_cast_fp16)[name = tensor("op_27663_cast_fp16")]; + tensor var_27664_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2713_cast_fp16)[name = tensor("op_27664_cast_fp16")]; + tensor var_27665_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2715_cast_fp16)[name = tensor("op_27665_cast_fp16")]; + tensor var_27666_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2717_cast_fp16)[name = tensor("op_27666_cast_fp16")]; + tensor var_27667_cast_fp16 = softmax(axis = var_26359, x = aw_chunk_2719_cast_fp16)[name = tensor("op_27667_cast_fp16")]; + tensor var_27669_equation_0 = const()[name = tensor("op_27669_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27669_cast_fp16 = einsum(equation = var_27669_equation_0, values = (var_27189_cast_fp16, var_27588_cast_fp16))[name = tensor("op_27669_cast_fp16")]; + tensor var_27671_equation_0 = const()[name = tensor("op_27671_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27671_cast_fp16 = einsum(equation = var_27671_equation_0, values = (var_27189_cast_fp16, var_27589_cast_fp16))[name = tensor("op_27671_cast_fp16")]; + tensor var_27673_equation_0 = const()[name = tensor("op_27673_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27673_cast_fp16 = einsum(equation = var_27673_equation_0, values = (var_27189_cast_fp16, var_27590_cast_fp16))[name = tensor("op_27673_cast_fp16")]; + tensor var_27675_equation_0 = const()[name = tensor("op_27675_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27675_cast_fp16 = einsum(equation = var_27675_equation_0, values = (var_27189_cast_fp16, var_27591_cast_fp16))[name = tensor("op_27675_cast_fp16")]; + tensor var_27677_equation_0 = const()[name = tensor("op_27677_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27677_cast_fp16 = einsum(equation = var_27677_equation_0, values = (var_27193_cast_fp16, var_27592_cast_fp16))[name = tensor("op_27677_cast_fp16")]; + tensor var_27679_equation_0 = const()[name = tensor("op_27679_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27679_cast_fp16 = einsum(equation = var_27679_equation_0, values = (var_27193_cast_fp16, var_27593_cast_fp16))[name = tensor("op_27679_cast_fp16")]; + tensor var_27681_equation_0 = const()[name = tensor("op_27681_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27681_cast_fp16 = einsum(equation = var_27681_equation_0, values = (var_27193_cast_fp16, var_27594_cast_fp16))[name = tensor("op_27681_cast_fp16")]; + tensor var_27683_equation_0 = const()[name = tensor("op_27683_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27683_cast_fp16 = einsum(equation = var_27683_equation_0, values = (var_27193_cast_fp16, var_27595_cast_fp16))[name = tensor("op_27683_cast_fp16")]; + tensor var_27685_equation_0 = const()[name = tensor("op_27685_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27685_cast_fp16 = einsum(equation = var_27685_equation_0, values = (var_27197_cast_fp16, var_27596_cast_fp16))[name = tensor("op_27685_cast_fp16")]; + tensor var_27687_equation_0 = const()[name = tensor("op_27687_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27687_cast_fp16 = einsum(equation = var_27687_equation_0, values = (var_27197_cast_fp16, var_27597_cast_fp16))[name = tensor("op_27687_cast_fp16")]; + tensor var_27689_equation_0 = const()[name = tensor("op_27689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27689_cast_fp16 = einsum(equation = var_27689_equation_0, values = (var_27197_cast_fp16, var_27598_cast_fp16))[name = tensor("op_27689_cast_fp16")]; + tensor var_27691_equation_0 = const()[name = tensor("op_27691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27691_cast_fp16 = einsum(equation = var_27691_equation_0, values = (var_27197_cast_fp16, var_27599_cast_fp16))[name = tensor("op_27691_cast_fp16")]; + tensor var_27693_equation_0 = const()[name = tensor("op_27693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27693_cast_fp16 = einsum(equation = var_27693_equation_0, values = (var_27201_cast_fp16, var_27600_cast_fp16))[name = tensor("op_27693_cast_fp16")]; + tensor var_27695_equation_0 = const()[name = tensor("op_27695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27695_cast_fp16 = einsum(equation = var_27695_equation_0, values = (var_27201_cast_fp16, var_27601_cast_fp16))[name = tensor("op_27695_cast_fp16")]; + tensor var_27697_equation_0 = const()[name = tensor("op_27697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27697_cast_fp16 = einsum(equation = var_27697_equation_0, values = (var_27201_cast_fp16, var_27602_cast_fp16))[name = tensor("op_27697_cast_fp16")]; + tensor var_27699_equation_0 = const()[name = tensor("op_27699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27699_cast_fp16 = einsum(equation = var_27699_equation_0, values = (var_27201_cast_fp16, var_27603_cast_fp16))[name = tensor("op_27699_cast_fp16")]; + tensor var_27701_equation_0 = const()[name = tensor("op_27701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27701_cast_fp16 = einsum(equation = var_27701_equation_0, values = (var_27205_cast_fp16, var_27604_cast_fp16))[name = tensor("op_27701_cast_fp16")]; + tensor var_27703_equation_0 = const()[name = tensor("op_27703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27703_cast_fp16 = einsum(equation = var_27703_equation_0, values = (var_27205_cast_fp16, var_27605_cast_fp16))[name = tensor("op_27703_cast_fp16")]; + tensor var_27705_equation_0 = const()[name = tensor("op_27705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27705_cast_fp16 = einsum(equation = var_27705_equation_0, values = (var_27205_cast_fp16, var_27606_cast_fp16))[name = tensor("op_27705_cast_fp16")]; + tensor var_27707_equation_0 = const()[name = tensor("op_27707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27707_cast_fp16 = einsum(equation = var_27707_equation_0, values = (var_27205_cast_fp16, var_27607_cast_fp16))[name = tensor("op_27707_cast_fp16")]; + tensor var_27709_equation_0 = const()[name = tensor("op_27709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27709_cast_fp16 = einsum(equation = var_27709_equation_0, values = (var_27209_cast_fp16, var_27608_cast_fp16))[name = tensor("op_27709_cast_fp16")]; + tensor var_27711_equation_0 = const()[name = tensor("op_27711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27711_cast_fp16 = einsum(equation = var_27711_equation_0, values = (var_27209_cast_fp16, var_27609_cast_fp16))[name = tensor("op_27711_cast_fp16")]; + tensor var_27713_equation_0 = const()[name = tensor("op_27713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27713_cast_fp16 = einsum(equation = var_27713_equation_0, values = (var_27209_cast_fp16, var_27610_cast_fp16))[name = tensor("op_27713_cast_fp16")]; + tensor var_27715_equation_0 = const()[name = tensor("op_27715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27715_cast_fp16 = einsum(equation = var_27715_equation_0, values = (var_27209_cast_fp16, var_27611_cast_fp16))[name = tensor("op_27715_cast_fp16")]; + tensor var_27717_equation_0 = const()[name = tensor("op_27717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27717_cast_fp16 = einsum(equation = var_27717_equation_0, values = (var_27213_cast_fp16, var_27612_cast_fp16))[name = tensor("op_27717_cast_fp16")]; + tensor var_27719_equation_0 = const()[name = tensor("op_27719_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27719_cast_fp16 = einsum(equation = var_27719_equation_0, values = (var_27213_cast_fp16, var_27613_cast_fp16))[name = tensor("op_27719_cast_fp16")]; + tensor var_27721_equation_0 = const()[name = tensor("op_27721_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27721_cast_fp16 = einsum(equation = var_27721_equation_0, values = (var_27213_cast_fp16, var_27614_cast_fp16))[name = tensor("op_27721_cast_fp16")]; + tensor var_27723_equation_0 = const()[name = tensor("op_27723_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27723_cast_fp16 = einsum(equation = var_27723_equation_0, values = (var_27213_cast_fp16, var_27615_cast_fp16))[name = tensor("op_27723_cast_fp16")]; + tensor var_27725_equation_0 = const()[name = tensor("op_27725_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27725_cast_fp16 = einsum(equation = var_27725_equation_0, values = (var_27217_cast_fp16, var_27616_cast_fp16))[name = tensor("op_27725_cast_fp16")]; + tensor var_27727_equation_0 = const()[name = tensor("op_27727_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27727_cast_fp16 = einsum(equation = var_27727_equation_0, values = (var_27217_cast_fp16, var_27617_cast_fp16))[name = tensor("op_27727_cast_fp16")]; + tensor var_27729_equation_0 = const()[name = tensor("op_27729_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27729_cast_fp16 = einsum(equation = var_27729_equation_0, values = (var_27217_cast_fp16, var_27618_cast_fp16))[name = tensor("op_27729_cast_fp16")]; + tensor var_27731_equation_0 = const()[name = tensor("op_27731_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27731_cast_fp16 = einsum(equation = var_27731_equation_0, values = (var_27217_cast_fp16, var_27619_cast_fp16))[name = tensor("op_27731_cast_fp16")]; + tensor var_27733_equation_0 = const()[name = tensor("op_27733_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27733_cast_fp16 = einsum(equation = var_27733_equation_0, values = (var_27221_cast_fp16, var_27620_cast_fp16))[name = tensor("op_27733_cast_fp16")]; + tensor var_27735_equation_0 = const()[name = tensor("op_27735_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27735_cast_fp16 = einsum(equation = var_27735_equation_0, values = (var_27221_cast_fp16, var_27621_cast_fp16))[name = tensor("op_27735_cast_fp16")]; + tensor var_27737_equation_0 = const()[name = tensor("op_27737_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27737_cast_fp16 = einsum(equation = var_27737_equation_0, values = (var_27221_cast_fp16, var_27622_cast_fp16))[name = tensor("op_27737_cast_fp16")]; + tensor var_27739_equation_0 = const()[name = tensor("op_27739_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27739_cast_fp16 = einsum(equation = var_27739_equation_0, values = (var_27221_cast_fp16, var_27623_cast_fp16))[name = tensor("op_27739_cast_fp16")]; + tensor var_27741_equation_0 = const()[name = tensor("op_27741_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27741_cast_fp16 = einsum(equation = var_27741_equation_0, values = (var_27225_cast_fp16, var_27624_cast_fp16))[name = tensor("op_27741_cast_fp16")]; + tensor var_27743_equation_0 = const()[name = tensor("op_27743_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27743_cast_fp16 = einsum(equation = var_27743_equation_0, values = (var_27225_cast_fp16, var_27625_cast_fp16))[name = tensor("op_27743_cast_fp16")]; + tensor var_27745_equation_0 = const()[name = tensor("op_27745_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27745_cast_fp16 = einsum(equation = var_27745_equation_0, values = (var_27225_cast_fp16, var_27626_cast_fp16))[name = tensor("op_27745_cast_fp16")]; + tensor var_27747_equation_0 = const()[name = tensor("op_27747_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27747_cast_fp16 = einsum(equation = var_27747_equation_0, values = (var_27225_cast_fp16, var_27627_cast_fp16))[name = tensor("op_27747_cast_fp16")]; + tensor var_27749_equation_0 = const()[name = tensor("op_27749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27749_cast_fp16 = einsum(equation = var_27749_equation_0, values = (var_27229_cast_fp16, var_27628_cast_fp16))[name = tensor("op_27749_cast_fp16")]; + tensor var_27751_equation_0 = const()[name = tensor("op_27751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27751_cast_fp16 = einsum(equation = var_27751_equation_0, values = (var_27229_cast_fp16, var_27629_cast_fp16))[name = tensor("op_27751_cast_fp16")]; + tensor var_27753_equation_0 = const()[name = tensor("op_27753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27753_cast_fp16 = einsum(equation = var_27753_equation_0, values = (var_27229_cast_fp16, var_27630_cast_fp16))[name = tensor("op_27753_cast_fp16")]; + tensor var_27755_equation_0 = const()[name = tensor("op_27755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27755_cast_fp16 = einsum(equation = var_27755_equation_0, values = (var_27229_cast_fp16, var_27631_cast_fp16))[name = tensor("op_27755_cast_fp16")]; + tensor var_27757_equation_0 = const()[name = tensor("op_27757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27757_cast_fp16 = einsum(equation = var_27757_equation_0, values = (var_27233_cast_fp16, var_27632_cast_fp16))[name = tensor("op_27757_cast_fp16")]; + tensor var_27759_equation_0 = const()[name = tensor("op_27759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27759_cast_fp16 = einsum(equation = var_27759_equation_0, values = (var_27233_cast_fp16, var_27633_cast_fp16))[name = tensor("op_27759_cast_fp16")]; + tensor var_27761_equation_0 = const()[name = tensor("op_27761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27761_cast_fp16 = einsum(equation = var_27761_equation_0, values = (var_27233_cast_fp16, var_27634_cast_fp16))[name = tensor("op_27761_cast_fp16")]; + tensor var_27763_equation_0 = const()[name = tensor("op_27763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27763_cast_fp16 = einsum(equation = var_27763_equation_0, values = (var_27233_cast_fp16, var_27635_cast_fp16))[name = tensor("op_27763_cast_fp16")]; + tensor var_27765_equation_0 = const()[name = tensor("op_27765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27765_cast_fp16 = einsum(equation = var_27765_equation_0, values = (var_27237_cast_fp16, var_27636_cast_fp16))[name = tensor("op_27765_cast_fp16")]; + tensor var_27767_equation_0 = const()[name = tensor("op_27767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27767_cast_fp16 = einsum(equation = var_27767_equation_0, values = (var_27237_cast_fp16, var_27637_cast_fp16))[name = tensor("op_27767_cast_fp16")]; + tensor var_27769_equation_0 = const()[name = tensor("op_27769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27769_cast_fp16 = einsum(equation = var_27769_equation_0, values = (var_27237_cast_fp16, var_27638_cast_fp16))[name = tensor("op_27769_cast_fp16")]; + tensor var_27771_equation_0 = const()[name = tensor("op_27771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27771_cast_fp16 = einsum(equation = var_27771_equation_0, values = (var_27237_cast_fp16, var_27639_cast_fp16))[name = tensor("op_27771_cast_fp16")]; + tensor var_27773_equation_0 = const()[name = tensor("op_27773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27773_cast_fp16 = einsum(equation = var_27773_equation_0, values = (var_27241_cast_fp16, var_27640_cast_fp16))[name = tensor("op_27773_cast_fp16")]; + tensor var_27775_equation_0 = const()[name = tensor("op_27775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27775_cast_fp16 = einsum(equation = var_27775_equation_0, values = (var_27241_cast_fp16, var_27641_cast_fp16))[name = tensor("op_27775_cast_fp16")]; + tensor var_27777_equation_0 = const()[name = tensor("op_27777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27777_cast_fp16 = einsum(equation = var_27777_equation_0, values = (var_27241_cast_fp16, var_27642_cast_fp16))[name = tensor("op_27777_cast_fp16")]; + tensor var_27779_equation_0 = const()[name = tensor("op_27779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27779_cast_fp16 = einsum(equation = var_27779_equation_0, values = (var_27241_cast_fp16, var_27643_cast_fp16))[name = tensor("op_27779_cast_fp16")]; + tensor var_27781_equation_0 = const()[name = tensor("op_27781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27781_cast_fp16 = einsum(equation = var_27781_equation_0, values = (var_27245_cast_fp16, var_27644_cast_fp16))[name = tensor("op_27781_cast_fp16")]; + tensor var_27783_equation_0 = const()[name = tensor("op_27783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27783_cast_fp16 = einsum(equation = var_27783_equation_0, values = (var_27245_cast_fp16, var_27645_cast_fp16))[name = tensor("op_27783_cast_fp16")]; + tensor var_27785_equation_0 = const()[name = tensor("op_27785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27785_cast_fp16 = einsum(equation = var_27785_equation_0, values = (var_27245_cast_fp16, var_27646_cast_fp16))[name = tensor("op_27785_cast_fp16")]; + tensor var_27787_equation_0 = const()[name = tensor("op_27787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27787_cast_fp16 = einsum(equation = var_27787_equation_0, values = (var_27245_cast_fp16, var_27647_cast_fp16))[name = tensor("op_27787_cast_fp16")]; + tensor var_27789_equation_0 = const()[name = tensor("op_27789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27789_cast_fp16 = einsum(equation = var_27789_equation_0, values = (var_27249_cast_fp16, var_27648_cast_fp16))[name = tensor("op_27789_cast_fp16")]; + tensor var_27791_equation_0 = const()[name = tensor("op_27791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27791_cast_fp16 = einsum(equation = var_27791_equation_0, values = (var_27249_cast_fp16, var_27649_cast_fp16))[name = tensor("op_27791_cast_fp16")]; + tensor var_27793_equation_0 = const()[name = tensor("op_27793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27793_cast_fp16 = einsum(equation = var_27793_equation_0, values = (var_27249_cast_fp16, var_27650_cast_fp16))[name = tensor("op_27793_cast_fp16")]; + tensor var_27795_equation_0 = const()[name = tensor("op_27795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27795_cast_fp16 = einsum(equation = var_27795_equation_0, values = (var_27249_cast_fp16, var_27651_cast_fp16))[name = tensor("op_27795_cast_fp16")]; + tensor var_27797_equation_0 = const()[name = tensor("op_27797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27797_cast_fp16 = einsum(equation = var_27797_equation_0, values = (var_27253_cast_fp16, var_27652_cast_fp16))[name = tensor("op_27797_cast_fp16")]; + tensor var_27799_equation_0 = const()[name = tensor("op_27799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27799_cast_fp16 = einsum(equation = var_27799_equation_0, values = (var_27253_cast_fp16, var_27653_cast_fp16))[name = tensor("op_27799_cast_fp16")]; + tensor var_27801_equation_0 = const()[name = tensor("op_27801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27801_cast_fp16 = einsum(equation = var_27801_equation_0, values = (var_27253_cast_fp16, var_27654_cast_fp16))[name = tensor("op_27801_cast_fp16")]; + tensor var_27803_equation_0 = const()[name = tensor("op_27803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27803_cast_fp16 = einsum(equation = var_27803_equation_0, values = (var_27253_cast_fp16, var_27655_cast_fp16))[name = tensor("op_27803_cast_fp16")]; + tensor var_27805_equation_0 = const()[name = tensor("op_27805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27805_cast_fp16 = einsum(equation = var_27805_equation_0, values = (var_27257_cast_fp16, var_27656_cast_fp16))[name = tensor("op_27805_cast_fp16")]; + tensor var_27807_equation_0 = const()[name = tensor("op_27807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27807_cast_fp16 = einsum(equation = var_27807_equation_0, values = (var_27257_cast_fp16, var_27657_cast_fp16))[name = tensor("op_27807_cast_fp16")]; + tensor var_27809_equation_0 = const()[name = tensor("op_27809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27809_cast_fp16 = einsum(equation = var_27809_equation_0, values = (var_27257_cast_fp16, var_27658_cast_fp16))[name = tensor("op_27809_cast_fp16")]; + tensor var_27811_equation_0 = const()[name = tensor("op_27811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27811_cast_fp16 = einsum(equation = var_27811_equation_0, values = (var_27257_cast_fp16, var_27659_cast_fp16))[name = tensor("op_27811_cast_fp16")]; + tensor var_27813_equation_0 = const()[name = tensor("op_27813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27813_cast_fp16 = einsum(equation = var_27813_equation_0, values = (var_27261_cast_fp16, var_27660_cast_fp16))[name = tensor("op_27813_cast_fp16")]; + tensor var_27815_equation_0 = const()[name = tensor("op_27815_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27815_cast_fp16 = einsum(equation = var_27815_equation_0, values = (var_27261_cast_fp16, var_27661_cast_fp16))[name = tensor("op_27815_cast_fp16")]; + tensor var_27817_equation_0 = const()[name = tensor("op_27817_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27817_cast_fp16 = einsum(equation = var_27817_equation_0, values = (var_27261_cast_fp16, var_27662_cast_fp16))[name = tensor("op_27817_cast_fp16")]; + tensor var_27819_equation_0 = const()[name = tensor("op_27819_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27819_cast_fp16 = einsum(equation = var_27819_equation_0, values = (var_27261_cast_fp16, var_27663_cast_fp16))[name = tensor("op_27819_cast_fp16")]; + tensor var_27821_equation_0 = const()[name = tensor("op_27821_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27821_cast_fp16 = einsum(equation = var_27821_equation_0, values = (var_27265_cast_fp16, var_27664_cast_fp16))[name = tensor("op_27821_cast_fp16")]; + tensor var_27823_equation_0 = const()[name = tensor("op_27823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27823_cast_fp16 = einsum(equation = var_27823_equation_0, values = (var_27265_cast_fp16, var_27665_cast_fp16))[name = tensor("op_27823_cast_fp16")]; + tensor var_27825_equation_0 = const()[name = tensor("op_27825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27825_cast_fp16 = einsum(equation = var_27825_equation_0, values = (var_27265_cast_fp16, var_27666_cast_fp16))[name = tensor("op_27825_cast_fp16")]; + tensor var_27827_equation_0 = const()[name = tensor("op_27827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_27827_cast_fp16 = einsum(equation = var_27827_equation_0, values = (var_27265_cast_fp16, var_27667_cast_fp16))[name = tensor("op_27827_cast_fp16")]; + tensor var_27829_interleave_0 = const()[name = tensor("op_27829_interleave_0"), val = tensor(false)]; + tensor var_27829_cast_fp16 = concat(axis = var_26334, interleave = var_27829_interleave_0, values = (var_27669_cast_fp16, var_27671_cast_fp16, var_27673_cast_fp16, var_27675_cast_fp16))[name = tensor("op_27829_cast_fp16")]; + tensor var_27831_interleave_0 = const()[name = tensor("op_27831_interleave_0"), val = tensor(false)]; + tensor var_27831_cast_fp16 = concat(axis = var_26334, interleave = var_27831_interleave_0, values = (var_27677_cast_fp16, var_27679_cast_fp16, var_27681_cast_fp16, var_27683_cast_fp16))[name = tensor("op_27831_cast_fp16")]; + tensor var_27833_interleave_0 = const()[name = tensor("op_27833_interleave_0"), val = tensor(false)]; + tensor var_27833_cast_fp16 = concat(axis = var_26334, interleave = var_27833_interleave_0, values = (var_27685_cast_fp16, var_27687_cast_fp16, var_27689_cast_fp16, var_27691_cast_fp16))[name = tensor("op_27833_cast_fp16")]; + tensor var_27835_interleave_0 = const()[name = tensor("op_27835_interleave_0"), val = tensor(false)]; + tensor var_27835_cast_fp16 = concat(axis = var_26334, interleave = var_27835_interleave_0, values = (var_27693_cast_fp16, var_27695_cast_fp16, var_27697_cast_fp16, var_27699_cast_fp16))[name = tensor("op_27835_cast_fp16")]; + tensor var_27837_interleave_0 = const()[name = tensor("op_27837_interleave_0"), val = tensor(false)]; + tensor var_27837_cast_fp16 = concat(axis = var_26334, interleave = var_27837_interleave_0, values = (var_27701_cast_fp16, var_27703_cast_fp16, var_27705_cast_fp16, var_27707_cast_fp16))[name = tensor("op_27837_cast_fp16")]; + tensor var_27839_interleave_0 = const()[name = tensor("op_27839_interleave_0"), val = tensor(false)]; + tensor var_27839_cast_fp16 = concat(axis = var_26334, interleave = var_27839_interleave_0, values = (var_27709_cast_fp16, var_27711_cast_fp16, var_27713_cast_fp16, var_27715_cast_fp16))[name = tensor("op_27839_cast_fp16")]; + tensor var_27841_interleave_0 = const()[name = tensor("op_27841_interleave_0"), val = tensor(false)]; + tensor var_27841_cast_fp16 = concat(axis = var_26334, interleave = var_27841_interleave_0, values = (var_27717_cast_fp16, var_27719_cast_fp16, var_27721_cast_fp16, var_27723_cast_fp16))[name = tensor("op_27841_cast_fp16")]; + tensor var_27843_interleave_0 = const()[name = tensor("op_27843_interleave_0"), val = tensor(false)]; + tensor var_27843_cast_fp16 = concat(axis = var_26334, interleave = var_27843_interleave_0, values = (var_27725_cast_fp16, var_27727_cast_fp16, var_27729_cast_fp16, var_27731_cast_fp16))[name = tensor("op_27843_cast_fp16")]; + tensor var_27845_interleave_0 = const()[name = tensor("op_27845_interleave_0"), val = tensor(false)]; + tensor var_27845_cast_fp16 = concat(axis = var_26334, interleave = var_27845_interleave_0, values = (var_27733_cast_fp16, var_27735_cast_fp16, var_27737_cast_fp16, var_27739_cast_fp16))[name = tensor("op_27845_cast_fp16")]; + tensor var_27847_interleave_0 = const()[name = tensor("op_27847_interleave_0"), val = tensor(false)]; + tensor var_27847_cast_fp16 = concat(axis = var_26334, interleave = var_27847_interleave_0, values = (var_27741_cast_fp16, var_27743_cast_fp16, var_27745_cast_fp16, var_27747_cast_fp16))[name = tensor("op_27847_cast_fp16")]; + tensor var_27849_interleave_0 = const()[name = tensor("op_27849_interleave_0"), val = tensor(false)]; + tensor var_27849_cast_fp16 = concat(axis = var_26334, interleave = var_27849_interleave_0, values = (var_27749_cast_fp16, var_27751_cast_fp16, var_27753_cast_fp16, var_27755_cast_fp16))[name = tensor("op_27849_cast_fp16")]; + tensor var_27851_interleave_0 = const()[name = tensor("op_27851_interleave_0"), val = tensor(false)]; + tensor var_27851_cast_fp16 = concat(axis = var_26334, interleave = var_27851_interleave_0, values = (var_27757_cast_fp16, var_27759_cast_fp16, var_27761_cast_fp16, var_27763_cast_fp16))[name = tensor("op_27851_cast_fp16")]; + tensor var_27853_interleave_0 = const()[name = tensor("op_27853_interleave_0"), val = tensor(false)]; + tensor var_27853_cast_fp16 = concat(axis = var_26334, interleave = var_27853_interleave_0, values = (var_27765_cast_fp16, var_27767_cast_fp16, var_27769_cast_fp16, var_27771_cast_fp16))[name = tensor("op_27853_cast_fp16")]; + tensor var_27855_interleave_0 = const()[name = tensor("op_27855_interleave_0"), val = tensor(false)]; + tensor var_27855_cast_fp16 = concat(axis = var_26334, interleave = var_27855_interleave_0, values = (var_27773_cast_fp16, var_27775_cast_fp16, var_27777_cast_fp16, var_27779_cast_fp16))[name = tensor("op_27855_cast_fp16")]; + tensor var_27857_interleave_0 = const()[name = tensor("op_27857_interleave_0"), val = tensor(false)]; + tensor var_27857_cast_fp16 = concat(axis = var_26334, interleave = var_27857_interleave_0, values = (var_27781_cast_fp16, var_27783_cast_fp16, var_27785_cast_fp16, var_27787_cast_fp16))[name = tensor("op_27857_cast_fp16")]; + tensor var_27859_interleave_0 = const()[name = tensor("op_27859_interleave_0"), val = tensor(false)]; + tensor var_27859_cast_fp16 = concat(axis = var_26334, interleave = var_27859_interleave_0, values = (var_27789_cast_fp16, var_27791_cast_fp16, var_27793_cast_fp16, var_27795_cast_fp16))[name = tensor("op_27859_cast_fp16")]; + tensor var_27861_interleave_0 = const()[name = tensor("op_27861_interleave_0"), val = tensor(false)]; + tensor var_27861_cast_fp16 = concat(axis = var_26334, interleave = var_27861_interleave_0, values = (var_27797_cast_fp16, var_27799_cast_fp16, var_27801_cast_fp16, var_27803_cast_fp16))[name = tensor("op_27861_cast_fp16")]; + tensor var_27863_interleave_0 = const()[name = tensor("op_27863_interleave_0"), val = tensor(false)]; + tensor var_27863_cast_fp16 = concat(axis = var_26334, interleave = var_27863_interleave_0, values = (var_27805_cast_fp16, var_27807_cast_fp16, var_27809_cast_fp16, var_27811_cast_fp16))[name = tensor("op_27863_cast_fp16")]; + tensor var_27865_interleave_0 = const()[name = tensor("op_27865_interleave_0"), val = tensor(false)]; + tensor var_27865_cast_fp16 = concat(axis = var_26334, interleave = var_27865_interleave_0, values = (var_27813_cast_fp16, var_27815_cast_fp16, var_27817_cast_fp16, var_27819_cast_fp16))[name = tensor("op_27865_cast_fp16")]; + tensor var_27867_interleave_0 = const()[name = tensor("op_27867_interleave_0"), val = tensor(false)]; + tensor var_27867_cast_fp16 = concat(axis = var_26334, interleave = var_27867_interleave_0, values = (var_27821_cast_fp16, var_27823_cast_fp16, var_27825_cast_fp16, var_27827_cast_fp16))[name = tensor("op_27867_cast_fp16")]; + tensor input_327_interleave_0 = const()[name = tensor("input_327_interleave_0"), val = tensor(false)]; + tensor input_327_cast_fp16 = concat(axis = var_26359, interleave = input_327_interleave_0, values = (var_27829_cast_fp16, var_27831_cast_fp16, var_27833_cast_fp16, var_27835_cast_fp16, var_27837_cast_fp16, var_27839_cast_fp16, var_27841_cast_fp16, var_27843_cast_fp16, var_27845_cast_fp16, var_27847_cast_fp16, var_27849_cast_fp16, var_27851_cast_fp16, var_27853_cast_fp16, var_27855_cast_fp16, var_27857_cast_fp16, var_27859_cast_fp16, var_27861_cast_fp16, var_27863_cast_fp16, var_27865_cast_fp16, var_27867_cast_fp16))[name = tensor("input_327_cast_fp16")]; + tensor var_27875 = const()[name = tensor("op_27875"), val = tensor([1, 1])]; + tensor var_27877 = const()[name = tensor("op_27877"), val = tensor([1, 1])]; + tensor pretrained_out_199_pad_type_0 = const()[name = tensor("pretrained_out_199_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_199_pad_0 = const()[name = tensor("pretrained_out_199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186635008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187454272))), name = tensor("layers_16_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_16_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187454400)))]; + tensor pretrained_out_199_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_27877, groups = var_26359, pad = pretrained_out_199_pad_0, pad_type = pretrained_out_199_pad_type_0, strides = var_27875, weight = layers_16_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_327_cast_fp16)[name = tensor("pretrained_out_199_cast_fp16")]; + tensor var_27881 = const()[name = tensor("op_27881"), val = tensor([1, 1])]; + tensor var_27883 = const()[name = tensor("op_27883"), val = tensor([1, 1])]; + tensor input_329_pad_type_0 = const()[name = tensor("input_329_pad_type_0"), val = tensor("custom")]; + tensor input_329_pad_0 = const()[name = tensor("input_329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187457024)))]; + tensor input_329_cast_fp16 = conv(dilations = var_27883, groups = var_26359, pad = input_329_pad_0, pad_type = input_329_pad_type_0, strides = var_27881, weight = layers_16_self_attn_o_proj_loraA_weight_to_fp16, x = input_327_cast_fp16)[name = tensor("input_329_cast_fp16")]; + tensor var_27887 = const()[name = tensor("op_27887"), val = tensor([1, 1])]; + tensor var_27889 = const()[name = tensor("op_27889"), val = tensor([1, 1])]; + tensor lora_out_397_pad_type_0 = const()[name = tensor("lora_out_397_pad_type_0"), val = tensor("custom")]; + tensor lora_out_397_pad_0 = const()[name = tensor("lora_out_397_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_399_weight_0_to_fp16 = const()[name = tensor("lora_out_399_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187498048)))]; + tensor lora_out_399_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_27889, groups = var_26359, pad = lora_out_397_pad_0, pad_type = lora_out_397_pad_type_0, strides = var_27887, weight = lora_out_399_weight_0_to_fp16, x = input_329_cast_fp16)[name = tensor("lora_out_399_cast_fp16")]; + tensor obj_67_cast_fp16 = add(x = pretrained_out_199_cast_fp16, y = lora_out_399_cast_fp16)[name = tensor("obj_67_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = obj_67_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; + tensor var_27898 = const()[name = tensor("op_27898"), val = tensor([1])]; + tensor channels_mean_67_cast_fp16 = reduce_mean(axes = var_27898, keep_dims = var_26360, x = inputs_67_cast_fp16)[name = tensor("channels_mean_67_cast_fp16")]; + tensor zero_mean_67_cast_fp16 = sub(x = inputs_67_cast_fp16, y = channels_mean_67_cast_fp16)[name = tensor("zero_mean_67_cast_fp16")]; + tensor zero_mean_sq_67_cast_fp16 = mul(x = zero_mean_67_cast_fp16, y = zero_mean_67_cast_fp16)[name = tensor("zero_mean_sq_67_cast_fp16")]; + tensor var_27902 = const()[name = tensor("op_27902"), val = tensor([1])]; + tensor var_27903_cast_fp16 = reduce_mean(axes = var_27902, keep_dims = var_26360, x = zero_mean_sq_67_cast_fp16)[name = tensor("op_27903_cast_fp16")]; + tensor var_27904_to_fp16 = const()[name = tensor("op_27904_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_27905_cast_fp16 = add(x = var_27903_cast_fp16, y = var_27904_to_fp16)[name = tensor("op_27905_cast_fp16")]; + tensor denom_67_epsilon_0 = const()[name = tensor("denom_67_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_67_cast_fp16 = rsqrt(epsilon = denom_67_epsilon_0, x = var_27905_cast_fp16)[name = tensor("denom_67_cast_fp16")]; + tensor out_67_cast_fp16 = mul(x = zero_mean_67_cast_fp16, y = denom_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; + tensor input_331_gamma_0_to_fp16 = const()[name = tensor("input_331_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187539072)))]; + tensor input_331_beta_0_to_fp16 = const()[name = tensor("input_331_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187541696)))]; + tensor input_331_epsilon_0_to_fp16 = const()[name = tensor("input_331_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_331_cast_fp16 = batch_norm(beta = input_331_beta_0_to_fp16, epsilon = input_331_epsilon_0_to_fp16, gamma = input_331_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor("input_331_cast_fp16")]; + tensor var_27919 = const()[name = tensor("op_27919"), val = tensor([1, 1])]; + tensor var_27921 = const()[name = tensor("op_27921"), val = tensor([1, 1])]; + tensor pretrained_out_201_pad_type_0 = const()[name = tensor("pretrained_out_201_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_201_pad_0 = const()[name = tensor("pretrained_out_201_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(187544320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190821184))), name = tensor("layers_16_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_16_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_16_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190821312)))]; + tensor pretrained_out_201_cast_fp16 = conv(bias = layers_16_fc1_pretrained_bias_to_fp16, dilations = var_27921, groups = var_26359, pad = pretrained_out_201_pad_0, pad_type = pretrained_out_201_pad_type_0, strides = var_27919, weight = layers_16_fc1_pretrained_weight_to_fp16_palettized, x = input_331_cast_fp16)[name = tensor("pretrained_out_201_cast_fp16")]; + tensor var_27925 = const()[name = tensor("op_27925"), val = tensor([1, 1])]; + tensor var_27927 = const()[name = tensor("op_27927"), val = tensor([1, 1])]; + tensor input_333_pad_type_0 = const()[name = tensor("input_333_pad_type_0"), val = tensor("custom")]; + tensor input_333_pad_0 = const()[name = tensor("input_333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_16_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190831616)))]; + tensor input_333_cast_fp16 = conv(dilations = var_27927, groups = var_26359, pad = input_333_pad_0, pad_type = input_333_pad_type_0, strides = var_27925, weight = layers_16_fc1_loraA_weight_to_fp16, x = input_331_cast_fp16)[name = tensor("input_333_cast_fp16")]; + tensor var_27931 = const()[name = tensor("op_27931"), val = tensor([1, 1])]; + tensor var_27933 = const()[name = tensor("op_27933"), val = tensor([1, 1])]; + tensor lora_out_401_pad_type_0 = const()[name = tensor("lora_out_401_pad_type_0"), val = tensor("custom")]; + tensor lora_out_401_pad_0 = const()[name = tensor("lora_out_401_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_403_weight_0_to_fp16 = const()[name = tensor("lora_out_403_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190872640)))]; + tensor lora_out_403_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_27933, groups = var_26359, pad = lora_out_401_pad_0, pad_type = lora_out_401_pad_type_0, strides = var_27931, weight = lora_out_403_weight_0_to_fp16, x = input_333_cast_fp16)[name = tensor("lora_out_403_cast_fp16")]; + tensor input_335_cast_fp16 = add(x = pretrained_out_201_cast_fp16, y = lora_out_403_cast_fp16)[name = tensor("input_335_cast_fp16")]; + tensor input_337_mode_0 = const()[name = tensor("input_337_mode_0"), val = tensor("EXACT")]; + tensor input_337_cast_fp16 = gelu(mode = input_337_mode_0, x = input_335_cast_fp16)[name = tensor("input_337_cast_fp16")]; + tensor var_27945 = const()[name = tensor("op_27945"), val = tensor([1, 1])]; + tensor var_27947 = const()[name = tensor("op_27947"), val = tensor([1, 1])]; + tensor pretrained_out_203_pad_type_0 = const()[name = tensor("pretrained_out_203_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_203_pad_0 = const()[name = tensor("pretrained_out_203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(191036544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194313408))), name = tensor("layers_16_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_16_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_16_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194313536)))]; + tensor pretrained_out_203_cast_fp16 = conv(bias = layers_16_fc2_pretrained_bias_to_fp16, dilations = var_27947, groups = var_26359, pad = pretrained_out_203_pad_0, pad_type = pretrained_out_203_pad_type_0, strides = var_27945, weight = layers_16_fc2_pretrained_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = tensor("pretrained_out_203_cast_fp16")]; + tensor var_27951 = const()[name = tensor("op_27951"), val = tensor([1, 1])]; + tensor var_27953 = const()[name = tensor("op_27953"), val = tensor([1, 1])]; + tensor input_339_pad_type_0 = const()[name = tensor("input_339_pad_type_0"), val = tensor("custom")]; + tensor input_339_pad_0 = const()[name = tensor("input_339_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_16_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_16_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194316160)))]; + tensor input_339_cast_fp16 = conv(dilations = var_27953, groups = var_26359, pad = input_339_pad_0, pad_type = input_339_pad_type_0, strides = var_27951, weight = layers_16_fc2_loraA_weight_to_fp16, x = input_337_cast_fp16)[name = tensor("input_339_cast_fp16")]; + tensor var_27957 = const()[name = tensor("op_27957"), val = tensor([1, 1])]; + tensor var_27959 = const()[name = tensor("op_27959"), val = tensor([1, 1])]; + tensor lora_out_405_pad_type_0 = const()[name = tensor("lora_out_405_pad_type_0"), val = tensor("custom")]; + tensor lora_out_405_pad_0 = const()[name = tensor("lora_out_405_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_407_weight_0_to_fp16 = const()[name = tensor("lora_out_407_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194480064)))]; + tensor lora_out_407_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_27959, groups = var_26359, pad = lora_out_405_pad_0, pad_type = lora_out_405_pad_type_0, strides = var_27957, weight = lora_out_407_weight_0_to_fp16, x = input_339_cast_fp16)[name = tensor("lora_out_407_cast_fp16")]; + tensor hidden_states_37_cast_fp16 = add(x = pretrained_out_203_cast_fp16, y = lora_out_407_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = hidden_states_37_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; + tensor var_27969 = const()[name = tensor("op_27969"), val = tensor(3)]; + tensor var_27994 = const()[name = tensor("op_27994"), val = tensor(1)]; + tensor var_27995 = const()[name = tensor("op_27995"), val = tensor(true)]; + tensor var_28005 = const()[name = tensor("op_28005"), val = tensor([1])]; + tensor channels_mean_69_cast_fp16 = reduce_mean(axes = var_28005, keep_dims = var_27995, x = inputs_69_cast_fp16)[name = tensor("channels_mean_69_cast_fp16")]; + tensor zero_mean_69_cast_fp16 = sub(x = inputs_69_cast_fp16, y = channels_mean_69_cast_fp16)[name = tensor("zero_mean_69_cast_fp16")]; + tensor zero_mean_sq_69_cast_fp16 = mul(x = zero_mean_69_cast_fp16, y = zero_mean_69_cast_fp16)[name = tensor("zero_mean_sq_69_cast_fp16")]; + tensor var_28009 = const()[name = tensor("op_28009"), val = tensor([1])]; + tensor var_28010_cast_fp16 = reduce_mean(axes = var_28009, keep_dims = var_27995, x = zero_mean_sq_69_cast_fp16)[name = tensor("op_28010_cast_fp16")]; + tensor var_28011_to_fp16 = const()[name = tensor("op_28011_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_28012_cast_fp16 = add(x = var_28010_cast_fp16, y = var_28011_to_fp16)[name = tensor("op_28012_cast_fp16")]; + tensor denom_69_epsilon_0 = const()[name = tensor("denom_69_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_69_cast_fp16 = rsqrt(epsilon = denom_69_epsilon_0, x = var_28012_cast_fp16)[name = tensor("denom_69_cast_fp16")]; + tensor out_69_cast_fp16 = mul(x = zero_mean_69_cast_fp16, y = denom_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; + tensor obj_69_gamma_0_to_fp16 = const()[name = tensor("obj_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194521088)))]; + tensor obj_69_beta_0_to_fp16 = const()[name = tensor("obj_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194523712)))]; + tensor obj_69_epsilon_0_to_fp16 = const()[name = tensor("obj_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_69_cast_fp16 = batch_norm(beta = obj_69_beta_0_to_fp16, epsilon = obj_69_epsilon_0_to_fp16, gamma = obj_69_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor("obj_69_cast_fp16")]; + tensor var_28030 = const()[name = tensor("op_28030"), val = tensor([1, 1])]; + tensor var_28032 = const()[name = tensor("op_28032"), val = tensor([1, 1])]; + tensor pretrained_out_205_pad_type_0 = const()[name = tensor("pretrained_out_205_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_205_pad_0 = const()[name = tensor("pretrained_out_205_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194526336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195345600))), name = tensor("layers_17_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_17_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195345728)))]; + tensor pretrained_out_205_cast_fp16 = conv(bias = layers_17_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_28032, groups = var_27994, pad = pretrained_out_205_pad_0, pad_type = pretrained_out_205_pad_type_0, strides = var_28030, weight = layers_17_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_69_cast_fp16)[name = tensor("pretrained_out_205_cast_fp16")]; + tensor var_28036 = const()[name = tensor("op_28036"), val = tensor([1, 1])]; + tensor var_28038 = const()[name = tensor("op_28038"), val = tensor([1, 1])]; + tensor input_341_pad_type_0 = const()[name = tensor("input_341_pad_type_0"), val = tensor("custom")]; + tensor input_341_pad_0 = const()[name = tensor("input_341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195348352)))]; + tensor input_341_cast_fp16 = conv(dilations = var_28038, groups = var_27994, pad = input_341_pad_0, pad_type = input_341_pad_type_0, strides = var_28036, weight = layers_17_self_attn_q_proj_loraA_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor("input_341_cast_fp16")]; + tensor var_28042 = const()[name = tensor("op_28042"), val = tensor([1, 1])]; + tensor var_28044 = const()[name = tensor("op_28044"), val = tensor([1, 1])]; + tensor lora_out_409_pad_type_0 = const()[name = tensor("lora_out_409_pad_type_0"), val = tensor("custom")]; + tensor lora_out_409_pad_0 = const()[name = tensor("lora_out_409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_411_weight_0_to_fp16 = const()[name = tensor("lora_out_411_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195389376)))]; + tensor lora_out_411_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_28044, groups = var_27994, pad = lora_out_409_pad_0, pad_type = lora_out_409_pad_type_0, strides = var_28042, weight = lora_out_411_weight_0_to_fp16, x = input_341_cast_fp16)[name = tensor("lora_out_411_cast_fp16")]; + tensor query_35_cast_fp16 = add(x = pretrained_out_205_cast_fp16, y = lora_out_411_cast_fp16)[name = tensor("query_35_cast_fp16")]; + tensor var_28054 = const()[name = tensor("op_28054"), val = tensor([1, 1])]; + tensor var_28056 = const()[name = tensor("op_28056"), val = tensor([1, 1])]; + tensor pretrained_out_207_pad_type_0 = const()[name = tensor("pretrained_out_207_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_207_pad_0 = const()[name = tensor("pretrained_out_207_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(195430400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196249664))), name = tensor("layers_17_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_207_cast_fp16 = conv(dilations = var_28056, groups = var_27994, pad = pretrained_out_207_pad_0, pad_type = pretrained_out_207_pad_type_0, strides = var_28054, weight = layers_17_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_69_cast_fp16)[name = tensor("pretrained_out_207_cast_fp16")]; + tensor var_28060 = const()[name = tensor("op_28060"), val = tensor([1, 1])]; + tensor var_28062 = const()[name = tensor("op_28062"), val = tensor([1, 1])]; + tensor input_343_pad_type_0 = const()[name = tensor("input_343_pad_type_0"), val = tensor("custom")]; + tensor input_343_pad_0 = const()[name = tensor("input_343_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196249792)))]; + tensor input_343_cast_fp16 = conv(dilations = var_28062, groups = var_27994, pad = input_343_pad_0, pad_type = input_343_pad_type_0, strides = var_28060, weight = layers_17_self_attn_k_proj_loraA_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor("input_343_cast_fp16")]; + tensor var_28066 = const()[name = tensor("op_28066"), val = tensor([1, 1])]; + tensor var_28068 = const()[name = tensor("op_28068"), val = tensor([1, 1])]; + tensor lora_out_413_pad_type_0 = const()[name = tensor("lora_out_413_pad_type_0"), val = tensor("custom")]; + tensor lora_out_413_pad_0 = const()[name = tensor("lora_out_413_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_415_weight_0_to_fp16 = const()[name = tensor("lora_out_415_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196290816)))]; + tensor lora_out_415_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_28068, groups = var_27994, pad = lora_out_413_pad_0, pad_type = lora_out_413_pad_type_0, strides = var_28066, weight = lora_out_415_weight_0_to_fp16, x = input_343_cast_fp16)[name = tensor("lora_out_415_cast_fp16")]; + tensor key_35_cast_fp16 = add(x = pretrained_out_207_cast_fp16, y = lora_out_415_cast_fp16)[name = tensor("key_35_cast_fp16")]; + tensor var_28079 = const()[name = tensor("op_28079"), val = tensor([1, 1])]; + tensor var_28081 = const()[name = tensor("op_28081"), val = tensor([1, 1])]; + tensor pretrained_out_209_pad_type_0 = const()[name = tensor("pretrained_out_209_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_209_pad_0 = const()[name = tensor("pretrained_out_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196331840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197151104))), name = tensor("layers_17_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_17_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197151232)))]; + tensor pretrained_out_209_cast_fp16 = conv(bias = layers_17_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_28081, groups = var_27994, pad = pretrained_out_209_pad_0, pad_type = pretrained_out_209_pad_type_0, strides = var_28079, weight = layers_17_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_69_cast_fp16)[name = tensor("pretrained_out_209_cast_fp16")]; + tensor var_28085 = const()[name = tensor("op_28085"), val = tensor([1, 1])]; + tensor var_28087 = const()[name = tensor("op_28087"), val = tensor([1, 1])]; + tensor input_345_pad_type_0 = const()[name = tensor("input_345_pad_type_0"), val = tensor("custom")]; + tensor input_345_pad_0 = const()[name = tensor("input_345_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197153856)))]; + tensor input_345_cast_fp16 = conv(dilations = var_28087, groups = var_27994, pad = input_345_pad_0, pad_type = input_345_pad_type_0, strides = var_28085, weight = layers_17_self_attn_v_proj_loraA_weight_to_fp16, x = obj_69_cast_fp16)[name = tensor("input_345_cast_fp16")]; + tensor var_28091 = const()[name = tensor("op_28091"), val = tensor([1, 1])]; + tensor var_28093 = const()[name = tensor("op_28093"), val = tensor([1, 1])]; + tensor lora_out_417_pad_type_0 = const()[name = tensor("lora_out_417_pad_type_0"), val = tensor("custom")]; + tensor lora_out_417_pad_0 = const()[name = tensor("lora_out_417_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_419_weight_0_to_fp16 = const()[name = tensor("lora_out_419_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197194880)))]; + tensor lora_out_419_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_28093, groups = var_27994, pad = lora_out_417_pad_0, pad_type = lora_out_417_pad_type_0, strides = var_28091, weight = lora_out_419_weight_0_to_fp16, x = input_345_cast_fp16)[name = tensor("lora_out_419_cast_fp16")]; + tensor value_35_cast_fp16 = add(x = pretrained_out_209_cast_fp16, y = lora_out_419_cast_fp16)[name = tensor("value_35_cast_fp16")]; + tensor var_28103_begin_0 = const()[name = tensor("op_28103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28103_end_0 = const()[name = tensor("op_28103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28103_end_mask_0 = const()[name = tensor("op_28103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28103_cast_fp16 = slice_by_index(begin = var_28103_begin_0, end = var_28103_end_0, end_mask = var_28103_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28103_cast_fp16")]; + tensor var_28107_begin_0 = const()[name = tensor("op_28107_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28107_end_0 = const()[name = tensor("op_28107_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_28107_end_mask_0 = const()[name = tensor("op_28107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28107_cast_fp16 = slice_by_index(begin = var_28107_begin_0, end = var_28107_end_0, end_mask = var_28107_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28107_cast_fp16")]; + tensor var_28111_begin_0 = const()[name = tensor("op_28111_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28111_end_0 = const()[name = tensor("op_28111_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_28111_end_mask_0 = const()[name = tensor("op_28111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28111_cast_fp16 = slice_by_index(begin = var_28111_begin_0, end = var_28111_end_0, end_mask = var_28111_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28111_cast_fp16")]; + tensor var_28115_begin_0 = const()[name = tensor("op_28115_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28115_end_0 = const()[name = tensor("op_28115_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_28115_end_mask_0 = const()[name = tensor("op_28115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28115_cast_fp16 = slice_by_index(begin = var_28115_begin_0, end = var_28115_end_0, end_mask = var_28115_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28115_cast_fp16")]; + tensor var_28119_begin_0 = const()[name = tensor("op_28119_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28119_end_0 = const()[name = tensor("op_28119_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_28119_end_mask_0 = const()[name = tensor("op_28119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28119_cast_fp16 = slice_by_index(begin = var_28119_begin_0, end = var_28119_end_0, end_mask = var_28119_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28119_cast_fp16")]; + tensor var_28123_begin_0 = const()[name = tensor("op_28123_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28123_end_0 = const()[name = tensor("op_28123_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_28123_end_mask_0 = const()[name = tensor("op_28123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28123_cast_fp16 = slice_by_index(begin = var_28123_begin_0, end = var_28123_end_0, end_mask = var_28123_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28123_cast_fp16")]; + tensor var_28127_begin_0 = const()[name = tensor("op_28127_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28127_end_0 = const()[name = tensor("op_28127_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_28127_end_mask_0 = const()[name = tensor("op_28127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28127_cast_fp16 = slice_by_index(begin = var_28127_begin_0, end = var_28127_end_0, end_mask = var_28127_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28127_cast_fp16")]; + tensor var_28131_begin_0 = const()[name = tensor("op_28131_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28131_end_0 = const()[name = tensor("op_28131_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_28131_end_mask_0 = const()[name = tensor("op_28131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28131_cast_fp16 = slice_by_index(begin = var_28131_begin_0, end = var_28131_end_0, end_mask = var_28131_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28131_cast_fp16")]; + tensor var_28135_begin_0 = const()[name = tensor("op_28135_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28135_end_0 = const()[name = tensor("op_28135_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_28135_end_mask_0 = const()[name = tensor("op_28135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28135_cast_fp16 = slice_by_index(begin = var_28135_begin_0, end = var_28135_end_0, end_mask = var_28135_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28135_cast_fp16")]; + tensor var_28139_begin_0 = const()[name = tensor("op_28139_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28139_end_0 = const()[name = tensor("op_28139_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_28139_end_mask_0 = const()[name = tensor("op_28139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28139_cast_fp16 = slice_by_index(begin = var_28139_begin_0, end = var_28139_end_0, end_mask = var_28139_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28139_cast_fp16")]; + tensor var_28143_begin_0 = const()[name = tensor("op_28143_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28143_end_0 = const()[name = tensor("op_28143_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_28143_end_mask_0 = const()[name = tensor("op_28143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28143_cast_fp16 = slice_by_index(begin = var_28143_begin_0, end = var_28143_end_0, end_mask = var_28143_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28143_cast_fp16")]; + tensor var_28147_begin_0 = const()[name = tensor("op_28147_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28147_end_0 = const()[name = tensor("op_28147_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_28147_end_mask_0 = const()[name = tensor("op_28147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28147_cast_fp16 = slice_by_index(begin = var_28147_begin_0, end = var_28147_end_0, end_mask = var_28147_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28147_cast_fp16")]; + tensor var_28151_begin_0 = const()[name = tensor("op_28151_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28151_end_0 = const()[name = tensor("op_28151_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_28151_end_mask_0 = const()[name = tensor("op_28151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28151_cast_fp16 = slice_by_index(begin = var_28151_begin_0, end = var_28151_end_0, end_mask = var_28151_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28151_cast_fp16")]; + tensor var_28155_begin_0 = const()[name = tensor("op_28155_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28155_end_0 = const()[name = tensor("op_28155_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_28155_end_mask_0 = const()[name = tensor("op_28155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28155_cast_fp16 = slice_by_index(begin = var_28155_begin_0, end = var_28155_end_0, end_mask = var_28155_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28155_cast_fp16")]; + tensor var_28159_begin_0 = const()[name = tensor("op_28159_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28159_end_0 = const()[name = tensor("op_28159_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_28159_end_mask_0 = const()[name = tensor("op_28159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28159_cast_fp16 = slice_by_index(begin = var_28159_begin_0, end = var_28159_end_0, end_mask = var_28159_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28159_cast_fp16")]; + tensor var_28163_begin_0 = const()[name = tensor("op_28163_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28163_end_0 = const()[name = tensor("op_28163_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_28163_end_mask_0 = const()[name = tensor("op_28163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28163_cast_fp16 = slice_by_index(begin = var_28163_begin_0, end = var_28163_end_0, end_mask = var_28163_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28163_cast_fp16")]; + tensor var_28167_begin_0 = const()[name = tensor("op_28167_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28167_end_0 = const()[name = tensor("op_28167_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_28167_end_mask_0 = const()[name = tensor("op_28167_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28167_cast_fp16 = slice_by_index(begin = var_28167_begin_0, end = var_28167_end_0, end_mask = var_28167_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28167_cast_fp16")]; + tensor var_28171_begin_0 = const()[name = tensor("op_28171_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28171_end_0 = const()[name = tensor("op_28171_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_28171_end_mask_0 = const()[name = tensor("op_28171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28171_cast_fp16 = slice_by_index(begin = var_28171_begin_0, end = var_28171_end_0, end_mask = var_28171_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28171_cast_fp16")]; + tensor var_28175_begin_0 = const()[name = tensor("op_28175_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28175_end_0 = const()[name = tensor("op_28175_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_28175_end_mask_0 = const()[name = tensor("op_28175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28175_cast_fp16 = slice_by_index(begin = var_28175_begin_0, end = var_28175_end_0, end_mask = var_28175_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28175_cast_fp16")]; + tensor var_28179_begin_0 = const()[name = tensor("op_28179_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28179_end_0 = const()[name = tensor("op_28179_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_28179_end_mask_0 = const()[name = tensor("op_28179_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28179_cast_fp16 = slice_by_index(begin = var_28179_begin_0, end = var_28179_end_0, end_mask = var_28179_end_mask_0, x = query_35_cast_fp16)[name = tensor("op_28179_cast_fp16")]; + tensor var_28188_begin_0 = const()[name = tensor("op_28188_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28188_end_0 = const()[name = tensor("op_28188_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28188_end_mask_0 = const()[name = tensor("op_28188_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28188_cast_fp16 = slice_by_index(begin = var_28188_begin_0, end = var_28188_end_0, end_mask = var_28188_end_mask_0, x = var_28103_cast_fp16)[name = tensor("op_28188_cast_fp16")]; + tensor var_28195_begin_0 = const()[name = tensor("op_28195_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28195_end_0 = const()[name = tensor("op_28195_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28195_end_mask_0 = const()[name = tensor("op_28195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28195_cast_fp16 = slice_by_index(begin = var_28195_begin_0, end = var_28195_end_0, end_mask = var_28195_end_mask_0, x = var_28103_cast_fp16)[name = tensor("op_28195_cast_fp16")]; + tensor var_28202_begin_0 = const()[name = tensor("op_28202_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28202_end_0 = const()[name = tensor("op_28202_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28202_end_mask_0 = const()[name = tensor("op_28202_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28202_cast_fp16 = slice_by_index(begin = var_28202_begin_0, end = var_28202_end_0, end_mask = var_28202_end_mask_0, x = var_28103_cast_fp16)[name = tensor("op_28202_cast_fp16")]; + tensor var_28209_begin_0 = const()[name = tensor("op_28209_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28209_end_0 = const()[name = tensor("op_28209_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28209_end_mask_0 = const()[name = tensor("op_28209_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28209_cast_fp16 = slice_by_index(begin = var_28209_begin_0, end = var_28209_end_0, end_mask = var_28209_end_mask_0, x = var_28103_cast_fp16)[name = tensor("op_28209_cast_fp16")]; + tensor var_28216_begin_0 = const()[name = tensor("op_28216_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28216_end_0 = const()[name = tensor("op_28216_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28216_end_mask_0 = const()[name = tensor("op_28216_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28216_cast_fp16 = slice_by_index(begin = var_28216_begin_0, end = var_28216_end_0, end_mask = var_28216_end_mask_0, x = var_28107_cast_fp16)[name = tensor("op_28216_cast_fp16")]; + tensor var_28223_begin_0 = const()[name = tensor("op_28223_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28223_end_0 = const()[name = tensor("op_28223_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28223_end_mask_0 = const()[name = tensor("op_28223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28223_cast_fp16 = slice_by_index(begin = var_28223_begin_0, end = var_28223_end_0, end_mask = var_28223_end_mask_0, x = var_28107_cast_fp16)[name = tensor("op_28223_cast_fp16")]; + tensor var_28230_begin_0 = const()[name = tensor("op_28230_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28230_end_0 = const()[name = tensor("op_28230_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28230_end_mask_0 = const()[name = tensor("op_28230_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28230_cast_fp16 = slice_by_index(begin = var_28230_begin_0, end = var_28230_end_0, end_mask = var_28230_end_mask_0, x = var_28107_cast_fp16)[name = tensor("op_28230_cast_fp16")]; + tensor var_28237_begin_0 = const()[name = tensor("op_28237_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28237_end_0 = const()[name = tensor("op_28237_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28237_end_mask_0 = const()[name = tensor("op_28237_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28237_cast_fp16 = slice_by_index(begin = var_28237_begin_0, end = var_28237_end_0, end_mask = var_28237_end_mask_0, x = var_28107_cast_fp16)[name = tensor("op_28237_cast_fp16")]; + tensor var_28244_begin_0 = const()[name = tensor("op_28244_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28244_end_0 = const()[name = tensor("op_28244_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28244_end_mask_0 = const()[name = tensor("op_28244_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28244_cast_fp16 = slice_by_index(begin = var_28244_begin_0, end = var_28244_end_0, end_mask = var_28244_end_mask_0, x = var_28111_cast_fp16)[name = tensor("op_28244_cast_fp16")]; + tensor var_28251_begin_0 = const()[name = tensor("op_28251_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28251_end_0 = const()[name = tensor("op_28251_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28251_end_mask_0 = const()[name = tensor("op_28251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28251_cast_fp16 = slice_by_index(begin = var_28251_begin_0, end = var_28251_end_0, end_mask = var_28251_end_mask_0, x = var_28111_cast_fp16)[name = tensor("op_28251_cast_fp16")]; + tensor var_28258_begin_0 = const()[name = tensor("op_28258_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28258_end_0 = const()[name = tensor("op_28258_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28258_end_mask_0 = const()[name = tensor("op_28258_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28258_cast_fp16 = slice_by_index(begin = var_28258_begin_0, end = var_28258_end_0, end_mask = var_28258_end_mask_0, x = var_28111_cast_fp16)[name = tensor("op_28258_cast_fp16")]; + tensor var_28265_begin_0 = const()[name = tensor("op_28265_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28265_end_0 = const()[name = tensor("op_28265_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28265_end_mask_0 = const()[name = tensor("op_28265_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28265_cast_fp16 = slice_by_index(begin = var_28265_begin_0, end = var_28265_end_0, end_mask = var_28265_end_mask_0, x = var_28111_cast_fp16)[name = tensor("op_28265_cast_fp16")]; + tensor var_28272_begin_0 = const()[name = tensor("op_28272_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28272_end_0 = const()[name = tensor("op_28272_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28272_end_mask_0 = const()[name = tensor("op_28272_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28272_cast_fp16 = slice_by_index(begin = var_28272_begin_0, end = var_28272_end_0, end_mask = var_28272_end_mask_0, x = var_28115_cast_fp16)[name = tensor("op_28272_cast_fp16")]; + tensor var_28279_begin_0 = const()[name = tensor("op_28279_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28279_end_0 = const()[name = tensor("op_28279_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28279_end_mask_0 = const()[name = tensor("op_28279_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28279_cast_fp16 = slice_by_index(begin = var_28279_begin_0, end = var_28279_end_0, end_mask = var_28279_end_mask_0, x = var_28115_cast_fp16)[name = tensor("op_28279_cast_fp16")]; + tensor var_28286_begin_0 = const()[name = tensor("op_28286_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28286_end_0 = const()[name = tensor("op_28286_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28286_end_mask_0 = const()[name = tensor("op_28286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28286_cast_fp16 = slice_by_index(begin = var_28286_begin_0, end = var_28286_end_0, end_mask = var_28286_end_mask_0, x = var_28115_cast_fp16)[name = tensor("op_28286_cast_fp16")]; + tensor var_28293_begin_0 = const()[name = tensor("op_28293_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28293_end_0 = const()[name = tensor("op_28293_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28293_end_mask_0 = const()[name = tensor("op_28293_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28293_cast_fp16 = slice_by_index(begin = var_28293_begin_0, end = var_28293_end_0, end_mask = var_28293_end_mask_0, x = var_28115_cast_fp16)[name = tensor("op_28293_cast_fp16")]; + tensor var_28300_begin_0 = const()[name = tensor("op_28300_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28300_end_0 = const()[name = tensor("op_28300_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28300_end_mask_0 = const()[name = tensor("op_28300_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28300_cast_fp16 = slice_by_index(begin = var_28300_begin_0, end = var_28300_end_0, end_mask = var_28300_end_mask_0, x = var_28119_cast_fp16)[name = tensor("op_28300_cast_fp16")]; + tensor var_28307_begin_0 = const()[name = tensor("op_28307_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28307_end_0 = const()[name = tensor("op_28307_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28307_end_mask_0 = const()[name = tensor("op_28307_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28307_cast_fp16 = slice_by_index(begin = var_28307_begin_0, end = var_28307_end_0, end_mask = var_28307_end_mask_0, x = var_28119_cast_fp16)[name = tensor("op_28307_cast_fp16")]; + tensor var_28314_begin_0 = const()[name = tensor("op_28314_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28314_end_0 = const()[name = tensor("op_28314_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28314_end_mask_0 = const()[name = tensor("op_28314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28314_cast_fp16 = slice_by_index(begin = var_28314_begin_0, end = var_28314_end_0, end_mask = var_28314_end_mask_0, x = var_28119_cast_fp16)[name = tensor("op_28314_cast_fp16")]; + tensor var_28321_begin_0 = const()[name = tensor("op_28321_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28321_end_0 = const()[name = tensor("op_28321_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28321_end_mask_0 = const()[name = tensor("op_28321_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28321_cast_fp16 = slice_by_index(begin = var_28321_begin_0, end = var_28321_end_0, end_mask = var_28321_end_mask_0, x = var_28119_cast_fp16)[name = tensor("op_28321_cast_fp16")]; + tensor var_28328_begin_0 = const()[name = tensor("op_28328_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28328_end_0 = const()[name = tensor("op_28328_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28328_end_mask_0 = const()[name = tensor("op_28328_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28328_cast_fp16 = slice_by_index(begin = var_28328_begin_0, end = var_28328_end_0, end_mask = var_28328_end_mask_0, x = var_28123_cast_fp16)[name = tensor("op_28328_cast_fp16")]; + tensor var_28335_begin_0 = const()[name = tensor("op_28335_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28335_end_0 = const()[name = tensor("op_28335_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28335_end_mask_0 = const()[name = tensor("op_28335_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28335_cast_fp16 = slice_by_index(begin = var_28335_begin_0, end = var_28335_end_0, end_mask = var_28335_end_mask_0, x = var_28123_cast_fp16)[name = tensor("op_28335_cast_fp16")]; + tensor var_28342_begin_0 = const()[name = tensor("op_28342_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28342_end_0 = const()[name = tensor("op_28342_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28342_end_mask_0 = const()[name = tensor("op_28342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28342_cast_fp16 = slice_by_index(begin = var_28342_begin_0, end = var_28342_end_0, end_mask = var_28342_end_mask_0, x = var_28123_cast_fp16)[name = tensor("op_28342_cast_fp16")]; + tensor var_28349_begin_0 = const()[name = tensor("op_28349_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28349_end_0 = const()[name = tensor("op_28349_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28349_end_mask_0 = const()[name = tensor("op_28349_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28349_cast_fp16 = slice_by_index(begin = var_28349_begin_0, end = var_28349_end_0, end_mask = var_28349_end_mask_0, x = var_28123_cast_fp16)[name = tensor("op_28349_cast_fp16")]; + tensor var_28356_begin_0 = const()[name = tensor("op_28356_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28356_end_0 = const()[name = tensor("op_28356_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28356_end_mask_0 = const()[name = tensor("op_28356_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28356_cast_fp16 = slice_by_index(begin = var_28356_begin_0, end = var_28356_end_0, end_mask = var_28356_end_mask_0, x = var_28127_cast_fp16)[name = tensor("op_28356_cast_fp16")]; + tensor var_28363_begin_0 = const()[name = tensor("op_28363_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28363_end_0 = const()[name = tensor("op_28363_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28363_end_mask_0 = const()[name = tensor("op_28363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28363_cast_fp16 = slice_by_index(begin = var_28363_begin_0, end = var_28363_end_0, end_mask = var_28363_end_mask_0, x = var_28127_cast_fp16)[name = tensor("op_28363_cast_fp16")]; + tensor var_28370_begin_0 = const()[name = tensor("op_28370_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28370_end_0 = const()[name = tensor("op_28370_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28370_end_mask_0 = const()[name = tensor("op_28370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28370_cast_fp16 = slice_by_index(begin = var_28370_begin_0, end = var_28370_end_0, end_mask = var_28370_end_mask_0, x = var_28127_cast_fp16)[name = tensor("op_28370_cast_fp16")]; + tensor var_28377_begin_0 = const()[name = tensor("op_28377_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28377_end_0 = const()[name = tensor("op_28377_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28377_end_mask_0 = const()[name = tensor("op_28377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28377_cast_fp16 = slice_by_index(begin = var_28377_begin_0, end = var_28377_end_0, end_mask = var_28377_end_mask_0, x = var_28127_cast_fp16)[name = tensor("op_28377_cast_fp16")]; + tensor var_28384_begin_0 = const()[name = tensor("op_28384_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28384_end_0 = const()[name = tensor("op_28384_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28384_end_mask_0 = const()[name = tensor("op_28384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28384_cast_fp16 = slice_by_index(begin = var_28384_begin_0, end = var_28384_end_0, end_mask = var_28384_end_mask_0, x = var_28131_cast_fp16)[name = tensor("op_28384_cast_fp16")]; + tensor var_28391_begin_0 = const()[name = tensor("op_28391_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28391_end_0 = const()[name = tensor("op_28391_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28391_end_mask_0 = const()[name = tensor("op_28391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28391_cast_fp16 = slice_by_index(begin = var_28391_begin_0, end = var_28391_end_0, end_mask = var_28391_end_mask_0, x = var_28131_cast_fp16)[name = tensor("op_28391_cast_fp16")]; + tensor var_28398_begin_0 = const()[name = tensor("op_28398_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28398_end_0 = const()[name = tensor("op_28398_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28398_end_mask_0 = const()[name = tensor("op_28398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28398_cast_fp16 = slice_by_index(begin = var_28398_begin_0, end = var_28398_end_0, end_mask = var_28398_end_mask_0, x = var_28131_cast_fp16)[name = tensor("op_28398_cast_fp16")]; + tensor var_28405_begin_0 = const()[name = tensor("op_28405_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28405_end_0 = const()[name = tensor("op_28405_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28405_end_mask_0 = const()[name = tensor("op_28405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28405_cast_fp16 = slice_by_index(begin = var_28405_begin_0, end = var_28405_end_0, end_mask = var_28405_end_mask_0, x = var_28131_cast_fp16)[name = tensor("op_28405_cast_fp16")]; + tensor var_28412_begin_0 = const()[name = tensor("op_28412_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28412_end_0 = const()[name = tensor("op_28412_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28412_end_mask_0 = const()[name = tensor("op_28412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28412_cast_fp16 = slice_by_index(begin = var_28412_begin_0, end = var_28412_end_0, end_mask = var_28412_end_mask_0, x = var_28135_cast_fp16)[name = tensor("op_28412_cast_fp16")]; + tensor var_28419_begin_0 = const()[name = tensor("op_28419_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28419_end_0 = const()[name = tensor("op_28419_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28419_end_mask_0 = const()[name = tensor("op_28419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28419_cast_fp16 = slice_by_index(begin = var_28419_begin_0, end = var_28419_end_0, end_mask = var_28419_end_mask_0, x = var_28135_cast_fp16)[name = tensor("op_28419_cast_fp16")]; + tensor var_28426_begin_0 = const()[name = tensor("op_28426_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28426_end_0 = const()[name = tensor("op_28426_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28426_end_mask_0 = const()[name = tensor("op_28426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28426_cast_fp16 = slice_by_index(begin = var_28426_begin_0, end = var_28426_end_0, end_mask = var_28426_end_mask_0, x = var_28135_cast_fp16)[name = tensor("op_28426_cast_fp16")]; + tensor var_28433_begin_0 = const()[name = tensor("op_28433_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28433_end_0 = const()[name = tensor("op_28433_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28433_end_mask_0 = const()[name = tensor("op_28433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28433_cast_fp16 = slice_by_index(begin = var_28433_begin_0, end = var_28433_end_0, end_mask = var_28433_end_mask_0, x = var_28135_cast_fp16)[name = tensor("op_28433_cast_fp16")]; + tensor var_28440_begin_0 = const()[name = tensor("op_28440_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28440_end_0 = const()[name = tensor("op_28440_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28440_end_mask_0 = const()[name = tensor("op_28440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28440_cast_fp16 = slice_by_index(begin = var_28440_begin_0, end = var_28440_end_0, end_mask = var_28440_end_mask_0, x = var_28139_cast_fp16)[name = tensor("op_28440_cast_fp16")]; + tensor var_28447_begin_0 = const()[name = tensor("op_28447_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28447_end_0 = const()[name = tensor("op_28447_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28447_end_mask_0 = const()[name = tensor("op_28447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28447_cast_fp16 = slice_by_index(begin = var_28447_begin_0, end = var_28447_end_0, end_mask = var_28447_end_mask_0, x = var_28139_cast_fp16)[name = tensor("op_28447_cast_fp16")]; + tensor var_28454_begin_0 = const()[name = tensor("op_28454_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28454_end_0 = const()[name = tensor("op_28454_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28454_end_mask_0 = const()[name = tensor("op_28454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28454_cast_fp16 = slice_by_index(begin = var_28454_begin_0, end = var_28454_end_0, end_mask = var_28454_end_mask_0, x = var_28139_cast_fp16)[name = tensor("op_28454_cast_fp16")]; + tensor var_28461_begin_0 = const()[name = tensor("op_28461_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28461_end_0 = const()[name = tensor("op_28461_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28461_end_mask_0 = const()[name = tensor("op_28461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28461_cast_fp16 = slice_by_index(begin = var_28461_begin_0, end = var_28461_end_0, end_mask = var_28461_end_mask_0, x = var_28139_cast_fp16)[name = tensor("op_28461_cast_fp16")]; + tensor var_28468_begin_0 = const()[name = tensor("op_28468_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28468_end_0 = const()[name = tensor("op_28468_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28468_end_mask_0 = const()[name = tensor("op_28468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28468_cast_fp16 = slice_by_index(begin = var_28468_begin_0, end = var_28468_end_0, end_mask = var_28468_end_mask_0, x = var_28143_cast_fp16)[name = tensor("op_28468_cast_fp16")]; + tensor var_28475_begin_0 = const()[name = tensor("op_28475_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28475_end_0 = const()[name = tensor("op_28475_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28475_end_mask_0 = const()[name = tensor("op_28475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28475_cast_fp16 = slice_by_index(begin = var_28475_begin_0, end = var_28475_end_0, end_mask = var_28475_end_mask_0, x = var_28143_cast_fp16)[name = tensor("op_28475_cast_fp16")]; + tensor var_28482_begin_0 = const()[name = tensor("op_28482_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28482_end_0 = const()[name = tensor("op_28482_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28482_end_mask_0 = const()[name = tensor("op_28482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28482_cast_fp16 = slice_by_index(begin = var_28482_begin_0, end = var_28482_end_0, end_mask = var_28482_end_mask_0, x = var_28143_cast_fp16)[name = tensor("op_28482_cast_fp16")]; + tensor var_28489_begin_0 = const()[name = tensor("op_28489_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28489_end_0 = const()[name = tensor("op_28489_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28489_end_mask_0 = const()[name = tensor("op_28489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28489_cast_fp16 = slice_by_index(begin = var_28489_begin_0, end = var_28489_end_0, end_mask = var_28489_end_mask_0, x = var_28143_cast_fp16)[name = tensor("op_28489_cast_fp16")]; + tensor var_28496_begin_0 = const()[name = tensor("op_28496_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28496_end_0 = const()[name = tensor("op_28496_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28496_end_mask_0 = const()[name = tensor("op_28496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28496_cast_fp16 = slice_by_index(begin = var_28496_begin_0, end = var_28496_end_0, end_mask = var_28496_end_mask_0, x = var_28147_cast_fp16)[name = tensor("op_28496_cast_fp16")]; + tensor var_28503_begin_0 = const()[name = tensor("op_28503_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28503_end_0 = const()[name = tensor("op_28503_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28503_end_mask_0 = const()[name = tensor("op_28503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28503_cast_fp16 = slice_by_index(begin = var_28503_begin_0, end = var_28503_end_0, end_mask = var_28503_end_mask_0, x = var_28147_cast_fp16)[name = tensor("op_28503_cast_fp16")]; + tensor var_28510_begin_0 = const()[name = tensor("op_28510_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28510_end_0 = const()[name = tensor("op_28510_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28510_end_mask_0 = const()[name = tensor("op_28510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28510_cast_fp16 = slice_by_index(begin = var_28510_begin_0, end = var_28510_end_0, end_mask = var_28510_end_mask_0, x = var_28147_cast_fp16)[name = tensor("op_28510_cast_fp16")]; + tensor var_28517_begin_0 = const()[name = tensor("op_28517_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28517_end_0 = const()[name = tensor("op_28517_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28517_end_mask_0 = const()[name = tensor("op_28517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28517_cast_fp16 = slice_by_index(begin = var_28517_begin_0, end = var_28517_end_0, end_mask = var_28517_end_mask_0, x = var_28147_cast_fp16)[name = tensor("op_28517_cast_fp16")]; + tensor var_28524_begin_0 = const()[name = tensor("op_28524_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28524_end_0 = const()[name = tensor("op_28524_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28524_end_mask_0 = const()[name = tensor("op_28524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28524_cast_fp16 = slice_by_index(begin = var_28524_begin_0, end = var_28524_end_0, end_mask = var_28524_end_mask_0, x = var_28151_cast_fp16)[name = tensor("op_28524_cast_fp16")]; + tensor var_28531_begin_0 = const()[name = tensor("op_28531_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28531_end_0 = const()[name = tensor("op_28531_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28531_end_mask_0 = const()[name = tensor("op_28531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28531_cast_fp16 = slice_by_index(begin = var_28531_begin_0, end = var_28531_end_0, end_mask = var_28531_end_mask_0, x = var_28151_cast_fp16)[name = tensor("op_28531_cast_fp16")]; + tensor var_28538_begin_0 = const()[name = tensor("op_28538_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28538_end_0 = const()[name = tensor("op_28538_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28538_end_mask_0 = const()[name = tensor("op_28538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28538_cast_fp16 = slice_by_index(begin = var_28538_begin_0, end = var_28538_end_0, end_mask = var_28538_end_mask_0, x = var_28151_cast_fp16)[name = tensor("op_28538_cast_fp16")]; + tensor var_28545_begin_0 = const()[name = tensor("op_28545_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28545_end_0 = const()[name = tensor("op_28545_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28545_end_mask_0 = const()[name = tensor("op_28545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28545_cast_fp16 = slice_by_index(begin = var_28545_begin_0, end = var_28545_end_0, end_mask = var_28545_end_mask_0, x = var_28151_cast_fp16)[name = tensor("op_28545_cast_fp16")]; + tensor var_28552_begin_0 = const()[name = tensor("op_28552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28552_end_0 = const()[name = tensor("op_28552_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28552_end_mask_0 = const()[name = tensor("op_28552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28552_cast_fp16 = slice_by_index(begin = var_28552_begin_0, end = var_28552_end_0, end_mask = var_28552_end_mask_0, x = var_28155_cast_fp16)[name = tensor("op_28552_cast_fp16")]; + tensor var_28559_begin_0 = const()[name = tensor("op_28559_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28559_end_0 = const()[name = tensor("op_28559_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28559_end_mask_0 = const()[name = tensor("op_28559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28559_cast_fp16 = slice_by_index(begin = var_28559_begin_0, end = var_28559_end_0, end_mask = var_28559_end_mask_0, x = var_28155_cast_fp16)[name = tensor("op_28559_cast_fp16")]; + tensor var_28566_begin_0 = const()[name = tensor("op_28566_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28566_end_0 = const()[name = tensor("op_28566_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28566_end_mask_0 = const()[name = tensor("op_28566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28566_cast_fp16 = slice_by_index(begin = var_28566_begin_0, end = var_28566_end_0, end_mask = var_28566_end_mask_0, x = var_28155_cast_fp16)[name = tensor("op_28566_cast_fp16")]; + tensor var_28573_begin_0 = const()[name = tensor("op_28573_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28573_end_0 = const()[name = tensor("op_28573_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28573_end_mask_0 = const()[name = tensor("op_28573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28573_cast_fp16 = slice_by_index(begin = var_28573_begin_0, end = var_28573_end_0, end_mask = var_28573_end_mask_0, x = var_28155_cast_fp16)[name = tensor("op_28573_cast_fp16")]; + tensor var_28580_begin_0 = const()[name = tensor("op_28580_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28580_end_0 = const()[name = tensor("op_28580_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28580_end_mask_0 = const()[name = tensor("op_28580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28580_cast_fp16 = slice_by_index(begin = var_28580_begin_0, end = var_28580_end_0, end_mask = var_28580_end_mask_0, x = var_28159_cast_fp16)[name = tensor("op_28580_cast_fp16")]; + tensor var_28587_begin_0 = const()[name = tensor("op_28587_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28587_end_0 = const()[name = tensor("op_28587_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28587_end_mask_0 = const()[name = tensor("op_28587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28587_cast_fp16 = slice_by_index(begin = var_28587_begin_0, end = var_28587_end_0, end_mask = var_28587_end_mask_0, x = var_28159_cast_fp16)[name = tensor("op_28587_cast_fp16")]; + tensor var_28594_begin_0 = const()[name = tensor("op_28594_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28594_end_0 = const()[name = tensor("op_28594_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28594_end_mask_0 = const()[name = tensor("op_28594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28594_cast_fp16 = slice_by_index(begin = var_28594_begin_0, end = var_28594_end_0, end_mask = var_28594_end_mask_0, x = var_28159_cast_fp16)[name = tensor("op_28594_cast_fp16")]; + tensor var_28601_begin_0 = const()[name = tensor("op_28601_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28601_end_0 = const()[name = tensor("op_28601_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28601_end_mask_0 = const()[name = tensor("op_28601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28601_cast_fp16 = slice_by_index(begin = var_28601_begin_0, end = var_28601_end_0, end_mask = var_28601_end_mask_0, x = var_28159_cast_fp16)[name = tensor("op_28601_cast_fp16")]; + tensor var_28608_begin_0 = const()[name = tensor("op_28608_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28608_end_0 = const()[name = tensor("op_28608_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28608_end_mask_0 = const()[name = tensor("op_28608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28608_cast_fp16 = slice_by_index(begin = var_28608_begin_0, end = var_28608_end_0, end_mask = var_28608_end_mask_0, x = var_28163_cast_fp16)[name = tensor("op_28608_cast_fp16")]; + tensor var_28615_begin_0 = const()[name = tensor("op_28615_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28615_end_0 = const()[name = tensor("op_28615_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28615_end_mask_0 = const()[name = tensor("op_28615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28615_cast_fp16 = slice_by_index(begin = var_28615_begin_0, end = var_28615_end_0, end_mask = var_28615_end_mask_0, x = var_28163_cast_fp16)[name = tensor("op_28615_cast_fp16")]; + tensor var_28622_begin_0 = const()[name = tensor("op_28622_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28622_end_0 = const()[name = tensor("op_28622_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28622_end_mask_0 = const()[name = tensor("op_28622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28622_cast_fp16 = slice_by_index(begin = var_28622_begin_0, end = var_28622_end_0, end_mask = var_28622_end_mask_0, x = var_28163_cast_fp16)[name = tensor("op_28622_cast_fp16")]; + tensor var_28629_begin_0 = const()[name = tensor("op_28629_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28629_end_0 = const()[name = tensor("op_28629_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28629_end_mask_0 = const()[name = tensor("op_28629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28629_cast_fp16 = slice_by_index(begin = var_28629_begin_0, end = var_28629_end_0, end_mask = var_28629_end_mask_0, x = var_28163_cast_fp16)[name = tensor("op_28629_cast_fp16")]; + tensor var_28636_begin_0 = const()[name = tensor("op_28636_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28636_end_0 = const()[name = tensor("op_28636_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28636_end_mask_0 = const()[name = tensor("op_28636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28636_cast_fp16 = slice_by_index(begin = var_28636_begin_0, end = var_28636_end_0, end_mask = var_28636_end_mask_0, x = var_28167_cast_fp16)[name = tensor("op_28636_cast_fp16")]; + tensor var_28643_begin_0 = const()[name = tensor("op_28643_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28643_end_0 = const()[name = tensor("op_28643_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28643_end_mask_0 = const()[name = tensor("op_28643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28643_cast_fp16 = slice_by_index(begin = var_28643_begin_0, end = var_28643_end_0, end_mask = var_28643_end_mask_0, x = var_28167_cast_fp16)[name = tensor("op_28643_cast_fp16")]; + tensor var_28650_begin_0 = const()[name = tensor("op_28650_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28650_end_0 = const()[name = tensor("op_28650_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28650_end_mask_0 = const()[name = tensor("op_28650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28650_cast_fp16 = slice_by_index(begin = var_28650_begin_0, end = var_28650_end_0, end_mask = var_28650_end_mask_0, x = var_28167_cast_fp16)[name = tensor("op_28650_cast_fp16")]; + tensor var_28657_begin_0 = const()[name = tensor("op_28657_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28657_end_0 = const()[name = tensor("op_28657_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28657_end_mask_0 = const()[name = tensor("op_28657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28657_cast_fp16 = slice_by_index(begin = var_28657_begin_0, end = var_28657_end_0, end_mask = var_28657_end_mask_0, x = var_28167_cast_fp16)[name = tensor("op_28657_cast_fp16")]; + tensor var_28664_begin_0 = const()[name = tensor("op_28664_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28664_end_0 = const()[name = tensor("op_28664_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28664_end_mask_0 = const()[name = tensor("op_28664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28664_cast_fp16 = slice_by_index(begin = var_28664_begin_0, end = var_28664_end_0, end_mask = var_28664_end_mask_0, x = var_28171_cast_fp16)[name = tensor("op_28664_cast_fp16")]; + tensor var_28671_begin_0 = const()[name = tensor("op_28671_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28671_end_0 = const()[name = tensor("op_28671_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28671_end_mask_0 = const()[name = tensor("op_28671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28671_cast_fp16 = slice_by_index(begin = var_28671_begin_0, end = var_28671_end_0, end_mask = var_28671_end_mask_0, x = var_28171_cast_fp16)[name = tensor("op_28671_cast_fp16")]; + tensor var_28678_begin_0 = const()[name = tensor("op_28678_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28678_end_0 = const()[name = tensor("op_28678_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28678_end_mask_0 = const()[name = tensor("op_28678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28678_cast_fp16 = slice_by_index(begin = var_28678_begin_0, end = var_28678_end_0, end_mask = var_28678_end_mask_0, x = var_28171_cast_fp16)[name = tensor("op_28678_cast_fp16")]; + tensor var_28685_begin_0 = const()[name = tensor("op_28685_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28685_end_0 = const()[name = tensor("op_28685_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28685_end_mask_0 = const()[name = tensor("op_28685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28685_cast_fp16 = slice_by_index(begin = var_28685_begin_0, end = var_28685_end_0, end_mask = var_28685_end_mask_0, x = var_28171_cast_fp16)[name = tensor("op_28685_cast_fp16")]; + tensor var_28692_begin_0 = const()[name = tensor("op_28692_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28692_end_0 = const()[name = tensor("op_28692_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28692_end_mask_0 = const()[name = tensor("op_28692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28692_cast_fp16 = slice_by_index(begin = var_28692_begin_0, end = var_28692_end_0, end_mask = var_28692_end_mask_0, x = var_28175_cast_fp16)[name = tensor("op_28692_cast_fp16")]; + tensor var_28699_begin_0 = const()[name = tensor("op_28699_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28699_end_0 = const()[name = tensor("op_28699_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28699_end_mask_0 = const()[name = tensor("op_28699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28699_cast_fp16 = slice_by_index(begin = var_28699_begin_0, end = var_28699_end_0, end_mask = var_28699_end_mask_0, x = var_28175_cast_fp16)[name = tensor("op_28699_cast_fp16")]; + tensor var_28706_begin_0 = const()[name = tensor("op_28706_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28706_end_0 = const()[name = tensor("op_28706_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28706_end_mask_0 = const()[name = tensor("op_28706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28706_cast_fp16 = slice_by_index(begin = var_28706_begin_0, end = var_28706_end_0, end_mask = var_28706_end_mask_0, x = var_28175_cast_fp16)[name = tensor("op_28706_cast_fp16")]; + tensor var_28713_begin_0 = const()[name = tensor("op_28713_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28713_end_0 = const()[name = tensor("op_28713_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28713_end_mask_0 = const()[name = tensor("op_28713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28713_cast_fp16 = slice_by_index(begin = var_28713_begin_0, end = var_28713_end_0, end_mask = var_28713_end_mask_0, x = var_28175_cast_fp16)[name = tensor("op_28713_cast_fp16")]; + tensor var_28720_begin_0 = const()[name = tensor("op_28720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28720_end_0 = const()[name = tensor("op_28720_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_28720_end_mask_0 = const()[name = tensor("op_28720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28720_cast_fp16 = slice_by_index(begin = var_28720_begin_0, end = var_28720_end_0, end_mask = var_28720_end_mask_0, x = var_28179_cast_fp16)[name = tensor("op_28720_cast_fp16")]; + tensor var_28727_begin_0 = const()[name = tensor("op_28727_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_28727_end_0 = const()[name = tensor("op_28727_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_28727_end_mask_0 = const()[name = tensor("op_28727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28727_cast_fp16 = slice_by_index(begin = var_28727_begin_0, end = var_28727_end_0, end_mask = var_28727_end_mask_0, x = var_28179_cast_fp16)[name = tensor("op_28727_cast_fp16")]; + tensor var_28734_begin_0 = const()[name = tensor("op_28734_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_28734_end_0 = const()[name = tensor("op_28734_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_28734_end_mask_0 = const()[name = tensor("op_28734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28734_cast_fp16 = slice_by_index(begin = var_28734_begin_0, end = var_28734_end_0, end_mask = var_28734_end_mask_0, x = var_28179_cast_fp16)[name = tensor("op_28734_cast_fp16")]; + tensor var_28741_begin_0 = const()[name = tensor("op_28741_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_28741_end_0 = const()[name = tensor("op_28741_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28741_end_mask_0 = const()[name = tensor("op_28741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28741_cast_fp16 = slice_by_index(begin = var_28741_begin_0, end = var_28741_end_0, end_mask = var_28741_end_mask_0, x = var_28179_cast_fp16)[name = tensor("op_28741_cast_fp16")]; + tensor k_35_perm_0 = const()[name = tensor("k_35_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_28746_begin_0 = const()[name = tensor("op_28746_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28746_end_0 = const()[name = tensor("op_28746_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_28746_end_mask_0 = const()[name = tensor("op_28746_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_14 = transpose(perm = k_35_perm_0, x = key_35_cast_fp16)[name = tensor("transpose_14")]; + tensor var_28746_cast_fp16 = slice_by_index(begin = var_28746_begin_0, end = var_28746_end_0, end_mask = var_28746_end_mask_0, x = transpose_14)[name = tensor("op_28746_cast_fp16")]; + tensor var_28750_begin_0 = const()[name = tensor("op_28750_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_28750_end_0 = const()[name = tensor("op_28750_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_28750_end_mask_0 = const()[name = tensor("op_28750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28750_cast_fp16 = slice_by_index(begin = var_28750_begin_0, end = var_28750_end_0, end_mask = var_28750_end_mask_0, x = transpose_14)[name = tensor("op_28750_cast_fp16")]; + tensor var_28754_begin_0 = const()[name = tensor("op_28754_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_28754_end_0 = const()[name = tensor("op_28754_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_28754_end_mask_0 = const()[name = tensor("op_28754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28754_cast_fp16 = slice_by_index(begin = var_28754_begin_0, end = var_28754_end_0, end_mask = var_28754_end_mask_0, x = transpose_14)[name = tensor("op_28754_cast_fp16")]; + tensor var_28758_begin_0 = const()[name = tensor("op_28758_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_28758_end_0 = const()[name = tensor("op_28758_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_28758_end_mask_0 = const()[name = tensor("op_28758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28758_cast_fp16 = slice_by_index(begin = var_28758_begin_0, end = var_28758_end_0, end_mask = var_28758_end_mask_0, x = transpose_14)[name = tensor("op_28758_cast_fp16")]; + tensor var_28762_begin_0 = const()[name = tensor("op_28762_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_28762_end_0 = const()[name = tensor("op_28762_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_28762_end_mask_0 = const()[name = tensor("op_28762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28762_cast_fp16 = slice_by_index(begin = var_28762_begin_0, end = var_28762_end_0, end_mask = var_28762_end_mask_0, x = transpose_14)[name = tensor("op_28762_cast_fp16")]; + tensor var_28766_begin_0 = const()[name = tensor("op_28766_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_28766_end_0 = const()[name = tensor("op_28766_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_28766_end_mask_0 = const()[name = tensor("op_28766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28766_cast_fp16 = slice_by_index(begin = var_28766_begin_0, end = var_28766_end_0, end_mask = var_28766_end_mask_0, x = transpose_14)[name = tensor("op_28766_cast_fp16")]; + tensor var_28770_begin_0 = const()[name = tensor("op_28770_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_28770_end_0 = const()[name = tensor("op_28770_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_28770_end_mask_0 = const()[name = tensor("op_28770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28770_cast_fp16 = slice_by_index(begin = var_28770_begin_0, end = var_28770_end_0, end_mask = var_28770_end_mask_0, x = transpose_14)[name = tensor("op_28770_cast_fp16")]; + tensor var_28774_begin_0 = const()[name = tensor("op_28774_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_28774_end_0 = const()[name = tensor("op_28774_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_28774_end_mask_0 = const()[name = tensor("op_28774_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28774_cast_fp16 = slice_by_index(begin = var_28774_begin_0, end = var_28774_end_0, end_mask = var_28774_end_mask_0, x = transpose_14)[name = tensor("op_28774_cast_fp16")]; + tensor var_28778_begin_0 = const()[name = tensor("op_28778_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_28778_end_0 = const()[name = tensor("op_28778_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_28778_end_mask_0 = const()[name = tensor("op_28778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28778_cast_fp16 = slice_by_index(begin = var_28778_begin_0, end = var_28778_end_0, end_mask = var_28778_end_mask_0, x = transpose_14)[name = tensor("op_28778_cast_fp16")]; + tensor var_28782_begin_0 = const()[name = tensor("op_28782_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_28782_end_0 = const()[name = tensor("op_28782_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_28782_end_mask_0 = const()[name = tensor("op_28782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28782_cast_fp16 = slice_by_index(begin = var_28782_begin_0, end = var_28782_end_0, end_mask = var_28782_end_mask_0, x = transpose_14)[name = tensor("op_28782_cast_fp16")]; + tensor var_28786_begin_0 = const()[name = tensor("op_28786_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_28786_end_0 = const()[name = tensor("op_28786_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_28786_end_mask_0 = const()[name = tensor("op_28786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28786_cast_fp16 = slice_by_index(begin = var_28786_begin_0, end = var_28786_end_0, end_mask = var_28786_end_mask_0, x = transpose_14)[name = tensor("op_28786_cast_fp16")]; + tensor var_28790_begin_0 = const()[name = tensor("op_28790_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_28790_end_0 = const()[name = tensor("op_28790_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_28790_end_mask_0 = const()[name = tensor("op_28790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28790_cast_fp16 = slice_by_index(begin = var_28790_begin_0, end = var_28790_end_0, end_mask = var_28790_end_mask_0, x = transpose_14)[name = tensor("op_28790_cast_fp16")]; + tensor var_28794_begin_0 = const()[name = tensor("op_28794_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_28794_end_0 = const()[name = tensor("op_28794_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_28794_end_mask_0 = const()[name = tensor("op_28794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28794_cast_fp16 = slice_by_index(begin = var_28794_begin_0, end = var_28794_end_0, end_mask = var_28794_end_mask_0, x = transpose_14)[name = tensor("op_28794_cast_fp16")]; + tensor var_28798_begin_0 = const()[name = tensor("op_28798_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_28798_end_0 = const()[name = tensor("op_28798_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_28798_end_mask_0 = const()[name = tensor("op_28798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28798_cast_fp16 = slice_by_index(begin = var_28798_begin_0, end = var_28798_end_0, end_mask = var_28798_end_mask_0, x = transpose_14)[name = tensor("op_28798_cast_fp16")]; + tensor var_28802_begin_0 = const()[name = tensor("op_28802_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_28802_end_0 = const()[name = tensor("op_28802_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_28802_end_mask_0 = const()[name = tensor("op_28802_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28802_cast_fp16 = slice_by_index(begin = var_28802_begin_0, end = var_28802_end_0, end_mask = var_28802_end_mask_0, x = transpose_14)[name = tensor("op_28802_cast_fp16")]; + tensor var_28806_begin_0 = const()[name = tensor("op_28806_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_28806_end_0 = const()[name = tensor("op_28806_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_28806_end_mask_0 = const()[name = tensor("op_28806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28806_cast_fp16 = slice_by_index(begin = var_28806_begin_0, end = var_28806_end_0, end_mask = var_28806_end_mask_0, x = transpose_14)[name = tensor("op_28806_cast_fp16")]; + tensor var_28810_begin_0 = const()[name = tensor("op_28810_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_28810_end_0 = const()[name = tensor("op_28810_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_28810_end_mask_0 = const()[name = tensor("op_28810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28810_cast_fp16 = slice_by_index(begin = var_28810_begin_0, end = var_28810_end_0, end_mask = var_28810_end_mask_0, x = transpose_14)[name = tensor("op_28810_cast_fp16")]; + tensor var_28814_begin_0 = const()[name = tensor("op_28814_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_28814_end_0 = const()[name = tensor("op_28814_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_28814_end_mask_0 = const()[name = tensor("op_28814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28814_cast_fp16 = slice_by_index(begin = var_28814_begin_0, end = var_28814_end_0, end_mask = var_28814_end_mask_0, x = transpose_14)[name = tensor("op_28814_cast_fp16")]; + tensor var_28818_begin_0 = const()[name = tensor("op_28818_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_28818_end_0 = const()[name = tensor("op_28818_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_28818_end_mask_0 = const()[name = tensor("op_28818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28818_cast_fp16 = slice_by_index(begin = var_28818_begin_0, end = var_28818_end_0, end_mask = var_28818_end_mask_0, x = transpose_14)[name = tensor("op_28818_cast_fp16")]; + tensor var_28822_begin_0 = const()[name = tensor("op_28822_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_28822_end_0 = const()[name = tensor("op_28822_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_28822_end_mask_0 = const()[name = tensor("op_28822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_28822_cast_fp16 = slice_by_index(begin = var_28822_begin_0, end = var_28822_end_0, end_mask = var_28822_end_mask_0, x = transpose_14)[name = tensor("op_28822_cast_fp16")]; + tensor var_28824_begin_0 = const()[name = tensor("op_28824_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_28824_end_0 = const()[name = tensor("op_28824_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_28824_end_mask_0 = const()[name = tensor("op_28824_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28824_cast_fp16 = slice_by_index(begin = var_28824_begin_0, end = var_28824_end_0, end_mask = var_28824_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28824_cast_fp16")]; + tensor var_28828_begin_0 = const()[name = tensor("op_28828_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_28828_end_0 = const()[name = tensor("op_28828_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_28828_end_mask_0 = const()[name = tensor("op_28828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28828_cast_fp16 = slice_by_index(begin = var_28828_begin_0, end = var_28828_end_0, end_mask = var_28828_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28828_cast_fp16")]; + tensor var_28832_begin_0 = const()[name = tensor("op_28832_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_28832_end_0 = const()[name = tensor("op_28832_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_28832_end_mask_0 = const()[name = tensor("op_28832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28832_cast_fp16 = slice_by_index(begin = var_28832_begin_0, end = var_28832_end_0, end_mask = var_28832_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28832_cast_fp16")]; + tensor var_28836_begin_0 = const()[name = tensor("op_28836_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_28836_end_0 = const()[name = tensor("op_28836_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_28836_end_mask_0 = const()[name = tensor("op_28836_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28836_cast_fp16 = slice_by_index(begin = var_28836_begin_0, end = var_28836_end_0, end_mask = var_28836_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28836_cast_fp16")]; + tensor var_28840_begin_0 = const()[name = tensor("op_28840_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_28840_end_0 = const()[name = tensor("op_28840_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_28840_end_mask_0 = const()[name = tensor("op_28840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28840_cast_fp16 = slice_by_index(begin = var_28840_begin_0, end = var_28840_end_0, end_mask = var_28840_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28840_cast_fp16")]; + tensor var_28844_begin_0 = const()[name = tensor("op_28844_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_28844_end_0 = const()[name = tensor("op_28844_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_28844_end_mask_0 = const()[name = tensor("op_28844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28844_cast_fp16 = slice_by_index(begin = var_28844_begin_0, end = var_28844_end_0, end_mask = var_28844_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28844_cast_fp16")]; + tensor var_28848_begin_0 = const()[name = tensor("op_28848_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_28848_end_0 = const()[name = tensor("op_28848_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_28848_end_mask_0 = const()[name = tensor("op_28848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28848_cast_fp16 = slice_by_index(begin = var_28848_begin_0, end = var_28848_end_0, end_mask = var_28848_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28848_cast_fp16")]; + tensor var_28852_begin_0 = const()[name = tensor("op_28852_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_28852_end_0 = const()[name = tensor("op_28852_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_28852_end_mask_0 = const()[name = tensor("op_28852_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28852_cast_fp16 = slice_by_index(begin = var_28852_begin_0, end = var_28852_end_0, end_mask = var_28852_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28852_cast_fp16")]; + tensor var_28856_begin_0 = const()[name = tensor("op_28856_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_28856_end_0 = const()[name = tensor("op_28856_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_28856_end_mask_0 = const()[name = tensor("op_28856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28856_cast_fp16 = slice_by_index(begin = var_28856_begin_0, end = var_28856_end_0, end_mask = var_28856_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28856_cast_fp16")]; + tensor var_28860_begin_0 = const()[name = tensor("op_28860_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_28860_end_0 = const()[name = tensor("op_28860_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_28860_end_mask_0 = const()[name = tensor("op_28860_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28860_cast_fp16 = slice_by_index(begin = var_28860_begin_0, end = var_28860_end_0, end_mask = var_28860_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28860_cast_fp16")]; + tensor var_28864_begin_0 = const()[name = tensor("op_28864_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_28864_end_0 = const()[name = tensor("op_28864_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_28864_end_mask_0 = const()[name = tensor("op_28864_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28864_cast_fp16 = slice_by_index(begin = var_28864_begin_0, end = var_28864_end_0, end_mask = var_28864_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28864_cast_fp16")]; + tensor var_28868_begin_0 = const()[name = tensor("op_28868_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_28868_end_0 = const()[name = tensor("op_28868_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_28868_end_mask_0 = const()[name = tensor("op_28868_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28868_cast_fp16 = slice_by_index(begin = var_28868_begin_0, end = var_28868_end_0, end_mask = var_28868_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28868_cast_fp16")]; + tensor var_28872_begin_0 = const()[name = tensor("op_28872_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_28872_end_0 = const()[name = tensor("op_28872_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_28872_end_mask_0 = const()[name = tensor("op_28872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28872_cast_fp16 = slice_by_index(begin = var_28872_begin_0, end = var_28872_end_0, end_mask = var_28872_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28872_cast_fp16")]; + tensor var_28876_begin_0 = const()[name = tensor("op_28876_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_28876_end_0 = const()[name = tensor("op_28876_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_28876_end_mask_0 = const()[name = tensor("op_28876_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28876_cast_fp16 = slice_by_index(begin = var_28876_begin_0, end = var_28876_end_0, end_mask = var_28876_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28876_cast_fp16")]; + tensor var_28880_begin_0 = const()[name = tensor("op_28880_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_28880_end_0 = const()[name = tensor("op_28880_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_28880_end_mask_0 = const()[name = tensor("op_28880_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28880_cast_fp16 = slice_by_index(begin = var_28880_begin_0, end = var_28880_end_0, end_mask = var_28880_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28880_cast_fp16")]; + tensor var_28884_begin_0 = const()[name = tensor("op_28884_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_28884_end_0 = const()[name = tensor("op_28884_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_28884_end_mask_0 = const()[name = tensor("op_28884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28884_cast_fp16 = slice_by_index(begin = var_28884_begin_0, end = var_28884_end_0, end_mask = var_28884_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28884_cast_fp16")]; + tensor var_28888_begin_0 = const()[name = tensor("op_28888_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_28888_end_0 = const()[name = tensor("op_28888_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_28888_end_mask_0 = const()[name = tensor("op_28888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28888_cast_fp16 = slice_by_index(begin = var_28888_begin_0, end = var_28888_end_0, end_mask = var_28888_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28888_cast_fp16")]; + tensor var_28892_begin_0 = const()[name = tensor("op_28892_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_28892_end_0 = const()[name = tensor("op_28892_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_28892_end_mask_0 = const()[name = tensor("op_28892_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28892_cast_fp16 = slice_by_index(begin = var_28892_begin_0, end = var_28892_end_0, end_mask = var_28892_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28892_cast_fp16")]; + tensor var_28896_begin_0 = const()[name = tensor("op_28896_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_28896_end_0 = const()[name = tensor("op_28896_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_28896_end_mask_0 = const()[name = tensor("op_28896_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28896_cast_fp16 = slice_by_index(begin = var_28896_begin_0, end = var_28896_end_0, end_mask = var_28896_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28896_cast_fp16")]; + tensor var_28900_begin_0 = const()[name = tensor("op_28900_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_28900_end_0 = const()[name = tensor("op_28900_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_28900_end_mask_0 = const()[name = tensor("op_28900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_28900_cast_fp16 = slice_by_index(begin = var_28900_begin_0, end = var_28900_end_0, end_mask = var_28900_end_mask_0, x = value_35_cast_fp16)[name = tensor("op_28900_cast_fp16")]; + tensor var_28904_equation_0 = const()[name = tensor("op_28904_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28904_cast_fp16 = einsum(equation = var_28904_equation_0, values = (var_28746_cast_fp16, var_28188_cast_fp16))[name = tensor("op_28904_cast_fp16")]; + tensor var_28905_to_fp16 = const()[name = tensor("op_28905_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2721_cast_fp16 = mul(x = var_28904_cast_fp16, y = var_28905_to_fp16)[name = tensor("aw_chunk_2721_cast_fp16")]; + tensor var_28908_equation_0 = const()[name = tensor("op_28908_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28908_cast_fp16 = einsum(equation = var_28908_equation_0, values = (var_28746_cast_fp16, var_28195_cast_fp16))[name = tensor("op_28908_cast_fp16")]; + tensor var_28909_to_fp16 = const()[name = tensor("op_28909_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2723_cast_fp16 = mul(x = var_28908_cast_fp16, y = var_28909_to_fp16)[name = tensor("aw_chunk_2723_cast_fp16")]; + tensor var_28912_equation_0 = const()[name = tensor("op_28912_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28912_cast_fp16 = einsum(equation = var_28912_equation_0, values = (var_28746_cast_fp16, var_28202_cast_fp16))[name = tensor("op_28912_cast_fp16")]; + tensor var_28913_to_fp16 = const()[name = tensor("op_28913_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2725_cast_fp16 = mul(x = var_28912_cast_fp16, y = var_28913_to_fp16)[name = tensor("aw_chunk_2725_cast_fp16")]; + tensor var_28916_equation_0 = const()[name = tensor("op_28916_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28916_cast_fp16 = einsum(equation = var_28916_equation_0, values = (var_28746_cast_fp16, var_28209_cast_fp16))[name = tensor("op_28916_cast_fp16")]; + tensor var_28917_to_fp16 = const()[name = tensor("op_28917_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2727_cast_fp16 = mul(x = var_28916_cast_fp16, y = var_28917_to_fp16)[name = tensor("aw_chunk_2727_cast_fp16")]; + tensor var_28920_equation_0 = const()[name = tensor("op_28920_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28920_cast_fp16 = einsum(equation = var_28920_equation_0, values = (var_28750_cast_fp16, var_28216_cast_fp16))[name = tensor("op_28920_cast_fp16")]; + tensor var_28921_to_fp16 = const()[name = tensor("op_28921_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2729_cast_fp16 = mul(x = var_28920_cast_fp16, y = var_28921_to_fp16)[name = tensor("aw_chunk_2729_cast_fp16")]; + tensor var_28924_equation_0 = const()[name = tensor("op_28924_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28924_cast_fp16 = einsum(equation = var_28924_equation_0, values = (var_28750_cast_fp16, var_28223_cast_fp16))[name = tensor("op_28924_cast_fp16")]; + tensor var_28925_to_fp16 = const()[name = tensor("op_28925_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2731_cast_fp16 = mul(x = var_28924_cast_fp16, y = var_28925_to_fp16)[name = tensor("aw_chunk_2731_cast_fp16")]; + tensor var_28928_equation_0 = const()[name = tensor("op_28928_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28928_cast_fp16 = einsum(equation = var_28928_equation_0, values = (var_28750_cast_fp16, var_28230_cast_fp16))[name = tensor("op_28928_cast_fp16")]; + tensor var_28929_to_fp16 = const()[name = tensor("op_28929_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2733_cast_fp16 = mul(x = var_28928_cast_fp16, y = var_28929_to_fp16)[name = tensor("aw_chunk_2733_cast_fp16")]; + tensor var_28932_equation_0 = const()[name = tensor("op_28932_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28932_cast_fp16 = einsum(equation = var_28932_equation_0, values = (var_28750_cast_fp16, var_28237_cast_fp16))[name = tensor("op_28932_cast_fp16")]; + tensor var_28933_to_fp16 = const()[name = tensor("op_28933_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2735_cast_fp16 = mul(x = var_28932_cast_fp16, y = var_28933_to_fp16)[name = tensor("aw_chunk_2735_cast_fp16")]; + tensor var_28936_equation_0 = const()[name = tensor("op_28936_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28936_cast_fp16 = einsum(equation = var_28936_equation_0, values = (var_28754_cast_fp16, var_28244_cast_fp16))[name = tensor("op_28936_cast_fp16")]; + tensor var_28937_to_fp16 = const()[name = tensor("op_28937_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2737_cast_fp16 = mul(x = var_28936_cast_fp16, y = var_28937_to_fp16)[name = tensor("aw_chunk_2737_cast_fp16")]; + tensor var_28940_equation_0 = const()[name = tensor("op_28940_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28940_cast_fp16 = einsum(equation = var_28940_equation_0, values = (var_28754_cast_fp16, var_28251_cast_fp16))[name = tensor("op_28940_cast_fp16")]; + tensor var_28941_to_fp16 = const()[name = tensor("op_28941_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2739_cast_fp16 = mul(x = var_28940_cast_fp16, y = var_28941_to_fp16)[name = tensor("aw_chunk_2739_cast_fp16")]; + tensor var_28944_equation_0 = const()[name = tensor("op_28944_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28944_cast_fp16 = einsum(equation = var_28944_equation_0, values = (var_28754_cast_fp16, var_28258_cast_fp16))[name = tensor("op_28944_cast_fp16")]; + tensor var_28945_to_fp16 = const()[name = tensor("op_28945_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2741_cast_fp16 = mul(x = var_28944_cast_fp16, y = var_28945_to_fp16)[name = tensor("aw_chunk_2741_cast_fp16")]; + tensor var_28948_equation_0 = const()[name = tensor("op_28948_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28948_cast_fp16 = einsum(equation = var_28948_equation_0, values = (var_28754_cast_fp16, var_28265_cast_fp16))[name = tensor("op_28948_cast_fp16")]; + tensor var_28949_to_fp16 = const()[name = tensor("op_28949_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2743_cast_fp16 = mul(x = var_28948_cast_fp16, y = var_28949_to_fp16)[name = tensor("aw_chunk_2743_cast_fp16")]; + tensor var_28952_equation_0 = const()[name = tensor("op_28952_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28952_cast_fp16 = einsum(equation = var_28952_equation_0, values = (var_28758_cast_fp16, var_28272_cast_fp16))[name = tensor("op_28952_cast_fp16")]; + tensor var_28953_to_fp16 = const()[name = tensor("op_28953_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2745_cast_fp16 = mul(x = var_28952_cast_fp16, y = var_28953_to_fp16)[name = tensor("aw_chunk_2745_cast_fp16")]; + tensor var_28956_equation_0 = const()[name = tensor("op_28956_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28956_cast_fp16 = einsum(equation = var_28956_equation_0, values = (var_28758_cast_fp16, var_28279_cast_fp16))[name = tensor("op_28956_cast_fp16")]; + tensor var_28957_to_fp16 = const()[name = tensor("op_28957_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2747_cast_fp16 = mul(x = var_28956_cast_fp16, y = var_28957_to_fp16)[name = tensor("aw_chunk_2747_cast_fp16")]; + tensor var_28960_equation_0 = const()[name = tensor("op_28960_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28960_cast_fp16 = einsum(equation = var_28960_equation_0, values = (var_28758_cast_fp16, var_28286_cast_fp16))[name = tensor("op_28960_cast_fp16")]; + tensor var_28961_to_fp16 = const()[name = tensor("op_28961_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2749_cast_fp16 = mul(x = var_28960_cast_fp16, y = var_28961_to_fp16)[name = tensor("aw_chunk_2749_cast_fp16")]; + tensor var_28964_equation_0 = const()[name = tensor("op_28964_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28964_cast_fp16 = einsum(equation = var_28964_equation_0, values = (var_28758_cast_fp16, var_28293_cast_fp16))[name = tensor("op_28964_cast_fp16")]; + tensor var_28965_to_fp16 = const()[name = tensor("op_28965_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2751_cast_fp16 = mul(x = var_28964_cast_fp16, y = var_28965_to_fp16)[name = tensor("aw_chunk_2751_cast_fp16")]; + tensor var_28968_equation_0 = const()[name = tensor("op_28968_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28968_cast_fp16 = einsum(equation = var_28968_equation_0, values = (var_28762_cast_fp16, var_28300_cast_fp16))[name = tensor("op_28968_cast_fp16")]; + tensor var_28969_to_fp16 = const()[name = tensor("op_28969_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2753_cast_fp16 = mul(x = var_28968_cast_fp16, y = var_28969_to_fp16)[name = tensor("aw_chunk_2753_cast_fp16")]; + tensor var_28972_equation_0 = const()[name = tensor("op_28972_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28972_cast_fp16 = einsum(equation = var_28972_equation_0, values = (var_28762_cast_fp16, var_28307_cast_fp16))[name = tensor("op_28972_cast_fp16")]; + tensor var_28973_to_fp16 = const()[name = tensor("op_28973_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2755_cast_fp16 = mul(x = var_28972_cast_fp16, y = var_28973_to_fp16)[name = tensor("aw_chunk_2755_cast_fp16")]; + tensor var_28976_equation_0 = const()[name = tensor("op_28976_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28976_cast_fp16 = einsum(equation = var_28976_equation_0, values = (var_28762_cast_fp16, var_28314_cast_fp16))[name = tensor("op_28976_cast_fp16")]; + tensor var_28977_to_fp16 = const()[name = tensor("op_28977_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2757_cast_fp16 = mul(x = var_28976_cast_fp16, y = var_28977_to_fp16)[name = tensor("aw_chunk_2757_cast_fp16")]; + tensor var_28980_equation_0 = const()[name = tensor("op_28980_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28980_cast_fp16 = einsum(equation = var_28980_equation_0, values = (var_28762_cast_fp16, var_28321_cast_fp16))[name = tensor("op_28980_cast_fp16")]; + tensor var_28981_to_fp16 = const()[name = tensor("op_28981_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2759_cast_fp16 = mul(x = var_28980_cast_fp16, y = var_28981_to_fp16)[name = tensor("aw_chunk_2759_cast_fp16")]; + tensor var_28984_equation_0 = const()[name = tensor("op_28984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28984_cast_fp16 = einsum(equation = var_28984_equation_0, values = (var_28766_cast_fp16, var_28328_cast_fp16))[name = tensor("op_28984_cast_fp16")]; + tensor var_28985_to_fp16 = const()[name = tensor("op_28985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2761_cast_fp16 = mul(x = var_28984_cast_fp16, y = var_28985_to_fp16)[name = tensor("aw_chunk_2761_cast_fp16")]; + tensor var_28988_equation_0 = const()[name = tensor("op_28988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28988_cast_fp16 = einsum(equation = var_28988_equation_0, values = (var_28766_cast_fp16, var_28335_cast_fp16))[name = tensor("op_28988_cast_fp16")]; + tensor var_28989_to_fp16 = const()[name = tensor("op_28989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2763_cast_fp16 = mul(x = var_28988_cast_fp16, y = var_28989_to_fp16)[name = tensor("aw_chunk_2763_cast_fp16")]; + tensor var_28992_equation_0 = const()[name = tensor("op_28992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28992_cast_fp16 = einsum(equation = var_28992_equation_0, values = (var_28766_cast_fp16, var_28342_cast_fp16))[name = tensor("op_28992_cast_fp16")]; + tensor var_28993_to_fp16 = const()[name = tensor("op_28993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2765_cast_fp16 = mul(x = var_28992_cast_fp16, y = var_28993_to_fp16)[name = tensor("aw_chunk_2765_cast_fp16")]; + tensor var_28996_equation_0 = const()[name = tensor("op_28996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_28996_cast_fp16 = einsum(equation = var_28996_equation_0, values = (var_28766_cast_fp16, var_28349_cast_fp16))[name = tensor("op_28996_cast_fp16")]; + tensor var_28997_to_fp16 = const()[name = tensor("op_28997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2767_cast_fp16 = mul(x = var_28996_cast_fp16, y = var_28997_to_fp16)[name = tensor("aw_chunk_2767_cast_fp16")]; + tensor var_29000_equation_0 = const()[name = tensor("op_29000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29000_cast_fp16 = einsum(equation = var_29000_equation_0, values = (var_28770_cast_fp16, var_28356_cast_fp16))[name = tensor("op_29000_cast_fp16")]; + tensor var_29001_to_fp16 = const()[name = tensor("op_29001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2769_cast_fp16 = mul(x = var_29000_cast_fp16, y = var_29001_to_fp16)[name = tensor("aw_chunk_2769_cast_fp16")]; + tensor var_29004_equation_0 = const()[name = tensor("op_29004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29004_cast_fp16 = einsum(equation = var_29004_equation_0, values = (var_28770_cast_fp16, var_28363_cast_fp16))[name = tensor("op_29004_cast_fp16")]; + tensor var_29005_to_fp16 = const()[name = tensor("op_29005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2771_cast_fp16 = mul(x = var_29004_cast_fp16, y = var_29005_to_fp16)[name = tensor("aw_chunk_2771_cast_fp16")]; + tensor var_29008_equation_0 = const()[name = tensor("op_29008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29008_cast_fp16 = einsum(equation = var_29008_equation_0, values = (var_28770_cast_fp16, var_28370_cast_fp16))[name = tensor("op_29008_cast_fp16")]; + tensor var_29009_to_fp16 = const()[name = tensor("op_29009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2773_cast_fp16 = mul(x = var_29008_cast_fp16, y = var_29009_to_fp16)[name = tensor("aw_chunk_2773_cast_fp16")]; + tensor var_29012_equation_0 = const()[name = tensor("op_29012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29012_cast_fp16 = einsum(equation = var_29012_equation_0, values = (var_28770_cast_fp16, var_28377_cast_fp16))[name = tensor("op_29012_cast_fp16")]; + tensor var_29013_to_fp16 = const()[name = tensor("op_29013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2775_cast_fp16 = mul(x = var_29012_cast_fp16, y = var_29013_to_fp16)[name = tensor("aw_chunk_2775_cast_fp16")]; + tensor var_29016_equation_0 = const()[name = tensor("op_29016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29016_cast_fp16 = einsum(equation = var_29016_equation_0, values = (var_28774_cast_fp16, var_28384_cast_fp16))[name = tensor("op_29016_cast_fp16")]; + tensor var_29017_to_fp16 = const()[name = tensor("op_29017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2777_cast_fp16 = mul(x = var_29016_cast_fp16, y = var_29017_to_fp16)[name = tensor("aw_chunk_2777_cast_fp16")]; + tensor var_29020_equation_0 = const()[name = tensor("op_29020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29020_cast_fp16 = einsum(equation = var_29020_equation_0, values = (var_28774_cast_fp16, var_28391_cast_fp16))[name = tensor("op_29020_cast_fp16")]; + tensor var_29021_to_fp16 = const()[name = tensor("op_29021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2779_cast_fp16 = mul(x = var_29020_cast_fp16, y = var_29021_to_fp16)[name = tensor("aw_chunk_2779_cast_fp16")]; + tensor var_29024_equation_0 = const()[name = tensor("op_29024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29024_cast_fp16 = einsum(equation = var_29024_equation_0, values = (var_28774_cast_fp16, var_28398_cast_fp16))[name = tensor("op_29024_cast_fp16")]; + tensor var_29025_to_fp16 = const()[name = tensor("op_29025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2781_cast_fp16 = mul(x = var_29024_cast_fp16, y = var_29025_to_fp16)[name = tensor("aw_chunk_2781_cast_fp16")]; + tensor var_29028_equation_0 = const()[name = tensor("op_29028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29028_cast_fp16 = einsum(equation = var_29028_equation_0, values = (var_28774_cast_fp16, var_28405_cast_fp16))[name = tensor("op_29028_cast_fp16")]; + tensor var_29029_to_fp16 = const()[name = tensor("op_29029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2783_cast_fp16 = mul(x = var_29028_cast_fp16, y = var_29029_to_fp16)[name = tensor("aw_chunk_2783_cast_fp16")]; + tensor var_29032_equation_0 = const()[name = tensor("op_29032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29032_cast_fp16 = einsum(equation = var_29032_equation_0, values = (var_28778_cast_fp16, var_28412_cast_fp16))[name = tensor("op_29032_cast_fp16")]; + tensor var_29033_to_fp16 = const()[name = tensor("op_29033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2785_cast_fp16 = mul(x = var_29032_cast_fp16, y = var_29033_to_fp16)[name = tensor("aw_chunk_2785_cast_fp16")]; + tensor var_29036_equation_0 = const()[name = tensor("op_29036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29036_cast_fp16 = einsum(equation = var_29036_equation_0, values = (var_28778_cast_fp16, var_28419_cast_fp16))[name = tensor("op_29036_cast_fp16")]; + tensor var_29037_to_fp16 = const()[name = tensor("op_29037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2787_cast_fp16 = mul(x = var_29036_cast_fp16, y = var_29037_to_fp16)[name = tensor("aw_chunk_2787_cast_fp16")]; + tensor var_29040_equation_0 = const()[name = tensor("op_29040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29040_cast_fp16 = einsum(equation = var_29040_equation_0, values = (var_28778_cast_fp16, var_28426_cast_fp16))[name = tensor("op_29040_cast_fp16")]; + tensor var_29041_to_fp16 = const()[name = tensor("op_29041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2789_cast_fp16 = mul(x = var_29040_cast_fp16, y = var_29041_to_fp16)[name = tensor("aw_chunk_2789_cast_fp16")]; + tensor var_29044_equation_0 = const()[name = tensor("op_29044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29044_cast_fp16 = einsum(equation = var_29044_equation_0, values = (var_28778_cast_fp16, var_28433_cast_fp16))[name = tensor("op_29044_cast_fp16")]; + tensor var_29045_to_fp16 = const()[name = tensor("op_29045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2791_cast_fp16 = mul(x = var_29044_cast_fp16, y = var_29045_to_fp16)[name = tensor("aw_chunk_2791_cast_fp16")]; + tensor var_29048_equation_0 = const()[name = tensor("op_29048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29048_cast_fp16 = einsum(equation = var_29048_equation_0, values = (var_28782_cast_fp16, var_28440_cast_fp16))[name = tensor("op_29048_cast_fp16")]; + tensor var_29049_to_fp16 = const()[name = tensor("op_29049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2793_cast_fp16 = mul(x = var_29048_cast_fp16, y = var_29049_to_fp16)[name = tensor("aw_chunk_2793_cast_fp16")]; + tensor var_29052_equation_0 = const()[name = tensor("op_29052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29052_cast_fp16 = einsum(equation = var_29052_equation_0, values = (var_28782_cast_fp16, var_28447_cast_fp16))[name = tensor("op_29052_cast_fp16")]; + tensor var_29053_to_fp16 = const()[name = tensor("op_29053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2795_cast_fp16 = mul(x = var_29052_cast_fp16, y = var_29053_to_fp16)[name = tensor("aw_chunk_2795_cast_fp16")]; + tensor var_29056_equation_0 = const()[name = tensor("op_29056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29056_cast_fp16 = einsum(equation = var_29056_equation_0, values = (var_28782_cast_fp16, var_28454_cast_fp16))[name = tensor("op_29056_cast_fp16")]; + tensor var_29057_to_fp16 = const()[name = tensor("op_29057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2797_cast_fp16 = mul(x = var_29056_cast_fp16, y = var_29057_to_fp16)[name = tensor("aw_chunk_2797_cast_fp16")]; + tensor var_29060_equation_0 = const()[name = tensor("op_29060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29060_cast_fp16 = einsum(equation = var_29060_equation_0, values = (var_28782_cast_fp16, var_28461_cast_fp16))[name = tensor("op_29060_cast_fp16")]; + tensor var_29061_to_fp16 = const()[name = tensor("op_29061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2799_cast_fp16 = mul(x = var_29060_cast_fp16, y = var_29061_to_fp16)[name = tensor("aw_chunk_2799_cast_fp16")]; + tensor var_29064_equation_0 = const()[name = tensor("op_29064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29064_cast_fp16 = einsum(equation = var_29064_equation_0, values = (var_28786_cast_fp16, var_28468_cast_fp16))[name = tensor("op_29064_cast_fp16")]; + tensor var_29065_to_fp16 = const()[name = tensor("op_29065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2801_cast_fp16 = mul(x = var_29064_cast_fp16, y = var_29065_to_fp16)[name = tensor("aw_chunk_2801_cast_fp16")]; + tensor var_29068_equation_0 = const()[name = tensor("op_29068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29068_cast_fp16 = einsum(equation = var_29068_equation_0, values = (var_28786_cast_fp16, var_28475_cast_fp16))[name = tensor("op_29068_cast_fp16")]; + tensor var_29069_to_fp16 = const()[name = tensor("op_29069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2803_cast_fp16 = mul(x = var_29068_cast_fp16, y = var_29069_to_fp16)[name = tensor("aw_chunk_2803_cast_fp16")]; + tensor var_29072_equation_0 = const()[name = tensor("op_29072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29072_cast_fp16 = einsum(equation = var_29072_equation_0, values = (var_28786_cast_fp16, var_28482_cast_fp16))[name = tensor("op_29072_cast_fp16")]; + tensor var_29073_to_fp16 = const()[name = tensor("op_29073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2805_cast_fp16 = mul(x = var_29072_cast_fp16, y = var_29073_to_fp16)[name = tensor("aw_chunk_2805_cast_fp16")]; + tensor var_29076_equation_0 = const()[name = tensor("op_29076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29076_cast_fp16 = einsum(equation = var_29076_equation_0, values = (var_28786_cast_fp16, var_28489_cast_fp16))[name = tensor("op_29076_cast_fp16")]; + tensor var_29077_to_fp16 = const()[name = tensor("op_29077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2807_cast_fp16 = mul(x = var_29076_cast_fp16, y = var_29077_to_fp16)[name = tensor("aw_chunk_2807_cast_fp16")]; + tensor var_29080_equation_0 = const()[name = tensor("op_29080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29080_cast_fp16 = einsum(equation = var_29080_equation_0, values = (var_28790_cast_fp16, var_28496_cast_fp16))[name = tensor("op_29080_cast_fp16")]; + tensor var_29081_to_fp16 = const()[name = tensor("op_29081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2809_cast_fp16 = mul(x = var_29080_cast_fp16, y = var_29081_to_fp16)[name = tensor("aw_chunk_2809_cast_fp16")]; + tensor var_29084_equation_0 = const()[name = tensor("op_29084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29084_cast_fp16 = einsum(equation = var_29084_equation_0, values = (var_28790_cast_fp16, var_28503_cast_fp16))[name = tensor("op_29084_cast_fp16")]; + tensor var_29085_to_fp16 = const()[name = tensor("op_29085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2811_cast_fp16 = mul(x = var_29084_cast_fp16, y = var_29085_to_fp16)[name = tensor("aw_chunk_2811_cast_fp16")]; + tensor var_29088_equation_0 = const()[name = tensor("op_29088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29088_cast_fp16 = einsum(equation = var_29088_equation_0, values = (var_28790_cast_fp16, var_28510_cast_fp16))[name = tensor("op_29088_cast_fp16")]; + tensor var_29089_to_fp16 = const()[name = tensor("op_29089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2813_cast_fp16 = mul(x = var_29088_cast_fp16, y = var_29089_to_fp16)[name = tensor("aw_chunk_2813_cast_fp16")]; + tensor var_29092_equation_0 = const()[name = tensor("op_29092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29092_cast_fp16 = einsum(equation = var_29092_equation_0, values = (var_28790_cast_fp16, var_28517_cast_fp16))[name = tensor("op_29092_cast_fp16")]; + tensor var_29093_to_fp16 = const()[name = tensor("op_29093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2815_cast_fp16 = mul(x = var_29092_cast_fp16, y = var_29093_to_fp16)[name = tensor("aw_chunk_2815_cast_fp16")]; + tensor var_29096_equation_0 = const()[name = tensor("op_29096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29096_cast_fp16 = einsum(equation = var_29096_equation_0, values = (var_28794_cast_fp16, var_28524_cast_fp16))[name = tensor("op_29096_cast_fp16")]; + tensor var_29097_to_fp16 = const()[name = tensor("op_29097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2817_cast_fp16 = mul(x = var_29096_cast_fp16, y = var_29097_to_fp16)[name = tensor("aw_chunk_2817_cast_fp16")]; + tensor var_29100_equation_0 = const()[name = tensor("op_29100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29100_cast_fp16 = einsum(equation = var_29100_equation_0, values = (var_28794_cast_fp16, var_28531_cast_fp16))[name = tensor("op_29100_cast_fp16")]; + tensor var_29101_to_fp16 = const()[name = tensor("op_29101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2819_cast_fp16 = mul(x = var_29100_cast_fp16, y = var_29101_to_fp16)[name = tensor("aw_chunk_2819_cast_fp16")]; + tensor var_29104_equation_0 = const()[name = tensor("op_29104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29104_cast_fp16 = einsum(equation = var_29104_equation_0, values = (var_28794_cast_fp16, var_28538_cast_fp16))[name = tensor("op_29104_cast_fp16")]; + tensor var_29105_to_fp16 = const()[name = tensor("op_29105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2821_cast_fp16 = mul(x = var_29104_cast_fp16, y = var_29105_to_fp16)[name = tensor("aw_chunk_2821_cast_fp16")]; + tensor var_29108_equation_0 = const()[name = tensor("op_29108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29108_cast_fp16 = einsum(equation = var_29108_equation_0, values = (var_28794_cast_fp16, var_28545_cast_fp16))[name = tensor("op_29108_cast_fp16")]; + tensor var_29109_to_fp16 = const()[name = tensor("op_29109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2823_cast_fp16 = mul(x = var_29108_cast_fp16, y = var_29109_to_fp16)[name = tensor("aw_chunk_2823_cast_fp16")]; + tensor var_29112_equation_0 = const()[name = tensor("op_29112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29112_cast_fp16 = einsum(equation = var_29112_equation_0, values = (var_28798_cast_fp16, var_28552_cast_fp16))[name = tensor("op_29112_cast_fp16")]; + tensor var_29113_to_fp16 = const()[name = tensor("op_29113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2825_cast_fp16 = mul(x = var_29112_cast_fp16, y = var_29113_to_fp16)[name = tensor("aw_chunk_2825_cast_fp16")]; + tensor var_29116_equation_0 = const()[name = tensor("op_29116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29116_cast_fp16 = einsum(equation = var_29116_equation_0, values = (var_28798_cast_fp16, var_28559_cast_fp16))[name = tensor("op_29116_cast_fp16")]; + tensor var_29117_to_fp16 = const()[name = tensor("op_29117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2827_cast_fp16 = mul(x = var_29116_cast_fp16, y = var_29117_to_fp16)[name = tensor("aw_chunk_2827_cast_fp16")]; + tensor var_29120_equation_0 = const()[name = tensor("op_29120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29120_cast_fp16 = einsum(equation = var_29120_equation_0, values = (var_28798_cast_fp16, var_28566_cast_fp16))[name = tensor("op_29120_cast_fp16")]; + tensor var_29121_to_fp16 = const()[name = tensor("op_29121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2829_cast_fp16 = mul(x = var_29120_cast_fp16, y = var_29121_to_fp16)[name = tensor("aw_chunk_2829_cast_fp16")]; + tensor var_29124_equation_0 = const()[name = tensor("op_29124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29124_cast_fp16 = einsum(equation = var_29124_equation_0, values = (var_28798_cast_fp16, var_28573_cast_fp16))[name = tensor("op_29124_cast_fp16")]; + tensor var_29125_to_fp16 = const()[name = tensor("op_29125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2831_cast_fp16 = mul(x = var_29124_cast_fp16, y = var_29125_to_fp16)[name = tensor("aw_chunk_2831_cast_fp16")]; + tensor var_29128_equation_0 = const()[name = tensor("op_29128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29128_cast_fp16 = einsum(equation = var_29128_equation_0, values = (var_28802_cast_fp16, var_28580_cast_fp16))[name = tensor("op_29128_cast_fp16")]; + tensor var_29129_to_fp16 = const()[name = tensor("op_29129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2833_cast_fp16 = mul(x = var_29128_cast_fp16, y = var_29129_to_fp16)[name = tensor("aw_chunk_2833_cast_fp16")]; + tensor var_29132_equation_0 = const()[name = tensor("op_29132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29132_cast_fp16 = einsum(equation = var_29132_equation_0, values = (var_28802_cast_fp16, var_28587_cast_fp16))[name = tensor("op_29132_cast_fp16")]; + tensor var_29133_to_fp16 = const()[name = tensor("op_29133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2835_cast_fp16 = mul(x = var_29132_cast_fp16, y = var_29133_to_fp16)[name = tensor("aw_chunk_2835_cast_fp16")]; + tensor var_29136_equation_0 = const()[name = tensor("op_29136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29136_cast_fp16 = einsum(equation = var_29136_equation_0, values = (var_28802_cast_fp16, var_28594_cast_fp16))[name = tensor("op_29136_cast_fp16")]; + tensor var_29137_to_fp16 = const()[name = tensor("op_29137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2837_cast_fp16 = mul(x = var_29136_cast_fp16, y = var_29137_to_fp16)[name = tensor("aw_chunk_2837_cast_fp16")]; + tensor var_29140_equation_0 = const()[name = tensor("op_29140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29140_cast_fp16 = einsum(equation = var_29140_equation_0, values = (var_28802_cast_fp16, var_28601_cast_fp16))[name = tensor("op_29140_cast_fp16")]; + tensor var_29141_to_fp16 = const()[name = tensor("op_29141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2839_cast_fp16 = mul(x = var_29140_cast_fp16, y = var_29141_to_fp16)[name = tensor("aw_chunk_2839_cast_fp16")]; + tensor var_29144_equation_0 = const()[name = tensor("op_29144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29144_cast_fp16 = einsum(equation = var_29144_equation_0, values = (var_28806_cast_fp16, var_28608_cast_fp16))[name = tensor("op_29144_cast_fp16")]; + tensor var_29145_to_fp16 = const()[name = tensor("op_29145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2841_cast_fp16 = mul(x = var_29144_cast_fp16, y = var_29145_to_fp16)[name = tensor("aw_chunk_2841_cast_fp16")]; + tensor var_29148_equation_0 = const()[name = tensor("op_29148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29148_cast_fp16 = einsum(equation = var_29148_equation_0, values = (var_28806_cast_fp16, var_28615_cast_fp16))[name = tensor("op_29148_cast_fp16")]; + tensor var_29149_to_fp16 = const()[name = tensor("op_29149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2843_cast_fp16 = mul(x = var_29148_cast_fp16, y = var_29149_to_fp16)[name = tensor("aw_chunk_2843_cast_fp16")]; + tensor var_29152_equation_0 = const()[name = tensor("op_29152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29152_cast_fp16 = einsum(equation = var_29152_equation_0, values = (var_28806_cast_fp16, var_28622_cast_fp16))[name = tensor("op_29152_cast_fp16")]; + tensor var_29153_to_fp16 = const()[name = tensor("op_29153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2845_cast_fp16 = mul(x = var_29152_cast_fp16, y = var_29153_to_fp16)[name = tensor("aw_chunk_2845_cast_fp16")]; + tensor var_29156_equation_0 = const()[name = tensor("op_29156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29156_cast_fp16 = einsum(equation = var_29156_equation_0, values = (var_28806_cast_fp16, var_28629_cast_fp16))[name = tensor("op_29156_cast_fp16")]; + tensor var_29157_to_fp16 = const()[name = tensor("op_29157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2847_cast_fp16 = mul(x = var_29156_cast_fp16, y = var_29157_to_fp16)[name = tensor("aw_chunk_2847_cast_fp16")]; + tensor var_29160_equation_0 = const()[name = tensor("op_29160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29160_cast_fp16 = einsum(equation = var_29160_equation_0, values = (var_28810_cast_fp16, var_28636_cast_fp16))[name = tensor("op_29160_cast_fp16")]; + tensor var_29161_to_fp16 = const()[name = tensor("op_29161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2849_cast_fp16 = mul(x = var_29160_cast_fp16, y = var_29161_to_fp16)[name = tensor("aw_chunk_2849_cast_fp16")]; + tensor var_29164_equation_0 = const()[name = tensor("op_29164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29164_cast_fp16 = einsum(equation = var_29164_equation_0, values = (var_28810_cast_fp16, var_28643_cast_fp16))[name = tensor("op_29164_cast_fp16")]; + tensor var_29165_to_fp16 = const()[name = tensor("op_29165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2851_cast_fp16 = mul(x = var_29164_cast_fp16, y = var_29165_to_fp16)[name = tensor("aw_chunk_2851_cast_fp16")]; + tensor var_29168_equation_0 = const()[name = tensor("op_29168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29168_cast_fp16 = einsum(equation = var_29168_equation_0, values = (var_28810_cast_fp16, var_28650_cast_fp16))[name = tensor("op_29168_cast_fp16")]; + tensor var_29169_to_fp16 = const()[name = tensor("op_29169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2853_cast_fp16 = mul(x = var_29168_cast_fp16, y = var_29169_to_fp16)[name = tensor("aw_chunk_2853_cast_fp16")]; + tensor var_29172_equation_0 = const()[name = tensor("op_29172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29172_cast_fp16 = einsum(equation = var_29172_equation_0, values = (var_28810_cast_fp16, var_28657_cast_fp16))[name = tensor("op_29172_cast_fp16")]; + tensor var_29173_to_fp16 = const()[name = tensor("op_29173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2855_cast_fp16 = mul(x = var_29172_cast_fp16, y = var_29173_to_fp16)[name = tensor("aw_chunk_2855_cast_fp16")]; + tensor var_29176_equation_0 = const()[name = tensor("op_29176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29176_cast_fp16 = einsum(equation = var_29176_equation_0, values = (var_28814_cast_fp16, var_28664_cast_fp16))[name = tensor("op_29176_cast_fp16")]; + tensor var_29177_to_fp16 = const()[name = tensor("op_29177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2857_cast_fp16 = mul(x = var_29176_cast_fp16, y = var_29177_to_fp16)[name = tensor("aw_chunk_2857_cast_fp16")]; + tensor var_29180_equation_0 = const()[name = tensor("op_29180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29180_cast_fp16 = einsum(equation = var_29180_equation_0, values = (var_28814_cast_fp16, var_28671_cast_fp16))[name = tensor("op_29180_cast_fp16")]; + tensor var_29181_to_fp16 = const()[name = tensor("op_29181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2859_cast_fp16 = mul(x = var_29180_cast_fp16, y = var_29181_to_fp16)[name = tensor("aw_chunk_2859_cast_fp16")]; + tensor var_29184_equation_0 = const()[name = tensor("op_29184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29184_cast_fp16 = einsum(equation = var_29184_equation_0, values = (var_28814_cast_fp16, var_28678_cast_fp16))[name = tensor("op_29184_cast_fp16")]; + tensor var_29185_to_fp16 = const()[name = tensor("op_29185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2861_cast_fp16 = mul(x = var_29184_cast_fp16, y = var_29185_to_fp16)[name = tensor("aw_chunk_2861_cast_fp16")]; + tensor var_29188_equation_0 = const()[name = tensor("op_29188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29188_cast_fp16 = einsum(equation = var_29188_equation_0, values = (var_28814_cast_fp16, var_28685_cast_fp16))[name = tensor("op_29188_cast_fp16")]; + tensor var_29189_to_fp16 = const()[name = tensor("op_29189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2863_cast_fp16 = mul(x = var_29188_cast_fp16, y = var_29189_to_fp16)[name = tensor("aw_chunk_2863_cast_fp16")]; + tensor var_29192_equation_0 = const()[name = tensor("op_29192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29192_cast_fp16 = einsum(equation = var_29192_equation_0, values = (var_28818_cast_fp16, var_28692_cast_fp16))[name = tensor("op_29192_cast_fp16")]; + tensor var_29193_to_fp16 = const()[name = tensor("op_29193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2865_cast_fp16 = mul(x = var_29192_cast_fp16, y = var_29193_to_fp16)[name = tensor("aw_chunk_2865_cast_fp16")]; + tensor var_29196_equation_0 = const()[name = tensor("op_29196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29196_cast_fp16 = einsum(equation = var_29196_equation_0, values = (var_28818_cast_fp16, var_28699_cast_fp16))[name = tensor("op_29196_cast_fp16")]; + tensor var_29197_to_fp16 = const()[name = tensor("op_29197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2867_cast_fp16 = mul(x = var_29196_cast_fp16, y = var_29197_to_fp16)[name = tensor("aw_chunk_2867_cast_fp16")]; + tensor var_29200_equation_0 = const()[name = tensor("op_29200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29200_cast_fp16 = einsum(equation = var_29200_equation_0, values = (var_28818_cast_fp16, var_28706_cast_fp16))[name = tensor("op_29200_cast_fp16")]; + tensor var_29201_to_fp16 = const()[name = tensor("op_29201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2869_cast_fp16 = mul(x = var_29200_cast_fp16, y = var_29201_to_fp16)[name = tensor("aw_chunk_2869_cast_fp16")]; + tensor var_29204_equation_0 = const()[name = tensor("op_29204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29204_cast_fp16 = einsum(equation = var_29204_equation_0, values = (var_28818_cast_fp16, var_28713_cast_fp16))[name = tensor("op_29204_cast_fp16")]; + tensor var_29205_to_fp16 = const()[name = tensor("op_29205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2871_cast_fp16 = mul(x = var_29204_cast_fp16, y = var_29205_to_fp16)[name = tensor("aw_chunk_2871_cast_fp16")]; + tensor var_29208_equation_0 = const()[name = tensor("op_29208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29208_cast_fp16 = einsum(equation = var_29208_equation_0, values = (var_28822_cast_fp16, var_28720_cast_fp16))[name = tensor("op_29208_cast_fp16")]; + tensor var_29209_to_fp16 = const()[name = tensor("op_29209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2873_cast_fp16 = mul(x = var_29208_cast_fp16, y = var_29209_to_fp16)[name = tensor("aw_chunk_2873_cast_fp16")]; + tensor var_29212_equation_0 = const()[name = tensor("op_29212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29212_cast_fp16 = einsum(equation = var_29212_equation_0, values = (var_28822_cast_fp16, var_28727_cast_fp16))[name = tensor("op_29212_cast_fp16")]; + tensor var_29213_to_fp16 = const()[name = tensor("op_29213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2875_cast_fp16 = mul(x = var_29212_cast_fp16, y = var_29213_to_fp16)[name = tensor("aw_chunk_2875_cast_fp16")]; + tensor var_29216_equation_0 = const()[name = tensor("op_29216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29216_cast_fp16 = einsum(equation = var_29216_equation_0, values = (var_28822_cast_fp16, var_28734_cast_fp16))[name = tensor("op_29216_cast_fp16")]; + tensor var_29217_to_fp16 = const()[name = tensor("op_29217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2877_cast_fp16 = mul(x = var_29216_cast_fp16, y = var_29217_to_fp16)[name = tensor("aw_chunk_2877_cast_fp16")]; + tensor var_29220_equation_0 = const()[name = tensor("op_29220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_29220_cast_fp16 = einsum(equation = var_29220_equation_0, values = (var_28822_cast_fp16, var_28741_cast_fp16))[name = tensor("op_29220_cast_fp16")]; + tensor var_29221_to_fp16 = const()[name = tensor("op_29221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2879_cast_fp16 = mul(x = var_29220_cast_fp16, y = var_29221_to_fp16)[name = tensor("aw_chunk_2879_cast_fp16")]; + tensor var_29223_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2721_cast_fp16)[name = tensor("op_29223_cast_fp16")]; + tensor var_29224_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2723_cast_fp16)[name = tensor("op_29224_cast_fp16")]; + tensor var_29225_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2725_cast_fp16)[name = tensor("op_29225_cast_fp16")]; + tensor var_29226_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2727_cast_fp16)[name = tensor("op_29226_cast_fp16")]; + tensor var_29227_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2729_cast_fp16)[name = tensor("op_29227_cast_fp16")]; + tensor var_29228_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2731_cast_fp16)[name = tensor("op_29228_cast_fp16")]; + tensor var_29229_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2733_cast_fp16)[name = tensor("op_29229_cast_fp16")]; + tensor var_29230_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2735_cast_fp16)[name = tensor("op_29230_cast_fp16")]; + tensor var_29231_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2737_cast_fp16)[name = tensor("op_29231_cast_fp16")]; + tensor var_29232_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2739_cast_fp16)[name = tensor("op_29232_cast_fp16")]; + tensor var_29233_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2741_cast_fp16)[name = tensor("op_29233_cast_fp16")]; + tensor var_29234_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2743_cast_fp16)[name = tensor("op_29234_cast_fp16")]; + tensor var_29235_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2745_cast_fp16)[name = tensor("op_29235_cast_fp16")]; + tensor var_29236_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2747_cast_fp16)[name = tensor("op_29236_cast_fp16")]; + tensor var_29237_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2749_cast_fp16)[name = tensor("op_29237_cast_fp16")]; + tensor var_29238_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2751_cast_fp16)[name = tensor("op_29238_cast_fp16")]; + tensor var_29239_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2753_cast_fp16)[name = tensor("op_29239_cast_fp16")]; + tensor var_29240_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2755_cast_fp16)[name = tensor("op_29240_cast_fp16")]; + tensor var_29241_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2757_cast_fp16)[name = tensor("op_29241_cast_fp16")]; + tensor var_29242_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2759_cast_fp16)[name = tensor("op_29242_cast_fp16")]; + tensor var_29243_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2761_cast_fp16)[name = tensor("op_29243_cast_fp16")]; + tensor var_29244_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2763_cast_fp16)[name = tensor("op_29244_cast_fp16")]; + tensor var_29245_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2765_cast_fp16)[name = tensor("op_29245_cast_fp16")]; + tensor var_29246_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2767_cast_fp16)[name = tensor("op_29246_cast_fp16")]; + tensor var_29247_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2769_cast_fp16)[name = tensor("op_29247_cast_fp16")]; + tensor var_29248_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2771_cast_fp16)[name = tensor("op_29248_cast_fp16")]; + tensor var_29249_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2773_cast_fp16)[name = tensor("op_29249_cast_fp16")]; + tensor var_29250_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2775_cast_fp16)[name = tensor("op_29250_cast_fp16")]; + tensor var_29251_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2777_cast_fp16)[name = tensor("op_29251_cast_fp16")]; + tensor var_29252_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2779_cast_fp16)[name = tensor("op_29252_cast_fp16")]; + tensor var_29253_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2781_cast_fp16)[name = tensor("op_29253_cast_fp16")]; + tensor var_29254_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2783_cast_fp16)[name = tensor("op_29254_cast_fp16")]; + tensor var_29255_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2785_cast_fp16)[name = tensor("op_29255_cast_fp16")]; + tensor var_29256_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2787_cast_fp16)[name = tensor("op_29256_cast_fp16")]; + tensor var_29257_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2789_cast_fp16)[name = tensor("op_29257_cast_fp16")]; + tensor var_29258_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2791_cast_fp16)[name = tensor("op_29258_cast_fp16")]; + tensor var_29259_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2793_cast_fp16)[name = tensor("op_29259_cast_fp16")]; + tensor var_29260_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2795_cast_fp16)[name = tensor("op_29260_cast_fp16")]; + tensor var_29261_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2797_cast_fp16)[name = tensor("op_29261_cast_fp16")]; + tensor var_29262_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2799_cast_fp16)[name = tensor("op_29262_cast_fp16")]; + tensor var_29263_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2801_cast_fp16)[name = tensor("op_29263_cast_fp16")]; + tensor var_29264_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2803_cast_fp16)[name = tensor("op_29264_cast_fp16")]; + tensor var_29265_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2805_cast_fp16)[name = tensor("op_29265_cast_fp16")]; + tensor var_29266_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2807_cast_fp16)[name = tensor("op_29266_cast_fp16")]; + tensor var_29267_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2809_cast_fp16)[name = tensor("op_29267_cast_fp16")]; + tensor var_29268_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2811_cast_fp16)[name = tensor("op_29268_cast_fp16")]; + tensor var_29269_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2813_cast_fp16)[name = tensor("op_29269_cast_fp16")]; + tensor var_29270_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2815_cast_fp16)[name = tensor("op_29270_cast_fp16")]; + tensor var_29271_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2817_cast_fp16)[name = tensor("op_29271_cast_fp16")]; + tensor var_29272_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2819_cast_fp16)[name = tensor("op_29272_cast_fp16")]; + tensor var_29273_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2821_cast_fp16)[name = tensor("op_29273_cast_fp16")]; + tensor var_29274_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2823_cast_fp16)[name = tensor("op_29274_cast_fp16")]; + tensor var_29275_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2825_cast_fp16)[name = tensor("op_29275_cast_fp16")]; + tensor var_29276_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2827_cast_fp16)[name = tensor("op_29276_cast_fp16")]; + tensor var_29277_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2829_cast_fp16)[name = tensor("op_29277_cast_fp16")]; + tensor var_29278_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2831_cast_fp16)[name = tensor("op_29278_cast_fp16")]; + tensor var_29279_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2833_cast_fp16)[name = tensor("op_29279_cast_fp16")]; + tensor var_29280_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2835_cast_fp16)[name = tensor("op_29280_cast_fp16")]; + tensor var_29281_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2837_cast_fp16)[name = tensor("op_29281_cast_fp16")]; + tensor var_29282_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2839_cast_fp16)[name = tensor("op_29282_cast_fp16")]; + tensor var_29283_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2841_cast_fp16)[name = tensor("op_29283_cast_fp16")]; + tensor var_29284_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2843_cast_fp16)[name = tensor("op_29284_cast_fp16")]; + tensor var_29285_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2845_cast_fp16)[name = tensor("op_29285_cast_fp16")]; + tensor var_29286_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2847_cast_fp16)[name = tensor("op_29286_cast_fp16")]; + tensor var_29287_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2849_cast_fp16)[name = tensor("op_29287_cast_fp16")]; + tensor var_29288_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2851_cast_fp16)[name = tensor("op_29288_cast_fp16")]; + tensor var_29289_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2853_cast_fp16)[name = tensor("op_29289_cast_fp16")]; + tensor var_29290_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2855_cast_fp16)[name = tensor("op_29290_cast_fp16")]; + tensor var_29291_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2857_cast_fp16)[name = tensor("op_29291_cast_fp16")]; + tensor var_29292_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2859_cast_fp16)[name = tensor("op_29292_cast_fp16")]; + tensor var_29293_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2861_cast_fp16)[name = tensor("op_29293_cast_fp16")]; + tensor var_29294_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2863_cast_fp16)[name = tensor("op_29294_cast_fp16")]; + tensor var_29295_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2865_cast_fp16)[name = tensor("op_29295_cast_fp16")]; + tensor var_29296_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2867_cast_fp16)[name = tensor("op_29296_cast_fp16")]; + tensor var_29297_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2869_cast_fp16)[name = tensor("op_29297_cast_fp16")]; + tensor var_29298_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2871_cast_fp16)[name = tensor("op_29298_cast_fp16")]; + tensor var_29299_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2873_cast_fp16)[name = tensor("op_29299_cast_fp16")]; + tensor var_29300_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2875_cast_fp16)[name = tensor("op_29300_cast_fp16")]; + tensor var_29301_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2877_cast_fp16)[name = tensor("op_29301_cast_fp16")]; + tensor var_29302_cast_fp16 = softmax(axis = var_27994, x = aw_chunk_2879_cast_fp16)[name = tensor("op_29302_cast_fp16")]; + tensor var_29304_equation_0 = const()[name = tensor("op_29304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29304_cast_fp16 = einsum(equation = var_29304_equation_0, values = (var_28824_cast_fp16, var_29223_cast_fp16))[name = tensor("op_29304_cast_fp16")]; + tensor var_29306_equation_0 = const()[name = tensor("op_29306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29306_cast_fp16 = einsum(equation = var_29306_equation_0, values = (var_28824_cast_fp16, var_29224_cast_fp16))[name = tensor("op_29306_cast_fp16")]; + tensor var_29308_equation_0 = const()[name = tensor("op_29308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29308_cast_fp16 = einsum(equation = var_29308_equation_0, values = (var_28824_cast_fp16, var_29225_cast_fp16))[name = tensor("op_29308_cast_fp16")]; + tensor var_29310_equation_0 = const()[name = tensor("op_29310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29310_cast_fp16 = einsum(equation = var_29310_equation_0, values = (var_28824_cast_fp16, var_29226_cast_fp16))[name = tensor("op_29310_cast_fp16")]; + tensor var_29312_equation_0 = const()[name = tensor("op_29312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29312_cast_fp16 = einsum(equation = var_29312_equation_0, values = (var_28828_cast_fp16, var_29227_cast_fp16))[name = tensor("op_29312_cast_fp16")]; + tensor var_29314_equation_0 = const()[name = tensor("op_29314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29314_cast_fp16 = einsum(equation = var_29314_equation_0, values = (var_28828_cast_fp16, var_29228_cast_fp16))[name = tensor("op_29314_cast_fp16")]; + tensor var_29316_equation_0 = const()[name = tensor("op_29316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29316_cast_fp16 = einsum(equation = var_29316_equation_0, values = (var_28828_cast_fp16, var_29229_cast_fp16))[name = tensor("op_29316_cast_fp16")]; + tensor var_29318_equation_0 = const()[name = tensor("op_29318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29318_cast_fp16 = einsum(equation = var_29318_equation_0, values = (var_28828_cast_fp16, var_29230_cast_fp16))[name = tensor("op_29318_cast_fp16")]; + tensor var_29320_equation_0 = const()[name = tensor("op_29320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29320_cast_fp16 = einsum(equation = var_29320_equation_0, values = (var_28832_cast_fp16, var_29231_cast_fp16))[name = tensor("op_29320_cast_fp16")]; + tensor var_29322_equation_0 = const()[name = tensor("op_29322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29322_cast_fp16 = einsum(equation = var_29322_equation_0, values = (var_28832_cast_fp16, var_29232_cast_fp16))[name = tensor("op_29322_cast_fp16")]; + tensor var_29324_equation_0 = const()[name = tensor("op_29324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29324_cast_fp16 = einsum(equation = var_29324_equation_0, values = (var_28832_cast_fp16, var_29233_cast_fp16))[name = tensor("op_29324_cast_fp16")]; + tensor var_29326_equation_0 = const()[name = tensor("op_29326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29326_cast_fp16 = einsum(equation = var_29326_equation_0, values = (var_28832_cast_fp16, var_29234_cast_fp16))[name = tensor("op_29326_cast_fp16")]; + tensor var_29328_equation_0 = const()[name = tensor("op_29328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29328_cast_fp16 = einsum(equation = var_29328_equation_0, values = (var_28836_cast_fp16, var_29235_cast_fp16))[name = tensor("op_29328_cast_fp16")]; + tensor var_29330_equation_0 = const()[name = tensor("op_29330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29330_cast_fp16 = einsum(equation = var_29330_equation_0, values = (var_28836_cast_fp16, var_29236_cast_fp16))[name = tensor("op_29330_cast_fp16")]; + tensor var_29332_equation_0 = const()[name = tensor("op_29332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29332_cast_fp16 = einsum(equation = var_29332_equation_0, values = (var_28836_cast_fp16, var_29237_cast_fp16))[name = tensor("op_29332_cast_fp16")]; + tensor var_29334_equation_0 = const()[name = tensor("op_29334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29334_cast_fp16 = einsum(equation = var_29334_equation_0, values = (var_28836_cast_fp16, var_29238_cast_fp16))[name = tensor("op_29334_cast_fp16")]; + tensor var_29336_equation_0 = const()[name = tensor("op_29336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29336_cast_fp16 = einsum(equation = var_29336_equation_0, values = (var_28840_cast_fp16, var_29239_cast_fp16))[name = tensor("op_29336_cast_fp16")]; + tensor var_29338_equation_0 = const()[name = tensor("op_29338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29338_cast_fp16 = einsum(equation = var_29338_equation_0, values = (var_28840_cast_fp16, var_29240_cast_fp16))[name = tensor("op_29338_cast_fp16")]; + tensor var_29340_equation_0 = const()[name = tensor("op_29340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29340_cast_fp16 = einsum(equation = var_29340_equation_0, values = (var_28840_cast_fp16, var_29241_cast_fp16))[name = tensor("op_29340_cast_fp16")]; + tensor var_29342_equation_0 = const()[name = tensor("op_29342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29342_cast_fp16 = einsum(equation = var_29342_equation_0, values = (var_28840_cast_fp16, var_29242_cast_fp16))[name = tensor("op_29342_cast_fp16")]; + tensor var_29344_equation_0 = const()[name = tensor("op_29344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29344_cast_fp16 = einsum(equation = var_29344_equation_0, values = (var_28844_cast_fp16, var_29243_cast_fp16))[name = tensor("op_29344_cast_fp16")]; + tensor var_29346_equation_0 = const()[name = tensor("op_29346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29346_cast_fp16 = einsum(equation = var_29346_equation_0, values = (var_28844_cast_fp16, var_29244_cast_fp16))[name = tensor("op_29346_cast_fp16")]; + tensor var_29348_equation_0 = const()[name = tensor("op_29348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29348_cast_fp16 = einsum(equation = var_29348_equation_0, values = (var_28844_cast_fp16, var_29245_cast_fp16))[name = tensor("op_29348_cast_fp16")]; + tensor var_29350_equation_0 = const()[name = tensor("op_29350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29350_cast_fp16 = einsum(equation = var_29350_equation_0, values = (var_28844_cast_fp16, var_29246_cast_fp16))[name = tensor("op_29350_cast_fp16")]; + tensor var_29352_equation_0 = const()[name = tensor("op_29352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29352_cast_fp16 = einsum(equation = var_29352_equation_0, values = (var_28848_cast_fp16, var_29247_cast_fp16))[name = tensor("op_29352_cast_fp16")]; + tensor var_29354_equation_0 = const()[name = tensor("op_29354_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29354_cast_fp16 = einsum(equation = var_29354_equation_0, values = (var_28848_cast_fp16, var_29248_cast_fp16))[name = tensor("op_29354_cast_fp16")]; + tensor var_29356_equation_0 = const()[name = tensor("op_29356_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29356_cast_fp16 = einsum(equation = var_29356_equation_0, values = (var_28848_cast_fp16, var_29249_cast_fp16))[name = tensor("op_29356_cast_fp16")]; + tensor var_29358_equation_0 = const()[name = tensor("op_29358_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29358_cast_fp16 = einsum(equation = var_29358_equation_0, values = (var_28848_cast_fp16, var_29250_cast_fp16))[name = tensor("op_29358_cast_fp16")]; + tensor var_29360_equation_0 = const()[name = tensor("op_29360_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29360_cast_fp16 = einsum(equation = var_29360_equation_0, values = (var_28852_cast_fp16, var_29251_cast_fp16))[name = tensor("op_29360_cast_fp16")]; + tensor var_29362_equation_0 = const()[name = tensor("op_29362_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29362_cast_fp16 = einsum(equation = var_29362_equation_0, values = (var_28852_cast_fp16, var_29252_cast_fp16))[name = tensor("op_29362_cast_fp16")]; + tensor var_29364_equation_0 = const()[name = tensor("op_29364_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29364_cast_fp16 = einsum(equation = var_29364_equation_0, values = (var_28852_cast_fp16, var_29253_cast_fp16))[name = tensor("op_29364_cast_fp16")]; + tensor var_29366_equation_0 = const()[name = tensor("op_29366_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29366_cast_fp16 = einsum(equation = var_29366_equation_0, values = (var_28852_cast_fp16, var_29254_cast_fp16))[name = tensor("op_29366_cast_fp16")]; + tensor var_29368_equation_0 = const()[name = tensor("op_29368_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29368_cast_fp16 = einsum(equation = var_29368_equation_0, values = (var_28856_cast_fp16, var_29255_cast_fp16))[name = tensor("op_29368_cast_fp16")]; + tensor var_29370_equation_0 = const()[name = tensor("op_29370_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29370_cast_fp16 = einsum(equation = var_29370_equation_0, values = (var_28856_cast_fp16, var_29256_cast_fp16))[name = tensor("op_29370_cast_fp16")]; + tensor var_29372_equation_0 = const()[name = tensor("op_29372_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29372_cast_fp16 = einsum(equation = var_29372_equation_0, values = (var_28856_cast_fp16, var_29257_cast_fp16))[name = tensor("op_29372_cast_fp16")]; + tensor var_29374_equation_0 = const()[name = tensor("op_29374_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29374_cast_fp16 = einsum(equation = var_29374_equation_0, values = (var_28856_cast_fp16, var_29258_cast_fp16))[name = tensor("op_29374_cast_fp16")]; + tensor var_29376_equation_0 = const()[name = tensor("op_29376_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29376_cast_fp16 = einsum(equation = var_29376_equation_0, values = (var_28860_cast_fp16, var_29259_cast_fp16))[name = tensor("op_29376_cast_fp16")]; + tensor var_29378_equation_0 = const()[name = tensor("op_29378_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29378_cast_fp16 = einsum(equation = var_29378_equation_0, values = (var_28860_cast_fp16, var_29260_cast_fp16))[name = tensor("op_29378_cast_fp16")]; + tensor var_29380_equation_0 = const()[name = tensor("op_29380_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29380_cast_fp16 = einsum(equation = var_29380_equation_0, values = (var_28860_cast_fp16, var_29261_cast_fp16))[name = tensor("op_29380_cast_fp16")]; + tensor var_29382_equation_0 = const()[name = tensor("op_29382_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29382_cast_fp16 = einsum(equation = var_29382_equation_0, values = (var_28860_cast_fp16, var_29262_cast_fp16))[name = tensor("op_29382_cast_fp16")]; + tensor var_29384_equation_0 = const()[name = tensor("op_29384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29384_cast_fp16 = einsum(equation = var_29384_equation_0, values = (var_28864_cast_fp16, var_29263_cast_fp16))[name = tensor("op_29384_cast_fp16")]; + tensor var_29386_equation_0 = const()[name = tensor("op_29386_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29386_cast_fp16 = einsum(equation = var_29386_equation_0, values = (var_28864_cast_fp16, var_29264_cast_fp16))[name = tensor("op_29386_cast_fp16")]; + tensor var_29388_equation_0 = const()[name = tensor("op_29388_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29388_cast_fp16 = einsum(equation = var_29388_equation_0, values = (var_28864_cast_fp16, var_29265_cast_fp16))[name = tensor("op_29388_cast_fp16")]; + tensor var_29390_equation_0 = const()[name = tensor("op_29390_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29390_cast_fp16 = einsum(equation = var_29390_equation_0, values = (var_28864_cast_fp16, var_29266_cast_fp16))[name = tensor("op_29390_cast_fp16")]; + tensor var_29392_equation_0 = const()[name = tensor("op_29392_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29392_cast_fp16 = einsum(equation = var_29392_equation_0, values = (var_28868_cast_fp16, var_29267_cast_fp16))[name = tensor("op_29392_cast_fp16")]; + tensor var_29394_equation_0 = const()[name = tensor("op_29394_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29394_cast_fp16 = einsum(equation = var_29394_equation_0, values = (var_28868_cast_fp16, var_29268_cast_fp16))[name = tensor("op_29394_cast_fp16")]; + tensor var_29396_equation_0 = const()[name = tensor("op_29396_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29396_cast_fp16 = einsum(equation = var_29396_equation_0, values = (var_28868_cast_fp16, var_29269_cast_fp16))[name = tensor("op_29396_cast_fp16")]; + tensor var_29398_equation_0 = const()[name = tensor("op_29398_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29398_cast_fp16 = einsum(equation = var_29398_equation_0, values = (var_28868_cast_fp16, var_29270_cast_fp16))[name = tensor("op_29398_cast_fp16")]; + tensor var_29400_equation_0 = const()[name = tensor("op_29400_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29400_cast_fp16 = einsum(equation = var_29400_equation_0, values = (var_28872_cast_fp16, var_29271_cast_fp16))[name = tensor("op_29400_cast_fp16")]; + tensor var_29402_equation_0 = const()[name = tensor("op_29402_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29402_cast_fp16 = einsum(equation = var_29402_equation_0, values = (var_28872_cast_fp16, var_29272_cast_fp16))[name = tensor("op_29402_cast_fp16")]; + tensor var_29404_equation_0 = const()[name = tensor("op_29404_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29404_cast_fp16 = einsum(equation = var_29404_equation_0, values = (var_28872_cast_fp16, var_29273_cast_fp16))[name = tensor("op_29404_cast_fp16")]; + tensor var_29406_equation_0 = const()[name = tensor("op_29406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29406_cast_fp16 = einsum(equation = var_29406_equation_0, values = (var_28872_cast_fp16, var_29274_cast_fp16))[name = tensor("op_29406_cast_fp16")]; + tensor var_29408_equation_0 = const()[name = tensor("op_29408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29408_cast_fp16 = einsum(equation = var_29408_equation_0, values = (var_28876_cast_fp16, var_29275_cast_fp16))[name = tensor("op_29408_cast_fp16")]; + tensor var_29410_equation_0 = const()[name = tensor("op_29410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29410_cast_fp16 = einsum(equation = var_29410_equation_0, values = (var_28876_cast_fp16, var_29276_cast_fp16))[name = tensor("op_29410_cast_fp16")]; + tensor var_29412_equation_0 = const()[name = tensor("op_29412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29412_cast_fp16 = einsum(equation = var_29412_equation_0, values = (var_28876_cast_fp16, var_29277_cast_fp16))[name = tensor("op_29412_cast_fp16")]; + tensor var_29414_equation_0 = const()[name = tensor("op_29414_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29414_cast_fp16 = einsum(equation = var_29414_equation_0, values = (var_28876_cast_fp16, var_29278_cast_fp16))[name = tensor("op_29414_cast_fp16")]; + tensor var_29416_equation_0 = const()[name = tensor("op_29416_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29416_cast_fp16 = einsum(equation = var_29416_equation_0, values = (var_28880_cast_fp16, var_29279_cast_fp16))[name = tensor("op_29416_cast_fp16")]; + tensor var_29418_equation_0 = const()[name = tensor("op_29418_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29418_cast_fp16 = einsum(equation = var_29418_equation_0, values = (var_28880_cast_fp16, var_29280_cast_fp16))[name = tensor("op_29418_cast_fp16")]; + tensor var_29420_equation_0 = const()[name = tensor("op_29420_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29420_cast_fp16 = einsum(equation = var_29420_equation_0, values = (var_28880_cast_fp16, var_29281_cast_fp16))[name = tensor("op_29420_cast_fp16")]; + tensor var_29422_equation_0 = const()[name = tensor("op_29422_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29422_cast_fp16 = einsum(equation = var_29422_equation_0, values = (var_28880_cast_fp16, var_29282_cast_fp16))[name = tensor("op_29422_cast_fp16")]; + tensor var_29424_equation_0 = const()[name = tensor("op_29424_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29424_cast_fp16 = einsum(equation = var_29424_equation_0, values = (var_28884_cast_fp16, var_29283_cast_fp16))[name = tensor("op_29424_cast_fp16")]; + tensor var_29426_equation_0 = const()[name = tensor("op_29426_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29426_cast_fp16 = einsum(equation = var_29426_equation_0, values = (var_28884_cast_fp16, var_29284_cast_fp16))[name = tensor("op_29426_cast_fp16")]; + tensor var_29428_equation_0 = const()[name = tensor("op_29428_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29428_cast_fp16 = einsum(equation = var_29428_equation_0, values = (var_28884_cast_fp16, var_29285_cast_fp16))[name = tensor("op_29428_cast_fp16")]; + tensor var_29430_equation_0 = const()[name = tensor("op_29430_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29430_cast_fp16 = einsum(equation = var_29430_equation_0, values = (var_28884_cast_fp16, var_29286_cast_fp16))[name = tensor("op_29430_cast_fp16")]; + tensor var_29432_equation_0 = const()[name = tensor("op_29432_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29432_cast_fp16 = einsum(equation = var_29432_equation_0, values = (var_28888_cast_fp16, var_29287_cast_fp16))[name = tensor("op_29432_cast_fp16")]; + tensor var_29434_equation_0 = const()[name = tensor("op_29434_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29434_cast_fp16 = einsum(equation = var_29434_equation_0, values = (var_28888_cast_fp16, var_29288_cast_fp16))[name = tensor("op_29434_cast_fp16")]; + tensor var_29436_equation_0 = const()[name = tensor("op_29436_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29436_cast_fp16 = einsum(equation = var_29436_equation_0, values = (var_28888_cast_fp16, var_29289_cast_fp16))[name = tensor("op_29436_cast_fp16")]; + tensor var_29438_equation_0 = const()[name = tensor("op_29438_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29438_cast_fp16 = einsum(equation = var_29438_equation_0, values = (var_28888_cast_fp16, var_29290_cast_fp16))[name = tensor("op_29438_cast_fp16")]; + tensor var_29440_equation_0 = const()[name = tensor("op_29440_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29440_cast_fp16 = einsum(equation = var_29440_equation_0, values = (var_28892_cast_fp16, var_29291_cast_fp16))[name = tensor("op_29440_cast_fp16")]; + tensor var_29442_equation_0 = const()[name = tensor("op_29442_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29442_cast_fp16 = einsum(equation = var_29442_equation_0, values = (var_28892_cast_fp16, var_29292_cast_fp16))[name = tensor("op_29442_cast_fp16")]; + tensor var_29444_equation_0 = const()[name = tensor("op_29444_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29444_cast_fp16 = einsum(equation = var_29444_equation_0, values = (var_28892_cast_fp16, var_29293_cast_fp16))[name = tensor("op_29444_cast_fp16")]; + tensor var_29446_equation_0 = const()[name = tensor("op_29446_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29446_cast_fp16 = einsum(equation = var_29446_equation_0, values = (var_28892_cast_fp16, var_29294_cast_fp16))[name = tensor("op_29446_cast_fp16")]; + tensor var_29448_equation_0 = const()[name = tensor("op_29448_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29448_cast_fp16 = einsum(equation = var_29448_equation_0, values = (var_28896_cast_fp16, var_29295_cast_fp16))[name = tensor("op_29448_cast_fp16")]; + tensor var_29450_equation_0 = const()[name = tensor("op_29450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29450_cast_fp16 = einsum(equation = var_29450_equation_0, values = (var_28896_cast_fp16, var_29296_cast_fp16))[name = tensor("op_29450_cast_fp16")]; + tensor var_29452_equation_0 = const()[name = tensor("op_29452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29452_cast_fp16 = einsum(equation = var_29452_equation_0, values = (var_28896_cast_fp16, var_29297_cast_fp16))[name = tensor("op_29452_cast_fp16")]; + tensor var_29454_equation_0 = const()[name = tensor("op_29454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29454_cast_fp16 = einsum(equation = var_29454_equation_0, values = (var_28896_cast_fp16, var_29298_cast_fp16))[name = tensor("op_29454_cast_fp16")]; + tensor var_29456_equation_0 = const()[name = tensor("op_29456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29456_cast_fp16 = einsum(equation = var_29456_equation_0, values = (var_28900_cast_fp16, var_29299_cast_fp16))[name = tensor("op_29456_cast_fp16")]; + tensor var_29458_equation_0 = const()[name = tensor("op_29458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29458_cast_fp16 = einsum(equation = var_29458_equation_0, values = (var_28900_cast_fp16, var_29300_cast_fp16))[name = tensor("op_29458_cast_fp16")]; + tensor var_29460_equation_0 = const()[name = tensor("op_29460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29460_cast_fp16 = einsum(equation = var_29460_equation_0, values = (var_28900_cast_fp16, var_29301_cast_fp16))[name = tensor("op_29460_cast_fp16")]; + tensor var_29462_equation_0 = const()[name = tensor("op_29462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_29462_cast_fp16 = einsum(equation = var_29462_equation_0, values = (var_28900_cast_fp16, var_29302_cast_fp16))[name = tensor("op_29462_cast_fp16")]; + tensor var_29464_interleave_0 = const()[name = tensor("op_29464_interleave_0"), val = tensor(false)]; + tensor var_29464_cast_fp16 = concat(axis = var_27969, interleave = var_29464_interleave_0, values = (var_29304_cast_fp16, var_29306_cast_fp16, var_29308_cast_fp16, var_29310_cast_fp16))[name = tensor("op_29464_cast_fp16")]; + tensor var_29466_interleave_0 = const()[name = tensor("op_29466_interleave_0"), val = tensor(false)]; + tensor var_29466_cast_fp16 = concat(axis = var_27969, interleave = var_29466_interleave_0, values = (var_29312_cast_fp16, var_29314_cast_fp16, var_29316_cast_fp16, var_29318_cast_fp16))[name = tensor("op_29466_cast_fp16")]; + tensor var_29468_interleave_0 = const()[name = tensor("op_29468_interleave_0"), val = tensor(false)]; + tensor var_29468_cast_fp16 = concat(axis = var_27969, interleave = var_29468_interleave_0, values = (var_29320_cast_fp16, var_29322_cast_fp16, var_29324_cast_fp16, var_29326_cast_fp16))[name = tensor("op_29468_cast_fp16")]; + tensor var_29470_interleave_0 = const()[name = tensor("op_29470_interleave_0"), val = tensor(false)]; + tensor var_29470_cast_fp16 = concat(axis = var_27969, interleave = var_29470_interleave_0, values = (var_29328_cast_fp16, var_29330_cast_fp16, var_29332_cast_fp16, var_29334_cast_fp16))[name = tensor("op_29470_cast_fp16")]; + tensor var_29472_interleave_0 = const()[name = tensor("op_29472_interleave_0"), val = tensor(false)]; + tensor var_29472_cast_fp16 = concat(axis = var_27969, interleave = var_29472_interleave_0, values = (var_29336_cast_fp16, var_29338_cast_fp16, var_29340_cast_fp16, var_29342_cast_fp16))[name = tensor("op_29472_cast_fp16")]; + tensor var_29474_interleave_0 = const()[name = tensor("op_29474_interleave_0"), val = tensor(false)]; + tensor var_29474_cast_fp16 = concat(axis = var_27969, interleave = var_29474_interleave_0, values = (var_29344_cast_fp16, var_29346_cast_fp16, var_29348_cast_fp16, var_29350_cast_fp16))[name = tensor("op_29474_cast_fp16")]; + tensor var_29476_interleave_0 = const()[name = tensor("op_29476_interleave_0"), val = tensor(false)]; + tensor var_29476_cast_fp16 = concat(axis = var_27969, interleave = var_29476_interleave_0, values = (var_29352_cast_fp16, var_29354_cast_fp16, var_29356_cast_fp16, var_29358_cast_fp16))[name = tensor("op_29476_cast_fp16")]; + tensor var_29478_interleave_0 = const()[name = tensor("op_29478_interleave_0"), val = tensor(false)]; + tensor var_29478_cast_fp16 = concat(axis = var_27969, interleave = var_29478_interleave_0, values = (var_29360_cast_fp16, var_29362_cast_fp16, var_29364_cast_fp16, var_29366_cast_fp16))[name = tensor("op_29478_cast_fp16")]; + tensor var_29480_interleave_0 = const()[name = tensor("op_29480_interleave_0"), val = tensor(false)]; + tensor var_29480_cast_fp16 = concat(axis = var_27969, interleave = var_29480_interleave_0, values = (var_29368_cast_fp16, var_29370_cast_fp16, var_29372_cast_fp16, var_29374_cast_fp16))[name = tensor("op_29480_cast_fp16")]; + tensor var_29482_interleave_0 = const()[name = tensor("op_29482_interleave_0"), val = tensor(false)]; + tensor var_29482_cast_fp16 = concat(axis = var_27969, interleave = var_29482_interleave_0, values = (var_29376_cast_fp16, var_29378_cast_fp16, var_29380_cast_fp16, var_29382_cast_fp16))[name = tensor("op_29482_cast_fp16")]; + tensor var_29484_interleave_0 = const()[name = tensor("op_29484_interleave_0"), val = tensor(false)]; + tensor var_29484_cast_fp16 = concat(axis = var_27969, interleave = var_29484_interleave_0, values = (var_29384_cast_fp16, var_29386_cast_fp16, var_29388_cast_fp16, var_29390_cast_fp16))[name = tensor("op_29484_cast_fp16")]; + tensor var_29486_interleave_0 = const()[name = tensor("op_29486_interleave_0"), val = tensor(false)]; + tensor var_29486_cast_fp16 = concat(axis = var_27969, interleave = var_29486_interleave_0, values = (var_29392_cast_fp16, var_29394_cast_fp16, var_29396_cast_fp16, var_29398_cast_fp16))[name = tensor("op_29486_cast_fp16")]; + tensor var_29488_interleave_0 = const()[name = tensor("op_29488_interleave_0"), val = tensor(false)]; + tensor var_29488_cast_fp16 = concat(axis = var_27969, interleave = var_29488_interleave_0, values = (var_29400_cast_fp16, var_29402_cast_fp16, var_29404_cast_fp16, var_29406_cast_fp16))[name = tensor("op_29488_cast_fp16")]; + tensor var_29490_interleave_0 = const()[name = tensor("op_29490_interleave_0"), val = tensor(false)]; + tensor var_29490_cast_fp16 = concat(axis = var_27969, interleave = var_29490_interleave_0, values = (var_29408_cast_fp16, var_29410_cast_fp16, var_29412_cast_fp16, var_29414_cast_fp16))[name = tensor("op_29490_cast_fp16")]; + tensor var_29492_interleave_0 = const()[name = tensor("op_29492_interleave_0"), val = tensor(false)]; + tensor var_29492_cast_fp16 = concat(axis = var_27969, interleave = var_29492_interleave_0, values = (var_29416_cast_fp16, var_29418_cast_fp16, var_29420_cast_fp16, var_29422_cast_fp16))[name = tensor("op_29492_cast_fp16")]; + tensor var_29494_interleave_0 = const()[name = tensor("op_29494_interleave_0"), val = tensor(false)]; + tensor var_29494_cast_fp16 = concat(axis = var_27969, interleave = var_29494_interleave_0, values = (var_29424_cast_fp16, var_29426_cast_fp16, var_29428_cast_fp16, var_29430_cast_fp16))[name = tensor("op_29494_cast_fp16")]; + tensor var_29496_interleave_0 = const()[name = tensor("op_29496_interleave_0"), val = tensor(false)]; + tensor var_29496_cast_fp16 = concat(axis = var_27969, interleave = var_29496_interleave_0, values = (var_29432_cast_fp16, var_29434_cast_fp16, var_29436_cast_fp16, var_29438_cast_fp16))[name = tensor("op_29496_cast_fp16")]; + tensor var_29498_interleave_0 = const()[name = tensor("op_29498_interleave_0"), val = tensor(false)]; + tensor var_29498_cast_fp16 = concat(axis = var_27969, interleave = var_29498_interleave_0, values = (var_29440_cast_fp16, var_29442_cast_fp16, var_29444_cast_fp16, var_29446_cast_fp16))[name = tensor("op_29498_cast_fp16")]; + tensor var_29500_interleave_0 = const()[name = tensor("op_29500_interleave_0"), val = tensor(false)]; + tensor var_29500_cast_fp16 = concat(axis = var_27969, interleave = var_29500_interleave_0, values = (var_29448_cast_fp16, var_29450_cast_fp16, var_29452_cast_fp16, var_29454_cast_fp16))[name = tensor("op_29500_cast_fp16")]; + tensor var_29502_interleave_0 = const()[name = tensor("op_29502_interleave_0"), val = tensor(false)]; + tensor var_29502_cast_fp16 = concat(axis = var_27969, interleave = var_29502_interleave_0, values = (var_29456_cast_fp16, var_29458_cast_fp16, var_29460_cast_fp16, var_29462_cast_fp16))[name = tensor("op_29502_cast_fp16")]; + tensor input_347_interleave_0 = const()[name = tensor("input_347_interleave_0"), val = tensor(false)]; + tensor input_347_cast_fp16 = concat(axis = var_27994, interleave = input_347_interleave_0, values = (var_29464_cast_fp16, var_29466_cast_fp16, var_29468_cast_fp16, var_29470_cast_fp16, var_29472_cast_fp16, var_29474_cast_fp16, var_29476_cast_fp16, var_29478_cast_fp16, var_29480_cast_fp16, var_29482_cast_fp16, var_29484_cast_fp16, var_29486_cast_fp16, var_29488_cast_fp16, var_29490_cast_fp16, var_29492_cast_fp16, var_29494_cast_fp16, var_29496_cast_fp16, var_29498_cast_fp16, var_29500_cast_fp16, var_29502_cast_fp16))[name = tensor("input_347_cast_fp16")]; + tensor var_29510 = const()[name = tensor("op_29510"), val = tensor([1, 1])]; + tensor var_29512 = const()[name = tensor("op_29512"), val = tensor([1, 1])]; + tensor pretrained_out_211_pad_type_0 = const()[name = tensor("pretrained_out_211_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_211_pad_0 = const()[name = tensor("pretrained_out_211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197235904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198055168))), name = tensor("layers_17_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_17_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_17_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198055296)))]; + tensor pretrained_out_211_cast_fp16 = conv(bias = layers_17_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_29512, groups = var_27994, pad = pretrained_out_211_pad_0, pad_type = pretrained_out_211_pad_type_0, strides = var_29510, weight = layers_17_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = tensor("pretrained_out_211_cast_fp16")]; + tensor var_29516 = const()[name = tensor("op_29516"), val = tensor([1, 1])]; + tensor var_29518 = const()[name = tensor("op_29518"), val = tensor([1, 1])]; + tensor input_349_pad_type_0 = const()[name = tensor("input_349_pad_type_0"), val = tensor("custom")]; + tensor input_349_pad_0 = const()[name = tensor("input_349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_17_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198057920)))]; + tensor input_349_cast_fp16 = conv(dilations = var_29518, groups = var_27994, pad = input_349_pad_0, pad_type = input_349_pad_type_0, strides = var_29516, weight = layers_17_self_attn_o_proj_loraA_weight_to_fp16, x = input_347_cast_fp16)[name = tensor("input_349_cast_fp16")]; + tensor var_29522 = const()[name = tensor("op_29522"), val = tensor([1, 1])]; + tensor var_29524 = const()[name = tensor("op_29524"), val = tensor([1, 1])]; + tensor lora_out_421_pad_type_0 = const()[name = tensor("lora_out_421_pad_type_0"), val = tensor("custom")]; + tensor lora_out_421_pad_0 = const()[name = tensor("lora_out_421_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_423_weight_0_to_fp16 = const()[name = tensor("lora_out_423_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198098944)))]; + tensor lora_out_423_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_29524, groups = var_27994, pad = lora_out_421_pad_0, pad_type = lora_out_421_pad_type_0, strides = var_29522, weight = lora_out_423_weight_0_to_fp16, x = input_349_cast_fp16)[name = tensor("lora_out_423_cast_fp16")]; + tensor obj_71_cast_fp16 = add(x = pretrained_out_211_cast_fp16, y = lora_out_423_cast_fp16)[name = tensor("obj_71_cast_fp16")]; + tensor inputs_71_cast_fp16 = add(x = inputs_69_cast_fp16, y = obj_71_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; + tensor var_29533 = const()[name = tensor("op_29533"), val = tensor([1])]; + tensor channels_mean_71_cast_fp16 = reduce_mean(axes = var_29533, keep_dims = var_27995, x = inputs_71_cast_fp16)[name = tensor("channels_mean_71_cast_fp16")]; + tensor zero_mean_71_cast_fp16 = sub(x = inputs_71_cast_fp16, y = channels_mean_71_cast_fp16)[name = tensor("zero_mean_71_cast_fp16")]; + tensor zero_mean_sq_71_cast_fp16 = mul(x = zero_mean_71_cast_fp16, y = zero_mean_71_cast_fp16)[name = tensor("zero_mean_sq_71_cast_fp16")]; + tensor var_29537 = const()[name = tensor("op_29537"), val = tensor([1])]; + tensor var_29538_cast_fp16 = reduce_mean(axes = var_29537, keep_dims = var_27995, x = zero_mean_sq_71_cast_fp16)[name = tensor("op_29538_cast_fp16")]; + tensor var_29539_to_fp16 = const()[name = tensor("op_29539_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_29540_cast_fp16 = add(x = var_29538_cast_fp16, y = var_29539_to_fp16)[name = tensor("op_29540_cast_fp16")]; + tensor denom_71_epsilon_0 = const()[name = tensor("denom_71_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_71_cast_fp16 = rsqrt(epsilon = denom_71_epsilon_0, x = var_29540_cast_fp16)[name = tensor("denom_71_cast_fp16")]; + tensor out_71_cast_fp16 = mul(x = zero_mean_71_cast_fp16, y = denom_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; + tensor input_351_gamma_0_to_fp16 = const()[name = tensor("input_351_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198139968)))]; + tensor input_351_beta_0_to_fp16 = const()[name = tensor("input_351_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198142592)))]; + tensor input_351_epsilon_0_to_fp16 = const()[name = tensor("input_351_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_351_cast_fp16 = batch_norm(beta = input_351_beta_0_to_fp16, epsilon = input_351_epsilon_0_to_fp16, gamma = input_351_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor("input_351_cast_fp16")]; + tensor var_29554 = const()[name = tensor("op_29554"), val = tensor([1, 1])]; + tensor var_29556 = const()[name = tensor("op_29556"), val = tensor([1, 1])]; + tensor pretrained_out_213_pad_type_0 = const()[name = tensor("pretrained_out_213_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_213_pad_0 = const()[name = tensor("pretrained_out_213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198145216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201422080))), name = tensor("layers_17_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_17_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_17_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201422208)))]; + tensor pretrained_out_213_cast_fp16 = conv(bias = layers_17_fc1_pretrained_bias_to_fp16, dilations = var_29556, groups = var_27994, pad = pretrained_out_213_pad_0, pad_type = pretrained_out_213_pad_type_0, strides = var_29554, weight = layers_17_fc1_pretrained_weight_to_fp16_palettized, x = input_351_cast_fp16)[name = tensor("pretrained_out_213_cast_fp16")]; + tensor var_29560 = const()[name = tensor("op_29560"), val = tensor([1, 1])]; + tensor var_29562 = const()[name = tensor("op_29562"), val = tensor([1, 1])]; + tensor input_353_pad_type_0 = const()[name = tensor("input_353_pad_type_0"), val = tensor("custom")]; + tensor input_353_pad_0 = const()[name = tensor("input_353_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_17_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201432512)))]; + tensor input_353_cast_fp16 = conv(dilations = var_29562, groups = var_27994, pad = input_353_pad_0, pad_type = input_353_pad_type_0, strides = var_29560, weight = layers_17_fc1_loraA_weight_to_fp16, x = input_351_cast_fp16)[name = tensor("input_353_cast_fp16")]; + tensor var_29566 = const()[name = tensor("op_29566"), val = tensor([1, 1])]; + tensor var_29568 = const()[name = tensor("op_29568"), val = tensor([1, 1])]; + tensor lora_out_425_pad_type_0 = const()[name = tensor("lora_out_425_pad_type_0"), val = tensor("custom")]; + tensor lora_out_425_pad_0 = const()[name = tensor("lora_out_425_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_427_weight_0_to_fp16 = const()[name = tensor("lora_out_427_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201473536)))]; + tensor lora_out_427_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_29568, groups = var_27994, pad = lora_out_425_pad_0, pad_type = lora_out_425_pad_type_0, strides = var_29566, weight = lora_out_427_weight_0_to_fp16, x = input_353_cast_fp16)[name = tensor("lora_out_427_cast_fp16")]; + tensor input_355_cast_fp16 = add(x = pretrained_out_213_cast_fp16, y = lora_out_427_cast_fp16)[name = tensor("input_355_cast_fp16")]; + tensor input_357_mode_0 = const()[name = tensor("input_357_mode_0"), val = tensor("EXACT")]; + tensor input_357_cast_fp16 = gelu(mode = input_357_mode_0, x = input_355_cast_fp16)[name = tensor("input_357_cast_fp16")]; + tensor var_29580 = const()[name = tensor("op_29580"), val = tensor([1, 1])]; + tensor var_29582 = const()[name = tensor("op_29582"), val = tensor([1, 1])]; + tensor pretrained_out_215_pad_type_0 = const()[name = tensor("pretrained_out_215_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_215_pad_0 = const()[name = tensor("pretrained_out_215_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201637440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204914304))), name = tensor("layers_17_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_17_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_17_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204914432)))]; + tensor pretrained_out_215_cast_fp16 = conv(bias = layers_17_fc2_pretrained_bias_to_fp16, dilations = var_29582, groups = var_27994, pad = pretrained_out_215_pad_0, pad_type = pretrained_out_215_pad_type_0, strides = var_29580, weight = layers_17_fc2_pretrained_weight_to_fp16_palettized, x = input_357_cast_fp16)[name = tensor("pretrained_out_215_cast_fp16")]; + tensor var_29586 = const()[name = tensor("op_29586"), val = tensor([1, 1])]; + tensor var_29588 = const()[name = tensor("op_29588"), val = tensor([1, 1])]; + tensor input_359_pad_type_0 = const()[name = tensor("input_359_pad_type_0"), val = tensor("custom")]; + tensor input_359_pad_0 = const()[name = tensor("input_359_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_17_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_17_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204917056)))]; + tensor input_359_cast_fp16 = conv(dilations = var_29588, groups = var_27994, pad = input_359_pad_0, pad_type = input_359_pad_type_0, strides = var_29586, weight = layers_17_fc2_loraA_weight_to_fp16, x = input_357_cast_fp16)[name = tensor("input_359_cast_fp16")]; + tensor var_29592 = const()[name = tensor("op_29592"), val = tensor([1, 1])]; + tensor var_29594 = const()[name = tensor("op_29594"), val = tensor([1, 1])]; + tensor lora_out_429_pad_type_0 = const()[name = tensor("lora_out_429_pad_type_0"), val = tensor("custom")]; + tensor lora_out_429_pad_0 = const()[name = tensor("lora_out_429_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_431_weight_0_to_fp16 = const()[name = tensor("lora_out_431_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205080960)))]; + tensor lora_out_431_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_29594, groups = var_27994, pad = lora_out_429_pad_0, pad_type = lora_out_429_pad_type_0, strides = var_29592, weight = lora_out_431_weight_0_to_fp16, x = input_359_cast_fp16)[name = tensor("lora_out_431_cast_fp16")]; + tensor hidden_states_39_cast_fp16 = add(x = pretrained_out_215_cast_fp16, y = lora_out_431_cast_fp16)[name = tensor("hidden_states_39_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = hidden_states_39_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; + tensor var_29604 = const()[name = tensor("op_29604"), val = tensor(3)]; + tensor var_29629 = const()[name = tensor("op_29629"), val = tensor(1)]; + tensor var_29630 = const()[name = tensor("op_29630"), val = tensor(true)]; + tensor var_29640 = const()[name = tensor("op_29640"), val = tensor([1])]; + tensor channels_mean_73_cast_fp16 = reduce_mean(axes = var_29640, keep_dims = var_29630, x = inputs_73_cast_fp16)[name = tensor("channels_mean_73_cast_fp16")]; + tensor zero_mean_73_cast_fp16 = sub(x = inputs_73_cast_fp16, y = channels_mean_73_cast_fp16)[name = tensor("zero_mean_73_cast_fp16")]; + tensor zero_mean_sq_73_cast_fp16 = mul(x = zero_mean_73_cast_fp16, y = zero_mean_73_cast_fp16)[name = tensor("zero_mean_sq_73_cast_fp16")]; + tensor var_29644 = const()[name = tensor("op_29644"), val = tensor([1])]; + tensor var_29645_cast_fp16 = reduce_mean(axes = var_29644, keep_dims = var_29630, x = zero_mean_sq_73_cast_fp16)[name = tensor("op_29645_cast_fp16")]; + tensor var_29646_to_fp16 = const()[name = tensor("op_29646_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_29647_cast_fp16 = add(x = var_29645_cast_fp16, y = var_29646_to_fp16)[name = tensor("op_29647_cast_fp16")]; + tensor denom_73_epsilon_0 = const()[name = tensor("denom_73_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_73_cast_fp16 = rsqrt(epsilon = denom_73_epsilon_0, x = var_29647_cast_fp16)[name = tensor("denom_73_cast_fp16")]; + tensor out_73_cast_fp16 = mul(x = zero_mean_73_cast_fp16, y = denom_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; + tensor obj_73_gamma_0_to_fp16 = const()[name = tensor("obj_73_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205121984)))]; + tensor obj_73_beta_0_to_fp16 = const()[name = tensor("obj_73_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205124608)))]; + tensor obj_73_epsilon_0_to_fp16 = const()[name = tensor("obj_73_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_73_cast_fp16 = batch_norm(beta = obj_73_beta_0_to_fp16, epsilon = obj_73_epsilon_0_to_fp16, gamma = obj_73_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor("obj_73_cast_fp16")]; + tensor var_29665 = const()[name = tensor("op_29665"), val = tensor([1, 1])]; + tensor var_29667 = const()[name = tensor("op_29667"), val = tensor([1, 1])]; + tensor pretrained_out_217_pad_type_0 = const()[name = tensor("pretrained_out_217_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_217_pad_0 = const()[name = tensor("pretrained_out_217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205127232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205946496))), name = tensor("layers_18_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_18_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205946624)))]; + tensor pretrained_out_217_cast_fp16 = conv(bias = layers_18_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_29667, groups = var_29629, pad = pretrained_out_217_pad_0, pad_type = pretrained_out_217_pad_type_0, strides = var_29665, weight = layers_18_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_73_cast_fp16)[name = tensor("pretrained_out_217_cast_fp16")]; + tensor var_29671 = const()[name = tensor("op_29671"), val = tensor([1, 1])]; + tensor var_29673 = const()[name = tensor("op_29673"), val = tensor([1, 1])]; + tensor input_361_pad_type_0 = const()[name = tensor("input_361_pad_type_0"), val = tensor("custom")]; + tensor input_361_pad_0 = const()[name = tensor("input_361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205949248)))]; + tensor input_361_cast_fp16 = conv(dilations = var_29673, groups = var_29629, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = var_29671, weight = layers_18_self_attn_q_proj_loraA_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor("input_361_cast_fp16")]; + tensor var_29677 = const()[name = tensor("op_29677"), val = tensor([1, 1])]; + tensor var_29679 = const()[name = tensor("op_29679"), val = tensor([1, 1])]; + tensor lora_out_433_pad_type_0 = const()[name = tensor("lora_out_433_pad_type_0"), val = tensor("custom")]; + tensor lora_out_433_pad_0 = const()[name = tensor("lora_out_433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_435_weight_0_to_fp16 = const()[name = tensor("lora_out_435_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(205990272)))]; + tensor lora_out_435_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_29679, groups = var_29629, pad = lora_out_433_pad_0, pad_type = lora_out_433_pad_type_0, strides = var_29677, weight = lora_out_435_weight_0_to_fp16, x = input_361_cast_fp16)[name = tensor("lora_out_435_cast_fp16")]; + tensor query_37_cast_fp16 = add(x = pretrained_out_217_cast_fp16, y = lora_out_435_cast_fp16)[name = tensor("query_37_cast_fp16")]; + tensor var_29689 = const()[name = tensor("op_29689"), val = tensor([1, 1])]; + tensor var_29691 = const()[name = tensor("op_29691"), val = tensor([1, 1])]; + tensor pretrained_out_219_pad_type_0 = const()[name = tensor("pretrained_out_219_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_219_pad_0 = const()[name = tensor("pretrained_out_219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206031296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206850560))), name = tensor("layers_18_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_219_cast_fp16 = conv(dilations = var_29691, groups = var_29629, pad = pretrained_out_219_pad_0, pad_type = pretrained_out_219_pad_type_0, strides = var_29689, weight = layers_18_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_73_cast_fp16)[name = tensor("pretrained_out_219_cast_fp16")]; + tensor var_29695 = const()[name = tensor("op_29695"), val = tensor([1, 1])]; + tensor var_29697 = const()[name = tensor("op_29697"), val = tensor([1, 1])]; + tensor input_363_pad_type_0 = const()[name = tensor("input_363_pad_type_0"), val = tensor("custom")]; + tensor input_363_pad_0 = const()[name = tensor("input_363_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206850688)))]; + tensor input_363_cast_fp16 = conv(dilations = var_29697, groups = var_29629, pad = input_363_pad_0, pad_type = input_363_pad_type_0, strides = var_29695, weight = layers_18_self_attn_k_proj_loraA_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor("input_363_cast_fp16")]; + tensor var_29701 = const()[name = tensor("op_29701"), val = tensor([1, 1])]; + tensor var_29703 = const()[name = tensor("op_29703"), val = tensor([1, 1])]; + tensor lora_out_437_pad_type_0 = const()[name = tensor("lora_out_437_pad_type_0"), val = tensor("custom")]; + tensor lora_out_437_pad_0 = const()[name = tensor("lora_out_437_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_439_weight_0_to_fp16 = const()[name = tensor("lora_out_439_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206891712)))]; + tensor lora_out_439_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_29703, groups = var_29629, pad = lora_out_437_pad_0, pad_type = lora_out_437_pad_type_0, strides = var_29701, weight = lora_out_439_weight_0_to_fp16, x = input_363_cast_fp16)[name = tensor("lora_out_439_cast_fp16")]; + tensor key_37_cast_fp16 = add(x = pretrained_out_219_cast_fp16, y = lora_out_439_cast_fp16)[name = tensor("key_37_cast_fp16")]; + tensor var_29714 = const()[name = tensor("op_29714"), val = tensor([1, 1])]; + tensor var_29716 = const()[name = tensor("op_29716"), val = tensor([1, 1])]; + tensor pretrained_out_221_pad_type_0 = const()[name = tensor("pretrained_out_221_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_221_pad_0 = const()[name = tensor("pretrained_out_221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(206932736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207752000))), name = tensor("layers_18_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_18_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207752128)))]; + tensor pretrained_out_221_cast_fp16 = conv(bias = layers_18_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_29716, groups = var_29629, pad = pretrained_out_221_pad_0, pad_type = pretrained_out_221_pad_type_0, strides = var_29714, weight = layers_18_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_73_cast_fp16)[name = tensor("pretrained_out_221_cast_fp16")]; + tensor var_29720 = const()[name = tensor("op_29720"), val = tensor([1, 1])]; + tensor var_29722 = const()[name = tensor("op_29722"), val = tensor([1, 1])]; + tensor input_365_pad_type_0 = const()[name = tensor("input_365_pad_type_0"), val = tensor("custom")]; + tensor input_365_pad_0 = const()[name = tensor("input_365_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207754752)))]; + tensor input_365_cast_fp16 = conv(dilations = var_29722, groups = var_29629, pad = input_365_pad_0, pad_type = input_365_pad_type_0, strides = var_29720, weight = layers_18_self_attn_v_proj_loraA_weight_to_fp16, x = obj_73_cast_fp16)[name = tensor("input_365_cast_fp16")]; + tensor var_29726 = const()[name = tensor("op_29726"), val = tensor([1, 1])]; + tensor var_29728 = const()[name = tensor("op_29728"), val = tensor([1, 1])]; + tensor lora_out_441_pad_type_0 = const()[name = tensor("lora_out_441_pad_type_0"), val = tensor("custom")]; + tensor lora_out_441_pad_0 = const()[name = tensor("lora_out_441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_443_weight_0_to_fp16 = const()[name = tensor("lora_out_443_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207795776)))]; + tensor lora_out_443_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_29728, groups = var_29629, pad = lora_out_441_pad_0, pad_type = lora_out_441_pad_type_0, strides = var_29726, weight = lora_out_443_weight_0_to_fp16, x = input_365_cast_fp16)[name = tensor("lora_out_443_cast_fp16")]; + tensor value_37_cast_fp16 = add(x = pretrained_out_221_cast_fp16, y = lora_out_443_cast_fp16)[name = tensor("value_37_cast_fp16")]; + tensor var_29738_begin_0 = const()[name = tensor("op_29738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29738_end_0 = const()[name = tensor("op_29738_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29738_end_mask_0 = const()[name = tensor("op_29738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29738_cast_fp16 = slice_by_index(begin = var_29738_begin_0, end = var_29738_end_0, end_mask = var_29738_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29738_cast_fp16")]; + tensor var_29742_begin_0 = const()[name = tensor("op_29742_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_29742_end_0 = const()[name = tensor("op_29742_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_29742_end_mask_0 = const()[name = tensor("op_29742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29742_cast_fp16 = slice_by_index(begin = var_29742_begin_0, end = var_29742_end_0, end_mask = var_29742_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29742_cast_fp16")]; + tensor var_29746_begin_0 = const()[name = tensor("op_29746_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_29746_end_0 = const()[name = tensor("op_29746_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_29746_end_mask_0 = const()[name = tensor("op_29746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29746_cast_fp16 = slice_by_index(begin = var_29746_begin_0, end = var_29746_end_0, end_mask = var_29746_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29746_cast_fp16")]; + tensor var_29750_begin_0 = const()[name = tensor("op_29750_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_29750_end_0 = const()[name = tensor("op_29750_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_29750_end_mask_0 = const()[name = tensor("op_29750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29750_cast_fp16 = slice_by_index(begin = var_29750_begin_0, end = var_29750_end_0, end_mask = var_29750_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29750_cast_fp16")]; + tensor var_29754_begin_0 = const()[name = tensor("op_29754_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_29754_end_0 = const()[name = tensor("op_29754_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_29754_end_mask_0 = const()[name = tensor("op_29754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29754_cast_fp16 = slice_by_index(begin = var_29754_begin_0, end = var_29754_end_0, end_mask = var_29754_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29754_cast_fp16")]; + tensor var_29758_begin_0 = const()[name = tensor("op_29758_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_29758_end_0 = const()[name = tensor("op_29758_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_29758_end_mask_0 = const()[name = tensor("op_29758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29758_cast_fp16 = slice_by_index(begin = var_29758_begin_0, end = var_29758_end_0, end_mask = var_29758_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29758_cast_fp16")]; + tensor var_29762_begin_0 = const()[name = tensor("op_29762_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_29762_end_0 = const()[name = tensor("op_29762_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_29762_end_mask_0 = const()[name = tensor("op_29762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29762_cast_fp16 = slice_by_index(begin = var_29762_begin_0, end = var_29762_end_0, end_mask = var_29762_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29762_cast_fp16")]; + tensor var_29766_begin_0 = const()[name = tensor("op_29766_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_29766_end_0 = const()[name = tensor("op_29766_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_29766_end_mask_0 = const()[name = tensor("op_29766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29766_cast_fp16 = slice_by_index(begin = var_29766_begin_0, end = var_29766_end_0, end_mask = var_29766_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29766_cast_fp16")]; + tensor var_29770_begin_0 = const()[name = tensor("op_29770_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_29770_end_0 = const()[name = tensor("op_29770_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_29770_end_mask_0 = const()[name = tensor("op_29770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29770_cast_fp16 = slice_by_index(begin = var_29770_begin_0, end = var_29770_end_0, end_mask = var_29770_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29770_cast_fp16")]; + tensor var_29774_begin_0 = const()[name = tensor("op_29774_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_29774_end_0 = const()[name = tensor("op_29774_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_29774_end_mask_0 = const()[name = tensor("op_29774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29774_cast_fp16 = slice_by_index(begin = var_29774_begin_0, end = var_29774_end_0, end_mask = var_29774_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29774_cast_fp16")]; + tensor var_29778_begin_0 = const()[name = tensor("op_29778_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_29778_end_0 = const()[name = tensor("op_29778_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_29778_end_mask_0 = const()[name = tensor("op_29778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29778_cast_fp16 = slice_by_index(begin = var_29778_begin_0, end = var_29778_end_0, end_mask = var_29778_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29778_cast_fp16")]; + tensor var_29782_begin_0 = const()[name = tensor("op_29782_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_29782_end_0 = const()[name = tensor("op_29782_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_29782_end_mask_0 = const()[name = tensor("op_29782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29782_cast_fp16 = slice_by_index(begin = var_29782_begin_0, end = var_29782_end_0, end_mask = var_29782_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29782_cast_fp16")]; + tensor var_29786_begin_0 = const()[name = tensor("op_29786_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_29786_end_0 = const()[name = tensor("op_29786_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_29786_end_mask_0 = const()[name = tensor("op_29786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29786_cast_fp16 = slice_by_index(begin = var_29786_begin_0, end = var_29786_end_0, end_mask = var_29786_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29786_cast_fp16")]; + tensor var_29790_begin_0 = const()[name = tensor("op_29790_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_29790_end_0 = const()[name = tensor("op_29790_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_29790_end_mask_0 = const()[name = tensor("op_29790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29790_cast_fp16 = slice_by_index(begin = var_29790_begin_0, end = var_29790_end_0, end_mask = var_29790_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29790_cast_fp16")]; + tensor var_29794_begin_0 = const()[name = tensor("op_29794_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_29794_end_0 = const()[name = tensor("op_29794_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_29794_end_mask_0 = const()[name = tensor("op_29794_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29794_cast_fp16 = slice_by_index(begin = var_29794_begin_0, end = var_29794_end_0, end_mask = var_29794_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29794_cast_fp16")]; + tensor var_29798_begin_0 = const()[name = tensor("op_29798_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_29798_end_0 = const()[name = tensor("op_29798_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_29798_end_mask_0 = const()[name = tensor("op_29798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29798_cast_fp16 = slice_by_index(begin = var_29798_begin_0, end = var_29798_end_0, end_mask = var_29798_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29798_cast_fp16")]; + tensor var_29802_begin_0 = const()[name = tensor("op_29802_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_29802_end_0 = const()[name = tensor("op_29802_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_29802_end_mask_0 = const()[name = tensor("op_29802_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29802_cast_fp16 = slice_by_index(begin = var_29802_begin_0, end = var_29802_end_0, end_mask = var_29802_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29802_cast_fp16")]; + tensor var_29806_begin_0 = const()[name = tensor("op_29806_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_29806_end_0 = const()[name = tensor("op_29806_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_29806_end_mask_0 = const()[name = tensor("op_29806_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29806_cast_fp16 = slice_by_index(begin = var_29806_begin_0, end = var_29806_end_0, end_mask = var_29806_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29806_cast_fp16")]; + tensor var_29810_begin_0 = const()[name = tensor("op_29810_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_29810_end_0 = const()[name = tensor("op_29810_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_29810_end_mask_0 = const()[name = tensor("op_29810_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29810_cast_fp16 = slice_by_index(begin = var_29810_begin_0, end = var_29810_end_0, end_mask = var_29810_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29810_cast_fp16")]; + tensor var_29814_begin_0 = const()[name = tensor("op_29814_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_29814_end_0 = const()[name = tensor("op_29814_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_29814_end_mask_0 = const()[name = tensor("op_29814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_29814_cast_fp16 = slice_by_index(begin = var_29814_begin_0, end = var_29814_end_0, end_mask = var_29814_end_mask_0, x = query_37_cast_fp16)[name = tensor("op_29814_cast_fp16")]; + tensor var_29823_begin_0 = const()[name = tensor("op_29823_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29823_end_0 = const()[name = tensor("op_29823_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29823_end_mask_0 = const()[name = tensor("op_29823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29823_cast_fp16 = slice_by_index(begin = var_29823_begin_0, end = var_29823_end_0, end_mask = var_29823_end_mask_0, x = var_29738_cast_fp16)[name = tensor("op_29823_cast_fp16")]; + tensor var_29830_begin_0 = const()[name = tensor("op_29830_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29830_end_0 = const()[name = tensor("op_29830_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29830_end_mask_0 = const()[name = tensor("op_29830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29830_cast_fp16 = slice_by_index(begin = var_29830_begin_0, end = var_29830_end_0, end_mask = var_29830_end_mask_0, x = var_29738_cast_fp16)[name = tensor("op_29830_cast_fp16")]; + tensor var_29837_begin_0 = const()[name = tensor("op_29837_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29837_end_0 = const()[name = tensor("op_29837_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29837_end_mask_0 = const()[name = tensor("op_29837_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29837_cast_fp16 = slice_by_index(begin = var_29837_begin_0, end = var_29837_end_0, end_mask = var_29837_end_mask_0, x = var_29738_cast_fp16)[name = tensor("op_29837_cast_fp16")]; + tensor var_29844_begin_0 = const()[name = tensor("op_29844_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29844_end_0 = const()[name = tensor("op_29844_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29844_end_mask_0 = const()[name = tensor("op_29844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29844_cast_fp16 = slice_by_index(begin = var_29844_begin_0, end = var_29844_end_0, end_mask = var_29844_end_mask_0, x = var_29738_cast_fp16)[name = tensor("op_29844_cast_fp16")]; + tensor var_29851_begin_0 = const()[name = tensor("op_29851_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29851_end_0 = const()[name = tensor("op_29851_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29851_end_mask_0 = const()[name = tensor("op_29851_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29851_cast_fp16 = slice_by_index(begin = var_29851_begin_0, end = var_29851_end_0, end_mask = var_29851_end_mask_0, x = var_29742_cast_fp16)[name = tensor("op_29851_cast_fp16")]; + tensor var_29858_begin_0 = const()[name = tensor("op_29858_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29858_end_0 = const()[name = tensor("op_29858_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29858_end_mask_0 = const()[name = tensor("op_29858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29858_cast_fp16 = slice_by_index(begin = var_29858_begin_0, end = var_29858_end_0, end_mask = var_29858_end_mask_0, x = var_29742_cast_fp16)[name = tensor("op_29858_cast_fp16")]; + tensor var_29865_begin_0 = const()[name = tensor("op_29865_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29865_end_0 = const()[name = tensor("op_29865_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29865_end_mask_0 = const()[name = tensor("op_29865_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29865_cast_fp16 = slice_by_index(begin = var_29865_begin_0, end = var_29865_end_0, end_mask = var_29865_end_mask_0, x = var_29742_cast_fp16)[name = tensor("op_29865_cast_fp16")]; + tensor var_29872_begin_0 = const()[name = tensor("op_29872_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29872_end_0 = const()[name = tensor("op_29872_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29872_end_mask_0 = const()[name = tensor("op_29872_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29872_cast_fp16 = slice_by_index(begin = var_29872_begin_0, end = var_29872_end_0, end_mask = var_29872_end_mask_0, x = var_29742_cast_fp16)[name = tensor("op_29872_cast_fp16")]; + tensor var_29879_begin_0 = const()[name = tensor("op_29879_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29879_end_0 = const()[name = tensor("op_29879_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29879_end_mask_0 = const()[name = tensor("op_29879_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29879_cast_fp16 = slice_by_index(begin = var_29879_begin_0, end = var_29879_end_0, end_mask = var_29879_end_mask_0, x = var_29746_cast_fp16)[name = tensor("op_29879_cast_fp16")]; + tensor var_29886_begin_0 = const()[name = tensor("op_29886_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29886_end_0 = const()[name = tensor("op_29886_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29886_end_mask_0 = const()[name = tensor("op_29886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29886_cast_fp16 = slice_by_index(begin = var_29886_begin_0, end = var_29886_end_0, end_mask = var_29886_end_mask_0, x = var_29746_cast_fp16)[name = tensor("op_29886_cast_fp16")]; + tensor var_29893_begin_0 = const()[name = tensor("op_29893_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29893_end_0 = const()[name = tensor("op_29893_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29893_end_mask_0 = const()[name = tensor("op_29893_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29893_cast_fp16 = slice_by_index(begin = var_29893_begin_0, end = var_29893_end_0, end_mask = var_29893_end_mask_0, x = var_29746_cast_fp16)[name = tensor("op_29893_cast_fp16")]; + tensor var_29900_begin_0 = const()[name = tensor("op_29900_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29900_end_0 = const()[name = tensor("op_29900_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29900_end_mask_0 = const()[name = tensor("op_29900_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29900_cast_fp16 = slice_by_index(begin = var_29900_begin_0, end = var_29900_end_0, end_mask = var_29900_end_mask_0, x = var_29746_cast_fp16)[name = tensor("op_29900_cast_fp16")]; + tensor var_29907_begin_0 = const()[name = tensor("op_29907_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29907_end_0 = const()[name = tensor("op_29907_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29907_end_mask_0 = const()[name = tensor("op_29907_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29907_cast_fp16 = slice_by_index(begin = var_29907_begin_0, end = var_29907_end_0, end_mask = var_29907_end_mask_0, x = var_29750_cast_fp16)[name = tensor("op_29907_cast_fp16")]; + tensor var_29914_begin_0 = const()[name = tensor("op_29914_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29914_end_0 = const()[name = tensor("op_29914_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29914_end_mask_0 = const()[name = tensor("op_29914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29914_cast_fp16 = slice_by_index(begin = var_29914_begin_0, end = var_29914_end_0, end_mask = var_29914_end_mask_0, x = var_29750_cast_fp16)[name = tensor("op_29914_cast_fp16")]; + tensor var_29921_begin_0 = const()[name = tensor("op_29921_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29921_end_0 = const()[name = tensor("op_29921_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29921_end_mask_0 = const()[name = tensor("op_29921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29921_cast_fp16 = slice_by_index(begin = var_29921_begin_0, end = var_29921_end_0, end_mask = var_29921_end_mask_0, x = var_29750_cast_fp16)[name = tensor("op_29921_cast_fp16")]; + tensor var_29928_begin_0 = const()[name = tensor("op_29928_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29928_end_0 = const()[name = tensor("op_29928_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29928_end_mask_0 = const()[name = tensor("op_29928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29928_cast_fp16 = slice_by_index(begin = var_29928_begin_0, end = var_29928_end_0, end_mask = var_29928_end_mask_0, x = var_29750_cast_fp16)[name = tensor("op_29928_cast_fp16")]; + tensor var_29935_begin_0 = const()[name = tensor("op_29935_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29935_end_0 = const()[name = tensor("op_29935_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29935_end_mask_0 = const()[name = tensor("op_29935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29935_cast_fp16 = slice_by_index(begin = var_29935_begin_0, end = var_29935_end_0, end_mask = var_29935_end_mask_0, x = var_29754_cast_fp16)[name = tensor("op_29935_cast_fp16")]; + tensor var_29942_begin_0 = const()[name = tensor("op_29942_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29942_end_0 = const()[name = tensor("op_29942_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29942_end_mask_0 = const()[name = tensor("op_29942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29942_cast_fp16 = slice_by_index(begin = var_29942_begin_0, end = var_29942_end_0, end_mask = var_29942_end_mask_0, x = var_29754_cast_fp16)[name = tensor("op_29942_cast_fp16")]; + tensor var_29949_begin_0 = const()[name = tensor("op_29949_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29949_end_0 = const()[name = tensor("op_29949_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29949_end_mask_0 = const()[name = tensor("op_29949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29949_cast_fp16 = slice_by_index(begin = var_29949_begin_0, end = var_29949_end_0, end_mask = var_29949_end_mask_0, x = var_29754_cast_fp16)[name = tensor("op_29949_cast_fp16")]; + tensor var_29956_begin_0 = const()[name = tensor("op_29956_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29956_end_0 = const()[name = tensor("op_29956_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29956_end_mask_0 = const()[name = tensor("op_29956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29956_cast_fp16 = slice_by_index(begin = var_29956_begin_0, end = var_29956_end_0, end_mask = var_29956_end_mask_0, x = var_29754_cast_fp16)[name = tensor("op_29956_cast_fp16")]; + tensor var_29963_begin_0 = const()[name = tensor("op_29963_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29963_end_0 = const()[name = tensor("op_29963_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29963_end_mask_0 = const()[name = tensor("op_29963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29963_cast_fp16 = slice_by_index(begin = var_29963_begin_0, end = var_29963_end_0, end_mask = var_29963_end_mask_0, x = var_29758_cast_fp16)[name = tensor("op_29963_cast_fp16")]; + tensor var_29970_begin_0 = const()[name = tensor("op_29970_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29970_end_0 = const()[name = tensor("op_29970_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29970_end_mask_0 = const()[name = tensor("op_29970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29970_cast_fp16 = slice_by_index(begin = var_29970_begin_0, end = var_29970_end_0, end_mask = var_29970_end_mask_0, x = var_29758_cast_fp16)[name = tensor("op_29970_cast_fp16")]; + tensor var_29977_begin_0 = const()[name = tensor("op_29977_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_29977_end_0 = const()[name = tensor("op_29977_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_29977_end_mask_0 = const()[name = tensor("op_29977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29977_cast_fp16 = slice_by_index(begin = var_29977_begin_0, end = var_29977_end_0, end_mask = var_29977_end_mask_0, x = var_29758_cast_fp16)[name = tensor("op_29977_cast_fp16")]; + tensor var_29984_begin_0 = const()[name = tensor("op_29984_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_29984_end_0 = const()[name = tensor("op_29984_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_29984_end_mask_0 = const()[name = tensor("op_29984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29984_cast_fp16 = slice_by_index(begin = var_29984_begin_0, end = var_29984_end_0, end_mask = var_29984_end_mask_0, x = var_29758_cast_fp16)[name = tensor("op_29984_cast_fp16")]; + tensor var_29991_begin_0 = const()[name = tensor("op_29991_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_29991_end_0 = const()[name = tensor("op_29991_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_29991_end_mask_0 = const()[name = tensor("op_29991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29991_cast_fp16 = slice_by_index(begin = var_29991_begin_0, end = var_29991_end_0, end_mask = var_29991_end_mask_0, x = var_29762_cast_fp16)[name = tensor("op_29991_cast_fp16")]; + tensor var_29998_begin_0 = const()[name = tensor("op_29998_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_29998_end_0 = const()[name = tensor("op_29998_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_29998_end_mask_0 = const()[name = tensor("op_29998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_29998_cast_fp16 = slice_by_index(begin = var_29998_begin_0, end = var_29998_end_0, end_mask = var_29998_end_mask_0, x = var_29762_cast_fp16)[name = tensor("op_29998_cast_fp16")]; + tensor var_30005_begin_0 = const()[name = tensor("op_30005_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30005_end_0 = const()[name = tensor("op_30005_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30005_end_mask_0 = const()[name = tensor("op_30005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30005_cast_fp16 = slice_by_index(begin = var_30005_begin_0, end = var_30005_end_0, end_mask = var_30005_end_mask_0, x = var_29762_cast_fp16)[name = tensor("op_30005_cast_fp16")]; + tensor var_30012_begin_0 = const()[name = tensor("op_30012_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30012_end_0 = const()[name = tensor("op_30012_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30012_end_mask_0 = const()[name = tensor("op_30012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30012_cast_fp16 = slice_by_index(begin = var_30012_begin_0, end = var_30012_end_0, end_mask = var_30012_end_mask_0, x = var_29762_cast_fp16)[name = tensor("op_30012_cast_fp16")]; + tensor var_30019_begin_0 = const()[name = tensor("op_30019_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30019_end_0 = const()[name = tensor("op_30019_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30019_end_mask_0 = const()[name = tensor("op_30019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30019_cast_fp16 = slice_by_index(begin = var_30019_begin_0, end = var_30019_end_0, end_mask = var_30019_end_mask_0, x = var_29766_cast_fp16)[name = tensor("op_30019_cast_fp16")]; + tensor var_30026_begin_0 = const()[name = tensor("op_30026_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30026_end_0 = const()[name = tensor("op_30026_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30026_end_mask_0 = const()[name = tensor("op_30026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30026_cast_fp16 = slice_by_index(begin = var_30026_begin_0, end = var_30026_end_0, end_mask = var_30026_end_mask_0, x = var_29766_cast_fp16)[name = tensor("op_30026_cast_fp16")]; + tensor var_30033_begin_0 = const()[name = tensor("op_30033_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30033_end_0 = const()[name = tensor("op_30033_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30033_end_mask_0 = const()[name = tensor("op_30033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30033_cast_fp16 = slice_by_index(begin = var_30033_begin_0, end = var_30033_end_0, end_mask = var_30033_end_mask_0, x = var_29766_cast_fp16)[name = tensor("op_30033_cast_fp16")]; + tensor var_30040_begin_0 = const()[name = tensor("op_30040_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30040_end_0 = const()[name = tensor("op_30040_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30040_end_mask_0 = const()[name = tensor("op_30040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30040_cast_fp16 = slice_by_index(begin = var_30040_begin_0, end = var_30040_end_0, end_mask = var_30040_end_mask_0, x = var_29766_cast_fp16)[name = tensor("op_30040_cast_fp16")]; + tensor var_30047_begin_0 = const()[name = tensor("op_30047_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30047_end_0 = const()[name = tensor("op_30047_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30047_end_mask_0 = const()[name = tensor("op_30047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30047_cast_fp16 = slice_by_index(begin = var_30047_begin_0, end = var_30047_end_0, end_mask = var_30047_end_mask_0, x = var_29770_cast_fp16)[name = tensor("op_30047_cast_fp16")]; + tensor var_30054_begin_0 = const()[name = tensor("op_30054_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30054_end_0 = const()[name = tensor("op_30054_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30054_end_mask_0 = const()[name = tensor("op_30054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30054_cast_fp16 = slice_by_index(begin = var_30054_begin_0, end = var_30054_end_0, end_mask = var_30054_end_mask_0, x = var_29770_cast_fp16)[name = tensor("op_30054_cast_fp16")]; + tensor var_30061_begin_0 = const()[name = tensor("op_30061_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30061_end_0 = const()[name = tensor("op_30061_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30061_end_mask_0 = const()[name = tensor("op_30061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30061_cast_fp16 = slice_by_index(begin = var_30061_begin_0, end = var_30061_end_0, end_mask = var_30061_end_mask_0, x = var_29770_cast_fp16)[name = tensor("op_30061_cast_fp16")]; + tensor var_30068_begin_0 = const()[name = tensor("op_30068_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30068_end_0 = const()[name = tensor("op_30068_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30068_end_mask_0 = const()[name = tensor("op_30068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30068_cast_fp16 = slice_by_index(begin = var_30068_begin_0, end = var_30068_end_0, end_mask = var_30068_end_mask_0, x = var_29770_cast_fp16)[name = tensor("op_30068_cast_fp16")]; + tensor var_30075_begin_0 = const()[name = tensor("op_30075_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30075_end_0 = const()[name = tensor("op_30075_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30075_end_mask_0 = const()[name = tensor("op_30075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30075_cast_fp16 = slice_by_index(begin = var_30075_begin_0, end = var_30075_end_0, end_mask = var_30075_end_mask_0, x = var_29774_cast_fp16)[name = tensor("op_30075_cast_fp16")]; + tensor var_30082_begin_0 = const()[name = tensor("op_30082_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30082_end_0 = const()[name = tensor("op_30082_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30082_end_mask_0 = const()[name = tensor("op_30082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30082_cast_fp16 = slice_by_index(begin = var_30082_begin_0, end = var_30082_end_0, end_mask = var_30082_end_mask_0, x = var_29774_cast_fp16)[name = tensor("op_30082_cast_fp16")]; + tensor var_30089_begin_0 = const()[name = tensor("op_30089_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30089_end_0 = const()[name = tensor("op_30089_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30089_end_mask_0 = const()[name = tensor("op_30089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30089_cast_fp16 = slice_by_index(begin = var_30089_begin_0, end = var_30089_end_0, end_mask = var_30089_end_mask_0, x = var_29774_cast_fp16)[name = tensor("op_30089_cast_fp16")]; + tensor var_30096_begin_0 = const()[name = tensor("op_30096_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30096_end_0 = const()[name = tensor("op_30096_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30096_end_mask_0 = const()[name = tensor("op_30096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30096_cast_fp16 = slice_by_index(begin = var_30096_begin_0, end = var_30096_end_0, end_mask = var_30096_end_mask_0, x = var_29774_cast_fp16)[name = tensor("op_30096_cast_fp16")]; + tensor var_30103_begin_0 = const()[name = tensor("op_30103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30103_end_0 = const()[name = tensor("op_30103_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30103_end_mask_0 = const()[name = tensor("op_30103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30103_cast_fp16 = slice_by_index(begin = var_30103_begin_0, end = var_30103_end_0, end_mask = var_30103_end_mask_0, x = var_29778_cast_fp16)[name = tensor("op_30103_cast_fp16")]; + tensor var_30110_begin_0 = const()[name = tensor("op_30110_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30110_end_0 = const()[name = tensor("op_30110_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30110_end_mask_0 = const()[name = tensor("op_30110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30110_cast_fp16 = slice_by_index(begin = var_30110_begin_0, end = var_30110_end_0, end_mask = var_30110_end_mask_0, x = var_29778_cast_fp16)[name = tensor("op_30110_cast_fp16")]; + tensor var_30117_begin_0 = const()[name = tensor("op_30117_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30117_end_0 = const()[name = tensor("op_30117_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30117_end_mask_0 = const()[name = tensor("op_30117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30117_cast_fp16 = slice_by_index(begin = var_30117_begin_0, end = var_30117_end_0, end_mask = var_30117_end_mask_0, x = var_29778_cast_fp16)[name = tensor("op_30117_cast_fp16")]; + tensor var_30124_begin_0 = const()[name = tensor("op_30124_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30124_end_0 = const()[name = tensor("op_30124_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30124_end_mask_0 = const()[name = tensor("op_30124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30124_cast_fp16 = slice_by_index(begin = var_30124_begin_0, end = var_30124_end_0, end_mask = var_30124_end_mask_0, x = var_29778_cast_fp16)[name = tensor("op_30124_cast_fp16")]; + tensor var_30131_begin_0 = const()[name = tensor("op_30131_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30131_end_0 = const()[name = tensor("op_30131_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30131_end_mask_0 = const()[name = tensor("op_30131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30131_cast_fp16 = slice_by_index(begin = var_30131_begin_0, end = var_30131_end_0, end_mask = var_30131_end_mask_0, x = var_29782_cast_fp16)[name = tensor("op_30131_cast_fp16")]; + tensor var_30138_begin_0 = const()[name = tensor("op_30138_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30138_end_0 = const()[name = tensor("op_30138_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30138_end_mask_0 = const()[name = tensor("op_30138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30138_cast_fp16 = slice_by_index(begin = var_30138_begin_0, end = var_30138_end_0, end_mask = var_30138_end_mask_0, x = var_29782_cast_fp16)[name = tensor("op_30138_cast_fp16")]; + tensor var_30145_begin_0 = const()[name = tensor("op_30145_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30145_end_0 = const()[name = tensor("op_30145_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30145_end_mask_0 = const()[name = tensor("op_30145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30145_cast_fp16 = slice_by_index(begin = var_30145_begin_0, end = var_30145_end_0, end_mask = var_30145_end_mask_0, x = var_29782_cast_fp16)[name = tensor("op_30145_cast_fp16")]; + tensor var_30152_begin_0 = const()[name = tensor("op_30152_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30152_end_0 = const()[name = tensor("op_30152_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30152_end_mask_0 = const()[name = tensor("op_30152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30152_cast_fp16 = slice_by_index(begin = var_30152_begin_0, end = var_30152_end_0, end_mask = var_30152_end_mask_0, x = var_29782_cast_fp16)[name = tensor("op_30152_cast_fp16")]; + tensor var_30159_begin_0 = const()[name = tensor("op_30159_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30159_end_0 = const()[name = tensor("op_30159_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30159_end_mask_0 = const()[name = tensor("op_30159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30159_cast_fp16 = slice_by_index(begin = var_30159_begin_0, end = var_30159_end_0, end_mask = var_30159_end_mask_0, x = var_29786_cast_fp16)[name = tensor("op_30159_cast_fp16")]; + tensor var_30166_begin_0 = const()[name = tensor("op_30166_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30166_end_0 = const()[name = tensor("op_30166_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30166_end_mask_0 = const()[name = tensor("op_30166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30166_cast_fp16 = slice_by_index(begin = var_30166_begin_0, end = var_30166_end_0, end_mask = var_30166_end_mask_0, x = var_29786_cast_fp16)[name = tensor("op_30166_cast_fp16")]; + tensor var_30173_begin_0 = const()[name = tensor("op_30173_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30173_end_0 = const()[name = tensor("op_30173_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30173_end_mask_0 = const()[name = tensor("op_30173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30173_cast_fp16 = slice_by_index(begin = var_30173_begin_0, end = var_30173_end_0, end_mask = var_30173_end_mask_0, x = var_29786_cast_fp16)[name = tensor("op_30173_cast_fp16")]; + tensor var_30180_begin_0 = const()[name = tensor("op_30180_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30180_end_0 = const()[name = tensor("op_30180_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30180_end_mask_0 = const()[name = tensor("op_30180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30180_cast_fp16 = slice_by_index(begin = var_30180_begin_0, end = var_30180_end_0, end_mask = var_30180_end_mask_0, x = var_29786_cast_fp16)[name = tensor("op_30180_cast_fp16")]; + tensor var_30187_begin_0 = const()[name = tensor("op_30187_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30187_end_0 = const()[name = tensor("op_30187_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30187_end_mask_0 = const()[name = tensor("op_30187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30187_cast_fp16 = slice_by_index(begin = var_30187_begin_0, end = var_30187_end_0, end_mask = var_30187_end_mask_0, x = var_29790_cast_fp16)[name = tensor("op_30187_cast_fp16")]; + tensor var_30194_begin_0 = const()[name = tensor("op_30194_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30194_end_0 = const()[name = tensor("op_30194_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30194_end_mask_0 = const()[name = tensor("op_30194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30194_cast_fp16 = slice_by_index(begin = var_30194_begin_0, end = var_30194_end_0, end_mask = var_30194_end_mask_0, x = var_29790_cast_fp16)[name = tensor("op_30194_cast_fp16")]; + tensor var_30201_begin_0 = const()[name = tensor("op_30201_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30201_end_0 = const()[name = tensor("op_30201_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30201_end_mask_0 = const()[name = tensor("op_30201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30201_cast_fp16 = slice_by_index(begin = var_30201_begin_0, end = var_30201_end_0, end_mask = var_30201_end_mask_0, x = var_29790_cast_fp16)[name = tensor("op_30201_cast_fp16")]; + tensor var_30208_begin_0 = const()[name = tensor("op_30208_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30208_end_0 = const()[name = tensor("op_30208_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30208_end_mask_0 = const()[name = tensor("op_30208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30208_cast_fp16 = slice_by_index(begin = var_30208_begin_0, end = var_30208_end_0, end_mask = var_30208_end_mask_0, x = var_29790_cast_fp16)[name = tensor("op_30208_cast_fp16")]; + tensor var_30215_begin_0 = const()[name = tensor("op_30215_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30215_end_0 = const()[name = tensor("op_30215_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30215_end_mask_0 = const()[name = tensor("op_30215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30215_cast_fp16 = slice_by_index(begin = var_30215_begin_0, end = var_30215_end_0, end_mask = var_30215_end_mask_0, x = var_29794_cast_fp16)[name = tensor("op_30215_cast_fp16")]; + tensor var_30222_begin_0 = const()[name = tensor("op_30222_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30222_end_0 = const()[name = tensor("op_30222_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30222_end_mask_0 = const()[name = tensor("op_30222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30222_cast_fp16 = slice_by_index(begin = var_30222_begin_0, end = var_30222_end_0, end_mask = var_30222_end_mask_0, x = var_29794_cast_fp16)[name = tensor("op_30222_cast_fp16")]; + tensor var_30229_begin_0 = const()[name = tensor("op_30229_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30229_end_0 = const()[name = tensor("op_30229_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30229_end_mask_0 = const()[name = tensor("op_30229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30229_cast_fp16 = slice_by_index(begin = var_30229_begin_0, end = var_30229_end_0, end_mask = var_30229_end_mask_0, x = var_29794_cast_fp16)[name = tensor("op_30229_cast_fp16")]; + tensor var_30236_begin_0 = const()[name = tensor("op_30236_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30236_end_0 = const()[name = tensor("op_30236_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30236_end_mask_0 = const()[name = tensor("op_30236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30236_cast_fp16 = slice_by_index(begin = var_30236_begin_0, end = var_30236_end_0, end_mask = var_30236_end_mask_0, x = var_29794_cast_fp16)[name = tensor("op_30236_cast_fp16")]; + tensor var_30243_begin_0 = const()[name = tensor("op_30243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30243_end_0 = const()[name = tensor("op_30243_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30243_end_mask_0 = const()[name = tensor("op_30243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30243_cast_fp16 = slice_by_index(begin = var_30243_begin_0, end = var_30243_end_0, end_mask = var_30243_end_mask_0, x = var_29798_cast_fp16)[name = tensor("op_30243_cast_fp16")]; + tensor var_30250_begin_0 = const()[name = tensor("op_30250_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30250_end_0 = const()[name = tensor("op_30250_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30250_end_mask_0 = const()[name = tensor("op_30250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30250_cast_fp16 = slice_by_index(begin = var_30250_begin_0, end = var_30250_end_0, end_mask = var_30250_end_mask_0, x = var_29798_cast_fp16)[name = tensor("op_30250_cast_fp16")]; + tensor var_30257_begin_0 = const()[name = tensor("op_30257_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30257_end_0 = const()[name = tensor("op_30257_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30257_end_mask_0 = const()[name = tensor("op_30257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30257_cast_fp16 = slice_by_index(begin = var_30257_begin_0, end = var_30257_end_0, end_mask = var_30257_end_mask_0, x = var_29798_cast_fp16)[name = tensor("op_30257_cast_fp16")]; + tensor var_30264_begin_0 = const()[name = tensor("op_30264_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30264_end_0 = const()[name = tensor("op_30264_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30264_end_mask_0 = const()[name = tensor("op_30264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30264_cast_fp16 = slice_by_index(begin = var_30264_begin_0, end = var_30264_end_0, end_mask = var_30264_end_mask_0, x = var_29798_cast_fp16)[name = tensor("op_30264_cast_fp16")]; + tensor var_30271_begin_0 = const()[name = tensor("op_30271_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30271_end_0 = const()[name = tensor("op_30271_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30271_end_mask_0 = const()[name = tensor("op_30271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30271_cast_fp16 = slice_by_index(begin = var_30271_begin_0, end = var_30271_end_0, end_mask = var_30271_end_mask_0, x = var_29802_cast_fp16)[name = tensor("op_30271_cast_fp16")]; + tensor var_30278_begin_0 = const()[name = tensor("op_30278_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30278_end_0 = const()[name = tensor("op_30278_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30278_end_mask_0 = const()[name = tensor("op_30278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30278_cast_fp16 = slice_by_index(begin = var_30278_begin_0, end = var_30278_end_0, end_mask = var_30278_end_mask_0, x = var_29802_cast_fp16)[name = tensor("op_30278_cast_fp16")]; + tensor var_30285_begin_0 = const()[name = tensor("op_30285_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30285_end_0 = const()[name = tensor("op_30285_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30285_end_mask_0 = const()[name = tensor("op_30285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30285_cast_fp16 = slice_by_index(begin = var_30285_begin_0, end = var_30285_end_0, end_mask = var_30285_end_mask_0, x = var_29802_cast_fp16)[name = tensor("op_30285_cast_fp16")]; + tensor var_30292_begin_0 = const()[name = tensor("op_30292_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30292_end_0 = const()[name = tensor("op_30292_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30292_end_mask_0 = const()[name = tensor("op_30292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30292_cast_fp16 = slice_by_index(begin = var_30292_begin_0, end = var_30292_end_0, end_mask = var_30292_end_mask_0, x = var_29802_cast_fp16)[name = tensor("op_30292_cast_fp16")]; + tensor var_30299_begin_0 = const()[name = tensor("op_30299_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30299_end_0 = const()[name = tensor("op_30299_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30299_end_mask_0 = const()[name = tensor("op_30299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30299_cast_fp16 = slice_by_index(begin = var_30299_begin_0, end = var_30299_end_0, end_mask = var_30299_end_mask_0, x = var_29806_cast_fp16)[name = tensor("op_30299_cast_fp16")]; + tensor var_30306_begin_0 = const()[name = tensor("op_30306_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30306_end_0 = const()[name = tensor("op_30306_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30306_end_mask_0 = const()[name = tensor("op_30306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30306_cast_fp16 = slice_by_index(begin = var_30306_begin_0, end = var_30306_end_0, end_mask = var_30306_end_mask_0, x = var_29806_cast_fp16)[name = tensor("op_30306_cast_fp16")]; + tensor var_30313_begin_0 = const()[name = tensor("op_30313_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30313_end_0 = const()[name = tensor("op_30313_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30313_end_mask_0 = const()[name = tensor("op_30313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30313_cast_fp16 = slice_by_index(begin = var_30313_begin_0, end = var_30313_end_0, end_mask = var_30313_end_mask_0, x = var_29806_cast_fp16)[name = tensor("op_30313_cast_fp16")]; + tensor var_30320_begin_0 = const()[name = tensor("op_30320_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30320_end_0 = const()[name = tensor("op_30320_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30320_end_mask_0 = const()[name = tensor("op_30320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30320_cast_fp16 = slice_by_index(begin = var_30320_begin_0, end = var_30320_end_0, end_mask = var_30320_end_mask_0, x = var_29806_cast_fp16)[name = tensor("op_30320_cast_fp16")]; + tensor var_30327_begin_0 = const()[name = tensor("op_30327_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30327_end_0 = const()[name = tensor("op_30327_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30327_end_mask_0 = const()[name = tensor("op_30327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30327_cast_fp16 = slice_by_index(begin = var_30327_begin_0, end = var_30327_end_0, end_mask = var_30327_end_mask_0, x = var_29810_cast_fp16)[name = tensor("op_30327_cast_fp16")]; + tensor var_30334_begin_0 = const()[name = tensor("op_30334_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30334_end_0 = const()[name = tensor("op_30334_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30334_end_mask_0 = const()[name = tensor("op_30334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30334_cast_fp16 = slice_by_index(begin = var_30334_begin_0, end = var_30334_end_0, end_mask = var_30334_end_mask_0, x = var_29810_cast_fp16)[name = tensor("op_30334_cast_fp16")]; + tensor var_30341_begin_0 = const()[name = tensor("op_30341_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30341_end_0 = const()[name = tensor("op_30341_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30341_end_mask_0 = const()[name = tensor("op_30341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30341_cast_fp16 = slice_by_index(begin = var_30341_begin_0, end = var_30341_end_0, end_mask = var_30341_end_mask_0, x = var_29810_cast_fp16)[name = tensor("op_30341_cast_fp16")]; + tensor var_30348_begin_0 = const()[name = tensor("op_30348_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30348_end_0 = const()[name = tensor("op_30348_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30348_end_mask_0 = const()[name = tensor("op_30348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30348_cast_fp16 = slice_by_index(begin = var_30348_begin_0, end = var_30348_end_0, end_mask = var_30348_end_mask_0, x = var_29810_cast_fp16)[name = tensor("op_30348_cast_fp16")]; + tensor var_30355_begin_0 = const()[name = tensor("op_30355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30355_end_0 = const()[name = tensor("op_30355_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_30355_end_mask_0 = const()[name = tensor("op_30355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30355_cast_fp16 = slice_by_index(begin = var_30355_begin_0, end = var_30355_end_0, end_mask = var_30355_end_mask_0, x = var_29814_cast_fp16)[name = tensor("op_30355_cast_fp16")]; + tensor var_30362_begin_0 = const()[name = tensor("op_30362_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_30362_end_0 = const()[name = tensor("op_30362_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_30362_end_mask_0 = const()[name = tensor("op_30362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30362_cast_fp16 = slice_by_index(begin = var_30362_begin_0, end = var_30362_end_0, end_mask = var_30362_end_mask_0, x = var_29814_cast_fp16)[name = tensor("op_30362_cast_fp16")]; + tensor var_30369_begin_0 = const()[name = tensor("op_30369_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_30369_end_0 = const()[name = tensor("op_30369_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_30369_end_mask_0 = const()[name = tensor("op_30369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30369_cast_fp16 = slice_by_index(begin = var_30369_begin_0, end = var_30369_end_0, end_mask = var_30369_end_mask_0, x = var_29814_cast_fp16)[name = tensor("op_30369_cast_fp16")]; + tensor var_30376_begin_0 = const()[name = tensor("op_30376_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_30376_end_0 = const()[name = tensor("op_30376_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30376_end_mask_0 = const()[name = tensor("op_30376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30376_cast_fp16 = slice_by_index(begin = var_30376_begin_0, end = var_30376_end_0, end_mask = var_30376_end_mask_0, x = var_29814_cast_fp16)[name = tensor("op_30376_cast_fp16")]; + tensor k_37_perm_0 = const()[name = tensor("k_37_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_30381_begin_0 = const()[name = tensor("op_30381_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30381_end_0 = const()[name = tensor("op_30381_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_30381_end_mask_0 = const()[name = tensor("op_30381_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_13 = transpose(perm = k_37_perm_0, x = key_37_cast_fp16)[name = tensor("transpose_13")]; + tensor var_30381_cast_fp16 = slice_by_index(begin = var_30381_begin_0, end = var_30381_end_0, end_mask = var_30381_end_mask_0, x = transpose_13)[name = tensor("op_30381_cast_fp16")]; + tensor var_30385_begin_0 = const()[name = tensor("op_30385_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_30385_end_0 = const()[name = tensor("op_30385_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_30385_end_mask_0 = const()[name = tensor("op_30385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30385_cast_fp16 = slice_by_index(begin = var_30385_begin_0, end = var_30385_end_0, end_mask = var_30385_end_mask_0, x = transpose_13)[name = tensor("op_30385_cast_fp16")]; + tensor var_30389_begin_0 = const()[name = tensor("op_30389_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_30389_end_0 = const()[name = tensor("op_30389_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_30389_end_mask_0 = const()[name = tensor("op_30389_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30389_cast_fp16 = slice_by_index(begin = var_30389_begin_0, end = var_30389_end_0, end_mask = var_30389_end_mask_0, x = transpose_13)[name = tensor("op_30389_cast_fp16")]; + tensor var_30393_begin_0 = const()[name = tensor("op_30393_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_30393_end_0 = const()[name = tensor("op_30393_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_30393_end_mask_0 = const()[name = tensor("op_30393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30393_cast_fp16 = slice_by_index(begin = var_30393_begin_0, end = var_30393_end_0, end_mask = var_30393_end_mask_0, x = transpose_13)[name = tensor("op_30393_cast_fp16")]; + tensor var_30397_begin_0 = const()[name = tensor("op_30397_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_30397_end_0 = const()[name = tensor("op_30397_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_30397_end_mask_0 = const()[name = tensor("op_30397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30397_cast_fp16 = slice_by_index(begin = var_30397_begin_0, end = var_30397_end_0, end_mask = var_30397_end_mask_0, x = transpose_13)[name = tensor("op_30397_cast_fp16")]; + tensor var_30401_begin_0 = const()[name = tensor("op_30401_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_30401_end_0 = const()[name = tensor("op_30401_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_30401_end_mask_0 = const()[name = tensor("op_30401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30401_cast_fp16 = slice_by_index(begin = var_30401_begin_0, end = var_30401_end_0, end_mask = var_30401_end_mask_0, x = transpose_13)[name = tensor("op_30401_cast_fp16")]; + tensor var_30405_begin_0 = const()[name = tensor("op_30405_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_30405_end_0 = const()[name = tensor("op_30405_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_30405_end_mask_0 = const()[name = tensor("op_30405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30405_cast_fp16 = slice_by_index(begin = var_30405_begin_0, end = var_30405_end_0, end_mask = var_30405_end_mask_0, x = transpose_13)[name = tensor("op_30405_cast_fp16")]; + tensor var_30409_begin_0 = const()[name = tensor("op_30409_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_30409_end_0 = const()[name = tensor("op_30409_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_30409_end_mask_0 = const()[name = tensor("op_30409_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30409_cast_fp16 = slice_by_index(begin = var_30409_begin_0, end = var_30409_end_0, end_mask = var_30409_end_mask_0, x = transpose_13)[name = tensor("op_30409_cast_fp16")]; + tensor var_30413_begin_0 = const()[name = tensor("op_30413_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_30413_end_0 = const()[name = tensor("op_30413_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_30413_end_mask_0 = const()[name = tensor("op_30413_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30413_cast_fp16 = slice_by_index(begin = var_30413_begin_0, end = var_30413_end_0, end_mask = var_30413_end_mask_0, x = transpose_13)[name = tensor("op_30413_cast_fp16")]; + tensor var_30417_begin_0 = const()[name = tensor("op_30417_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_30417_end_0 = const()[name = tensor("op_30417_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_30417_end_mask_0 = const()[name = tensor("op_30417_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30417_cast_fp16 = slice_by_index(begin = var_30417_begin_0, end = var_30417_end_0, end_mask = var_30417_end_mask_0, x = transpose_13)[name = tensor("op_30417_cast_fp16")]; + tensor var_30421_begin_0 = const()[name = tensor("op_30421_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_30421_end_0 = const()[name = tensor("op_30421_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_30421_end_mask_0 = const()[name = tensor("op_30421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30421_cast_fp16 = slice_by_index(begin = var_30421_begin_0, end = var_30421_end_0, end_mask = var_30421_end_mask_0, x = transpose_13)[name = tensor("op_30421_cast_fp16")]; + tensor var_30425_begin_0 = const()[name = tensor("op_30425_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_30425_end_0 = const()[name = tensor("op_30425_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_30425_end_mask_0 = const()[name = tensor("op_30425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30425_cast_fp16 = slice_by_index(begin = var_30425_begin_0, end = var_30425_end_0, end_mask = var_30425_end_mask_0, x = transpose_13)[name = tensor("op_30425_cast_fp16")]; + tensor var_30429_begin_0 = const()[name = tensor("op_30429_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_30429_end_0 = const()[name = tensor("op_30429_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_30429_end_mask_0 = const()[name = tensor("op_30429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30429_cast_fp16 = slice_by_index(begin = var_30429_begin_0, end = var_30429_end_0, end_mask = var_30429_end_mask_0, x = transpose_13)[name = tensor("op_30429_cast_fp16")]; + tensor var_30433_begin_0 = const()[name = tensor("op_30433_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_30433_end_0 = const()[name = tensor("op_30433_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_30433_end_mask_0 = const()[name = tensor("op_30433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30433_cast_fp16 = slice_by_index(begin = var_30433_begin_0, end = var_30433_end_0, end_mask = var_30433_end_mask_0, x = transpose_13)[name = tensor("op_30433_cast_fp16")]; + tensor var_30437_begin_0 = const()[name = tensor("op_30437_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_30437_end_0 = const()[name = tensor("op_30437_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_30437_end_mask_0 = const()[name = tensor("op_30437_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30437_cast_fp16 = slice_by_index(begin = var_30437_begin_0, end = var_30437_end_0, end_mask = var_30437_end_mask_0, x = transpose_13)[name = tensor("op_30437_cast_fp16")]; + tensor var_30441_begin_0 = const()[name = tensor("op_30441_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_30441_end_0 = const()[name = tensor("op_30441_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_30441_end_mask_0 = const()[name = tensor("op_30441_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30441_cast_fp16 = slice_by_index(begin = var_30441_begin_0, end = var_30441_end_0, end_mask = var_30441_end_mask_0, x = transpose_13)[name = tensor("op_30441_cast_fp16")]; + tensor var_30445_begin_0 = const()[name = tensor("op_30445_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_30445_end_0 = const()[name = tensor("op_30445_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_30445_end_mask_0 = const()[name = tensor("op_30445_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30445_cast_fp16 = slice_by_index(begin = var_30445_begin_0, end = var_30445_end_0, end_mask = var_30445_end_mask_0, x = transpose_13)[name = tensor("op_30445_cast_fp16")]; + tensor var_30449_begin_0 = const()[name = tensor("op_30449_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_30449_end_0 = const()[name = tensor("op_30449_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_30449_end_mask_0 = const()[name = tensor("op_30449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30449_cast_fp16 = slice_by_index(begin = var_30449_begin_0, end = var_30449_end_0, end_mask = var_30449_end_mask_0, x = transpose_13)[name = tensor("op_30449_cast_fp16")]; + tensor var_30453_begin_0 = const()[name = tensor("op_30453_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_30453_end_0 = const()[name = tensor("op_30453_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_30453_end_mask_0 = const()[name = tensor("op_30453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30453_cast_fp16 = slice_by_index(begin = var_30453_begin_0, end = var_30453_end_0, end_mask = var_30453_end_mask_0, x = transpose_13)[name = tensor("op_30453_cast_fp16")]; + tensor var_30457_begin_0 = const()[name = tensor("op_30457_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_30457_end_0 = const()[name = tensor("op_30457_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_30457_end_mask_0 = const()[name = tensor("op_30457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_30457_cast_fp16 = slice_by_index(begin = var_30457_begin_0, end = var_30457_end_0, end_mask = var_30457_end_mask_0, x = transpose_13)[name = tensor("op_30457_cast_fp16")]; + tensor var_30459_begin_0 = const()[name = tensor("op_30459_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_30459_end_0 = const()[name = tensor("op_30459_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_30459_end_mask_0 = const()[name = tensor("op_30459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30459_cast_fp16 = slice_by_index(begin = var_30459_begin_0, end = var_30459_end_0, end_mask = var_30459_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30459_cast_fp16")]; + tensor var_30463_begin_0 = const()[name = tensor("op_30463_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_30463_end_0 = const()[name = tensor("op_30463_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_30463_end_mask_0 = const()[name = tensor("op_30463_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30463_cast_fp16 = slice_by_index(begin = var_30463_begin_0, end = var_30463_end_0, end_mask = var_30463_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30463_cast_fp16")]; + tensor var_30467_begin_0 = const()[name = tensor("op_30467_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_30467_end_0 = const()[name = tensor("op_30467_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_30467_end_mask_0 = const()[name = tensor("op_30467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30467_cast_fp16 = slice_by_index(begin = var_30467_begin_0, end = var_30467_end_0, end_mask = var_30467_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30467_cast_fp16")]; + tensor var_30471_begin_0 = const()[name = tensor("op_30471_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_30471_end_0 = const()[name = tensor("op_30471_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_30471_end_mask_0 = const()[name = tensor("op_30471_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30471_cast_fp16 = slice_by_index(begin = var_30471_begin_0, end = var_30471_end_0, end_mask = var_30471_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30471_cast_fp16")]; + tensor var_30475_begin_0 = const()[name = tensor("op_30475_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_30475_end_0 = const()[name = tensor("op_30475_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_30475_end_mask_0 = const()[name = tensor("op_30475_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30475_cast_fp16 = slice_by_index(begin = var_30475_begin_0, end = var_30475_end_0, end_mask = var_30475_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30475_cast_fp16")]; + tensor var_30479_begin_0 = const()[name = tensor("op_30479_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_30479_end_0 = const()[name = tensor("op_30479_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_30479_end_mask_0 = const()[name = tensor("op_30479_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30479_cast_fp16 = slice_by_index(begin = var_30479_begin_0, end = var_30479_end_0, end_mask = var_30479_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30479_cast_fp16")]; + tensor var_30483_begin_0 = const()[name = tensor("op_30483_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_30483_end_0 = const()[name = tensor("op_30483_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_30483_end_mask_0 = const()[name = tensor("op_30483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30483_cast_fp16 = slice_by_index(begin = var_30483_begin_0, end = var_30483_end_0, end_mask = var_30483_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30483_cast_fp16")]; + tensor var_30487_begin_0 = const()[name = tensor("op_30487_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_30487_end_0 = const()[name = tensor("op_30487_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_30487_end_mask_0 = const()[name = tensor("op_30487_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30487_cast_fp16 = slice_by_index(begin = var_30487_begin_0, end = var_30487_end_0, end_mask = var_30487_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30487_cast_fp16")]; + tensor var_30491_begin_0 = const()[name = tensor("op_30491_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_30491_end_0 = const()[name = tensor("op_30491_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_30491_end_mask_0 = const()[name = tensor("op_30491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30491_cast_fp16 = slice_by_index(begin = var_30491_begin_0, end = var_30491_end_0, end_mask = var_30491_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30491_cast_fp16")]; + tensor var_30495_begin_0 = const()[name = tensor("op_30495_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_30495_end_0 = const()[name = tensor("op_30495_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_30495_end_mask_0 = const()[name = tensor("op_30495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30495_cast_fp16 = slice_by_index(begin = var_30495_begin_0, end = var_30495_end_0, end_mask = var_30495_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30495_cast_fp16")]; + tensor var_30499_begin_0 = const()[name = tensor("op_30499_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_30499_end_0 = const()[name = tensor("op_30499_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_30499_end_mask_0 = const()[name = tensor("op_30499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30499_cast_fp16 = slice_by_index(begin = var_30499_begin_0, end = var_30499_end_0, end_mask = var_30499_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30499_cast_fp16")]; + tensor var_30503_begin_0 = const()[name = tensor("op_30503_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_30503_end_0 = const()[name = tensor("op_30503_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_30503_end_mask_0 = const()[name = tensor("op_30503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30503_cast_fp16 = slice_by_index(begin = var_30503_begin_0, end = var_30503_end_0, end_mask = var_30503_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30503_cast_fp16")]; + tensor var_30507_begin_0 = const()[name = tensor("op_30507_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_30507_end_0 = const()[name = tensor("op_30507_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_30507_end_mask_0 = const()[name = tensor("op_30507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30507_cast_fp16 = slice_by_index(begin = var_30507_begin_0, end = var_30507_end_0, end_mask = var_30507_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30507_cast_fp16")]; + tensor var_30511_begin_0 = const()[name = tensor("op_30511_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_30511_end_0 = const()[name = tensor("op_30511_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_30511_end_mask_0 = const()[name = tensor("op_30511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30511_cast_fp16 = slice_by_index(begin = var_30511_begin_0, end = var_30511_end_0, end_mask = var_30511_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30511_cast_fp16")]; + tensor var_30515_begin_0 = const()[name = tensor("op_30515_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_30515_end_0 = const()[name = tensor("op_30515_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_30515_end_mask_0 = const()[name = tensor("op_30515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30515_cast_fp16 = slice_by_index(begin = var_30515_begin_0, end = var_30515_end_0, end_mask = var_30515_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30515_cast_fp16")]; + tensor var_30519_begin_0 = const()[name = tensor("op_30519_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_30519_end_0 = const()[name = tensor("op_30519_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_30519_end_mask_0 = const()[name = tensor("op_30519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30519_cast_fp16 = slice_by_index(begin = var_30519_begin_0, end = var_30519_end_0, end_mask = var_30519_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30519_cast_fp16")]; + tensor var_30523_begin_0 = const()[name = tensor("op_30523_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_30523_end_0 = const()[name = tensor("op_30523_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_30523_end_mask_0 = const()[name = tensor("op_30523_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30523_cast_fp16 = slice_by_index(begin = var_30523_begin_0, end = var_30523_end_0, end_mask = var_30523_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30523_cast_fp16")]; + tensor var_30527_begin_0 = const()[name = tensor("op_30527_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_30527_end_0 = const()[name = tensor("op_30527_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_30527_end_mask_0 = const()[name = tensor("op_30527_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30527_cast_fp16 = slice_by_index(begin = var_30527_begin_0, end = var_30527_end_0, end_mask = var_30527_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30527_cast_fp16")]; + tensor var_30531_begin_0 = const()[name = tensor("op_30531_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_30531_end_0 = const()[name = tensor("op_30531_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_30531_end_mask_0 = const()[name = tensor("op_30531_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30531_cast_fp16 = slice_by_index(begin = var_30531_begin_0, end = var_30531_end_0, end_mask = var_30531_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30531_cast_fp16")]; + tensor var_30535_begin_0 = const()[name = tensor("op_30535_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_30535_end_0 = const()[name = tensor("op_30535_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_30535_end_mask_0 = const()[name = tensor("op_30535_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_30535_cast_fp16 = slice_by_index(begin = var_30535_begin_0, end = var_30535_end_0, end_mask = var_30535_end_mask_0, x = value_37_cast_fp16)[name = tensor("op_30535_cast_fp16")]; + tensor var_30539_equation_0 = const()[name = tensor("op_30539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30539_cast_fp16 = einsum(equation = var_30539_equation_0, values = (var_30381_cast_fp16, var_29823_cast_fp16))[name = tensor("op_30539_cast_fp16")]; + tensor var_30540_to_fp16 = const()[name = tensor("op_30540_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2881_cast_fp16 = mul(x = var_30539_cast_fp16, y = var_30540_to_fp16)[name = tensor("aw_chunk_2881_cast_fp16")]; + tensor var_30543_equation_0 = const()[name = tensor("op_30543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30543_cast_fp16 = einsum(equation = var_30543_equation_0, values = (var_30381_cast_fp16, var_29830_cast_fp16))[name = tensor("op_30543_cast_fp16")]; + tensor var_30544_to_fp16 = const()[name = tensor("op_30544_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2883_cast_fp16 = mul(x = var_30543_cast_fp16, y = var_30544_to_fp16)[name = tensor("aw_chunk_2883_cast_fp16")]; + tensor var_30547_equation_0 = const()[name = tensor("op_30547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30547_cast_fp16 = einsum(equation = var_30547_equation_0, values = (var_30381_cast_fp16, var_29837_cast_fp16))[name = tensor("op_30547_cast_fp16")]; + tensor var_30548_to_fp16 = const()[name = tensor("op_30548_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2885_cast_fp16 = mul(x = var_30547_cast_fp16, y = var_30548_to_fp16)[name = tensor("aw_chunk_2885_cast_fp16")]; + tensor var_30551_equation_0 = const()[name = tensor("op_30551_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30551_cast_fp16 = einsum(equation = var_30551_equation_0, values = (var_30381_cast_fp16, var_29844_cast_fp16))[name = tensor("op_30551_cast_fp16")]; + tensor var_30552_to_fp16 = const()[name = tensor("op_30552_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2887_cast_fp16 = mul(x = var_30551_cast_fp16, y = var_30552_to_fp16)[name = tensor("aw_chunk_2887_cast_fp16")]; + tensor var_30555_equation_0 = const()[name = tensor("op_30555_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30555_cast_fp16 = einsum(equation = var_30555_equation_0, values = (var_30385_cast_fp16, var_29851_cast_fp16))[name = tensor("op_30555_cast_fp16")]; + tensor var_30556_to_fp16 = const()[name = tensor("op_30556_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2889_cast_fp16 = mul(x = var_30555_cast_fp16, y = var_30556_to_fp16)[name = tensor("aw_chunk_2889_cast_fp16")]; + tensor var_30559_equation_0 = const()[name = tensor("op_30559_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30559_cast_fp16 = einsum(equation = var_30559_equation_0, values = (var_30385_cast_fp16, var_29858_cast_fp16))[name = tensor("op_30559_cast_fp16")]; + tensor var_30560_to_fp16 = const()[name = tensor("op_30560_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2891_cast_fp16 = mul(x = var_30559_cast_fp16, y = var_30560_to_fp16)[name = tensor("aw_chunk_2891_cast_fp16")]; + tensor var_30563_equation_0 = const()[name = tensor("op_30563_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30563_cast_fp16 = einsum(equation = var_30563_equation_0, values = (var_30385_cast_fp16, var_29865_cast_fp16))[name = tensor("op_30563_cast_fp16")]; + tensor var_30564_to_fp16 = const()[name = tensor("op_30564_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2893_cast_fp16 = mul(x = var_30563_cast_fp16, y = var_30564_to_fp16)[name = tensor("aw_chunk_2893_cast_fp16")]; + tensor var_30567_equation_0 = const()[name = tensor("op_30567_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30567_cast_fp16 = einsum(equation = var_30567_equation_0, values = (var_30385_cast_fp16, var_29872_cast_fp16))[name = tensor("op_30567_cast_fp16")]; + tensor var_30568_to_fp16 = const()[name = tensor("op_30568_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2895_cast_fp16 = mul(x = var_30567_cast_fp16, y = var_30568_to_fp16)[name = tensor("aw_chunk_2895_cast_fp16")]; + tensor var_30571_equation_0 = const()[name = tensor("op_30571_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30571_cast_fp16 = einsum(equation = var_30571_equation_0, values = (var_30389_cast_fp16, var_29879_cast_fp16))[name = tensor("op_30571_cast_fp16")]; + tensor var_30572_to_fp16 = const()[name = tensor("op_30572_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2897_cast_fp16 = mul(x = var_30571_cast_fp16, y = var_30572_to_fp16)[name = tensor("aw_chunk_2897_cast_fp16")]; + tensor var_30575_equation_0 = const()[name = tensor("op_30575_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30575_cast_fp16 = einsum(equation = var_30575_equation_0, values = (var_30389_cast_fp16, var_29886_cast_fp16))[name = tensor("op_30575_cast_fp16")]; + tensor var_30576_to_fp16 = const()[name = tensor("op_30576_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2899_cast_fp16 = mul(x = var_30575_cast_fp16, y = var_30576_to_fp16)[name = tensor("aw_chunk_2899_cast_fp16")]; + tensor var_30579_equation_0 = const()[name = tensor("op_30579_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30579_cast_fp16 = einsum(equation = var_30579_equation_0, values = (var_30389_cast_fp16, var_29893_cast_fp16))[name = tensor("op_30579_cast_fp16")]; + tensor var_30580_to_fp16 = const()[name = tensor("op_30580_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2901_cast_fp16 = mul(x = var_30579_cast_fp16, y = var_30580_to_fp16)[name = tensor("aw_chunk_2901_cast_fp16")]; + tensor var_30583_equation_0 = const()[name = tensor("op_30583_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30583_cast_fp16 = einsum(equation = var_30583_equation_0, values = (var_30389_cast_fp16, var_29900_cast_fp16))[name = tensor("op_30583_cast_fp16")]; + tensor var_30584_to_fp16 = const()[name = tensor("op_30584_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2903_cast_fp16 = mul(x = var_30583_cast_fp16, y = var_30584_to_fp16)[name = tensor("aw_chunk_2903_cast_fp16")]; + tensor var_30587_equation_0 = const()[name = tensor("op_30587_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30587_cast_fp16 = einsum(equation = var_30587_equation_0, values = (var_30393_cast_fp16, var_29907_cast_fp16))[name = tensor("op_30587_cast_fp16")]; + tensor var_30588_to_fp16 = const()[name = tensor("op_30588_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2905_cast_fp16 = mul(x = var_30587_cast_fp16, y = var_30588_to_fp16)[name = tensor("aw_chunk_2905_cast_fp16")]; + tensor var_30591_equation_0 = const()[name = tensor("op_30591_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30591_cast_fp16 = einsum(equation = var_30591_equation_0, values = (var_30393_cast_fp16, var_29914_cast_fp16))[name = tensor("op_30591_cast_fp16")]; + tensor var_30592_to_fp16 = const()[name = tensor("op_30592_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2907_cast_fp16 = mul(x = var_30591_cast_fp16, y = var_30592_to_fp16)[name = tensor("aw_chunk_2907_cast_fp16")]; + tensor var_30595_equation_0 = const()[name = tensor("op_30595_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30595_cast_fp16 = einsum(equation = var_30595_equation_0, values = (var_30393_cast_fp16, var_29921_cast_fp16))[name = tensor("op_30595_cast_fp16")]; + tensor var_30596_to_fp16 = const()[name = tensor("op_30596_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2909_cast_fp16 = mul(x = var_30595_cast_fp16, y = var_30596_to_fp16)[name = tensor("aw_chunk_2909_cast_fp16")]; + tensor var_30599_equation_0 = const()[name = tensor("op_30599_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30599_cast_fp16 = einsum(equation = var_30599_equation_0, values = (var_30393_cast_fp16, var_29928_cast_fp16))[name = tensor("op_30599_cast_fp16")]; + tensor var_30600_to_fp16 = const()[name = tensor("op_30600_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2911_cast_fp16 = mul(x = var_30599_cast_fp16, y = var_30600_to_fp16)[name = tensor("aw_chunk_2911_cast_fp16")]; + tensor var_30603_equation_0 = const()[name = tensor("op_30603_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30603_cast_fp16 = einsum(equation = var_30603_equation_0, values = (var_30397_cast_fp16, var_29935_cast_fp16))[name = tensor("op_30603_cast_fp16")]; + tensor var_30604_to_fp16 = const()[name = tensor("op_30604_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2913_cast_fp16 = mul(x = var_30603_cast_fp16, y = var_30604_to_fp16)[name = tensor("aw_chunk_2913_cast_fp16")]; + tensor var_30607_equation_0 = const()[name = tensor("op_30607_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30607_cast_fp16 = einsum(equation = var_30607_equation_0, values = (var_30397_cast_fp16, var_29942_cast_fp16))[name = tensor("op_30607_cast_fp16")]; + tensor var_30608_to_fp16 = const()[name = tensor("op_30608_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2915_cast_fp16 = mul(x = var_30607_cast_fp16, y = var_30608_to_fp16)[name = tensor("aw_chunk_2915_cast_fp16")]; + tensor var_30611_equation_0 = const()[name = tensor("op_30611_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30611_cast_fp16 = einsum(equation = var_30611_equation_0, values = (var_30397_cast_fp16, var_29949_cast_fp16))[name = tensor("op_30611_cast_fp16")]; + tensor var_30612_to_fp16 = const()[name = tensor("op_30612_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2917_cast_fp16 = mul(x = var_30611_cast_fp16, y = var_30612_to_fp16)[name = tensor("aw_chunk_2917_cast_fp16")]; + tensor var_30615_equation_0 = const()[name = tensor("op_30615_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30615_cast_fp16 = einsum(equation = var_30615_equation_0, values = (var_30397_cast_fp16, var_29956_cast_fp16))[name = tensor("op_30615_cast_fp16")]; + tensor var_30616_to_fp16 = const()[name = tensor("op_30616_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2919_cast_fp16 = mul(x = var_30615_cast_fp16, y = var_30616_to_fp16)[name = tensor("aw_chunk_2919_cast_fp16")]; + tensor var_30619_equation_0 = const()[name = tensor("op_30619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30619_cast_fp16 = einsum(equation = var_30619_equation_0, values = (var_30401_cast_fp16, var_29963_cast_fp16))[name = tensor("op_30619_cast_fp16")]; + tensor var_30620_to_fp16 = const()[name = tensor("op_30620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2921_cast_fp16 = mul(x = var_30619_cast_fp16, y = var_30620_to_fp16)[name = tensor("aw_chunk_2921_cast_fp16")]; + tensor var_30623_equation_0 = const()[name = tensor("op_30623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30623_cast_fp16 = einsum(equation = var_30623_equation_0, values = (var_30401_cast_fp16, var_29970_cast_fp16))[name = tensor("op_30623_cast_fp16")]; + tensor var_30624_to_fp16 = const()[name = tensor("op_30624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2923_cast_fp16 = mul(x = var_30623_cast_fp16, y = var_30624_to_fp16)[name = tensor("aw_chunk_2923_cast_fp16")]; + tensor var_30627_equation_0 = const()[name = tensor("op_30627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30627_cast_fp16 = einsum(equation = var_30627_equation_0, values = (var_30401_cast_fp16, var_29977_cast_fp16))[name = tensor("op_30627_cast_fp16")]; + tensor var_30628_to_fp16 = const()[name = tensor("op_30628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2925_cast_fp16 = mul(x = var_30627_cast_fp16, y = var_30628_to_fp16)[name = tensor("aw_chunk_2925_cast_fp16")]; + tensor var_30631_equation_0 = const()[name = tensor("op_30631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30631_cast_fp16 = einsum(equation = var_30631_equation_0, values = (var_30401_cast_fp16, var_29984_cast_fp16))[name = tensor("op_30631_cast_fp16")]; + tensor var_30632_to_fp16 = const()[name = tensor("op_30632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2927_cast_fp16 = mul(x = var_30631_cast_fp16, y = var_30632_to_fp16)[name = tensor("aw_chunk_2927_cast_fp16")]; + tensor var_30635_equation_0 = const()[name = tensor("op_30635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30635_cast_fp16 = einsum(equation = var_30635_equation_0, values = (var_30405_cast_fp16, var_29991_cast_fp16))[name = tensor("op_30635_cast_fp16")]; + tensor var_30636_to_fp16 = const()[name = tensor("op_30636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2929_cast_fp16 = mul(x = var_30635_cast_fp16, y = var_30636_to_fp16)[name = tensor("aw_chunk_2929_cast_fp16")]; + tensor var_30639_equation_0 = const()[name = tensor("op_30639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30639_cast_fp16 = einsum(equation = var_30639_equation_0, values = (var_30405_cast_fp16, var_29998_cast_fp16))[name = tensor("op_30639_cast_fp16")]; + tensor var_30640_to_fp16 = const()[name = tensor("op_30640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2931_cast_fp16 = mul(x = var_30639_cast_fp16, y = var_30640_to_fp16)[name = tensor("aw_chunk_2931_cast_fp16")]; + tensor var_30643_equation_0 = const()[name = tensor("op_30643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30643_cast_fp16 = einsum(equation = var_30643_equation_0, values = (var_30405_cast_fp16, var_30005_cast_fp16))[name = tensor("op_30643_cast_fp16")]; + tensor var_30644_to_fp16 = const()[name = tensor("op_30644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2933_cast_fp16 = mul(x = var_30643_cast_fp16, y = var_30644_to_fp16)[name = tensor("aw_chunk_2933_cast_fp16")]; + tensor var_30647_equation_0 = const()[name = tensor("op_30647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30647_cast_fp16 = einsum(equation = var_30647_equation_0, values = (var_30405_cast_fp16, var_30012_cast_fp16))[name = tensor("op_30647_cast_fp16")]; + tensor var_30648_to_fp16 = const()[name = tensor("op_30648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2935_cast_fp16 = mul(x = var_30647_cast_fp16, y = var_30648_to_fp16)[name = tensor("aw_chunk_2935_cast_fp16")]; + tensor var_30651_equation_0 = const()[name = tensor("op_30651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30651_cast_fp16 = einsum(equation = var_30651_equation_0, values = (var_30409_cast_fp16, var_30019_cast_fp16))[name = tensor("op_30651_cast_fp16")]; + tensor var_30652_to_fp16 = const()[name = tensor("op_30652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2937_cast_fp16 = mul(x = var_30651_cast_fp16, y = var_30652_to_fp16)[name = tensor("aw_chunk_2937_cast_fp16")]; + tensor var_30655_equation_0 = const()[name = tensor("op_30655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30655_cast_fp16 = einsum(equation = var_30655_equation_0, values = (var_30409_cast_fp16, var_30026_cast_fp16))[name = tensor("op_30655_cast_fp16")]; + tensor var_30656_to_fp16 = const()[name = tensor("op_30656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2939_cast_fp16 = mul(x = var_30655_cast_fp16, y = var_30656_to_fp16)[name = tensor("aw_chunk_2939_cast_fp16")]; + tensor var_30659_equation_0 = const()[name = tensor("op_30659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30659_cast_fp16 = einsum(equation = var_30659_equation_0, values = (var_30409_cast_fp16, var_30033_cast_fp16))[name = tensor("op_30659_cast_fp16")]; + tensor var_30660_to_fp16 = const()[name = tensor("op_30660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2941_cast_fp16 = mul(x = var_30659_cast_fp16, y = var_30660_to_fp16)[name = tensor("aw_chunk_2941_cast_fp16")]; + tensor var_30663_equation_0 = const()[name = tensor("op_30663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30663_cast_fp16 = einsum(equation = var_30663_equation_0, values = (var_30409_cast_fp16, var_30040_cast_fp16))[name = tensor("op_30663_cast_fp16")]; + tensor var_30664_to_fp16 = const()[name = tensor("op_30664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2943_cast_fp16 = mul(x = var_30663_cast_fp16, y = var_30664_to_fp16)[name = tensor("aw_chunk_2943_cast_fp16")]; + tensor var_30667_equation_0 = const()[name = tensor("op_30667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30667_cast_fp16 = einsum(equation = var_30667_equation_0, values = (var_30413_cast_fp16, var_30047_cast_fp16))[name = tensor("op_30667_cast_fp16")]; + tensor var_30668_to_fp16 = const()[name = tensor("op_30668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2945_cast_fp16 = mul(x = var_30667_cast_fp16, y = var_30668_to_fp16)[name = tensor("aw_chunk_2945_cast_fp16")]; + tensor var_30671_equation_0 = const()[name = tensor("op_30671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30671_cast_fp16 = einsum(equation = var_30671_equation_0, values = (var_30413_cast_fp16, var_30054_cast_fp16))[name = tensor("op_30671_cast_fp16")]; + tensor var_30672_to_fp16 = const()[name = tensor("op_30672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2947_cast_fp16 = mul(x = var_30671_cast_fp16, y = var_30672_to_fp16)[name = tensor("aw_chunk_2947_cast_fp16")]; + tensor var_30675_equation_0 = const()[name = tensor("op_30675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30675_cast_fp16 = einsum(equation = var_30675_equation_0, values = (var_30413_cast_fp16, var_30061_cast_fp16))[name = tensor("op_30675_cast_fp16")]; + tensor var_30676_to_fp16 = const()[name = tensor("op_30676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2949_cast_fp16 = mul(x = var_30675_cast_fp16, y = var_30676_to_fp16)[name = tensor("aw_chunk_2949_cast_fp16")]; + tensor var_30679_equation_0 = const()[name = tensor("op_30679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30679_cast_fp16 = einsum(equation = var_30679_equation_0, values = (var_30413_cast_fp16, var_30068_cast_fp16))[name = tensor("op_30679_cast_fp16")]; + tensor var_30680_to_fp16 = const()[name = tensor("op_30680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2951_cast_fp16 = mul(x = var_30679_cast_fp16, y = var_30680_to_fp16)[name = tensor("aw_chunk_2951_cast_fp16")]; + tensor var_30683_equation_0 = const()[name = tensor("op_30683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30683_cast_fp16 = einsum(equation = var_30683_equation_0, values = (var_30417_cast_fp16, var_30075_cast_fp16))[name = tensor("op_30683_cast_fp16")]; + tensor var_30684_to_fp16 = const()[name = tensor("op_30684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2953_cast_fp16 = mul(x = var_30683_cast_fp16, y = var_30684_to_fp16)[name = tensor("aw_chunk_2953_cast_fp16")]; + tensor var_30687_equation_0 = const()[name = tensor("op_30687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30687_cast_fp16 = einsum(equation = var_30687_equation_0, values = (var_30417_cast_fp16, var_30082_cast_fp16))[name = tensor("op_30687_cast_fp16")]; + tensor var_30688_to_fp16 = const()[name = tensor("op_30688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2955_cast_fp16 = mul(x = var_30687_cast_fp16, y = var_30688_to_fp16)[name = tensor("aw_chunk_2955_cast_fp16")]; + tensor var_30691_equation_0 = const()[name = tensor("op_30691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30691_cast_fp16 = einsum(equation = var_30691_equation_0, values = (var_30417_cast_fp16, var_30089_cast_fp16))[name = tensor("op_30691_cast_fp16")]; + tensor var_30692_to_fp16 = const()[name = tensor("op_30692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2957_cast_fp16 = mul(x = var_30691_cast_fp16, y = var_30692_to_fp16)[name = tensor("aw_chunk_2957_cast_fp16")]; + tensor var_30695_equation_0 = const()[name = tensor("op_30695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30695_cast_fp16 = einsum(equation = var_30695_equation_0, values = (var_30417_cast_fp16, var_30096_cast_fp16))[name = tensor("op_30695_cast_fp16")]; + tensor var_30696_to_fp16 = const()[name = tensor("op_30696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2959_cast_fp16 = mul(x = var_30695_cast_fp16, y = var_30696_to_fp16)[name = tensor("aw_chunk_2959_cast_fp16")]; + tensor var_30699_equation_0 = const()[name = tensor("op_30699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30699_cast_fp16 = einsum(equation = var_30699_equation_0, values = (var_30421_cast_fp16, var_30103_cast_fp16))[name = tensor("op_30699_cast_fp16")]; + tensor var_30700_to_fp16 = const()[name = tensor("op_30700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2961_cast_fp16 = mul(x = var_30699_cast_fp16, y = var_30700_to_fp16)[name = tensor("aw_chunk_2961_cast_fp16")]; + tensor var_30703_equation_0 = const()[name = tensor("op_30703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30703_cast_fp16 = einsum(equation = var_30703_equation_0, values = (var_30421_cast_fp16, var_30110_cast_fp16))[name = tensor("op_30703_cast_fp16")]; + tensor var_30704_to_fp16 = const()[name = tensor("op_30704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2963_cast_fp16 = mul(x = var_30703_cast_fp16, y = var_30704_to_fp16)[name = tensor("aw_chunk_2963_cast_fp16")]; + tensor var_30707_equation_0 = const()[name = tensor("op_30707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30707_cast_fp16 = einsum(equation = var_30707_equation_0, values = (var_30421_cast_fp16, var_30117_cast_fp16))[name = tensor("op_30707_cast_fp16")]; + tensor var_30708_to_fp16 = const()[name = tensor("op_30708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2965_cast_fp16 = mul(x = var_30707_cast_fp16, y = var_30708_to_fp16)[name = tensor("aw_chunk_2965_cast_fp16")]; + tensor var_30711_equation_0 = const()[name = tensor("op_30711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30711_cast_fp16 = einsum(equation = var_30711_equation_0, values = (var_30421_cast_fp16, var_30124_cast_fp16))[name = tensor("op_30711_cast_fp16")]; + tensor var_30712_to_fp16 = const()[name = tensor("op_30712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2967_cast_fp16 = mul(x = var_30711_cast_fp16, y = var_30712_to_fp16)[name = tensor("aw_chunk_2967_cast_fp16")]; + tensor var_30715_equation_0 = const()[name = tensor("op_30715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30715_cast_fp16 = einsum(equation = var_30715_equation_0, values = (var_30425_cast_fp16, var_30131_cast_fp16))[name = tensor("op_30715_cast_fp16")]; + tensor var_30716_to_fp16 = const()[name = tensor("op_30716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2969_cast_fp16 = mul(x = var_30715_cast_fp16, y = var_30716_to_fp16)[name = tensor("aw_chunk_2969_cast_fp16")]; + tensor var_30719_equation_0 = const()[name = tensor("op_30719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30719_cast_fp16 = einsum(equation = var_30719_equation_0, values = (var_30425_cast_fp16, var_30138_cast_fp16))[name = tensor("op_30719_cast_fp16")]; + tensor var_30720_to_fp16 = const()[name = tensor("op_30720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2971_cast_fp16 = mul(x = var_30719_cast_fp16, y = var_30720_to_fp16)[name = tensor("aw_chunk_2971_cast_fp16")]; + tensor var_30723_equation_0 = const()[name = tensor("op_30723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30723_cast_fp16 = einsum(equation = var_30723_equation_0, values = (var_30425_cast_fp16, var_30145_cast_fp16))[name = tensor("op_30723_cast_fp16")]; + tensor var_30724_to_fp16 = const()[name = tensor("op_30724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2973_cast_fp16 = mul(x = var_30723_cast_fp16, y = var_30724_to_fp16)[name = tensor("aw_chunk_2973_cast_fp16")]; + tensor var_30727_equation_0 = const()[name = tensor("op_30727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30727_cast_fp16 = einsum(equation = var_30727_equation_0, values = (var_30425_cast_fp16, var_30152_cast_fp16))[name = tensor("op_30727_cast_fp16")]; + tensor var_30728_to_fp16 = const()[name = tensor("op_30728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2975_cast_fp16 = mul(x = var_30727_cast_fp16, y = var_30728_to_fp16)[name = tensor("aw_chunk_2975_cast_fp16")]; + tensor var_30731_equation_0 = const()[name = tensor("op_30731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30731_cast_fp16 = einsum(equation = var_30731_equation_0, values = (var_30429_cast_fp16, var_30159_cast_fp16))[name = tensor("op_30731_cast_fp16")]; + tensor var_30732_to_fp16 = const()[name = tensor("op_30732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2977_cast_fp16 = mul(x = var_30731_cast_fp16, y = var_30732_to_fp16)[name = tensor("aw_chunk_2977_cast_fp16")]; + tensor var_30735_equation_0 = const()[name = tensor("op_30735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30735_cast_fp16 = einsum(equation = var_30735_equation_0, values = (var_30429_cast_fp16, var_30166_cast_fp16))[name = tensor("op_30735_cast_fp16")]; + tensor var_30736_to_fp16 = const()[name = tensor("op_30736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2979_cast_fp16 = mul(x = var_30735_cast_fp16, y = var_30736_to_fp16)[name = tensor("aw_chunk_2979_cast_fp16")]; + tensor var_30739_equation_0 = const()[name = tensor("op_30739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30739_cast_fp16 = einsum(equation = var_30739_equation_0, values = (var_30429_cast_fp16, var_30173_cast_fp16))[name = tensor("op_30739_cast_fp16")]; + tensor var_30740_to_fp16 = const()[name = tensor("op_30740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2981_cast_fp16 = mul(x = var_30739_cast_fp16, y = var_30740_to_fp16)[name = tensor("aw_chunk_2981_cast_fp16")]; + tensor var_30743_equation_0 = const()[name = tensor("op_30743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30743_cast_fp16 = einsum(equation = var_30743_equation_0, values = (var_30429_cast_fp16, var_30180_cast_fp16))[name = tensor("op_30743_cast_fp16")]; + tensor var_30744_to_fp16 = const()[name = tensor("op_30744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2983_cast_fp16 = mul(x = var_30743_cast_fp16, y = var_30744_to_fp16)[name = tensor("aw_chunk_2983_cast_fp16")]; + tensor var_30747_equation_0 = const()[name = tensor("op_30747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30747_cast_fp16 = einsum(equation = var_30747_equation_0, values = (var_30433_cast_fp16, var_30187_cast_fp16))[name = tensor("op_30747_cast_fp16")]; + tensor var_30748_to_fp16 = const()[name = tensor("op_30748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2985_cast_fp16 = mul(x = var_30747_cast_fp16, y = var_30748_to_fp16)[name = tensor("aw_chunk_2985_cast_fp16")]; + tensor var_30751_equation_0 = const()[name = tensor("op_30751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30751_cast_fp16 = einsum(equation = var_30751_equation_0, values = (var_30433_cast_fp16, var_30194_cast_fp16))[name = tensor("op_30751_cast_fp16")]; + tensor var_30752_to_fp16 = const()[name = tensor("op_30752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2987_cast_fp16 = mul(x = var_30751_cast_fp16, y = var_30752_to_fp16)[name = tensor("aw_chunk_2987_cast_fp16")]; + tensor var_30755_equation_0 = const()[name = tensor("op_30755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30755_cast_fp16 = einsum(equation = var_30755_equation_0, values = (var_30433_cast_fp16, var_30201_cast_fp16))[name = tensor("op_30755_cast_fp16")]; + tensor var_30756_to_fp16 = const()[name = tensor("op_30756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2989_cast_fp16 = mul(x = var_30755_cast_fp16, y = var_30756_to_fp16)[name = tensor("aw_chunk_2989_cast_fp16")]; + tensor var_30759_equation_0 = const()[name = tensor("op_30759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30759_cast_fp16 = einsum(equation = var_30759_equation_0, values = (var_30433_cast_fp16, var_30208_cast_fp16))[name = tensor("op_30759_cast_fp16")]; + tensor var_30760_to_fp16 = const()[name = tensor("op_30760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2991_cast_fp16 = mul(x = var_30759_cast_fp16, y = var_30760_to_fp16)[name = tensor("aw_chunk_2991_cast_fp16")]; + tensor var_30763_equation_0 = const()[name = tensor("op_30763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30763_cast_fp16 = einsum(equation = var_30763_equation_0, values = (var_30437_cast_fp16, var_30215_cast_fp16))[name = tensor("op_30763_cast_fp16")]; + tensor var_30764_to_fp16 = const()[name = tensor("op_30764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2993_cast_fp16 = mul(x = var_30763_cast_fp16, y = var_30764_to_fp16)[name = tensor("aw_chunk_2993_cast_fp16")]; + tensor var_30767_equation_0 = const()[name = tensor("op_30767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30767_cast_fp16 = einsum(equation = var_30767_equation_0, values = (var_30437_cast_fp16, var_30222_cast_fp16))[name = tensor("op_30767_cast_fp16")]; + tensor var_30768_to_fp16 = const()[name = tensor("op_30768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2995_cast_fp16 = mul(x = var_30767_cast_fp16, y = var_30768_to_fp16)[name = tensor("aw_chunk_2995_cast_fp16")]; + tensor var_30771_equation_0 = const()[name = tensor("op_30771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30771_cast_fp16 = einsum(equation = var_30771_equation_0, values = (var_30437_cast_fp16, var_30229_cast_fp16))[name = tensor("op_30771_cast_fp16")]; + tensor var_30772_to_fp16 = const()[name = tensor("op_30772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2997_cast_fp16 = mul(x = var_30771_cast_fp16, y = var_30772_to_fp16)[name = tensor("aw_chunk_2997_cast_fp16")]; + tensor var_30775_equation_0 = const()[name = tensor("op_30775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30775_cast_fp16 = einsum(equation = var_30775_equation_0, values = (var_30437_cast_fp16, var_30236_cast_fp16))[name = tensor("op_30775_cast_fp16")]; + tensor var_30776_to_fp16 = const()[name = tensor("op_30776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_2999_cast_fp16 = mul(x = var_30775_cast_fp16, y = var_30776_to_fp16)[name = tensor("aw_chunk_2999_cast_fp16")]; + tensor var_30779_equation_0 = const()[name = tensor("op_30779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30779_cast_fp16 = einsum(equation = var_30779_equation_0, values = (var_30441_cast_fp16, var_30243_cast_fp16))[name = tensor("op_30779_cast_fp16")]; + tensor var_30780_to_fp16 = const()[name = tensor("op_30780_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3001_cast_fp16 = mul(x = var_30779_cast_fp16, y = var_30780_to_fp16)[name = tensor("aw_chunk_3001_cast_fp16")]; + tensor var_30783_equation_0 = const()[name = tensor("op_30783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30783_cast_fp16 = einsum(equation = var_30783_equation_0, values = (var_30441_cast_fp16, var_30250_cast_fp16))[name = tensor("op_30783_cast_fp16")]; + tensor var_30784_to_fp16 = const()[name = tensor("op_30784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3003_cast_fp16 = mul(x = var_30783_cast_fp16, y = var_30784_to_fp16)[name = tensor("aw_chunk_3003_cast_fp16")]; + tensor var_30787_equation_0 = const()[name = tensor("op_30787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30787_cast_fp16 = einsum(equation = var_30787_equation_0, values = (var_30441_cast_fp16, var_30257_cast_fp16))[name = tensor("op_30787_cast_fp16")]; + tensor var_30788_to_fp16 = const()[name = tensor("op_30788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3005_cast_fp16 = mul(x = var_30787_cast_fp16, y = var_30788_to_fp16)[name = tensor("aw_chunk_3005_cast_fp16")]; + tensor var_30791_equation_0 = const()[name = tensor("op_30791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30791_cast_fp16 = einsum(equation = var_30791_equation_0, values = (var_30441_cast_fp16, var_30264_cast_fp16))[name = tensor("op_30791_cast_fp16")]; + tensor var_30792_to_fp16 = const()[name = tensor("op_30792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3007_cast_fp16 = mul(x = var_30791_cast_fp16, y = var_30792_to_fp16)[name = tensor("aw_chunk_3007_cast_fp16")]; + tensor var_30795_equation_0 = const()[name = tensor("op_30795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30795_cast_fp16 = einsum(equation = var_30795_equation_0, values = (var_30445_cast_fp16, var_30271_cast_fp16))[name = tensor("op_30795_cast_fp16")]; + tensor var_30796_to_fp16 = const()[name = tensor("op_30796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3009_cast_fp16 = mul(x = var_30795_cast_fp16, y = var_30796_to_fp16)[name = tensor("aw_chunk_3009_cast_fp16")]; + tensor var_30799_equation_0 = const()[name = tensor("op_30799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30799_cast_fp16 = einsum(equation = var_30799_equation_0, values = (var_30445_cast_fp16, var_30278_cast_fp16))[name = tensor("op_30799_cast_fp16")]; + tensor var_30800_to_fp16 = const()[name = tensor("op_30800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3011_cast_fp16 = mul(x = var_30799_cast_fp16, y = var_30800_to_fp16)[name = tensor("aw_chunk_3011_cast_fp16")]; + tensor var_30803_equation_0 = const()[name = tensor("op_30803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30803_cast_fp16 = einsum(equation = var_30803_equation_0, values = (var_30445_cast_fp16, var_30285_cast_fp16))[name = tensor("op_30803_cast_fp16")]; + tensor var_30804_to_fp16 = const()[name = tensor("op_30804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3013_cast_fp16 = mul(x = var_30803_cast_fp16, y = var_30804_to_fp16)[name = tensor("aw_chunk_3013_cast_fp16")]; + tensor var_30807_equation_0 = const()[name = tensor("op_30807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30807_cast_fp16 = einsum(equation = var_30807_equation_0, values = (var_30445_cast_fp16, var_30292_cast_fp16))[name = tensor("op_30807_cast_fp16")]; + tensor var_30808_to_fp16 = const()[name = tensor("op_30808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3015_cast_fp16 = mul(x = var_30807_cast_fp16, y = var_30808_to_fp16)[name = tensor("aw_chunk_3015_cast_fp16")]; + tensor var_30811_equation_0 = const()[name = tensor("op_30811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30811_cast_fp16 = einsum(equation = var_30811_equation_0, values = (var_30449_cast_fp16, var_30299_cast_fp16))[name = tensor("op_30811_cast_fp16")]; + tensor var_30812_to_fp16 = const()[name = tensor("op_30812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3017_cast_fp16 = mul(x = var_30811_cast_fp16, y = var_30812_to_fp16)[name = tensor("aw_chunk_3017_cast_fp16")]; + tensor var_30815_equation_0 = const()[name = tensor("op_30815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30815_cast_fp16 = einsum(equation = var_30815_equation_0, values = (var_30449_cast_fp16, var_30306_cast_fp16))[name = tensor("op_30815_cast_fp16")]; + tensor var_30816_to_fp16 = const()[name = tensor("op_30816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3019_cast_fp16 = mul(x = var_30815_cast_fp16, y = var_30816_to_fp16)[name = tensor("aw_chunk_3019_cast_fp16")]; + tensor var_30819_equation_0 = const()[name = tensor("op_30819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30819_cast_fp16 = einsum(equation = var_30819_equation_0, values = (var_30449_cast_fp16, var_30313_cast_fp16))[name = tensor("op_30819_cast_fp16")]; + tensor var_30820_to_fp16 = const()[name = tensor("op_30820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3021_cast_fp16 = mul(x = var_30819_cast_fp16, y = var_30820_to_fp16)[name = tensor("aw_chunk_3021_cast_fp16")]; + tensor var_30823_equation_0 = const()[name = tensor("op_30823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30823_cast_fp16 = einsum(equation = var_30823_equation_0, values = (var_30449_cast_fp16, var_30320_cast_fp16))[name = tensor("op_30823_cast_fp16")]; + tensor var_30824_to_fp16 = const()[name = tensor("op_30824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3023_cast_fp16 = mul(x = var_30823_cast_fp16, y = var_30824_to_fp16)[name = tensor("aw_chunk_3023_cast_fp16")]; + tensor var_30827_equation_0 = const()[name = tensor("op_30827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30827_cast_fp16 = einsum(equation = var_30827_equation_0, values = (var_30453_cast_fp16, var_30327_cast_fp16))[name = tensor("op_30827_cast_fp16")]; + tensor var_30828_to_fp16 = const()[name = tensor("op_30828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3025_cast_fp16 = mul(x = var_30827_cast_fp16, y = var_30828_to_fp16)[name = tensor("aw_chunk_3025_cast_fp16")]; + tensor var_30831_equation_0 = const()[name = tensor("op_30831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30831_cast_fp16 = einsum(equation = var_30831_equation_0, values = (var_30453_cast_fp16, var_30334_cast_fp16))[name = tensor("op_30831_cast_fp16")]; + tensor var_30832_to_fp16 = const()[name = tensor("op_30832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3027_cast_fp16 = mul(x = var_30831_cast_fp16, y = var_30832_to_fp16)[name = tensor("aw_chunk_3027_cast_fp16")]; + tensor var_30835_equation_0 = const()[name = tensor("op_30835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30835_cast_fp16 = einsum(equation = var_30835_equation_0, values = (var_30453_cast_fp16, var_30341_cast_fp16))[name = tensor("op_30835_cast_fp16")]; + tensor var_30836_to_fp16 = const()[name = tensor("op_30836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3029_cast_fp16 = mul(x = var_30835_cast_fp16, y = var_30836_to_fp16)[name = tensor("aw_chunk_3029_cast_fp16")]; + tensor var_30839_equation_0 = const()[name = tensor("op_30839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30839_cast_fp16 = einsum(equation = var_30839_equation_0, values = (var_30453_cast_fp16, var_30348_cast_fp16))[name = tensor("op_30839_cast_fp16")]; + tensor var_30840_to_fp16 = const()[name = tensor("op_30840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3031_cast_fp16 = mul(x = var_30839_cast_fp16, y = var_30840_to_fp16)[name = tensor("aw_chunk_3031_cast_fp16")]; + tensor var_30843_equation_0 = const()[name = tensor("op_30843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30843_cast_fp16 = einsum(equation = var_30843_equation_0, values = (var_30457_cast_fp16, var_30355_cast_fp16))[name = tensor("op_30843_cast_fp16")]; + tensor var_30844_to_fp16 = const()[name = tensor("op_30844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3033_cast_fp16 = mul(x = var_30843_cast_fp16, y = var_30844_to_fp16)[name = tensor("aw_chunk_3033_cast_fp16")]; + tensor var_30847_equation_0 = const()[name = tensor("op_30847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30847_cast_fp16 = einsum(equation = var_30847_equation_0, values = (var_30457_cast_fp16, var_30362_cast_fp16))[name = tensor("op_30847_cast_fp16")]; + tensor var_30848_to_fp16 = const()[name = tensor("op_30848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3035_cast_fp16 = mul(x = var_30847_cast_fp16, y = var_30848_to_fp16)[name = tensor("aw_chunk_3035_cast_fp16")]; + tensor var_30851_equation_0 = const()[name = tensor("op_30851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30851_cast_fp16 = einsum(equation = var_30851_equation_0, values = (var_30457_cast_fp16, var_30369_cast_fp16))[name = tensor("op_30851_cast_fp16")]; + tensor var_30852_to_fp16 = const()[name = tensor("op_30852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3037_cast_fp16 = mul(x = var_30851_cast_fp16, y = var_30852_to_fp16)[name = tensor("aw_chunk_3037_cast_fp16")]; + tensor var_30855_equation_0 = const()[name = tensor("op_30855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_30855_cast_fp16 = einsum(equation = var_30855_equation_0, values = (var_30457_cast_fp16, var_30376_cast_fp16))[name = tensor("op_30855_cast_fp16")]; + tensor var_30856_to_fp16 = const()[name = tensor("op_30856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3039_cast_fp16 = mul(x = var_30855_cast_fp16, y = var_30856_to_fp16)[name = tensor("aw_chunk_3039_cast_fp16")]; + tensor var_30858_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2881_cast_fp16)[name = tensor("op_30858_cast_fp16")]; + tensor var_30859_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2883_cast_fp16)[name = tensor("op_30859_cast_fp16")]; + tensor var_30860_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2885_cast_fp16)[name = tensor("op_30860_cast_fp16")]; + tensor var_30861_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2887_cast_fp16)[name = tensor("op_30861_cast_fp16")]; + tensor var_30862_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2889_cast_fp16)[name = tensor("op_30862_cast_fp16")]; + tensor var_30863_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2891_cast_fp16)[name = tensor("op_30863_cast_fp16")]; + tensor var_30864_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2893_cast_fp16)[name = tensor("op_30864_cast_fp16")]; + tensor var_30865_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2895_cast_fp16)[name = tensor("op_30865_cast_fp16")]; + tensor var_30866_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2897_cast_fp16)[name = tensor("op_30866_cast_fp16")]; + tensor var_30867_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2899_cast_fp16)[name = tensor("op_30867_cast_fp16")]; + tensor var_30868_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2901_cast_fp16)[name = tensor("op_30868_cast_fp16")]; + tensor var_30869_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2903_cast_fp16)[name = tensor("op_30869_cast_fp16")]; + tensor var_30870_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2905_cast_fp16)[name = tensor("op_30870_cast_fp16")]; + tensor var_30871_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2907_cast_fp16)[name = tensor("op_30871_cast_fp16")]; + tensor var_30872_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2909_cast_fp16)[name = tensor("op_30872_cast_fp16")]; + tensor var_30873_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2911_cast_fp16)[name = tensor("op_30873_cast_fp16")]; + tensor var_30874_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2913_cast_fp16)[name = tensor("op_30874_cast_fp16")]; + tensor var_30875_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2915_cast_fp16)[name = tensor("op_30875_cast_fp16")]; + tensor var_30876_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2917_cast_fp16)[name = tensor("op_30876_cast_fp16")]; + tensor var_30877_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2919_cast_fp16)[name = tensor("op_30877_cast_fp16")]; + tensor var_30878_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2921_cast_fp16)[name = tensor("op_30878_cast_fp16")]; + tensor var_30879_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2923_cast_fp16)[name = tensor("op_30879_cast_fp16")]; + tensor var_30880_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2925_cast_fp16)[name = tensor("op_30880_cast_fp16")]; + tensor var_30881_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2927_cast_fp16)[name = tensor("op_30881_cast_fp16")]; + tensor var_30882_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2929_cast_fp16)[name = tensor("op_30882_cast_fp16")]; + tensor var_30883_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2931_cast_fp16)[name = tensor("op_30883_cast_fp16")]; + tensor var_30884_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2933_cast_fp16)[name = tensor("op_30884_cast_fp16")]; + tensor var_30885_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2935_cast_fp16)[name = tensor("op_30885_cast_fp16")]; + tensor var_30886_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2937_cast_fp16)[name = tensor("op_30886_cast_fp16")]; + tensor var_30887_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2939_cast_fp16)[name = tensor("op_30887_cast_fp16")]; + tensor var_30888_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2941_cast_fp16)[name = tensor("op_30888_cast_fp16")]; + tensor var_30889_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2943_cast_fp16)[name = tensor("op_30889_cast_fp16")]; + tensor var_30890_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2945_cast_fp16)[name = tensor("op_30890_cast_fp16")]; + tensor var_30891_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2947_cast_fp16)[name = tensor("op_30891_cast_fp16")]; + tensor var_30892_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2949_cast_fp16)[name = tensor("op_30892_cast_fp16")]; + tensor var_30893_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2951_cast_fp16)[name = tensor("op_30893_cast_fp16")]; + tensor var_30894_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2953_cast_fp16)[name = tensor("op_30894_cast_fp16")]; + tensor var_30895_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2955_cast_fp16)[name = tensor("op_30895_cast_fp16")]; + tensor var_30896_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2957_cast_fp16)[name = tensor("op_30896_cast_fp16")]; + tensor var_30897_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2959_cast_fp16)[name = tensor("op_30897_cast_fp16")]; + tensor var_30898_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2961_cast_fp16)[name = tensor("op_30898_cast_fp16")]; + tensor var_30899_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2963_cast_fp16)[name = tensor("op_30899_cast_fp16")]; + tensor var_30900_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2965_cast_fp16)[name = tensor("op_30900_cast_fp16")]; + tensor var_30901_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2967_cast_fp16)[name = tensor("op_30901_cast_fp16")]; + tensor var_30902_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2969_cast_fp16)[name = tensor("op_30902_cast_fp16")]; + tensor var_30903_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2971_cast_fp16)[name = tensor("op_30903_cast_fp16")]; + tensor var_30904_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2973_cast_fp16)[name = tensor("op_30904_cast_fp16")]; + tensor var_30905_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2975_cast_fp16)[name = tensor("op_30905_cast_fp16")]; + tensor var_30906_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2977_cast_fp16)[name = tensor("op_30906_cast_fp16")]; + tensor var_30907_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2979_cast_fp16)[name = tensor("op_30907_cast_fp16")]; + tensor var_30908_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2981_cast_fp16)[name = tensor("op_30908_cast_fp16")]; + tensor var_30909_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2983_cast_fp16)[name = tensor("op_30909_cast_fp16")]; + tensor var_30910_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2985_cast_fp16)[name = tensor("op_30910_cast_fp16")]; + tensor var_30911_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2987_cast_fp16)[name = tensor("op_30911_cast_fp16")]; + tensor var_30912_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2989_cast_fp16)[name = tensor("op_30912_cast_fp16")]; + tensor var_30913_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2991_cast_fp16)[name = tensor("op_30913_cast_fp16")]; + tensor var_30914_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2993_cast_fp16)[name = tensor("op_30914_cast_fp16")]; + tensor var_30915_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2995_cast_fp16)[name = tensor("op_30915_cast_fp16")]; + tensor var_30916_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2997_cast_fp16)[name = tensor("op_30916_cast_fp16")]; + tensor var_30917_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_2999_cast_fp16)[name = tensor("op_30917_cast_fp16")]; + tensor var_30918_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3001_cast_fp16)[name = tensor("op_30918_cast_fp16")]; + tensor var_30919_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3003_cast_fp16)[name = tensor("op_30919_cast_fp16")]; + tensor var_30920_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3005_cast_fp16)[name = tensor("op_30920_cast_fp16")]; + tensor var_30921_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3007_cast_fp16)[name = tensor("op_30921_cast_fp16")]; + tensor var_30922_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3009_cast_fp16)[name = tensor("op_30922_cast_fp16")]; + tensor var_30923_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3011_cast_fp16)[name = tensor("op_30923_cast_fp16")]; + tensor var_30924_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3013_cast_fp16)[name = tensor("op_30924_cast_fp16")]; + tensor var_30925_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3015_cast_fp16)[name = tensor("op_30925_cast_fp16")]; + tensor var_30926_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3017_cast_fp16)[name = tensor("op_30926_cast_fp16")]; + tensor var_30927_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3019_cast_fp16)[name = tensor("op_30927_cast_fp16")]; + tensor var_30928_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3021_cast_fp16)[name = tensor("op_30928_cast_fp16")]; + tensor var_30929_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3023_cast_fp16)[name = tensor("op_30929_cast_fp16")]; + tensor var_30930_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3025_cast_fp16)[name = tensor("op_30930_cast_fp16")]; + tensor var_30931_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3027_cast_fp16)[name = tensor("op_30931_cast_fp16")]; + tensor var_30932_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3029_cast_fp16)[name = tensor("op_30932_cast_fp16")]; + tensor var_30933_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3031_cast_fp16)[name = tensor("op_30933_cast_fp16")]; + tensor var_30934_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3033_cast_fp16)[name = tensor("op_30934_cast_fp16")]; + tensor var_30935_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3035_cast_fp16)[name = tensor("op_30935_cast_fp16")]; + tensor var_30936_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3037_cast_fp16)[name = tensor("op_30936_cast_fp16")]; + tensor var_30937_cast_fp16 = softmax(axis = var_29629, x = aw_chunk_3039_cast_fp16)[name = tensor("op_30937_cast_fp16")]; + tensor var_30939_equation_0 = const()[name = tensor("op_30939_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30939_cast_fp16 = einsum(equation = var_30939_equation_0, values = (var_30459_cast_fp16, var_30858_cast_fp16))[name = tensor("op_30939_cast_fp16")]; + tensor var_30941_equation_0 = const()[name = tensor("op_30941_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30941_cast_fp16 = einsum(equation = var_30941_equation_0, values = (var_30459_cast_fp16, var_30859_cast_fp16))[name = tensor("op_30941_cast_fp16")]; + tensor var_30943_equation_0 = const()[name = tensor("op_30943_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30943_cast_fp16 = einsum(equation = var_30943_equation_0, values = (var_30459_cast_fp16, var_30860_cast_fp16))[name = tensor("op_30943_cast_fp16")]; + tensor var_30945_equation_0 = const()[name = tensor("op_30945_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30945_cast_fp16 = einsum(equation = var_30945_equation_0, values = (var_30459_cast_fp16, var_30861_cast_fp16))[name = tensor("op_30945_cast_fp16")]; + tensor var_30947_equation_0 = const()[name = tensor("op_30947_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30947_cast_fp16 = einsum(equation = var_30947_equation_0, values = (var_30463_cast_fp16, var_30862_cast_fp16))[name = tensor("op_30947_cast_fp16")]; + tensor var_30949_equation_0 = const()[name = tensor("op_30949_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30949_cast_fp16 = einsum(equation = var_30949_equation_0, values = (var_30463_cast_fp16, var_30863_cast_fp16))[name = tensor("op_30949_cast_fp16")]; + tensor var_30951_equation_0 = const()[name = tensor("op_30951_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30951_cast_fp16 = einsum(equation = var_30951_equation_0, values = (var_30463_cast_fp16, var_30864_cast_fp16))[name = tensor("op_30951_cast_fp16")]; + tensor var_30953_equation_0 = const()[name = tensor("op_30953_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30953_cast_fp16 = einsum(equation = var_30953_equation_0, values = (var_30463_cast_fp16, var_30865_cast_fp16))[name = tensor("op_30953_cast_fp16")]; + tensor var_30955_equation_0 = const()[name = tensor("op_30955_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30955_cast_fp16 = einsum(equation = var_30955_equation_0, values = (var_30467_cast_fp16, var_30866_cast_fp16))[name = tensor("op_30955_cast_fp16")]; + tensor var_30957_equation_0 = const()[name = tensor("op_30957_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30957_cast_fp16 = einsum(equation = var_30957_equation_0, values = (var_30467_cast_fp16, var_30867_cast_fp16))[name = tensor("op_30957_cast_fp16")]; + tensor var_30959_equation_0 = const()[name = tensor("op_30959_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30959_cast_fp16 = einsum(equation = var_30959_equation_0, values = (var_30467_cast_fp16, var_30868_cast_fp16))[name = tensor("op_30959_cast_fp16")]; + tensor var_30961_equation_0 = const()[name = tensor("op_30961_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30961_cast_fp16 = einsum(equation = var_30961_equation_0, values = (var_30467_cast_fp16, var_30869_cast_fp16))[name = tensor("op_30961_cast_fp16")]; + tensor var_30963_equation_0 = const()[name = tensor("op_30963_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30963_cast_fp16 = einsum(equation = var_30963_equation_0, values = (var_30471_cast_fp16, var_30870_cast_fp16))[name = tensor("op_30963_cast_fp16")]; + tensor var_30965_equation_0 = const()[name = tensor("op_30965_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30965_cast_fp16 = einsum(equation = var_30965_equation_0, values = (var_30471_cast_fp16, var_30871_cast_fp16))[name = tensor("op_30965_cast_fp16")]; + tensor var_30967_equation_0 = const()[name = tensor("op_30967_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30967_cast_fp16 = einsum(equation = var_30967_equation_0, values = (var_30471_cast_fp16, var_30872_cast_fp16))[name = tensor("op_30967_cast_fp16")]; + tensor var_30969_equation_0 = const()[name = tensor("op_30969_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30969_cast_fp16 = einsum(equation = var_30969_equation_0, values = (var_30471_cast_fp16, var_30873_cast_fp16))[name = tensor("op_30969_cast_fp16")]; + tensor var_30971_equation_0 = const()[name = tensor("op_30971_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30971_cast_fp16 = einsum(equation = var_30971_equation_0, values = (var_30475_cast_fp16, var_30874_cast_fp16))[name = tensor("op_30971_cast_fp16")]; + tensor var_30973_equation_0 = const()[name = tensor("op_30973_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30973_cast_fp16 = einsum(equation = var_30973_equation_0, values = (var_30475_cast_fp16, var_30875_cast_fp16))[name = tensor("op_30973_cast_fp16")]; + tensor var_30975_equation_0 = const()[name = tensor("op_30975_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30975_cast_fp16 = einsum(equation = var_30975_equation_0, values = (var_30475_cast_fp16, var_30876_cast_fp16))[name = tensor("op_30975_cast_fp16")]; + tensor var_30977_equation_0 = const()[name = tensor("op_30977_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30977_cast_fp16 = einsum(equation = var_30977_equation_0, values = (var_30475_cast_fp16, var_30877_cast_fp16))[name = tensor("op_30977_cast_fp16")]; + tensor var_30979_equation_0 = const()[name = tensor("op_30979_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30979_cast_fp16 = einsum(equation = var_30979_equation_0, values = (var_30479_cast_fp16, var_30878_cast_fp16))[name = tensor("op_30979_cast_fp16")]; + tensor var_30981_equation_0 = const()[name = tensor("op_30981_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30981_cast_fp16 = einsum(equation = var_30981_equation_0, values = (var_30479_cast_fp16, var_30879_cast_fp16))[name = tensor("op_30981_cast_fp16")]; + tensor var_30983_equation_0 = const()[name = tensor("op_30983_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30983_cast_fp16 = einsum(equation = var_30983_equation_0, values = (var_30479_cast_fp16, var_30880_cast_fp16))[name = tensor("op_30983_cast_fp16")]; + tensor var_30985_equation_0 = const()[name = tensor("op_30985_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30985_cast_fp16 = einsum(equation = var_30985_equation_0, values = (var_30479_cast_fp16, var_30881_cast_fp16))[name = tensor("op_30985_cast_fp16")]; + tensor var_30987_equation_0 = const()[name = tensor("op_30987_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30987_cast_fp16 = einsum(equation = var_30987_equation_0, values = (var_30483_cast_fp16, var_30882_cast_fp16))[name = tensor("op_30987_cast_fp16")]; + tensor var_30989_equation_0 = const()[name = tensor("op_30989_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30989_cast_fp16 = einsum(equation = var_30989_equation_0, values = (var_30483_cast_fp16, var_30883_cast_fp16))[name = tensor("op_30989_cast_fp16")]; + tensor var_30991_equation_0 = const()[name = tensor("op_30991_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30991_cast_fp16 = einsum(equation = var_30991_equation_0, values = (var_30483_cast_fp16, var_30884_cast_fp16))[name = tensor("op_30991_cast_fp16")]; + tensor var_30993_equation_0 = const()[name = tensor("op_30993_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30993_cast_fp16 = einsum(equation = var_30993_equation_0, values = (var_30483_cast_fp16, var_30885_cast_fp16))[name = tensor("op_30993_cast_fp16")]; + tensor var_30995_equation_0 = const()[name = tensor("op_30995_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30995_cast_fp16 = einsum(equation = var_30995_equation_0, values = (var_30487_cast_fp16, var_30886_cast_fp16))[name = tensor("op_30995_cast_fp16")]; + tensor var_30997_equation_0 = const()[name = tensor("op_30997_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30997_cast_fp16 = einsum(equation = var_30997_equation_0, values = (var_30487_cast_fp16, var_30887_cast_fp16))[name = tensor("op_30997_cast_fp16")]; + tensor var_30999_equation_0 = const()[name = tensor("op_30999_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_30999_cast_fp16 = einsum(equation = var_30999_equation_0, values = (var_30487_cast_fp16, var_30888_cast_fp16))[name = tensor("op_30999_cast_fp16")]; + tensor var_31001_equation_0 = const()[name = tensor("op_31001_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31001_cast_fp16 = einsum(equation = var_31001_equation_0, values = (var_30487_cast_fp16, var_30889_cast_fp16))[name = tensor("op_31001_cast_fp16")]; + tensor var_31003_equation_0 = const()[name = tensor("op_31003_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31003_cast_fp16 = einsum(equation = var_31003_equation_0, values = (var_30491_cast_fp16, var_30890_cast_fp16))[name = tensor("op_31003_cast_fp16")]; + tensor var_31005_equation_0 = const()[name = tensor("op_31005_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31005_cast_fp16 = einsum(equation = var_31005_equation_0, values = (var_30491_cast_fp16, var_30891_cast_fp16))[name = tensor("op_31005_cast_fp16")]; + tensor var_31007_equation_0 = const()[name = tensor("op_31007_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31007_cast_fp16 = einsum(equation = var_31007_equation_0, values = (var_30491_cast_fp16, var_30892_cast_fp16))[name = tensor("op_31007_cast_fp16")]; + tensor var_31009_equation_0 = const()[name = tensor("op_31009_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31009_cast_fp16 = einsum(equation = var_31009_equation_0, values = (var_30491_cast_fp16, var_30893_cast_fp16))[name = tensor("op_31009_cast_fp16")]; + tensor var_31011_equation_0 = const()[name = tensor("op_31011_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31011_cast_fp16 = einsum(equation = var_31011_equation_0, values = (var_30495_cast_fp16, var_30894_cast_fp16))[name = tensor("op_31011_cast_fp16")]; + tensor var_31013_equation_0 = const()[name = tensor("op_31013_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31013_cast_fp16 = einsum(equation = var_31013_equation_0, values = (var_30495_cast_fp16, var_30895_cast_fp16))[name = tensor("op_31013_cast_fp16")]; + tensor var_31015_equation_0 = const()[name = tensor("op_31015_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31015_cast_fp16 = einsum(equation = var_31015_equation_0, values = (var_30495_cast_fp16, var_30896_cast_fp16))[name = tensor("op_31015_cast_fp16")]; + tensor var_31017_equation_0 = const()[name = tensor("op_31017_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31017_cast_fp16 = einsum(equation = var_31017_equation_0, values = (var_30495_cast_fp16, var_30897_cast_fp16))[name = tensor("op_31017_cast_fp16")]; + tensor var_31019_equation_0 = const()[name = tensor("op_31019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31019_cast_fp16 = einsum(equation = var_31019_equation_0, values = (var_30499_cast_fp16, var_30898_cast_fp16))[name = tensor("op_31019_cast_fp16")]; + tensor var_31021_equation_0 = const()[name = tensor("op_31021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31021_cast_fp16 = einsum(equation = var_31021_equation_0, values = (var_30499_cast_fp16, var_30899_cast_fp16))[name = tensor("op_31021_cast_fp16")]; + tensor var_31023_equation_0 = const()[name = tensor("op_31023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31023_cast_fp16 = einsum(equation = var_31023_equation_0, values = (var_30499_cast_fp16, var_30900_cast_fp16))[name = tensor("op_31023_cast_fp16")]; + tensor var_31025_equation_0 = const()[name = tensor("op_31025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31025_cast_fp16 = einsum(equation = var_31025_equation_0, values = (var_30499_cast_fp16, var_30901_cast_fp16))[name = tensor("op_31025_cast_fp16")]; + tensor var_31027_equation_0 = const()[name = tensor("op_31027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31027_cast_fp16 = einsum(equation = var_31027_equation_0, values = (var_30503_cast_fp16, var_30902_cast_fp16))[name = tensor("op_31027_cast_fp16")]; + tensor var_31029_equation_0 = const()[name = tensor("op_31029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31029_cast_fp16 = einsum(equation = var_31029_equation_0, values = (var_30503_cast_fp16, var_30903_cast_fp16))[name = tensor("op_31029_cast_fp16")]; + tensor var_31031_equation_0 = const()[name = tensor("op_31031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31031_cast_fp16 = einsum(equation = var_31031_equation_0, values = (var_30503_cast_fp16, var_30904_cast_fp16))[name = tensor("op_31031_cast_fp16")]; + tensor var_31033_equation_0 = const()[name = tensor("op_31033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31033_cast_fp16 = einsum(equation = var_31033_equation_0, values = (var_30503_cast_fp16, var_30905_cast_fp16))[name = tensor("op_31033_cast_fp16")]; + tensor var_31035_equation_0 = const()[name = tensor("op_31035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31035_cast_fp16 = einsum(equation = var_31035_equation_0, values = (var_30507_cast_fp16, var_30906_cast_fp16))[name = tensor("op_31035_cast_fp16")]; + tensor var_31037_equation_0 = const()[name = tensor("op_31037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31037_cast_fp16 = einsum(equation = var_31037_equation_0, values = (var_30507_cast_fp16, var_30907_cast_fp16))[name = tensor("op_31037_cast_fp16")]; + tensor var_31039_equation_0 = const()[name = tensor("op_31039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31039_cast_fp16 = einsum(equation = var_31039_equation_0, values = (var_30507_cast_fp16, var_30908_cast_fp16))[name = tensor("op_31039_cast_fp16")]; + tensor var_31041_equation_0 = const()[name = tensor("op_31041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31041_cast_fp16 = einsum(equation = var_31041_equation_0, values = (var_30507_cast_fp16, var_30909_cast_fp16))[name = tensor("op_31041_cast_fp16")]; + tensor var_31043_equation_0 = const()[name = tensor("op_31043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31043_cast_fp16 = einsum(equation = var_31043_equation_0, values = (var_30511_cast_fp16, var_30910_cast_fp16))[name = tensor("op_31043_cast_fp16")]; + tensor var_31045_equation_0 = const()[name = tensor("op_31045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31045_cast_fp16 = einsum(equation = var_31045_equation_0, values = (var_30511_cast_fp16, var_30911_cast_fp16))[name = tensor("op_31045_cast_fp16")]; + tensor var_31047_equation_0 = const()[name = tensor("op_31047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31047_cast_fp16 = einsum(equation = var_31047_equation_0, values = (var_30511_cast_fp16, var_30912_cast_fp16))[name = tensor("op_31047_cast_fp16")]; + tensor var_31049_equation_0 = const()[name = tensor("op_31049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31049_cast_fp16 = einsum(equation = var_31049_equation_0, values = (var_30511_cast_fp16, var_30913_cast_fp16))[name = tensor("op_31049_cast_fp16")]; + tensor var_31051_equation_0 = const()[name = tensor("op_31051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31051_cast_fp16 = einsum(equation = var_31051_equation_0, values = (var_30515_cast_fp16, var_30914_cast_fp16))[name = tensor("op_31051_cast_fp16")]; + tensor var_31053_equation_0 = const()[name = tensor("op_31053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31053_cast_fp16 = einsum(equation = var_31053_equation_0, values = (var_30515_cast_fp16, var_30915_cast_fp16))[name = tensor("op_31053_cast_fp16")]; + tensor var_31055_equation_0 = const()[name = tensor("op_31055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31055_cast_fp16 = einsum(equation = var_31055_equation_0, values = (var_30515_cast_fp16, var_30916_cast_fp16))[name = tensor("op_31055_cast_fp16")]; + tensor var_31057_equation_0 = const()[name = tensor("op_31057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31057_cast_fp16 = einsum(equation = var_31057_equation_0, values = (var_30515_cast_fp16, var_30917_cast_fp16))[name = tensor("op_31057_cast_fp16")]; + tensor var_31059_equation_0 = const()[name = tensor("op_31059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31059_cast_fp16 = einsum(equation = var_31059_equation_0, values = (var_30519_cast_fp16, var_30918_cast_fp16))[name = tensor("op_31059_cast_fp16")]; + tensor var_31061_equation_0 = const()[name = tensor("op_31061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31061_cast_fp16 = einsum(equation = var_31061_equation_0, values = (var_30519_cast_fp16, var_30919_cast_fp16))[name = tensor("op_31061_cast_fp16")]; + tensor var_31063_equation_0 = const()[name = tensor("op_31063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31063_cast_fp16 = einsum(equation = var_31063_equation_0, values = (var_30519_cast_fp16, var_30920_cast_fp16))[name = tensor("op_31063_cast_fp16")]; + tensor var_31065_equation_0 = const()[name = tensor("op_31065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31065_cast_fp16 = einsum(equation = var_31065_equation_0, values = (var_30519_cast_fp16, var_30921_cast_fp16))[name = tensor("op_31065_cast_fp16")]; + tensor var_31067_equation_0 = const()[name = tensor("op_31067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31067_cast_fp16 = einsum(equation = var_31067_equation_0, values = (var_30523_cast_fp16, var_30922_cast_fp16))[name = tensor("op_31067_cast_fp16")]; + tensor var_31069_equation_0 = const()[name = tensor("op_31069_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31069_cast_fp16 = einsum(equation = var_31069_equation_0, values = (var_30523_cast_fp16, var_30923_cast_fp16))[name = tensor("op_31069_cast_fp16")]; + tensor var_31071_equation_0 = const()[name = tensor("op_31071_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31071_cast_fp16 = einsum(equation = var_31071_equation_0, values = (var_30523_cast_fp16, var_30924_cast_fp16))[name = tensor("op_31071_cast_fp16")]; + tensor var_31073_equation_0 = const()[name = tensor("op_31073_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31073_cast_fp16 = einsum(equation = var_31073_equation_0, values = (var_30523_cast_fp16, var_30925_cast_fp16))[name = tensor("op_31073_cast_fp16")]; + tensor var_31075_equation_0 = const()[name = tensor("op_31075_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31075_cast_fp16 = einsum(equation = var_31075_equation_0, values = (var_30527_cast_fp16, var_30926_cast_fp16))[name = tensor("op_31075_cast_fp16")]; + tensor var_31077_equation_0 = const()[name = tensor("op_31077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31077_cast_fp16 = einsum(equation = var_31077_equation_0, values = (var_30527_cast_fp16, var_30927_cast_fp16))[name = tensor("op_31077_cast_fp16")]; + tensor var_31079_equation_0 = const()[name = tensor("op_31079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31079_cast_fp16 = einsum(equation = var_31079_equation_0, values = (var_30527_cast_fp16, var_30928_cast_fp16))[name = tensor("op_31079_cast_fp16")]; + tensor var_31081_equation_0 = const()[name = tensor("op_31081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31081_cast_fp16 = einsum(equation = var_31081_equation_0, values = (var_30527_cast_fp16, var_30929_cast_fp16))[name = tensor("op_31081_cast_fp16")]; + tensor var_31083_equation_0 = const()[name = tensor("op_31083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31083_cast_fp16 = einsum(equation = var_31083_equation_0, values = (var_30531_cast_fp16, var_30930_cast_fp16))[name = tensor("op_31083_cast_fp16")]; + tensor var_31085_equation_0 = const()[name = tensor("op_31085_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31085_cast_fp16 = einsum(equation = var_31085_equation_0, values = (var_30531_cast_fp16, var_30931_cast_fp16))[name = tensor("op_31085_cast_fp16")]; + tensor var_31087_equation_0 = const()[name = tensor("op_31087_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31087_cast_fp16 = einsum(equation = var_31087_equation_0, values = (var_30531_cast_fp16, var_30932_cast_fp16))[name = tensor("op_31087_cast_fp16")]; + tensor var_31089_equation_0 = const()[name = tensor("op_31089_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31089_cast_fp16 = einsum(equation = var_31089_equation_0, values = (var_30531_cast_fp16, var_30933_cast_fp16))[name = tensor("op_31089_cast_fp16")]; + tensor var_31091_equation_0 = const()[name = tensor("op_31091_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31091_cast_fp16 = einsum(equation = var_31091_equation_0, values = (var_30535_cast_fp16, var_30934_cast_fp16))[name = tensor("op_31091_cast_fp16")]; + tensor var_31093_equation_0 = const()[name = tensor("op_31093_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31093_cast_fp16 = einsum(equation = var_31093_equation_0, values = (var_30535_cast_fp16, var_30935_cast_fp16))[name = tensor("op_31093_cast_fp16")]; + tensor var_31095_equation_0 = const()[name = tensor("op_31095_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31095_cast_fp16 = einsum(equation = var_31095_equation_0, values = (var_30535_cast_fp16, var_30936_cast_fp16))[name = tensor("op_31095_cast_fp16")]; + tensor var_31097_equation_0 = const()[name = tensor("op_31097_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_31097_cast_fp16 = einsum(equation = var_31097_equation_0, values = (var_30535_cast_fp16, var_30937_cast_fp16))[name = tensor("op_31097_cast_fp16")]; + tensor var_31099_interleave_0 = const()[name = tensor("op_31099_interleave_0"), val = tensor(false)]; + tensor var_31099_cast_fp16 = concat(axis = var_29604, interleave = var_31099_interleave_0, values = (var_30939_cast_fp16, var_30941_cast_fp16, var_30943_cast_fp16, var_30945_cast_fp16))[name = tensor("op_31099_cast_fp16")]; + tensor var_31101_interleave_0 = const()[name = tensor("op_31101_interleave_0"), val = tensor(false)]; + tensor var_31101_cast_fp16 = concat(axis = var_29604, interleave = var_31101_interleave_0, values = (var_30947_cast_fp16, var_30949_cast_fp16, var_30951_cast_fp16, var_30953_cast_fp16))[name = tensor("op_31101_cast_fp16")]; + tensor var_31103_interleave_0 = const()[name = tensor("op_31103_interleave_0"), val = tensor(false)]; + tensor var_31103_cast_fp16 = concat(axis = var_29604, interleave = var_31103_interleave_0, values = (var_30955_cast_fp16, var_30957_cast_fp16, var_30959_cast_fp16, var_30961_cast_fp16))[name = tensor("op_31103_cast_fp16")]; + tensor var_31105_interleave_0 = const()[name = tensor("op_31105_interleave_0"), val = tensor(false)]; + tensor var_31105_cast_fp16 = concat(axis = var_29604, interleave = var_31105_interleave_0, values = (var_30963_cast_fp16, var_30965_cast_fp16, var_30967_cast_fp16, var_30969_cast_fp16))[name = tensor("op_31105_cast_fp16")]; + tensor var_31107_interleave_0 = const()[name = tensor("op_31107_interleave_0"), val = tensor(false)]; + tensor var_31107_cast_fp16 = concat(axis = var_29604, interleave = var_31107_interleave_0, values = (var_30971_cast_fp16, var_30973_cast_fp16, var_30975_cast_fp16, var_30977_cast_fp16))[name = tensor("op_31107_cast_fp16")]; + tensor var_31109_interleave_0 = const()[name = tensor("op_31109_interleave_0"), val = tensor(false)]; + tensor var_31109_cast_fp16 = concat(axis = var_29604, interleave = var_31109_interleave_0, values = (var_30979_cast_fp16, var_30981_cast_fp16, var_30983_cast_fp16, var_30985_cast_fp16))[name = tensor("op_31109_cast_fp16")]; + tensor var_31111_interleave_0 = const()[name = tensor("op_31111_interleave_0"), val = tensor(false)]; + tensor var_31111_cast_fp16 = concat(axis = var_29604, interleave = var_31111_interleave_0, values = (var_30987_cast_fp16, var_30989_cast_fp16, var_30991_cast_fp16, var_30993_cast_fp16))[name = tensor("op_31111_cast_fp16")]; + tensor var_31113_interleave_0 = const()[name = tensor("op_31113_interleave_0"), val = tensor(false)]; + tensor var_31113_cast_fp16 = concat(axis = var_29604, interleave = var_31113_interleave_0, values = (var_30995_cast_fp16, var_30997_cast_fp16, var_30999_cast_fp16, var_31001_cast_fp16))[name = tensor("op_31113_cast_fp16")]; + tensor var_31115_interleave_0 = const()[name = tensor("op_31115_interleave_0"), val = tensor(false)]; + tensor var_31115_cast_fp16 = concat(axis = var_29604, interleave = var_31115_interleave_0, values = (var_31003_cast_fp16, var_31005_cast_fp16, var_31007_cast_fp16, var_31009_cast_fp16))[name = tensor("op_31115_cast_fp16")]; + tensor var_31117_interleave_0 = const()[name = tensor("op_31117_interleave_0"), val = tensor(false)]; + tensor var_31117_cast_fp16 = concat(axis = var_29604, interleave = var_31117_interleave_0, values = (var_31011_cast_fp16, var_31013_cast_fp16, var_31015_cast_fp16, var_31017_cast_fp16))[name = tensor("op_31117_cast_fp16")]; + tensor var_31119_interleave_0 = const()[name = tensor("op_31119_interleave_0"), val = tensor(false)]; + tensor var_31119_cast_fp16 = concat(axis = var_29604, interleave = var_31119_interleave_0, values = (var_31019_cast_fp16, var_31021_cast_fp16, var_31023_cast_fp16, var_31025_cast_fp16))[name = tensor("op_31119_cast_fp16")]; + tensor var_31121_interleave_0 = const()[name = tensor("op_31121_interleave_0"), val = tensor(false)]; + tensor var_31121_cast_fp16 = concat(axis = var_29604, interleave = var_31121_interleave_0, values = (var_31027_cast_fp16, var_31029_cast_fp16, var_31031_cast_fp16, var_31033_cast_fp16))[name = tensor("op_31121_cast_fp16")]; + tensor var_31123_interleave_0 = const()[name = tensor("op_31123_interleave_0"), val = tensor(false)]; + tensor var_31123_cast_fp16 = concat(axis = var_29604, interleave = var_31123_interleave_0, values = (var_31035_cast_fp16, var_31037_cast_fp16, var_31039_cast_fp16, var_31041_cast_fp16))[name = tensor("op_31123_cast_fp16")]; + tensor var_31125_interleave_0 = const()[name = tensor("op_31125_interleave_0"), val = tensor(false)]; + tensor var_31125_cast_fp16 = concat(axis = var_29604, interleave = var_31125_interleave_0, values = (var_31043_cast_fp16, var_31045_cast_fp16, var_31047_cast_fp16, var_31049_cast_fp16))[name = tensor("op_31125_cast_fp16")]; + tensor var_31127_interleave_0 = const()[name = tensor("op_31127_interleave_0"), val = tensor(false)]; + tensor var_31127_cast_fp16 = concat(axis = var_29604, interleave = var_31127_interleave_0, values = (var_31051_cast_fp16, var_31053_cast_fp16, var_31055_cast_fp16, var_31057_cast_fp16))[name = tensor("op_31127_cast_fp16")]; + tensor var_31129_interleave_0 = const()[name = tensor("op_31129_interleave_0"), val = tensor(false)]; + tensor var_31129_cast_fp16 = concat(axis = var_29604, interleave = var_31129_interleave_0, values = (var_31059_cast_fp16, var_31061_cast_fp16, var_31063_cast_fp16, var_31065_cast_fp16))[name = tensor("op_31129_cast_fp16")]; + tensor var_31131_interleave_0 = const()[name = tensor("op_31131_interleave_0"), val = tensor(false)]; + tensor var_31131_cast_fp16 = concat(axis = var_29604, interleave = var_31131_interleave_0, values = (var_31067_cast_fp16, var_31069_cast_fp16, var_31071_cast_fp16, var_31073_cast_fp16))[name = tensor("op_31131_cast_fp16")]; + tensor var_31133_interleave_0 = const()[name = tensor("op_31133_interleave_0"), val = tensor(false)]; + tensor var_31133_cast_fp16 = concat(axis = var_29604, interleave = var_31133_interleave_0, values = (var_31075_cast_fp16, var_31077_cast_fp16, var_31079_cast_fp16, var_31081_cast_fp16))[name = tensor("op_31133_cast_fp16")]; + tensor var_31135_interleave_0 = const()[name = tensor("op_31135_interleave_0"), val = tensor(false)]; + tensor var_31135_cast_fp16 = concat(axis = var_29604, interleave = var_31135_interleave_0, values = (var_31083_cast_fp16, var_31085_cast_fp16, var_31087_cast_fp16, var_31089_cast_fp16))[name = tensor("op_31135_cast_fp16")]; + tensor var_31137_interleave_0 = const()[name = tensor("op_31137_interleave_0"), val = tensor(false)]; + tensor var_31137_cast_fp16 = concat(axis = var_29604, interleave = var_31137_interleave_0, values = (var_31091_cast_fp16, var_31093_cast_fp16, var_31095_cast_fp16, var_31097_cast_fp16))[name = tensor("op_31137_cast_fp16")]; + tensor input_367_interleave_0 = const()[name = tensor("input_367_interleave_0"), val = tensor(false)]; + tensor input_367_cast_fp16 = concat(axis = var_29629, interleave = input_367_interleave_0, values = (var_31099_cast_fp16, var_31101_cast_fp16, var_31103_cast_fp16, var_31105_cast_fp16, var_31107_cast_fp16, var_31109_cast_fp16, var_31111_cast_fp16, var_31113_cast_fp16, var_31115_cast_fp16, var_31117_cast_fp16, var_31119_cast_fp16, var_31121_cast_fp16, var_31123_cast_fp16, var_31125_cast_fp16, var_31127_cast_fp16, var_31129_cast_fp16, var_31131_cast_fp16, var_31133_cast_fp16, var_31135_cast_fp16, var_31137_cast_fp16))[name = tensor("input_367_cast_fp16")]; + tensor var_31145 = const()[name = tensor("op_31145"), val = tensor([1, 1])]; + tensor var_31147 = const()[name = tensor("op_31147"), val = tensor([1, 1])]; + tensor pretrained_out_223_pad_type_0 = const()[name = tensor("pretrained_out_223_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_223_pad_0 = const()[name = tensor("pretrained_out_223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207836800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208656064))), name = tensor("layers_18_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_18_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_18_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208656192)))]; + tensor pretrained_out_223_cast_fp16 = conv(bias = layers_18_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_31147, groups = var_29629, pad = pretrained_out_223_pad_0, pad_type = pretrained_out_223_pad_type_0, strides = var_31145, weight = layers_18_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_367_cast_fp16)[name = tensor("pretrained_out_223_cast_fp16")]; + tensor var_31151 = const()[name = tensor("op_31151"), val = tensor([1, 1])]; + tensor var_31153 = const()[name = tensor("op_31153"), val = tensor([1, 1])]; + tensor input_369_pad_type_0 = const()[name = tensor("input_369_pad_type_0"), val = tensor("custom")]; + tensor input_369_pad_0 = const()[name = tensor("input_369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_18_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208658816)))]; + tensor input_369_cast_fp16 = conv(dilations = var_31153, groups = var_29629, pad = input_369_pad_0, pad_type = input_369_pad_type_0, strides = var_31151, weight = layers_18_self_attn_o_proj_loraA_weight_to_fp16, x = input_367_cast_fp16)[name = tensor("input_369_cast_fp16")]; + tensor var_31157 = const()[name = tensor("op_31157"), val = tensor([1, 1])]; + tensor var_31159 = const()[name = tensor("op_31159"), val = tensor([1, 1])]; + tensor lora_out_445_pad_type_0 = const()[name = tensor("lora_out_445_pad_type_0"), val = tensor("custom")]; + tensor lora_out_445_pad_0 = const()[name = tensor("lora_out_445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_447_weight_0_to_fp16 = const()[name = tensor("lora_out_447_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208699840)))]; + tensor lora_out_447_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_31159, groups = var_29629, pad = lora_out_445_pad_0, pad_type = lora_out_445_pad_type_0, strides = var_31157, weight = lora_out_447_weight_0_to_fp16, x = input_369_cast_fp16)[name = tensor("lora_out_447_cast_fp16")]; + tensor obj_75_cast_fp16 = add(x = pretrained_out_223_cast_fp16, y = lora_out_447_cast_fp16)[name = tensor("obj_75_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_75_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; + tensor var_31168 = const()[name = tensor("op_31168"), val = tensor([1])]; + tensor channels_mean_75_cast_fp16 = reduce_mean(axes = var_31168, keep_dims = var_29630, x = inputs_75_cast_fp16)[name = tensor("channels_mean_75_cast_fp16")]; + tensor zero_mean_75_cast_fp16 = sub(x = inputs_75_cast_fp16, y = channels_mean_75_cast_fp16)[name = tensor("zero_mean_75_cast_fp16")]; + tensor zero_mean_sq_75_cast_fp16 = mul(x = zero_mean_75_cast_fp16, y = zero_mean_75_cast_fp16)[name = tensor("zero_mean_sq_75_cast_fp16")]; + tensor var_31172 = const()[name = tensor("op_31172"), val = tensor([1])]; + tensor var_31173_cast_fp16 = reduce_mean(axes = var_31172, keep_dims = var_29630, x = zero_mean_sq_75_cast_fp16)[name = tensor("op_31173_cast_fp16")]; + tensor var_31174_to_fp16 = const()[name = tensor("op_31174_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_31175_cast_fp16 = add(x = var_31173_cast_fp16, y = var_31174_to_fp16)[name = tensor("op_31175_cast_fp16")]; + tensor denom_75_epsilon_0 = const()[name = tensor("denom_75_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_75_cast_fp16 = rsqrt(epsilon = denom_75_epsilon_0, x = var_31175_cast_fp16)[name = tensor("denom_75_cast_fp16")]; + tensor out_75_cast_fp16 = mul(x = zero_mean_75_cast_fp16, y = denom_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; + tensor input_371_gamma_0_to_fp16 = const()[name = tensor("input_371_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208740864)))]; + tensor input_371_beta_0_to_fp16 = const()[name = tensor("input_371_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208743488)))]; + tensor input_371_epsilon_0_to_fp16 = const()[name = tensor("input_371_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_371_cast_fp16 = batch_norm(beta = input_371_beta_0_to_fp16, epsilon = input_371_epsilon_0_to_fp16, gamma = input_371_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor("input_371_cast_fp16")]; + tensor var_31189 = const()[name = tensor("op_31189"), val = tensor([1, 1])]; + tensor var_31191 = const()[name = tensor("op_31191"), val = tensor([1, 1])]; + tensor pretrained_out_225_pad_type_0 = const()[name = tensor("pretrained_out_225_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_225_pad_0 = const()[name = tensor("pretrained_out_225_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208746112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212022976))), name = tensor("layers_18_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_18_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_18_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212023104)))]; + tensor pretrained_out_225_cast_fp16 = conv(bias = layers_18_fc1_pretrained_bias_to_fp16, dilations = var_31191, groups = var_29629, pad = pretrained_out_225_pad_0, pad_type = pretrained_out_225_pad_type_0, strides = var_31189, weight = layers_18_fc1_pretrained_weight_to_fp16_palettized, x = input_371_cast_fp16)[name = tensor("pretrained_out_225_cast_fp16")]; + tensor var_31195 = const()[name = tensor("op_31195"), val = tensor([1, 1])]; + tensor var_31197 = const()[name = tensor("op_31197"), val = tensor([1, 1])]; + tensor input_373_pad_type_0 = const()[name = tensor("input_373_pad_type_0"), val = tensor("custom")]; + tensor input_373_pad_0 = const()[name = tensor("input_373_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_18_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212033408)))]; + tensor input_373_cast_fp16 = conv(dilations = var_31197, groups = var_29629, pad = input_373_pad_0, pad_type = input_373_pad_type_0, strides = var_31195, weight = layers_18_fc1_loraA_weight_to_fp16, x = input_371_cast_fp16)[name = tensor("input_373_cast_fp16")]; + tensor var_31201 = const()[name = tensor("op_31201"), val = tensor([1, 1])]; + tensor var_31203 = const()[name = tensor("op_31203"), val = tensor([1, 1])]; + tensor lora_out_449_pad_type_0 = const()[name = tensor("lora_out_449_pad_type_0"), val = tensor("custom")]; + tensor lora_out_449_pad_0 = const()[name = tensor("lora_out_449_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_451_weight_0_to_fp16 = const()[name = tensor("lora_out_451_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212074432)))]; + tensor lora_out_451_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_31203, groups = var_29629, pad = lora_out_449_pad_0, pad_type = lora_out_449_pad_type_0, strides = var_31201, weight = lora_out_451_weight_0_to_fp16, x = input_373_cast_fp16)[name = tensor("lora_out_451_cast_fp16")]; + tensor input_375_cast_fp16 = add(x = pretrained_out_225_cast_fp16, y = lora_out_451_cast_fp16)[name = tensor("input_375_cast_fp16")]; + tensor input_377_mode_0 = const()[name = tensor("input_377_mode_0"), val = tensor("EXACT")]; + tensor input_377_cast_fp16 = gelu(mode = input_377_mode_0, x = input_375_cast_fp16)[name = tensor("input_377_cast_fp16")]; + tensor var_31215 = const()[name = tensor("op_31215"), val = tensor([1, 1])]; + tensor var_31217 = const()[name = tensor("op_31217"), val = tensor([1, 1])]; + tensor pretrained_out_227_pad_type_0 = const()[name = tensor("pretrained_out_227_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_227_pad_0 = const()[name = tensor("pretrained_out_227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212238336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215515200))), name = tensor("layers_18_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_18_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_18_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215515328)))]; + tensor pretrained_out_227_cast_fp16 = conv(bias = layers_18_fc2_pretrained_bias_to_fp16, dilations = var_31217, groups = var_29629, pad = pretrained_out_227_pad_0, pad_type = pretrained_out_227_pad_type_0, strides = var_31215, weight = layers_18_fc2_pretrained_weight_to_fp16_palettized, x = input_377_cast_fp16)[name = tensor("pretrained_out_227_cast_fp16")]; + tensor var_31221 = const()[name = tensor("op_31221"), val = tensor([1, 1])]; + tensor var_31223 = const()[name = tensor("op_31223"), val = tensor([1, 1])]; + tensor input_379_pad_type_0 = const()[name = tensor("input_379_pad_type_0"), val = tensor("custom")]; + tensor input_379_pad_0 = const()[name = tensor("input_379_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_18_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_18_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215517952)))]; + tensor input_379_cast_fp16 = conv(dilations = var_31223, groups = var_29629, pad = input_379_pad_0, pad_type = input_379_pad_type_0, strides = var_31221, weight = layers_18_fc2_loraA_weight_to_fp16, x = input_377_cast_fp16)[name = tensor("input_379_cast_fp16")]; + tensor var_31227 = const()[name = tensor("op_31227"), val = tensor([1, 1])]; + tensor var_31229 = const()[name = tensor("op_31229"), val = tensor([1, 1])]; + tensor lora_out_453_pad_type_0 = const()[name = tensor("lora_out_453_pad_type_0"), val = tensor("custom")]; + tensor lora_out_453_pad_0 = const()[name = tensor("lora_out_453_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_455_weight_0_to_fp16 = const()[name = tensor("lora_out_455_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215681856)))]; + tensor lora_out_455_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_31229, groups = var_29629, pad = lora_out_453_pad_0, pad_type = lora_out_453_pad_type_0, strides = var_31227, weight = lora_out_455_weight_0_to_fp16, x = input_379_cast_fp16)[name = tensor("lora_out_455_cast_fp16")]; + tensor hidden_states_41_cast_fp16 = add(x = pretrained_out_227_cast_fp16, y = lora_out_455_cast_fp16)[name = tensor("hidden_states_41_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = hidden_states_41_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; + tensor var_31239 = const()[name = tensor("op_31239"), val = tensor(3)]; + tensor var_31264 = const()[name = tensor("op_31264"), val = tensor(1)]; + tensor var_31265 = const()[name = tensor("op_31265"), val = tensor(true)]; + tensor var_31275 = const()[name = tensor("op_31275"), val = tensor([1])]; + tensor channels_mean_77_cast_fp16 = reduce_mean(axes = var_31275, keep_dims = var_31265, x = inputs_77_cast_fp16)[name = tensor("channels_mean_77_cast_fp16")]; + tensor zero_mean_77_cast_fp16 = sub(x = inputs_77_cast_fp16, y = channels_mean_77_cast_fp16)[name = tensor("zero_mean_77_cast_fp16")]; + tensor zero_mean_sq_77_cast_fp16 = mul(x = zero_mean_77_cast_fp16, y = zero_mean_77_cast_fp16)[name = tensor("zero_mean_sq_77_cast_fp16")]; + tensor var_31279 = const()[name = tensor("op_31279"), val = tensor([1])]; + tensor var_31280_cast_fp16 = reduce_mean(axes = var_31279, keep_dims = var_31265, x = zero_mean_sq_77_cast_fp16)[name = tensor("op_31280_cast_fp16")]; + tensor var_31281_to_fp16 = const()[name = tensor("op_31281_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_31282_cast_fp16 = add(x = var_31280_cast_fp16, y = var_31281_to_fp16)[name = tensor("op_31282_cast_fp16")]; + tensor denom_77_epsilon_0 = const()[name = tensor("denom_77_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_77_cast_fp16 = rsqrt(epsilon = denom_77_epsilon_0, x = var_31282_cast_fp16)[name = tensor("denom_77_cast_fp16")]; + tensor out_77_cast_fp16 = mul(x = zero_mean_77_cast_fp16, y = denom_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; + tensor obj_77_gamma_0_to_fp16 = const()[name = tensor("obj_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215722880)))]; + tensor obj_77_beta_0_to_fp16 = const()[name = tensor("obj_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215725504)))]; + tensor obj_77_epsilon_0_to_fp16 = const()[name = tensor("obj_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_77_cast_fp16 = batch_norm(beta = obj_77_beta_0_to_fp16, epsilon = obj_77_epsilon_0_to_fp16, gamma = obj_77_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor("obj_77_cast_fp16")]; + tensor var_31300 = const()[name = tensor("op_31300"), val = tensor([1, 1])]; + tensor var_31302 = const()[name = tensor("op_31302"), val = tensor([1, 1])]; + tensor pretrained_out_229_pad_type_0 = const()[name = tensor("pretrained_out_229_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_229_pad_0 = const()[name = tensor("pretrained_out_229_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215728128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216547392))), name = tensor("layers_19_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_19_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216547520)))]; + tensor pretrained_out_229_cast_fp16 = conv(bias = layers_19_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_31302, groups = var_31264, pad = pretrained_out_229_pad_0, pad_type = pretrained_out_229_pad_type_0, strides = var_31300, weight = layers_19_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_77_cast_fp16)[name = tensor("pretrained_out_229_cast_fp16")]; + tensor var_31306 = const()[name = tensor("op_31306"), val = tensor([1, 1])]; + tensor var_31308 = const()[name = tensor("op_31308"), val = tensor([1, 1])]; + tensor input_381_pad_type_0 = const()[name = tensor("input_381_pad_type_0"), val = tensor("custom")]; + tensor input_381_pad_0 = const()[name = tensor("input_381_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216550144)))]; + tensor input_381_cast_fp16 = conv(dilations = var_31308, groups = var_31264, pad = input_381_pad_0, pad_type = input_381_pad_type_0, strides = var_31306, weight = layers_19_self_attn_q_proj_loraA_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor("input_381_cast_fp16")]; + tensor var_31312 = const()[name = tensor("op_31312"), val = tensor([1, 1])]; + tensor var_31314 = const()[name = tensor("op_31314"), val = tensor([1, 1])]; + tensor lora_out_457_pad_type_0 = const()[name = tensor("lora_out_457_pad_type_0"), val = tensor("custom")]; + tensor lora_out_457_pad_0 = const()[name = tensor("lora_out_457_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_459_weight_0_to_fp16 = const()[name = tensor("lora_out_459_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216591168)))]; + tensor lora_out_459_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_31314, groups = var_31264, pad = lora_out_457_pad_0, pad_type = lora_out_457_pad_type_0, strides = var_31312, weight = lora_out_459_weight_0_to_fp16, x = input_381_cast_fp16)[name = tensor("lora_out_459_cast_fp16")]; + tensor query_39_cast_fp16 = add(x = pretrained_out_229_cast_fp16, y = lora_out_459_cast_fp16)[name = tensor("query_39_cast_fp16")]; + tensor var_31324 = const()[name = tensor("op_31324"), val = tensor([1, 1])]; + tensor var_31326 = const()[name = tensor("op_31326"), val = tensor([1, 1])]; + tensor pretrained_out_231_pad_type_0 = const()[name = tensor("pretrained_out_231_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_231_pad_0 = const()[name = tensor("pretrained_out_231_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(216632192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217451456))), name = tensor("layers_19_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_231_cast_fp16 = conv(dilations = var_31326, groups = var_31264, pad = pretrained_out_231_pad_0, pad_type = pretrained_out_231_pad_type_0, strides = var_31324, weight = layers_19_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_77_cast_fp16)[name = tensor("pretrained_out_231_cast_fp16")]; + tensor var_31330 = const()[name = tensor("op_31330"), val = tensor([1, 1])]; + tensor var_31332 = const()[name = tensor("op_31332"), val = tensor([1, 1])]; + tensor input_383_pad_type_0 = const()[name = tensor("input_383_pad_type_0"), val = tensor("custom")]; + tensor input_383_pad_0 = const()[name = tensor("input_383_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217451584)))]; + tensor input_383_cast_fp16 = conv(dilations = var_31332, groups = var_31264, pad = input_383_pad_0, pad_type = input_383_pad_type_0, strides = var_31330, weight = layers_19_self_attn_k_proj_loraA_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor("input_383_cast_fp16")]; + tensor var_31336 = const()[name = tensor("op_31336"), val = tensor([1, 1])]; + tensor var_31338 = const()[name = tensor("op_31338"), val = tensor([1, 1])]; + tensor lora_out_461_pad_type_0 = const()[name = tensor("lora_out_461_pad_type_0"), val = tensor("custom")]; + tensor lora_out_461_pad_0 = const()[name = tensor("lora_out_461_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_463_weight_0_to_fp16 = const()[name = tensor("lora_out_463_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217492608)))]; + tensor lora_out_463_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_31338, groups = var_31264, pad = lora_out_461_pad_0, pad_type = lora_out_461_pad_type_0, strides = var_31336, weight = lora_out_463_weight_0_to_fp16, x = input_383_cast_fp16)[name = tensor("lora_out_463_cast_fp16")]; + tensor key_39_cast_fp16 = add(x = pretrained_out_231_cast_fp16, y = lora_out_463_cast_fp16)[name = tensor("key_39_cast_fp16")]; + tensor var_31349 = const()[name = tensor("op_31349"), val = tensor([1, 1])]; + tensor var_31351 = const()[name = tensor("op_31351"), val = tensor([1, 1])]; + tensor pretrained_out_233_pad_type_0 = const()[name = tensor("pretrained_out_233_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_233_pad_0 = const()[name = tensor("pretrained_out_233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217533632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218352896))), name = tensor("layers_19_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_19_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218353024)))]; + tensor pretrained_out_233_cast_fp16 = conv(bias = layers_19_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_31351, groups = var_31264, pad = pretrained_out_233_pad_0, pad_type = pretrained_out_233_pad_type_0, strides = var_31349, weight = layers_19_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_77_cast_fp16)[name = tensor("pretrained_out_233_cast_fp16")]; + tensor var_31355 = const()[name = tensor("op_31355"), val = tensor([1, 1])]; + tensor var_31357 = const()[name = tensor("op_31357"), val = tensor([1, 1])]; + tensor input_385_pad_type_0 = const()[name = tensor("input_385_pad_type_0"), val = tensor("custom")]; + tensor input_385_pad_0 = const()[name = tensor("input_385_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218355648)))]; + tensor input_385_cast_fp16 = conv(dilations = var_31357, groups = var_31264, pad = input_385_pad_0, pad_type = input_385_pad_type_0, strides = var_31355, weight = layers_19_self_attn_v_proj_loraA_weight_to_fp16, x = obj_77_cast_fp16)[name = tensor("input_385_cast_fp16")]; + tensor var_31361 = const()[name = tensor("op_31361"), val = tensor([1, 1])]; + tensor var_31363 = const()[name = tensor("op_31363"), val = tensor([1, 1])]; + tensor lora_out_465_pad_type_0 = const()[name = tensor("lora_out_465_pad_type_0"), val = tensor("custom")]; + tensor lora_out_465_pad_0 = const()[name = tensor("lora_out_465_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_467_weight_0_to_fp16 = const()[name = tensor("lora_out_467_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218396672)))]; + tensor lora_out_467_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_31363, groups = var_31264, pad = lora_out_465_pad_0, pad_type = lora_out_465_pad_type_0, strides = var_31361, weight = lora_out_467_weight_0_to_fp16, x = input_385_cast_fp16)[name = tensor("lora_out_467_cast_fp16")]; + tensor value_39_cast_fp16 = add(x = pretrained_out_233_cast_fp16, y = lora_out_467_cast_fp16)[name = tensor("value_39_cast_fp16")]; + tensor var_31373_begin_0 = const()[name = tensor("op_31373_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31373_end_0 = const()[name = tensor("op_31373_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31373_end_mask_0 = const()[name = tensor("op_31373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31373_cast_fp16 = slice_by_index(begin = var_31373_begin_0, end = var_31373_end_0, end_mask = var_31373_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31373_cast_fp16")]; + tensor var_31377_begin_0 = const()[name = tensor("op_31377_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_31377_end_0 = const()[name = tensor("op_31377_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_31377_end_mask_0 = const()[name = tensor("op_31377_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31377_cast_fp16 = slice_by_index(begin = var_31377_begin_0, end = var_31377_end_0, end_mask = var_31377_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31377_cast_fp16")]; + tensor var_31381_begin_0 = const()[name = tensor("op_31381_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_31381_end_0 = const()[name = tensor("op_31381_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_31381_end_mask_0 = const()[name = tensor("op_31381_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31381_cast_fp16 = slice_by_index(begin = var_31381_begin_0, end = var_31381_end_0, end_mask = var_31381_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31381_cast_fp16")]; + tensor var_31385_begin_0 = const()[name = tensor("op_31385_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_31385_end_0 = const()[name = tensor("op_31385_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_31385_end_mask_0 = const()[name = tensor("op_31385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31385_cast_fp16 = slice_by_index(begin = var_31385_begin_0, end = var_31385_end_0, end_mask = var_31385_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31385_cast_fp16")]; + tensor var_31389_begin_0 = const()[name = tensor("op_31389_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_31389_end_0 = const()[name = tensor("op_31389_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_31389_end_mask_0 = const()[name = tensor("op_31389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31389_cast_fp16 = slice_by_index(begin = var_31389_begin_0, end = var_31389_end_0, end_mask = var_31389_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31389_cast_fp16")]; + tensor var_31393_begin_0 = const()[name = tensor("op_31393_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_31393_end_0 = const()[name = tensor("op_31393_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_31393_end_mask_0 = const()[name = tensor("op_31393_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31393_cast_fp16 = slice_by_index(begin = var_31393_begin_0, end = var_31393_end_0, end_mask = var_31393_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31393_cast_fp16")]; + tensor var_31397_begin_0 = const()[name = tensor("op_31397_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_31397_end_0 = const()[name = tensor("op_31397_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_31397_end_mask_0 = const()[name = tensor("op_31397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31397_cast_fp16 = slice_by_index(begin = var_31397_begin_0, end = var_31397_end_0, end_mask = var_31397_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31397_cast_fp16")]; + tensor var_31401_begin_0 = const()[name = tensor("op_31401_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_31401_end_0 = const()[name = tensor("op_31401_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_31401_end_mask_0 = const()[name = tensor("op_31401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31401_cast_fp16 = slice_by_index(begin = var_31401_begin_0, end = var_31401_end_0, end_mask = var_31401_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31401_cast_fp16")]; + tensor var_31405_begin_0 = const()[name = tensor("op_31405_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_31405_end_0 = const()[name = tensor("op_31405_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_31405_end_mask_0 = const()[name = tensor("op_31405_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31405_cast_fp16 = slice_by_index(begin = var_31405_begin_0, end = var_31405_end_0, end_mask = var_31405_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31405_cast_fp16")]; + tensor var_31409_begin_0 = const()[name = tensor("op_31409_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_31409_end_0 = const()[name = tensor("op_31409_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_31409_end_mask_0 = const()[name = tensor("op_31409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31409_cast_fp16 = slice_by_index(begin = var_31409_begin_0, end = var_31409_end_0, end_mask = var_31409_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31409_cast_fp16")]; + tensor var_31413_begin_0 = const()[name = tensor("op_31413_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_31413_end_0 = const()[name = tensor("op_31413_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_31413_end_mask_0 = const()[name = tensor("op_31413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31413_cast_fp16 = slice_by_index(begin = var_31413_begin_0, end = var_31413_end_0, end_mask = var_31413_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31413_cast_fp16")]; + tensor var_31417_begin_0 = const()[name = tensor("op_31417_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_31417_end_0 = const()[name = tensor("op_31417_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_31417_end_mask_0 = const()[name = tensor("op_31417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31417_cast_fp16 = slice_by_index(begin = var_31417_begin_0, end = var_31417_end_0, end_mask = var_31417_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31417_cast_fp16")]; + tensor var_31421_begin_0 = const()[name = tensor("op_31421_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_31421_end_0 = const()[name = tensor("op_31421_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_31421_end_mask_0 = const()[name = tensor("op_31421_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31421_cast_fp16 = slice_by_index(begin = var_31421_begin_0, end = var_31421_end_0, end_mask = var_31421_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31421_cast_fp16")]; + tensor var_31425_begin_0 = const()[name = tensor("op_31425_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_31425_end_0 = const()[name = tensor("op_31425_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_31425_end_mask_0 = const()[name = tensor("op_31425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31425_cast_fp16 = slice_by_index(begin = var_31425_begin_0, end = var_31425_end_0, end_mask = var_31425_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31425_cast_fp16")]; + tensor var_31429_begin_0 = const()[name = tensor("op_31429_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_31429_end_0 = const()[name = tensor("op_31429_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_31429_end_mask_0 = const()[name = tensor("op_31429_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31429_cast_fp16 = slice_by_index(begin = var_31429_begin_0, end = var_31429_end_0, end_mask = var_31429_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31429_cast_fp16")]; + tensor var_31433_begin_0 = const()[name = tensor("op_31433_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_31433_end_0 = const()[name = tensor("op_31433_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_31433_end_mask_0 = const()[name = tensor("op_31433_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31433_cast_fp16 = slice_by_index(begin = var_31433_begin_0, end = var_31433_end_0, end_mask = var_31433_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31433_cast_fp16")]; + tensor var_31437_begin_0 = const()[name = tensor("op_31437_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_31437_end_0 = const()[name = tensor("op_31437_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_31437_end_mask_0 = const()[name = tensor("op_31437_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31437_cast_fp16 = slice_by_index(begin = var_31437_begin_0, end = var_31437_end_0, end_mask = var_31437_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31437_cast_fp16")]; + tensor var_31441_begin_0 = const()[name = tensor("op_31441_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_31441_end_0 = const()[name = tensor("op_31441_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_31441_end_mask_0 = const()[name = tensor("op_31441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31441_cast_fp16 = slice_by_index(begin = var_31441_begin_0, end = var_31441_end_0, end_mask = var_31441_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31441_cast_fp16")]; + tensor var_31445_begin_0 = const()[name = tensor("op_31445_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_31445_end_0 = const()[name = tensor("op_31445_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_31445_end_mask_0 = const()[name = tensor("op_31445_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31445_cast_fp16 = slice_by_index(begin = var_31445_begin_0, end = var_31445_end_0, end_mask = var_31445_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31445_cast_fp16")]; + tensor var_31449_begin_0 = const()[name = tensor("op_31449_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_31449_end_0 = const()[name = tensor("op_31449_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_31449_end_mask_0 = const()[name = tensor("op_31449_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_31449_cast_fp16 = slice_by_index(begin = var_31449_begin_0, end = var_31449_end_0, end_mask = var_31449_end_mask_0, x = query_39_cast_fp16)[name = tensor("op_31449_cast_fp16")]; + tensor var_31458_begin_0 = const()[name = tensor("op_31458_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31458_end_0 = const()[name = tensor("op_31458_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31458_end_mask_0 = const()[name = tensor("op_31458_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31458_cast_fp16 = slice_by_index(begin = var_31458_begin_0, end = var_31458_end_0, end_mask = var_31458_end_mask_0, x = var_31373_cast_fp16)[name = tensor("op_31458_cast_fp16")]; + tensor var_31465_begin_0 = const()[name = tensor("op_31465_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31465_end_0 = const()[name = tensor("op_31465_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31465_end_mask_0 = const()[name = tensor("op_31465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31465_cast_fp16 = slice_by_index(begin = var_31465_begin_0, end = var_31465_end_0, end_mask = var_31465_end_mask_0, x = var_31373_cast_fp16)[name = tensor("op_31465_cast_fp16")]; + tensor var_31472_begin_0 = const()[name = tensor("op_31472_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31472_end_0 = const()[name = tensor("op_31472_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31472_end_mask_0 = const()[name = tensor("op_31472_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31472_cast_fp16 = slice_by_index(begin = var_31472_begin_0, end = var_31472_end_0, end_mask = var_31472_end_mask_0, x = var_31373_cast_fp16)[name = tensor("op_31472_cast_fp16")]; + tensor var_31479_begin_0 = const()[name = tensor("op_31479_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31479_end_0 = const()[name = tensor("op_31479_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31479_end_mask_0 = const()[name = tensor("op_31479_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31479_cast_fp16 = slice_by_index(begin = var_31479_begin_0, end = var_31479_end_0, end_mask = var_31479_end_mask_0, x = var_31373_cast_fp16)[name = tensor("op_31479_cast_fp16")]; + tensor var_31486_begin_0 = const()[name = tensor("op_31486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31486_end_0 = const()[name = tensor("op_31486_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31486_end_mask_0 = const()[name = tensor("op_31486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31486_cast_fp16 = slice_by_index(begin = var_31486_begin_0, end = var_31486_end_0, end_mask = var_31486_end_mask_0, x = var_31377_cast_fp16)[name = tensor("op_31486_cast_fp16")]; + tensor var_31493_begin_0 = const()[name = tensor("op_31493_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31493_end_0 = const()[name = tensor("op_31493_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31493_end_mask_0 = const()[name = tensor("op_31493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31493_cast_fp16 = slice_by_index(begin = var_31493_begin_0, end = var_31493_end_0, end_mask = var_31493_end_mask_0, x = var_31377_cast_fp16)[name = tensor("op_31493_cast_fp16")]; + tensor var_31500_begin_0 = const()[name = tensor("op_31500_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31500_end_0 = const()[name = tensor("op_31500_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31500_end_mask_0 = const()[name = tensor("op_31500_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31500_cast_fp16 = slice_by_index(begin = var_31500_begin_0, end = var_31500_end_0, end_mask = var_31500_end_mask_0, x = var_31377_cast_fp16)[name = tensor("op_31500_cast_fp16")]; + tensor var_31507_begin_0 = const()[name = tensor("op_31507_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31507_end_0 = const()[name = tensor("op_31507_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31507_end_mask_0 = const()[name = tensor("op_31507_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31507_cast_fp16 = slice_by_index(begin = var_31507_begin_0, end = var_31507_end_0, end_mask = var_31507_end_mask_0, x = var_31377_cast_fp16)[name = tensor("op_31507_cast_fp16")]; + tensor var_31514_begin_0 = const()[name = tensor("op_31514_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31514_end_0 = const()[name = tensor("op_31514_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31514_end_mask_0 = const()[name = tensor("op_31514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31514_cast_fp16 = slice_by_index(begin = var_31514_begin_0, end = var_31514_end_0, end_mask = var_31514_end_mask_0, x = var_31381_cast_fp16)[name = tensor("op_31514_cast_fp16")]; + tensor var_31521_begin_0 = const()[name = tensor("op_31521_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31521_end_0 = const()[name = tensor("op_31521_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31521_end_mask_0 = const()[name = tensor("op_31521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31521_cast_fp16 = slice_by_index(begin = var_31521_begin_0, end = var_31521_end_0, end_mask = var_31521_end_mask_0, x = var_31381_cast_fp16)[name = tensor("op_31521_cast_fp16")]; + tensor var_31528_begin_0 = const()[name = tensor("op_31528_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31528_end_0 = const()[name = tensor("op_31528_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31528_end_mask_0 = const()[name = tensor("op_31528_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31528_cast_fp16 = slice_by_index(begin = var_31528_begin_0, end = var_31528_end_0, end_mask = var_31528_end_mask_0, x = var_31381_cast_fp16)[name = tensor("op_31528_cast_fp16")]; + tensor var_31535_begin_0 = const()[name = tensor("op_31535_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31535_end_0 = const()[name = tensor("op_31535_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31535_end_mask_0 = const()[name = tensor("op_31535_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31535_cast_fp16 = slice_by_index(begin = var_31535_begin_0, end = var_31535_end_0, end_mask = var_31535_end_mask_0, x = var_31381_cast_fp16)[name = tensor("op_31535_cast_fp16")]; + tensor var_31542_begin_0 = const()[name = tensor("op_31542_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31542_end_0 = const()[name = tensor("op_31542_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31542_end_mask_0 = const()[name = tensor("op_31542_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31542_cast_fp16 = slice_by_index(begin = var_31542_begin_0, end = var_31542_end_0, end_mask = var_31542_end_mask_0, x = var_31385_cast_fp16)[name = tensor("op_31542_cast_fp16")]; + tensor var_31549_begin_0 = const()[name = tensor("op_31549_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31549_end_0 = const()[name = tensor("op_31549_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31549_end_mask_0 = const()[name = tensor("op_31549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31549_cast_fp16 = slice_by_index(begin = var_31549_begin_0, end = var_31549_end_0, end_mask = var_31549_end_mask_0, x = var_31385_cast_fp16)[name = tensor("op_31549_cast_fp16")]; + tensor var_31556_begin_0 = const()[name = tensor("op_31556_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31556_end_0 = const()[name = tensor("op_31556_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31556_end_mask_0 = const()[name = tensor("op_31556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31556_cast_fp16 = slice_by_index(begin = var_31556_begin_0, end = var_31556_end_0, end_mask = var_31556_end_mask_0, x = var_31385_cast_fp16)[name = tensor("op_31556_cast_fp16")]; + tensor var_31563_begin_0 = const()[name = tensor("op_31563_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31563_end_0 = const()[name = tensor("op_31563_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31563_end_mask_0 = const()[name = tensor("op_31563_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31563_cast_fp16 = slice_by_index(begin = var_31563_begin_0, end = var_31563_end_0, end_mask = var_31563_end_mask_0, x = var_31385_cast_fp16)[name = tensor("op_31563_cast_fp16")]; + tensor var_31570_begin_0 = const()[name = tensor("op_31570_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31570_end_0 = const()[name = tensor("op_31570_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31570_end_mask_0 = const()[name = tensor("op_31570_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31570_cast_fp16 = slice_by_index(begin = var_31570_begin_0, end = var_31570_end_0, end_mask = var_31570_end_mask_0, x = var_31389_cast_fp16)[name = tensor("op_31570_cast_fp16")]; + tensor var_31577_begin_0 = const()[name = tensor("op_31577_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31577_end_0 = const()[name = tensor("op_31577_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31577_end_mask_0 = const()[name = tensor("op_31577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31577_cast_fp16 = slice_by_index(begin = var_31577_begin_0, end = var_31577_end_0, end_mask = var_31577_end_mask_0, x = var_31389_cast_fp16)[name = tensor("op_31577_cast_fp16")]; + tensor var_31584_begin_0 = const()[name = tensor("op_31584_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31584_end_0 = const()[name = tensor("op_31584_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31584_end_mask_0 = const()[name = tensor("op_31584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31584_cast_fp16 = slice_by_index(begin = var_31584_begin_0, end = var_31584_end_0, end_mask = var_31584_end_mask_0, x = var_31389_cast_fp16)[name = tensor("op_31584_cast_fp16")]; + tensor var_31591_begin_0 = const()[name = tensor("op_31591_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31591_end_0 = const()[name = tensor("op_31591_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31591_end_mask_0 = const()[name = tensor("op_31591_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31591_cast_fp16 = slice_by_index(begin = var_31591_begin_0, end = var_31591_end_0, end_mask = var_31591_end_mask_0, x = var_31389_cast_fp16)[name = tensor("op_31591_cast_fp16")]; + tensor var_31598_begin_0 = const()[name = tensor("op_31598_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31598_end_0 = const()[name = tensor("op_31598_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31598_end_mask_0 = const()[name = tensor("op_31598_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31598_cast_fp16 = slice_by_index(begin = var_31598_begin_0, end = var_31598_end_0, end_mask = var_31598_end_mask_0, x = var_31393_cast_fp16)[name = tensor("op_31598_cast_fp16")]; + tensor var_31605_begin_0 = const()[name = tensor("op_31605_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31605_end_0 = const()[name = tensor("op_31605_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31605_end_mask_0 = const()[name = tensor("op_31605_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31605_cast_fp16 = slice_by_index(begin = var_31605_begin_0, end = var_31605_end_0, end_mask = var_31605_end_mask_0, x = var_31393_cast_fp16)[name = tensor("op_31605_cast_fp16")]; + tensor var_31612_begin_0 = const()[name = tensor("op_31612_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31612_end_0 = const()[name = tensor("op_31612_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31612_end_mask_0 = const()[name = tensor("op_31612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31612_cast_fp16 = slice_by_index(begin = var_31612_begin_0, end = var_31612_end_0, end_mask = var_31612_end_mask_0, x = var_31393_cast_fp16)[name = tensor("op_31612_cast_fp16")]; + tensor var_31619_begin_0 = const()[name = tensor("op_31619_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31619_end_0 = const()[name = tensor("op_31619_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31619_end_mask_0 = const()[name = tensor("op_31619_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31619_cast_fp16 = slice_by_index(begin = var_31619_begin_0, end = var_31619_end_0, end_mask = var_31619_end_mask_0, x = var_31393_cast_fp16)[name = tensor("op_31619_cast_fp16")]; + tensor var_31626_begin_0 = const()[name = tensor("op_31626_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31626_end_0 = const()[name = tensor("op_31626_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31626_end_mask_0 = const()[name = tensor("op_31626_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31626_cast_fp16 = slice_by_index(begin = var_31626_begin_0, end = var_31626_end_0, end_mask = var_31626_end_mask_0, x = var_31397_cast_fp16)[name = tensor("op_31626_cast_fp16")]; + tensor var_31633_begin_0 = const()[name = tensor("op_31633_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31633_end_0 = const()[name = tensor("op_31633_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31633_end_mask_0 = const()[name = tensor("op_31633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31633_cast_fp16 = slice_by_index(begin = var_31633_begin_0, end = var_31633_end_0, end_mask = var_31633_end_mask_0, x = var_31397_cast_fp16)[name = tensor("op_31633_cast_fp16")]; + tensor var_31640_begin_0 = const()[name = tensor("op_31640_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31640_end_0 = const()[name = tensor("op_31640_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31640_end_mask_0 = const()[name = tensor("op_31640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31640_cast_fp16 = slice_by_index(begin = var_31640_begin_0, end = var_31640_end_0, end_mask = var_31640_end_mask_0, x = var_31397_cast_fp16)[name = tensor("op_31640_cast_fp16")]; + tensor var_31647_begin_0 = const()[name = tensor("op_31647_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31647_end_0 = const()[name = tensor("op_31647_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31647_end_mask_0 = const()[name = tensor("op_31647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31647_cast_fp16 = slice_by_index(begin = var_31647_begin_0, end = var_31647_end_0, end_mask = var_31647_end_mask_0, x = var_31397_cast_fp16)[name = tensor("op_31647_cast_fp16")]; + tensor var_31654_begin_0 = const()[name = tensor("op_31654_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31654_end_0 = const()[name = tensor("op_31654_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31654_end_mask_0 = const()[name = tensor("op_31654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31654_cast_fp16 = slice_by_index(begin = var_31654_begin_0, end = var_31654_end_0, end_mask = var_31654_end_mask_0, x = var_31401_cast_fp16)[name = tensor("op_31654_cast_fp16")]; + tensor var_31661_begin_0 = const()[name = tensor("op_31661_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31661_end_0 = const()[name = tensor("op_31661_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31661_end_mask_0 = const()[name = tensor("op_31661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31661_cast_fp16 = slice_by_index(begin = var_31661_begin_0, end = var_31661_end_0, end_mask = var_31661_end_mask_0, x = var_31401_cast_fp16)[name = tensor("op_31661_cast_fp16")]; + tensor var_31668_begin_0 = const()[name = tensor("op_31668_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31668_end_0 = const()[name = tensor("op_31668_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31668_end_mask_0 = const()[name = tensor("op_31668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31668_cast_fp16 = slice_by_index(begin = var_31668_begin_0, end = var_31668_end_0, end_mask = var_31668_end_mask_0, x = var_31401_cast_fp16)[name = tensor("op_31668_cast_fp16")]; + tensor var_31675_begin_0 = const()[name = tensor("op_31675_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31675_end_0 = const()[name = tensor("op_31675_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31675_end_mask_0 = const()[name = tensor("op_31675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31675_cast_fp16 = slice_by_index(begin = var_31675_begin_0, end = var_31675_end_0, end_mask = var_31675_end_mask_0, x = var_31401_cast_fp16)[name = tensor("op_31675_cast_fp16")]; + tensor var_31682_begin_0 = const()[name = tensor("op_31682_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31682_end_0 = const()[name = tensor("op_31682_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31682_end_mask_0 = const()[name = tensor("op_31682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31682_cast_fp16 = slice_by_index(begin = var_31682_begin_0, end = var_31682_end_0, end_mask = var_31682_end_mask_0, x = var_31405_cast_fp16)[name = tensor("op_31682_cast_fp16")]; + tensor var_31689_begin_0 = const()[name = tensor("op_31689_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31689_end_0 = const()[name = tensor("op_31689_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31689_end_mask_0 = const()[name = tensor("op_31689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31689_cast_fp16 = slice_by_index(begin = var_31689_begin_0, end = var_31689_end_0, end_mask = var_31689_end_mask_0, x = var_31405_cast_fp16)[name = tensor("op_31689_cast_fp16")]; + tensor var_31696_begin_0 = const()[name = tensor("op_31696_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31696_end_0 = const()[name = tensor("op_31696_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31696_end_mask_0 = const()[name = tensor("op_31696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31696_cast_fp16 = slice_by_index(begin = var_31696_begin_0, end = var_31696_end_0, end_mask = var_31696_end_mask_0, x = var_31405_cast_fp16)[name = tensor("op_31696_cast_fp16")]; + tensor var_31703_begin_0 = const()[name = tensor("op_31703_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31703_end_0 = const()[name = tensor("op_31703_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31703_end_mask_0 = const()[name = tensor("op_31703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31703_cast_fp16 = slice_by_index(begin = var_31703_begin_0, end = var_31703_end_0, end_mask = var_31703_end_mask_0, x = var_31405_cast_fp16)[name = tensor("op_31703_cast_fp16")]; + tensor var_31710_begin_0 = const()[name = tensor("op_31710_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31710_end_0 = const()[name = tensor("op_31710_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31710_end_mask_0 = const()[name = tensor("op_31710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31710_cast_fp16 = slice_by_index(begin = var_31710_begin_0, end = var_31710_end_0, end_mask = var_31710_end_mask_0, x = var_31409_cast_fp16)[name = tensor("op_31710_cast_fp16")]; + tensor var_31717_begin_0 = const()[name = tensor("op_31717_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31717_end_0 = const()[name = tensor("op_31717_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31717_end_mask_0 = const()[name = tensor("op_31717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31717_cast_fp16 = slice_by_index(begin = var_31717_begin_0, end = var_31717_end_0, end_mask = var_31717_end_mask_0, x = var_31409_cast_fp16)[name = tensor("op_31717_cast_fp16")]; + tensor var_31724_begin_0 = const()[name = tensor("op_31724_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31724_end_0 = const()[name = tensor("op_31724_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31724_end_mask_0 = const()[name = tensor("op_31724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31724_cast_fp16 = slice_by_index(begin = var_31724_begin_0, end = var_31724_end_0, end_mask = var_31724_end_mask_0, x = var_31409_cast_fp16)[name = tensor("op_31724_cast_fp16")]; + tensor var_31731_begin_0 = const()[name = tensor("op_31731_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31731_end_0 = const()[name = tensor("op_31731_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31731_end_mask_0 = const()[name = tensor("op_31731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31731_cast_fp16 = slice_by_index(begin = var_31731_begin_0, end = var_31731_end_0, end_mask = var_31731_end_mask_0, x = var_31409_cast_fp16)[name = tensor("op_31731_cast_fp16")]; + tensor var_31738_begin_0 = const()[name = tensor("op_31738_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31738_end_0 = const()[name = tensor("op_31738_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31738_end_mask_0 = const()[name = tensor("op_31738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31738_cast_fp16 = slice_by_index(begin = var_31738_begin_0, end = var_31738_end_0, end_mask = var_31738_end_mask_0, x = var_31413_cast_fp16)[name = tensor("op_31738_cast_fp16")]; + tensor var_31745_begin_0 = const()[name = tensor("op_31745_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31745_end_0 = const()[name = tensor("op_31745_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31745_end_mask_0 = const()[name = tensor("op_31745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31745_cast_fp16 = slice_by_index(begin = var_31745_begin_0, end = var_31745_end_0, end_mask = var_31745_end_mask_0, x = var_31413_cast_fp16)[name = tensor("op_31745_cast_fp16")]; + tensor var_31752_begin_0 = const()[name = tensor("op_31752_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31752_end_0 = const()[name = tensor("op_31752_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31752_end_mask_0 = const()[name = tensor("op_31752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31752_cast_fp16 = slice_by_index(begin = var_31752_begin_0, end = var_31752_end_0, end_mask = var_31752_end_mask_0, x = var_31413_cast_fp16)[name = tensor("op_31752_cast_fp16")]; + tensor var_31759_begin_0 = const()[name = tensor("op_31759_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31759_end_0 = const()[name = tensor("op_31759_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31759_end_mask_0 = const()[name = tensor("op_31759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31759_cast_fp16 = slice_by_index(begin = var_31759_begin_0, end = var_31759_end_0, end_mask = var_31759_end_mask_0, x = var_31413_cast_fp16)[name = tensor("op_31759_cast_fp16")]; + tensor var_31766_begin_0 = const()[name = tensor("op_31766_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31766_end_0 = const()[name = tensor("op_31766_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31766_end_mask_0 = const()[name = tensor("op_31766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31766_cast_fp16 = slice_by_index(begin = var_31766_begin_0, end = var_31766_end_0, end_mask = var_31766_end_mask_0, x = var_31417_cast_fp16)[name = tensor("op_31766_cast_fp16")]; + tensor var_31773_begin_0 = const()[name = tensor("op_31773_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31773_end_0 = const()[name = tensor("op_31773_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31773_end_mask_0 = const()[name = tensor("op_31773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31773_cast_fp16 = slice_by_index(begin = var_31773_begin_0, end = var_31773_end_0, end_mask = var_31773_end_mask_0, x = var_31417_cast_fp16)[name = tensor("op_31773_cast_fp16")]; + tensor var_31780_begin_0 = const()[name = tensor("op_31780_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31780_end_0 = const()[name = tensor("op_31780_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31780_end_mask_0 = const()[name = tensor("op_31780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31780_cast_fp16 = slice_by_index(begin = var_31780_begin_0, end = var_31780_end_0, end_mask = var_31780_end_mask_0, x = var_31417_cast_fp16)[name = tensor("op_31780_cast_fp16")]; + tensor var_31787_begin_0 = const()[name = tensor("op_31787_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31787_end_0 = const()[name = tensor("op_31787_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31787_end_mask_0 = const()[name = tensor("op_31787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31787_cast_fp16 = slice_by_index(begin = var_31787_begin_0, end = var_31787_end_0, end_mask = var_31787_end_mask_0, x = var_31417_cast_fp16)[name = tensor("op_31787_cast_fp16")]; + tensor var_31794_begin_0 = const()[name = tensor("op_31794_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31794_end_0 = const()[name = tensor("op_31794_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31794_end_mask_0 = const()[name = tensor("op_31794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31794_cast_fp16 = slice_by_index(begin = var_31794_begin_0, end = var_31794_end_0, end_mask = var_31794_end_mask_0, x = var_31421_cast_fp16)[name = tensor("op_31794_cast_fp16")]; + tensor var_31801_begin_0 = const()[name = tensor("op_31801_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31801_end_0 = const()[name = tensor("op_31801_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31801_end_mask_0 = const()[name = tensor("op_31801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31801_cast_fp16 = slice_by_index(begin = var_31801_begin_0, end = var_31801_end_0, end_mask = var_31801_end_mask_0, x = var_31421_cast_fp16)[name = tensor("op_31801_cast_fp16")]; + tensor var_31808_begin_0 = const()[name = tensor("op_31808_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31808_end_0 = const()[name = tensor("op_31808_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31808_end_mask_0 = const()[name = tensor("op_31808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31808_cast_fp16 = slice_by_index(begin = var_31808_begin_0, end = var_31808_end_0, end_mask = var_31808_end_mask_0, x = var_31421_cast_fp16)[name = tensor("op_31808_cast_fp16")]; + tensor var_31815_begin_0 = const()[name = tensor("op_31815_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31815_end_0 = const()[name = tensor("op_31815_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31815_end_mask_0 = const()[name = tensor("op_31815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31815_cast_fp16 = slice_by_index(begin = var_31815_begin_0, end = var_31815_end_0, end_mask = var_31815_end_mask_0, x = var_31421_cast_fp16)[name = tensor("op_31815_cast_fp16")]; + tensor var_31822_begin_0 = const()[name = tensor("op_31822_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31822_end_0 = const()[name = tensor("op_31822_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31822_end_mask_0 = const()[name = tensor("op_31822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31822_cast_fp16 = slice_by_index(begin = var_31822_begin_0, end = var_31822_end_0, end_mask = var_31822_end_mask_0, x = var_31425_cast_fp16)[name = tensor("op_31822_cast_fp16")]; + tensor var_31829_begin_0 = const()[name = tensor("op_31829_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31829_end_0 = const()[name = tensor("op_31829_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31829_end_mask_0 = const()[name = tensor("op_31829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31829_cast_fp16 = slice_by_index(begin = var_31829_begin_0, end = var_31829_end_0, end_mask = var_31829_end_mask_0, x = var_31425_cast_fp16)[name = tensor("op_31829_cast_fp16")]; + tensor var_31836_begin_0 = const()[name = tensor("op_31836_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31836_end_0 = const()[name = tensor("op_31836_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31836_end_mask_0 = const()[name = tensor("op_31836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31836_cast_fp16 = slice_by_index(begin = var_31836_begin_0, end = var_31836_end_0, end_mask = var_31836_end_mask_0, x = var_31425_cast_fp16)[name = tensor("op_31836_cast_fp16")]; + tensor var_31843_begin_0 = const()[name = tensor("op_31843_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31843_end_0 = const()[name = tensor("op_31843_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31843_end_mask_0 = const()[name = tensor("op_31843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31843_cast_fp16 = slice_by_index(begin = var_31843_begin_0, end = var_31843_end_0, end_mask = var_31843_end_mask_0, x = var_31425_cast_fp16)[name = tensor("op_31843_cast_fp16")]; + tensor var_31850_begin_0 = const()[name = tensor("op_31850_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31850_end_0 = const()[name = tensor("op_31850_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31850_end_mask_0 = const()[name = tensor("op_31850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31850_cast_fp16 = slice_by_index(begin = var_31850_begin_0, end = var_31850_end_0, end_mask = var_31850_end_mask_0, x = var_31429_cast_fp16)[name = tensor("op_31850_cast_fp16")]; + tensor var_31857_begin_0 = const()[name = tensor("op_31857_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31857_end_0 = const()[name = tensor("op_31857_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31857_end_mask_0 = const()[name = tensor("op_31857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31857_cast_fp16 = slice_by_index(begin = var_31857_begin_0, end = var_31857_end_0, end_mask = var_31857_end_mask_0, x = var_31429_cast_fp16)[name = tensor("op_31857_cast_fp16")]; + tensor var_31864_begin_0 = const()[name = tensor("op_31864_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31864_end_0 = const()[name = tensor("op_31864_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31864_end_mask_0 = const()[name = tensor("op_31864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31864_cast_fp16 = slice_by_index(begin = var_31864_begin_0, end = var_31864_end_0, end_mask = var_31864_end_mask_0, x = var_31429_cast_fp16)[name = tensor("op_31864_cast_fp16")]; + tensor var_31871_begin_0 = const()[name = tensor("op_31871_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31871_end_0 = const()[name = tensor("op_31871_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31871_end_mask_0 = const()[name = tensor("op_31871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31871_cast_fp16 = slice_by_index(begin = var_31871_begin_0, end = var_31871_end_0, end_mask = var_31871_end_mask_0, x = var_31429_cast_fp16)[name = tensor("op_31871_cast_fp16")]; + tensor var_31878_begin_0 = const()[name = tensor("op_31878_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31878_end_0 = const()[name = tensor("op_31878_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31878_end_mask_0 = const()[name = tensor("op_31878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31878_cast_fp16 = slice_by_index(begin = var_31878_begin_0, end = var_31878_end_0, end_mask = var_31878_end_mask_0, x = var_31433_cast_fp16)[name = tensor("op_31878_cast_fp16")]; + tensor var_31885_begin_0 = const()[name = tensor("op_31885_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31885_end_0 = const()[name = tensor("op_31885_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31885_end_mask_0 = const()[name = tensor("op_31885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31885_cast_fp16 = slice_by_index(begin = var_31885_begin_0, end = var_31885_end_0, end_mask = var_31885_end_mask_0, x = var_31433_cast_fp16)[name = tensor("op_31885_cast_fp16")]; + tensor var_31892_begin_0 = const()[name = tensor("op_31892_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31892_end_0 = const()[name = tensor("op_31892_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31892_end_mask_0 = const()[name = tensor("op_31892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31892_cast_fp16 = slice_by_index(begin = var_31892_begin_0, end = var_31892_end_0, end_mask = var_31892_end_mask_0, x = var_31433_cast_fp16)[name = tensor("op_31892_cast_fp16")]; + tensor var_31899_begin_0 = const()[name = tensor("op_31899_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31899_end_0 = const()[name = tensor("op_31899_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31899_end_mask_0 = const()[name = tensor("op_31899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31899_cast_fp16 = slice_by_index(begin = var_31899_begin_0, end = var_31899_end_0, end_mask = var_31899_end_mask_0, x = var_31433_cast_fp16)[name = tensor("op_31899_cast_fp16")]; + tensor var_31906_begin_0 = const()[name = tensor("op_31906_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31906_end_0 = const()[name = tensor("op_31906_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31906_end_mask_0 = const()[name = tensor("op_31906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31906_cast_fp16 = slice_by_index(begin = var_31906_begin_0, end = var_31906_end_0, end_mask = var_31906_end_mask_0, x = var_31437_cast_fp16)[name = tensor("op_31906_cast_fp16")]; + tensor var_31913_begin_0 = const()[name = tensor("op_31913_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31913_end_0 = const()[name = tensor("op_31913_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31913_end_mask_0 = const()[name = tensor("op_31913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31913_cast_fp16 = slice_by_index(begin = var_31913_begin_0, end = var_31913_end_0, end_mask = var_31913_end_mask_0, x = var_31437_cast_fp16)[name = tensor("op_31913_cast_fp16")]; + tensor var_31920_begin_0 = const()[name = tensor("op_31920_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31920_end_0 = const()[name = tensor("op_31920_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31920_end_mask_0 = const()[name = tensor("op_31920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31920_cast_fp16 = slice_by_index(begin = var_31920_begin_0, end = var_31920_end_0, end_mask = var_31920_end_mask_0, x = var_31437_cast_fp16)[name = tensor("op_31920_cast_fp16")]; + tensor var_31927_begin_0 = const()[name = tensor("op_31927_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31927_end_0 = const()[name = tensor("op_31927_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31927_end_mask_0 = const()[name = tensor("op_31927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31927_cast_fp16 = slice_by_index(begin = var_31927_begin_0, end = var_31927_end_0, end_mask = var_31927_end_mask_0, x = var_31437_cast_fp16)[name = tensor("op_31927_cast_fp16")]; + tensor var_31934_begin_0 = const()[name = tensor("op_31934_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31934_end_0 = const()[name = tensor("op_31934_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31934_end_mask_0 = const()[name = tensor("op_31934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31934_cast_fp16 = slice_by_index(begin = var_31934_begin_0, end = var_31934_end_0, end_mask = var_31934_end_mask_0, x = var_31441_cast_fp16)[name = tensor("op_31934_cast_fp16")]; + tensor var_31941_begin_0 = const()[name = tensor("op_31941_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31941_end_0 = const()[name = tensor("op_31941_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31941_end_mask_0 = const()[name = tensor("op_31941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31941_cast_fp16 = slice_by_index(begin = var_31941_begin_0, end = var_31941_end_0, end_mask = var_31941_end_mask_0, x = var_31441_cast_fp16)[name = tensor("op_31941_cast_fp16")]; + tensor var_31948_begin_0 = const()[name = tensor("op_31948_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31948_end_0 = const()[name = tensor("op_31948_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31948_end_mask_0 = const()[name = tensor("op_31948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31948_cast_fp16 = slice_by_index(begin = var_31948_begin_0, end = var_31948_end_0, end_mask = var_31948_end_mask_0, x = var_31441_cast_fp16)[name = tensor("op_31948_cast_fp16")]; + tensor var_31955_begin_0 = const()[name = tensor("op_31955_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31955_end_0 = const()[name = tensor("op_31955_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31955_end_mask_0 = const()[name = tensor("op_31955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31955_cast_fp16 = slice_by_index(begin = var_31955_begin_0, end = var_31955_end_0, end_mask = var_31955_end_mask_0, x = var_31441_cast_fp16)[name = tensor("op_31955_cast_fp16")]; + tensor var_31962_begin_0 = const()[name = tensor("op_31962_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31962_end_0 = const()[name = tensor("op_31962_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31962_end_mask_0 = const()[name = tensor("op_31962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31962_cast_fp16 = slice_by_index(begin = var_31962_begin_0, end = var_31962_end_0, end_mask = var_31962_end_mask_0, x = var_31445_cast_fp16)[name = tensor("op_31962_cast_fp16")]; + tensor var_31969_begin_0 = const()[name = tensor("op_31969_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31969_end_0 = const()[name = tensor("op_31969_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31969_end_mask_0 = const()[name = tensor("op_31969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31969_cast_fp16 = slice_by_index(begin = var_31969_begin_0, end = var_31969_end_0, end_mask = var_31969_end_mask_0, x = var_31445_cast_fp16)[name = tensor("op_31969_cast_fp16")]; + tensor var_31976_begin_0 = const()[name = tensor("op_31976_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_31976_end_0 = const()[name = tensor("op_31976_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_31976_end_mask_0 = const()[name = tensor("op_31976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31976_cast_fp16 = slice_by_index(begin = var_31976_begin_0, end = var_31976_end_0, end_mask = var_31976_end_mask_0, x = var_31445_cast_fp16)[name = tensor("op_31976_cast_fp16")]; + tensor var_31983_begin_0 = const()[name = tensor("op_31983_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_31983_end_0 = const()[name = tensor("op_31983_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_31983_end_mask_0 = const()[name = tensor("op_31983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31983_cast_fp16 = slice_by_index(begin = var_31983_begin_0, end = var_31983_end_0, end_mask = var_31983_end_mask_0, x = var_31445_cast_fp16)[name = tensor("op_31983_cast_fp16")]; + tensor var_31990_begin_0 = const()[name = tensor("op_31990_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_31990_end_0 = const()[name = tensor("op_31990_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_31990_end_mask_0 = const()[name = tensor("op_31990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31990_cast_fp16 = slice_by_index(begin = var_31990_begin_0, end = var_31990_end_0, end_mask = var_31990_end_mask_0, x = var_31449_cast_fp16)[name = tensor("op_31990_cast_fp16")]; + tensor var_31997_begin_0 = const()[name = tensor("op_31997_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_31997_end_0 = const()[name = tensor("op_31997_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_31997_end_mask_0 = const()[name = tensor("op_31997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_31997_cast_fp16 = slice_by_index(begin = var_31997_begin_0, end = var_31997_end_0, end_mask = var_31997_end_mask_0, x = var_31449_cast_fp16)[name = tensor("op_31997_cast_fp16")]; + tensor var_32004_begin_0 = const()[name = tensor("op_32004_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_32004_end_0 = const()[name = tensor("op_32004_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_32004_end_mask_0 = const()[name = tensor("op_32004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32004_cast_fp16 = slice_by_index(begin = var_32004_begin_0, end = var_32004_end_0, end_mask = var_32004_end_mask_0, x = var_31449_cast_fp16)[name = tensor("op_32004_cast_fp16")]; + tensor var_32011_begin_0 = const()[name = tensor("op_32011_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_32011_end_0 = const()[name = tensor("op_32011_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32011_end_mask_0 = const()[name = tensor("op_32011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32011_cast_fp16 = slice_by_index(begin = var_32011_begin_0, end = var_32011_end_0, end_mask = var_32011_end_mask_0, x = var_31449_cast_fp16)[name = tensor("op_32011_cast_fp16")]; + tensor k_39_perm_0 = const()[name = tensor("k_39_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_32016_begin_0 = const()[name = tensor("op_32016_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32016_end_0 = const()[name = tensor("op_32016_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_32016_end_mask_0 = const()[name = tensor("op_32016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_12 = transpose(perm = k_39_perm_0, x = key_39_cast_fp16)[name = tensor("transpose_12")]; + tensor var_32016_cast_fp16 = slice_by_index(begin = var_32016_begin_0, end = var_32016_end_0, end_mask = var_32016_end_mask_0, x = transpose_12)[name = tensor("op_32016_cast_fp16")]; + tensor var_32020_begin_0 = const()[name = tensor("op_32020_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_32020_end_0 = const()[name = tensor("op_32020_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_32020_end_mask_0 = const()[name = tensor("op_32020_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32020_cast_fp16 = slice_by_index(begin = var_32020_begin_0, end = var_32020_end_0, end_mask = var_32020_end_mask_0, x = transpose_12)[name = tensor("op_32020_cast_fp16")]; + tensor var_32024_begin_0 = const()[name = tensor("op_32024_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_32024_end_0 = const()[name = tensor("op_32024_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_32024_end_mask_0 = const()[name = tensor("op_32024_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32024_cast_fp16 = slice_by_index(begin = var_32024_begin_0, end = var_32024_end_0, end_mask = var_32024_end_mask_0, x = transpose_12)[name = tensor("op_32024_cast_fp16")]; + tensor var_32028_begin_0 = const()[name = tensor("op_32028_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_32028_end_0 = const()[name = tensor("op_32028_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_32028_end_mask_0 = const()[name = tensor("op_32028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32028_cast_fp16 = slice_by_index(begin = var_32028_begin_0, end = var_32028_end_0, end_mask = var_32028_end_mask_0, x = transpose_12)[name = tensor("op_32028_cast_fp16")]; + tensor var_32032_begin_0 = const()[name = tensor("op_32032_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_32032_end_0 = const()[name = tensor("op_32032_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_32032_end_mask_0 = const()[name = tensor("op_32032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32032_cast_fp16 = slice_by_index(begin = var_32032_begin_0, end = var_32032_end_0, end_mask = var_32032_end_mask_0, x = transpose_12)[name = tensor("op_32032_cast_fp16")]; + tensor var_32036_begin_0 = const()[name = tensor("op_32036_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_32036_end_0 = const()[name = tensor("op_32036_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_32036_end_mask_0 = const()[name = tensor("op_32036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32036_cast_fp16 = slice_by_index(begin = var_32036_begin_0, end = var_32036_end_0, end_mask = var_32036_end_mask_0, x = transpose_12)[name = tensor("op_32036_cast_fp16")]; + tensor var_32040_begin_0 = const()[name = tensor("op_32040_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_32040_end_0 = const()[name = tensor("op_32040_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_32040_end_mask_0 = const()[name = tensor("op_32040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32040_cast_fp16 = slice_by_index(begin = var_32040_begin_0, end = var_32040_end_0, end_mask = var_32040_end_mask_0, x = transpose_12)[name = tensor("op_32040_cast_fp16")]; + tensor var_32044_begin_0 = const()[name = tensor("op_32044_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_32044_end_0 = const()[name = tensor("op_32044_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_32044_end_mask_0 = const()[name = tensor("op_32044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32044_cast_fp16 = slice_by_index(begin = var_32044_begin_0, end = var_32044_end_0, end_mask = var_32044_end_mask_0, x = transpose_12)[name = tensor("op_32044_cast_fp16")]; + tensor var_32048_begin_0 = const()[name = tensor("op_32048_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_32048_end_0 = const()[name = tensor("op_32048_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_32048_end_mask_0 = const()[name = tensor("op_32048_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32048_cast_fp16 = slice_by_index(begin = var_32048_begin_0, end = var_32048_end_0, end_mask = var_32048_end_mask_0, x = transpose_12)[name = tensor("op_32048_cast_fp16")]; + tensor var_32052_begin_0 = const()[name = tensor("op_32052_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_32052_end_0 = const()[name = tensor("op_32052_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_32052_end_mask_0 = const()[name = tensor("op_32052_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32052_cast_fp16 = slice_by_index(begin = var_32052_begin_0, end = var_32052_end_0, end_mask = var_32052_end_mask_0, x = transpose_12)[name = tensor("op_32052_cast_fp16")]; + tensor var_32056_begin_0 = const()[name = tensor("op_32056_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_32056_end_0 = const()[name = tensor("op_32056_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_32056_end_mask_0 = const()[name = tensor("op_32056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32056_cast_fp16 = slice_by_index(begin = var_32056_begin_0, end = var_32056_end_0, end_mask = var_32056_end_mask_0, x = transpose_12)[name = tensor("op_32056_cast_fp16")]; + tensor var_32060_begin_0 = const()[name = tensor("op_32060_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_32060_end_0 = const()[name = tensor("op_32060_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_32060_end_mask_0 = const()[name = tensor("op_32060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32060_cast_fp16 = slice_by_index(begin = var_32060_begin_0, end = var_32060_end_0, end_mask = var_32060_end_mask_0, x = transpose_12)[name = tensor("op_32060_cast_fp16")]; + tensor var_32064_begin_0 = const()[name = tensor("op_32064_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_32064_end_0 = const()[name = tensor("op_32064_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_32064_end_mask_0 = const()[name = tensor("op_32064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32064_cast_fp16 = slice_by_index(begin = var_32064_begin_0, end = var_32064_end_0, end_mask = var_32064_end_mask_0, x = transpose_12)[name = tensor("op_32064_cast_fp16")]; + tensor var_32068_begin_0 = const()[name = tensor("op_32068_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_32068_end_0 = const()[name = tensor("op_32068_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_32068_end_mask_0 = const()[name = tensor("op_32068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32068_cast_fp16 = slice_by_index(begin = var_32068_begin_0, end = var_32068_end_0, end_mask = var_32068_end_mask_0, x = transpose_12)[name = tensor("op_32068_cast_fp16")]; + tensor var_32072_begin_0 = const()[name = tensor("op_32072_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_32072_end_0 = const()[name = tensor("op_32072_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_32072_end_mask_0 = const()[name = tensor("op_32072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32072_cast_fp16 = slice_by_index(begin = var_32072_begin_0, end = var_32072_end_0, end_mask = var_32072_end_mask_0, x = transpose_12)[name = tensor("op_32072_cast_fp16")]; + tensor var_32076_begin_0 = const()[name = tensor("op_32076_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_32076_end_0 = const()[name = tensor("op_32076_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_32076_end_mask_0 = const()[name = tensor("op_32076_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32076_cast_fp16 = slice_by_index(begin = var_32076_begin_0, end = var_32076_end_0, end_mask = var_32076_end_mask_0, x = transpose_12)[name = tensor("op_32076_cast_fp16")]; + tensor var_32080_begin_0 = const()[name = tensor("op_32080_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_32080_end_0 = const()[name = tensor("op_32080_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_32080_end_mask_0 = const()[name = tensor("op_32080_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32080_cast_fp16 = slice_by_index(begin = var_32080_begin_0, end = var_32080_end_0, end_mask = var_32080_end_mask_0, x = transpose_12)[name = tensor("op_32080_cast_fp16")]; + tensor var_32084_begin_0 = const()[name = tensor("op_32084_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_32084_end_0 = const()[name = tensor("op_32084_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_32084_end_mask_0 = const()[name = tensor("op_32084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32084_cast_fp16 = slice_by_index(begin = var_32084_begin_0, end = var_32084_end_0, end_mask = var_32084_end_mask_0, x = transpose_12)[name = tensor("op_32084_cast_fp16")]; + tensor var_32088_begin_0 = const()[name = tensor("op_32088_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_32088_end_0 = const()[name = tensor("op_32088_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_32088_end_mask_0 = const()[name = tensor("op_32088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32088_cast_fp16 = slice_by_index(begin = var_32088_begin_0, end = var_32088_end_0, end_mask = var_32088_end_mask_0, x = transpose_12)[name = tensor("op_32088_cast_fp16")]; + tensor var_32092_begin_0 = const()[name = tensor("op_32092_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_32092_end_0 = const()[name = tensor("op_32092_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_32092_end_mask_0 = const()[name = tensor("op_32092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_32092_cast_fp16 = slice_by_index(begin = var_32092_begin_0, end = var_32092_end_0, end_mask = var_32092_end_mask_0, x = transpose_12)[name = tensor("op_32092_cast_fp16")]; + tensor var_32094_begin_0 = const()[name = tensor("op_32094_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_32094_end_0 = const()[name = tensor("op_32094_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_32094_end_mask_0 = const()[name = tensor("op_32094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32094_cast_fp16 = slice_by_index(begin = var_32094_begin_0, end = var_32094_end_0, end_mask = var_32094_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32094_cast_fp16")]; + tensor var_32098_begin_0 = const()[name = tensor("op_32098_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_32098_end_0 = const()[name = tensor("op_32098_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_32098_end_mask_0 = const()[name = tensor("op_32098_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32098_cast_fp16 = slice_by_index(begin = var_32098_begin_0, end = var_32098_end_0, end_mask = var_32098_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32098_cast_fp16")]; + tensor var_32102_begin_0 = const()[name = tensor("op_32102_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_32102_end_0 = const()[name = tensor("op_32102_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_32102_end_mask_0 = const()[name = tensor("op_32102_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32102_cast_fp16 = slice_by_index(begin = var_32102_begin_0, end = var_32102_end_0, end_mask = var_32102_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32102_cast_fp16")]; + tensor var_32106_begin_0 = const()[name = tensor("op_32106_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_32106_end_0 = const()[name = tensor("op_32106_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_32106_end_mask_0 = const()[name = tensor("op_32106_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32106_cast_fp16 = slice_by_index(begin = var_32106_begin_0, end = var_32106_end_0, end_mask = var_32106_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32106_cast_fp16")]; + tensor var_32110_begin_0 = const()[name = tensor("op_32110_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_32110_end_0 = const()[name = tensor("op_32110_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_32110_end_mask_0 = const()[name = tensor("op_32110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32110_cast_fp16 = slice_by_index(begin = var_32110_begin_0, end = var_32110_end_0, end_mask = var_32110_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32110_cast_fp16")]; + tensor var_32114_begin_0 = const()[name = tensor("op_32114_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_32114_end_0 = const()[name = tensor("op_32114_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_32114_end_mask_0 = const()[name = tensor("op_32114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32114_cast_fp16 = slice_by_index(begin = var_32114_begin_0, end = var_32114_end_0, end_mask = var_32114_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32114_cast_fp16")]; + tensor var_32118_begin_0 = const()[name = tensor("op_32118_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_32118_end_0 = const()[name = tensor("op_32118_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_32118_end_mask_0 = const()[name = tensor("op_32118_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32118_cast_fp16 = slice_by_index(begin = var_32118_begin_0, end = var_32118_end_0, end_mask = var_32118_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32118_cast_fp16")]; + tensor var_32122_begin_0 = const()[name = tensor("op_32122_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_32122_end_0 = const()[name = tensor("op_32122_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_32122_end_mask_0 = const()[name = tensor("op_32122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32122_cast_fp16 = slice_by_index(begin = var_32122_begin_0, end = var_32122_end_0, end_mask = var_32122_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32122_cast_fp16")]; + tensor var_32126_begin_0 = const()[name = tensor("op_32126_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_32126_end_0 = const()[name = tensor("op_32126_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_32126_end_mask_0 = const()[name = tensor("op_32126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32126_cast_fp16 = slice_by_index(begin = var_32126_begin_0, end = var_32126_end_0, end_mask = var_32126_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32126_cast_fp16")]; + tensor var_32130_begin_0 = const()[name = tensor("op_32130_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_32130_end_0 = const()[name = tensor("op_32130_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_32130_end_mask_0 = const()[name = tensor("op_32130_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32130_cast_fp16 = slice_by_index(begin = var_32130_begin_0, end = var_32130_end_0, end_mask = var_32130_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32130_cast_fp16")]; + tensor var_32134_begin_0 = const()[name = tensor("op_32134_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_32134_end_0 = const()[name = tensor("op_32134_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_32134_end_mask_0 = const()[name = tensor("op_32134_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32134_cast_fp16 = slice_by_index(begin = var_32134_begin_0, end = var_32134_end_0, end_mask = var_32134_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32134_cast_fp16")]; + tensor var_32138_begin_0 = const()[name = tensor("op_32138_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_32138_end_0 = const()[name = tensor("op_32138_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_32138_end_mask_0 = const()[name = tensor("op_32138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32138_cast_fp16 = slice_by_index(begin = var_32138_begin_0, end = var_32138_end_0, end_mask = var_32138_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32138_cast_fp16")]; + tensor var_32142_begin_0 = const()[name = tensor("op_32142_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_32142_end_0 = const()[name = tensor("op_32142_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_32142_end_mask_0 = const()[name = tensor("op_32142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32142_cast_fp16 = slice_by_index(begin = var_32142_begin_0, end = var_32142_end_0, end_mask = var_32142_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32142_cast_fp16")]; + tensor var_32146_begin_0 = const()[name = tensor("op_32146_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_32146_end_0 = const()[name = tensor("op_32146_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_32146_end_mask_0 = const()[name = tensor("op_32146_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32146_cast_fp16 = slice_by_index(begin = var_32146_begin_0, end = var_32146_end_0, end_mask = var_32146_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32146_cast_fp16")]; + tensor var_32150_begin_0 = const()[name = tensor("op_32150_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_32150_end_0 = const()[name = tensor("op_32150_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_32150_end_mask_0 = const()[name = tensor("op_32150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32150_cast_fp16 = slice_by_index(begin = var_32150_begin_0, end = var_32150_end_0, end_mask = var_32150_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32150_cast_fp16")]; + tensor var_32154_begin_0 = const()[name = tensor("op_32154_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_32154_end_0 = const()[name = tensor("op_32154_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_32154_end_mask_0 = const()[name = tensor("op_32154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32154_cast_fp16 = slice_by_index(begin = var_32154_begin_0, end = var_32154_end_0, end_mask = var_32154_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32154_cast_fp16")]; + tensor var_32158_begin_0 = const()[name = tensor("op_32158_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_32158_end_0 = const()[name = tensor("op_32158_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_32158_end_mask_0 = const()[name = tensor("op_32158_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32158_cast_fp16 = slice_by_index(begin = var_32158_begin_0, end = var_32158_end_0, end_mask = var_32158_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32158_cast_fp16")]; + tensor var_32162_begin_0 = const()[name = tensor("op_32162_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_32162_end_0 = const()[name = tensor("op_32162_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_32162_end_mask_0 = const()[name = tensor("op_32162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32162_cast_fp16 = slice_by_index(begin = var_32162_begin_0, end = var_32162_end_0, end_mask = var_32162_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32162_cast_fp16")]; + tensor var_32166_begin_0 = const()[name = tensor("op_32166_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_32166_end_0 = const()[name = tensor("op_32166_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_32166_end_mask_0 = const()[name = tensor("op_32166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32166_cast_fp16 = slice_by_index(begin = var_32166_begin_0, end = var_32166_end_0, end_mask = var_32166_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32166_cast_fp16")]; + tensor var_32170_begin_0 = const()[name = tensor("op_32170_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_32170_end_0 = const()[name = tensor("op_32170_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_32170_end_mask_0 = const()[name = tensor("op_32170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_32170_cast_fp16 = slice_by_index(begin = var_32170_begin_0, end = var_32170_end_0, end_mask = var_32170_end_mask_0, x = value_39_cast_fp16)[name = tensor("op_32170_cast_fp16")]; + tensor var_32174_equation_0 = const()[name = tensor("op_32174_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32174_cast_fp16 = einsum(equation = var_32174_equation_0, values = (var_32016_cast_fp16, var_31458_cast_fp16))[name = tensor("op_32174_cast_fp16")]; + tensor var_32175_to_fp16 = const()[name = tensor("op_32175_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3041_cast_fp16 = mul(x = var_32174_cast_fp16, y = var_32175_to_fp16)[name = tensor("aw_chunk_3041_cast_fp16")]; + tensor var_32178_equation_0 = const()[name = tensor("op_32178_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32178_cast_fp16 = einsum(equation = var_32178_equation_0, values = (var_32016_cast_fp16, var_31465_cast_fp16))[name = tensor("op_32178_cast_fp16")]; + tensor var_32179_to_fp16 = const()[name = tensor("op_32179_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3043_cast_fp16 = mul(x = var_32178_cast_fp16, y = var_32179_to_fp16)[name = tensor("aw_chunk_3043_cast_fp16")]; + tensor var_32182_equation_0 = const()[name = tensor("op_32182_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32182_cast_fp16 = einsum(equation = var_32182_equation_0, values = (var_32016_cast_fp16, var_31472_cast_fp16))[name = tensor("op_32182_cast_fp16")]; + tensor var_32183_to_fp16 = const()[name = tensor("op_32183_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3045_cast_fp16 = mul(x = var_32182_cast_fp16, y = var_32183_to_fp16)[name = tensor("aw_chunk_3045_cast_fp16")]; + tensor var_32186_equation_0 = const()[name = tensor("op_32186_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32186_cast_fp16 = einsum(equation = var_32186_equation_0, values = (var_32016_cast_fp16, var_31479_cast_fp16))[name = tensor("op_32186_cast_fp16")]; + tensor var_32187_to_fp16 = const()[name = tensor("op_32187_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3047_cast_fp16 = mul(x = var_32186_cast_fp16, y = var_32187_to_fp16)[name = tensor("aw_chunk_3047_cast_fp16")]; + tensor var_32190_equation_0 = const()[name = tensor("op_32190_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32190_cast_fp16 = einsum(equation = var_32190_equation_0, values = (var_32020_cast_fp16, var_31486_cast_fp16))[name = tensor("op_32190_cast_fp16")]; + tensor var_32191_to_fp16 = const()[name = tensor("op_32191_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3049_cast_fp16 = mul(x = var_32190_cast_fp16, y = var_32191_to_fp16)[name = tensor("aw_chunk_3049_cast_fp16")]; + tensor var_32194_equation_0 = const()[name = tensor("op_32194_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32194_cast_fp16 = einsum(equation = var_32194_equation_0, values = (var_32020_cast_fp16, var_31493_cast_fp16))[name = tensor("op_32194_cast_fp16")]; + tensor var_32195_to_fp16 = const()[name = tensor("op_32195_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3051_cast_fp16 = mul(x = var_32194_cast_fp16, y = var_32195_to_fp16)[name = tensor("aw_chunk_3051_cast_fp16")]; + tensor var_32198_equation_0 = const()[name = tensor("op_32198_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32198_cast_fp16 = einsum(equation = var_32198_equation_0, values = (var_32020_cast_fp16, var_31500_cast_fp16))[name = tensor("op_32198_cast_fp16")]; + tensor var_32199_to_fp16 = const()[name = tensor("op_32199_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3053_cast_fp16 = mul(x = var_32198_cast_fp16, y = var_32199_to_fp16)[name = tensor("aw_chunk_3053_cast_fp16")]; + tensor var_32202_equation_0 = const()[name = tensor("op_32202_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32202_cast_fp16 = einsum(equation = var_32202_equation_0, values = (var_32020_cast_fp16, var_31507_cast_fp16))[name = tensor("op_32202_cast_fp16")]; + tensor var_32203_to_fp16 = const()[name = tensor("op_32203_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3055_cast_fp16 = mul(x = var_32202_cast_fp16, y = var_32203_to_fp16)[name = tensor("aw_chunk_3055_cast_fp16")]; + tensor var_32206_equation_0 = const()[name = tensor("op_32206_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32206_cast_fp16 = einsum(equation = var_32206_equation_0, values = (var_32024_cast_fp16, var_31514_cast_fp16))[name = tensor("op_32206_cast_fp16")]; + tensor var_32207_to_fp16 = const()[name = tensor("op_32207_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3057_cast_fp16 = mul(x = var_32206_cast_fp16, y = var_32207_to_fp16)[name = tensor("aw_chunk_3057_cast_fp16")]; + tensor var_32210_equation_0 = const()[name = tensor("op_32210_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32210_cast_fp16 = einsum(equation = var_32210_equation_0, values = (var_32024_cast_fp16, var_31521_cast_fp16))[name = tensor("op_32210_cast_fp16")]; + tensor var_32211_to_fp16 = const()[name = tensor("op_32211_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3059_cast_fp16 = mul(x = var_32210_cast_fp16, y = var_32211_to_fp16)[name = tensor("aw_chunk_3059_cast_fp16")]; + tensor var_32214_equation_0 = const()[name = tensor("op_32214_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32214_cast_fp16 = einsum(equation = var_32214_equation_0, values = (var_32024_cast_fp16, var_31528_cast_fp16))[name = tensor("op_32214_cast_fp16")]; + tensor var_32215_to_fp16 = const()[name = tensor("op_32215_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3061_cast_fp16 = mul(x = var_32214_cast_fp16, y = var_32215_to_fp16)[name = tensor("aw_chunk_3061_cast_fp16")]; + tensor var_32218_equation_0 = const()[name = tensor("op_32218_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32218_cast_fp16 = einsum(equation = var_32218_equation_0, values = (var_32024_cast_fp16, var_31535_cast_fp16))[name = tensor("op_32218_cast_fp16")]; + tensor var_32219_to_fp16 = const()[name = tensor("op_32219_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3063_cast_fp16 = mul(x = var_32218_cast_fp16, y = var_32219_to_fp16)[name = tensor("aw_chunk_3063_cast_fp16")]; + tensor var_32222_equation_0 = const()[name = tensor("op_32222_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32222_cast_fp16 = einsum(equation = var_32222_equation_0, values = (var_32028_cast_fp16, var_31542_cast_fp16))[name = tensor("op_32222_cast_fp16")]; + tensor var_32223_to_fp16 = const()[name = tensor("op_32223_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3065_cast_fp16 = mul(x = var_32222_cast_fp16, y = var_32223_to_fp16)[name = tensor("aw_chunk_3065_cast_fp16")]; + tensor var_32226_equation_0 = const()[name = tensor("op_32226_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32226_cast_fp16 = einsum(equation = var_32226_equation_0, values = (var_32028_cast_fp16, var_31549_cast_fp16))[name = tensor("op_32226_cast_fp16")]; + tensor var_32227_to_fp16 = const()[name = tensor("op_32227_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3067_cast_fp16 = mul(x = var_32226_cast_fp16, y = var_32227_to_fp16)[name = tensor("aw_chunk_3067_cast_fp16")]; + tensor var_32230_equation_0 = const()[name = tensor("op_32230_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32230_cast_fp16 = einsum(equation = var_32230_equation_0, values = (var_32028_cast_fp16, var_31556_cast_fp16))[name = tensor("op_32230_cast_fp16")]; + tensor var_32231_to_fp16 = const()[name = tensor("op_32231_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3069_cast_fp16 = mul(x = var_32230_cast_fp16, y = var_32231_to_fp16)[name = tensor("aw_chunk_3069_cast_fp16")]; + tensor var_32234_equation_0 = const()[name = tensor("op_32234_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32234_cast_fp16 = einsum(equation = var_32234_equation_0, values = (var_32028_cast_fp16, var_31563_cast_fp16))[name = tensor("op_32234_cast_fp16")]; + tensor var_32235_to_fp16 = const()[name = tensor("op_32235_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3071_cast_fp16 = mul(x = var_32234_cast_fp16, y = var_32235_to_fp16)[name = tensor("aw_chunk_3071_cast_fp16")]; + tensor var_32238_equation_0 = const()[name = tensor("op_32238_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32238_cast_fp16 = einsum(equation = var_32238_equation_0, values = (var_32032_cast_fp16, var_31570_cast_fp16))[name = tensor("op_32238_cast_fp16")]; + tensor var_32239_to_fp16 = const()[name = tensor("op_32239_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3073_cast_fp16 = mul(x = var_32238_cast_fp16, y = var_32239_to_fp16)[name = tensor("aw_chunk_3073_cast_fp16")]; + tensor var_32242_equation_0 = const()[name = tensor("op_32242_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32242_cast_fp16 = einsum(equation = var_32242_equation_0, values = (var_32032_cast_fp16, var_31577_cast_fp16))[name = tensor("op_32242_cast_fp16")]; + tensor var_32243_to_fp16 = const()[name = tensor("op_32243_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3075_cast_fp16 = mul(x = var_32242_cast_fp16, y = var_32243_to_fp16)[name = tensor("aw_chunk_3075_cast_fp16")]; + tensor var_32246_equation_0 = const()[name = tensor("op_32246_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32246_cast_fp16 = einsum(equation = var_32246_equation_0, values = (var_32032_cast_fp16, var_31584_cast_fp16))[name = tensor("op_32246_cast_fp16")]; + tensor var_32247_to_fp16 = const()[name = tensor("op_32247_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3077_cast_fp16 = mul(x = var_32246_cast_fp16, y = var_32247_to_fp16)[name = tensor("aw_chunk_3077_cast_fp16")]; + tensor var_32250_equation_0 = const()[name = tensor("op_32250_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32250_cast_fp16 = einsum(equation = var_32250_equation_0, values = (var_32032_cast_fp16, var_31591_cast_fp16))[name = tensor("op_32250_cast_fp16")]; + tensor var_32251_to_fp16 = const()[name = tensor("op_32251_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3079_cast_fp16 = mul(x = var_32250_cast_fp16, y = var_32251_to_fp16)[name = tensor("aw_chunk_3079_cast_fp16")]; + tensor var_32254_equation_0 = const()[name = tensor("op_32254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32254_cast_fp16 = einsum(equation = var_32254_equation_0, values = (var_32036_cast_fp16, var_31598_cast_fp16))[name = tensor("op_32254_cast_fp16")]; + tensor var_32255_to_fp16 = const()[name = tensor("op_32255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3081_cast_fp16 = mul(x = var_32254_cast_fp16, y = var_32255_to_fp16)[name = tensor("aw_chunk_3081_cast_fp16")]; + tensor var_32258_equation_0 = const()[name = tensor("op_32258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32258_cast_fp16 = einsum(equation = var_32258_equation_0, values = (var_32036_cast_fp16, var_31605_cast_fp16))[name = tensor("op_32258_cast_fp16")]; + tensor var_32259_to_fp16 = const()[name = tensor("op_32259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3083_cast_fp16 = mul(x = var_32258_cast_fp16, y = var_32259_to_fp16)[name = tensor("aw_chunk_3083_cast_fp16")]; + tensor var_32262_equation_0 = const()[name = tensor("op_32262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32262_cast_fp16 = einsum(equation = var_32262_equation_0, values = (var_32036_cast_fp16, var_31612_cast_fp16))[name = tensor("op_32262_cast_fp16")]; + tensor var_32263_to_fp16 = const()[name = tensor("op_32263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3085_cast_fp16 = mul(x = var_32262_cast_fp16, y = var_32263_to_fp16)[name = tensor("aw_chunk_3085_cast_fp16")]; + tensor var_32266_equation_0 = const()[name = tensor("op_32266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32266_cast_fp16 = einsum(equation = var_32266_equation_0, values = (var_32036_cast_fp16, var_31619_cast_fp16))[name = tensor("op_32266_cast_fp16")]; + tensor var_32267_to_fp16 = const()[name = tensor("op_32267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3087_cast_fp16 = mul(x = var_32266_cast_fp16, y = var_32267_to_fp16)[name = tensor("aw_chunk_3087_cast_fp16")]; + tensor var_32270_equation_0 = const()[name = tensor("op_32270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32270_cast_fp16 = einsum(equation = var_32270_equation_0, values = (var_32040_cast_fp16, var_31626_cast_fp16))[name = tensor("op_32270_cast_fp16")]; + tensor var_32271_to_fp16 = const()[name = tensor("op_32271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3089_cast_fp16 = mul(x = var_32270_cast_fp16, y = var_32271_to_fp16)[name = tensor("aw_chunk_3089_cast_fp16")]; + tensor var_32274_equation_0 = const()[name = tensor("op_32274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32274_cast_fp16 = einsum(equation = var_32274_equation_0, values = (var_32040_cast_fp16, var_31633_cast_fp16))[name = tensor("op_32274_cast_fp16")]; + tensor var_32275_to_fp16 = const()[name = tensor("op_32275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3091_cast_fp16 = mul(x = var_32274_cast_fp16, y = var_32275_to_fp16)[name = tensor("aw_chunk_3091_cast_fp16")]; + tensor var_32278_equation_0 = const()[name = tensor("op_32278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32278_cast_fp16 = einsum(equation = var_32278_equation_0, values = (var_32040_cast_fp16, var_31640_cast_fp16))[name = tensor("op_32278_cast_fp16")]; + tensor var_32279_to_fp16 = const()[name = tensor("op_32279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3093_cast_fp16 = mul(x = var_32278_cast_fp16, y = var_32279_to_fp16)[name = tensor("aw_chunk_3093_cast_fp16")]; + tensor var_32282_equation_0 = const()[name = tensor("op_32282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32282_cast_fp16 = einsum(equation = var_32282_equation_0, values = (var_32040_cast_fp16, var_31647_cast_fp16))[name = tensor("op_32282_cast_fp16")]; + tensor var_32283_to_fp16 = const()[name = tensor("op_32283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3095_cast_fp16 = mul(x = var_32282_cast_fp16, y = var_32283_to_fp16)[name = tensor("aw_chunk_3095_cast_fp16")]; + tensor var_32286_equation_0 = const()[name = tensor("op_32286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32286_cast_fp16 = einsum(equation = var_32286_equation_0, values = (var_32044_cast_fp16, var_31654_cast_fp16))[name = tensor("op_32286_cast_fp16")]; + tensor var_32287_to_fp16 = const()[name = tensor("op_32287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3097_cast_fp16 = mul(x = var_32286_cast_fp16, y = var_32287_to_fp16)[name = tensor("aw_chunk_3097_cast_fp16")]; + tensor var_32290_equation_0 = const()[name = tensor("op_32290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32290_cast_fp16 = einsum(equation = var_32290_equation_0, values = (var_32044_cast_fp16, var_31661_cast_fp16))[name = tensor("op_32290_cast_fp16")]; + tensor var_32291_to_fp16 = const()[name = tensor("op_32291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3099_cast_fp16 = mul(x = var_32290_cast_fp16, y = var_32291_to_fp16)[name = tensor("aw_chunk_3099_cast_fp16")]; + tensor var_32294_equation_0 = const()[name = tensor("op_32294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32294_cast_fp16 = einsum(equation = var_32294_equation_0, values = (var_32044_cast_fp16, var_31668_cast_fp16))[name = tensor("op_32294_cast_fp16")]; + tensor var_32295_to_fp16 = const()[name = tensor("op_32295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3101_cast_fp16 = mul(x = var_32294_cast_fp16, y = var_32295_to_fp16)[name = tensor("aw_chunk_3101_cast_fp16")]; + tensor var_32298_equation_0 = const()[name = tensor("op_32298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32298_cast_fp16 = einsum(equation = var_32298_equation_0, values = (var_32044_cast_fp16, var_31675_cast_fp16))[name = tensor("op_32298_cast_fp16")]; + tensor var_32299_to_fp16 = const()[name = tensor("op_32299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3103_cast_fp16 = mul(x = var_32298_cast_fp16, y = var_32299_to_fp16)[name = tensor("aw_chunk_3103_cast_fp16")]; + tensor var_32302_equation_0 = const()[name = tensor("op_32302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32302_cast_fp16 = einsum(equation = var_32302_equation_0, values = (var_32048_cast_fp16, var_31682_cast_fp16))[name = tensor("op_32302_cast_fp16")]; + tensor var_32303_to_fp16 = const()[name = tensor("op_32303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3105_cast_fp16 = mul(x = var_32302_cast_fp16, y = var_32303_to_fp16)[name = tensor("aw_chunk_3105_cast_fp16")]; + tensor var_32306_equation_0 = const()[name = tensor("op_32306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32306_cast_fp16 = einsum(equation = var_32306_equation_0, values = (var_32048_cast_fp16, var_31689_cast_fp16))[name = tensor("op_32306_cast_fp16")]; + tensor var_32307_to_fp16 = const()[name = tensor("op_32307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3107_cast_fp16 = mul(x = var_32306_cast_fp16, y = var_32307_to_fp16)[name = tensor("aw_chunk_3107_cast_fp16")]; + tensor var_32310_equation_0 = const()[name = tensor("op_32310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32310_cast_fp16 = einsum(equation = var_32310_equation_0, values = (var_32048_cast_fp16, var_31696_cast_fp16))[name = tensor("op_32310_cast_fp16")]; + tensor var_32311_to_fp16 = const()[name = tensor("op_32311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3109_cast_fp16 = mul(x = var_32310_cast_fp16, y = var_32311_to_fp16)[name = tensor("aw_chunk_3109_cast_fp16")]; + tensor var_32314_equation_0 = const()[name = tensor("op_32314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32314_cast_fp16 = einsum(equation = var_32314_equation_0, values = (var_32048_cast_fp16, var_31703_cast_fp16))[name = tensor("op_32314_cast_fp16")]; + tensor var_32315_to_fp16 = const()[name = tensor("op_32315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3111_cast_fp16 = mul(x = var_32314_cast_fp16, y = var_32315_to_fp16)[name = tensor("aw_chunk_3111_cast_fp16")]; + tensor var_32318_equation_0 = const()[name = tensor("op_32318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32318_cast_fp16 = einsum(equation = var_32318_equation_0, values = (var_32052_cast_fp16, var_31710_cast_fp16))[name = tensor("op_32318_cast_fp16")]; + tensor var_32319_to_fp16 = const()[name = tensor("op_32319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3113_cast_fp16 = mul(x = var_32318_cast_fp16, y = var_32319_to_fp16)[name = tensor("aw_chunk_3113_cast_fp16")]; + tensor var_32322_equation_0 = const()[name = tensor("op_32322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32322_cast_fp16 = einsum(equation = var_32322_equation_0, values = (var_32052_cast_fp16, var_31717_cast_fp16))[name = tensor("op_32322_cast_fp16")]; + tensor var_32323_to_fp16 = const()[name = tensor("op_32323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3115_cast_fp16 = mul(x = var_32322_cast_fp16, y = var_32323_to_fp16)[name = tensor("aw_chunk_3115_cast_fp16")]; + tensor var_32326_equation_0 = const()[name = tensor("op_32326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32326_cast_fp16 = einsum(equation = var_32326_equation_0, values = (var_32052_cast_fp16, var_31724_cast_fp16))[name = tensor("op_32326_cast_fp16")]; + tensor var_32327_to_fp16 = const()[name = tensor("op_32327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3117_cast_fp16 = mul(x = var_32326_cast_fp16, y = var_32327_to_fp16)[name = tensor("aw_chunk_3117_cast_fp16")]; + tensor var_32330_equation_0 = const()[name = tensor("op_32330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32330_cast_fp16 = einsum(equation = var_32330_equation_0, values = (var_32052_cast_fp16, var_31731_cast_fp16))[name = tensor("op_32330_cast_fp16")]; + tensor var_32331_to_fp16 = const()[name = tensor("op_32331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3119_cast_fp16 = mul(x = var_32330_cast_fp16, y = var_32331_to_fp16)[name = tensor("aw_chunk_3119_cast_fp16")]; + tensor var_32334_equation_0 = const()[name = tensor("op_32334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32334_cast_fp16 = einsum(equation = var_32334_equation_0, values = (var_32056_cast_fp16, var_31738_cast_fp16))[name = tensor("op_32334_cast_fp16")]; + tensor var_32335_to_fp16 = const()[name = tensor("op_32335_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3121_cast_fp16 = mul(x = var_32334_cast_fp16, y = var_32335_to_fp16)[name = tensor("aw_chunk_3121_cast_fp16")]; + tensor var_32338_equation_0 = const()[name = tensor("op_32338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32338_cast_fp16 = einsum(equation = var_32338_equation_0, values = (var_32056_cast_fp16, var_31745_cast_fp16))[name = tensor("op_32338_cast_fp16")]; + tensor var_32339_to_fp16 = const()[name = tensor("op_32339_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3123_cast_fp16 = mul(x = var_32338_cast_fp16, y = var_32339_to_fp16)[name = tensor("aw_chunk_3123_cast_fp16")]; + tensor var_32342_equation_0 = const()[name = tensor("op_32342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32342_cast_fp16 = einsum(equation = var_32342_equation_0, values = (var_32056_cast_fp16, var_31752_cast_fp16))[name = tensor("op_32342_cast_fp16")]; + tensor var_32343_to_fp16 = const()[name = tensor("op_32343_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3125_cast_fp16 = mul(x = var_32342_cast_fp16, y = var_32343_to_fp16)[name = tensor("aw_chunk_3125_cast_fp16")]; + tensor var_32346_equation_0 = const()[name = tensor("op_32346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32346_cast_fp16 = einsum(equation = var_32346_equation_0, values = (var_32056_cast_fp16, var_31759_cast_fp16))[name = tensor("op_32346_cast_fp16")]; + tensor var_32347_to_fp16 = const()[name = tensor("op_32347_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3127_cast_fp16 = mul(x = var_32346_cast_fp16, y = var_32347_to_fp16)[name = tensor("aw_chunk_3127_cast_fp16")]; + tensor var_32350_equation_0 = const()[name = tensor("op_32350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32350_cast_fp16 = einsum(equation = var_32350_equation_0, values = (var_32060_cast_fp16, var_31766_cast_fp16))[name = tensor("op_32350_cast_fp16")]; + tensor var_32351_to_fp16 = const()[name = tensor("op_32351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3129_cast_fp16 = mul(x = var_32350_cast_fp16, y = var_32351_to_fp16)[name = tensor("aw_chunk_3129_cast_fp16")]; + tensor var_32354_equation_0 = const()[name = tensor("op_32354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32354_cast_fp16 = einsum(equation = var_32354_equation_0, values = (var_32060_cast_fp16, var_31773_cast_fp16))[name = tensor("op_32354_cast_fp16")]; + tensor var_32355_to_fp16 = const()[name = tensor("op_32355_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3131_cast_fp16 = mul(x = var_32354_cast_fp16, y = var_32355_to_fp16)[name = tensor("aw_chunk_3131_cast_fp16")]; + tensor var_32358_equation_0 = const()[name = tensor("op_32358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32358_cast_fp16 = einsum(equation = var_32358_equation_0, values = (var_32060_cast_fp16, var_31780_cast_fp16))[name = tensor("op_32358_cast_fp16")]; + tensor var_32359_to_fp16 = const()[name = tensor("op_32359_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3133_cast_fp16 = mul(x = var_32358_cast_fp16, y = var_32359_to_fp16)[name = tensor("aw_chunk_3133_cast_fp16")]; + tensor var_32362_equation_0 = const()[name = tensor("op_32362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32362_cast_fp16 = einsum(equation = var_32362_equation_0, values = (var_32060_cast_fp16, var_31787_cast_fp16))[name = tensor("op_32362_cast_fp16")]; + tensor var_32363_to_fp16 = const()[name = tensor("op_32363_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3135_cast_fp16 = mul(x = var_32362_cast_fp16, y = var_32363_to_fp16)[name = tensor("aw_chunk_3135_cast_fp16")]; + tensor var_32366_equation_0 = const()[name = tensor("op_32366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32366_cast_fp16 = einsum(equation = var_32366_equation_0, values = (var_32064_cast_fp16, var_31794_cast_fp16))[name = tensor("op_32366_cast_fp16")]; + tensor var_32367_to_fp16 = const()[name = tensor("op_32367_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3137_cast_fp16 = mul(x = var_32366_cast_fp16, y = var_32367_to_fp16)[name = tensor("aw_chunk_3137_cast_fp16")]; + tensor var_32370_equation_0 = const()[name = tensor("op_32370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32370_cast_fp16 = einsum(equation = var_32370_equation_0, values = (var_32064_cast_fp16, var_31801_cast_fp16))[name = tensor("op_32370_cast_fp16")]; + tensor var_32371_to_fp16 = const()[name = tensor("op_32371_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3139_cast_fp16 = mul(x = var_32370_cast_fp16, y = var_32371_to_fp16)[name = tensor("aw_chunk_3139_cast_fp16")]; + tensor var_32374_equation_0 = const()[name = tensor("op_32374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32374_cast_fp16 = einsum(equation = var_32374_equation_0, values = (var_32064_cast_fp16, var_31808_cast_fp16))[name = tensor("op_32374_cast_fp16")]; + tensor var_32375_to_fp16 = const()[name = tensor("op_32375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3141_cast_fp16 = mul(x = var_32374_cast_fp16, y = var_32375_to_fp16)[name = tensor("aw_chunk_3141_cast_fp16")]; + tensor var_32378_equation_0 = const()[name = tensor("op_32378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32378_cast_fp16 = einsum(equation = var_32378_equation_0, values = (var_32064_cast_fp16, var_31815_cast_fp16))[name = tensor("op_32378_cast_fp16")]; + tensor var_32379_to_fp16 = const()[name = tensor("op_32379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3143_cast_fp16 = mul(x = var_32378_cast_fp16, y = var_32379_to_fp16)[name = tensor("aw_chunk_3143_cast_fp16")]; + tensor var_32382_equation_0 = const()[name = tensor("op_32382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32382_cast_fp16 = einsum(equation = var_32382_equation_0, values = (var_32068_cast_fp16, var_31822_cast_fp16))[name = tensor("op_32382_cast_fp16")]; + tensor var_32383_to_fp16 = const()[name = tensor("op_32383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3145_cast_fp16 = mul(x = var_32382_cast_fp16, y = var_32383_to_fp16)[name = tensor("aw_chunk_3145_cast_fp16")]; + tensor var_32386_equation_0 = const()[name = tensor("op_32386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32386_cast_fp16 = einsum(equation = var_32386_equation_0, values = (var_32068_cast_fp16, var_31829_cast_fp16))[name = tensor("op_32386_cast_fp16")]; + tensor var_32387_to_fp16 = const()[name = tensor("op_32387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3147_cast_fp16 = mul(x = var_32386_cast_fp16, y = var_32387_to_fp16)[name = tensor("aw_chunk_3147_cast_fp16")]; + tensor var_32390_equation_0 = const()[name = tensor("op_32390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32390_cast_fp16 = einsum(equation = var_32390_equation_0, values = (var_32068_cast_fp16, var_31836_cast_fp16))[name = tensor("op_32390_cast_fp16")]; + tensor var_32391_to_fp16 = const()[name = tensor("op_32391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3149_cast_fp16 = mul(x = var_32390_cast_fp16, y = var_32391_to_fp16)[name = tensor("aw_chunk_3149_cast_fp16")]; + tensor var_32394_equation_0 = const()[name = tensor("op_32394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32394_cast_fp16 = einsum(equation = var_32394_equation_0, values = (var_32068_cast_fp16, var_31843_cast_fp16))[name = tensor("op_32394_cast_fp16")]; + tensor var_32395_to_fp16 = const()[name = tensor("op_32395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3151_cast_fp16 = mul(x = var_32394_cast_fp16, y = var_32395_to_fp16)[name = tensor("aw_chunk_3151_cast_fp16")]; + tensor var_32398_equation_0 = const()[name = tensor("op_32398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32398_cast_fp16 = einsum(equation = var_32398_equation_0, values = (var_32072_cast_fp16, var_31850_cast_fp16))[name = tensor("op_32398_cast_fp16")]; + tensor var_32399_to_fp16 = const()[name = tensor("op_32399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3153_cast_fp16 = mul(x = var_32398_cast_fp16, y = var_32399_to_fp16)[name = tensor("aw_chunk_3153_cast_fp16")]; + tensor var_32402_equation_0 = const()[name = tensor("op_32402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32402_cast_fp16 = einsum(equation = var_32402_equation_0, values = (var_32072_cast_fp16, var_31857_cast_fp16))[name = tensor("op_32402_cast_fp16")]; + tensor var_32403_to_fp16 = const()[name = tensor("op_32403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3155_cast_fp16 = mul(x = var_32402_cast_fp16, y = var_32403_to_fp16)[name = tensor("aw_chunk_3155_cast_fp16")]; + tensor var_32406_equation_0 = const()[name = tensor("op_32406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32406_cast_fp16 = einsum(equation = var_32406_equation_0, values = (var_32072_cast_fp16, var_31864_cast_fp16))[name = tensor("op_32406_cast_fp16")]; + tensor var_32407_to_fp16 = const()[name = tensor("op_32407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3157_cast_fp16 = mul(x = var_32406_cast_fp16, y = var_32407_to_fp16)[name = tensor("aw_chunk_3157_cast_fp16")]; + tensor var_32410_equation_0 = const()[name = tensor("op_32410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32410_cast_fp16 = einsum(equation = var_32410_equation_0, values = (var_32072_cast_fp16, var_31871_cast_fp16))[name = tensor("op_32410_cast_fp16")]; + tensor var_32411_to_fp16 = const()[name = tensor("op_32411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3159_cast_fp16 = mul(x = var_32410_cast_fp16, y = var_32411_to_fp16)[name = tensor("aw_chunk_3159_cast_fp16")]; + tensor var_32414_equation_0 = const()[name = tensor("op_32414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32414_cast_fp16 = einsum(equation = var_32414_equation_0, values = (var_32076_cast_fp16, var_31878_cast_fp16))[name = tensor("op_32414_cast_fp16")]; + tensor var_32415_to_fp16 = const()[name = tensor("op_32415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3161_cast_fp16 = mul(x = var_32414_cast_fp16, y = var_32415_to_fp16)[name = tensor("aw_chunk_3161_cast_fp16")]; + tensor var_32418_equation_0 = const()[name = tensor("op_32418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32418_cast_fp16 = einsum(equation = var_32418_equation_0, values = (var_32076_cast_fp16, var_31885_cast_fp16))[name = tensor("op_32418_cast_fp16")]; + tensor var_32419_to_fp16 = const()[name = tensor("op_32419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3163_cast_fp16 = mul(x = var_32418_cast_fp16, y = var_32419_to_fp16)[name = tensor("aw_chunk_3163_cast_fp16")]; + tensor var_32422_equation_0 = const()[name = tensor("op_32422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32422_cast_fp16 = einsum(equation = var_32422_equation_0, values = (var_32076_cast_fp16, var_31892_cast_fp16))[name = tensor("op_32422_cast_fp16")]; + tensor var_32423_to_fp16 = const()[name = tensor("op_32423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3165_cast_fp16 = mul(x = var_32422_cast_fp16, y = var_32423_to_fp16)[name = tensor("aw_chunk_3165_cast_fp16")]; + tensor var_32426_equation_0 = const()[name = tensor("op_32426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32426_cast_fp16 = einsum(equation = var_32426_equation_0, values = (var_32076_cast_fp16, var_31899_cast_fp16))[name = tensor("op_32426_cast_fp16")]; + tensor var_32427_to_fp16 = const()[name = tensor("op_32427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3167_cast_fp16 = mul(x = var_32426_cast_fp16, y = var_32427_to_fp16)[name = tensor("aw_chunk_3167_cast_fp16")]; + tensor var_32430_equation_0 = const()[name = tensor("op_32430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32430_cast_fp16 = einsum(equation = var_32430_equation_0, values = (var_32080_cast_fp16, var_31906_cast_fp16))[name = tensor("op_32430_cast_fp16")]; + tensor var_32431_to_fp16 = const()[name = tensor("op_32431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3169_cast_fp16 = mul(x = var_32430_cast_fp16, y = var_32431_to_fp16)[name = tensor("aw_chunk_3169_cast_fp16")]; + tensor var_32434_equation_0 = const()[name = tensor("op_32434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32434_cast_fp16 = einsum(equation = var_32434_equation_0, values = (var_32080_cast_fp16, var_31913_cast_fp16))[name = tensor("op_32434_cast_fp16")]; + tensor var_32435_to_fp16 = const()[name = tensor("op_32435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3171_cast_fp16 = mul(x = var_32434_cast_fp16, y = var_32435_to_fp16)[name = tensor("aw_chunk_3171_cast_fp16")]; + tensor var_32438_equation_0 = const()[name = tensor("op_32438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32438_cast_fp16 = einsum(equation = var_32438_equation_0, values = (var_32080_cast_fp16, var_31920_cast_fp16))[name = tensor("op_32438_cast_fp16")]; + tensor var_32439_to_fp16 = const()[name = tensor("op_32439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3173_cast_fp16 = mul(x = var_32438_cast_fp16, y = var_32439_to_fp16)[name = tensor("aw_chunk_3173_cast_fp16")]; + tensor var_32442_equation_0 = const()[name = tensor("op_32442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32442_cast_fp16 = einsum(equation = var_32442_equation_0, values = (var_32080_cast_fp16, var_31927_cast_fp16))[name = tensor("op_32442_cast_fp16")]; + tensor var_32443_to_fp16 = const()[name = tensor("op_32443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3175_cast_fp16 = mul(x = var_32442_cast_fp16, y = var_32443_to_fp16)[name = tensor("aw_chunk_3175_cast_fp16")]; + tensor var_32446_equation_0 = const()[name = tensor("op_32446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32446_cast_fp16 = einsum(equation = var_32446_equation_0, values = (var_32084_cast_fp16, var_31934_cast_fp16))[name = tensor("op_32446_cast_fp16")]; + tensor var_32447_to_fp16 = const()[name = tensor("op_32447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3177_cast_fp16 = mul(x = var_32446_cast_fp16, y = var_32447_to_fp16)[name = tensor("aw_chunk_3177_cast_fp16")]; + tensor var_32450_equation_0 = const()[name = tensor("op_32450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32450_cast_fp16 = einsum(equation = var_32450_equation_0, values = (var_32084_cast_fp16, var_31941_cast_fp16))[name = tensor("op_32450_cast_fp16")]; + tensor var_32451_to_fp16 = const()[name = tensor("op_32451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3179_cast_fp16 = mul(x = var_32450_cast_fp16, y = var_32451_to_fp16)[name = tensor("aw_chunk_3179_cast_fp16")]; + tensor var_32454_equation_0 = const()[name = tensor("op_32454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32454_cast_fp16 = einsum(equation = var_32454_equation_0, values = (var_32084_cast_fp16, var_31948_cast_fp16))[name = tensor("op_32454_cast_fp16")]; + tensor var_32455_to_fp16 = const()[name = tensor("op_32455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3181_cast_fp16 = mul(x = var_32454_cast_fp16, y = var_32455_to_fp16)[name = tensor("aw_chunk_3181_cast_fp16")]; + tensor var_32458_equation_0 = const()[name = tensor("op_32458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32458_cast_fp16 = einsum(equation = var_32458_equation_0, values = (var_32084_cast_fp16, var_31955_cast_fp16))[name = tensor("op_32458_cast_fp16")]; + tensor var_32459_to_fp16 = const()[name = tensor("op_32459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3183_cast_fp16 = mul(x = var_32458_cast_fp16, y = var_32459_to_fp16)[name = tensor("aw_chunk_3183_cast_fp16")]; + tensor var_32462_equation_0 = const()[name = tensor("op_32462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32462_cast_fp16 = einsum(equation = var_32462_equation_0, values = (var_32088_cast_fp16, var_31962_cast_fp16))[name = tensor("op_32462_cast_fp16")]; + tensor var_32463_to_fp16 = const()[name = tensor("op_32463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3185_cast_fp16 = mul(x = var_32462_cast_fp16, y = var_32463_to_fp16)[name = tensor("aw_chunk_3185_cast_fp16")]; + tensor var_32466_equation_0 = const()[name = tensor("op_32466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32466_cast_fp16 = einsum(equation = var_32466_equation_0, values = (var_32088_cast_fp16, var_31969_cast_fp16))[name = tensor("op_32466_cast_fp16")]; + tensor var_32467_to_fp16 = const()[name = tensor("op_32467_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3187_cast_fp16 = mul(x = var_32466_cast_fp16, y = var_32467_to_fp16)[name = tensor("aw_chunk_3187_cast_fp16")]; + tensor var_32470_equation_0 = const()[name = tensor("op_32470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32470_cast_fp16 = einsum(equation = var_32470_equation_0, values = (var_32088_cast_fp16, var_31976_cast_fp16))[name = tensor("op_32470_cast_fp16")]; + tensor var_32471_to_fp16 = const()[name = tensor("op_32471_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3189_cast_fp16 = mul(x = var_32470_cast_fp16, y = var_32471_to_fp16)[name = tensor("aw_chunk_3189_cast_fp16")]; + tensor var_32474_equation_0 = const()[name = tensor("op_32474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32474_cast_fp16 = einsum(equation = var_32474_equation_0, values = (var_32088_cast_fp16, var_31983_cast_fp16))[name = tensor("op_32474_cast_fp16")]; + tensor var_32475_to_fp16 = const()[name = tensor("op_32475_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3191_cast_fp16 = mul(x = var_32474_cast_fp16, y = var_32475_to_fp16)[name = tensor("aw_chunk_3191_cast_fp16")]; + tensor var_32478_equation_0 = const()[name = tensor("op_32478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32478_cast_fp16 = einsum(equation = var_32478_equation_0, values = (var_32092_cast_fp16, var_31990_cast_fp16))[name = tensor("op_32478_cast_fp16")]; + tensor var_32479_to_fp16 = const()[name = tensor("op_32479_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3193_cast_fp16 = mul(x = var_32478_cast_fp16, y = var_32479_to_fp16)[name = tensor("aw_chunk_3193_cast_fp16")]; + tensor var_32482_equation_0 = const()[name = tensor("op_32482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32482_cast_fp16 = einsum(equation = var_32482_equation_0, values = (var_32092_cast_fp16, var_31997_cast_fp16))[name = tensor("op_32482_cast_fp16")]; + tensor var_32483_to_fp16 = const()[name = tensor("op_32483_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3195_cast_fp16 = mul(x = var_32482_cast_fp16, y = var_32483_to_fp16)[name = tensor("aw_chunk_3195_cast_fp16")]; + tensor var_32486_equation_0 = const()[name = tensor("op_32486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32486_cast_fp16 = einsum(equation = var_32486_equation_0, values = (var_32092_cast_fp16, var_32004_cast_fp16))[name = tensor("op_32486_cast_fp16")]; + tensor var_32487_to_fp16 = const()[name = tensor("op_32487_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3197_cast_fp16 = mul(x = var_32486_cast_fp16, y = var_32487_to_fp16)[name = tensor("aw_chunk_3197_cast_fp16")]; + tensor var_32490_equation_0 = const()[name = tensor("op_32490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_32490_cast_fp16 = einsum(equation = var_32490_equation_0, values = (var_32092_cast_fp16, var_32011_cast_fp16))[name = tensor("op_32490_cast_fp16")]; + tensor var_32491_to_fp16 = const()[name = tensor("op_32491_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3199_cast_fp16 = mul(x = var_32490_cast_fp16, y = var_32491_to_fp16)[name = tensor("aw_chunk_3199_cast_fp16")]; + tensor var_32493_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3041_cast_fp16)[name = tensor("op_32493_cast_fp16")]; + tensor var_32494_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3043_cast_fp16)[name = tensor("op_32494_cast_fp16")]; + tensor var_32495_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3045_cast_fp16)[name = tensor("op_32495_cast_fp16")]; + tensor var_32496_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3047_cast_fp16)[name = tensor("op_32496_cast_fp16")]; + tensor var_32497_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3049_cast_fp16)[name = tensor("op_32497_cast_fp16")]; + tensor var_32498_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3051_cast_fp16)[name = tensor("op_32498_cast_fp16")]; + tensor var_32499_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3053_cast_fp16)[name = tensor("op_32499_cast_fp16")]; + tensor var_32500_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3055_cast_fp16)[name = tensor("op_32500_cast_fp16")]; + tensor var_32501_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3057_cast_fp16)[name = tensor("op_32501_cast_fp16")]; + tensor var_32502_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3059_cast_fp16)[name = tensor("op_32502_cast_fp16")]; + tensor var_32503_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3061_cast_fp16)[name = tensor("op_32503_cast_fp16")]; + tensor var_32504_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3063_cast_fp16)[name = tensor("op_32504_cast_fp16")]; + tensor var_32505_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3065_cast_fp16)[name = tensor("op_32505_cast_fp16")]; + tensor var_32506_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3067_cast_fp16)[name = tensor("op_32506_cast_fp16")]; + tensor var_32507_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3069_cast_fp16)[name = tensor("op_32507_cast_fp16")]; + tensor var_32508_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3071_cast_fp16)[name = tensor("op_32508_cast_fp16")]; + tensor var_32509_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3073_cast_fp16)[name = tensor("op_32509_cast_fp16")]; + tensor var_32510_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3075_cast_fp16)[name = tensor("op_32510_cast_fp16")]; + tensor var_32511_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3077_cast_fp16)[name = tensor("op_32511_cast_fp16")]; + tensor var_32512_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3079_cast_fp16)[name = tensor("op_32512_cast_fp16")]; + tensor var_32513_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3081_cast_fp16)[name = tensor("op_32513_cast_fp16")]; + tensor var_32514_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3083_cast_fp16)[name = tensor("op_32514_cast_fp16")]; + tensor var_32515_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3085_cast_fp16)[name = tensor("op_32515_cast_fp16")]; + tensor var_32516_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3087_cast_fp16)[name = tensor("op_32516_cast_fp16")]; + tensor var_32517_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3089_cast_fp16)[name = tensor("op_32517_cast_fp16")]; + tensor var_32518_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3091_cast_fp16)[name = tensor("op_32518_cast_fp16")]; + tensor var_32519_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3093_cast_fp16)[name = tensor("op_32519_cast_fp16")]; + tensor var_32520_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3095_cast_fp16)[name = tensor("op_32520_cast_fp16")]; + tensor var_32521_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3097_cast_fp16)[name = tensor("op_32521_cast_fp16")]; + tensor var_32522_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3099_cast_fp16)[name = tensor("op_32522_cast_fp16")]; + tensor var_32523_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3101_cast_fp16)[name = tensor("op_32523_cast_fp16")]; + tensor var_32524_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3103_cast_fp16)[name = tensor("op_32524_cast_fp16")]; + tensor var_32525_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3105_cast_fp16)[name = tensor("op_32525_cast_fp16")]; + tensor var_32526_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3107_cast_fp16)[name = tensor("op_32526_cast_fp16")]; + tensor var_32527_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3109_cast_fp16)[name = tensor("op_32527_cast_fp16")]; + tensor var_32528_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3111_cast_fp16)[name = tensor("op_32528_cast_fp16")]; + tensor var_32529_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3113_cast_fp16)[name = tensor("op_32529_cast_fp16")]; + tensor var_32530_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3115_cast_fp16)[name = tensor("op_32530_cast_fp16")]; + tensor var_32531_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3117_cast_fp16)[name = tensor("op_32531_cast_fp16")]; + tensor var_32532_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3119_cast_fp16)[name = tensor("op_32532_cast_fp16")]; + tensor var_32533_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3121_cast_fp16)[name = tensor("op_32533_cast_fp16")]; + tensor var_32534_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3123_cast_fp16)[name = tensor("op_32534_cast_fp16")]; + tensor var_32535_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3125_cast_fp16)[name = tensor("op_32535_cast_fp16")]; + tensor var_32536_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3127_cast_fp16)[name = tensor("op_32536_cast_fp16")]; + tensor var_32537_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3129_cast_fp16)[name = tensor("op_32537_cast_fp16")]; + tensor var_32538_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3131_cast_fp16)[name = tensor("op_32538_cast_fp16")]; + tensor var_32539_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3133_cast_fp16)[name = tensor("op_32539_cast_fp16")]; + tensor var_32540_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3135_cast_fp16)[name = tensor("op_32540_cast_fp16")]; + tensor var_32541_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3137_cast_fp16)[name = tensor("op_32541_cast_fp16")]; + tensor var_32542_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3139_cast_fp16)[name = tensor("op_32542_cast_fp16")]; + tensor var_32543_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3141_cast_fp16)[name = tensor("op_32543_cast_fp16")]; + tensor var_32544_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3143_cast_fp16)[name = tensor("op_32544_cast_fp16")]; + tensor var_32545_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3145_cast_fp16)[name = tensor("op_32545_cast_fp16")]; + tensor var_32546_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3147_cast_fp16)[name = tensor("op_32546_cast_fp16")]; + tensor var_32547_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3149_cast_fp16)[name = tensor("op_32547_cast_fp16")]; + tensor var_32548_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3151_cast_fp16)[name = tensor("op_32548_cast_fp16")]; + tensor var_32549_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3153_cast_fp16)[name = tensor("op_32549_cast_fp16")]; + tensor var_32550_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3155_cast_fp16)[name = tensor("op_32550_cast_fp16")]; + tensor var_32551_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3157_cast_fp16)[name = tensor("op_32551_cast_fp16")]; + tensor var_32552_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3159_cast_fp16)[name = tensor("op_32552_cast_fp16")]; + tensor var_32553_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3161_cast_fp16)[name = tensor("op_32553_cast_fp16")]; + tensor var_32554_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3163_cast_fp16)[name = tensor("op_32554_cast_fp16")]; + tensor var_32555_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3165_cast_fp16)[name = tensor("op_32555_cast_fp16")]; + tensor var_32556_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3167_cast_fp16)[name = tensor("op_32556_cast_fp16")]; + tensor var_32557_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3169_cast_fp16)[name = tensor("op_32557_cast_fp16")]; + tensor var_32558_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3171_cast_fp16)[name = tensor("op_32558_cast_fp16")]; + tensor var_32559_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3173_cast_fp16)[name = tensor("op_32559_cast_fp16")]; + tensor var_32560_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3175_cast_fp16)[name = tensor("op_32560_cast_fp16")]; + tensor var_32561_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3177_cast_fp16)[name = tensor("op_32561_cast_fp16")]; + tensor var_32562_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3179_cast_fp16)[name = tensor("op_32562_cast_fp16")]; + tensor var_32563_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3181_cast_fp16)[name = tensor("op_32563_cast_fp16")]; + tensor var_32564_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3183_cast_fp16)[name = tensor("op_32564_cast_fp16")]; + tensor var_32565_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3185_cast_fp16)[name = tensor("op_32565_cast_fp16")]; + tensor var_32566_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3187_cast_fp16)[name = tensor("op_32566_cast_fp16")]; + tensor var_32567_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3189_cast_fp16)[name = tensor("op_32567_cast_fp16")]; + tensor var_32568_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3191_cast_fp16)[name = tensor("op_32568_cast_fp16")]; + tensor var_32569_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3193_cast_fp16)[name = tensor("op_32569_cast_fp16")]; + tensor var_32570_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3195_cast_fp16)[name = tensor("op_32570_cast_fp16")]; + tensor var_32571_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3197_cast_fp16)[name = tensor("op_32571_cast_fp16")]; + tensor var_32572_cast_fp16 = softmax(axis = var_31264, x = aw_chunk_3199_cast_fp16)[name = tensor("op_32572_cast_fp16")]; + tensor var_32574_equation_0 = const()[name = tensor("op_32574_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32574_cast_fp16 = einsum(equation = var_32574_equation_0, values = (var_32094_cast_fp16, var_32493_cast_fp16))[name = tensor("op_32574_cast_fp16")]; + tensor var_32576_equation_0 = const()[name = tensor("op_32576_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32576_cast_fp16 = einsum(equation = var_32576_equation_0, values = (var_32094_cast_fp16, var_32494_cast_fp16))[name = tensor("op_32576_cast_fp16")]; + tensor var_32578_equation_0 = const()[name = tensor("op_32578_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32578_cast_fp16 = einsum(equation = var_32578_equation_0, values = (var_32094_cast_fp16, var_32495_cast_fp16))[name = tensor("op_32578_cast_fp16")]; + tensor var_32580_equation_0 = const()[name = tensor("op_32580_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32580_cast_fp16 = einsum(equation = var_32580_equation_0, values = (var_32094_cast_fp16, var_32496_cast_fp16))[name = tensor("op_32580_cast_fp16")]; + tensor var_32582_equation_0 = const()[name = tensor("op_32582_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32582_cast_fp16 = einsum(equation = var_32582_equation_0, values = (var_32098_cast_fp16, var_32497_cast_fp16))[name = tensor("op_32582_cast_fp16")]; + tensor var_32584_equation_0 = const()[name = tensor("op_32584_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32584_cast_fp16 = einsum(equation = var_32584_equation_0, values = (var_32098_cast_fp16, var_32498_cast_fp16))[name = tensor("op_32584_cast_fp16")]; + tensor var_32586_equation_0 = const()[name = tensor("op_32586_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32586_cast_fp16 = einsum(equation = var_32586_equation_0, values = (var_32098_cast_fp16, var_32499_cast_fp16))[name = tensor("op_32586_cast_fp16")]; + tensor var_32588_equation_0 = const()[name = tensor("op_32588_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32588_cast_fp16 = einsum(equation = var_32588_equation_0, values = (var_32098_cast_fp16, var_32500_cast_fp16))[name = tensor("op_32588_cast_fp16")]; + tensor var_32590_equation_0 = const()[name = tensor("op_32590_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32590_cast_fp16 = einsum(equation = var_32590_equation_0, values = (var_32102_cast_fp16, var_32501_cast_fp16))[name = tensor("op_32590_cast_fp16")]; + tensor var_32592_equation_0 = const()[name = tensor("op_32592_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32592_cast_fp16 = einsum(equation = var_32592_equation_0, values = (var_32102_cast_fp16, var_32502_cast_fp16))[name = tensor("op_32592_cast_fp16")]; + tensor var_32594_equation_0 = const()[name = tensor("op_32594_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32594_cast_fp16 = einsum(equation = var_32594_equation_0, values = (var_32102_cast_fp16, var_32503_cast_fp16))[name = tensor("op_32594_cast_fp16")]; + tensor var_32596_equation_0 = const()[name = tensor("op_32596_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32596_cast_fp16 = einsum(equation = var_32596_equation_0, values = (var_32102_cast_fp16, var_32504_cast_fp16))[name = tensor("op_32596_cast_fp16")]; + tensor var_32598_equation_0 = const()[name = tensor("op_32598_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32598_cast_fp16 = einsum(equation = var_32598_equation_0, values = (var_32106_cast_fp16, var_32505_cast_fp16))[name = tensor("op_32598_cast_fp16")]; + tensor var_32600_equation_0 = const()[name = tensor("op_32600_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32600_cast_fp16 = einsum(equation = var_32600_equation_0, values = (var_32106_cast_fp16, var_32506_cast_fp16))[name = tensor("op_32600_cast_fp16")]; + tensor var_32602_equation_0 = const()[name = tensor("op_32602_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32602_cast_fp16 = einsum(equation = var_32602_equation_0, values = (var_32106_cast_fp16, var_32507_cast_fp16))[name = tensor("op_32602_cast_fp16")]; + tensor var_32604_equation_0 = const()[name = tensor("op_32604_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32604_cast_fp16 = einsum(equation = var_32604_equation_0, values = (var_32106_cast_fp16, var_32508_cast_fp16))[name = tensor("op_32604_cast_fp16")]; + tensor var_32606_equation_0 = const()[name = tensor("op_32606_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32606_cast_fp16 = einsum(equation = var_32606_equation_0, values = (var_32110_cast_fp16, var_32509_cast_fp16))[name = tensor("op_32606_cast_fp16")]; + tensor var_32608_equation_0 = const()[name = tensor("op_32608_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32608_cast_fp16 = einsum(equation = var_32608_equation_0, values = (var_32110_cast_fp16, var_32510_cast_fp16))[name = tensor("op_32608_cast_fp16")]; + tensor var_32610_equation_0 = const()[name = tensor("op_32610_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32610_cast_fp16 = einsum(equation = var_32610_equation_0, values = (var_32110_cast_fp16, var_32511_cast_fp16))[name = tensor("op_32610_cast_fp16")]; + tensor var_32612_equation_0 = const()[name = tensor("op_32612_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32612_cast_fp16 = einsum(equation = var_32612_equation_0, values = (var_32110_cast_fp16, var_32512_cast_fp16))[name = tensor("op_32612_cast_fp16")]; + tensor var_32614_equation_0 = const()[name = tensor("op_32614_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32614_cast_fp16 = einsum(equation = var_32614_equation_0, values = (var_32114_cast_fp16, var_32513_cast_fp16))[name = tensor("op_32614_cast_fp16")]; + tensor var_32616_equation_0 = const()[name = tensor("op_32616_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32616_cast_fp16 = einsum(equation = var_32616_equation_0, values = (var_32114_cast_fp16, var_32514_cast_fp16))[name = tensor("op_32616_cast_fp16")]; + tensor var_32618_equation_0 = const()[name = tensor("op_32618_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32618_cast_fp16 = einsum(equation = var_32618_equation_0, values = (var_32114_cast_fp16, var_32515_cast_fp16))[name = tensor("op_32618_cast_fp16")]; + tensor var_32620_equation_0 = const()[name = tensor("op_32620_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32620_cast_fp16 = einsum(equation = var_32620_equation_0, values = (var_32114_cast_fp16, var_32516_cast_fp16))[name = tensor("op_32620_cast_fp16")]; + tensor var_32622_equation_0 = const()[name = tensor("op_32622_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32622_cast_fp16 = einsum(equation = var_32622_equation_0, values = (var_32118_cast_fp16, var_32517_cast_fp16))[name = tensor("op_32622_cast_fp16")]; + tensor var_32624_equation_0 = const()[name = tensor("op_32624_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32624_cast_fp16 = einsum(equation = var_32624_equation_0, values = (var_32118_cast_fp16, var_32518_cast_fp16))[name = tensor("op_32624_cast_fp16")]; + tensor var_32626_equation_0 = const()[name = tensor("op_32626_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32626_cast_fp16 = einsum(equation = var_32626_equation_0, values = (var_32118_cast_fp16, var_32519_cast_fp16))[name = tensor("op_32626_cast_fp16")]; + tensor var_32628_equation_0 = const()[name = tensor("op_32628_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32628_cast_fp16 = einsum(equation = var_32628_equation_0, values = (var_32118_cast_fp16, var_32520_cast_fp16))[name = tensor("op_32628_cast_fp16")]; + tensor var_32630_equation_0 = const()[name = tensor("op_32630_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32630_cast_fp16 = einsum(equation = var_32630_equation_0, values = (var_32122_cast_fp16, var_32521_cast_fp16))[name = tensor("op_32630_cast_fp16")]; + tensor var_32632_equation_0 = const()[name = tensor("op_32632_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32632_cast_fp16 = einsum(equation = var_32632_equation_0, values = (var_32122_cast_fp16, var_32522_cast_fp16))[name = tensor("op_32632_cast_fp16")]; + tensor var_32634_equation_0 = const()[name = tensor("op_32634_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32634_cast_fp16 = einsum(equation = var_32634_equation_0, values = (var_32122_cast_fp16, var_32523_cast_fp16))[name = tensor("op_32634_cast_fp16")]; + tensor var_32636_equation_0 = const()[name = tensor("op_32636_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32636_cast_fp16 = einsum(equation = var_32636_equation_0, values = (var_32122_cast_fp16, var_32524_cast_fp16))[name = tensor("op_32636_cast_fp16")]; + tensor var_32638_equation_0 = const()[name = tensor("op_32638_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32638_cast_fp16 = einsum(equation = var_32638_equation_0, values = (var_32126_cast_fp16, var_32525_cast_fp16))[name = tensor("op_32638_cast_fp16")]; + tensor var_32640_equation_0 = const()[name = tensor("op_32640_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32640_cast_fp16 = einsum(equation = var_32640_equation_0, values = (var_32126_cast_fp16, var_32526_cast_fp16))[name = tensor("op_32640_cast_fp16")]; + tensor var_32642_equation_0 = const()[name = tensor("op_32642_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32642_cast_fp16 = einsum(equation = var_32642_equation_0, values = (var_32126_cast_fp16, var_32527_cast_fp16))[name = tensor("op_32642_cast_fp16")]; + tensor var_32644_equation_0 = const()[name = tensor("op_32644_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32644_cast_fp16 = einsum(equation = var_32644_equation_0, values = (var_32126_cast_fp16, var_32528_cast_fp16))[name = tensor("op_32644_cast_fp16")]; + tensor var_32646_equation_0 = const()[name = tensor("op_32646_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32646_cast_fp16 = einsum(equation = var_32646_equation_0, values = (var_32130_cast_fp16, var_32529_cast_fp16))[name = tensor("op_32646_cast_fp16")]; + tensor var_32648_equation_0 = const()[name = tensor("op_32648_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32648_cast_fp16 = einsum(equation = var_32648_equation_0, values = (var_32130_cast_fp16, var_32530_cast_fp16))[name = tensor("op_32648_cast_fp16")]; + tensor var_32650_equation_0 = const()[name = tensor("op_32650_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32650_cast_fp16 = einsum(equation = var_32650_equation_0, values = (var_32130_cast_fp16, var_32531_cast_fp16))[name = tensor("op_32650_cast_fp16")]; + tensor var_32652_equation_0 = const()[name = tensor("op_32652_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32652_cast_fp16 = einsum(equation = var_32652_equation_0, values = (var_32130_cast_fp16, var_32532_cast_fp16))[name = tensor("op_32652_cast_fp16")]; + tensor var_32654_equation_0 = const()[name = tensor("op_32654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32654_cast_fp16 = einsum(equation = var_32654_equation_0, values = (var_32134_cast_fp16, var_32533_cast_fp16))[name = tensor("op_32654_cast_fp16")]; + tensor var_32656_equation_0 = const()[name = tensor("op_32656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32656_cast_fp16 = einsum(equation = var_32656_equation_0, values = (var_32134_cast_fp16, var_32534_cast_fp16))[name = tensor("op_32656_cast_fp16")]; + tensor var_32658_equation_0 = const()[name = tensor("op_32658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32658_cast_fp16 = einsum(equation = var_32658_equation_0, values = (var_32134_cast_fp16, var_32535_cast_fp16))[name = tensor("op_32658_cast_fp16")]; + tensor var_32660_equation_0 = const()[name = tensor("op_32660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32660_cast_fp16 = einsum(equation = var_32660_equation_0, values = (var_32134_cast_fp16, var_32536_cast_fp16))[name = tensor("op_32660_cast_fp16")]; + tensor var_32662_equation_0 = const()[name = tensor("op_32662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32662_cast_fp16 = einsum(equation = var_32662_equation_0, values = (var_32138_cast_fp16, var_32537_cast_fp16))[name = tensor("op_32662_cast_fp16")]; + tensor var_32664_equation_0 = const()[name = tensor("op_32664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32664_cast_fp16 = einsum(equation = var_32664_equation_0, values = (var_32138_cast_fp16, var_32538_cast_fp16))[name = tensor("op_32664_cast_fp16")]; + tensor var_32666_equation_0 = const()[name = tensor("op_32666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32666_cast_fp16 = einsum(equation = var_32666_equation_0, values = (var_32138_cast_fp16, var_32539_cast_fp16))[name = tensor("op_32666_cast_fp16")]; + tensor var_32668_equation_0 = const()[name = tensor("op_32668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32668_cast_fp16 = einsum(equation = var_32668_equation_0, values = (var_32138_cast_fp16, var_32540_cast_fp16))[name = tensor("op_32668_cast_fp16")]; + tensor var_32670_equation_0 = const()[name = tensor("op_32670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32670_cast_fp16 = einsum(equation = var_32670_equation_0, values = (var_32142_cast_fp16, var_32541_cast_fp16))[name = tensor("op_32670_cast_fp16")]; + tensor var_32672_equation_0 = const()[name = tensor("op_32672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32672_cast_fp16 = einsum(equation = var_32672_equation_0, values = (var_32142_cast_fp16, var_32542_cast_fp16))[name = tensor("op_32672_cast_fp16")]; + tensor var_32674_equation_0 = const()[name = tensor("op_32674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32674_cast_fp16 = einsum(equation = var_32674_equation_0, values = (var_32142_cast_fp16, var_32543_cast_fp16))[name = tensor("op_32674_cast_fp16")]; + tensor var_32676_equation_0 = const()[name = tensor("op_32676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32676_cast_fp16 = einsum(equation = var_32676_equation_0, values = (var_32142_cast_fp16, var_32544_cast_fp16))[name = tensor("op_32676_cast_fp16")]; + tensor var_32678_equation_0 = const()[name = tensor("op_32678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32678_cast_fp16 = einsum(equation = var_32678_equation_0, values = (var_32146_cast_fp16, var_32545_cast_fp16))[name = tensor("op_32678_cast_fp16")]; + tensor var_32680_equation_0 = const()[name = tensor("op_32680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32680_cast_fp16 = einsum(equation = var_32680_equation_0, values = (var_32146_cast_fp16, var_32546_cast_fp16))[name = tensor("op_32680_cast_fp16")]; + tensor var_32682_equation_0 = const()[name = tensor("op_32682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32682_cast_fp16 = einsum(equation = var_32682_equation_0, values = (var_32146_cast_fp16, var_32547_cast_fp16))[name = tensor("op_32682_cast_fp16")]; + tensor var_32684_equation_0 = const()[name = tensor("op_32684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32684_cast_fp16 = einsum(equation = var_32684_equation_0, values = (var_32146_cast_fp16, var_32548_cast_fp16))[name = tensor("op_32684_cast_fp16")]; + tensor var_32686_equation_0 = const()[name = tensor("op_32686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32686_cast_fp16 = einsum(equation = var_32686_equation_0, values = (var_32150_cast_fp16, var_32549_cast_fp16))[name = tensor("op_32686_cast_fp16")]; + tensor var_32688_equation_0 = const()[name = tensor("op_32688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32688_cast_fp16 = einsum(equation = var_32688_equation_0, values = (var_32150_cast_fp16, var_32550_cast_fp16))[name = tensor("op_32688_cast_fp16")]; + tensor var_32690_equation_0 = const()[name = tensor("op_32690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32690_cast_fp16 = einsum(equation = var_32690_equation_0, values = (var_32150_cast_fp16, var_32551_cast_fp16))[name = tensor("op_32690_cast_fp16")]; + tensor var_32692_equation_0 = const()[name = tensor("op_32692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32692_cast_fp16 = einsum(equation = var_32692_equation_0, values = (var_32150_cast_fp16, var_32552_cast_fp16))[name = tensor("op_32692_cast_fp16")]; + tensor var_32694_equation_0 = const()[name = tensor("op_32694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32694_cast_fp16 = einsum(equation = var_32694_equation_0, values = (var_32154_cast_fp16, var_32553_cast_fp16))[name = tensor("op_32694_cast_fp16")]; + tensor var_32696_equation_0 = const()[name = tensor("op_32696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32696_cast_fp16 = einsum(equation = var_32696_equation_0, values = (var_32154_cast_fp16, var_32554_cast_fp16))[name = tensor("op_32696_cast_fp16")]; + tensor var_32698_equation_0 = const()[name = tensor("op_32698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32698_cast_fp16 = einsum(equation = var_32698_equation_0, values = (var_32154_cast_fp16, var_32555_cast_fp16))[name = tensor("op_32698_cast_fp16")]; + tensor var_32700_equation_0 = const()[name = tensor("op_32700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32700_cast_fp16 = einsum(equation = var_32700_equation_0, values = (var_32154_cast_fp16, var_32556_cast_fp16))[name = tensor("op_32700_cast_fp16")]; + tensor var_32702_equation_0 = const()[name = tensor("op_32702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32702_cast_fp16 = einsum(equation = var_32702_equation_0, values = (var_32158_cast_fp16, var_32557_cast_fp16))[name = tensor("op_32702_cast_fp16")]; + tensor var_32704_equation_0 = const()[name = tensor("op_32704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32704_cast_fp16 = einsum(equation = var_32704_equation_0, values = (var_32158_cast_fp16, var_32558_cast_fp16))[name = tensor("op_32704_cast_fp16")]; + tensor var_32706_equation_0 = const()[name = tensor("op_32706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32706_cast_fp16 = einsum(equation = var_32706_equation_0, values = (var_32158_cast_fp16, var_32559_cast_fp16))[name = tensor("op_32706_cast_fp16")]; + tensor var_32708_equation_0 = const()[name = tensor("op_32708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32708_cast_fp16 = einsum(equation = var_32708_equation_0, values = (var_32158_cast_fp16, var_32560_cast_fp16))[name = tensor("op_32708_cast_fp16")]; + tensor var_32710_equation_0 = const()[name = tensor("op_32710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32710_cast_fp16 = einsum(equation = var_32710_equation_0, values = (var_32162_cast_fp16, var_32561_cast_fp16))[name = tensor("op_32710_cast_fp16")]; + tensor var_32712_equation_0 = const()[name = tensor("op_32712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32712_cast_fp16 = einsum(equation = var_32712_equation_0, values = (var_32162_cast_fp16, var_32562_cast_fp16))[name = tensor("op_32712_cast_fp16")]; + tensor var_32714_equation_0 = const()[name = tensor("op_32714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32714_cast_fp16 = einsum(equation = var_32714_equation_0, values = (var_32162_cast_fp16, var_32563_cast_fp16))[name = tensor("op_32714_cast_fp16")]; + tensor var_32716_equation_0 = const()[name = tensor("op_32716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32716_cast_fp16 = einsum(equation = var_32716_equation_0, values = (var_32162_cast_fp16, var_32564_cast_fp16))[name = tensor("op_32716_cast_fp16")]; + tensor var_32718_equation_0 = const()[name = tensor("op_32718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32718_cast_fp16 = einsum(equation = var_32718_equation_0, values = (var_32166_cast_fp16, var_32565_cast_fp16))[name = tensor("op_32718_cast_fp16")]; + tensor var_32720_equation_0 = const()[name = tensor("op_32720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32720_cast_fp16 = einsum(equation = var_32720_equation_0, values = (var_32166_cast_fp16, var_32566_cast_fp16))[name = tensor("op_32720_cast_fp16")]; + tensor var_32722_equation_0 = const()[name = tensor("op_32722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32722_cast_fp16 = einsum(equation = var_32722_equation_0, values = (var_32166_cast_fp16, var_32567_cast_fp16))[name = tensor("op_32722_cast_fp16")]; + tensor var_32724_equation_0 = const()[name = tensor("op_32724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32724_cast_fp16 = einsum(equation = var_32724_equation_0, values = (var_32166_cast_fp16, var_32568_cast_fp16))[name = tensor("op_32724_cast_fp16")]; + tensor var_32726_equation_0 = const()[name = tensor("op_32726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32726_cast_fp16 = einsum(equation = var_32726_equation_0, values = (var_32170_cast_fp16, var_32569_cast_fp16))[name = tensor("op_32726_cast_fp16")]; + tensor var_32728_equation_0 = const()[name = tensor("op_32728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32728_cast_fp16 = einsum(equation = var_32728_equation_0, values = (var_32170_cast_fp16, var_32570_cast_fp16))[name = tensor("op_32728_cast_fp16")]; + tensor var_32730_equation_0 = const()[name = tensor("op_32730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32730_cast_fp16 = einsum(equation = var_32730_equation_0, values = (var_32170_cast_fp16, var_32571_cast_fp16))[name = tensor("op_32730_cast_fp16")]; + tensor var_32732_equation_0 = const()[name = tensor("op_32732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_32732_cast_fp16 = einsum(equation = var_32732_equation_0, values = (var_32170_cast_fp16, var_32572_cast_fp16))[name = tensor("op_32732_cast_fp16")]; + tensor var_32734_interleave_0 = const()[name = tensor("op_32734_interleave_0"), val = tensor(false)]; + tensor var_32734_cast_fp16 = concat(axis = var_31239, interleave = var_32734_interleave_0, values = (var_32574_cast_fp16, var_32576_cast_fp16, var_32578_cast_fp16, var_32580_cast_fp16))[name = tensor("op_32734_cast_fp16")]; + tensor var_32736_interleave_0 = const()[name = tensor("op_32736_interleave_0"), val = tensor(false)]; + tensor var_32736_cast_fp16 = concat(axis = var_31239, interleave = var_32736_interleave_0, values = (var_32582_cast_fp16, var_32584_cast_fp16, var_32586_cast_fp16, var_32588_cast_fp16))[name = tensor("op_32736_cast_fp16")]; + tensor var_32738_interleave_0 = const()[name = tensor("op_32738_interleave_0"), val = tensor(false)]; + tensor var_32738_cast_fp16 = concat(axis = var_31239, interleave = var_32738_interleave_0, values = (var_32590_cast_fp16, var_32592_cast_fp16, var_32594_cast_fp16, var_32596_cast_fp16))[name = tensor("op_32738_cast_fp16")]; + tensor var_32740_interleave_0 = const()[name = tensor("op_32740_interleave_0"), val = tensor(false)]; + tensor var_32740_cast_fp16 = concat(axis = var_31239, interleave = var_32740_interleave_0, values = (var_32598_cast_fp16, var_32600_cast_fp16, var_32602_cast_fp16, var_32604_cast_fp16))[name = tensor("op_32740_cast_fp16")]; + tensor var_32742_interleave_0 = const()[name = tensor("op_32742_interleave_0"), val = tensor(false)]; + tensor var_32742_cast_fp16 = concat(axis = var_31239, interleave = var_32742_interleave_0, values = (var_32606_cast_fp16, var_32608_cast_fp16, var_32610_cast_fp16, var_32612_cast_fp16))[name = tensor("op_32742_cast_fp16")]; + tensor var_32744_interleave_0 = const()[name = tensor("op_32744_interleave_0"), val = tensor(false)]; + tensor var_32744_cast_fp16 = concat(axis = var_31239, interleave = var_32744_interleave_0, values = (var_32614_cast_fp16, var_32616_cast_fp16, var_32618_cast_fp16, var_32620_cast_fp16))[name = tensor("op_32744_cast_fp16")]; + tensor var_32746_interleave_0 = const()[name = tensor("op_32746_interleave_0"), val = tensor(false)]; + tensor var_32746_cast_fp16 = concat(axis = var_31239, interleave = var_32746_interleave_0, values = (var_32622_cast_fp16, var_32624_cast_fp16, var_32626_cast_fp16, var_32628_cast_fp16))[name = tensor("op_32746_cast_fp16")]; + tensor var_32748_interleave_0 = const()[name = tensor("op_32748_interleave_0"), val = tensor(false)]; + tensor var_32748_cast_fp16 = concat(axis = var_31239, interleave = var_32748_interleave_0, values = (var_32630_cast_fp16, var_32632_cast_fp16, var_32634_cast_fp16, var_32636_cast_fp16))[name = tensor("op_32748_cast_fp16")]; + tensor var_32750_interleave_0 = const()[name = tensor("op_32750_interleave_0"), val = tensor(false)]; + tensor var_32750_cast_fp16 = concat(axis = var_31239, interleave = var_32750_interleave_0, values = (var_32638_cast_fp16, var_32640_cast_fp16, var_32642_cast_fp16, var_32644_cast_fp16))[name = tensor("op_32750_cast_fp16")]; + tensor var_32752_interleave_0 = const()[name = tensor("op_32752_interleave_0"), val = tensor(false)]; + tensor var_32752_cast_fp16 = concat(axis = var_31239, interleave = var_32752_interleave_0, values = (var_32646_cast_fp16, var_32648_cast_fp16, var_32650_cast_fp16, var_32652_cast_fp16))[name = tensor("op_32752_cast_fp16")]; + tensor var_32754_interleave_0 = const()[name = tensor("op_32754_interleave_0"), val = tensor(false)]; + tensor var_32754_cast_fp16 = concat(axis = var_31239, interleave = var_32754_interleave_0, values = (var_32654_cast_fp16, var_32656_cast_fp16, var_32658_cast_fp16, var_32660_cast_fp16))[name = tensor("op_32754_cast_fp16")]; + tensor var_32756_interleave_0 = const()[name = tensor("op_32756_interleave_0"), val = tensor(false)]; + tensor var_32756_cast_fp16 = concat(axis = var_31239, interleave = var_32756_interleave_0, values = (var_32662_cast_fp16, var_32664_cast_fp16, var_32666_cast_fp16, var_32668_cast_fp16))[name = tensor("op_32756_cast_fp16")]; + tensor var_32758_interleave_0 = const()[name = tensor("op_32758_interleave_0"), val = tensor(false)]; + tensor var_32758_cast_fp16 = concat(axis = var_31239, interleave = var_32758_interleave_0, values = (var_32670_cast_fp16, var_32672_cast_fp16, var_32674_cast_fp16, var_32676_cast_fp16))[name = tensor("op_32758_cast_fp16")]; + tensor var_32760_interleave_0 = const()[name = tensor("op_32760_interleave_0"), val = tensor(false)]; + tensor var_32760_cast_fp16 = concat(axis = var_31239, interleave = var_32760_interleave_0, values = (var_32678_cast_fp16, var_32680_cast_fp16, var_32682_cast_fp16, var_32684_cast_fp16))[name = tensor("op_32760_cast_fp16")]; + tensor var_32762_interleave_0 = const()[name = tensor("op_32762_interleave_0"), val = tensor(false)]; + tensor var_32762_cast_fp16 = concat(axis = var_31239, interleave = var_32762_interleave_0, values = (var_32686_cast_fp16, var_32688_cast_fp16, var_32690_cast_fp16, var_32692_cast_fp16))[name = tensor("op_32762_cast_fp16")]; + tensor var_32764_interleave_0 = const()[name = tensor("op_32764_interleave_0"), val = tensor(false)]; + tensor var_32764_cast_fp16 = concat(axis = var_31239, interleave = var_32764_interleave_0, values = (var_32694_cast_fp16, var_32696_cast_fp16, var_32698_cast_fp16, var_32700_cast_fp16))[name = tensor("op_32764_cast_fp16")]; + tensor var_32766_interleave_0 = const()[name = tensor("op_32766_interleave_0"), val = tensor(false)]; + tensor var_32766_cast_fp16 = concat(axis = var_31239, interleave = var_32766_interleave_0, values = (var_32702_cast_fp16, var_32704_cast_fp16, var_32706_cast_fp16, var_32708_cast_fp16))[name = tensor("op_32766_cast_fp16")]; + tensor var_32768_interleave_0 = const()[name = tensor("op_32768_interleave_0"), val = tensor(false)]; + tensor var_32768_cast_fp16 = concat(axis = var_31239, interleave = var_32768_interleave_0, values = (var_32710_cast_fp16, var_32712_cast_fp16, var_32714_cast_fp16, var_32716_cast_fp16))[name = tensor("op_32768_cast_fp16")]; + tensor var_32770_interleave_0 = const()[name = tensor("op_32770_interleave_0"), val = tensor(false)]; + tensor var_32770_cast_fp16 = concat(axis = var_31239, interleave = var_32770_interleave_0, values = (var_32718_cast_fp16, var_32720_cast_fp16, var_32722_cast_fp16, var_32724_cast_fp16))[name = tensor("op_32770_cast_fp16")]; + tensor var_32772_interleave_0 = const()[name = tensor("op_32772_interleave_0"), val = tensor(false)]; + tensor var_32772_cast_fp16 = concat(axis = var_31239, interleave = var_32772_interleave_0, values = (var_32726_cast_fp16, var_32728_cast_fp16, var_32730_cast_fp16, var_32732_cast_fp16))[name = tensor("op_32772_cast_fp16")]; + tensor input_387_interleave_0 = const()[name = tensor("input_387_interleave_0"), val = tensor(false)]; + tensor input_387_cast_fp16 = concat(axis = var_31264, interleave = input_387_interleave_0, values = (var_32734_cast_fp16, var_32736_cast_fp16, var_32738_cast_fp16, var_32740_cast_fp16, var_32742_cast_fp16, var_32744_cast_fp16, var_32746_cast_fp16, var_32748_cast_fp16, var_32750_cast_fp16, var_32752_cast_fp16, var_32754_cast_fp16, var_32756_cast_fp16, var_32758_cast_fp16, var_32760_cast_fp16, var_32762_cast_fp16, var_32764_cast_fp16, var_32766_cast_fp16, var_32768_cast_fp16, var_32770_cast_fp16, var_32772_cast_fp16))[name = tensor("input_387_cast_fp16")]; + tensor var_32780 = const()[name = tensor("op_32780"), val = tensor([1, 1])]; + tensor var_32782 = const()[name = tensor("op_32782"), val = tensor([1, 1])]; + tensor pretrained_out_235_pad_type_0 = const()[name = tensor("pretrained_out_235_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_235_pad_0 = const()[name = tensor("pretrained_out_235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(218437696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219256960))), name = tensor("layers_19_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_19_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_19_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219257088)))]; + tensor pretrained_out_235_cast_fp16 = conv(bias = layers_19_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_32782, groups = var_31264, pad = pretrained_out_235_pad_0, pad_type = pretrained_out_235_pad_type_0, strides = var_32780, weight = layers_19_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_387_cast_fp16)[name = tensor("pretrained_out_235_cast_fp16")]; + tensor var_32786 = const()[name = tensor("op_32786"), val = tensor([1, 1])]; + tensor var_32788 = const()[name = tensor("op_32788"), val = tensor([1, 1])]; + tensor input_389_pad_type_0 = const()[name = tensor("input_389_pad_type_0"), val = tensor("custom")]; + tensor input_389_pad_0 = const()[name = tensor("input_389_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_19_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219259712)))]; + tensor input_389_cast_fp16 = conv(dilations = var_32788, groups = var_31264, pad = input_389_pad_0, pad_type = input_389_pad_type_0, strides = var_32786, weight = layers_19_self_attn_o_proj_loraA_weight_to_fp16, x = input_387_cast_fp16)[name = tensor("input_389_cast_fp16")]; + tensor var_32792 = const()[name = tensor("op_32792"), val = tensor([1, 1])]; + tensor var_32794 = const()[name = tensor("op_32794"), val = tensor([1, 1])]; + tensor lora_out_469_pad_type_0 = const()[name = tensor("lora_out_469_pad_type_0"), val = tensor("custom")]; + tensor lora_out_469_pad_0 = const()[name = tensor("lora_out_469_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_471_weight_0_to_fp16 = const()[name = tensor("lora_out_471_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219300736)))]; + tensor lora_out_471_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_32794, groups = var_31264, pad = lora_out_469_pad_0, pad_type = lora_out_469_pad_type_0, strides = var_32792, weight = lora_out_471_weight_0_to_fp16, x = input_389_cast_fp16)[name = tensor("lora_out_471_cast_fp16")]; + tensor obj_79_cast_fp16 = add(x = pretrained_out_235_cast_fp16, y = lora_out_471_cast_fp16)[name = tensor("obj_79_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = obj_79_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; + tensor var_32803 = const()[name = tensor("op_32803"), val = tensor([1])]; + tensor channels_mean_79_cast_fp16 = reduce_mean(axes = var_32803, keep_dims = var_31265, x = inputs_79_cast_fp16)[name = tensor("channels_mean_79_cast_fp16")]; + tensor zero_mean_79_cast_fp16 = sub(x = inputs_79_cast_fp16, y = channels_mean_79_cast_fp16)[name = tensor("zero_mean_79_cast_fp16")]; + tensor zero_mean_sq_79_cast_fp16 = mul(x = zero_mean_79_cast_fp16, y = zero_mean_79_cast_fp16)[name = tensor("zero_mean_sq_79_cast_fp16")]; + tensor var_32807 = const()[name = tensor("op_32807"), val = tensor([1])]; + tensor var_32808_cast_fp16 = reduce_mean(axes = var_32807, keep_dims = var_31265, x = zero_mean_sq_79_cast_fp16)[name = tensor("op_32808_cast_fp16")]; + tensor var_32809_to_fp16 = const()[name = tensor("op_32809_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_32810_cast_fp16 = add(x = var_32808_cast_fp16, y = var_32809_to_fp16)[name = tensor("op_32810_cast_fp16")]; + tensor denom_79_epsilon_0 = const()[name = tensor("denom_79_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_79_cast_fp16 = rsqrt(epsilon = denom_79_epsilon_0, x = var_32810_cast_fp16)[name = tensor("denom_79_cast_fp16")]; + tensor out_79_cast_fp16 = mul(x = zero_mean_79_cast_fp16, y = denom_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; + tensor input_391_gamma_0_to_fp16 = const()[name = tensor("input_391_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219341760)))]; + tensor input_391_beta_0_to_fp16 = const()[name = tensor("input_391_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219344384)))]; + tensor input_391_epsilon_0_to_fp16 = const()[name = tensor("input_391_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_391_cast_fp16 = batch_norm(beta = input_391_beta_0_to_fp16, epsilon = input_391_epsilon_0_to_fp16, gamma = input_391_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor("input_391_cast_fp16")]; + tensor var_32824 = const()[name = tensor("op_32824"), val = tensor([1, 1])]; + tensor var_32826 = const()[name = tensor("op_32826"), val = tensor([1, 1])]; + tensor pretrained_out_237_pad_type_0 = const()[name = tensor("pretrained_out_237_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_237_pad_0 = const()[name = tensor("pretrained_out_237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219347008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222623872))), name = tensor("layers_19_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_19_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_19_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222624000)))]; + tensor pretrained_out_237_cast_fp16 = conv(bias = layers_19_fc1_pretrained_bias_to_fp16, dilations = var_32826, groups = var_31264, pad = pretrained_out_237_pad_0, pad_type = pretrained_out_237_pad_type_0, strides = var_32824, weight = layers_19_fc1_pretrained_weight_to_fp16_palettized, x = input_391_cast_fp16)[name = tensor("pretrained_out_237_cast_fp16")]; + tensor var_32830 = const()[name = tensor("op_32830"), val = tensor([1, 1])]; + tensor var_32832 = const()[name = tensor("op_32832"), val = tensor([1, 1])]; + tensor input_393_pad_type_0 = const()[name = tensor("input_393_pad_type_0"), val = tensor("custom")]; + tensor input_393_pad_0 = const()[name = tensor("input_393_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_19_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222634304)))]; + tensor input_393_cast_fp16 = conv(dilations = var_32832, groups = var_31264, pad = input_393_pad_0, pad_type = input_393_pad_type_0, strides = var_32830, weight = layers_19_fc1_loraA_weight_to_fp16, x = input_391_cast_fp16)[name = tensor("input_393_cast_fp16")]; + tensor var_32836 = const()[name = tensor("op_32836"), val = tensor([1, 1])]; + tensor var_32838 = const()[name = tensor("op_32838"), val = tensor([1, 1])]; + tensor lora_out_473_pad_type_0 = const()[name = tensor("lora_out_473_pad_type_0"), val = tensor("custom")]; + tensor lora_out_473_pad_0 = const()[name = tensor("lora_out_473_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_475_weight_0_to_fp16 = const()[name = tensor("lora_out_475_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222675328)))]; + tensor lora_out_475_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_32838, groups = var_31264, pad = lora_out_473_pad_0, pad_type = lora_out_473_pad_type_0, strides = var_32836, weight = lora_out_475_weight_0_to_fp16, x = input_393_cast_fp16)[name = tensor("lora_out_475_cast_fp16")]; + tensor input_395_cast_fp16 = add(x = pretrained_out_237_cast_fp16, y = lora_out_475_cast_fp16)[name = tensor("input_395_cast_fp16")]; + tensor input_397_mode_0 = const()[name = tensor("input_397_mode_0"), val = tensor("EXACT")]; + tensor input_397_cast_fp16 = gelu(mode = input_397_mode_0, x = input_395_cast_fp16)[name = tensor("input_397_cast_fp16")]; + tensor var_32850 = const()[name = tensor("op_32850"), val = tensor([1, 1])]; + tensor var_32852 = const()[name = tensor("op_32852"), val = tensor([1, 1])]; + tensor pretrained_out_239_pad_type_0 = const()[name = tensor("pretrained_out_239_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_239_pad_0 = const()[name = tensor("pretrained_out_239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222839232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226116096))), name = tensor("layers_19_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_19_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_19_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226116224)))]; + tensor pretrained_out_239_cast_fp16 = conv(bias = layers_19_fc2_pretrained_bias_to_fp16, dilations = var_32852, groups = var_31264, pad = pretrained_out_239_pad_0, pad_type = pretrained_out_239_pad_type_0, strides = var_32850, weight = layers_19_fc2_pretrained_weight_to_fp16_palettized, x = input_397_cast_fp16)[name = tensor("pretrained_out_239_cast_fp16")]; + tensor var_32856 = const()[name = tensor("op_32856"), val = tensor([1, 1])]; + tensor var_32858 = const()[name = tensor("op_32858"), val = tensor([1, 1])]; + tensor input_399_pad_type_0 = const()[name = tensor("input_399_pad_type_0"), val = tensor("custom")]; + tensor input_399_pad_0 = const()[name = tensor("input_399_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_19_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_19_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226118848)))]; + tensor input_399_cast_fp16 = conv(dilations = var_32858, groups = var_31264, pad = input_399_pad_0, pad_type = input_399_pad_type_0, strides = var_32856, weight = layers_19_fc2_loraA_weight_to_fp16, x = input_397_cast_fp16)[name = tensor("input_399_cast_fp16")]; + tensor var_32862 = const()[name = tensor("op_32862"), val = tensor([1, 1])]; + tensor var_32864 = const()[name = tensor("op_32864"), val = tensor([1, 1])]; + tensor lora_out_477_pad_type_0 = const()[name = tensor("lora_out_477_pad_type_0"), val = tensor("custom")]; + tensor lora_out_477_pad_0 = const()[name = tensor("lora_out_477_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_479_weight_0_to_fp16 = const()[name = tensor("lora_out_479_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226282752)))]; + tensor lora_out_479_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_32864, groups = var_31264, pad = lora_out_477_pad_0, pad_type = lora_out_477_pad_type_0, strides = var_32862, weight = lora_out_479_weight_0_to_fp16, x = input_399_cast_fp16)[name = tensor("lora_out_479_cast_fp16")]; + tensor hidden_states_43_cast_fp16 = add(x = pretrained_out_239_cast_fp16, y = lora_out_479_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor inputs_81_cast_fp16 = add(x = inputs_79_cast_fp16, y = hidden_states_43_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; + tensor var_32874 = const()[name = tensor("op_32874"), val = tensor(3)]; + tensor var_32899 = const()[name = tensor("op_32899"), val = tensor(1)]; + tensor var_32900 = const()[name = tensor("op_32900"), val = tensor(true)]; + tensor var_32910 = const()[name = tensor("op_32910"), val = tensor([1])]; + tensor channels_mean_81_cast_fp16 = reduce_mean(axes = var_32910, keep_dims = var_32900, x = inputs_81_cast_fp16)[name = tensor("channels_mean_81_cast_fp16")]; + tensor zero_mean_81_cast_fp16 = sub(x = inputs_81_cast_fp16, y = channels_mean_81_cast_fp16)[name = tensor("zero_mean_81_cast_fp16")]; + tensor zero_mean_sq_81_cast_fp16 = mul(x = zero_mean_81_cast_fp16, y = zero_mean_81_cast_fp16)[name = tensor("zero_mean_sq_81_cast_fp16")]; + tensor var_32914 = const()[name = tensor("op_32914"), val = tensor([1])]; + tensor var_32915_cast_fp16 = reduce_mean(axes = var_32914, keep_dims = var_32900, x = zero_mean_sq_81_cast_fp16)[name = tensor("op_32915_cast_fp16")]; + tensor var_32916_to_fp16 = const()[name = tensor("op_32916_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_32917_cast_fp16 = add(x = var_32915_cast_fp16, y = var_32916_to_fp16)[name = tensor("op_32917_cast_fp16")]; + tensor denom_81_epsilon_0 = const()[name = tensor("denom_81_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_81_cast_fp16 = rsqrt(epsilon = denom_81_epsilon_0, x = var_32917_cast_fp16)[name = tensor("denom_81_cast_fp16")]; + tensor out_81_cast_fp16 = mul(x = zero_mean_81_cast_fp16, y = denom_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; + tensor obj_81_gamma_0_to_fp16 = const()[name = tensor("obj_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226323776)))]; + tensor obj_81_beta_0_to_fp16 = const()[name = tensor("obj_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226326400)))]; + tensor obj_81_epsilon_0_to_fp16 = const()[name = tensor("obj_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_81_cast_fp16 = batch_norm(beta = obj_81_beta_0_to_fp16, epsilon = obj_81_epsilon_0_to_fp16, gamma = obj_81_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor("obj_81_cast_fp16")]; + tensor var_32935 = const()[name = tensor("op_32935"), val = tensor([1, 1])]; + tensor var_32937 = const()[name = tensor("op_32937"), val = tensor([1, 1])]; + tensor pretrained_out_241_pad_type_0 = const()[name = tensor("pretrained_out_241_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_241_pad_0 = const()[name = tensor("pretrained_out_241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226329024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227148288))), name = tensor("layers_20_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_20_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227148416)))]; + tensor pretrained_out_241_cast_fp16 = conv(bias = layers_20_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_32937, groups = var_32899, pad = pretrained_out_241_pad_0, pad_type = pretrained_out_241_pad_type_0, strides = var_32935, weight = layers_20_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_81_cast_fp16)[name = tensor("pretrained_out_241_cast_fp16")]; + tensor var_32941 = const()[name = tensor("op_32941"), val = tensor([1, 1])]; + tensor var_32943 = const()[name = tensor("op_32943"), val = tensor([1, 1])]; + tensor input_401_pad_type_0 = const()[name = tensor("input_401_pad_type_0"), val = tensor("custom")]; + tensor input_401_pad_0 = const()[name = tensor("input_401_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227151040)))]; + tensor input_401_cast_fp16 = conv(dilations = var_32943, groups = var_32899, pad = input_401_pad_0, pad_type = input_401_pad_type_0, strides = var_32941, weight = layers_20_self_attn_q_proj_loraA_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor("input_401_cast_fp16")]; + tensor var_32947 = const()[name = tensor("op_32947"), val = tensor([1, 1])]; + tensor var_32949 = const()[name = tensor("op_32949"), val = tensor([1, 1])]; + tensor lora_out_481_pad_type_0 = const()[name = tensor("lora_out_481_pad_type_0"), val = tensor("custom")]; + tensor lora_out_481_pad_0 = const()[name = tensor("lora_out_481_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_483_weight_0_to_fp16 = const()[name = tensor("lora_out_483_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227192064)))]; + tensor lora_out_483_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_32949, groups = var_32899, pad = lora_out_481_pad_0, pad_type = lora_out_481_pad_type_0, strides = var_32947, weight = lora_out_483_weight_0_to_fp16, x = input_401_cast_fp16)[name = tensor("lora_out_483_cast_fp16")]; + tensor query_41_cast_fp16 = add(x = pretrained_out_241_cast_fp16, y = lora_out_483_cast_fp16)[name = tensor("query_41_cast_fp16")]; + tensor var_32959 = const()[name = tensor("op_32959"), val = tensor([1, 1])]; + tensor var_32961 = const()[name = tensor("op_32961"), val = tensor([1, 1])]; + tensor pretrained_out_243_pad_type_0 = const()[name = tensor("pretrained_out_243_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_243_pad_0 = const()[name = tensor("pretrained_out_243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227233088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228052352))), name = tensor("layers_20_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_243_cast_fp16 = conv(dilations = var_32961, groups = var_32899, pad = pretrained_out_243_pad_0, pad_type = pretrained_out_243_pad_type_0, strides = var_32959, weight = layers_20_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_81_cast_fp16)[name = tensor("pretrained_out_243_cast_fp16")]; + tensor var_32965 = const()[name = tensor("op_32965"), val = tensor([1, 1])]; + tensor var_32967 = const()[name = tensor("op_32967"), val = tensor([1, 1])]; + tensor input_403_pad_type_0 = const()[name = tensor("input_403_pad_type_0"), val = tensor("custom")]; + tensor input_403_pad_0 = const()[name = tensor("input_403_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228052480)))]; + tensor input_403_cast_fp16 = conv(dilations = var_32967, groups = var_32899, pad = input_403_pad_0, pad_type = input_403_pad_type_0, strides = var_32965, weight = layers_20_self_attn_k_proj_loraA_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor("input_403_cast_fp16")]; + tensor var_32971 = const()[name = tensor("op_32971"), val = tensor([1, 1])]; + tensor var_32973 = const()[name = tensor("op_32973"), val = tensor([1, 1])]; + tensor lora_out_485_pad_type_0 = const()[name = tensor("lora_out_485_pad_type_0"), val = tensor("custom")]; + tensor lora_out_485_pad_0 = const()[name = tensor("lora_out_485_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_487_weight_0_to_fp16 = const()[name = tensor("lora_out_487_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228093504)))]; + tensor lora_out_487_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_32973, groups = var_32899, pad = lora_out_485_pad_0, pad_type = lora_out_485_pad_type_0, strides = var_32971, weight = lora_out_487_weight_0_to_fp16, x = input_403_cast_fp16)[name = tensor("lora_out_487_cast_fp16")]; + tensor key_41_cast_fp16 = add(x = pretrained_out_243_cast_fp16, y = lora_out_487_cast_fp16)[name = tensor("key_41_cast_fp16")]; + tensor var_32984 = const()[name = tensor("op_32984"), val = tensor([1, 1])]; + tensor var_32986 = const()[name = tensor("op_32986"), val = tensor([1, 1])]; + tensor pretrained_out_245_pad_type_0 = const()[name = tensor("pretrained_out_245_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_245_pad_0 = const()[name = tensor("pretrained_out_245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228134528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228953792))), name = tensor("layers_20_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_20_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228953920)))]; + tensor pretrained_out_245_cast_fp16 = conv(bias = layers_20_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_32986, groups = var_32899, pad = pretrained_out_245_pad_0, pad_type = pretrained_out_245_pad_type_0, strides = var_32984, weight = layers_20_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_81_cast_fp16)[name = tensor("pretrained_out_245_cast_fp16")]; + tensor var_32990 = const()[name = tensor("op_32990"), val = tensor([1, 1])]; + tensor var_32992 = const()[name = tensor("op_32992"), val = tensor([1, 1])]; + tensor input_405_pad_type_0 = const()[name = tensor("input_405_pad_type_0"), val = tensor("custom")]; + tensor input_405_pad_0 = const()[name = tensor("input_405_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228956544)))]; + tensor input_405_cast_fp16 = conv(dilations = var_32992, groups = var_32899, pad = input_405_pad_0, pad_type = input_405_pad_type_0, strides = var_32990, weight = layers_20_self_attn_v_proj_loraA_weight_to_fp16, x = obj_81_cast_fp16)[name = tensor("input_405_cast_fp16")]; + tensor var_32996 = const()[name = tensor("op_32996"), val = tensor([1, 1])]; + tensor var_32998 = const()[name = tensor("op_32998"), val = tensor([1, 1])]; + tensor lora_out_489_pad_type_0 = const()[name = tensor("lora_out_489_pad_type_0"), val = tensor("custom")]; + tensor lora_out_489_pad_0 = const()[name = tensor("lora_out_489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_491_weight_0_to_fp16 = const()[name = tensor("lora_out_491_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228997568)))]; + tensor lora_out_491_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_32998, groups = var_32899, pad = lora_out_489_pad_0, pad_type = lora_out_489_pad_type_0, strides = var_32996, weight = lora_out_491_weight_0_to_fp16, x = input_405_cast_fp16)[name = tensor("lora_out_491_cast_fp16")]; + tensor value_41_cast_fp16 = add(x = pretrained_out_245_cast_fp16, y = lora_out_491_cast_fp16)[name = tensor("value_41_cast_fp16")]; + tensor var_33008_begin_0 = const()[name = tensor("op_33008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33008_end_0 = const()[name = tensor("op_33008_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33008_end_mask_0 = const()[name = tensor("op_33008_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33008_cast_fp16 = slice_by_index(begin = var_33008_begin_0, end = var_33008_end_0, end_mask = var_33008_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33008_cast_fp16")]; + tensor var_33012_begin_0 = const()[name = tensor("op_33012_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33012_end_0 = const()[name = tensor("op_33012_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33012_end_mask_0 = const()[name = tensor("op_33012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33012_cast_fp16 = slice_by_index(begin = var_33012_begin_0, end = var_33012_end_0, end_mask = var_33012_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33012_cast_fp16")]; + tensor var_33016_begin_0 = const()[name = tensor("op_33016_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33016_end_0 = const()[name = tensor("op_33016_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33016_end_mask_0 = const()[name = tensor("op_33016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33016_cast_fp16 = slice_by_index(begin = var_33016_begin_0, end = var_33016_end_0, end_mask = var_33016_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33016_cast_fp16")]; + tensor var_33020_begin_0 = const()[name = tensor("op_33020_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33020_end_0 = const()[name = tensor("op_33020_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33020_end_mask_0 = const()[name = tensor("op_33020_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33020_cast_fp16 = slice_by_index(begin = var_33020_begin_0, end = var_33020_end_0, end_mask = var_33020_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33020_cast_fp16")]; + tensor var_33024_begin_0 = const()[name = tensor("op_33024_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33024_end_0 = const()[name = tensor("op_33024_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33024_end_mask_0 = const()[name = tensor("op_33024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33024_cast_fp16 = slice_by_index(begin = var_33024_begin_0, end = var_33024_end_0, end_mask = var_33024_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33024_cast_fp16")]; + tensor var_33028_begin_0 = const()[name = tensor("op_33028_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33028_end_0 = const()[name = tensor("op_33028_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33028_end_mask_0 = const()[name = tensor("op_33028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33028_cast_fp16 = slice_by_index(begin = var_33028_begin_0, end = var_33028_end_0, end_mask = var_33028_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33028_cast_fp16")]; + tensor var_33032_begin_0 = const()[name = tensor("op_33032_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33032_end_0 = const()[name = tensor("op_33032_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33032_end_mask_0 = const()[name = tensor("op_33032_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33032_cast_fp16 = slice_by_index(begin = var_33032_begin_0, end = var_33032_end_0, end_mask = var_33032_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33032_cast_fp16")]; + tensor var_33036_begin_0 = const()[name = tensor("op_33036_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33036_end_0 = const()[name = tensor("op_33036_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33036_end_mask_0 = const()[name = tensor("op_33036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33036_cast_fp16 = slice_by_index(begin = var_33036_begin_0, end = var_33036_end_0, end_mask = var_33036_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33036_cast_fp16")]; + tensor var_33040_begin_0 = const()[name = tensor("op_33040_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33040_end_0 = const()[name = tensor("op_33040_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33040_end_mask_0 = const()[name = tensor("op_33040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33040_cast_fp16 = slice_by_index(begin = var_33040_begin_0, end = var_33040_end_0, end_mask = var_33040_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33040_cast_fp16")]; + tensor var_33044_begin_0 = const()[name = tensor("op_33044_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33044_end_0 = const()[name = tensor("op_33044_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33044_end_mask_0 = const()[name = tensor("op_33044_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33044_cast_fp16 = slice_by_index(begin = var_33044_begin_0, end = var_33044_end_0, end_mask = var_33044_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33044_cast_fp16")]; + tensor var_33048_begin_0 = const()[name = tensor("op_33048_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33048_end_0 = const()[name = tensor("op_33048_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33048_end_mask_0 = const()[name = tensor("op_33048_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33048_cast_fp16 = slice_by_index(begin = var_33048_begin_0, end = var_33048_end_0, end_mask = var_33048_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33048_cast_fp16")]; + tensor var_33052_begin_0 = const()[name = tensor("op_33052_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33052_end_0 = const()[name = tensor("op_33052_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33052_end_mask_0 = const()[name = tensor("op_33052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33052_cast_fp16 = slice_by_index(begin = var_33052_begin_0, end = var_33052_end_0, end_mask = var_33052_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33052_cast_fp16")]; + tensor var_33056_begin_0 = const()[name = tensor("op_33056_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33056_end_0 = const()[name = tensor("op_33056_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33056_end_mask_0 = const()[name = tensor("op_33056_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33056_cast_fp16 = slice_by_index(begin = var_33056_begin_0, end = var_33056_end_0, end_mask = var_33056_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33056_cast_fp16")]; + tensor var_33060_begin_0 = const()[name = tensor("op_33060_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33060_end_0 = const()[name = tensor("op_33060_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33060_end_mask_0 = const()[name = tensor("op_33060_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33060_cast_fp16 = slice_by_index(begin = var_33060_begin_0, end = var_33060_end_0, end_mask = var_33060_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33060_cast_fp16")]; + tensor var_33064_begin_0 = const()[name = tensor("op_33064_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33064_end_0 = const()[name = tensor("op_33064_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33064_end_mask_0 = const()[name = tensor("op_33064_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33064_cast_fp16 = slice_by_index(begin = var_33064_begin_0, end = var_33064_end_0, end_mask = var_33064_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33064_cast_fp16")]; + tensor var_33068_begin_0 = const()[name = tensor("op_33068_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33068_end_0 = const()[name = tensor("op_33068_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33068_end_mask_0 = const()[name = tensor("op_33068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33068_cast_fp16 = slice_by_index(begin = var_33068_begin_0, end = var_33068_end_0, end_mask = var_33068_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33068_cast_fp16")]; + tensor var_33072_begin_0 = const()[name = tensor("op_33072_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33072_end_0 = const()[name = tensor("op_33072_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33072_end_mask_0 = const()[name = tensor("op_33072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33072_cast_fp16 = slice_by_index(begin = var_33072_begin_0, end = var_33072_end_0, end_mask = var_33072_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33072_cast_fp16")]; + tensor var_33076_begin_0 = const()[name = tensor("op_33076_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33076_end_0 = const()[name = tensor("op_33076_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33076_end_mask_0 = const()[name = tensor("op_33076_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33076_cast_fp16 = slice_by_index(begin = var_33076_begin_0, end = var_33076_end_0, end_mask = var_33076_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33076_cast_fp16")]; + tensor var_33080_begin_0 = const()[name = tensor("op_33080_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33080_end_0 = const()[name = tensor("op_33080_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33080_end_mask_0 = const()[name = tensor("op_33080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33080_cast_fp16 = slice_by_index(begin = var_33080_begin_0, end = var_33080_end_0, end_mask = var_33080_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33080_cast_fp16")]; + tensor var_33084_begin_0 = const()[name = tensor("op_33084_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33084_end_0 = const()[name = tensor("op_33084_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33084_end_mask_0 = const()[name = tensor("op_33084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33084_cast_fp16 = slice_by_index(begin = var_33084_begin_0, end = var_33084_end_0, end_mask = var_33084_end_mask_0, x = query_41_cast_fp16)[name = tensor("op_33084_cast_fp16")]; + tensor var_33093_begin_0 = const()[name = tensor("op_33093_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33093_end_0 = const()[name = tensor("op_33093_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33093_end_mask_0 = const()[name = tensor("op_33093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33093_cast_fp16 = slice_by_index(begin = var_33093_begin_0, end = var_33093_end_0, end_mask = var_33093_end_mask_0, x = var_33008_cast_fp16)[name = tensor("op_33093_cast_fp16")]; + tensor var_33100_begin_0 = const()[name = tensor("op_33100_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33100_end_0 = const()[name = tensor("op_33100_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33100_end_mask_0 = const()[name = tensor("op_33100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33100_cast_fp16 = slice_by_index(begin = var_33100_begin_0, end = var_33100_end_0, end_mask = var_33100_end_mask_0, x = var_33008_cast_fp16)[name = tensor("op_33100_cast_fp16")]; + tensor var_33107_begin_0 = const()[name = tensor("op_33107_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33107_end_0 = const()[name = tensor("op_33107_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33107_end_mask_0 = const()[name = tensor("op_33107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33107_cast_fp16 = slice_by_index(begin = var_33107_begin_0, end = var_33107_end_0, end_mask = var_33107_end_mask_0, x = var_33008_cast_fp16)[name = tensor("op_33107_cast_fp16")]; + tensor var_33114_begin_0 = const()[name = tensor("op_33114_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33114_end_0 = const()[name = tensor("op_33114_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33114_end_mask_0 = const()[name = tensor("op_33114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33114_cast_fp16 = slice_by_index(begin = var_33114_begin_0, end = var_33114_end_0, end_mask = var_33114_end_mask_0, x = var_33008_cast_fp16)[name = tensor("op_33114_cast_fp16")]; + tensor var_33121_begin_0 = const()[name = tensor("op_33121_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33121_end_0 = const()[name = tensor("op_33121_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33121_end_mask_0 = const()[name = tensor("op_33121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33121_cast_fp16 = slice_by_index(begin = var_33121_begin_0, end = var_33121_end_0, end_mask = var_33121_end_mask_0, x = var_33012_cast_fp16)[name = tensor("op_33121_cast_fp16")]; + tensor var_33128_begin_0 = const()[name = tensor("op_33128_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33128_end_0 = const()[name = tensor("op_33128_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33128_end_mask_0 = const()[name = tensor("op_33128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33128_cast_fp16 = slice_by_index(begin = var_33128_begin_0, end = var_33128_end_0, end_mask = var_33128_end_mask_0, x = var_33012_cast_fp16)[name = tensor("op_33128_cast_fp16")]; + tensor var_33135_begin_0 = const()[name = tensor("op_33135_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33135_end_0 = const()[name = tensor("op_33135_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33135_end_mask_0 = const()[name = tensor("op_33135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33135_cast_fp16 = slice_by_index(begin = var_33135_begin_0, end = var_33135_end_0, end_mask = var_33135_end_mask_0, x = var_33012_cast_fp16)[name = tensor("op_33135_cast_fp16")]; + tensor var_33142_begin_0 = const()[name = tensor("op_33142_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33142_end_0 = const()[name = tensor("op_33142_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33142_end_mask_0 = const()[name = tensor("op_33142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33142_cast_fp16 = slice_by_index(begin = var_33142_begin_0, end = var_33142_end_0, end_mask = var_33142_end_mask_0, x = var_33012_cast_fp16)[name = tensor("op_33142_cast_fp16")]; + tensor var_33149_begin_0 = const()[name = tensor("op_33149_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33149_end_0 = const()[name = tensor("op_33149_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33149_end_mask_0 = const()[name = tensor("op_33149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33149_cast_fp16 = slice_by_index(begin = var_33149_begin_0, end = var_33149_end_0, end_mask = var_33149_end_mask_0, x = var_33016_cast_fp16)[name = tensor("op_33149_cast_fp16")]; + tensor var_33156_begin_0 = const()[name = tensor("op_33156_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33156_end_0 = const()[name = tensor("op_33156_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33156_end_mask_0 = const()[name = tensor("op_33156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33156_cast_fp16 = slice_by_index(begin = var_33156_begin_0, end = var_33156_end_0, end_mask = var_33156_end_mask_0, x = var_33016_cast_fp16)[name = tensor("op_33156_cast_fp16")]; + tensor var_33163_begin_0 = const()[name = tensor("op_33163_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33163_end_0 = const()[name = tensor("op_33163_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33163_end_mask_0 = const()[name = tensor("op_33163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33163_cast_fp16 = slice_by_index(begin = var_33163_begin_0, end = var_33163_end_0, end_mask = var_33163_end_mask_0, x = var_33016_cast_fp16)[name = tensor("op_33163_cast_fp16")]; + tensor var_33170_begin_0 = const()[name = tensor("op_33170_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33170_end_0 = const()[name = tensor("op_33170_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33170_end_mask_0 = const()[name = tensor("op_33170_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33170_cast_fp16 = slice_by_index(begin = var_33170_begin_0, end = var_33170_end_0, end_mask = var_33170_end_mask_0, x = var_33016_cast_fp16)[name = tensor("op_33170_cast_fp16")]; + tensor var_33177_begin_0 = const()[name = tensor("op_33177_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33177_end_0 = const()[name = tensor("op_33177_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33177_end_mask_0 = const()[name = tensor("op_33177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33177_cast_fp16 = slice_by_index(begin = var_33177_begin_0, end = var_33177_end_0, end_mask = var_33177_end_mask_0, x = var_33020_cast_fp16)[name = tensor("op_33177_cast_fp16")]; + tensor var_33184_begin_0 = const()[name = tensor("op_33184_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33184_end_0 = const()[name = tensor("op_33184_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33184_end_mask_0 = const()[name = tensor("op_33184_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33184_cast_fp16 = slice_by_index(begin = var_33184_begin_0, end = var_33184_end_0, end_mask = var_33184_end_mask_0, x = var_33020_cast_fp16)[name = tensor("op_33184_cast_fp16")]; + tensor var_33191_begin_0 = const()[name = tensor("op_33191_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33191_end_0 = const()[name = tensor("op_33191_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33191_end_mask_0 = const()[name = tensor("op_33191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33191_cast_fp16 = slice_by_index(begin = var_33191_begin_0, end = var_33191_end_0, end_mask = var_33191_end_mask_0, x = var_33020_cast_fp16)[name = tensor("op_33191_cast_fp16")]; + tensor var_33198_begin_0 = const()[name = tensor("op_33198_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33198_end_0 = const()[name = tensor("op_33198_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33198_end_mask_0 = const()[name = tensor("op_33198_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33198_cast_fp16 = slice_by_index(begin = var_33198_begin_0, end = var_33198_end_0, end_mask = var_33198_end_mask_0, x = var_33020_cast_fp16)[name = tensor("op_33198_cast_fp16")]; + tensor var_33205_begin_0 = const()[name = tensor("op_33205_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33205_end_0 = const()[name = tensor("op_33205_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33205_end_mask_0 = const()[name = tensor("op_33205_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33205_cast_fp16 = slice_by_index(begin = var_33205_begin_0, end = var_33205_end_0, end_mask = var_33205_end_mask_0, x = var_33024_cast_fp16)[name = tensor("op_33205_cast_fp16")]; + tensor var_33212_begin_0 = const()[name = tensor("op_33212_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33212_end_0 = const()[name = tensor("op_33212_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33212_end_mask_0 = const()[name = tensor("op_33212_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33212_cast_fp16 = slice_by_index(begin = var_33212_begin_0, end = var_33212_end_0, end_mask = var_33212_end_mask_0, x = var_33024_cast_fp16)[name = tensor("op_33212_cast_fp16")]; + tensor var_33219_begin_0 = const()[name = tensor("op_33219_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33219_end_0 = const()[name = tensor("op_33219_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33219_end_mask_0 = const()[name = tensor("op_33219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33219_cast_fp16 = slice_by_index(begin = var_33219_begin_0, end = var_33219_end_0, end_mask = var_33219_end_mask_0, x = var_33024_cast_fp16)[name = tensor("op_33219_cast_fp16")]; + tensor var_33226_begin_0 = const()[name = tensor("op_33226_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33226_end_0 = const()[name = tensor("op_33226_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33226_end_mask_0 = const()[name = tensor("op_33226_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33226_cast_fp16 = slice_by_index(begin = var_33226_begin_0, end = var_33226_end_0, end_mask = var_33226_end_mask_0, x = var_33024_cast_fp16)[name = tensor("op_33226_cast_fp16")]; + tensor var_33233_begin_0 = const()[name = tensor("op_33233_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33233_end_0 = const()[name = tensor("op_33233_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33233_end_mask_0 = const()[name = tensor("op_33233_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33233_cast_fp16 = slice_by_index(begin = var_33233_begin_0, end = var_33233_end_0, end_mask = var_33233_end_mask_0, x = var_33028_cast_fp16)[name = tensor("op_33233_cast_fp16")]; + tensor var_33240_begin_0 = const()[name = tensor("op_33240_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33240_end_0 = const()[name = tensor("op_33240_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33240_end_mask_0 = const()[name = tensor("op_33240_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33240_cast_fp16 = slice_by_index(begin = var_33240_begin_0, end = var_33240_end_0, end_mask = var_33240_end_mask_0, x = var_33028_cast_fp16)[name = tensor("op_33240_cast_fp16")]; + tensor var_33247_begin_0 = const()[name = tensor("op_33247_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33247_end_0 = const()[name = tensor("op_33247_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33247_end_mask_0 = const()[name = tensor("op_33247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33247_cast_fp16 = slice_by_index(begin = var_33247_begin_0, end = var_33247_end_0, end_mask = var_33247_end_mask_0, x = var_33028_cast_fp16)[name = tensor("op_33247_cast_fp16")]; + tensor var_33254_begin_0 = const()[name = tensor("op_33254_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33254_end_0 = const()[name = tensor("op_33254_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33254_end_mask_0 = const()[name = tensor("op_33254_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33254_cast_fp16 = slice_by_index(begin = var_33254_begin_0, end = var_33254_end_0, end_mask = var_33254_end_mask_0, x = var_33028_cast_fp16)[name = tensor("op_33254_cast_fp16")]; + tensor var_33261_begin_0 = const()[name = tensor("op_33261_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33261_end_0 = const()[name = tensor("op_33261_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33261_end_mask_0 = const()[name = tensor("op_33261_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33261_cast_fp16 = slice_by_index(begin = var_33261_begin_0, end = var_33261_end_0, end_mask = var_33261_end_mask_0, x = var_33032_cast_fp16)[name = tensor("op_33261_cast_fp16")]; + tensor var_33268_begin_0 = const()[name = tensor("op_33268_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33268_end_0 = const()[name = tensor("op_33268_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33268_end_mask_0 = const()[name = tensor("op_33268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33268_cast_fp16 = slice_by_index(begin = var_33268_begin_0, end = var_33268_end_0, end_mask = var_33268_end_mask_0, x = var_33032_cast_fp16)[name = tensor("op_33268_cast_fp16")]; + tensor var_33275_begin_0 = const()[name = tensor("op_33275_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33275_end_0 = const()[name = tensor("op_33275_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33275_end_mask_0 = const()[name = tensor("op_33275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33275_cast_fp16 = slice_by_index(begin = var_33275_begin_0, end = var_33275_end_0, end_mask = var_33275_end_mask_0, x = var_33032_cast_fp16)[name = tensor("op_33275_cast_fp16")]; + tensor var_33282_begin_0 = const()[name = tensor("op_33282_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33282_end_0 = const()[name = tensor("op_33282_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33282_end_mask_0 = const()[name = tensor("op_33282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33282_cast_fp16 = slice_by_index(begin = var_33282_begin_0, end = var_33282_end_0, end_mask = var_33282_end_mask_0, x = var_33032_cast_fp16)[name = tensor("op_33282_cast_fp16")]; + tensor var_33289_begin_0 = const()[name = tensor("op_33289_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33289_end_0 = const()[name = tensor("op_33289_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33289_end_mask_0 = const()[name = tensor("op_33289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33289_cast_fp16 = slice_by_index(begin = var_33289_begin_0, end = var_33289_end_0, end_mask = var_33289_end_mask_0, x = var_33036_cast_fp16)[name = tensor("op_33289_cast_fp16")]; + tensor var_33296_begin_0 = const()[name = tensor("op_33296_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33296_end_0 = const()[name = tensor("op_33296_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33296_end_mask_0 = const()[name = tensor("op_33296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33296_cast_fp16 = slice_by_index(begin = var_33296_begin_0, end = var_33296_end_0, end_mask = var_33296_end_mask_0, x = var_33036_cast_fp16)[name = tensor("op_33296_cast_fp16")]; + tensor var_33303_begin_0 = const()[name = tensor("op_33303_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33303_end_0 = const()[name = tensor("op_33303_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33303_end_mask_0 = const()[name = tensor("op_33303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33303_cast_fp16 = slice_by_index(begin = var_33303_begin_0, end = var_33303_end_0, end_mask = var_33303_end_mask_0, x = var_33036_cast_fp16)[name = tensor("op_33303_cast_fp16")]; + tensor var_33310_begin_0 = const()[name = tensor("op_33310_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33310_end_0 = const()[name = tensor("op_33310_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33310_end_mask_0 = const()[name = tensor("op_33310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33310_cast_fp16 = slice_by_index(begin = var_33310_begin_0, end = var_33310_end_0, end_mask = var_33310_end_mask_0, x = var_33036_cast_fp16)[name = tensor("op_33310_cast_fp16")]; + tensor var_33317_begin_0 = const()[name = tensor("op_33317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33317_end_0 = const()[name = tensor("op_33317_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33317_end_mask_0 = const()[name = tensor("op_33317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33317_cast_fp16 = slice_by_index(begin = var_33317_begin_0, end = var_33317_end_0, end_mask = var_33317_end_mask_0, x = var_33040_cast_fp16)[name = tensor("op_33317_cast_fp16")]; + tensor var_33324_begin_0 = const()[name = tensor("op_33324_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33324_end_0 = const()[name = tensor("op_33324_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33324_end_mask_0 = const()[name = tensor("op_33324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33324_cast_fp16 = slice_by_index(begin = var_33324_begin_0, end = var_33324_end_0, end_mask = var_33324_end_mask_0, x = var_33040_cast_fp16)[name = tensor("op_33324_cast_fp16")]; + tensor var_33331_begin_0 = const()[name = tensor("op_33331_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33331_end_0 = const()[name = tensor("op_33331_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33331_end_mask_0 = const()[name = tensor("op_33331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33331_cast_fp16 = slice_by_index(begin = var_33331_begin_0, end = var_33331_end_0, end_mask = var_33331_end_mask_0, x = var_33040_cast_fp16)[name = tensor("op_33331_cast_fp16")]; + tensor var_33338_begin_0 = const()[name = tensor("op_33338_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33338_end_0 = const()[name = tensor("op_33338_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33338_end_mask_0 = const()[name = tensor("op_33338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33338_cast_fp16 = slice_by_index(begin = var_33338_begin_0, end = var_33338_end_0, end_mask = var_33338_end_mask_0, x = var_33040_cast_fp16)[name = tensor("op_33338_cast_fp16")]; + tensor var_33345_begin_0 = const()[name = tensor("op_33345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33345_end_0 = const()[name = tensor("op_33345_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33345_end_mask_0 = const()[name = tensor("op_33345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33345_cast_fp16 = slice_by_index(begin = var_33345_begin_0, end = var_33345_end_0, end_mask = var_33345_end_mask_0, x = var_33044_cast_fp16)[name = tensor("op_33345_cast_fp16")]; + tensor var_33352_begin_0 = const()[name = tensor("op_33352_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33352_end_0 = const()[name = tensor("op_33352_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33352_end_mask_0 = const()[name = tensor("op_33352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33352_cast_fp16 = slice_by_index(begin = var_33352_begin_0, end = var_33352_end_0, end_mask = var_33352_end_mask_0, x = var_33044_cast_fp16)[name = tensor("op_33352_cast_fp16")]; + tensor var_33359_begin_0 = const()[name = tensor("op_33359_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33359_end_0 = const()[name = tensor("op_33359_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33359_end_mask_0 = const()[name = tensor("op_33359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33359_cast_fp16 = slice_by_index(begin = var_33359_begin_0, end = var_33359_end_0, end_mask = var_33359_end_mask_0, x = var_33044_cast_fp16)[name = tensor("op_33359_cast_fp16")]; + tensor var_33366_begin_0 = const()[name = tensor("op_33366_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33366_end_0 = const()[name = tensor("op_33366_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33366_end_mask_0 = const()[name = tensor("op_33366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33366_cast_fp16 = slice_by_index(begin = var_33366_begin_0, end = var_33366_end_0, end_mask = var_33366_end_mask_0, x = var_33044_cast_fp16)[name = tensor("op_33366_cast_fp16")]; + tensor var_33373_begin_0 = const()[name = tensor("op_33373_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33373_end_0 = const()[name = tensor("op_33373_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33373_end_mask_0 = const()[name = tensor("op_33373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33373_cast_fp16 = slice_by_index(begin = var_33373_begin_0, end = var_33373_end_0, end_mask = var_33373_end_mask_0, x = var_33048_cast_fp16)[name = tensor("op_33373_cast_fp16")]; + tensor var_33380_begin_0 = const()[name = tensor("op_33380_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33380_end_0 = const()[name = tensor("op_33380_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33380_end_mask_0 = const()[name = tensor("op_33380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33380_cast_fp16 = slice_by_index(begin = var_33380_begin_0, end = var_33380_end_0, end_mask = var_33380_end_mask_0, x = var_33048_cast_fp16)[name = tensor("op_33380_cast_fp16")]; + tensor var_33387_begin_0 = const()[name = tensor("op_33387_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33387_end_0 = const()[name = tensor("op_33387_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33387_end_mask_0 = const()[name = tensor("op_33387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33387_cast_fp16 = slice_by_index(begin = var_33387_begin_0, end = var_33387_end_0, end_mask = var_33387_end_mask_0, x = var_33048_cast_fp16)[name = tensor("op_33387_cast_fp16")]; + tensor var_33394_begin_0 = const()[name = tensor("op_33394_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33394_end_0 = const()[name = tensor("op_33394_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33394_end_mask_0 = const()[name = tensor("op_33394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33394_cast_fp16 = slice_by_index(begin = var_33394_begin_0, end = var_33394_end_0, end_mask = var_33394_end_mask_0, x = var_33048_cast_fp16)[name = tensor("op_33394_cast_fp16")]; + tensor var_33401_begin_0 = const()[name = tensor("op_33401_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33401_end_0 = const()[name = tensor("op_33401_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33401_end_mask_0 = const()[name = tensor("op_33401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33401_cast_fp16 = slice_by_index(begin = var_33401_begin_0, end = var_33401_end_0, end_mask = var_33401_end_mask_0, x = var_33052_cast_fp16)[name = tensor("op_33401_cast_fp16")]; + tensor var_33408_begin_0 = const()[name = tensor("op_33408_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33408_end_0 = const()[name = tensor("op_33408_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33408_end_mask_0 = const()[name = tensor("op_33408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33408_cast_fp16 = slice_by_index(begin = var_33408_begin_0, end = var_33408_end_0, end_mask = var_33408_end_mask_0, x = var_33052_cast_fp16)[name = tensor("op_33408_cast_fp16")]; + tensor var_33415_begin_0 = const()[name = tensor("op_33415_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33415_end_0 = const()[name = tensor("op_33415_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33415_end_mask_0 = const()[name = tensor("op_33415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33415_cast_fp16 = slice_by_index(begin = var_33415_begin_0, end = var_33415_end_0, end_mask = var_33415_end_mask_0, x = var_33052_cast_fp16)[name = tensor("op_33415_cast_fp16")]; + tensor var_33422_begin_0 = const()[name = tensor("op_33422_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33422_end_0 = const()[name = tensor("op_33422_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33422_end_mask_0 = const()[name = tensor("op_33422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33422_cast_fp16 = slice_by_index(begin = var_33422_begin_0, end = var_33422_end_0, end_mask = var_33422_end_mask_0, x = var_33052_cast_fp16)[name = tensor("op_33422_cast_fp16")]; + tensor var_33429_begin_0 = const()[name = tensor("op_33429_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33429_end_0 = const()[name = tensor("op_33429_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33429_end_mask_0 = const()[name = tensor("op_33429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33429_cast_fp16 = slice_by_index(begin = var_33429_begin_0, end = var_33429_end_0, end_mask = var_33429_end_mask_0, x = var_33056_cast_fp16)[name = tensor("op_33429_cast_fp16")]; + tensor var_33436_begin_0 = const()[name = tensor("op_33436_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33436_end_0 = const()[name = tensor("op_33436_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33436_end_mask_0 = const()[name = tensor("op_33436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33436_cast_fp16 = slice_by_index(begin = var_33436_begin_0, end = var_33436_end_0, end_mask = var_33436_end_mask_0, x = var_33056_cast_fp16)[name = tensor("op_33436_cast_fp16")]; + tensor var_33443_begin_0 = const()[name = tensor("op_33443_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33443_end_0 = const()[name = tensor("op_33443_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33443_end_mask_0 = const()[name = tensor("op_33443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33443_cast_fp16 = slice_by_index(begin = var_33443_begin_0, end = var_33443_end_0, end_mask = var_33443_end_mask_0, x = var_33056_cast_fp16)[name = tensor("op_33443_cast_fp16")]; + tensor var_33450_begin_0 = const()[name = tensor("op_33450_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33450_end_0 = const()[name = tensor("op_33450_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33450_end_mask_0 = const()[name = tensor("op_33450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33450_cast_fp16 = slice_by_index(begin = var_33450_begin_0, end = var_33450_end_0, end_mask = var_33450_end_mask_0, x = var_33056_cast_fp16)[name = tensor("op_33450_cast_fp16")]; + tensor var_33457_begin_0 = const()[name = tensor("op_33457_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33457_end_0 = const()[name = tensor("op_33457_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33457_end_mask_0 = const()[name = tensor("op_33457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33457_cast_fp16 = slice_by_index(begin = var_33457_begin_0, end = var_33457_end_0, end_mask = var_33457_end_mask_0, x = var_33060_cast_fp16)[name = tensor("op_33457_cast_fp16")]; + tensor var_33464_begin_0 = const()[name = tensor("op_33464_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33464_end_0 = const()[name = tensor("op_33464_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33464_end_mask_0 = const()[name = tensor("op_33464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33464_cast_fp16 = slice_by_index(begin = var_33464_begin_0, end = var_33464_end_0, end_mask = var_33464_end_mask_0, x = var_33060_cast_fp16)[name = tensor("op_33464_cast_fp16")]; + tensor var_33471_begin_0 = const()[name = tensor("op_33471_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33471_end_0 = const()[name = tensor("op_33471_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33471_end_mask_0 = const()[name = tensor("op_33471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33471_cast_fp16 = slice_by_index(begin = var_33471_begin_0, end = var_33471_end_0, end_mask = var_33471_end_mask_0, x = var_33060_cast_fp16)[name = tensor("op_33471_cast_fp16")]; + tensor var_33478_begin_0 = const()[name = tensor("op_33478_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33478_end_0 = const()[name = tensor("op_33478_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33478_end_mask_0 = const()[name = tensor("op_33478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33478_cast_fp16 = slice_by_index(begin = var_33478_begin_0, end = var_33478_end_0, end_mask = var_33478_end_mask_0, x = var_33060_cast_fp16)[name = tensor("op_33478_cast_fp16")]; + tensor var_33485_begin_0 = const()[name = tensor("op_33485_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33485_end_0 = const()[name = tensor("op_33485_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33485_end_mask_0 = const()[name = tensor("op_33485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33485_cast_fp16 = slice_by_index(begin = var_33485_begin_0, end = var_33485_end_0, end_mask = var_33485_end_mask_0, x = var_33064_cast_fp16)[name = tensor("op_33485_cast_fp16")]; + tensor var_33492_begin_0 = const()[name = tensor("op_33492_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33492_end_0 = const()[name = tensor("op_33492_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33492_end_mask_0 = const()[name = tensor("op_33492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33492_cast_fp16 = slice_by_index(begin = var_33492_begin_0, end = var_33492_end_0, end_mask = var_33492_end_mask_0, x = var_33064_cast_fp16)[name = tensor("op_33492_cast_fp16")]; + tensor var_33499_begin_0 = const()[name = tensor("op_33499_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33499_end_0 = const()[name = tensor("op_33499_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33499_end_mask_0 = const()[name = tensor("op_33499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33499_cast_fp16 = slice_by_index(begin = var_33499_begin_0, end = var_33499_end_0, end_mask = var_33499_end_mask_0, x = var_33064_cast_fp16)[name = tensor("op_33499_cast_fp16")]; + tensor var_33506_begin_0 = const()[name = tensor("op_33506_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33506_end_0 = const()[name = tensor("op_33506_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33506_end_mask_0 = const()[name = tensor("op_33506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33506_cast_fp16 = slice_by_index(begin = var_33506_begin_0, end = var_33506_end_0, end_mask = var_33506_end_mask_0, x = var_33064_cast_fp16)[name = tensor("op_33506_cast_fp16")]; + tensor var_33513_begin_0 = const()[name = tensor("op_33513_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33513_end_0 = const()[name = tensor("op_33513_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33513_end_mask_0 = const()[name = tensor("op_33513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33513_cast_fp16 = slice_by_index(begin = var_33513_begin_0, end = var_33513_end_0, end_mask = var_33513_end_mask_0, x = var_33068_cast_fp16)[name = tensor("op_33513_cast_fp16")]; + tensor var_33520_begin_0 = const()[name = tensor("op_33520_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33520_end_0 = const()[name = tensor("op_33520_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33520_end_mask_0 = const()[name = tensor("op_33520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33520_cast_fp16 = slice_by_index(begin = var_33520_begin_0, end = var_33520_end_0, end_mask = var_33520_end_mask_0, x = var_33068_cast_fp16)[name = tensor("op_33520_cast_fp16")]; + tensor var_33527_begin_0 = const()[name = tensor("op_33527_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33527_end_0 = const()[name = tensor("op_33527_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33527_end_mask_0 = const()[name = tensor("op_33527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33527_cast_fp16 = slice_by_index(begin = var_33527_begin_0, end = var_33527_end_0, end_mask = var_33527_end_mask_0, x = var_33068_cast_fp16)[name = tensor("op_33527_cast_fp16")]; + tensor var_33534_begin_0 = const()[name = tensor("op_33534_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33534_end_0 = const()[name = tensor("op_33534_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33534_end_mask_0 = const()[name = tensor("op_33534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33534_cast_fp16 = slice_by_index(begin = var_33534_begin_0, end = var_33534_end_0, end_mask = var_33534_end_mask_0, x = var_33068_cast_fp16)[name = tensor("op_33534_cast_fp16")]; + tensor var_33541_begin_0 = const()[name = tensor("op_33541_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33541_end_0 = const()[name = tensor("op_33541_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33541_end_mask_0 = const()[name = tensor("op_33541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33541_cast_fp16 = slice_by_index(begin = var_33541_begin_0, end = var_33541_end_0, end_mask = var_33541_end_mask_0, x = var_33072_cast_fp16)[name = tensor("op_33541_cast_fp16")]; + tensor var_33548_begin_0 = const()[name = tensor("op_33548_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33548_end_0 = const()[name = tensor("op_33548_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33548_end_mask_0 = const()[name = tensor("op_33548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33548_cast_fp16 = slice_by_index(begin = var_33548_begin_0, end = var_33548_end_0, end_mask = var_33548_end_mask_0, x = var_33072_cast_fp16)[name = tensor("op_33548_cast_fp16")]; + tensor var_33555_begin_0 = const()[name = tensor("op_33555_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33555_end_0 = const()[name = tensor("op_33555_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33555_end_mask_0 = const()[name = tensor("op_33555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33555_cast_fp16 = slice_by_index(begin = var_33555_begin_0, end = var_33555_end_0, end_mask = var_33555_end_mask_0, x = var_33072_cast_fp16)[name = tensor("op_33555_cast_fp16")]; + tensor var_33562_begin_0 = const()[name = tensor("op_33562_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33562_end_0 = const()[name = tensor("op_33562_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33562_end_mask_0 = const()[name = tensor("op_33562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33562_cast_fp16 = slice_by_index(begin = var_33562_begin_0, end = var_33562_end_0, end_mask = var_33562_end_mask_0, x = var_33072_cast_fp16)[name = tensor("op_33562_cast_fp16")]; + tensor var_33569_begin_0 = const()[name = tensor("op_33569_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33569_end_0 = const()[name = tensor("op_33569_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33569_end_mask_0 = const()[name = tensor("op_33569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33569_cast_fp16 = slice_by_index(begin = var_33569_begin_0, end = var_33569_end_0, end_mask = var_33569_end_mask_0, x = var_33076_cast_fp16)[name = tensor("op_33569_cast_fp16")]; + tensor var_33576_begin_0 = const()[name = tensor("op_33576_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33576_end_0 = const()[name = tensor("op_33576_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33576_end_mask_0 = const()[name = tensor("op_33576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33576_cast_fp16 = slice_by_index(begin = var_33576_begin_0, end = var_33576_end_0, end_mask = var_33576_end_mask_0, x = var_33076_cast_fp16)[name = tensor("op_33576_cast_fp16")]; + tensor var_33583_begin_0 = const()[name = tensor("op_33583_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33583_end_0 = const()[name = tensor("op_33583_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33583_end_mask_0 = const()[name = tensor("op_33583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33583_cast_fp16 = slice_by_index(begin = var_33583_begin_0, end = var_33583_end_0, end_mask = var_33583_end_mask_0, x = var_33076_cast_fp16)[name = tensor("op_33583_cast_fp16")]; + tensor var_33590_begin_0 = const()[name = tensor("op_33590_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33590_end_0 = const()[name = tensor("op_33590_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33590_end_mask_0 = const()[name = tensor("op_33590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33590_cast_fp16 = slice_by_index(begin = var_33590_begin_0, end = var_33590_end_0, end_mask = var_33590_end_mask_0, x = var_33076_cast_fp16)[name = tensor("op_33590_cast_fp16")]; + tensor var_33597_begin_0 = const()[name = tensor("op_33597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33597_end_0 = const()[name = tensor("op_33597_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33597_end_mask_0 = const()[name = tensor("op_33597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33597_cast_fp16 = slice_by_index(begin = var_33597_begin_0, end = var_33597_end_0, end_mask = var_33597_end_mask_0, x = var_33080_cast_fp16)[name = tensor("op_33597_cast_fp16")]; + tensor var_33604_begin_0 = const()[name = tensor("op_33604_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33604_end_0 = const()[name = tensor("op_33604_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33604_end_mask_0 = const()[name = tensor("op_33604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33604_cast_fp16 = slice_by_index(begin = var_33604_begin_0, end = var_33604_end_0, end_mask = var_33604_end_mask_0, x = var_33080_cast_fp16)[name = tensor("op_33604_cast_fp16")]; + tensor var_33611_begin_0 = const()[name = tensor("op_33611_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33611_end_0 = const()[name = tensor("op_33611_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33611_end_mask_0 = const()[name = tensor("op_33611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33611_cast_fp16 = slice_by_index(begin = var_33611_begin_0, end = var_33611_end_0, end_mask = var_33611_end_mask_0, x = var_33080_cast_fp16)[name = tensor("op_33611_cast_fp16")]; + tensor var_33618_begin_0 = const()[name = tensor("op_33618_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33618_end_0 = const()[name = tensor("op_33618_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33618_end_mask_0 = const()[name = tensor("op_33618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33618_cast_fp16 = slice_by_index(begin = var_33618_begin_0, end = var_33618_end_0, end_mask = var_33618_end_mask_0, x = var_33080_cast_fp16)[name = tensor("op_33618_cast_fp16")]; + tensor var_33625_begin_0 = const()[name = tensor("op_33625_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33625_end_0 = const()[name = tensor("op_33625_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_33625_end_mask_0 = const()[name = tensor("op_33625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33625_cast_fp16 = slice_by_index(begin = var_33625_begin_0, end = var_33625_end_0, end_mask = var_33625_end_mask_0, x = var_33084_cast_fp16)[name = tensor("op_33625_cast_fp16")]; + tensor var_33632_begin_0 = const()[name = tensor("op_33632_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_33632_end_0 = const()[name = tensor("op_33632_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_33632_end_mask_0 = const()[name = tensor("op_33632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33632_cast_fp16 = slice_by_index(begin = var_33632_begin_0, end = var_33632_end_0, end_mask = var_33632_end_mask_0, x = var_33084_cast_fp16)[name = tensor("op_33632_cast_fp16")]; + tensor var_33639_begin_0 = const()[name = tensor("op_33639_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_33639_end_0 = const()[name = tensor("op_33639_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_33639_end_mask_0 = const()[name = tensor("op_33639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33639_cast_fp16 = slice_by_index(begin = var_33639_begin_0, end = var_33639_end_0, end_mask = var_33639_end_mask_0, x = var_33084_cast_fp16)[name = tensor("op_33639_cast_fp16")]; + tensor var_33646_begin_0 = const()[name = tensor("op_33646_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_33646_end_0 = const()[name = tensor("op_33646_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33646_end_mask_0 = const()[name = tensor("op_33646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33646_cast_fp16 = slice_by_index(begin = var_33646_begin_0, end = var_33646_end_0, end_mask = var_33646_end_mask_0, x = var_33084_cast_fp16)[name = tensor("op_33646_cast_fp16")]; + tensor k_41_perm_0 = const()[name = tensor("k_41_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_33651_begin_0 = const()[name = tensor("op_33651_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33651_end_0 = const()[name = tensor("op_33651_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_33651_end_mask_0 = const()[name = tensor("op_33651_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_11 = transpose(perm = k_41_perm_0, x = key_41_cast_fp16)[name = tensor("transpose_11")]; + tensor var_33651_cast_fp16 = slice_by_index(begin = var_33651_begin_0, end = var_33651_end_0, end_mask = var_33651_end_mask_0, x = transpose_11)[name = tensor("op_33651_cast_fp16")]; + tensor var_33655_begin_0 = const()[name = tensor("op_33655_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_33655_end_0 = const()[name = tensor("op_33655_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_33655_end_mask_0 = const()[name = tensor("op_33655_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33655_cast_fp16 = slice_by_index(begin = var_33655_begin_0, end = var_33655_end_0, end_mask = var_33655_end_mask_0, x = transpose_11)[name = tensor("op_33655_cast_fp16")]; + tensor var_33659_begin_0 = const()[name = tensor("op_33659_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_33659_end_0 = const()[name = tensor("op_33659_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_33659_end_mask_0 = const()[name = tensor("op_33659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33659_cast_fp16 = slice_by_index(begin = var_33659_begin_0, end = var_33659_end_0, end_mask = var_33659_end_mask_0, x = transpose_11)[name = tensor("op_33659_cast_fp16")]; + tensor var_33663_begin_0 = const()[name = tensor("op_33663_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_33663_end_0 = const()[name = tensor("op_33663_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_33663_end_mask_0 = const()[name = tensor("op_33663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33663_cast_fp16 = slice_by_index(begin = var_33663_begin_0, end = var_33663_end_0, end_mask = var_33663_end_mask_0, x = transpose_11)[name = tensor("op_33663_cast_fp16")]; + tensor var_33667_begin_0 = const()[name = tensor("op_33667_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_33667_end_0 = const()[name = tensor("op_33667_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_33667_end_mask_0 = const()[name = tensor("op_33667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33667_cast_fp16 = slice_by_index(begin = var_33667_begin_0, end = var_33667_end_0, end_mask = var_33667_end_mask_0, x = transpose_11)[name = tensor("op_33667_cast_fp16")]; + tensor var_33671_begin_0 = const()[name = tensor("op_33671_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_33671_end_0 = const()[name = tensor("op_33671_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_33671_end_mask_0 = const()[name = tensor("op_33671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33671_cast_fp16 = slice_by_index(begin = var_33671_begin_0, end = var_33671_end_0, end_mask = var_33671_end_mask_0, x = transpose_11)[name = tensor("op_33671_cast_fp16")]; + tensor var_33675_begin_0 = const()[name = tensor("op_33675_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_33675_end_0 = const()[name = tensor("op_33675_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_33675_end_mask_0 = const()[name = tensor("op_33675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33675_cast_fp16 = slice_by_index(begin = var_33675_begin_0, end = var_33675_end_0, end_mask = var_33675_end_mask_0, x = transpose_11)[name = tensor("op_33675_cast_fp16")]; + tensor var_33679_begin_0 = const()[name = tensor("op_33679_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_33679_end_0 = const()[name = tensor("op_33679_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_33679_end_mask_0 = const()[name = tensor("op_33679_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33679_cast_fp16 = slice_by_index(begin = var_33679_begin_0, end = var_33679_end_0, end_mask = var_33679_end_mask_0, x = transpose_11)[name = tensor("op_33679_cast_fp16")]; + tensor var_33683_begin_0 = const()[name = tensor("op_33683_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_33683_end_0 = const()[name = tensor("op_33683_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_33683_end_mask_0 = const()[name = tensor("op_33683_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33683_cast_fp16 = slice_by_index(begin = var_33683_begin_0, end = var_33683_end_0, end_mask = var_33683_end_mask_0, x = transpose_11)[name = tensor("op_33683_cast_fp16")]; + tensor var_33687_begin_0 = const()[name = tensor("op_33687_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_33687_end_0 = const()[name = tensor("op_33687_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_33687_end_mask_0 = const()[name = tensor("op_33687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33687_cast_fp16 = slice_by_index(begin = var_33687_begin_0, end = var_33687_end_0, end_mask = var_33687_end_mask_0, x = transpose_11)[name = tensor("op_33687_cast_fp16")]; + tensor var_33691_begin_0 = const()[name = tensor("op_33691_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_33691_end_0 = const()[name = tensor("op_33691_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_33691_end_mask_0 = const()[name = tensor("op_33691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33691_cast_fp16 = slice_by_index(begin = var_33691_begin_0, end = var_33691_end_0, end_mask = var_33691_end_mask_0, x = transpose_11)[name = tensor("op_33691_cast_fp16")]; + tensor var_33695_begin_0 = const()[name = tensor("op_33695_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_33695_end_0 = const()[name = tensor("op_33695_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_33695_end_mask_0 = const()[name = tensor("op_33695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33695_cast_fp16 = slice_by_index(begin = var_33695_begin_0, end = var_33695_end_0, end_mask = var_33695_end_mask_0, x = transpose_11)[name = tensor("op_33695_cast_fp16")]; + tensor var_33699_begin_0 = const()[name = tensor("op_33699_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_33699_end_0 = const()[name = tensor("op_33699_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_33699_end_mask_0 = const()[name = tensor("op_33699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33699_cast_fp16 = slice_by_index(begin = var_33699_begin_0, end = var_33699_end_0, end_mask = var_33699_end_mask_0, x = transpose_11)[name = tensor("op_33699_cast_fp16")]; + tensor var_33703_begin_0 = const()[name = tensor("op_33703_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_33703_end_0 = const()[name = tensor("op_33703_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_33703_end_mask_0 = const()[name = tensor("op_33703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33703_cast_fp16 = slice_by_index(begin = var_33703_begin_0, end = var_33703_end_0, end_mask = var_33703_end_mask_0, x = transpose_11)[name = tensor("op_33703_cast_fp16")]; + tensor var_33707_begin_0 = const()[name = tensor("op_33707_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_33707_end_0 = const()[name = tensor("op_33707_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_33707_end_mask_0 = const()[name = tensor("op_33707_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33707_cast_fp16 = slice_by_index(begin = var_33707_begin_0, end = var_33707_end_0, end_mask = var_33707_end_mask_0, x = transpose_11)[name = tensor("op_33707_cast_fp16")]; + tensor var_33711_begin_0 = const()[name = tensor("op_33711_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_33711_end_0 = const()[name = tensor("op_33711_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_33711_end_mask_0 = const()[name = tensor("op_33711_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33711_cast_fp16 = slice_by_index(begin = var_33711_begin_0, end = var_33711_end_0, end_mask = var_33711_end_mask_0, x = transpose_11)[name = tensor("op_33711_cast_fp16")]; + tensor var_33715_begin_0 = const()[name = tensor("op_33715_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_33715_end_0 = const()[name = tensor("op_33715_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_33715_end_mask_0 = const()[name = tensor("op_33715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33715_cast_fp16 = slice_by_index(begin = var_33715_begin_0, end = var_33715_end_0, end_mask = var_33715_end_mask_0, x = transpose_11)[name = tensor("op_33715_cast_fp16")]; + tensor var_33719_begin_0 = const()[name = tensor("op_33719_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_33719_end_0 = const()[name = tensor("op_33719_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_33719_end_mask_0 = const()[name = tensor("op_33719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33719_cast_fp16 = slice_by_index(begin = var_33719_begin_0, end = var_33719_end_0, end_mask = var_33719_end_mask_0, x = transpose_11)[name = tensor("op_33719_cast_fp16")]; + tensor var_33723_begin_0 = const()[name = tensor("op_33723_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_33723_end_0 = const()[name = tensor("op_33723_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_33723_end_mask_0 = const()[name = tensor("op_33723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33723_cast_fp16 = slice_by_index(begin = var_33723_begin_0, end = var_33723_end_0, end_mask = var_33723_end_mask_0, x = transpose_11)[name = tensor("op_33723_cast_fp16")]; + tensor var_33727_begin_0 = const()[name = tensor("op_33727_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_33727_end_0 = const()[name = tensor("op_33727_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_33727_end_mask_0 = const()[name = tensor("op_33727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_33727_cast_fp16 = slice_by_index(begin = var_33727_begin_0, end = var_33727_end_0, end_mask = var_33727_end_mask_0, x = transpose_11)[name = tensor("op_33727_cast_fp16")]; + tensor var_33729_begin_0 = const()[name = tensor("op_33729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_33729_end_0 = const()[name = tensor("op_33729_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_33729_end_mask_0 = const()[name = tensor("op_33729_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33729_cast_fp16 = slice_by_index(begin = var_33729_begin_0, end = var_33729_end_0, end_mask = var_33729_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33729_cast_fp16")]; + tensor var_33733_begin_0 = const()[name = tensor("op_33733_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_33733_end_0 = const()[name = tensor("op_33733_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_33733_end_mask_0 = const()[name = tensor("op_33733_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33733_cast_fp16 = slice_by_index(begin = var_33733_begin_0, end = var_33733_end_0, end_mask = var_33733_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33733_cast_fp16")]; + tensor var_33737_begin_0 = const()[name = tensor("op_33737_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_33737_end_0 = const()[name = tensor("op_33737_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_33737_end_mask_0 = const()[name = tensor("op_33737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33737_cast_fp16 = slice_by_index(begin = var_33737_begin_0, end = var_33737_end_0, end_mask = var_33737_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33737_cast_fp16")]; + tensor var_33741_begin_0 = const()[name = tensor("op_33741_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_33741_end_0 = const()[name = tensor("op_33741_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_33741_end_mask_0 = const()[name = tensor("op_33741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33741_cast_fp16 = slice_by_index(begin = var_33741_begin_0, end = var_33741_end_0, end_mask = var_33741_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33741_cast_fp16")]; + tensor var_33745_begin_0 = const()[name = tensor("op_33745_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_33745_end_0 = const()[name = tensor("op_33745_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_33745_end_mask_0 = const()[name = tensor("op_33745_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33745_cast_fp16 = slice_by_index(begin = var_33745_begin_0, end = var_33745_end_0, end_mask = var_33745_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33745_cast_fp16")]; + tensor var_33749_begin_0 = const()[name = tensor("op_33749_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_33749_end_0 = const()[name = tensor("op_33749_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_33749_end_mask_0 = const()[name = tensor("op_33749_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33749_cast_fp16 = slice_by_index(begin = var_33749_begin_0, end = var_33749_end_0, end_mask = var_33749_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33749_cast_fp16")]; + tensor var_33753_begin_0 = const()[name = tensor("op_33753_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_33753_end_0 = const()[name = tensor("op_33753_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_33753_end_mask_0 = const()[name = tensor("op_33753_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33753_cast_fp16 = slice_by_index(begin = var_33753_begin_0, end = var_33753_end_0, end_mask = var_33753_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33753_cast_fp16")]; + tensor var_33757_begin_0 = const()[name = tensor("op_33757_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_33757_end_0 = const()[name = tensor("op_33757_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_33757_end_mask_0 = const()[name = tensor("op_33757_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33757_cast_fp16 = slice_by_index(begin = var_33757_begin_0, end = var_33757_end_0, end_mask = var_33757_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33757_cast_fp16")]; + tensor var_33761_begin_0 = const()[name = tensor("op_33761_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_33761_end_0 = const()[name = tensor("op_33761_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_33761_end_mask_0 = const()[name = tensor("op_33761_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33761_cast_fp16 = slice_by_index(begin = var_33761_begin_0, end = var_33761_end_0, end_mask = var_33761_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33761_cast_fp16")]; + tensor var_33765_begin_0 = const()[name = tensor("op_33765_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_33765_end_0 = const()[name = tensor("op_33765_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_33765_end_mask_0 = const()[name = tensor("op_33765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33765_cast_fp16 = slice_by_index(begin = var_33765_begin_0, end = var_33765_end_0, end_mask = var_33765_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33765_cast_fp16")]; + tensor var_33769_begin_0 = const()[name = tensor("op_33769_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_33769_end_0 = const()[name = tensor("op_33769_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_33769_end_mask_0 = const()[name = tensor("op_33769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33769_cast_fp16 = slice_by_index(begin = var_33769_begin_0, end = var_33769_end_0, end_mask = var_33769_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33769_cast_fp16")]; + tensor var_33773_begin_0 = const()[name = tensor("op_33773_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_33773_end_0 = const()[name = tensor("op_33773_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_33773_end_mask_0 = const()[name = tensor("op_33773_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33773_cast_fp16 = slice_by_index(begin = var_33773_begin_0, end = var_33773_end_0, end_mask = var_33773_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33773_cast_fp16")]; + tensor var_33777_begin_0 = const()[name = tensor("op_33777_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_33777_end_0 = const()[name = tensor("op_33777_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_33777_end_mask_0 = const()[name = tensor("op_33777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33777_cast_fp16 = slice_by_index(begin = var_33777_begin_0, end = var_33777_end_0, end_mask = var_33777_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33777_cast_fp16")]; + tensor var_33781_begin_0 = const()[name = tensor("op_33781_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_33781_end_0 = const()[name = tensor("op_33781_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_33781_end_mask_0 = const()[name = tensor("op_33781_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33781_cast_fp16 = slice_by_index(begin = var_33781_begin_0, end = var_33781_end_0, end_mask = var_33781_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33781_cast_fp16")]; + tensor var_33785_begin_0 = const()[name = tensor("op_33785_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_33785_end_0 = const()[name = tensor("op_33785_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_33785_end_mask_0 = const()[name = tensor("op_33785_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33785_cast_fp16 = slice_by_index(begin = var_33785_begin_0, end = var_33785_end_0, end_mask = var_33785_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33785_cast_fp16")]; + tensor var_33789_begin_0 = const()[name = tensor("op_33789_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_33789_end_0 = const()[name = tensor("op_33789_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_33789_end_mask_0 = const()[name = tensor("op_33789_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33789_cast_fp16 = slice_by_index(begin = var_33789_begin_0, end = var_33789_end_0, end_mask = var_33789_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33789_cast_fp16")]; + tensor var_33793_begin_0 = const()[name = tensor("op_33793_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_33793_end_0 = const()[name = tensor("op_33793_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_33793_end_mask_0 = const()[name = tensor("op_33793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33793_cast_fp16 = slice_by_index(begin = var_33793_begin_0, end = var_33793_end_0, end_mask = var_33793_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33793_cast_fp16")]; + tensor var_33797_begin_0 = const()[name = tensor("op_33797_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_33797_end_0 = const()[name = tensor("op_33797_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_33797_end_mask_0 = const()[name = tensor("op_33797_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33797_cast_fp16 = slice_by_index(begin = var_33797_begin_0, end = var_33797_end_0, end_mask = var_33797_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33797_cast_fp16")]; + tensor var_33801_begin_0 = const()[name = tensor("op_33801_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_33801_end_0 = const()[name = tensor("op_33801_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_33801_end_mask_0 = const()[name = tensor("op_33801_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33801_cast_fp16 = slice_by_index(begin = var_33801_begin_0, end = var_33801_end_0, end_mask = var_33801_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33801_cast_fp16")]; + tensor var_33805_begin_0 = const()[name = tensor("op_33805_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_33805_end_0 = const()[name = tensor("op_33805_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_33805_end_mask_0 = const()[name = tensor("op_33805_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_33805_cast_fp16 = slice_by_index(begin = var_33805_begin_0, end = var_33805_end_0, end_mask = var_33805_end_mask_0, x = value_41_cast_fp16)[name = tensor("op_33805_cast_fp16")]; + tensor var_33809_equation_0 = const()[name = tensor("op_33809_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33809_cast_fp16 = einsum(equation = var_33809_equation_0, values = (var_33651_cast_fp16, var_33093_cast_fp16))[name = tensor("op_33809_cast_fp16")]; + tensor var_33810_to_fp16 = const()[name = tensor("op_33810_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3201_cast_fp16 = mul(x = var_33809_cast_fp16, y = var_33810_to_fp16)[name = tensor("aw_chunk_3201_cast_fp16")]; + tensor var_33813_equation_0 = const()[name = tensor("op_33813_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33813_cast_fp16 = einsum(equation = var_33813_equation_0, values = (var_33651_cast_fp16, var_33100_cast_fp16))[name = tensor("op_33813_cast_fp16")]; + tensor var_33814_to_fp16 = const()[name = tensor("op_33814_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3203_cast_fp16 = mul(x = var_33813_cast_fp16, y = var_33814_to_fp16)[name = tensor("aw_chunk_3203_cast_fp16")]; + tensor var_33817_equation_0 = const()[name = tensor("op_33817_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33817_cast_fp16 = einsum(equation = var_33817_equation_0, values = (var_33651_cast_fp16, var_33107_cast_fp16))[name = tensor("op_33817_cast_fp16")]; + tensor var_33818_to_fp16 = const()[name = tensor("op_33818_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3205_cast_fp16 = mul(x = var_33817_cast_fp16, y = var_33818_to_fp16)[name = tensor("aw_chunk_3205_cast_fp16")]; + tensor var_33821_equation_0 = const()[name = tensor("op_33821_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33821_cast_fp16 = einsum(equation = var_33821_equation_0, values = (var_33651_cast_fp16, var_33114_cast_fp16))[name = tensor("op_33821_cast_fp16")]; + tensor var_33822_to_fp16 = const()[name = tensor("op_33822_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3207_cast_fp16 = mul(x = var_33821_cast_fp16, y = var_33822_to_fp16)[name = tensor("aw_chunk_3207_cast_fp16")]; + tensor var_33825_equation_0 = const()[name = tensor("op_33825_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33825_cast_fp16 = einsum(equation = var_33825_equation_0, values = (var_33655_cast_fp16, var_33121_cast_fp16))[name = tensor("op_33825_cast_fp16")]; + tensor var_33826_to_fp16 = const()[name = tensor("op_33826_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3209_cast_fp16 = mul(x = var_33825_cast_fp16, y = var_33826_to_fp16)[name = tensor("aw_chunk_3209_cast_fp16")]; + tensor var_33829_equation_0 = const()[name = tensor("op_33829_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33829_cast_fp16 = einsum(equation = var_33829_equation_0, values = (var_33655_cast_fp16, var_33128_cast_fp16))[name = tensor("op_33829_cast_fp16")]; + tensor var_33830_to_fp16 = const()[name = tensor("op_33830_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3211_cast_fp16 = mul(x = var_33829_cast_fp16, y = var_33830_to_fp16)[name = tensor("aw_chunk_3211_cast_fp16")]; + tensor var_33833_equation_0 = const()[name = tensor("op_33833_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33833_cast_fp16 = einsum(equation = var_33833_equation_0, values = (var_33655_cast_fp16, var_33135_cast_fp16))[name = tensor("op_33833_cast_fp16")]; + tensor var_33834_to_fp16 = const()[name = tensor("op_33834_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3213_cast_fp16 = mul(x = var_33833_cast_fp16, y = var_33834_to_fp16)[name = tensor("aw_chunk_3213_cast_fp16")]; + tensor var_33837_equation_0 = const()[name = tensor("op_33837_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33837_cast_fp16 = einsum(equation = var_33837_equation_0, values = (var_33655_cast_fp16, var_33142_cast_fp16))[name = tensor("op_33837_cast_fp16")]; + tensor var_33838_to_fp16 = const()[name = tensor("op_33838_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3215_cast_fp16 = mul(x = var_33837_cast_fp16, y = var_33838_to_fp16)[name = tensor("aw_chunk_3215_cast_fp16")]; + tensor var_33841_equation_0 = const()[name = tensor("op_33841_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33841_cast_fp16 = einsum(equation = var_33841_equation_0, values = (var_33659_cast_fp16, var_33149_cast_fp16))[name = tensor("op_33841_cast_fp16")]; + tensor var_33842_to_fp16 = const()[name = tensor("op_33842_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3217_cast_fp16 = mul(x = var_33841_cast_fp16, y = var_33842_to_fp16)[name = tensor("aw_chunk_3217_cast_fp16")]; + tensor var_33845_equation_0 = const()[name = tensor("op_33845_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33845_cast_fp16 = einsum(equation = var_33845_equation_0, values = (var_33659_cast_fp16, var_33156_cast_fp16))[name = tensor("op_33845_cast_fp16")]; + tensor var_33846_to_fp16 = const()[name = tensor("op_33846_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3219_cast_fp16 = mul(x = var_33845_cast_fp16, y = var_33846_to_fp16)[name = tensor("aw_chunk_3219_cast_fp16")]; + tensor var_33849_equation_0 = const()[name = tensor("op_33849_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33849_cast_fp16 = einsum(equation = var_33849_equation_0, values = (var_33659_cast_fp16, var_33163_cast_fp16))[name = tensor("op_33849_cast_fp16")]; + tensor var_33850_to_fp16 = const()[name = tensor("op_33850_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3221_cast_fp16 = mul(x = var_33849_cast_fp16, y = var_33850_to_fp16)[name = tensor("aw_chunk_3221_cast_fp16")]; + tensor var_33853_equation_0 = const()[name = tensor("op_33853_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33853_cast_fp16 = einsum(equation = var_33853_equation_0, values = (var_33659_cast_fp16, var_33170_cast_fp16))[name = tensor("op_33853_cast_fp16")]; + tensor var_33854_to_fp16 = const()[name = tensor("op_33854_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3223_cast_fp16 = mul(x = var_33853_cast_fp16, y = var_33854_to_fp16)[name = tensor("aw_chunk_3223_cast_fp16")]; + tensor var_33857_equation_0 = const()[name = tensor("op_33857_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33857_cast_fp16 = einsum(equation = var_33857_equation_0, values = (var_33663_cast_fp16, var_33177_cast_fp16))[name = tensor("op_33857_cast_fp16")]; + tensor var_33858_to_fp16 = const()[name = tensor("op_33858_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3225_cast_fp16 = mul(x = var_33857_cast_fp16, y = var_33858_to_fp16)[name = tensor("aw_chunk_3225_cast_fp16")]; + tensor var_33861_equation_0 = const()[name = tensor("op_33861_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33861_cast_fp16 = einsum(equation = var_33861_equation_0, values = (var_33663_cast_fp16, var_33184_cast_fp16))[name = tensor("op_33861_cast_fp16")]; + tensor var_33862_to_fp16 = const()[name = tensor("op_33862_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3227_cast_fp16 = mul(x = var_33861_cast_fp16, y = var_33862_to_fp16)[name = tensor("aw_chunk_3227_cast_fp16")]; + tensor var_33865_equation_0 = const()[name = tensor("op_33865_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33865_cast_fp16 = einsum(equation = var_33865_equation_0, values = (var_33663_cast_fp16, var_33191_cast_fp16))[name = tensor("op_33865_cast_fp16")]; + tensor var_33866_to_fp16 = const()[name = tensor("op_33866_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3229_cast_fp16 = mul(x = var_33865_cast_fp16, y = var_33866_to_fp16)[name = tensor("aw_chunk_3229_cast_fp16")]; + tensor var_33869_equation_0 = const()[name = tensor("op_33869_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33869_cast_fp16 = einsum(equation = var_33869_equation_0, values = (var_33663_cast_fp16, var_33198_cast_fp16))[name = tensor("op_33869_cast_fp16")]; + tensor var_33870_to_fp16 = const()[name = tensor("op_33870_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3231_cast_fp16 = mul(x = var_33869_cast_fp16, y = var_33870_to_fp16)[name = tensor("aw_chunk_3231_cast_fp16")]; + tensor var_33873_equation_0 = const()[name = tensor("op_33873_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33873_cast_fp16 = einsum(equation = var_33873_equation_0, values = (var_33667_cast_fp16, var_33205_cast_fp16))[name = tensor("op_33873_cast_fp16")]; + tensor var_33874_to_fp16 = const()[name = tensor("op_33874_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3233_cast_fp16 = mul(x = var_33873_cast_fp16, y = var_33874_to_fp16)[name = tensor("aw_chunk_3233_cast_fp16")]; + tensor var_33877_equation_0 = const()[name = tensor("op_33877_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33877_cast_fp16 = einsum(equation = var_33877_equation_0, values = (var_33667_cast_fp16, var_33212_cast_fp16))[name = tensor("op_33877_cast_fp16")]; + tensor var_33878_to_fp16 = const()[name = tensor("op_33878_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3235_cast_fp16 = mul(x = var_33877_cast_fp16, y = var_33878_to_fp16)[name = tensor("aw_chunk_3235_cast_fp16")]; + tensor var_33881_equation_0 = const()[name = tensor("op_33881_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33881_cast_fp16 = einsum(equation = var_33881_equation_0, values = (var_33667_cast_fp16, var_33219_cast_fp16))[name = tensor("op_33881_cast_fp16")]; + tensor var_33882_to_fp16 = const()[name = tensor("op_33882_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3237_cast_fp16 = mul(x = var_33881_cast_fp16, y = var_33882_to_fp16)[name = tensor("aw_chunk_3237_cast_fp16")]; + tensor var_33885_equation_0 = const()[name = tensor("op_33885_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33885_cast_fp16 = einsum(equation = var_33885_equation_0, values = (var_33667_cast_fp16, var_33226_cast_fp16))[name = tensor("op_33885_cast_fp16")]; + tensor var_33886_to_fp16 = const()[name = tensor("op_33886_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3239_cast_fp16 = mul(x = var_33885_cast_fp16, y = var_33886_to_fp16)[name = tensor("aw_chunk_3239_cast_fp16")]; + tensor var_33889_equation_0 = const()[name = tensor("op_33889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33889_cast_fp16 = einsum(equation = var_33889_equation_0, values = (var_33671_cast_fp16, var_33233_cast_fp16))[name = tensor("op_33889_cast_fp16")]; + tensor var_33890_to_fp16 = const()[name = tensor("op_33890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3241_cast_fp16 = mul(x = var_33889_cast_fp16, y = var_33890_to_fp16)[name = tensor("aw_chunk_3241_cast_fp16")]; + tensor var_33893_equation_0 = const()[name = tensor("op_33893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33893_cast_fp16 = einsum(equation = var_33893_equation_0, values = (var_33671_cast_fp16, var_33240_cast_fp16))[name = tensor("op_33893_cast_fp16")]; + tensor var_33894_to_fp16 = const()[name = tensor("op_33894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3243_cast_fp16 = mul(x = var_33893_cast_fp16, y = var_33894_to_fp16)[name = tensor("aw_chunk_3243_cast_fp16")]; + tensor var_33897_equation_0 = const()[name = tensor("op_33897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33897_cast_fp16 = einsum(equation = var_33897_equation_0, values = (var_33671_cast_fp16, var_33247_cast_fp16))[name = tensor("op_33897_cast_fp16")]; + tensor var_33898_to_fp16 = const()[name = tensor("op_33898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3245_cast_fp16 = mul(x = var_33897_cast_fp16, y = var_33898_to_fp16)[name = tensor("aw_chunk_3245_cast_fp16")]; + tensor var_33901_equation_0 = const()[name = tensor("op_33901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33901_cast_fp16 = einsum(equation = var_33901_equation_0, values = (var_33671_cast_fp16, var_33254_cast_fp16))[name = tensor("op_33901_cast_fp16")]; + tensor var_33902_to_fp16 = const()[name = tensor("op_33902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3247_cast_fp16 = mul(x = var_33901_cast_fp16, y = var_33902_to_fp16)[name = tensor("aw_chunk_3247_cast_fp16")]; + tensor var_33905_equation_0 = const()[name = tensor("op_33905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33905_cast_fp16 = einsum(equation = var_33905_equation_0, values = (var_33675_cast_fp16, var_33261_cast_fp16))[name = tensor("op_33905_cast_fp16")]; + tensor var_33906_to_fp16 = const()[name = tensor("op_33906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3249_cast_fp16 = mul(x = var_33905_cast_fp16, y = var_33906_to_fp16)[name = tensor("aw_chunk_3249_cast_fp16")]; + tensor var_33909_equation_0 = const()[name = tensor("op_33909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33909_cast_fp16 = einsum(equation = var_33909_equation_0, values = (var_33675_cast_fp16, var_33268_cast_fp16))[name = tensor("op_33909_cast_fp16")]; + tensor var_33910_to_fp16 = const()[name = tensor("op_33910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3251_cast_fp16 = mul(x = var_33909_cast_fp16, y = var_33910_to_fp16)[name = tensor("aw_chunk_3251_cast_fp16")]; + tensor var_33913_equation_0 = const()[name = tensor("op_33913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33913_cast_fp16 = einsum(equation = var_33913_equation_0, values = (var_33675_cast_fp16, var_33275_cast_fp16))[name = tensor("op_33913_cast_fp16")]; + tensor var_33914_to_fp16 = const()[name = tensor("op_33914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3253_cast_fp16 = mul(x = var_33913_cast_fp16, y = var_33914_to_fp16)[name = tensor("aw_chunk_3253_cast_fp16")]; + tensor var_33917_equation_0 = const()[name = tensor("op_33917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33917_cast_fp16 = einsum(equation = var_33917_equation_0, values = (var_33675_cast_fp16, var_33282_cast_fp16))[name = tensor("op_33917_cast_fp16")]; + tensor var_33918_to_fp16 = const()[name = tensor("op_33918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3255_cast_fp16 = mul(x = var_33917_cast_fp16, y = var_33918_to_fp16)[name = tensor("aw_chunk_3255_cast_fp16")]; + tensor var_33921_equation_0 = const()[name = tensor("op_33921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33921_cast_fp16 = einsum(equation = var_33921_equation_0, values = (var_33679_cast_fp16, var_33289_cast_fp16))[name = tensor("op_33921_cast_fp16")]; + tensor var_33922_to_fp16 = const()[name = tensor("op_33922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3257_cast_fp16 = mul(x = var_33921_cast_fp16, y = var_33922_to_fp16)[name = tensor("aw_chunk_3257_cast_fp16")]; + tensor var_33925_equation_0 = const()[name = tensor("op_33925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33925_cast_fp16 = einsum(equation = var_33925_equation_0, values = (var_33679_cast_fp16, var_33296_cast_fp16))[name = tensor("op_33925_cast_fp16")]; + tensor var_33926_to_fp16 = const()[name = tensor("op_33926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3259_cast_fp16 = mul(x = var_33925_cast_fp16, y = var_33926_to_fp16)[name = tensor("aw_chunk_3259_cast_fp16")]; + tensor var_33929_equation_0 = const()[name = tensor("op_33929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33929_cast_fp16 = einsum(equation = var_33929_equation_0, values = (var_33679_cast_fp16, var_33303_cast_fp16))[name = tensor("op_33929_cast_fp16")]; + tensor var_33930_to_fp16 = const()[name = tensor("op_33930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3261_cast_fp16 = mul(x = var_33929_cast_fp16, y = var_33930_to_fp16)[name = tensor("aw_chunk_3261_cast_fp16")]; + tensor var_33933_equation_0 = const()[name = tensor("op_33933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33933_cast_fp16 = einsum(equation = var_33933_equation_0, values = (var_33679_cast_fp16, var_33310_cast_fp16))[name = tensor("op_33933_cast_fp16")]; + tensor var_33934_to_fp16 = const()[name = tensor("op_33934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3263_cast_fp16 = mul(x = var_33933_cast_fp16, y = var_33934_to_fp16)[name = tensor("aw_chunk_3263_cast_fp16")]; + tensor var_33937_equation_0 = const()[name = tensor("op_33937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33937_cast_fp16 = einsum(equation = var_33937_equation_0, values = (var_33683_cast_fp16, var_33317_cast_fp16))[name = tensor("op_33937_cast_fp16")]; + tensor var_33938_to_fp16 = const()[name = tensor("op_33938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3265_cast_fp16 = mul(x = var_33937_cast_fp16, y = var_33938_to_fp16)[name = tensor("aw_chunk_3265_cast_fp16")]; + tensor var_33941_equation_0 = const()[name = tensor("op_33941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33941_cast_fp16 = einsum(equation = var_33941_equation_0, values = (var_33683_cast_fp16, var_33324_cast_fp16))[name = tensor("op_33941_cast_fp16")]; + tensor var_33942_to_fp16 = const()[name = tensor("op_33942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3267_cast_fp16 = mul(x = var_33941_cast_fp16, y = var_33942_to_fp16)[name = tensor("aw_chunk_3267_cast_fp16")]; + tensor var_33945_equation_0 = const()[name = tensor("op_33945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33945_cast_fp16 = einsum(equation = var_33945_equation_0, values = (var_33683_cast_fp16, var_33331_cast_fp16))[name = tensor("op_33945_cast_fp16")]; + tensor var_33946_to_fp16 = const()[name = tensor("op_33946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3269_cast_fp16 = mul(x = var_33945_cast_fp16, y = var_33946_to_fp16)[name = tensor("aw_chunk_3269_cast_fp16")]; + tensor var_33949_equation_0 = const()[name = tensor("op_33949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33949_cast_fp16 = einsum(equation = var_33949_equation_0, values = (var_33683_cast_fp16, var_33338_cast_fp16))[name = tensor("op_33949_cast_fp16")]; + tensor var_33950_to_fp16 = const()[name = tensor("op_33950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3271_cast_fp16 = mul(x = var_33949_cast_fp16, y = var_33950_to_fp16)[name = tensor("aw_chunk_3271_cast_fp16")]; + tensor var_33953_equation_0 = const()[name = tensor("op_33953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33953_cast_fp16 = einsum(equation = var_33953_equation_0, values = (var_33687_cast_fp16, var_33345_cast_fp16))[name = tensor("op_33953_cast_fp16")]; + tensor var_33954_to_fp16 = const()[name = tensor("op_33954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3273_cast_fp16 = mul(x = var_33953_cast_fp16, y = var_33954_to_fp16)[name = tensor("aw_chunk_3273_cast_fp16")]; + tensor var_33957_equation_0 = const()[name = tensor("op_33957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33957_cast_fp16 = einsum(equation = var_33957_equation_0, values = (var_33687_cast_fp16, var_33352_cast_fp16))[name = tensor("op_33957_cast_fp16")]; + tensor var_33958_to_fp16 = const()[name = tensor("op_33958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3275_cast_fp16 = mul(x = var_33957_cast_fp16, y = var_33958_to_fp16)[name = tensor("aw_chunk_3275_cast_fp16")]; + tensor var_33961_equation_0 = const()[name = tensor("op_33961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33961_cast_fp16 = einsum(equation = var_33961_equation_0, values = (var_33687_cast_fp16, var_33359_cast_fp16))[name = tensor("op_33961_cast_fp16")]; + tensor var_33962_to_fp16 = const()[name = tensor("op_33962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3277_cast_fp16 = mul(x = var_33961_cast_fp16, y = var_33962_to_fp16)[name = tensor("aw_chunk_3277_cast_fp16")]; + tensor var_33965_equation_0 = const()[name = tensor("op_33965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33965_cast_fp16 = einsum(equation = var_33965_equation_0, values = (var_33687_cast_fp16, var_33366_cast_fp16))[name = tensor("op_33965_cast_fp16")]; + tensor var_33966_to_fp16 = const()[name = tensor("op_33966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3279_cast_fp16 = mul(x = var_33965_cast_fp16, y = var_33966_to_fp16)[name = tensor("aw_chunk_3279_cast_fp16")]; + tensor var_33969_equation_0 = const()[name = tensor("op_33969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33969_cast_fp16 = einsum(equation = var_33969_equation_0, values = (var_33691_cast_fp16, var_33373_cast_fp16))[name = tensor("op_33969_cast_fp16")]; + tensor var_33970_to_fp16 = const()[name = tensor("op_33970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3281_cast_fp16 = mul(x = var_33969_cast_fp16, y = var_33970_to_fp16)[name = tensor("aw_chunk_3281_cast_fp16")]; + tensor var_33973_equation_0 = const()[name = tensor("op_33973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33973_cast_fp16 = einsum(equation = var_33973_equation_0, values = (var_33691_cast_fp16, var_33380_cast_fp16))[name = tensor("op_33973_cast_fp16")]; + tensor var_33974_to_fp16 = const()[name = tensor("op_33974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3283_cast_fp16 = mul(x = var_33973_cast_fp16, y = var_33974_to_fp16)[name = tensor("aw_chunk_3283_cast_fp16")]; + tensor var_33977_equation_0 = const()[name = tensor("op_33977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33977_cast_fp16 = einsum(equation = var_33977_equation_0, values = (var_33691_cast_fp16, var_33387_cast_fp16))[name = tensor("op_33977_cast_fp16")]; + tensor var_33978_to_fp16 = const()[name = tensor("op_33978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3285_cast_fp16 = mul(x = var_33977_cast_fp16, y = var_33978_to_fp16)[name = tensor("aw_chunk_3285_cast_fp16")]; + tensor var_33981_equation_0 = const()[name = tensor("op_33981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33981_cast_fp16 = einsum(equation = var_33981_equation_0, values = (var_33691_cast_fp16, var_33394_cast_fp16))[name = tensor("op_33981_cast_fp16")]; + tensor var_33982_to_fp16 = const()[name = tensor("op_33982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3287_cast_fp16 = mul(x = var_33981_cast_fp16, y = var_33982_to_fp16)[name = tensor("aw_chunk_3287_cast_fp16")]; + tensor var_33985_equation_0 = const()[name = tensor("op_33985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33985_cast_fp16 = einsum(equation = var_33985_equation_0, values = (var_33695_cast_fp16, var_33401_cast_fp16))[name = tensor("op_33985_cast_fp16")]; + tensor var_33986_to_fp16 = const()[name = tensor("op_33986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3289_cast_fp16 = mul(x = var_33985_cast_fp16, y = var_33986_to_fp16)[name = tensor("aw_chunk_3289_cast_fp16")]; + tensor var_33989_equation_0 = const()[name = tensor("op_33989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33989_cast_fp16 = einsum(equation = var_33989_equation_0, values = (var_33695_cast_fp16, var_33408_cast_fp16))[name = tensor("op_33989_cast_fp16")]; + tensor var_33990_to_fp16 = const()[name = tensor("op_33990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3291_cast_fp16 = mul(x = var_33989_cast_fp16, y = var_33990_to_fp16)[name = tensor("aw_chunk_3291_cast_fp16")]; + tensor var_33993_equation_0 = const()[name = tensor("op_33993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33993_cast_fp16 = einsum(equation = var_33993_equation_0, values = (var_33695_cast_fp16, var_33415_cast_fp16))[name = tensor("op_33993_cast_fp16")]; + tensor var_33994_to_fp16 = const()[name = tensor("op_33994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3293_cast_fp16 = mul(x = var_33993_cast_fp16, y = var_33994_to_fp16)[name = tensor("aw_chunk_3293_cast_fp16")]; + tensor var_33997_equation_0 = const()[name = tensor("op_33997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_33997_cast_fp16 = einsum(equation = var_33997_equation_0, values = (var_33695_cast_fp16, var_33422_cast_fp16))[name = tensor("op_33997_cast_fp16")]; + tensor var_33998_to_fp16 = const()[name = tensor("op_33998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3295_cast_fp16 = mul(x = var_33997_cast_fp16, y = var_33998_to_fp16)[name = tensor("aw_chunk_3295_cast_fp16")]; + tensor var_34001_equation_0 = const()[name = tensor("op_34001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34001_cast_fp16 = einsum(equation = var_34001_equation_0, values = (var_33699_cast_fp16, var_33429_cast_fp16))[name = tensor("op_34001_cast_fp16")]; + tensor var_34002_to_fp16 = const()[name = tensor("op_34002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3297_cast_fp16 = mul(x = var_34001_cast_fp16, y = var_34002_to_fp16)[name = tensor("aw_chunk_3297_cast_fp16")]; + tensor var_34005_equation_0 = const()[name = tensor("op_34005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34005_cast_fp16 = einsum(equation = var_34005_equation_0, values = (var_33699_cast_fp16, var_33436_cast_fp16))[name = tensor("op_34005_cast_fp16")]; + tensor var_34006_to_fp16 = const()[name = tensor("op_34006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3299_cast_fp16 = mul(x = var_34005_cast_fp16, y = var_34006_to_fp16)[name = tensor("aw_chunk_3299_cast_fp16")]; + tensor var_34009_equation_0 = const()[name = tensor("op_34009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34009_cast_fp16 = einsum(equation = var_34009_equation_0, values = (var_33699_cast_fp16, var_33443_cast_fp16))[name = tensor("op_34009_cast_fp16")]; + tensor var_34010_to_fp16 = const()[name = tensor("op_34010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3301_cast_fp16 = mul(x = var_34009_cast_fp16, y = var_34010_to_fp16)[name = tensor("aw_chunk_3301_cast_fp16")]; + tensor var_34013_equation_0 = const()[name = tensor("op_34013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34013_cast_fp16 = einsum(equation = var_34013_equation_0, values = (var_33699_cast_fp16, var_33450_cast_fp16))[name = tensor("op_34013_cast_fp16")]; + tensor var_34014_to_fp16 = const()[name = tensor("op_34014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3303_cast_fp16 = mul(x = var_34013_cast_fp16, y = var_34014_to_fp16)[name = tensor("aw_chunk_3303_cast_fp16")]; + tensor var_34017_equation_0 = const()[name = tensor("op_34017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34017_cast_fp16 = einsum(equation = var_34017_equation_0, values = (var_33703_cast_fp16, var_33457_cast_fp16))[name = tensor("op_34017_cast_fp16")]; + tensor var_34018_to_fp16 = const()[name = tensor("op_34018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3305_cast_fp16 = mul(x = var_34017_cast_fp16, y = var_34018_to_fp16)[name = tensor("aw_chunk_3305_cast_fp16")]; + tensor var_34021_equation_0 = const()[name = tensor("op_34021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34021_cast_fp16 = einsum(equation = var_34021_equation_0, values = (var_33703_cast_fp16, var_33464_cast_fp16))[name = tensor("op_34021_cast_fp16")]; + tensor var_34022_to_fp16 = const()[name = tensor("op_34022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3307_cast_fp16 = mul(x = var_34021_cast_fp16, y = var_34022_to_fp16)[name = tensor("aw_chunk_3307_cast_fp16")]; + tensor var_34025_equation_0 = const()[name = tensor("op_34025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34025_cast_fp16 = einsum(equation = var_34025_equation_0, values = (var_33703_cast_fp16, var_33471_cast_fp16))[name = tensor("op_34025_cast_fp16")]; + tensor var_34026_to_fp16 = const()[name = tensor("op_34026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3309_cast_fp16 = mul(x = var_34025_cast_fp16, y = var_34026_to_fp16)[name = tensor("aw_chunk_3309_cast_fp16")]; + tensor var_34029_equation_0 = const()[name = tensor("op_34029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34029_cast_fp16 = einsum(equation = var_34029_equation_0, values = (var_33703_cast_fp16, var_33478_cast_fp16))[name = tensor("op_34029_cast_fp16")]; + tensor var_34030_to_fp16 = const()[name = tensor("op_34030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3311_cast_fp16 = mul(x = var_34029_cast_fp16, y = var_34030_to_fp16)[name = tensor("aw_chunk_3311_cast_fp16")]; + tensor var_34033_equation_0 = const()[name = tensor("op_34033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34033_cast_fp16 = einsum(equation = var_34033_equation_0, values = (var_33707_cast_fp16, var_33485_cast_fp16))[name = tensor("op_34033_cast_fp16")]; + tensor var_34034_to_fp16 = const()[name = tensor("op_34034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3313_cast_fp16 = mul(x = var_34033_cast_fp16, y = var_34034_to_fp16)[name = tensor("aw_chunk_3313_cast_fp16")]; + tensor var_34037_equation_0 = const()[name = tensor("op_34037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34037_cast_fp16 = einsum(equation = var_34037_equation_0, values = (var_33707_cast_fp16, var_33492_cast_fp16))[name = tensor("op_34037_cast_fp16")]; + tensor var_34038_to_fp16 = const()[name = tensor("op_34038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3315_cast_fp16 = mul(x = var_34037_cast_fp16, y = var_34038_to_fp16)[name = tensor("aw_chunk_3315_cast_fp16")]; + tensor var_34041_equation_0 = const()[name = tensor("op_34041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34041_cast_fp16 = einsum(equation = var_34041_equation_0, values = (var_33707_cast_fp16, var_33499_cast_fp16))[name = tensor("op_34041_cast_fp16")]; + tensor var_34042_to_fp16 = const()[name = tensor("op_34042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3317_cast_fp16 = mul(x = var_34041_cast_fp16, y = var_34042_to_fp16)[name = tensor("aw_chunk_3317_cast_fp16")]; + tensor var_34045_equation_0 = const()[name = tensor("op_34045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34045_cast_fp16 = einsum(equation = var_34045_equation_0, values = (var_33707_cast_fp16, var_33506_cast_fp16))[name = tensor("op_34045_cast_fp16")]; + tensor var_34046_to_fp16 = const()[name = tensor("op_34046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3319_cast_fp16 = mul(x = var_34045_cast_fp16, y = var_34046_to_fp16)[name = tensor("aw_chunk_3319_cast_fp16")]; + tensor var_34049_equation_0 = const()[name = tensor("op_34049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34049_cast_fp16 = einsum(equation = var_34049_equation_0, values = (var_33711_cast_fp16, var_33513_cast_fp16))[name = tensor("op_34049_cast_fp16")]; + tensor var_34050_to_fp16 = const()[name = tensor("op_34050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3321_cast_fp16 = mul(x = var_34049_cast_fp16, y = var_34050_to_fp16)[name = tensor("aw_chunk_3321_cast_fp16")]; + tensor var_34053_equation_0 = const()[name = tensor("op_34053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34053_cast_fp16 = einsum(equation = var_34053_equation_0, values = (var_33711_cast_fp16, var_33520_cast_fp16))[name = tensor("op_34053_cast_fp16")]; + tensor var_34054_to_fp16 = const()[name = tensor("op_34054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3323_cast_fp16 = mul(x = var_34053_cast_fp16, y = var_34054_to_fp16)[name = tensor("aw_chunk_3323_cast_fp16")]; + tensor var_34057_equation_0 = const()[name = tensor("op_34057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34057_cast_fp16 = einsum(equation = var_34057_equation_0, values = (var_33711_cast_fp16, var_33527_cast_fp16))[name = tensor("op_34057_cast_fp16")]; + tensor var_34058_to_fp16 = const()[name = tensor("op_34058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3325_cast_fp16 = mul(x = var_34057_cast_fp16, y = var_34058_to_fp16)[name = tensor("aw_chunk_3325_cast_fp16")]; + tensor var_34061_equation_0 = const()[name = tensor("op_34061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34061_cast_fp16 = einsum(equation = var_34061_equation_0, values = (var_33711_cast_fp16, var_33534_cast_fp16))[name = tensor("op_34061_cast_fp16")]; + tensor var_34062_to_fp16 = const()[name = tensor("op_34062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3327_cast_fp16 = mul(x = var_34061_cast_fp16, y = var_34062_to_fp16)[name = tensor("aw_chunk_3327_cast_fp16")]; + tensor var_34065_equation_0 = const()[name = tensor("op_34065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34065_cast_fp16 = einsum(equation = var_34065_equation_0, values = (var_33715_cast_fp16, var_33541_cast_fp16))[name = tensor("op_34065_cast_fp16")]; + tensor var_34066_to_fp16 = const()[name = tensor("op_34066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3329_cast_fp16 = mul(x = var_34065_cast_fp16, y = var_34066_to_fp16)[name = tensor("aw_chunk_3329_cast_fp16")]; + tensor var_34069_equation_0 = const()[name = tensor("op_34069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34069_cast_fp16 = einsum(equation = var_34069_equation_0, values = (var_33715_cast_fp16, var_33548_cast_fp16))[name = tensor("op_34069_cast_fp16")]; + tensor var_34070_to_fp16 = const()[name = tensor("op_34070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3331_cast_fp16 = mul(x = var_34069_cast_fp16, y = var_34070_to_fp16)[name = tensor("aw_chunk_3331_cast_fp16")]; + tensor var_34073_equation_0 = const()[name = tensor("op_34073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34073_cast_fp16 = einsum(equation = var_34073_equation_0, values = (var_33715_cast_fp16, var_33555_cast_fp16))[name = tensor("op_34073_cast_fp16")]; + tensor var_34074_to_fp16 = const()[name = tensor("op_34074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3333_cast_fp16 = mul(x = var_34073_cast_fp16, y = var_34074_to_fp16)[name = tensor("aw_chunk_3333_cast_fp16")]; + tensor var_34077_equation_0 = const()[name = tensor("op_34077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34077_cast_fp16 = einsum(equation = var_34077_equation_0, values = (var_33715_cast_fp16, var_33562_cast_fp16))[name = tensor("op_34077_cast_fp16")]; + tensor var_34078_to_fp16 = const()[name = tensor("op_34078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3335_cast_fp16 = mul(x = var_34077_cast_fp16, y = var_34078_to_fp16)[name = tensor("aw_chunk_3335_cast_fp16")]; + tensor var_34081_equation_0 = const()[name = tensor("op_34081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34081_cast_fp16 = einsum(equation = var_34081_equation_0, values = (var_33719_cast_fp16, var_33569_cast_fp16))[name = tensor("op_34081_cast_fp16")]; + tensor var_34082_to_fp16 = const()[name = tensor("op_34082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3337_cast_fp16 = mul(x = var_34081_cast_fp16, y = var_34082_to_fp16)[name = tensor("aw_chunk_3337_cast_fp16")]; + tensor var_34085_equation_0 = const()[name = tensor("op_34085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34085_cast_fp16 = einsum(equation = var_34085_equation_0, values = (var_33719_cast_fp16, var_33576_cast_fp16))[name = tensor("op_34085_cast_fp16")]; + tensor var_34086_to_fp16 = const()[name = tensor("op_34086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3339_cast_fp16 = mul(x = var_34085_cast_fp16, y = var_34086_to_fp16)[name = tensor("aw_chunk_3339_cast_fp16")]; + tensor var_34089_equation_0 = const()[name = tensor("op_34089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34089_cast_fp16 = einsum(equation = var_34089_equation_0, values = (var_33719_cast_fp16, var_33583_cast_fp16))[name = tensor("op_34089_cast_fp16")]; + tensor var_34090_to_fp16 = const()[name = tensor("op_34090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3341_cast_fp16 = mul(x = var_34089_cast_fp16, y = var_34090_to_fp16)[name = tensor("aw_chunk_3341_cast_fp16")]; + tensor var_34093_equation_0 = const()[name = tensor("op_34093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34093_cast_fp16 = einsum(equation = var_34093_equation_0, values = (var_33719_cast_fp16, var_33590_cast_fp16))[name = tensor("op_34093_cast_fp16")]; + tensor var_34094_to_fp16 = const()[name = tensor("op_34094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3343_cast_fp16 = mul(x = var_34093_cast_fp16, y = var_34094_to_fp16)[name = tensor("aw_chunk_3343_cast_fp16")]; + tensor var_34097_equation_0 = const()[name = tensor("op_34097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34097_cast_fp16 = einsum(equation = var_34097_equation_0, values = (var_33723_cast_fp16, var_33597_cast_fp16))[name = tensor("op_34097_cast_fp16")]; + tensor var_34098_to_fp16 = const()[name = tensor("op_34098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3345_cast_fp16 = mul(x = var_34097_cast_fp16, y = var_34098_to_fp16)[name = tensor("aw_chunk_3345_cast_fp16")]; + tensor var_34101_equation_0 = const()[name = tensor("op_34101_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34101_cast_fp16 = einsum(equation = var_34101_equation_0, values = (var_33723_cast_fp16, var_33604_cast_fp16))[name = tensor("op_34101_cast_fp16")]; + tensor var_34102_to_fp16 = const()[name = tensor("op_34102_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3347_cast_fp16 = mul(x = var_34101_cast_fp16, y = var_34102_to_fp16)[name = tensor("aw_chunk_3347_cast_fp16")]; + tensor var_34105_equation_0 = const()[name = tensor("op_34105_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34105_cast_fp16 = einsum(equation = var_34105_equation_0, values = (var_33723_cast_fp16, var_33611_cast_fp16))[name = tensor("op_34105_cast_fp16")]; + tensor var_34106_to_fp16 = const()[name = tensor("op_34106_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3349_cast_fp16 = mul(x = var_34105_cast_fp16, y = var_34106_to_fp16)[name = tensor("aw_chunk_3349_cast_fp16")]; + tensor var_34109_equation_0 = const()[name = tensor("op_34109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34109_cast_fp16 = einsum(equation = var_34109_equation_0, values = (var_33723_cast_fp16, var_33618_cast_fp16))[name = tensor("op_34109_cast_fp16")]; + tensor var_34110_to_fp16 = const()[name = tensor("op_34110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3351_cast_fp16 = mul(x = var_34109_cast_fp16, y = var_34110_to_fp16)[name = tensor("aw_chunk_3351_cast_fp16")]; + tensor var_34113_equation_0 = const()[name = tensor("op_34113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34113_cast_fp16 = einsum(equation = var_34113_equation_0, values = (var_33727_cast_fp16, var_33625_cast_fp16))[name = tensor("op_34113_cast_fp16")]; + tensor var_34114_to_fp16 = const()[name = tensor("op_34114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3353_cast_fp16 = mul(x = var_34113_cast_fp16, y = var_34114_to_fp16)[name = tensor("aw_chunk_3353_cast_fp16")]; + tensor var_34117_equation_0 = const()[name = tensor("op_34117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34117_cast_fp16 = einsum(equation = var_34117_equation_0, values = (var_33727_cast_fp16, var_33632_cast_fp16))[name = tensor("op_34117_cast_fp16")]; + tensor var_34118_to_fp16 = const()[name = tensor("op_34118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3355_cast_fp16 = mul(x = var_34117_cast_fp16, y = var_34118_to_fp16)[name = tensor("aw_chunk_3355_cast_fp16")]; + tensor var_34121_equation_0 = const()[name = tensor("op_34121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34121_cast_fp16 = einsum(equation = var_34121_equation_0, values = (var_33727_cast_fp16, var_33639_cast_fp16))[name = tensor("op_34121_cast_fp16")]; + tensor var_34122_to_fp16 = const()[name = tensor("op_34122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3357_cast_fp16 = mul(x = var_34121_cast_fp16, y = var_34122_to_fp16)[name = tensor("aw_chunk_3357_cast_fp16")]; + tensor var_34125_equation_0 = const()[name = tensor("op_34125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_34125_cast_fp16 = einsum(equation = var_34125_equation_0, values = (var_33727_cast_fp16, var_33646_cast_fp16))[name = tensor("op_34125_cast_fp16")]; + tensor var_34126_to_fp16 = const()[name = tensor("op_34126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3359_cast_fp16 = mul(x = var_34125_cast_fp16, y = var_34126_to_fp16)[name = tensor("aw_chunk_3359_cast_fp16")]; + tensor var_34128_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3201_cast_fp16)[name = tensor("op_34128_cast_fp16")]; + tensor var_34129_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3203_cast_fp16)[name = tensor("op_34129_cast_fp16")]; + tensor var_34130_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3205_cast_fp16)[name = tensor("op_34130_cast_fp16")]; + tensor var_34131_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3207_cast_fp16)[name = tensor("op_34131_cast_fp16")]; + tensor var_34132_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3209_cast_fp16)[name = tensor("op_34132_cast_fp16")]; + tensor var_34133_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3211_cast_fp16)[name = tensor("op_34133_cast_fp16")]; + tensor var_34134_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3213_cast_fp16)[name = tensor("op_34134_cast_fp16")]; + tensor var_34135_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3215_cast_fp16)[name = tensor("op_34135_cast_fp16")]; + tensor var_34136_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3217_cast_fp16)[name = tensor("op_34136_cast_fp16")]; + tensor var_34137_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3219_cast_fp16)[name = tensor("op_34137_cast_fp16")]; + tensor var_34138_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3221_cast_fp16)[name = tensor("op_34138_cast_fp16")]; + tensor var_34139_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3223_cast_fp16)[name = tensor("op_34139_cast_fp16")]; + tensor var_34140_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3225_cast_fp16)[name = tensor("op_34140_cast_fp16")]; + tensor var_34141_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3227_cast_fp16)[name = tensor("op_34141_cast_fp16")]; + tensor var_34142_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3229_cast_fp16)[name = tensor("op_34142_cast_fp16")]; + tensor var_34143_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3231_cast_fp16)[name = tensor("op_34143_cast_fp16")]; + tensor var_34144_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3233_cast_fp16)[name = tensor("op_34144_cast_fp16")]; + tensor var_34145_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3235_cast_fp16)[name = tensor("op_34145_cast_fp16")]; + tensor var_34146_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3237_cast_fp16)[name = tensor("op_34146_cast_fp16")]; + tensor var_34147_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3239_cast_fp16)[name = tensor("op_34147_cast_fp16")]; + tensor var_34148_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3241_cast_fp16)[name = tensor("op_34148_cast_fp16")]; + tensor var_34149_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3243_cast_fp16)[name = tensor("op_34149_cast_fp16")]; + tensor var_34150_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3245_cast_fp16)[name = tensor("op_34150_cast_fp16")]; + tensor var_34151_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3247_cast_fp16)[name = tensor("op_34151_cast_fp16")]; + tensor var_34152_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3249_cast_fp16)[name = tensor("op_34152_cast_fp16")]; + tensor var_34153_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3251_cast_fp16)[name = tensor("op_34153_cast_fp16")]; + tensor var_34154_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3253_cast_fp16)[name = tensor("op_34154_cast_fp16")]; + tensor var_34155_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3255_cast_fp16)[name = tensor("op_34155_cast_fp16")]; + tensor var_34156_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3257_cast_fp16)[name = tensor("op_34156_cast_fp16")]; + tensor var_34157_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3259_cast_fp16)[name = tensor("op_34157_cast_fp16")]; + tensor var_34158_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3261_cast_fp16)[name = tensor("op_34158_cast_fp16")]; + tensor var_34159_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3263_cast_fp16)[name = tensor("op_34159_cast_fp16")]; + tensor var_34160_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3265_cast_fp16)[name = tensor("op_34160_cast_fp16")]; + tensor var_34161_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3267_cast_fp16)[name = tensor("op_34161_cast_fp16")]; + tensor var_34162_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3269_cast_fp16)[name = tensor("op_34162_cast_fp16")]; + tensor var_34163_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3271_cast_fp16)[name = tensor("op_34163_cast_fp16")]; + tensor var_34164_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3273_cast_fp16)[name = tensor("op_34164_cast_fp16")]; + tensor var_34165_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3275_cast_fp16)[name = tensor("op_34165_cast_fp16")]; + tensor var_34166_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3277_cast_fp16)[name = tensor("op_34166_cast_fp16")]; + tensor var_34167_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3279_cast_fp16)[name = tensor("op_34167_cast_fp16")]; + tensor var_34168_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3281_cast_fp16)[name = tensor("op_34168_cast_fp16")]; + tensor var_34169_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3283_cast_fp16)[name = tensor("op_34169_cast_fp16")]; + tensor var_34170_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3285_cast_fp16)[name = tensor("op_34170_cast_fp16")]; + tensor var_34171_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3287_cast_fp16)[name = tensor("op_34171_cast_fp16")]; + tensor var_34172_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3289_cast_fp16)[name = tensor("op_34172_cast_fp16")]; + tensor var_34173_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3291_cast_fp16)[name = tensor("op_34173_cast_fp16")]; + tensor var_34174_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3293_cast_fp16)[name = tensor("op_34174_cast_fp16")]; + tensor var_34175_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3295_cast_fp16)[name = tensor("op_34175_cast_fp16")]; + tensor var_34176_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3297_cast_fp16)[name = tensor("op_34176_cast_fp16")]; + tensor var_34177_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3299_cast_fp16)[name = tensor("op_34177_cast_fp16")]; + tensor var_34178_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3301_cast_fp16)[name = tensor("op_34178_cast_fp16")]; + tensor var_34179_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3303_cast_fp16)[name = tensor("op_34179_cast_fp16")]; + tensor var_34180_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3305_cast_fp16)[name = tensor("op_34180_cast_fp16")]; + tensor var_34181_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3307_cast_fp16)[name = tensor("op_34181_cast_fp16")]; + tensor var_34182_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3309_cast_fp16)[name = tensor("op_34182_cast_fp16")]; + tensor var_34183_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3311_cast_fp16)[name = tensor("op_34183_cast_fp16")]; + tensor var_34184_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3313_cast_fp16)[name = tensor("op_34184_cast_fp16")]; + tensor var_34185_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3315_cast_fp16)[name = tensor("op_34185_cast_fp16")]; + tensor var_34186_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3317_cast_fp16)[name = tensor("op_34186_cast_fp16")]; + tensor var_34187_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3319_cast_fp16)[name = tensor("op_34187_cast_fp16")]; + tensor var_34188_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3321_cast_fp16)[name = tensor("op_34188_cast_fp16")]; + tensor var_34189_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3323_cast_fp16)[name = tensor("op_34189_cast_fp16")]; + tensor var_34190_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3325_cast_fp16)[name = tensor("op_34190_cast_fp16")]; + tensor var_34191_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3327_cast_fp16)[name = tensor("op_34191_cast_fp16")]; + tensor var_34192_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3329_cast_fp16)[name = tensor("op_34192_cast_fp16")]; + tensor var_34193_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3331_cast_fp16)[name = tensor("op_34193_cast_fp16")]; + tensor var_34194_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3333_cast_fp16)[name = tensor("op_34194_cast_fp16")]; + tensor var_34195_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3335_cast_fp16)[name = tensor("op_34195_cast_fp16")]; + tensor var_34196_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3337_cast_fp16)[name = tensor("op_34196_cast_fp16")]; + tensor var_34197_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3339_cast_fp16)[name = tensor("op_34197_cast_fp16")]; + tensor var_34198_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3341_cast_fp16)[name = tensor("op_34198_cast_fp16")]; + tensor var_34199_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3343_cast_fp16)[name = tensor("op_34199_cast_fp16")]; + tensor var_34200_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3345_cast_fp16)[name = tensor("op_34200_cast_fp16")]; + tensor var_34201_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3347_cast_fp16)[name = tensor("op_34201_cast_fp16")]; + tensor var_34202_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3349_cast_fp16)[name = tensor("op_34202_cast_fp16")]; + tensor var_34203_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3351_cast_fp16)[name = tensor("op_34203_cast_fp16")]; + tensor var_34204_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3353_cast_fp16)[name = tensor("op_34204_cast_fp16")]; + tensor var_34205_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3355_cast_fp16)[name = tensor("op_34205_cast_fp16")]; + tensor var_34206_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3357_cast_fp16)[name = tensor("op_34206_cast_fp16")]; + tensor var_34207_cast_fp16 = softmax(axis = var_32899, x = aw_chunk_3359_cast_fp16)[name = tensor("op_34207_cast_fp16")]; + tensor var_34209_equation_0 = const()[name = tensor("op_34209_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34209_cast_fp16 = einsum(equation = var_34209_equation_0, values = (var_33729_cast_fp16, var_34128_cast_fp16))[name = tensor("op_34209_cast_fp16")]; + tensor var_34211_equation_0 = const()[name = tensor("op_34211_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34211_cast_fp16 = einsum(equation = var_34211_equation_0, values = (var_33729_cast_fp16, var_34129_cast_fp16))[name = tensor("op_34211_cast_fp16")]; + tensor var_34213_equation_0 = const()[name = tensor("op_34213_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34213_cast_fp16 = einsum(equation = var_34213_equation_0, values = (var_33729_cast_fp16, var_34130_cast_fp16))[name = tensor("op_34213_cast_fp16")]; + tensor var_34215_equation_0 = const()[name = tensor("op_34215_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34215_cast_fp16 = einsum(equation = var_34215_equation_0, values = (var_33729_cast_fp16, var_34131_cast_fp16))[name = tensor("op_34215_cast_fp16")]; + tensor var_34217_equation_0 = const()[name = tensor("op_34217_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34217_cast_fp16 = einsum(equation = var_34217_equation_0, values = (var_33733_cast_fp16, var_34132_cast_fp16))[name = tensor("op_34217_cast_fp16")]; + tensor var_34219_equation_0 = const()[name = tensor("op_34219_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34219_cast_fp16 = einsum(equation = var_34219_equation_0, values = (var_33733_cast_fp16, var_34133_cast_fp16))[name = tensor("op_34219_cast_fp16")]; + tensor var_34221_equation_0 = const()[name = tensor("op_34221_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34221_cast_fp16 = einsum(equation = var_34221_equation_0, values = (var_33733_cast_fp16, var_34134_cast_fp16))[name = tensor("op_34221_cast_fp16")]; + tensor var_34223_equation_0 = const()[name = tensor("op_34223_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34223_cast_fp16 = einsum(equation = var_34223_equation_0, values = (var_33733_cast_fp16, var_34135_cast_fp16))[name = tensor("op_34223_cast_fp16")]; + tensor var_34225_equation_0 = const()[name = tensor("op_34225_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34225_cast_fp16 = einsum(equation = var_34225_equation_0, values = (var_33737_cast_fp16, var_34136_cast_fp16))[name = tensor("op_34225_cast_fp16")]; + tensor var_34227_equation_0 = const()[name = tensor("op_34227_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34227_cast_fp16 = einsum(equation = var_34227_equation_0, values = (var_33737_cast_fp16, var_34137_cast_fp16))[name = tensor("op_34227_cast_fp16")]; + tensor var_34229_equation_0 = const()[name = tensor("op_34229_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34229_cast_fp16 = einsum(equation = var_34229_equation_0, values = (var_33737_cast_fp16, var_34138_cast_fp16))[name = tensor("op_34229_cast_fp16")]; + tensor var_34231_equation_0 = const()[name = tensor("op_34231_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34231_cast_fp16 = einsum(equation = var_34231_equation_0, values = (var_33737_cast_fp16, var_34139_cast_fp16))[name = tensor("op_34231_cast_fp16")]; + tensor var_34233_equation_0 = const()[name = tensor("op_34233_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34233_cast_fp16 = einsum(equation = var_34233_equation_0, values = (var_33741_cast_fp16, var_34140_cast_fp16))[name = tensor("op_34233_cast_fp16")]; + tensor var_34235_equation_0 = const()[name = tensor("op_34235_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34235_cast_fp16 = einsum(equation = var_34235_equation_0, values = (var_33741_cast_fp16, var_34141_cast_fp16))[name = tensor("op_34235_cast_fp16")]; + tensor var_34237_equation_0 = const()[name = tensor("op_34237_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34237_cast_fp16 = einsum(equation = var_34237_equation_0, values = (var_33741_cast_fp16, var_34142_cast_fp16))[name = tensor("op_34237_cast_fp16")]; + tensor var_34239_equation_0 = const()[name = tensor("op_34239_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34239_cast_fp16 = einsum(equation = var_34239_equation_0, values = (var_33741_cast_fp16, var_34143_cast_fp16))[name = tensor("op_34239_cast_fp16")]; + tensor var_34241_equation_0 = const()[name = tensor("op_34241_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34241_cast_fp16 = einsum(equation = var_34241_equation_0, values = (var_33745_cast_fp16, var_34144_cast_fp16))[name = tensor("op_34241_cast_fp16")]; + tensor var_34243_equation_0 = const()[name = tensor("op_34243_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34243_cast_fp16 = einsum(equation = var_34243_equation_0, values = (var_33745_cast_fp16, var_34145_cast_fp16))[name = tensor("op_34243_cast_fp16")]; + tensor var_34245_equation_0 = const()[name = tensor("op_34245_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34245_cast_fp16 = einsum(equation = var_34245_equation_0, values = (var_33745_cast_fp16, var_34146_cast_fp16))[name = tensor("op_34245_cast_fp16")]; + tensor var_34247_equation_0 = const()[name = tensor("op_34247_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34247_cast_fp16 = einsum(equation = var_34247_equation_0, values = (var_33745_cast_fp16, var_34147_cast_fp16))[name = tensor("op_34247_cast_fp16")]; + tensor var_34249_equation_0 = const()[name = tensor("op_34249_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34249_cast_fp16 = einsum(equation = var_34249_equation_0, values = (var_33749_cast_fp16, var_34148_cast_fp16))[name = tensor("op_34249_cast_fp16")]; + tensor var_34251_equation_0 = const()[name = tensor("op_34251_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34251_cast_fp16 = einsum(equation = var_34251_equation_0, values = (var_33749_cast_fp16, var_34149_cast_fp16))[name = tensor("op_34251_cast_fp16")]; + tensor var_34253_equation_0 = const()[name = tensor("op_34253_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34253_cast_fp16 = einsum(equation = var_34253_equation_0, values = (var_33749_cast_fp16, var_34150_cast_fp16))[name = tensor("op_34253_cast_fp16")]; + tensor var_34255_equation_0 = const()[name = tensor("op_34255_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34255_cast_fp16 = einsum(equation = var_34255_equation_0, values = (var_33749_cast_fp16, var_34151_cast_fp16))[name = tensor("op_34255_cast_fp16")]; + tensor var_34257_equation_0 = const()[name = tensor("op_34257_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34257_cast_fp16 = einsum(equation = var_34257_equation_0, values = (var_33753_cast_fp16, var_34152_cast_fp16))[name = tensor("op_34257_cast_fp16")]; + tensor var_34259_equation_0 = const()[name = tensor("op_34259_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34259_cast_fp16 = einsum(equation = var_34259_equation_0, values = (var_33753_cast_fp16, var_34153_cast_fp16))[name = tensor("op_34259_cast_fp16")]; + tensor var_34261_equation_0 = const()[name = tensor("op_34261_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34261_cast_fp16 = einsum(equation = var_34261_equation_0, values = (var_33753_cast_fp16, var_34154_cast_fp16))[name = tensor("op_34261_cast_fp16")]; + tensor var_34263_equation_0 = const()[name = tensor("op_34263_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34263_cast_fp16 = einsum(equation = var_34263_equation_0, values = (var_33753_cast_fp16, var_34155_cast_fp16))[name = tensor("op_34263_cast_fp16")]; + tensor var_34265_equation_0 = const()[name = tensor("op_34265_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34265_cast_fp16 = einsum(equation = var_34265_equation_0, values = (var_33757_cast_fp16, var_34156_cast_fp16))[name = tensor("op_34265_cast_fp16")]; + tensor var_34267_equation_0 = const()[name = tensor("op_34267_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34267_cast_fp16 = einsum(equation = var_34267_equation_0, values = (var_33757_cast_fp16, var_34157_cast_fp16))[name = tensor("op_34267_cast_fp16")]; + tensor var_34269_equation_0 = const()[name = tensor("op_34269_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34269_cast_fp16 = einsum(equation = var_34269_equation_0, values = (var_33757_cast_fp16, var_34158_cast_fp16))[name = tensor("op_34269_cast_fp16")]; + tensor var_34271_equation_0 = const()[name = tensor("op_34271_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34271_cast_fp16 = einsum(equation = var_34271_equation_0, values = (var_33757_cast_fp16, var_34159_cast_fp16))[name = tensor("op_34271_cast_fp16")]; + tensor var_34273_equation_0 = const()[name = tensor("op_34273_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34273_cast_fp16 = einsum(equation = var_34273_equation_0, values = (var_33761_cast_fp16, var_34160_cast_fp16))[name = tensor("op_34273_cast_fp16")]; + tensor var_34275_equation_0 = const()[name = tensor("op_34275_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34275_cast_fp16 = einsum(equation = var_34275_equation_0, values = (var_33761_cast_fp16, var_34161_cast_fp16))[name = tensor("op_34275_cast_fp16")]; + tensor var_34277_equation_0 = const()[name = tensor("op_34277_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34277_cast_fp16 = einsum(equation = var_34277_equation_0, values = (var_33761_cast_fp16, var_34162_cast_fp16))[name = tensor("op_34277_cast_fp16")]; + tensor var_34279_equation_0 = const()[name = tensor("op_34279_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34279_cast_fp16 = einsum(equation = var_34279_equation_0, values = (var_33761_cast_fp16, var_34163_cast_fp16))[name = tensor("op_34279_cast_fp16")]; + tensor var_34281_equation_0 = const()[name = tensor("op_34281_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34281_cast_fp16 = einsum(equation = var_34281_equation_0, values = (var_33765_cast_fp16, var_34164_cast_fp16))[name = tensor("op_34281_cast_fp16")]; + tensor var_34283_equation_0 = const()[name = tensor("op_34283_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34283_cast_fp16 = einsum(equation = var_34283_equation_0, values = (var_33765_cast_fp16, var_34165_cast_fp16))[name = tensor("op_34283_cast_fp16")]; + tensor var_34285_equation_0 = const()[name = tensor("op_34285_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34285_cast_fp16 = einsum(equation = var_34285_equation_0, values = (var_33765_cast_fp16, var_34166_cast_fp16))[name = tensor("op_34285_cast_fp16")]; + tensor var_34287_equation_0 = const()[name = tensor("op_34287_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34287_cast_fp16 = einsum(equation = var_34287_equation_0, values = (var_33765_cast_fp16, var_34167_cast_fp16))[name = tensor("op_34287_cast_fp16")]; + tensor var_34289_equation_0 = const()[name = tensor("op_34289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34289_cast_fp16 = einsum(equation = var_34289_equation_0, values = (var_33769_cast_fp16, var_34168_cast_fp16))[name = tensor("op_34289_cast_fp16")]; + tensor var_34291_equation_0 = const()[name = tensor("op_34291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34291_cast_fp16 = einsum(equation = var_34291_equation_0, values = (var_33769_cast_fp16, var_34169_cast_fp16))[name = tensor("op_34291_cast_fp16")]; + tensor var_34293_equation_0 = const()[name = tensor("op_34293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34293_cast_fp16 = einsum(equation = var_34293_equation_0, values = (var_33769_cast_fp16, var_34170_cast_fp16))[name = tensor("op_34293_cast_fp16")]; + tensor var_34295_equation_0 = const()[name = tensor("op_34295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34295_cast_fp16 = einsum(equation = var_34295_equation_0, values = (var_33769_cast_fp16, var_34171_cast_fp16))[name = tensor("op_34295_cast_fp16")]; + tensor var_34297_equation_0 = const()[name = tensor("op_34297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34297_cast_fp16 = einsum(equation = var_34297_equation_0, values = (var_33773_cast_fp16, var_34172_cast_fp16))[name = tensor("op_34297_cast_fp16")]; + tensor var_34299_equation_0 = const()[name = tensor("op_34299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34299_cast_fp16 = einsum(equation = var_34299_equation_0, values = (var_33773_cast_fp16, var_34173_cast_fp16))[name = tensor("op_34299_cast_fp16")]; + tensor var_34301_equation_0 = const()[name = tensor("op_34301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34301_cast_fp16 = einsum(equation = var_34301_equation_0, values = (var_33773_cast_fp16, var_34174_cast_fp16))[name = tensor("op_34301_cast_fp16")]; + tensor var_34303_equation_0 = const()[name = tensor("op_34303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34303_cast_fp16 = einsum(equation = var_34303_equation_0, values = (var_33773_cast_fp16, var_34175_cast_fp16))[name = tensor("op_34303_cast_fp16")]; + tensor var_34305_equation_0 = const()[name = tensor("op_34305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34305_cast_fp16 = einsum(equation = var_34305_equation_0, values = (var_33777_cast_fp16, var_34176_cast_fp16))[name = tensor("op_34305_cast_fp16")]; + tensor var_34307_equation_0 = const()[name = tensor("op_34307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34307_cast_fp16 = einsum(equation = var_34307_equation_0, values = (var_33777_cast_fp16, var_34177_cast_fp16))[name = tensor("op_34307_cast_fp16")]; + tensor var_34309_equation_0 = const()[name = tensor("op_34309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34309_cast_fp16 = einsum(equation = var_34309_equation_0, values = (var_33777_cast_fp16, var_34178_cast_fp16))[name = tensor("op_34309_cast_fp16")]; + tensor var_34311_equation_0 = const()[name = tensor("op_34311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34311_cast_fp16 = einsum(equation = var_34311_equation_0, values = (var_33777_cast_fp16, var_34179_cast_fp16))[name = tensor("op_34311_cast_fp16")]; + tensor var_34313_equation_0 = const()[name = tensor("op_34313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34313_cast_fp16 = einsum(equation = var_34313_equation_0, values = (var_33781_cast_fp16, var_34180_cast_fp16))[name = tensor("op_34313_cast_fp16")]; + tensor var_34315_equation_0 = const()[name = tensor("op_34315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34315_cast_fp16 = einsum(equation = var_34315_equation_0, values = (var_33781_cast_fp16, var_34181_cast_fp16))[name = tensor("op_34315_cast_fp16")]; + tensor var_34317_equation_0 = const()[name = tensor("op_34317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34317_cast_fp16 = einsum(equation = var_34317_equation_0, values = (var_33781_cast_fp16, var_34182_cast_fp16))[name = tensor("op_34317_cast_fp16")]; + tensor var_34319_equation_0 = const()[name = tensor("op_34319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34319_cast_fp16 = einsum(equation = var_34319_equation_0, values = (var_33781_cast_fp16, var_34183_cast_fp16))[name = tensor("op_34319_cast_fp16")]; + tensor var_34321_equation_0 = const()[name = tensor("op_34321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34321_cast_fp16 = einsum(equation = var_34321_equation_0, values = (var_33785_cast_fp16, var_34184_cast_fp16))[name = tensor("op_34321_cast_fp16")]; + tensor var_34323_equation_0 = const()[name = tensor("op_34323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34323_cast_fp16 = einsum(equation = var_34323_equation_0, values = (var_33785_cast_fp16, var_34185_cast_fp16))[name = tensor("op_34323_cast_fp16")]; + tensor var_34325_equation_0 = const()[name = tensor("op_34325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34325_cast_fp16 = einsum(equation = var_34325_equation_0, values = (var_33785_cast_fp16, var_34186_cast_fp16))[name = tensor("op_34325_cast_fp16")]; + tensor var_34327_equation_0 = const()[name = tensor("op_34327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34327_cast_fp16 = einsum(equation = var_34327_equation_0, values = (var_33785_cast_fp16, var_34187_cast_fp16))[name = tensor("op_34327_cast_fp16")]; + tensor var_34329_equation_0 = const()[name = tensor("op_34329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34329_cast_fp16 = einsum(equation = var_34329_equation_0, values = (var_33789_cast_fp16, var_34188_cast_fp16))[name = tensor("op_34329_cast_fp16")]; + tensor var_34331_equation_0 = const()[name = tensor("op_34331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34331_cast_fp16 = einsum(equation = var_34331_equation_0, values = (var_33789_cast_fp16, var_34189_cast_fp16))[name = tensor("op_34331_cast_fp16")]; + tensor var_34333_equation_0 = const()[name = tensor("op_34333_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34333_cast_fp16 = einsum(equation = var_34333_equation_0, values = (var_33789_cast_fp16, var_34190_cast_fp16))[name = tensor("op_34333_cast_fp16")]; + tensor var_34335_equation_0 = const()[name = tensor("op_34335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34335_cast_fp16 = einsum(equation = var_34335_equation_0, values = (var_33789_cast_fp16, var_34191_cast_fp16))[name = tensor("op_34335_cast_fp16")]; + tensor var_34337_equation_0 = const()[name = tensor("op_34337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34337_cast_fp16 = einsum(equation = var_34337_equation_0, values = (var_33793_cast_fp16, var_34192_cast_fp16))[name = tensor("op_34337_cast_fp16")]; + tensor var_34339_equation_0 = const()[name = tensor("op_34339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34339_cast_fp16 = einsum(equation = var_34339_equation_0, values = (var_33793_cast_fp16, var_34193_cast_fp16))[name = tensor("op_34339_cast_fp16")]; + tensor var_34341_equation_0 = const()[name = tensor("op_34341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34341_cast_fp16 = einsum(equation = var_34341_equation_0, values = (var_33793_cast_fp16, var_34194_cast_fp16))[name = tensor("op_34341_cast_fp16")]; + tensor var_34343_equation_0 = const()[name = tensor("op_34343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34343_cast_fp16 = einsum(equation = var_34343_equation_0, values = (var_33793_cast_fp16, var_34195_cast_fp16))[name = tensor("op_34343_cast_fp16")]; + tensor var_34345_equation_0 = const()[name = tensor("op_34345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34345_cast_fp16 = einsum(equation = var_34345_equation_0, values = (var_33797_cast_fp16, var_34196_cast_fp16))[name = tensor("op_34345_cast_fp16")]; + tensor var_34347_equation_0 = const()[name = tensor("op_34347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34347_cast_fp16 = einsum(equation = var_34347_equation_0, values = (var_33797_cast_fp16, var_34197_cast_fp16))[name = tensor("op_34347_cast_fp16")]; + tensor var_34349_equation_0 = const()[name = tensor("op_34349_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34349_cast_fp16 = einsum(equation = var_34349_equation_0, values = (var_33797_cast_fp16, var_34198_cast_fp16))[name = tensor("op_34349_cast_fp16")]; + tensor var_34351_equation_0 = const()[name = tensor("op_34351_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34351_cast_fp16 = einsum(equation = var_34351_equation_0, values = (var_33797_cast_fp16, var_34199_cast_fp16))[name = tensor("op_34351_cast_fp16")]; + tensor var_34353_equation_0 = const()[name = tensor("op_34353_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34353_cast_fp16 = einsum(equation = var_34353_equation_0, values = (var_33801_cast_fp16, var_34200_cast_fp16))[name = tensor("op_34353_cast_fp16")]; + tensor var_34355_equation_0 = const()[name = tensor("op_34355_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34355_cast_fp16 = einsum(equation = var_34355_equation_0, values = (var_33801_cast_fp16, var_34201_cast_fp16))[name = tensor("op_34355_cast_fp16")]; + tensor var_34357_equation_0 = const()[name = tensor("op_34357_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34357_cast_fp16 = einsum(equation = var_34357_equation_0, values = (var_33801_cast_fp16, var_34202_cast_fp16))[name = tensor("op_34357_cast_fp16")]; + tensor var_34359_equation_0 = const()[name = tensor("op_34359_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34359_cast_fp16 = einsum(equation = var_34359_equation_0, values = (var_33801_cast_fp16, var_34203_cast_fp16))[name = tensor("op_34359_cast_fp16")]; + tensor var_34361_equation_0 = const()[name = tensor("op_34361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34361_cast_fp16 = einsum(equation = var_34361_equation_0, values = (var_33805_cast_fp16, var_34204_cast_fp16))[name = tensor("op_34361_cast_fp16")]; + tensor var_34363_equation_0 = const()[name = tensor("op_34363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34363_cast_fp16 = einsum(equation = var_34363_equation_0, values = (var_33805_cast_fp16, var_34205_cast_fp16))[name = tensor("op_34363_cast_fp16")]; + tensor var_34365_equation_0 = const()[name = tensor("op_34365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34365_cast_fp16 = einsum(equation = var_34365_equation_0, values = (var_33805_cast_fp16, var_34206_cast_fp16))[name = tensor("op_34365_cast_fp16")]; + tensor var_34367_equation_0 = const()[name = tensor("op_34367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_34367_cast_fp16 = einsum(equation = var_34367_equation_0, values = (var_33805_cast_fp16, var_34207_cast_fp16))[name = tensor("op_34367_cast_fp16")]; + tensor var_34369_interleave_0 = const()[name = tensor("op_34369_interleave_0"), val = tensor(false)]; + tensor var_34369_cast_fp16 = concat(axis = var_32874, interleave = var_34369_interleave_0, values = (var_34209_cast_fp16, var_34211_cast_fp16, var_34213_cast_fp16, var_34215_cast_fp16))[name = tensor("op_34369_cast_fp16")]; + tensor var_34371_interleave_0 = const()[name = tensor("op_34371_interleave_0"), val = tensor(false)]; + tensor var_34371_cast_fp16 = concat(axis = var_32874, interleave = var_34371_interleave_0, values = (var_34217_cast_fp16, var_34219_cast_fp16, var_34221_cast_fp16, var_34223_cast_fp16))[name = tensor("op_34371_cast_fp16")]; + tensor var_34373_interleave_0 = const()[name = tensor("op_34373_interleave_0"), val = tensor(false)]; + tensor var_34373_cast_fp16 = concat(axis = var_32874, interleave = var_34373_interleave_0, values = (var_34225_cast_fp16, var_34227_cast_fp16, var_34229_cast_fp16, var_34231_cast_fp16))[name = tensor("op_34373_cast_fp16")]; + tensor var_34375_interleave_0 = const()[name = tensor("op_34375_interleave_0"), val = tensor(false)]; + tensor var_34375_cast_fp16 = concat(axis = var_32874, interleave = var_34375_interleave_0, values = (var_34233_cast_fp16, var_34235_cast_fp16, var_34237_cast_fp16, var_34239_cast_fp16))[name = tensor("op_34375_cast_fp16")]; + tensor var_34377_interleave_0 = const()[name = tensor("op_34377_interleave_0"), val = tensor(false)]; + tensor var_34377_cast_fp16 = concat(axis = var_32874, interleave = var_34377_interleave_0, values = (var_34241_cast_fp16, var_34243_cast_fp16, var_34245_cast_fp16, var_34247_cast_fp16))[name = tensor("op_34377_cast_fp16")]; + tensor var_34379_interleave_0 = const()[name = tensor("op_34379_interleave_0"), val = tensor(false)]; + tensor var_34379_cast_fp16 = concat(axis = var_32874, interleave = var_34379_interleave_0, values = (var_34249_cast_fp16, var_34251_cast_fp16, var_34253_cast_fp16, var_34255_cast_fp16))[name = tensor("op_34379_cast_fp16")]; + tensor var_34381_interleave_0 = const()[name = tensor("op_34381_interleave_0"), val = tensor(false)]; + tensor var_34381_cast_fp16 = concat(axis = var_32874, interleave = var_34381_interleave_0, values = (var_34257_cast_fp16, var_34259_cast_fp16, var_34261_cast_fp16, var_34263_cast_fp16))[name = tensor("op_34381_cast_fp16")]; + tensor var_34383_interleave_0 = const()[name = tensor("op_34383_interleave_0"), val = tensor(false)]; + tensor var_34383_cast_fp16 = concat(axis = var_32874, interleave = var_34383_interleave_0, values = (var_34265_cast_fp16, var_34267_cast_fp16, var_34269_cast_fp16, var_34271_cast_fp16))[name = tensor("op_34383_cast_fp16")]; + tensor var_34385_interleave_0 = const()[name = tensor("op_34385_interleave_0"), val = tensor(false)]; + tensor var_34385_cast_fp16 = concat(axis = var_32874, interleave = var_34385_interleave_0, values = (var_34273_cast_fp16, var_34275_cast_fp16, var_34277_cast_fp16, var_34279_cast_fp16))[name = tensor("op_34385_cast_fp16")]; + tensor var_34387_interleave_0 = const()[name = tensor("op_34387_interleave_0"), val = tensor(false)]; + tensor var_34387_cast_fp16 = concat(axis = var_32874, interleave = var_34387_interleave_0, values = (var_34281_cast_fp16, var_34283_cast_fp16, var_34285_cast_fp16, var_34287_cast_fp16))[name = tensor("op_34387_cast_fp16")]; + tensor var_34389_interleave_0 = const()[name = tensor("op_34389_interleave_0"), val = tensor(false)]; + tensor var_34389_cast_fp16 = concat(axis = var_32874, interleave = var_34389_interleave_0, values = (var_34289_cast_fp16, var_34291_cast_fp16, var_34293_cast_fp16, var_34295_cast_fp16))[name = tensor("op_34389_cast_fp16")]; + tensor var_34391_interleave_0 = const()[name = tensor("op_34391_interleave_0"), val = tensor(false)]; + tensor var_34391_cast_fp16 = concat(axis = var_32874, interleave = var_34391_interleave_0, values = (var_34297_cast_fp16, var_34299_cast_fp16, var_34301_cast_fp16, var_34303_cast_fp16))[name = tensor("op_34391_cast_fp16")]; + tensor var_34393_interleave_0 = const()[name = tensor("op_34393_interleave_0"), val = tensor(false)]; + tensor var_34393_cast_fp16 = concat(axis = var_32874, interleave = var_34393_interleave_0, values = (var_34305_cast_fp16, var_34307_cast_fp16, var_34309_cast_fp16, var_34311_cast_fp16))[name = tensor("op_34393_cast_fp16")]; + tensor var_34395_interleave_0 = const()[name = tensor("op_34395_interleave_0"), val = tensor(false)]; + tensor var_34395_cast_fp16 = concat(axis = var_32874, interleave = var_34395_interleave_0, values = (var_34313_cast_fp16, var_34315_cast_fp16, var_34317_cast_fp16, var_34319_cast_fp16))[name = tensor("op_34395_cast_fp16")]; + tensor var_34397_interleave_0 = const()[name = tensor("op_34397_interleave_0"), val = tensor(false)]; + tensor var_34397_cast_fp16 = concat(axis = var_32874, interleave = var_34397_interleave_0, values = (var_34321_cast_fp16, var_34323_cast_fp16, var_34325_cast_fp16, var_34327_cast_fp16))[name = tensor("op_34397_cast_fp16")]; + tensor var_34399_interleave_0 = const()[name = tensor("op_34399_interleave_0"), val = tensor(false)]; + tensor var_34399_cast_fp16 = concat(axis = var_32874, interleave = var_34399_interleave_0, values = (var_34329_cast_fp16, var_34331_cast_fp16, var_34333_cast_fp16, var_34335_cast_fp16))[name = tensor("op_34399_cast_fp16")]; + tensor var_34401_interleave_0 = const()[name = tensor("op_34401_interleave_0"), val = tensor(false)]; + tensor var_34401_cast_fp16 = concat(axis = var_32874, interleave = var_34401_interleave_0, values = (var_34337_cast_fp16, var_34339_cast_fp16, var_34341_cast_fp16, var_34343_cast_fp16))[name = tensor("op_34401_cast_fp16")]; + tensor var_34403_interleave_0 = const()[name = tensor("op_34403_interleave_0"), val = tensor(false)]; + tensor var_34403_cast_fp16 = concat(axis = var_32874, interleave = var_34403_interleave_0, values = (var_34345_cast_fp16, var_34347_cast_fp16, var_34349_cast_fp16, var_34351_cast_fp16))[name = tensor("op_34403_cast_fp16")]; + tensor var_34405_interleave_0 = const()[name = tensor("op_34405_interleave_0"), val = tensor(false)]; + tensor var_34405_cast_fp16 = concat(axis = var_32874, interleave = var_34405_interleave_0, values = (var_34353_cast_fp16, var_34355_cast_fp16, var_34357_cast_fp16, var_34359_cast_fp16))[name = tensor("op_34405_cast_fp16")]; + tensor var_34407_interleave_0 = const()[name = tensor("op_34407_interleave_0"), val = tensor(false)]; + tensor var_34407_cast_fp16 = concat(axis = var_32874, interleave = var_34407_interleave_0, values = (var_34361_cast_fp16, var_34363_cast_fp16, var_34365_cast_fp16, var_34367_cast_fp16))[name = tensor("op_34407_cast_fp16")]; + tensor input_407_interleave_0 = const()[name = tensor("input_407_interleave_0"), val = tensor(false)]; + tensor input_407_cast_fp16 = concat(axis = var_32899, interleave = input_407_interleave_0, values = (var_34369_cast_fp16, var_34371_cast_fp16, var_34373_cast_fp16, var_34375_cast_fp16, var_34377_cast_fp16, var_34379_cast_fp16, var_34381_cast_fp16, var_34383_cast_fp16, var_34385_cast_fp16, var_34387_cast_fp16, var_34389_cast_fp16, var_34391_cast_fp16, var_34393_cast_fp16, var_34395_cast_fp16, var_34397_cast_fp16, var_34399_cast_fp16, var_34401_cast_fp16, var_34403_cast_fp16, var_34405_cast_fp16, var_34407_cast_fp16))[name = tensor("input_407_cast_fp16")]; + tensor var_34415 = const()[name = tensor("op_34415"), val = tensor([1, 1])]; + tensor var_34417 = const()[name = tensor("op_34417"), val = tensor([1, 1])]; + tensor pretrained_out_247_pad_type_0 = const()[name = tensor("pretrained_out_247_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_247_pad_0 = const()[name = tensor("pretrained_out_247_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229038592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229857856))), name = tensor("layers_20_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_20_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_20_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229857984)))]; + tensor pretrained_out_247_cast_fp16 = conv(bias = layers_20_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_34417, groups = var_32899, pad = pretrained_out_247_pad_0, pad_type = pretrained_out_247_pad_type_0, strides = var_34415, weight = layers_20_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_407_cast_fp16)[name = tensor("pretrained_out_247_cast_fp16")]; + tensor var_34421 = const()[name = tensor("op_34421"), val = tensor([1, 1])]; + tensor var_34423 = const()[name = tensor("op_34423"), val = tensor([1, 1])]; + tensor input_409_pad_type_0 = const()[name = tensor("input_409_pad_type_0"), val = tensor("custom")]; + tensor input_409_pad_0 = const()[name = tensor("input_409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_20_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229860608)))]; + tensor input_409_cast_fp16 = conv(dilations = var_34423, groups = var_32899, pad = input_409_pad_0, pad_type = input_409_pad_type_0, strides = var_34421, weight = layers_20_self_attn_o_proj_loraA_weight_to_fp16, x = input_407_cast_fp16)[name = tensor("input_409_cast_fp16")]; + tensor var_34427 = const()[name = tensor("op_34427"), val = tensor([1, 1])]; + tensor var_34429 = const()[name = tensor("op_34429"), val = tensor([1, 1])]; + tensor lora_out_493_pad_type_0 = const()[name = tensor("lora_out_493_pad_type_0"), val = tensor("custom")]; + tensor lora_out_493_pad_0 = const()[name = tensor("lora_out_493_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_495_weight_0_to_fp16 = const()[name = tensor("lora_out_495_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229901632)))]; + tensor lora_out_495_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_34429, groups = var_32899, pad = lora_out_493_pad_0, pad_type = lora_out_493_pad_type_0, strides = var_34427, weight = lora_out_495_weight_0_to_fp16, x = input_409_cast_fp16)[name = tensor("lora_out_495_cast_fp16")]; + tensor obj_83_cast_fp16 = add(x = pretrained_out_247_cast_fp16, y = lora_out_495_cast_fp16)[name = tensor("obj_83_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = obj_83_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; + tensor var_34438 = const()[name = tensor("op_34438"), val = tensor([1])]; + tensor channels_mean_83_cast_fp16 = reduce_mean(axes = var_34438, keep_dims = var_32900, x = inputs_83_cast_fp16)[name = tensor("channels_mean_83_cast_fp16")]; + tensor zero_mean_83_cast_fp16 = sub(x = inputs_83_cast_fp16, y = channels_mean_83_cast_fp16)[name = tensor("zero_mean_83_cast_fp16")]; + tensor zero_mean_sq_83_cast_fp16 = mul(x = zero_mean_83_cast_fp16, y = zero_mean_83_cast_fp16)[name = tensor("zero_mean_sq_83_cast_fp16")]; + tensor var_34442 = const()[name = tensor("op_34442"), val = tensor([1])]; + tensor var_34443_cast_fp16 = reduce_mean(axes = var_34442, keep_dims = var_32900, x = zero_mean_sq_83_cast_fp16)[name = tensor("op_34443_cast_fp16")]; + tensor var_34444_to_fp16 = const()[name = tensor("op_34444_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_34445_cast_fp16 = add(x = var_34443_cast_fp16, y = var_34444_to_fp16)[name = tensor("op_34445_cast_fp16")]; + tensor denom_83_epsilon_0 = const()[name = tensor("denom_83_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_83_cast_fp16 = rsqrt(epsilon = denom_83_epsilon_0, x = var_34445_cast_fp16)[name = tensor("denom_83_cast_fp16")]; + tensor out_83_cast_fp16 = mul(x = zero_mean_83_cast_fp16, y = denom_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; + tensor input_411_gamma_0_to_fp16 = const()[name = tensor("input_411_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229942656)))]; + tensor input_411_beta_0_to_fp16 = const()[name = tensor("input_411_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229945280)))]; + tensor input_411_epsilon_0_to_fp16 = const()[name = tensor("input_411_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_411_cast_fp16 = batch_norm(beta = input_411_beta_0_to_fp16, epsilon = input_411_epsilon_0_to_fp16, gamma = input_411_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor("input_411_cast_fp16")]; + tensor var_34459 = const()[name = tensor("op_34459"), val = tensor([1, 1])]; + tensor var_34461 = const()[name = tensor("op_34461"), val = tensor([1, 1])]; + tensor pretrained_out_249_pad_type_0 = const()[name = tensor("pretrained_out_249_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_249_pad_0 = const()[name = tensor("pretrained_out_249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(229947904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233224768))), name = tensor("layers_20_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_20_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_20_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233224896)))]; + tensor pretrained_out_249_cast_fp16 = conv(bias = layers_20_fc1_pretrained_bias_to_fp16, dilations = var_34461, groups = var_32899, pad = pretrained_out_249_pad_0, pad_type = pretrained_out_249_pad_type_0, strides = var_34459, weight = layers_20_fc1_pretrained_weight_to_fp16_palettized, x = input_411_cast_fp16)[name = tensor("pretrained_out_249_cast_fp16")]; + tensor var_34465 = const()[name = tensor("op_34465"), val = tensor([1, 1])]; + tensor var_34467 = const()[name = tensor("op_34467"), val = tensor([1, 1])]; + tensor input_413_pad_type_0 = const()[name = tensor("input_413_pad_type_0"), val = tensor("custom")]; + tensor input_413_pad_0 = const()[name = tensor("input_413_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_20_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233235200)))]; + tensor input_413_cast_fp16 = conv(dilations = var_34467, groups = var_32899, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = var_34465, weight = layers_20_fc1_loraA_weight_to_fp16, x = input_411_cast_fp16)[name = tensor("input_413_cast_fp16")]; + tensor var_34471 = const()[name = tensor("op_34471"), val = tensor([1, 1])]; + tensor var_34473 = const()[name = tensor("op_34473"), val = tensor([1, 1])]; + tensor lora_out_497_pad_type_0 = const()[name = tensor("lora_out_497_pad_type_0"), val = tensor("custom")]; + tensor lora_out_497_pad_0 = const()[name = tensor("lora_out_497_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_499_weight_0_to_fp16 = const()[name = tensor("lora_out_499_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233276224)))]; + tensor lora_out_499_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_34473, groups = var_32899, pad = lora_out_497_pad_0, pad_type = lora_out_497_pad_type_0, strides = var_34471, weight = lora_out_499_weight_0_to_fp16, x = input_413_cast_fp16)[name = tensor("lora_out_499_cast_fp16")]; + tensor input_415_cast_fp16 = add(x = pretrained_out_249_cast_fp16, y = lora_out_499_cast_fp16)[name = tensor("input_415_cast_fp16")]; + tensor input_417_mode_0 = const()[name = tensor("input_417_mode_0"), val = tensor("EXACT")]; + tensor input_417_cast_fp16 = gelu(mode = input_417_mode_0, x = input_415_cast_fp16)[name = tensor("input_417_cast_fp16")]; + tensor var_34485 = const()[name = tensor("op_34485"), val = tensor([1, 1])]; + tensor var_34487 = const()[name = tensor("op_34487"), val = tensor([1, 1])]; + tensor pretrained_out_251_pad_type_0 = const()[name = tensor("pretrained_out_251_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_251_pad_0 = const()[name = tensor("pretrained_out_251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(233440128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236716992))), name = tensor("layers_20_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_20_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_20_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236717120)))]; + tensor pretrained_out_251_cast_fp16 = conv(bias = layers_20_fc2_pretrained_bias_to_fp16, dilations = var_34487, groups = var_32899, pad = pretrained_out_251_pad_0, pad_type = pretrained_out_251_pad_type_0, strides = var_34485, weight = layers_20_fc2_pretrained_weight_to_fp16_palettized, x = input_417_cast_fp16)[name = tensor("pretrained_out_251_cast_fp16")]; + tensor var_34491 = const()[name = tensor("op_34491"), val = tensor([1, 1])]; + tensor var_34493 = const()[name = tensor("op_34493"), val = tensor([1, 1])]; + tensor input_419_pad_type_0 = const()[name = tensor("input_419_pad_type_0"), val = tensor("custom")]; + tensor input_419_pad_0 = const()[name = tensor("input_419_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_20_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_20_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236719744)))]; + tensor input_419_cast_fp16 = conv(dilations = var_34493, groups = var_32899, pad = input_419_pad_0, pad_type = input_419_pad_type_0, strides = var_34491, weight = layers_20_fc2_loraA_weight_to_fp16, x = input_417_cast_fp16)[name = tensor("input_419_cast_fp16")]; + tensor var_34497 = const()[name = tensor("op_34497"), val = tensor([1, 1])]; + tensor var_34499 = const()[name = tensor("op_34499"), val = tensor([1, 1])]; + tensor lora_out_501_pad_type_0 = const()[name = tensor("lora_out_501_pad_type_0"), val = tensor("custom")]; + tensor lora_out_501_pad_0 = const()[name = tensor("lora_out_501_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_503_weight_0_to_fp16 = const()[name = tensor("lora_out_503_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236883648)))]; + tensor lora_out_503_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_34499, groups = var_32899, pad = lora_out_501_pad_0, pad_type = lora_out_501_pad_type_0, strides = var_34497, weight = lora_out_503_weight_0_to_fp16, x = input_419_cast_fp16)[name = tensor("lora_out_503_cast_fp16")]; + tensor hidden_states_45_cast_fp16 = add(x = pretrained_out_251_cast_fp16, y = lora_out_503_cast_fp16)[name = tensor("hidden_states_45_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = hidden_states_45_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; + tensor var_34509 = const()[name = tensor("op_34509"), val = tensor(3)]; + tensor var_34534 = const()[name = tensor("op_34534"), val = tensor(1)]; + tensor var_34535 = const()[name = tensor("op_34535"), val = tensor(true)]; + tensor var_34545 = const()[name = tensor("op_34545"), val = tensor([1])]; + tensor channels_mean_85_cast_fp16 = reduce_mean(axes = var_34545, keep_dims = var_34535, x = inputs_85_cast_fp16)[name = tensor("channels_mean_85_cast_fp16")]; + tensor zero_mean_85_cast_fp16 = sub(x = inputs_85_cast_fp16, y = channels_mean_85_cast_fp16)[name = tensor("zero_mean_85_cast_fp16")]; + tensor zero_mean_sq_85_cast_fp16 = mul(x = zero_mean_85_cast_fp16, y = zero_mean_85_cast_fp16)[name = tensor("zero_mean_sq_85_cast_fp16")]; + tensor var_34549 = const()[name = tensor("op_34549"), val = tensor([1])]; + tensor var_34550_cast_fp16 = reduce_mean(axes = var_34549, keep_dims = var_34535, x = zero_mean_sq_85_cast_fp16)[name = tensor("op_34550_cast_fp16")]; + tensor var_34551_to_fp16 = const()[name = tensor("op_34551_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_34552_cast_fp16 = add(x = var_34550_cast_fp16, y = var_34551_to_fp16)[name = tensor("op_34552_cast_fp16")]; + tensor denom_85_epsilon_0 = const()[name = tensor("denom_85_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_85_cast_fp16 = rsqrt(epsilon = denom_85_epsilon_0, x = var_34552_cast_fp16)[name = tensor("denom_85_cast_fp16")]; + tensor out_85_cast_fp16 = mul(x = zero_mean_85_cast_fp16, y = denom_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; + tensor obj_85_gamma_0_to_fp16 = const()[name = tensor("obj_85_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236924672)))]; + tensor obj_85_beta_0_to_fp16 = const()[name = tensor("obj_85_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236927296)))]; + tensor obj_85_epsilon_0_to_fp16 = const()[name = tensor("obj_85_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_85_cast_fp16 = batch_norm(beta = obj_85_beta_0_to_fp16, epsilon = obj_85_epsilon_0_to_fp16, gamma = obj_85_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor("obj_85_cast_fp16")]; + tensor var_34570 = const()[name = tensor("op_34570"), val = tensor([1, 1])]; + tensor var_34572 = const()[name = tensor("op_34572"), val = tensor([1, 1])]; + tensor pretrained_out_253_pad_type_0 = const()[name = tensor("pretrained_out_253_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_253_pad_0 = const()[name = tensor("pretrained_out_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236929920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237749184))), name = tensor("layers_21_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_21_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237749312)))]; + tensor pretrained_out_253_cast_fp16 = conv(bias = layers_21_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_34572, groups = var_34534, pad = pretrained_out_253_pad_0, pad_type = pretrained_out_253_pad_type_0, strides = var_34570, weight = layers_21_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("pretrained_out_253_cast_fp16")]; + tensor var_34576 = const()[name = tensor("op_34576"), val = tensor([1, 1])]; + tensor var_34578 = const()[name = tensor("op_34578"), val = tensor([1, 1])]; + tensor input_421_pad_type_0 = const()[name = tensor("input_421_pad_type_0"), val = tensor("custom")]; + tensor input_421_pad_0 = const()[name = tensor("input_421_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237751936)))]; + tensor input_421_cast_fp16 = conv(dilations = var_34578, groups = var_34534, pad = input_421_pad_0, pad_type = input_421_pad_type_0, strides = var_34576, weight = layers_21_self_attn_q_proj_loraA_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("input_421_cast_fp16")]; + tensor var_34582 = const()[name = tensor("op_34582"), val = tensor([1, 1])]; + tensor var_34584 = const()[name = tensor("op_34584"), val = tensor([1, 1])]; + tensor lora_out_505_pad_type_0 = const()[name = tensor("lora_out_505_pad_type_0"), val = tensor("custom")]; + tensor lora_out_505_pad_0 = const()[name = tensor("lora_out_505_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_507_weight_0_to_fp16 = const()[name = tensor("lora_out_507_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237792960)))]; + tensor lora_out_507_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_34584, groups = var_34534, pad = lora_out_505_pad_0, pad_type = lora_out_505_pad_type_0, strides = var_34582, weight = lora_out_507_weight_0_to_fp16, x = input_421_cast_fp16)[name = tensor("lora_out_507_cast_fp16")]; + tensor query_43_cast_fp16 = add(x = pretrained_out_253_cast_fp16, y = lora_out_507_cast_fp16)[name = tensor("query_43_cast_fp16")]; + tensor var_34594 = const()[name = tensor("op_34594"), val = tensor([1, 1])]; + tensor var_34596 = const()[name = tensor("op_34596"), val = tensor([1, 1])]; + tensor pretrained_out_255_pad_type_0 = const()[name = tensor("pretrained_out_255_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_255_pad_0 = const()[name = tensor("pretrained_out_255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237833984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238653248))), name = tensor("layers_21_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_255_cast_fp16 = conv(dilations = var_34596, groups = var_34534, pad = pretrained_out_255_pad_0, pad_type = pretrained_out_255_pad_type_0, strides = var_34594, weight = layers_21_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("pretrained_out_255_cast_fp16")]; + tensor var_34600 = const()[name = tensor("op_34600"), val = tensor([1, 1])]; + tensor var_34602 = const()[name = tensor("op_34602"), val = tensor([1, 1])]; + tensor input_423_pad_type_0 = const()[name = tensor("input_423_pad_type_0"), val = tensor("custom")]; + tensor input_423_pad_0 = const()[name = tensor("input_423_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238653376)))]; + tensor input_423_cast_fp16 = conv(dilations = var_34602, groups = var_34534, pad = input_423_pad_0, pad_type = input_423_pad_type_0, strides = var_34600, weight = layers_21_self_attn_k_proj_loraA_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("input_423_cast_fp16")]; + tensor var_34606 = const()[name = tensor("op_34606"), val = tensor([1, 1])]; + tensor var_34608 = const()[name = tensor("op_34608"), val = tensor([1, 1])]; + tensor lora_out_509_pad_type_0 = const()[name = tensor("lora_out_509_pad_type_0"), val = tensor("custom")]; + tensor lora_out_509_pad_0 = const()[name = tensor("lora_out_509_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_511_weight_0_to_fp16 = const()[name = tensor("lora_out_511_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238694400)))]; + tensor lora_out_511_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_34608, groups = var_34534, pad = lora_out_509_pad_0, pad_type = lora_out_509_pad_type_0, strides = var_34606, weight = lora_out_511_weight_0_to_fp16, x = input_423_cast_fp16)[name = tensor("lora_out_511_cast_fp16")]; + tensor key_43_cast_fp16 = add(x = pretrained_out_255_cast_fp16, y = lora_out_511_cast_fp16)[name = tensor("key_43_cast_fp16")]; + tensor var_34619 = const()[name = tensor("op_34619"), val = tensor([1, 1])]; + tensor var_34621 = const()[name = tensor("op_34621"), val = tensor([1, 1])]; + tensor pretrained_out_257_pad_type_0 = const()[name = tensor("pretrained_out_257_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_257_pad_0 = const()[name = tensor("pretrained_out_257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238735424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239554688))), name = tensor("layers_21_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_21_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239554816)))]; + tensor pretrained_out_257_cast_fp16 = conv(bias = layers_21_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_34621, groups = var_34534, pad = pretrained_out_257_pad_0, pad_type = pretrained_out_257_pad_type_0, strides = var_34619, weight = layers_21_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_85_cast_fp16)[name = tensor("pretrained_out_257_cast_fp16")]; + tensor var_34625 = const()[name = tensor("op_34625"), val = tensor([1, 1])]; + tensor var_34627 = const()[name = tensor("op_34627"), val = tensor([1, 1])]; + tensor input_425_pad_type_0 = const()[name = tensor("input_425_pad_type_0"), val = tensor("custom")]; + tensor input_425_pad_0 = const()[name = tensor("input_425_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239557440)))]; + tensor input_425_cast_fp16 = conv(dilations = var_34627, groups = var_34534, pad = input_425_pad_0, pad_type = input_425_pad_type_0, strides = var_34625, weight = layers_21_self_attn_v_proj_loraA_weight_to_fp16, x = obj_85_cast_fp16)[name = tensor("input_425_cast_fp16")]; + tensor var_34631 = const()[name = tensor("op_34631"), val = tensor([1, 1])]; + tensor var_34633 = const()[name = tensor("op_34633"), val = tensor([1, 1])]; + tensor lora_out_513_pad_type_0 = const()[name = tensor("lora_out_513_pad_type_0"), val = tensor("custom")]; + tensor lora_out_513_pad_0 = const()[name = tensor("lora_out_513_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_515_weight_0_to_fp16 = const()[name = tensor("lora_out_515_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239598464)))]; + tensor lora_out_515_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_34633, groups = var_34534, pad = lora_out_513_pad_0, pad_type = lora_out_513_pad_type_0, strides = var_34631, weight = lora_out_515_weight_0_to_fp16, x = input_425_cast_fp16)[name = tensor("lora_out_515_cast_fp16")]; + tensor value_43_cast_fp16 = add(x = pretrained_out_257_cast_fp16, y = lora_out_515_cast_fp16)[name = tensor("value_43_cast_fp16")]; + tensor var_34643_begin_0 = const()[name = tensor("op_34643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34643_end_0 = const()[name = tensor("op_34643_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34643_end_mask_0 = const()[name = tensor("op_34643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34643_cast_fp16 = slice_by_index(begin = var_34643_begin_0, end = var_34643_end_0, end_mask = var_34643_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34643_cast_fp16")]; + tensor var_34647_begin_0 = const()[name = tensor("op_34647_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_34647_end_0 = const()[name = tensor("op_34647_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_34647_end_mask_0 = const()[name = tensor("op_34647_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34647_cast_fp16 = slice_by_index(begin = var_34647_begin_0, end = var_34647_end_0, end_mask = var_34647_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34647_cast_fp16")]; + tensor var_34651_begin_0 = const()[name = tensor("op_34651_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_34651_end_0 = const()[name = tensor("op_34651_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_34651_end_mask_0 = const()[name = tensor("op_34651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34651_cast_fp16 = slice_by_index(begin = var_34651_begin_0, end = var_34651_end_0, end_mask = var_34651_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34651_cast_fp16")]; + tensor var_34655_begin_0 = const()[name = tensor("op_34655_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_34655_end_0 = const()[name = tensor("op_34655_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_34655_end_mask_0 = const()[name = tensor("op_34655_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34655_cast_fp16 = slice_by_index(begin = var_34655_begin_0, end = var_34655_end_0, end_mask = var_34655_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34655_cast_fp16")]; + tensor var_34659_begin_0 = const()[name = tensor("op_34659_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_34659_end_0 = const()[name = tensor("op_34659_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_34659_end_mask_0 = const()[name = tensor("op_34659_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34659_cast_fp16 = slice_by_index(begin = var_34659_begin_0, end = var_34659_end_0, end_mask = var_34659_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34659_cast_fp16")]; + tensor var_34663_begin_0 = const()[name = tensor("op_34663_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_34663_end_0 = const()[name = tensor("op_34663_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_34663_end_mask_0 = const()[name = tensor("op_34663_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34663_cast_fp16 = slice_by_index(begin = var_34663_begin_0, end = var_34663_end_0, end_mask = var_34663_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34663_cast_fp16")]; + tensor var_34667_begin_0 = const()[name = tensor("op_34667_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_34667_end_0 = const()[name = tensor("op_34667_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_34667_end_mask_0 = const()[name = tensor("op_34667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34667_cast_fp16 = slice_by_index(begin = var_34667_begin_0, end = var_34667_end_0, end_mask = var_34667_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34667_cast_fp16")]; + tensor var_34671_begin_0 = const()[name = tensor("op_34671_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_34671_end_0 = const()[name = tensor("op_34671_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_34671_end_mask_0 = const()[name = tensor("op_34671_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34671_cast_fp16 = slice_by_index(begin = var_34671_begin_0, end = var_34671_end_0, end_mask = var_34671_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34671_cast_fp16")]; + tensor var_34675_begin_0 = const()[name = tensor("op_34675_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_34675_end_0 = const()[name = tensor("op_34675_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_34675_end_mask_0 = const()[name = tensor("op_34675_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34675_cast_fp16 = slice_by_index(begin = var_34675_begin_0, end = var_34675_end_0, end_mask = var_34675_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34675_cast_fp16")]; + tensor var_34679_begin_0 = const()[name = tensor("op_34679_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_34679_end_0 = const()[name = tensor("op_34679_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_34679_end_mask_0 = const()[name = tensor("op_34679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34679_cast_fp16 = slice_by_index(begin = var_34679_begin_0, end = var_34679_end_0, end_mask = var_34679_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34679_cast_fp16")]; + tensor var_34683_begin_0 = const()[name = tensor("op_34683_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_34683_end_0 = const()[name = tensor("op_34683_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_34683_end_mask_0 = const()[name = tensor("op_34683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34683_cast_fp16 = slice_by_index(begin = var_34683_begin_0, end = var_34683_end_0, end_mask = var_34683_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34683_cast_fp16")]; + tensor var_34687_begin_0 = const()[name = tensor("op_34687_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_34687_end_0 = const()[name = tensor("op_34687_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_34687_end_mask_0 = const()[name = tensor("op_34687_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34687_cast_fp16 = slice_by_index(begin = var_34687_begin_0, end = var_34687_end_0, end_mask = var_34687_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34687_cast_fp16")]; + tensor var_34691_begin_0 = const()[name = tensor("op_34691_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_34691_end_0 = const()[name = tensor("op_34691_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_34691_end_mask_0 = const()[name = tensor("op_34691_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34691_cast_fp16 = slice_by_index(begin = var_34691_begin_0, end = var_34691_end_0, end_mask = var_34691_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34691_cast_fp16")]; + tensor var_34695_begin_0 = const()[name = tensor("op_34695_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_34695_end_0 = const()[name = tensor("op_34695_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_34695_end_mask_0 = const()[name = tensor("op_34695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34695_cast_fp16 = slice_by_index(begin = var_34695_begin_0, end = var_34695_end_0, end_mask = var_34695_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34695_cast_fp16")]; + tensor var_34699_begin_0 = const()[name = tensor("op_34699_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_34699_end_0 = const()[name = tensor("op_34699_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_34699_end_mask_0 = const()[name = tensor("op_34699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34699_cast_fp16 = slice_by_index(begin = var_34699_begin_0, end = var_34699_end_0, end_mask = var_34699_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34699_cast_fp16")]; + tensor var_34703_begin_0 = const()[name = tensor("op_34703_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_34703_end_0 = const()[name = tensor("op_34703_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_34703_end_mask_0 = const()[name = tensor("op_34703_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34703_cast_fp16 = slice_by_index(begin = var_34703_begin_0, end = var_34703_end_0, end_mask = var_34703_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34703_cast_fp16")]; + tensor var_34707_begin_0 = const()[name = tensor("op_34707_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_34707_end_0 = const()[name = tensor("op_34707_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_34707_end_mask_0 = const()[name = tensor("op_34707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34707_cast_fp16 = slice_by_index(begin = var_34707_begin_0, end = var_34707_end_0, end_mask = var_34707_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34707_cast_fp16")]; + tensor var_34711_begin_0 = const()[name = tensor("op_34711_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_34711_end_0 = const()[name = tensor("op_34711_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_34711_end_mask_0 = const()[name = tensor("op_34711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34711_cast_fp16 = slice_by_index(begin = var_34711_begin_0, end = var_34711_end_0, end_mask = var_34711_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34711_cast_fp16")]; + tensor var_34715_begin_0 = const()[name = tensor("op_34715_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_34715_end_0 = const()[name = tensor("op_34715_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_34715_end_mask_0 = const()[name = tensor("op_34715_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34715_cast_fp16 = slice_by_index(begin = var_34715_begin_0, end = var_34715_end_0, end_mask = var_34715_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34715_cast_fp16")]; + tensor var_34719_begin_0 = const()[name = tensor("op_34719_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_34719_end_0 = const()[name = tensor("op_34719_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_34719_end_mask_0 = const()[name = tensor("op_34719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_34719_cast_fp16 = slice_by_index(begin = var_34719_begin_0, end = var_34719_end_0, end_mask = var_34719_end_mask_0, x = query_43_cast_fp16)[name = tensor("op_34719_cast_fp16")]; + tensor var_34728_begin_0 = const()[name = tensor("op_34728_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34728_end_0 = const()[name = tensor("op_34728_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34728_end_mask_0 = const()[name = tensor("op_34728_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34728_cast_fp16 = slice_by_index(begin = var_34728_begin_0, end = var_34728_end_0, end_mask = var_34728_end_mask_0, x = var_34643_cast_fp16)[name = tensor("op_34728_cast_fp16")]; + tensor var_34735_begin_0 = const()[name = tensor("op_34735_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34735_end_0 = const()[name = tensor("op_34735_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34735_end_mask_0 = const()[name = tensor("op_34735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34735_cast_fp16 = slice_by_index(begin = var_34735_begin_0, end = var_34735_end_0, end_mask = var_34735_end_mask_0, x = var_34643_cast_fp16)[name = tensor("op_34735_cast_fp16")]; + tensor var_34742_begin_0 = const()[name = tensor("op_34742_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34742_end_0 = const()[name = tensor("op_34742_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34742_end_mask_0 = const()[name = tensor("op_34742_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34742_cast_fp16 = slice_by_index(begin = var_34742_begin_0, end = var_34742_end_0, end_mask = var_34742_end_mask_0, x = var_34643_cast_fp16)[name = tensor("op_34742_cast_fp16")]; + tensor var_34749_begin_0 = const()[name = tensor("op_34749_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34749_end_0 = const()[name = tensor("op_34749_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34749_end_mask_0 = const()[name = tensor("op_34749_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34749_cast_fp16 = slice_by_index(begin = var_34749_begin_0, end = var_34749_end_0, end_mask = var_34749_end_mask_0, x = var_34643_cast_fp16)[name = tensor("op_34749_cast_fp16")]; + tensor var_34756_begin_0 = const()[name = tensor("op_34756_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34756_end_0 = const()[name = tensor("op_34756_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34756_end_mask_0 = const()[name = tensor("op_34756_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34756_cast_fp16 = slice_by_index(begin = var_34756_begin_0, end = var_34756_end_0, end_mask = var_34756_end_mask_0, x = var_34647_cast_fp16)[name = tensor("op_34756_cast_fp16")]; + tensor var_34763_begin_0 = const()[name = tensor("op_34763_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34763_end_0 = const()[name = tensor("op_34763_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34763_end_mask_0 = const()[name = tensor("op_34763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34763_cast_fp16 = slice_by_index(begin = var_34763_begin_0, end = var_34763_end_0, end_mask = var_34763_end_mask_0, x = var_34647_cast_fp16)[name = tensor("op_34763_cast_fp16")]; + tensor var_34770_begin_0 = const()[name = tensor("op_34770_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34770_end_0 = const()[name = tensor("op_34770_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34770_end_mask_0 = const()[name = tensor("op_34770_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34770_cast_fp16 = slice_by_index(begin = var_34770_begin_0, end = var_34770_end_0, end_mask = var_34770_end_mask_0, x = var_34647_cast_fp16)[name = tensor("op_34770_cast_fp16")]; + tensor var_34777_begin_0 = const()[name = tensor("op_34777_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34777_end_0 = const()[name = tensor("op_34777_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34777_end_mask_0 = const()[name = tensor("op_34777_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34777_cast_fp16 = slice_by_index(begin = var_34777_begin_0, end = var_34777_end_0, end_mask = var_34777_end_mask_0, x = var_34647_cast_fp16)[name = tensor("op_34777_cast_fp16")]; + tensor var_34784_begin_0 = const()[name = tensor("op_34784_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34784_end_0 = const()[name = tensor("op_34784_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34784_end_mask_0 = const()[name = tensor("op_34784_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34784_cast_fp16 = slice_by_index(begin = var_34784_begin_0, end = var_34784_end_0, end_mask = var_34784_end_mask_0, x = var_34651_cast_fp16)[name = tensor("op_34784_cast_fp16")]; + tensor var_34791_begin_0 = const()[name = tensor("op_34791_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34791_end_0 = const()[name = tensor("op_34791_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34791_end_mask_0 = const()[name = tensor("op_34791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34791_cast_fp16 = slice_by_index(begin = var_34791_begin_0, end = var_34791_end_0, end_mask = var_34791_end_mask_0, x = var_34651_cast_fp16)[name = tensor("op_34791_cast_fp16")]; + tensor var_34798_begin_0 = const()[name = tensor("op_34798_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34798_end_0 = const()[name = tensor("op_34798_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34798_end_mask_0 = const()[name = tensor("op_34798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34798_cast_fp16 = slice_by_index(begin = var_34798_begin_0, end = var_34798_end_0, end_mask = var_34798_end_mask_0, x = var_34651_cast_fp16)[name = tensor("op_34798_cast_fp16")]; + tensor var_34805_begin_0 = const()[name = tensor("op_34805_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34805_end_0 = const()[name = tensor("op_34805_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34805_end_mask_0 = const()[name = tensor("op_34805_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34805_cast_fp16 = slice_by_index(begin = var_34805_begin_0, end = var_34805_end_0, end_mask = var_34805_end_mask_0, x = var_34651_cast_fp16)[name = tensor("op_34805_cast_fp16")]; + tensor var_34812_begin_0 = const()[name = tensor("op_34812_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34812_end_0 = const()[name = tensor("op_34812_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34812_end_mask_0 = const()[name = tensor("op_34812_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34812_cast_fp16 = slice_by_index(begin = var_34812_begin_0, end = var_34812_end_0, end_mask = var_34812_end_mask_0, x = var_34655_cast_fp16)[name = tensor("op_34812_cast_fp16")]; + tensor var_34819_begin_0 = const()[name = tensor("op_34819_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34819_end_0 = const()[name = tensor("op_34819_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34819_end_mask_0 = const()[name = tensor("op_34819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34819_cast_fp16 = slice_by_index(begin = var_34819_begin_0, end = var_34819_end_0, end_mask = var_34819_end_mask_0, x = var_34655_cast_fp16)[name = tensor("op_34819_cast_fp16")]; + tensor var_34826_begin_0 = const()[name = tensor("op_34826_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34826_end_0 = const()[name = tensor("op_34826_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34826_end_mask_0 = const()[name = tensor("op_34826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34826_cast_fp16 = slice_by_index(begin = var_34826_begin_0, end = var_34826_end_0, end_mask = var_34826_end_mask_0, x = var_34655_cast_fp16)[name = tensor("op_34826_cast_fp16")]; + tensor var_34833_begin_0 = const()[name = tensor("op_34833_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34833_end_0 = const()[name = tensor("op_34833_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34833_end_mask_0 = const()[name = tensor("op_34833_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34833_cast_fp16 = slice_by_index(begin = var_34833_begin_0, end = var_34833_end_0, end_mask = var_34833_end_mask_0, x = var_34655_cast_fp16)[name = tensor("op_34833_cast_fp16")]; + tensor var_34840_begin_0 = const()[name = tensor("op_34840_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34840_end_0 = const()[name = tensor("op_34840_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34840_end_mask_0 = const()[name = tensor("op_34840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34840_cast_fp16 = slice_by_index(begin = var_34840_begin_0, end = var_34840_end_0, end_mask = var_34840_end_mask_0, x = var_34659_cast_fp16)[name = tensor("op_34840_cast_fp16")]; + tensor var_34847_begin_0 = const()[name = tensor("op_34847_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34847_end_0 = const()[name = tensor("op_34847_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34847_end_mask_0 = const()[name = tensor("op_34847_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34847_cast_fp16 = slice_by_index(begin = var_34847_begin_0, end = var_34847_end_0, end_mask = var_34847_end_mask_0, x = var_34659_cast_fp16)[name = tensor("op_34847_cast_fp16")]; + tensor var_34854_begin_0 = const()[name = tensor("op_34854_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34854_end_0 = const()[name = tensor("op_34854_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34854_end_mask_0 = const()[name = tensor("op_34854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34854_cast_fp16 = slice_by_index(begin = var_34854_begin_0, end = var_34854_end_0, end_mask = var_34854_end_mask_0, x = var_34659_cast_fp16)[name = tensor("op_34854_cast_fp16")]; + tensor var_34861_begin_0 = const()[name = tensor("op_34861_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34861_end_0 = const()[name = tensor("op_34861_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34861_end_mask_0 = const()[name = tensor("op_34861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34861_cast_fp16 = slice_by_index(begin = var_34861_begin_0, end = var_34861_end_0, end_mask = var_34861_end_mask_0, x = var_34659_cast_fp16)[name = tensor("op_34861_cast_fp16")]; + tensor var_34868_begin_0 = const()[name = tensor("op_34868_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34868_end_0 = const()[name = tensor("op_34868_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34868_end_mask_0 = const()[name = tensor("op_34868_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34868_cast_fp16 = slice_by_index(begin = var_34868_begin_0, end = var_34868_end_0, end_mask = var_34868_end_mask_0, x = var_34663_cast_fp16)[name = tensor("op_34868_cast_fp16")]; + tensor var_34875_begin_0 = const()[name = tensor("op_34875_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34875_end_0 = const()[name = tensor("op_34875_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34875_end_mask_0 = const()[name = tensor("op_34875_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34875_cast_fp16 = slice_by_index(begin = var_34875_begin_0, end = var_34875_end_0, end_mask = var_34875_end_mask_0, x = var_34663_cast_fp16)[name = tensor("op_34875_cast_fp16")]; + tensor var_34882_begin_0 = const()[name = tensor("op_34882_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34882_end_0 = const()[name = tensor("op_34882_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34882_end_mask_0 = const()[name = tensor("op_34882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34882_cast_fp16 = slice_by_index(begin = var_34882_begin_0, end = var_34882_end_0, end_mask = var_34882_end_mask_0, x = var_34663_cast_fp16)[name = tensor("op_34882_cast_fp16")]; + tensor var_34889_begin_0 = const()[name = tensor("op_34889_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34889_end_0 = const()[name = tensor("op_34889_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34889_end_mask_0 = const()[name = tensor("op_34889_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34889_cast_fp16 = slice_by_index(begin = var_34889_begin_0, end = var_34889_end_0, end_mask = var_34889_end_mask_0, x = var_34663_cast_fp16)[name = tensor("op_34889_cast_fp16")]; + tensor var_34896_begin_0 = const()[name = tensor("op_34896_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34896_end_0 = const()[name = tensor("op_34896_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34896_end_mask_0 = const()[name = tensor("op_34896_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34896_cast_fp16 = slice_by_index(begin = var_34896_begin_0, end = var_34896_end_0, end_mask = var_34896_end_mask_0, x = var_34667_cast_fp16)[name = tensor("op_34896_cast_fp16")]; + tensor var_34903_begin_0 = const()[name = tensor("op_34903_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34903_end_0 = const()[name = tensor("op_34903_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34903_end_mask_0 = const()[name = tensor("op_34903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34903_cast_fp16 = slice_by_index(begin = var_34903_begin_0, end = var_34903_end_0, end_mask = var_34903_end_mask_0, x = var_34667_cast_fp16)[name = tensor("op_34903_cast_fp16")]; + tensor var_34910_begin_0 = const()[name = tensor("op_34910_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34910_end_0 = const()[name = tensor("op_34910_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34910_end_mask_0 = const()[name = tensor("op_34910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34910_cast_fp16 = slice_by_index(begin = var_34910_begin_0, end = var_34910_end_0, end_mask = var_34910_end_mask_0, x = var_34667_cast_fp16)[name = tensor("op_34910_cast_fp16")]; + tensor var_34917_begin_0 = const()[name = tensor("op_34917_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34917_end_0 = const()[name = tensor("op_34917_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34917_end_mask_0 = const()[name = tensor("op_34917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34917_cast_fp16 = slice_by_index(begin = var_34917_begin_0, end = var_34917_end_0, end_mask = var_34917_end_mask_0, x = var_34667_cast_fp16)[name = tensor("op_34917_cast_fp16")]; + tensor var_34924_begin_0 = const()[name = tensor("op_34924_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34924_end_0 = const()[name = tensor("op_34924_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34924_end_mask_0 = const()[name = tensor("op_34924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34924_cast_fp16 = slice_by_index(begin = var_34924_begin_0, end = var_34924_end_0, end_mask = var_34924_end_mask_0, x = var_34671_cast_fp16)[name = tensor("op_34924_cast_fp16")]; + tensor var_34931_begin_0 = const()[name = tensor("op_34931_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34931_end_0 = const()[name = tensor("op_34931_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34931_end_mask_0 = const()[name = tensor("op_34931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34931_cast_fp16 = slice_by_index(begin = var_34931_begin_0, end = var_34931_end_0, end_mask = var_34931_end_mask_0, x = var_34671_cast_fp16)[name = tensor("op_34931_cast_fp16")]; + tensor var_34938_begin_0 = const()[name = tensor("op_34938_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34938_end_0 = const()[name = tensor("op_34938_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34938_end_mask_0 = const()[name = tensor("op_34938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34938_cast_fp16 = slice_by_index(begin = var_34938_begin_0, end = var_34938_end_0, end_mask = var_34938_end_mask_0, x = var_34671_cast_fp16)[name = tensor("op_34938_cast_fp16")]; + tensor var_34945_begin_0 = const()[name = tensor("op_34945_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34945_end_0 = const()[name = tensor("op_34945_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34945_end_mask_0 = const()[name = tensor("op_34945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34945_cast_fp16 = slice_by_index(begin = var_34945_begin_0, end = var_34945_end_0, end_mask = var_34945_end_mask_0, x = var_34671_cast_fp16)[name = tensor("op_34945_cast_fp16")]; + tensor var_34952_begin_0 = const()[name = tensor("op_34952_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34952_end_0 = const()[name = tensor("op_34952_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34952_end_mask_0 = const()[name = tensor("op_34952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34952_cast_fp16 = slice_by_index(begin = var_34952_begin_0, end = var_34952_end_0, end_mask = var_34952_end_mask_0, x = var_34675_cast_fp16)[name = tensor("op_34952_cast_fp16")]; + tensor var_34959_begin_0 = const()[name = tensor("op_34959_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34959_end_0 = const()[name = tensor("op_34959_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34959_end_mask_0 = const()[name = tensor("op_34959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34959_cast_fp16 = slice_by_index(begin = var_34959_begin_0, end = var_34959_end_0, end_mask = var_34959_end_mask_0, x = var_34675_cast_fp16)[name = tensor("op_34959_cast_fp16")]; + tensor var_34966_begin_0 = const()[name = tensor("op_34966_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34966_end_0 = const()[name = tensor("op_34966_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34966_end_mask_0 = const()[name = tensor("op_34966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34966_cast_fp16 = slice_by_index(begin = var_34966_begin_0, end = var_34966_end_0, end_mask = var_34966_end_mask_0, x = var_34675_cast_fp16)[name = tensor("op_34966_cast_fp16")]; + tensor var_34973_begin_0 = const()[name = tensor("op_34973_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_34973_end_0 = const()[name = tensor("op_34973_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_34973_end_mask_0 = const()[name = tensor("op_34973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34973_cast_fp16 = slice_by_index(begin = var_34973_begin_0, end = var_34973_end_0, end_mask = var_34973_end_mask_0, x = var_34675_cast_fp16)[name = tensor("op_34973_cast_fp16")]; + tensor var_34980_begin_0 = const()[name = tensor("op_34980_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_34980_end_0 = const()[name = tensor("op_34980_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_34980_end_mask_0 = const()[name = tensor("op_34980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34980_cast_fp16 = slice_by_index(begin = var_34980_begin_0, end = var_34980_end_0, end_mask = var_34980_end_mask_0, x = var_34679_cast_fp16)[name = tensor("op_34980_cast_fp16")]; + tensor var_34987_begin_0 = const()[name = tensor("op_34987_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_34987_end_0 = const()[name = tensor("op_34987_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_34987_end_mask_0 = const()[name = tensor("op_34987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34987_cast_fp16 = slice_by_index(begin = var_34987_begin_0, end = var_34987_end_0, end_mask = var_34987_end_mask_0, x = var_34679_cast_fp16)[name = tensor("op_34987_cast_fp16")]; + tensor var_34994_begin_0 = const()[name = tensor("op_34994_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_34994_end_0 = const()[name = tensor("op_34994_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_34994_end_mask_0 = const()[name = tensor("op_34994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_34994_cast_fp16 = slice_by_index(begin = var_34994_begin_0, end = var_34994_end_0, end_mask = var_34994_end_mask_0, x = var_34679_cast_fp16)[name = tensor("op_34994_cast_fp16")]; + tensor var_35001_begin_0 = const()[name = tensor("op_35001_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35001_end_0 = const()[name = tensor("op_35001_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35001_end_mask_0 = const()[name = tensor("op_35001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35001_cast_fp16 = slice_by_index(begin = var_35001_begin_0, end = var_35001_end_0, end_mask = var_35001_end_mask_0, x = var_34679_cast_fp16)[name = tensor("op_35001_cast_fp16")]; + tensor var_35008_begin_0 = const()[name = tensor("op_35008_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35008_end_0 = const()[name = tensor("op_35008_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35008_end_mask_0 = const()[name = tensor("op_35008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35008_cast_fp16 = slice_by_index(begin = var_35008_begin_0, end = var_35008_end_0, end_mask = var_35008_end_mask_0, x = var_34683_cast_fp16)[name = tensor("op_35008_cast_fp16")]; + tensor var_35015_begin_0 = const()[name = tensor("op_35015_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35015_end_0 = const()[name = tensor("op_35015_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35015_end_mask_0 = const()[name = tensor("op_35015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35015_cast_fp16 = slice_by_index(begin = var_35015_begin_0, end = var_35015_end_0, end_mask = var_35015_end_mask_0, x = var_34683_cast_fp16)[name = tensor("op_35015_cast_fp16")]; + tensor var_35022_begin_0 = const()[name = tensor("op_35022_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35022_end_0 = const()[name = tensor("op_35022_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35022_end_mask_0 = const()[name = tensor("op_35022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35022_cast_fp16 = slice_by_index(begin = var_35022_begin_0, end = var_35022_end_0, end_mask = var_35022_end_mask_0, x = var_34683_cast_fp16)[name = tensor("op_35022_cast_fp16")]; + tensor var_35029_begin_0 = const()[name = tensor("op_35029_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35029_end_0 = const()[name = tensor("op_35029_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35029_end_mask_0 = const()[name = tensor("op_35029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35029_cast_fp16 = slice_by_index(begin = var_35029_begin_0, end = var_35029_end_0, end_mask = var_35029_end_mask_0, x = var_34683_cast_fp16)[name = tensor("op_35029_cast_fp16")]; + tensor var_35036_begin_0 = const()[name = tensor("op_35036_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35036_end_0 = const()[name = tensor("op_35036_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35036_end_mask_0 = const()[name = tensor("op_35036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35036_cast_fp16 = slice_by_index(begin = var_35036_begin_0, end = var_35036_end_0, end_mask = var_35036_end_mask_0, x = var_34687_cast_fp16)[name = tensor("op_35036_cast_fp16")]; + tensor var_35043_begin_0 = const()[name = tensor("op_35043_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35043_end_0 = const()[name = tensor("op_35043_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35043_end_mask_0 = const()[name = tensor("op_35043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35043_cast_fp16 = slice_by_index(begin = var_35043_begin_0, end = var_35043_end_0, end_mask = var_35043_end_mask_0, x = var_34687_cast_fp16)[name = tensor("op_35043_cast_fp16")]; + tensor var_35050_begin_0 = const()[name = tensor("op_35050_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35050_end_0 = const()[name = tensor("op_35050_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35050_end_mask_0 = const()[name = tensor("op_35050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35050_cast_fp16 = slice_by_index(begin = var_35050_begin_0, end = var_35050_end_0, end_mask = var_35050_end_mask_0, x = var_34687_cast_fp16)[name = tensor("op_35050_cast_fp16")]; + tensor var_35057_begin_0 = const()[name = tensor("op_35057_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35057_end_0 = const()[name = tensor("op_35057_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35057_end_mask_0 = const()[name = tensor("op_35057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35057_cast_fp16 = slice_by_index(begin = var_35057_begin_0, end = var_35057_end_0, end_mask = var_35057_end_mask_0, x = var_34687_cast_fp16)[name = tensor("op_35057_cast_fp16")]; + tensor var_35064_begin_0 = const()[name = tensor("op_35064_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35064_end_0 = const()[name = tensor("op_35064_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35064_end_mask_0 = const()[name = tensor("op_35064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35064_cast_fp16 = slice_by_index(begin = var_35064_begin_0, end = var_35064_end_0, end_mask = var_35064_end_mask_0, x = var_34691_cast_fp16)[name = tensor("op_35064_cast_fp16")]; + tensor var_35071_begin_0 = const()[name = tensor("op_35071_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35071_end_0 = const()[name = tensor("op_35071_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35071_end_mask_0 = const()[name = tensor("op_35071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35071_cast_fp16 = slice_by_index(begin = var_35071_begin_0, end = var_35071_end_0, end_mask = var_35071_end_mask_0, x = var_34691_cast_fp16)[name = tensor("op_35071_cast_fp16")]; + tensor var_35078_begin_0 = const()[name = tensor("op_35078_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35078_end_0 = const()[name = tensor("op_35078_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35078_end_mask_0 = const()[name = tensor("op_35078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35078_cast_fp16 = slice_by_index(begin = var_35078_begin_0, end = var_35078_end_0, end_mask = var_35078_end_mask_0, x = var_34691_cast_fp16)[name = tensor("op_35078_cast_fp16")]; + tensor var_35085_begin_0 = const()[name = tensor("op_35085_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35085_end_0 = const()[name = tensor("op_35085_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35085_end_mask_0 = const()[name = tensor("op_35085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35085_cast_fp16 = slice_by_index(begin = var_35085_begin_0, end = var_35085_end_0, end_mask = var_35085_end_mask_0, x = var_34691_cast_fp16)[name = tensor("op_35085_cast_fp16")]; + tensor var_35092_begin_0 = const()[name = tensor("op_35092_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35092_end_0 = const()[name = tensor("op_35092_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35092_end_mask_0 = const()[name = tensor("op_35092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35092_cast_fp16 = slice_by_index(begin = var_35092_begin_0, end = var_35092_end_0, end_mask = var_35092_end_mask_0, x = var_34695_cast_fp16)[name = tensor("op_35092_cast_fp16")]; + tensor var_35099_begin_0 = const()[name = tensor("op_35099_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35099_end_0 = const()[name = tensor("op_35099_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35099_end_mask_0 = const()[name = tensor("op_35099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35099_cast_fp16 = slice_by_index(begin = var_35099_begin_0, end = var_35099_end_0, end_mask = var_35099_end_mask_0, x = var_34695_cast_fp16)[name = tensor("op_35099_cast_fp16")]; + tensor var_35106_begin_0 = const()[name = tensor("op_35106_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35106_end_0 = const()[name = tensor("op_35106_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35106_end_mask_0 = const()[name = tensor("op_35106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35106_cast_fp16 = slice_by_index(begin = var_35106_begin_0, end = var_35106_end_0, end_mask = var_35106_end_mask_0, x = var_34695_cast_fp16)[name = tensor("op_35106_cast_fp16")]; + tensor var_35113_begin_0 = const()[name = tensor("op_35113_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35113_end_0 = const()[name = tensor("op_35113_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35113_end_mask_0 = const()[name = tensor("op_35113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35113_cast_fp16 = slice_by_index(begin = var_35113_begin_0, end = var_35113_end_0, end_mask = var_35113_end_mask_0, x = var_34695_cast_fp16)[name = tensor("op_35113_cast_fp16")]; + tensor var_35120_begin_0 = const()[name = tensor("op_35120_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35120_end_0 = const()[name = tensor("op_35120_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35120_end_mask_0 = const()[name = tensor("op_35120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35120_cast_fp16 = slice_by_index(begin = var_35120_begin_0, end = var_35120_end_0, end_mask = var_35120_end_mask_0, x = var_34699_cast_fp16)[name = tensor("op_35120_cast_fp16")]; + tensor var_35127_begin_0 = const()[name = tensor("op_35127_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35127_end_0 = const()[name = tensor("op_35127_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35127_end_mask_0 = const()[name = tensor("op_35127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35127_cast_fp16 = slice_by_index(begin = var_35127_begin_0, end = var_35127_end_0, end_mask = var_35127_end_mask_0, x = var_34699_cast_fp16)[name = tensor("op_35127_cast_fp16")]; + tensor var_35134_begin_0 = const()[name = tensor("op_35134_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35134_end_0 = const()[name = tensor("op_35134_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35134_end_mask_0 = const()[name = tensor("op_35134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35134_cast_fp16 = slice_by_index(begin = var_35134_begin_0, end = var_35134_end_0, end_mask = var_35134_end_mask_0, x = var_34699_cast_fp16)[name = tensor("op_35134_cast_fp16")]; + tensor var_35141_begin_0 = const()[name = tensor("op_35141_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35141_end_0 = const()[name = tensor("op_35141_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35141_end_mask_0 = const()[name = tensor("op_35141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35141_cast_fp16 = slice_by_index(begin = var_35141_begin_0, end = var_35141_end_0, end_mask = var_35141_end_mask_0, x = var_34699_cast_fp16)[name = tensor("op_35141_cast_fp16")]; + tensor var_35148_begin_0 = const()[name = tensor("op_35148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35148_end_0 = const()[name = tensor("op_35148_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35148_end_mask_0 = const()[name = tensor("op_35148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35148_cast_fp16 = slice_by_index(begin = var_35148_begin_0, end = var_35148_end_0, end_mask = var_35148_end_mask_0, x = var_34703_cast_fp16)[name = tensor("op_35148_cast_fp16")]; + tensor var_35155_begin_0 = const()[name = tensor("op_35155_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35155_end_0 = const()[name = tensor("op_35155_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35155_end_mask_0 = const()[name = tensor("op_35155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35155_cast_fp16 = slice_by_index(begin = var_35155_begin_0, end = var_35155_end_0, end_mask = var_35155_end_mask_0, x = var_34703_cast_fp16)[name = tensor("op_35155_cast_fp16")]; + tensor var_35162_begin_0 = const()[name = tensor("op_35162_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35162_end_0 = const()[name = tensor("op_35162_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35162_end_mask_0 = const()[name = tensor("op_35162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35162_cast_fp16 = slice_by_index(begin = var_35162_begin_0, end = var_35162_end_0, end_mask = var_35162_end_mask_0, x = var_34703_cast_fp16)[name = tensor("op_35162_cast_fp16")]; + tensor var_35169_begin_0 = const()[name = tensor("op_35169_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35169_end_0 = const()[name = tensor("op_35169_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35169_end_mask_0 = const()[name = tensor("op_35169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35169_cast_fp16 = slice_by_index(begin = var_35169_begin_0, end = var_35169_end_0, end_mask = var_35169_end_mask_0, x = var_34703_cast_fp16)[name = tensor("op_35169_cast_fp16")]; + tensor var_35176_begin_0 = const()[name = tensor("op_35176_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35176_end_0 = const()[name = tensor("op_35176_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35176_end_mask_0 = const()[name = tensor("op_35176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35176_cast_fp16 = slice_by_index(begin = var_35176_begin_0, end = var_35176_end_0, end_mask = var_35176_end_mask_0, x = var_34707_cast_fp16)[name = tensor("op_35176_cast_fp16")]; + tensor var_35183_begin_0 = const()[name = tensor("op_35183_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35183_end_0 = const()[name = tensor("op_35183_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35183_end_mask_0 = const()[name = tensor("op_35183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35183_cast_fp16 = slice_by_index(begin = var_35183_begin_0, end = var_35183_end_0, end_mask = var_35183_end_mask_0, x = var_34707_cast_fp16)[name = tensor("op_35183_cast_fp16")]; + tensor var_35190_begin_0 = const()[name = tensor("op_35190_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35190_end_0 = const()[name = tensor("op_35190_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35190_end_mask_0 = const()[name = tensor("op_35190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35190_cast_fp16 = slice_by_index(begin = var_35190_begin_0, end = var_35190_end_0, end_mask = var_35190_end_mask_0, x = var_34707_cast_fp16)[name = tensor("op_35190_cast_fp16")]; + tensor var_35197_begin_0 = const()[name = tensor("op_35197_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35197_end_0 = const()[name = tensor("op_35197_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35197_end_mask_0 = const()[name = tensor("op_35197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35197_cast_fp16 = slice_by_index(begin = var_35197_begin_0, end = var_35197_end_0, end_mask = var_35197_end_mask_0, x = var_34707_cast_fp16)[name = tensor("op_35197_cast_fp16")]; + tensor var_35204_begin_0 = const()[name = tensor("op_35204_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35204_end_0 = const()[name = tensor("op_35204_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35204_end_mask_0 = const()[name = tensor("op_35204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35204_cast_fp16 = slice_by_index(begin = var_35204_begin_0, end = var_35204_end_0, end_mask = var_35204_end_mask_0, x = var_34711_cast_fp16)[name = tensor("op_35204_cast_fp16")]; + tensor var_35211_begin_0 = const()[name = tensor("op_35211_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35211_end_0 = const()[name = tensor("op_35211_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35211_end_mask_0 = const()[name = tensor("op_35211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35211_cast_fp16 = slice_by_index(begin = var_35211_begin_0, end = var_35211_end_0, end_mask = var_35211_end_mask_0, x = var_34711_cast_fp16)[name = tensor("op_35211_cast_fp16")]; + tensor var_35218_begin_0 = const()[name = tensor("op_35218_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35218_end_0 = const()[name = tensor("op_35218_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35218_end_mask_0 = const()[name = tensor("op_35218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35218_cast_fp16 = slice_by_index(begin = var_35218_begin_0, end = var_35218_end_0, end_mask = var_35218_end_mask_0, x = var_34711_cast_fp16)[name = tensor("op_35218_cast_fp16")]; + tensor var_35225_begin_0 = const()[name = tensor("op_35225_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35225_end_0 = const()[name = tensor("op_35225_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35225_end_mask_0 = const()[name = tensor("op_35225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35225_cast_fp16 = slice_by_index(begin = var_35225_begin_0, end = var_35225_end_0, end_mask = var_35225_end_mask_0, x = var_34711_cast_fp16)[name = tensor("op_35225_cast_fp16")]; + tensor var_35232_begin_0 = const()[name = tensor("op_35232_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35232_end_0 = const()[name = tensor("op_35232_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35232_end_mask_0 = const()[name = tensor("op_35232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35232_cast_fp16 = slice_by_index(begin = var_35232_begin_0, end = var_35232_end_0, end_mask = var_35232_end_mask_0, x = var_34715_cast_fp16)[name = tensor("op_35232_cast_fp16")]; + tensor var_35239_begin_0 = const()[name = tensor("op_35239_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35239_end_0 = const()[name = tensor("op_35239_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35239_end_mask_0 = const()[name = tensor("op_35239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35239_cast_fp16 = slice_by_index(begin = var_35239_begin_0, end = var_35239_end_0, end_mask = var_35239_end_mask_0, x = var_34715_cast_fp16)[name = tensor("op_35239_cast_fp16")]; + tensor var_35246_begin_0 = const()[name = tensor("op_35246_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35246_end_0 = const()[name = tensor("op_35246_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35246_end_mask_0 = const()[name = tensor("op_35246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35246_cast_fp16 = slice_by_index(begin = var_35246_begin_0, end = var_35246_end_0, end_mask = var_35246_end_mask_0, x = var_34715_cast_fp16)[name = tensor("op_35246_cast_fp16")]; + tensor var_35253_begin_0 = const()[name = tensor("op_35253_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35253_end_0 = const()[name = tensor("op_35253_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35253_end_mask_0 = const()[name = tensor("op_35253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35253_cast_fp16 = slice_by_index(begin = var_35253_begin_0, end = var_35253_end_0, end_mask = var_35253_end_mask_0, x = var_34715_cast_fp16)[name = tensor("op_35253_cast_fp16")]; + tensor var_35260_begin_0 = const()[name = tensor("op_35260_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35260_end_0 = const()[name = tensor("op_35260_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_35260_end_mask_0 = const()[name = tensor("op_35260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35260_cast_fp16 = slice_by_index(begin = var_35260_begin_0, end = var_35260_end_0, end_mask = var_35260_end_mask_0, x = var_34719_cast_fp16)[name = tensor("op_35260_cast_fp16")]; + tensor var_35267_begin_0 = const()[name = tensor("op_35267_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_35267_end_0 = const()[name = tensor("op_35267_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_35267_end_mask_0 = const()[name = tensor("op_35267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35267_cast_fp16 = slice_by_index(begin = var_35267_begin_0, end = var_35267_end_0, end_mask = var_35267_end_mask_0, x = var_34719_cast_fp16)[name = tensor("op_35267_cast_fp16")]; + tensor var_35274_begin_0 = const()[name = tensor("op_35274_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_35274_end_0 = const()[name = tensor("op_35274_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_35274_end_mask_0 = const()[name = tensor("op_35274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35274_cast_fp16 = slice_by_index(begin = var_35274_begin_0, end = var_35274_end_0, end_mask = var_35274_end_mask_0, x = var_34719_cast_fp16)[name = tensor("op_35274_cast_fp16")]; + tensor var_35281_begin_0 = const()[name = tensor("op_35281_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_35281_end_0 = const()[name = tensor("op_35281_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35281_end_mask_0 = const()[name = tensor("op_35281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35281_cast_fp16 = slice_by_index(begin = var_35281_begin_0, end = var_35281_end_0, end_mask = var_35281_end_mask_0, x = var_34719_cast_fp16)[name = tensor("op_35281_cast_fp16")]; + tensor k_43_perm_0 = const()[name = tensor("k_43_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_35286_begin_0 = const()[name = tensor("op_35286_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35286_end_0 = const()[name = tensor("op_35286_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_35286_end_mask_0 = const()[name = tensor("op_35286_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_10 = transpose(perm = k_43_perm_0, x = key_43_cast_fp16)[name = tensor("transpose_10")]; + tensor var_35286_cast_fp16 = slice_by_index(begin = var_35286_begin_0, end = var_35286_end_0, end_mask = var_35286_end_mask_0, x = transpose_10)[name = tensor("op_35286_cast_fp16")]; + tensor var_35290_begin_0 = const()[name = tensor("op_35290_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_35290_end_0 = const()[name = tensor("op_35290_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_35290_end_mask_0 = const()[name = tensor("op_35290_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35290_cast_fp16 = slice_by_index(begin = var_35290_begin_0, end = var_35290_end_0, end_mask = var_35290_end_mask_0, x = transpose_10)[name = tensor("op_35290_cast_fp16")]; + tensor var_35294_begin_0 = const()[name = tensor("op_35294_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_35294_end_0 = const()[name = tensor("op_35294_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_35294_end_mask_0 = const()[name = tensor("op_35294_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35294_cast_fp16 = slice_by_index(begin = var_35294_begin_0, end = var_35294_end_0, end_mask = var_35294_end_mask_0, x = transpose_10)[name = tensor("op_35294_cast_fp16")]; + tensor var_35298_begin_0 = const()[name = tensor("op_35298_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_35298_end_0 = const()[name = tensor("op_35298_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_35298_end_mask_0 = const()[name = tensor("op_35298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35298_cast_fp16 = slice_by_index(begin = var_35298_begin_0, end = var_35298_end_0, end_mask = var_35298_end_mask_0, x = transpose_10)[name = tensor("op_35298_cast_fp16")]; + tensor var_35302_begin_0 = const()[name = tensor("op_35302_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_35302_end_0 = const()[name = tensor("op_35302_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_35302_end_mask_0 = const()[name = tensor("op_35302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35302_cast_fp16 = slice_by_index(begin = var_35302_begin_0, end = var_35302_end_0, end_mask = var_35302_end_mask_0, x = transpose_10)[name = tensor("op_35302_cast_fp16")]; + tensor var_35306_begin_0 = const()[name = tensor("op_35306_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_35306_end_0 = const()[name = tensor("op_35306_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_35306_end_mask_0 = const()[name = tensor("op_35306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35306_cast_fp16 = slice_by_index(begin = var_35306_begin_0, end = var_35306_end_0, end_mask = var_35306_end_mask_0, x = transpose_10)[name = tensor("op_35306_cast_fp16")]; + tensor var_35310_begin_0 = const()[name = tensor("op_35310_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_35310_end_0 = const()[name = tensor("op_35310_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_35310_end_mask_0 = const()[name = tensor("op_35310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35310_cast_fp16 = slice_by_index(begin = var_35310_begin_0, end = var_35310_end_0, end_mask = var_35310_end_mask_0, x = transpose_10)[name = tensor("op_35310_cast_fp16")]; + tensor var_35314_begin_0 = const()[name = tensor("op_35314_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_35314_end_0 = const()[name = tensor("op_35314_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_35314_end_mask_0 = const()[name = tensor("op_35314_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35314_cast_fp16 = slice_by_index(begin = var_35314_begin_0, end = var_35314_end_0, end_mask = var_35314_end_mask_0, x = transpose_10)[name = tensor("op_35314_cast_fp16")]; + tensor var_35318_begin_0 = const()[name = tensor("op_35318_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_35318_end_0 = const()[name = tensor("op_35318_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_35318_end_mask_0 = const()[name = tensor("op_35318_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35318_cast_fp16 = slice_by_index(begin = var_35318_begin_0, end = var_35318_end_0, end_mask = var_35318_end_mask_0, x = transpose_10)[name = tensor("op_35318_cast_fp16")]; + tensor var_35322_begin_0 = const()[name = tensor("op_35322_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_35322_end_0 = const()[name = tensor("op_35322_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_35322_end_mask_0 = const()[name = tensor("op_35322_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35322_cast_fp16 = slice_by_index(begin = var_35322_begin_0, end = var_35322_end_0, end_mask = var_35322_end_mask_0, x = transpose_10)[name = tensor("op_35322_cast_fp16")]; + tensor var_35326_begin_0 = const()[name = tensor("op_35326_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_35326_end_0 = const()[name = tensor("op_35326_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_35326_end_mask_0 = const()[name = tensor("op_35326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35326_cast_fp16 = slice_by_index(begin = var_35326_begin_0, end = var_35326_end_0, end_mask = var_35326_end_mask_0, x = transpose_10)[name = tensor("op_35326_cast_fp16")]; + tensor var_35330_begin_0 = const()[name = tensor("op_35330_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_35330_end_0 = const()[name = tensor("op_35330_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_35330_end_mask_0 = const()[name = tensor("op_35330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35330_cast_fp16 = slice_by_index(begin = var_35330_begin_0, end = var_35330_end_0, end_mask = var_35330_end_mask_0, x = transpose_10)[name = tensor("op_35330_cast_fp16")]; + tensor var_35334_begin_0 = const()[name = tensor("op_35334_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_35334_end_0 = const()[name = tensor("op_35334_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_35334_end_mask_0 = const()[name = tensor("op_35334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35334_cast_fp16 = slice_by_index(begin = var_35334_begin_0, end = var_35334_end_0, end_mask = var_35334_end_mask_0, x = transpose_10)[name = tensor("op_35334_cast_fp16")]; + tensor var_35338_begin_0 = const()[name = tensor("op_35338_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_35338_end_0 = const()[name = tensor("op_35338_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_35338_end_mask_0 = const()[name = tensor("op_35338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35338_cast_fp16 = slice_by_index(begin = var_35338_begin_0, end = var_35338_end_0, end_mask = var_35338_end_mask_0, x = transpose_10)[name = tensor("op_35338_cast_fp16")]; + tensor var_35342_begin_0 = const()[name = tensor("op_35342_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_35342_end_0 = const()[name = tensor("op_35342_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_35342_end_mask_0 = const()[name = tensor("op_35342_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35342_cast_fp16 = slice_by_index(begin = var_35342_begin_0, end = var_35342_end_0, end_mask = var_35342_end_mask_0, x = transpose_10)[name = tensor("op_35342_cast_fp16")]; + tensor var_35346_begin_0 = const()[name = tensor("op_35346_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_35346_end_0 = const()[name = tensor("op_35346_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_35346_end_mask_0 = const()[name = tensor("op_35346_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35346_cast_fp16 = slice_by_index(begin = var_35346_begin_0, end = var_35346_end_0, end_mask = var_35346_end_mask_0, x = transpose_10)[name = tensor("op_35346_cast_fp16")]; + tensor var_35350_begin_0 = const()[name = tensor("op_35350_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_35350_end_0 = const()[name = tensor("op_35350_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_35350_end_mask_0 = const()[name = tensor("op_35350_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35350_cast_fp16 = slice_by_index(begin = var_35350_begin_0, end = var_35350_end_0, end_mask = var_35350_end_mask_0, x = transpose_10)[name = tensor("op_35350_cast_fp16")]; + tensor var_35354_begin_0 = const()[name = tensor("op_35354_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_35354_end_0 = const()[name = tensor("op_35354_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_35354_end_mask_0 = const()[name = tensor("op_35354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35354_cast_fp16 = slice_by_index(begin = var_35354_begin_0, end = var_35354_end_0, end_mask = var_35354_end_mask_0, x = transpose_10)[name = tensor("op_35354_cast_fp16")]; + tensor var_35358_begin_0 = const()[name = tensor("op_35358_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_35358_end_0 = const()[name = tensor("op_35358_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_35358_end_mask_0 = const()[name = tensor("op_35358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35358_cast_fp16 = slice_by_index(begin = var_35358_begin_0, end = var_35358_end_0, end_mask = var_35358_end_mask_0, x = transpose_10)[name = tensor("op_35358_cast_fp16")]; + tensor var_35362_begin_0 = const()[name = tensor("op_35362_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_35362_end_0 = const()[name = tensor("op_35362_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_35362_end_mask_0 = const()[name = tensor("op_35362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_35362_cast_fp16 = slice_by_index(begin = var_35362_begin_0, end = var_35362_end_0, end_mask = var_35362_end_mask_0, x = transpose_10)[name = tensor("op_35362_cast_fp16")]; + tensor var_35364_begin_0 = const()[name = tensor("op_35364_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_35364_end_0 = const()[name = tensor("op_35364_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_35364_end_mask_0 = const()[name = tensor("op_35364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35364_cast_fp16 = slice_by_index(begin = var_35364_begin_0, end = var_35364_end_0, end_mask = var_35364_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35364_cast_fp16")]; + tensor var_35368_begin_0 = const()[name = tensor("op_35368_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_35368_end_0 = const()[name = tensor("op_35368_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_35368_end_mask_0 = const()[name = tensor("op_35368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35368_cast_fp16 = slice_by_index(begin = var_35368_begin_0, end = var_35368_end_0, end_mask = var_35368_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35368_cast_fp16")]; + tensor var_35372_begin_0 = const()[name = tensor("op_35372_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_35372_end_0 = const()[name = tensor("op_35372_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_35372_end_mask_0 = const()[name = tensor("op_35372_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35372_cast_fp16 = slice_by_index(begin = var_35372_begin_0, end = var_35372_end_0, end_mask = var_35372_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35372_cast_fp16")]; + tensor var_35376_begin_0 = const()[name = tensor("op_35376_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_35376_end_0 = const()[name = tensor("op_35376_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_35376_end_mask_0 = const()[name = tensor("op_35376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35376_cast_fp16 = slice_by_index(begin = var_35376_begin_0, end = var_35376_end_0, end_mask = var_35376_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35376_cast_fp16")]; + tensor var_35380_begin_0 = const()[name = tensor("op_35380_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_35380_end_0 = const()[name = tensor("op_35380_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_35380_end_mask_0 = const()[name = tensor("op_35380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35380_cast_fp16 = slice_by_index(begin = var_35380_begin_0, end = var_35380_end_0, end_mask = var_35380_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35380_cast_fp16")]; + tensor var_35384_begin_0 = const()[name = tensor("op_35384_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_35384_end_0 = const()[name = tensor("op_35384_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_35384_end_mask_0 = const()[name = tensor("op_35384_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35384_cast_fp16 = slice_by_index(begin = var_35384_begin_0, end = var_35384_end_0, end_mask = var_35384_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35384_cast_fp16")]; + tensor var_35388_begin_0 = const()[name = tensor("op_35388_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_35388_end_0 = const()[name = tensor("op_35388_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_35388_end_mask_0 = const()[name = tensor("op_35388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35388_cast_fp16 = slice_by_index(begin = var_35388_begin_0, end = var_35388_end_0, end_mask = var_35388_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35388_cast_fp16")]; + tensor var_35392_begin_0 = const()[name = tensor("op_35392_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_35392_end_0 = const()[name = tensor("op_35392_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_35392_end_mask_0 = const()[name = tensor("op_35392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35392_cast_fp16 = slice_by_index(begin = var_35392_begin_0, end = var_35392_end_0, end_mask = var_35392_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35392_cast_fp16")]; + tensor var_35396_begin_0 = const()[name = tensor("op_35396_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_35396_end_0 = const()[name = tensor("op_35396_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_35396_end_mask_0 = const()[name = tensor("op_35396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35396_cast_fp16 = slice_by_index(begin = var_35396_begin_0, end = var_35396_end_0, end_mask = var_35396_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35396_cast_fp16")]; + tensor var_35400_begin_0 = const()[name = tensor("op_35400_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_35400_end_0 = const()[name = tensor("op_35400_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_35400_end_mask_0 = const()[name = tensor("op_35400_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35400_cast_fp16 = slice_by_index(begin = var_35400_begin_0, end = var_35400_end_0, end_mask = var_35400_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35400_cast_fp16")]; + tensor var_35404_begin_0 = const()[name = tensor("op_35404_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_35404_end_0 = const()[name = tensor("op_35404_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_35404_end_mask_0 = const()[name = tensor("op_35404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35404_cast_fp16 = slice_by_index(begin = var_35404_begin_0, end = var_35404_end_0, end_mask = var_35404_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35404_cast_fp16")]; + tensor var_35408_begin_0 = const()[name = tensor("op_35408_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_35408_end_0 = const()[name = tensor("op_35408_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_35408_end_mask_0 = const()[name = tensor("op_35408_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35408_cast_fp16 = slice_by_index(begin = var_35408_begin_0, end = var_35408_end_0, end_mask = var_35408_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35408_cast_fp16")]; + tensor var_35412_begin_0 = const()[name = tensor("op_35412_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_35412_end_0 = const()[name = tensor("op_35412_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_35412_end_mask_0 = const()[name = tensor("op_35412_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35412_cast_fp16 = slice_by_index(begin = var_35412_begin_0, end = var_35412_end_0, end_mask = var_35412_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35412_cast_fp16")]; + tensor var_35416_begin_0 = const()[name = tensor("op_35416_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_35416_end_0 = const()[name = tensor("op_35416_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_35416_end_mask_0 = const()[name = tensor("op_35416_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35416_cast_fp16 = slice_by_index(begin = var_35416_begin_0, end = var_35416_end_0, end_mask = var_35416_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35416_cast_fp16")]; + tensor var_35420_begin_0 = const()[name = tensor("op_35420_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_35420_end_0 = const()[name = tensor("op_35420_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_35420_end_mask_0 = const()[name = tensor("op_35420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35420_cast_fp16 = slice_by_index(begin = var_35420_begin_0, end = var_35420_end_0, end_mask = var_35420_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35420_cast_fp16")]; + tensor var_35424_begin_0 = const()[name = tensor("op_35424_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_35424_end_0 = const()[name = tensor("op_35424_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_35424_end_mask_0 = const()[name = tensor("op_35424_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35424_cast_fp16 = slice_by_index(begin = var_35424_begin_0, end = var_35424_end_0, end_mask = var_35424_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35424_cast_fp16")]; + tensor var_35428_begin_0 = const()[name = tensor("op_35428_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_35428_end_0 = const()[name = tensor("op_35428_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_35428_end_mask_0 = const()[name = tensor("op_35428_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35428_cast_fp16 = slice_by_index(begin = var_35428_begin_0, end = var_35428_end_0, end_mask = var_35428_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35428_cast_fp16")]; + tensor var_35432_begin_0 = const()[name = tensor("op_35432_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_35432_end_0 = const()[name = tensor("op_35432_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_35432_end_mask_0 = const()[name = tensor("op_35432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35432_cast_fp16 = slice_by_index(begin = var_35432_begin_0, end = var_35432_end_0, end_mask = var_35432_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35432_cast_fp16")]; + tensor var_35436_begin_0 = const()[name = tensor("op_35436_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_35436_end_0 = const()[name = tensor("op_35436_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_35436_end_mask_0 = const()[name = tensor("op_35436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35436_cast_fp16 = slice_by_index(begin = var_35436_begin_0, end = var_35436_end_0, end_mask = var_35436_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35436_cast_fp16")]; + tensor var_35440_begin_0 = const()[name = tensor("op_35440_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_35440_end_0 = const()[name = tensor("op_35440_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_35440_end_mask_0 = const()[name = tensor("op_35440_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_35440_cast_fp16 = slice_by_index(begin = var_35440_begin_0, end = var_35440_end_0, end_mask = var_35440_end_mask_0, x = value_43_cast_fp16)[name = tensor("op_35440_cast_fp16")]; + tensor var_35444_equation_0 = const()[name = tensor("op_35444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35444_cast_fp16 = einsum(equation = var_35444_equation_0, values = (var_35286_cast_fp16, var_34728_cast_fp16))[name = tensor("op_35444_cast_fp16")]; + tensor var_35445_to_fp16 = const()[name = tensor("op_35445_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3361_cast_fp16 = mul(x = var_35444_cast_fp16, y = var_35445_to_fp16)[name = tensor("aw_chunk_3361_cast_fp16")]; + tensor var_35448_equation_0 = const()[name = tensor("op_35448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35448_cast_fp16 = einsum(equation = var_35448_equation_0, values = (var_35286_cast_fp16, var_34735_cast_fp16))[name = tensor("op_35448_cast_fp16")]; + tensor var_35449_to_fp16 = const()[name = tensor("op_35449_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3363_cast_fp16 = mul(x = var_35448_cast_fp16, y = var_35449_to_fp16)[name = tensor("aw_chunk_3363_cast_fp16")]; + tensor var_35452_equation_0 = const()[name = tensor("op_35452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35452_cast_fp16 = einsum(equation = var_35452_equation_0, values = (var_35286_cast_fp16, var_34742_cast_fp16))[name = tensor("op_35452_cast_fp16")]; + tensor var_35453_to_fp16 = const()[name = tensor("op_35453_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3365_cast_fp16 = mul(x = var_35452_cast_fp16, y = var_35453_to_fp16)[name = tensor("aw_chunk_3365_cast_fp16")]; + tensor var_35456_equation_0 = const()[name = tensor("op_35456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35456_cast_fp16 = einsum(equation = var_35456_equation_0, values = (var_35286_cast_fp16, var_34749_cast_fp16))[name = tensor("op_35456_cast_fp16")]; + tensor var_35457_to_fp16 = const()[name = tensor("op_35457_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3367_cast_fp16 = mul(x = var_35456_cast_fp16, y = var_35457_to_fp16)[name = tensor("aw_chunk_3367_cast_fp16")]; + tensor var_35460_equation_0 = const()[name = tensor("op_35460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35460_cast_fp16 = einsum(equation = var_35460_equation_0, values = (var_35290_cast_fp16, var_34756_cast_fp16))[name = tensor("op_35460_cast_fp16")]; + tensor var_35461_to_fp16 = const()[name = tensor("op_35461_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3369_cast_fp16 = mul(x = var_35460_cast_fp16, y = var_35461_to_fp16)[name = tensor("aw_chunk_3369_cast_fp16")]; + tensor var_35464_equation_0 = const()[name = tensor("op_35464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35464_cast_fp16 = einsum(equation = var_35464_equation_0, values = (var_35290_cast_fp16, var_34763_cast_fp16))[name = tensor("op_35464_cast_fp16")]; + tensor var_35465_to_fp16 = const()[name = tensor("op_35465_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3371_cast_fp16 = mul(x = var_35464_cast_fp16, y = var_35465_to_fp16)[name = tensor("aw_chunk_3371_cast_fp16")]; + tensor var_35468_equation_0 = const()[name = tensor("op_35468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35468_cast_fp16 = einsum(equation = var_35468_equation_0, values = (var_35290_cast_fp16, var_34770_cast_fp16))[name = tensor("op_35468_cast_fp16")]; + tensor var_35469_to_fp16 = const()[name = tensor("op_35469_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3373_cast_fp16 = mul(x = var_35468_cast_fp16, y = var_35469_to_fp16)[name = tensor("aw_chunk_3373_cast_fp16")]; + tensor var_35472_equation_0 = const()[name = tensor("op_35472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35472_cast_fp16 = einsum(equation = var_35472_equation_0, values = (var_35290_cast_fp16, var_34777_cast_fp16))[name = tensor("op_35472_cast_fp16")]; + tensor var_35473_to_fp16 = const()[name = tensor("op_35473_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3375_cast_fp16 = mul(x = var_35472_cast_fp16, y = var_35473_to_fp16)[name = tensor("aw_chunk_3375_cast_fp16")]; + tensor var_35476_equation_0 = const()[name = tensor("op_35476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35476_cast_fp16 = einsum(equation = var_35476_equation_0, values = (var_35294_cast_fp16, var_34784_cast_fp16))[name = tensor("op_35476_cast_fp16")]; + tensor var_35477_to_fp16 = const()[name = tensor("op_35477_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3377_cast_fp16 = mul(x = var_35476_cast_fp16, y = var_35477_to_fp16)[name = tensor("aw_chunk_3377_cast_fp16")]; + tensor var_35480_equation_0 = const()[name = tensor("op_35480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35480_cast_fp16 = einsum(equation = var_35480_equation_0, values = (var_35294_cast_fp16, var_34791_cast_fp16))[name = tensor("op_35480_cast_fp16")]; + tensor var_35481_to_fp16 = const()[name = tensor("op_35481_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3379_cast_fp16 = mul(x = var_35480_cast_fp16, y = var_35481_to_fp16)[name = tensor("aw_chunk_3379_cast_fp16")]; + tensor var_35484_equation_0 = const()[name = tensor("op_35484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35484_cast_fp16 = einsum(equation = var_35484_equation_0, values = (var_35294_cast_fp16, var_34798_cast_fp16))[name = tensor("op_35484_cast_fp16")]; + tensor var_35485_to_fp16 = const()[name = tensor("op_35485_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3381_cast_fp16 = mul(x = var_35484_cast_fp16, y = var_35485_to_fp16)[name = tensor("aw_chunk_3381_cast_fp16")]; + tensor var_35488_equation_0 = const()[name = tensor("op_35488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35488_cast_fp16 = einsum(equation = var_35488_equation_0, values = (var_35294_cast_fp16, var_34805_cast_fp16))[name = tensor("op_35488_cast_fp16")]; + tensor var_35489_to_fp16 = const()[name = tensor("op_35489_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3383_cast_fp16 = mul(x = var_35488_cast_fp16, y = var_35489_to_fp16)[name = tensor("aw_chunk_3383_cast_fp16")]; + tensor var_35492_equation_0 = const()[name = tensor("op_35492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35492_cast_fp16 = einsum(equation = var_35492_equation_0, values = (var_35298_cast_fp16, var_34812_cast_fp16))[name = tensor("op_35492_cast_fp16")]; + tensor var_35493_to_fp16 = const()[name = tensor("op_35493_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3385_cast_fp16 = mul(x = var_35492_cast_fp16, y = var_35493_to_fp16)[name = tensor("aw_chunk_3385_cast_fp16")]; + tensor var_35496_equation_0 = const()[name = tensor("op_35496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35496_cast_fp16 = einsum(equation = var_35496_equation_0, values = (var_35298_cast_fp16, var_34819_cast_fp16))[name = tensor("op_35496_cast_fp16")]; + tensor var_35497_to_fp16 = const()[name = tensor("op_35497_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3387_cast_fp16 = mul(x = var_35496_cast_fp16, y = var_35497_to_fp16)[name = tensor("aw_chunk_3387_cast_fp16")]; + tensor var_35500_equation_0 = const()[name = tensor("op_35500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35500_cast_fp16 = einsum(equation = var_35500_equation_0, values = (var_35298_cast_fp16, var_34826_cast_fp16))[name = tensor("op_35500_cast_fp16")]; + tensor var_35501_to_fp16 = const()[name = tensor("op_35501_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3389_cast_fp16 = mul(x = var_35500_cast_fp16, y = var_35501_to_fp16)[name = tensor("aw_chunk_3389_cast_fp16")]; + tensor var_35504_equation_0 = const()[name = tensor("op_35504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35504_cast_fp16 = einsum(equation = var_35504_equation_0, values = (var_35298_cast_fp16, var_34833_cast_fp16))[name = tensor("op_35504_cast_fp16")]; + tensor var_35505_to_fp16 = const()[name = tensor("op_35505_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3391_cast_fp16 = mul(x = var_35504_cast_fp16, y = var_35505_to_fp16)[name = tensor("aw_chunk_3391_cast_fp16")]; + tensor var_35508_equation_0 = const()[name = tensor("op_35508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35508_cast_fp16 = einsum(equation = var_35508_equation_0, values = (var_35302_cast_fp16, var_34840_cast_fp16))[name = tensor("op_35508_cast_fp16")]; + tensor var_35509_to_fp16 = const()[name = tensor("op_35509_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3393_cast_fp16 = mul(x = var_35508_cast_fp16, y = var_35509_to_fp16)[name = tensor("aw_chunk_3393_cast_fp16")]; + tensor var_35512_equation_0 = const()[name = tensor("op_35512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35512_cast_fp16 = einsum(equation = var_35512_equation_0, values = (var_35302_cast_fp16, var_34847_cast_fp16))[name = tensor("op_35512_cast_fp16")]; + tensor var_35513_to_fp16 = const()[name = tensor("op_35513_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3395_cast_fp16 = mul(x = var_35512_cast_fp16, y = var_35513_to_fp16)[name = tensor("aw_chunk_3395_cast_fp16")]; + tensor var_35516_equation_0 = const()[name = tensor("op_35516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35516_cast_fp16 = einsum(equation = var_35516_equation_0, values = (var_35302_cast_fp16, var_34854_cast_fp16))[name = tensor("op_35516_cast_fp16")]; + tensor var_35517_to_fp16 = const()[name = tensor("op_35517_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3397_cast_fp16 = mul(x = var_35516_cast_fp16, y = var_35517_to_fp16)[name = tensor("aw_chunk_3397_cast_fp16")]; + tensor var_35520_equation_0 = const()[name = tensor("op_35520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35520_cast_fp16 = einsum(equation = var_35520_equation_0, values = (var_35302_cast_fp16, var_34861_cast_fp16))[name = tensor("op_35520_cast_fp16")]; + tensor var_35521_to_fp16 = const()[name = tensor("op_35521_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3399_cast_fp16 = mul(x = var_35520_cast_fp16, y = var_35521_to_fp16)[name = tensor("aw_chunk_3399_cast_fp16")]; + tensor var_35524_equation_0 = const()[name = tensor("op_35524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35524_cast_fp16 = einsum(equation = var_35524_equation_0, values = (var_35306_cast_fp16, var_34868_cast_fp16))[name = tensor("op_35524_cast_fp16")]; + tensor var_35525_to_fp16 = const()[name = tensor("op_35525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3401_cast_fp16 = mul(x = var_35524_cast_fp16, y = var_35525_to_fp16)[name = tensor("aw_chunk_3401_cast_fp16")]; + tensor var_35528_equation_0 = const()[name = tensor("op_35528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35528_cast_fp16 = einsum(equation = var_35528_equation_0, values = (var_35306_cast_fp16, var_34875_cast_fp16))[name = tensor("op_35528_cast_fp16")]; + tensor var_35529_to_fp16 = const()[name = tensor("op_35529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3403_cast_fp16 = mul(x = var_35528_cast_fp16, y = var_35529_to_fp16)[name = tensor("aw_chunk_3403_cast_fp16")]; + tensor var_35532_equation_0 = const()[name = tensor("op_35532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35532_cast_fp16 = einsum(equation = var_35532_equation_0, values = (var_35306_cast_fp16, var_34882_cast_fp16))[name = tensor("op_35532_cast_fp16")]; + tensor var_35533_to_fp16 = const()[name = tensor("op_35533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3405_cast_fp16 = mul(x = var_35532_cast_fp16, y = var_35533_to_fp16)[name = tensor("aw_chunk_3405_cast_fp16")]; + tensor var_35536_equation_0 = const()[name = tensor("op_35536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35536_cast_fp16 = einsum(equation = var_35536_equation_0, values = (var_35306_cast_fp16, var_34889_cast_fp16))[name = tensor("op_35536_cast_fp16")]; + tensor var_35537_to_fp16 = const()[name = tensor("op_35537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3407_cast_fp16 = mul(x = var_35536_cast_fp16, y = var_35537_to_fp16)[name = tensor("aw_chunk_3407_cast_fp16")]; + tensor var_35540_equation_0 = const()[name = tensor("op_35540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35540_cast_fp16 = einsum(equation = var_35540_equation_0, values = (var_35310_cast_fp16, var_34896_cast_fp16))[name = tensor("op_35540_cast_fp16")]; + tensor var_35541_to_fp16 = const()[name = tensor("op_35541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3409_cast_fp16 = mul(x = var_35540_cast_fp16, y = var_35541_to_fp16)[name = tensor("aw_chunk_3409_cast_fp16")]; + tensor var_35544_equation_0 = const()[name = tensor("op_35544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35544_cast_fp16 = einsum(equation = var_35544_equation_0, values = (var_35310_cast_fp16, var_34903_cast_fp16))[name = tensor("op_35544_cast_fp16")]; + tensor var_35545_to_fp16 = const()[name = tensor("op_35545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3411_cast_fp16 = mul(x = var_35544_cast_fp16, y = var_35545_to_fp16)[name = tensor("aw_chunk_3411_cast_fp16")]; + tensor var_35548_equation_0 = const()[name = tensor("op_35548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35548_cast_fp16 = einsum(equation = var_35548_equation_0, values = (var_35310_cast_fp16, var_34910_cast_fp16))[name = tensor("op_35548_cast_fp16")]; + tensor var_35549_to_fp16 = const()[name = tensor("op_35549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3413_cast_fp16 = mul(x = var_35548_cast_fp16, y = var_35549_to_fp16)[name = tensor("aw_chunk_3413_cast_fp16")]; + tensor var_35552_equation_0 = const()[name = tensor("op_35552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35552_cast_fp16 = einsum(equation = var_35552_equation_0, values = (var_35310_cast_fp16, var_34917_cast_fp16))[name = tensor("op_35552_cast_fp16")]; + tensor var_35553_to_fp16 = const()[name = tensor("op_35553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3415_cast_fp16 = mul(x = var_35552_cast_fp16, y = var_35553_to_fp16)[name = tensor("aw_chunk_3415_cast_fp16")]; + tensor var_35556_equation_0 = const()[name = tensor("op_35556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35556_cast_fp16 = einsum(equation = var_35556_equation_0, values = (var_35314_cast_fp16, var_34924_cast_fp16))[name = tensor("op_35556_cast_fp16")]; + tensor var_35557_to_fp16 = const()[name = tensor("op_35557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3417_cast_fp16 = mul(x = var_35556_cast_fp16, y = var_35557_to_fp16)[name = tensor("aw_chunk_3417_cast_fp16")]; + tensor var_35560_equation_0 = const()[name = tensor("op_35560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35560_cast_fp16 = einsum(equation = var_35560_equation_0, values = (var_35314_cast_fp16, var_34931_cast_fp16))[name = tensor("op_35560_cast_fp16")]; + tensor var_35561_to_fp16 = const()[name = tensor("op_35561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3419_cast_fp16 = mul(x = var_35560_cast_fp16, y = var_35561_to_fp16)[name = tensor("aw_chunk_3419_cast_fp16")]; + tensor var_35564_equation_0 = const()[name = tensor("op_35564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35564_cast_fp16 = einsum(equation = var_35564_equation_0, values = (var_35314_cast_fp16, var_34938_cast_fp16))[name = tensor("op_35564_cast_fp16")]; + tensor var_35565_to_fp16 = const()[name = tensor("op_35565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3421_cast_fp16 = mul(x = var_35564_cast_fp16, y = var_35565_to_fp16)[name = tensor("aw_chunk_3421_cast_fp16")]; + tensor var_35568_equation_0 = const()[name = tensor("op_35568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35568_cast_fp16 = einsum(equation = var_35568_equation_0, values = (var_35314_cast_fp16, var_34945_cast_fp16))[name = tensor("op_35568_cast_fp16")]; + tensor var_35569_to_fp16 = const()[name = tensor("op_35569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3423_cast_fp16 = mul(x = var_35568_cast_fp16, y = var_35569_to_fp16)[name = tensor("aw_chunk_3423_cast_fp16")]; + tensor var_35572_equation_0 = const()[name = tensor("op_35572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35572_cast_fp16 = einsum(equation = var_35572_equation_0, values = (var_35318_cast_fp16, var_34952_cast_fp16))[name = tensor("op_35572_cast_fp16")]; + tensor var_35573_to_fp16 = const()[name = tensor("op_35573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3425_cast_fp16 = mul(x = var_35572_cast_fp16, y = var_35573_to_fp16)[name = tensor("aw_chunk_3425_cast_fp16")]; + tensor var_35576_equation_0 = const()[name = tensor("op_35576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35576_cast_fp16 = einsum(equation = var_35576_equation_0, values = (var_35318_cast_fp16, var_34959_cast_fp16))[name = tensor("op_35576_cast_fp16")]; + tensor var_35577_to_fp16 = const()[name = tensor("op_35577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3427_cast_fp16 = mul(x = var_35576_cast_fp16, y = var_35577_to_fp16)[name = tensor("aw_chunk_3427_cast_fp16")]; + tensor var_35580_equation_0 = const()[name = tensor("op_35580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35580_cast_fp16 = einsum(equation = var_35580_equation_0, values = (var_35318_cast_fp16, var_34966_cast_fp16))[name = tensor("op_35580_cast_fp16")]; + tensor var_35581_to_fp16 = const()[name = tensor("op_35581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3429_cast_fp16 = mul(x = var_35580_cast_fp16, y = var_35581_to_fp16)[name = tensor("aw_chunk_3429_cast_fp16")]; + tensor var_35584_equation_0 = const()[name = tensor("op_35584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35584_cast_fp16 = einsum(equation = var_35584_equation_0, values = (var_35318_cast_fp16, var_34973_cast_fp16))[name = tensor("op_35584_cast_fp16")]; + tensor var_35585_to_fp16 = const()[name = tensor("op_35585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3431_cast_fp16 = mul(x = var_35584_cast_fp16, y = var_35585_to_fp16)[name = tensor("aw_chunk_3431_cast_fp16")]; + tensor var_35588_equation_0 = const()[name = tensor("op_35588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35588_cast_fp16 = einsum(equation = var_35588_equation_0, values = (var_35322_cast_fp16, var_34980_cast_fp16))[name = tensor("op_35588_cast_fp16")]; + tensor var_35589_to_fp16 = const()[name = tensor("op_35589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3433_cast_fp16 = mul(x = var_35588_cast_fp16, y = var_35589_to_fp16)[name = tensor("aw_chunk_3433_cast_fp16")]; + tensor var_35592_equation_0 = const()[name = tensor("op_35592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35592_cast_fp16 = einsum(equation = var_35592_equation_0, values = (var_35322_cast_fp16, var_34987_cast_fp16))[name = tensor("op_35592_cast_fp16")]; + tensor var_35593_to_fp16 = const()[name = tensor("op_35593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3435_cast_fp16 = mul(x = var_35592_cast_fp16, y = var_35593_to_fp16)[name = tensor("aw_chunk_3435_cast_fp16")]; + tensor var_35596_equation_0 = const()[name = tensor("op_35596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35596_cast_fp16 = einsum(equation = var_35596_equation_0, values = (var_35322_cast_fp16, var_34994_cast_fp16))[name = tensor("op_35596_cast_fp16")]; + tensor var_35597_to_fp16 = const()[name = tensor("op_35597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3437_cast_fp16 = mul(x = var_35596_cast_fp16, y = var_35597_to_fp16)[name = tensor("aw_chunk_3437_cast_fp16")]; + tensor var_35600_equation_0 = const()[name = tensor("op_35600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35600_cast_fp16 = einsum(equation = var_35600_equation_0, values = (var_35322_cast_fp16, var_35001_cast_fp16))[name = tensor("op_35600_cast_fp16")]; + tensor var_35601_to_fp16 = const()[name = tensor("op_35601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3439_cast_fp16 = mul(x = var_35600_cast_fp16, y = var_35601_to_fp16)[name = tensor("aw_chunk_3439_cast_fp16")]; + tensor var_35604_equation_0 = const()[name = tensor("op_35604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35604_cast_fp16 = einsum(equation = var_35604_equation_0, values = (var_35326_cast_fp16, var_35008_cast_fp16))[name = tensor("op_35604_cast_fp16")]; + tensor var_35605_to_fp16 = const()[name = tensor("op_35605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3441_cast_fp16 = mul(x = var_35604_cast_fp16, y = var_35605_to_fp16)[name = tensor("aw_chunk_3441_cast_fp16")]; + tensor var_35608_equation_0 = const()[name = tensor("op_35608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35608_cast_fp16 = einsum(equation = var_35608_equation_0, values = (var_35326_cast_fp16, var_35015_cast_fp16))[name = tensor("op_35608_cast_fp16")]; + tensor var_35609_to_fp16 = const()[name = tensor("op_35609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3443_cast_fp16 = mul(x = var_35608_cast_fp16, y = var_35609_to_fp16)[name = tensor("aw_chunk_3443_cast_fp16")]; + tensor var_35612_equation_0 = const()[name = tensor("op_35612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35612_cast_fp16 = einsum(equation = var_35612_equation_0, values = (var_35326_cast_fp16, var_35022_cast_fp16))[name = tensor("op_35612_cast_fp16")]; + tensor var_35613_to_fp16 = const()[name = tensor("op_35613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3445_cast_fp16 = mul(x = var_35612_cast_fp16, y = var_35613_to_fp16)[name = tensor("aw_chunk_3445_cast_fp16")]; + tensor var_35616_equation_0 = const()[name = tensor("op_35616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35616_cast_fp16 = einsum(equation = var_35616_equation_0, values = (var_35326_cast_fp16, var_35029_cast_fp16))[name = tensor("op_35616_cast_fp16")]; + tensor var_35617_to_fp16 = const()[name = tensor("op_35617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3447_cast_fp16 = mul(x = var_35616_cast_fp16, y = var_35617_to_fp16)[name = tensor("aw_chunk_3447_cast_fp16")]; + tensor var_35620_equation_0 = const()[name = tensor("op_35620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35620_cast_fp16 = einsum(equation = var_35620_equation_0, values = (var_35330_cast_fp16, var_35036_cast_fp16))[name = tensor("op_35620_cast_fp16")]; + tensor var_35621_to_fp16 = const()[name = tensor("op_35621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3449_cast_fp16 = mul(x = var_35620_cast_fp16, y = var_35621_to_fp16)[name = tensor("aw_chunk_3449_cast_fp16")]; + tensor var_35624_equation_0 = const()[name = tensor("op_35624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35624_cast_fp16 = einsum(equation = var_35624_equation_0, values = (var_35330_cast_fp16, var_35043_cast_fp16))[name = tensor("op_35624_cast_fp16")]; + tensor var_35625_to_fp16 = const()[name = tensor("op_35625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3451_cast_fp16 = mul(x = var_35624_cast_fp16, y = var_35625_to_fp16)[name = tensor("aw_chunk_3451_cast_fp16")]; + tensor var_35628_equation_0 = const()[name = tensor("op_35628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35628_cast_fp16 = einsum(equation = var_35628_equation_0, values = (var_35330_cast_fp16, var_35050_cast_fp16))[name = tensor("op_35628_cast_fp16")]; + tensor var_35629_to_fp16 = const()[name = tensor("op_35629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3453_cast_fp16 = mul(x = var_35628_cast_fp16, y = var_35629_to_fp16)[name = tensor("aw_chunk_3453_cast_fp16")]; + tensor var_35632_equation_0 = const()[name = tensor("op_35632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35632_cast_fp16 = einsum(equation = var_35632_equation_0, values = (var_35330_cast_fp16, var_35057_cast_fp16))[name = tensor("op_35632_cast_fp16")]; + tensor var_35633_to_fp16 = const()[name = tensor("op_35633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3455_cast_fp16 = mul(x = var_35632_cast_fp16, y = var_35633_to_fp16)[name = tensor("aw_chunk_3455_cast_fp16")]; + tensor var_35636_equation_0 = const()[name = tensor("op_35636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35636_cast_fp16 = einsum(equation = var_35636_equation_0, values = (var_35334_cast_fp16, var_35064_cast_fp16))[name = tensor("op_35636_cast_fp16")]; + tensor var_35637_to_fp16 = const()[name = tensor("op_35637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3457_cast_fp16 = mul(x = var_35636_cast_fp16, y = var_35637_to_fp16)[name = tensor("aw_chunk_3457_cast_fp16")]; + tensor var_35640_equation_0 = const()[name = tensor("op_35640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35640_cast_fp16 = einsum(equation = var_35640_equation_0, values = (var_35334_cast_fp16, var_35071_cast_fp16))[name = tensor("op_35640_cast_fp16")]; + tensor var_35641_to_fp16 = const()[name = tensor("op_35641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3459_cast_fp16 = mul(x = var_35640_cast_fp16, y = var_35641_to_fp16)[name = tensor("aw_chunk_3459_cast_fp16")]; + tensor var_35644_equation_0 = const()[name = tensor("op_35644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35644_cast_fp16 = einsum(equation = var_35644_equation_0, values = (var_35334_cast_fp16, var_35078_cast_fp16))[name = tensor("op_35644_cast_fp16")]; + tensor var_35645_to_fp16 = const()[name = tensor("op_35645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3461_cast_fp16 = mul(x = var_35644_cast_fp16, y = var_35645_to_fp16)[name = tensor("aw_chunk_3461_cast_fp16")]; + tensor var_35648_equation_0 = const()[name = tensor("op_35648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35648_cast_fp16 = einsum(equation = var_35648_equation_0, values = (var_35334_cast_fp16, var_35085_cast_fp16))[name = tensor("op_35648_cast_fp16")]; + tensor var_35649_to_fp16 = const()[name = tensor("op_35649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3463_cast_fp16 = mul(x = var_35648_cast_fp16, y = var_35649_to_fp16)[name = tensor("aw_chunk_3463_cast_fp16")]; + tensor var_35652_equation_0 = const()[name = tensor("op_35652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35652_cast_fp16 = einsum(equation = var_35652_equation_0, values = (var_35338_cast_fp16, var_35092_cast_fp16))[name = tensor("op_35652_cast_fp16")]; + tensor var_35653_to_fp16 = const()[name = tensor("op_35653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3465_cast_fp16 = mul(x = var_35652_cast_fp16, y = var_35653_to_fp16)[name = tensor("aw_chunk_3465_cast_fp16")]; + tensor var_35656_equation_0 = const()[name = tensor("op_35656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35656_cast_fp16 = einsum(equation = var_35656_equation_0, values = (var_35338_cast_fp16, var_35099_cast_fp16))[name = tensor("op_35656_cast_fp16")]; + tensor var_35657_to_fp16 = const()[name = tensor("op_35657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3467_cast_fp16 = mul(x = var_35656_cast_fp16, y = var_35657_to_fp16)[name = tensor("aw_chunk_3467_cast_fp16")]; + tensor var_35660_equation_0 = const()[name = tensor("op_35660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35660_cast_fp16 = einsum(equation = var_35660_equation_0, values = (var_35338_cast_fp16, var_35106_cast_fp16))[name = tensor("op_35660_cast_fp16")]; + tensor var_35661_to_fp16 = const()[name = tensor("op_35661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3469_cast_fp16 = mul(x = var_35660_cast_fp16, y = var_35661_to_fp16)[name = tensor("aw_chunk_3469_cast_fp16")]; + tensor var_35664_equation_0 = const()[name = tensor("op_35664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35664_cast_fp16 = einsum(equation = var_35664_equation_0, values = (var_35338_cast_fp16, var_35113_cast_fp16))[name = tensor("op_35664_cast_fp16")]; + tensor var_35665_to_fp16 = const()[name = tensor("op_35665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3471_cast_fp16 = mul(x = var_35664_cast_fp16, y = var_35665_to_fp16)[name = tensor("aw_chunk_3471_cast_fp16")]; + tensor var_35668_equation_0 = const()[name = tensor("op_35668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35668_cast_fp16 = einsum(equation = var_35668_equation_0, values = (var_35342_cast_fp16, var_35120_cast_fp16))[name = tensor("op_35668_cast_fp16")]; + tensor var_35669_to_fp16 = const()[name = tensor("op_35669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3473_cast_fp16 = mul(x = var_35668_cast_fp16, y = var_35669_to_fp16)[name = tensor("aw_chunk_3473_cast_fp16")]; + tensor var_35672_equation_0 = const()[name = tensor("op_35672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35672_cast_fp16 = einsum(equation = var_35672_equation_0, values = (var_35342_cast_fp16, var_35127_cast_fp16))[name = tensor("op_35672_cast_fp16")]; + tensor var_35673_to_fp16 = const()[name = tensor("op_35673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3475_cast_fp16 = mul(x = var_35672_cast_fp16, y = var_35673_to_fp16)[name = tensor("aw_chunk_3475_cast_fp16")]; + tensor var_35676_equation_0 = const()[name = tensor("op_35676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35676_cast_fp16 = einsum(equation = var_35676_equation_0, values = (var_35342_cast_fp16, var_35134_cast_fp16))[name = tensor("op_35676_cast_fp16")]; + tensor var_35677_to_fp16 = const()[name = tensor("op_35677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3477_cast_fp16 = mul(x = var_35676_cast_fp16, y = var_35677_to_fp16)[name = tensor("aw_chunk_3477_cast_fp16")]; + tensor var_35680_equation_0 = const()[name = tensor("op_35680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35680_cast_fp16 = einsum(equation = var_35680_equation_0, values = (var_35342_cast_fp16, var_35141_cast_fp16))[name = tensor("op_35680_cast_fp16")]; + tensor var_35681_to_fp16 = const()[name = tensor("op_35681_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3479_cast_fp16 = mul(x = var_35680_cast_fp16, y = var_35681_to_fp16)[name = tensor("aw_chunk_3479_cast_fp16")]; + tensor var_35684_equation_0 = const()[name = tensor("op_35684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35684_cast_fp16 = einsum(equation = var_35684_equation_0, values = (var_35346_cast_fp16, var_35148_cast_fp16))[name = tensor("op_35684_cast_fp16")]; + tensor var_35685_to_fp16 = const()[name = tensor("op_35685_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3481_cast_fp16 = mul(x = var_35684_cast_fp16, y = var_35685_to_fp16)[name = tensor("aw_chunk_3481_cast_fp16")]; + tensor var_35688_equation_0 = const()[name = tensor("op_35688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35688_cast_fp16 = einsum(equation = var_35688_equation_0, values = (var_35346_cast_fp16, var_35155_cast_fp16))[name = tensor("op_35688_cast_fp16")]; + tensor var_35689_to_fp16 = const()[name = tensor("op_35689_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3483_cast_fp16 = mul(x = var_35688_cast_fp16, y = var_35689_to_fp16)[name = tensor("aw_chunk_3483_cast_fp16")]; + tensor var_35692_equation_0 = const()[name = tensor("op_35692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35692_cast_fp16 = einsum(equation = var_35692_equation_0, values = (var_35346_cast_fp16, var_35162_cast_fp16))[name = tensor("op_35692_cast_fp16")]; + tensor var_35693_to_fp16 = const()[name = tensor("op_35693_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3485_cast_fp16 = mul(x = var_35692_cast_fp16, y = var_35693_to_fp16)[name = tensor("aw_chunk_3485_cast_fp16")]; + tensor var_35696_equation_0 = const()[name = tensor("op_35696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35696_cast_fp16 = einsum(equation = var_35696_equation_0, values = (var_35346_cast_fp16, var_35169_cast_fp16))[name = tensor("op_35696_cast_fp16")]; + tensor var_35697_to_fp16 = const()[name = tensor("op_35697_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3487_cast_fp16 = mul(x = var_35696_cast_fp16, y = var_35697_to_fp16)[name = tensor("aw_chunk_3487_cast_fp16")]; + tensor var_35700_equation_0 = const()[name = tensor("op_35700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35700_cast_fp16 = einsum(equation = var_35700_equation_0, values = (var_35350_cast_fp16, var_35176_cast_fp16))[name = tensor("op_35700_cast_fp16")]; + tensor var_35701_to_fp16 = const()[name = tensor("op_35701_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3489_cast_fp16 = mul(x = var_35700_cast_fp16, y = var_35701_to_fp16)[name = tensor("aw_chunk_3489_cast_fp16")]; + tensor var_35704_equation_0 = const()[name = tensor("op_35704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35704_cast_fp16 = einsum(equation = var_35704_equation_0, values = (var_35350_cast_fp16, var_35183_cast_fp16))[name = tensor("op_35704_cast_fp16")]; + tensor var_35705_to_fp16 = const()[name = tensor("op_35705_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3491_cast_fp16 = mul(x = var_35704_cast_fp16, y = var_35705_to_fp16)[name = tensor("aw_chunk_3491_cast_fp16")]; + tensor var_35708_equation_0 = const()[name = tensor("op_35708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35708_cast_fp16 = einsum(equation = var_35708_equation_0, values = (var_35350_cast_fp16, var_35190_cast_fp16))[name = tensor("op_35708_cast_fp16")]; + tensor var_35709_to_fp16 = const()[name = tensor("op_35709_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3493_cast_fp16 = mul(x = var_35708_cast_fp16, y = var_35709_to_fp16)[name = tensor("aw_chunk_3493_cast_fp16")]; + tensor var_35712_equation_0 = const()[name = tensor("op_35712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35712_cast_fp16 = einsum(equation = var_35712_equation_0, values = (var_35350_cast_fp16, var_35197_cast_fp16))[name = tensor("op_35712_cast_fp16")]; + tensor var_35713_to_fp16 = const()[name = tensor("op_35713_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3495_cast_fp16 = mul(x = var_35712_cast_fp16, y = var_35713_to_fp16)[name = tensor("aw_chunk_3495_cast_fp16")]; + tensor var_35716_equation_0 = const()[name = tensor("op_35716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35716_cast_fp16 = einsum(equation = var_35716_equation_0, values = (var_35354_cast_fp16, var_35204_cast_fp16))[name = tensor("op_35716_cast_fp16")]; + tensor var_35717_to_fp16 = const()[name = tensor("op_35717_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3497_cast_fp16 = mul(x = var_35716_cast_fp16, y = var_35717_to_fp16)[name = tensor("aw_chunk_3497_cast_fp16")]; + tensor var_35720_equation_0 = const()[name = tensor("op_35720_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35720_cast_fp16 = einsum(equation = var_35720_equation_0, values = (var_35354_cast_fp16, var_35211_cast_fp16))[name = tensor("op_35720_cast_fp16")]; + tensor var_35721_to_fp16 = const()[name = tensor("op_35721_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3499_cast_fp16 = mul(x = var_35720_cast_fp16, y = var_35721_to_fp16)[name = tensor("aw_chunk_3499_cast_fp16")]; + tensor var_35724_equation_0 = const()[name = tensor("op_35724_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35724_cast_fp16 = einsum(equation = var_35724_equation_0, values = (var_35354_cast_fp16, var_35218_cast_fp16))[name = tensor("op_35724_cast_fp16")]; + tensor var_35725_to_fp16 = const()[name = tensor("op_35725_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3501_cast_fp16 = mul(x = var_35724_cast_fp16, y = var_35725_to_fp16)[name = tensor("aw_chunk_3501_cast_fp16")]; + tensor var_35728_equation_0 = const()[name = tensor("op_35728_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35728_cast_fp16 = einsum(equation = var_35728_equation_0, values = (var_35354_cast_fp16, var_35225_cast_fp16))[name = tensor("op_35728_cast_fp16")]; + tensor var_35729_to_fp16 = const()[name = tensor("op_35729_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3503_cast_fp16 = mul(x = var_35728_cast_fp16, y = var_35729_to_fp16)[name = tensor("aw_chunk_3503_cast_fp16")]; + tensor var_35732_equation_0 = const()[name = tensor("op_35732_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35732_cast_fp16 = einsum(equation = var_35732_equation_0, values = (var_35358_cast_fp16, var_35232_cast_fp16))[name = tensor("op_35732_cast_fp16")]; + tensor var_35733_to_fp16 = const()[name = tensor("op_35733_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3505_cast_fp16 = mul(x = var_35732_cast_fp16, y = var_35733_to_fp16)[name = tensor("aw_chunk_3505_cast_fp16")]; + tensor var_35736_equation_0 = const()[name = tensor("op_35736_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35736_cast_fp16 = einsum(equation = var_35736_equation_0, values = (var_35358_cast_fp16, var_35239_cast_fp16))[name = tensor("op_35736_cast_fp16")]; + tensor var_35737_to_fp16 = const()[name = tensor("op_35737_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3507_cast_fp16 = mul(x = var_35736_cast_fp16, y = var_35737_to_fp16)[name = tensor("aw_chunk_3507_cast_fp16")]; + tensor var_35740_equation_0 = const()[name = tensor("op_35740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35740_cast_fp16 = einsum(equation = var_35740_equation_0, values = (var_35358_cast_fp16, var_35246_cast_fp16))[name = tensor("op_35740_cast_fp16")]; + tensor var_35741_to_fp16 = const()[name = tensor("op_35741_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3509_cast_fp16 = mul(x = var_35740_cast_fp16, y = var_35741_to_fp16)[name = tensor("aw_chunk_3509_cast_fp16")]; + tensor var_35744_equation_0 = const()[name = tensor("op_35744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35744_cast_fp16 = einsum(equation = var_35744_equation_0, values = (var_35358_cast_fp16, var_35253_cast_fp16))[name = tensor("op_35744_cast_fp16")]; + tensor var_35745_to_fp16 = const()[name = tensor("op_35745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3511_cast_fp16 = mul(x = var_35744_cast_fp16, y = var_35745_to_fp16)[name = tensor("aw_chunk_3511_cast_fp16")]; + tensor var_35748_equation_0 = const()[name = tensor("op_35748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35748_cast_fp16 = einsum(equation = var_35748_equation_0, values = (var_35362_cast_fp16, var_35260_cast_fp16))[name = tensor("op_35748_cast_fp16")]; + tensor var_35749_to_fp16 = const()[name = tensor("op_35749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3513_cast_fp16 = mul(x = var_35748_cast_fp16, y = var_35749_to_fp16)[name = tensor("aw_chunk_3513_cast_fp16")]; + tensor var_35752_equation_0 = const()[name = tensor("op_35752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35752_cast_fp16 = einsum(equation = var_35752_equation_0, values = (var_35362_cast_fp16, var_35267_cast_fp16))[name = tensor("op_35752_cast_fp16")]; + tensor var_35753_to_fp16 = const()[name = tensor("op_35753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3515_cast_fp16 = mul(x = var_35752_cast_fp16, y = var_35753_to_fp16)[name = tensor("aw_chunk_3515_cast_fp16")]; + tensor var_35756_equation_0 = const()[name = tensor("op_35756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35756_cast_fp16 = einsum(equation = var_35756_equation_0, values = (var_35362_cast_fp16, var_35274_cast_fp16))[name = tensor("op_35756_cast_fp16")]; + tensor var_35757_to_fp16 = const()[name = tensor("op_35757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3517_cast_fp16 = mul(x = var_35756_cast_fp16, y = var_35757_to_fp16)[name = tensor("aw_chunk_3517_cast_fp16")]; + tensor var_35760_equation_0 = const()[name = tensor("op_35760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_35760_cast_fp16 = einsum(equation = var_35760_equation_0, values = (var_35362_cast_fp16, var_35281_cast_fp16))[name = tensor("op_35760_cast_fp16")]; + tensor var_35761_to_fp16 = const()[name = tensor("op_35761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3519_cast_fp16 = mul(x = var_35760_cast_fp16, y = var_35761_to_fp16)[name = tensor("aw_chunk_3519_cast_fp16")]; + tensor var_35763_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3361_cast_fp16)[name = tensor("op_35763_cast_fp16")]; + tensor var_35764_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3363_cast_fp16)[name = tensor("op_35764_cast_fp16")]; + tensor var_35765_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3365_cast_fp16)[name = tensor("op_35765_cast_fp16")]; + tensor var_35766_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3367_cast_fp16)[name = tensor("op_35766_cast_fp16")]; + tensor var_35767_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3369_cast_fp16)[name = tensor("op_35767_cast_fp16")]; + tensor var_35768_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3371_cast_fp16)[name = tensor("op_35768_cast_fp16")]; + tensor var_35769_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3373_cast_fp16)[name = tensor("op_35769_cast_fp16")]; + tensor var_35770_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3375_cast_fp16)[name = tensor("op_35770_cast_fp16")]; + tensor var_35771_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3377_cast_fp16)[name = tensor("op_35771_cast_fp16")]; + tensor var_35772_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3379_cast_fp16)[name = tensor("op_35772_cast_fp16")]; + tensor var_35773_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3381_cast_fp16)[name = tensor("op_35773_cast_fp16")]; + tensor var_35774_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3383_cast_fp16)[name = tensor("op_35774_cast_fp16")]; + tensor var_35775_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3385_cast_fp16)[name = tensor("op_35775_cast_fp16")]; + tensor var_35776_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3387_cast_fp16)[name = tensor("op_35776_cast_fp16")]; + tensor var_35777_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3389_cast_fp16)[name = tensor("op_35777_cast_fp16")]; + tensor var_35778_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3391_cast_fp16)[name = tensor("op_35778_cast_fp16")]; + tensor var_35779_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3393_cast_fp16)[name = tensor("op_35779_cast_fp16")]; + tensor var_35780_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3395_cast_fp16)[name = tensor("op_35780_cast_fp16")]; + tensor var_35781_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3397_cast_fp16)[name = tensor("op_35781_cast_fp16")]; + tensor var_35782_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3399_cast_fp16)[name = tensor("op_35782_cast_fp16")]; + tensor var_35783_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3401_cast_fp16)[name = tensor("op_35783_cast_fp16")]; + tensor var_35784_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3403_cast_fp16)[name = tensor("op_35784_cast_fp16")]; + tensor var_35785_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3405_cast_fp16)[name = tensor("op_35785_cast_fp16")]; + tensor var_35786_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3407_cast_fp16)[name = tensor("op_35786_cast_fp16")]; + tensor var_35787_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3409_cast_fp16)[name = tensor("op_35787_cast_fp16")]; + tensor var_35788_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3411_cast_fp16)[name = tensor("op_35788_cast_fp16")]; + tensor var_35789_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3413_cast_fp16)[name = tensor("op_35789_cast_fp16")]; + tensor var_35790_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3415_cast_fp16)[name = tensor("op_35790_cast_fp16")]; + tensor var_35791_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3417_cast_fp16)[name = tensor("op_35791_cast_fp16")]; + tensor var_35792_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3419_cast_fp16)[name = tensor("op_35792_cast_fp16")]; + tensor var_35793_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3421_cast_fp16)[name = tensor("op_35793_cast_fp16")]; + tensor var_35794_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3423_cast_fp16)[name = tensor("op_35794_cast_fp16")]; + tensor var_35795_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3425_cast_fp16)[name = tensor("op_35795_cast_fp16")]; + tensor var_35796_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3427_cast_fp16)[name = tensor("op_35796_cast_fp16")]; + tensor var_35797_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3429_cast_fp16)[name = tensor("op_35797_cast_fp16")]; + tensor var_35798_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3431_cast_fp16)[name = tensor("op_35798_cast_fp16")]; + tensor var_35799_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3433_cast_fp16)[name = tensor("op_35799_cast_fp16")]; + tensor var_35800_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3435_cast_fp16)[name = tensor("op_35800_cast_fp16")]; + tensor var_35801_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3437_cast_fp16)[name = tensor("op_35801_cast_fp16")]; + tensor var_35802_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3439_cast_fp16)[name = tensor("op_35802_cast_fp16")]; + tensor var_35803_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3441_cast_fp16)[name = tensor("op_35803_cast_fp16")]; + tensor var_35804_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3443_cast_fp16)[name = tensor("op_35804_cast_fp16")]; + tensor var_35805_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3445_cast_fp16)[name = tensor("op_35805_cast_fp16")]; + tensor var_35806_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3447_cast_fp16)[name = tensor("op_35806_cast_fp16")]; + tensor var_35807_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3449_cast_fp16)[name = tensor("op_35807_cast_fp16")]; + tensor var_35808_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3451_cast_fp16)[name = tensor("op_35808_cast_fp16")]; + tensor var_35809_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3453_cast_fp16)[name = tensor("op_35809_cast_fp16")]; + tensor var_35810_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3455_cast_fp16)[name = tensor("op_35810_cast_fp16")]; + tensor var_35811_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3457_cast_fp16)[name = tensor("op_35811_cast_fp16")]; + tensor var_35812_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3459_cast_fp16)[name = tensor("op_35812_cast_fp16")]; + tensor var_35813_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3461_cast_fp16)[name = tensor("op_35813_cast_fp16")]; + tensor var_35814_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3463_cast_fp16)[name = tensor("op_35814_cast_fp16")]; + tensor var_35815_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3465_cast_fp16)[name = tensor("op_35815_cast_fp16")]; + tensor var_35816_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3467_cast_fp16)[name = tensor("op_35816_cast_fp16")]; + tensor var_35817_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3469_cast_fp16)[name = tensor("op_35817_cast_fp16")]; + tensor var_35818_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3471_cast_fp16)[name = tensor("op_35818_cast_fp16")]; + tensor var_35819_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3473_cast_fp16)[name = tensor("op_35819_cast_fp16")]; + tensor var_35820_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3475_cast_fp16)[name = tensor("op_35820_cast_fp16")]; + tensor var_35821_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3477_cast_fp16)[name = tensor("op_35821_cast_fp16")]; + tensor var_35822_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3479_cast_fp16)[name = tensor("op_35822_cast_fp16")]; + tensor var_35823_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3481_cast_fp16)[name = tensor("op_35823_cast_fp16")]; + tensor var_35824_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3483_cast_fp16)[name = tensor("op_35824_cast_fp16")]; + tensor var_35825_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3485_cast_fp16)[name = tensor("op_35825_cast_fp16")]; + tensor var_35826_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3487_cast_fp16)[name = tensor("op_35826_cast_fp16")]; + tensor var_35827_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3489_cast_fp16)[name = tensor("op_35827_cast_fp16")]; + tensor var_35828_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3491_cast_fp16)[name = tensor("op_35828_cast_fp16")]; + tensor var_35829_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3493_cast_fp16)[name = tensor("op_35829_cast_fp16")]; + tensor var_35830_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3495_cast_fp16)[name = tensor("op_35830_cast_fp16")]; + tensor var_35831_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3497_cast_fp16)[name = tensor("op_35831_cast_fp16")]; + tensor var_35832_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3499_cast_fp16)[name = tensor("op_35832_cast_fp16")]; + tensor var_35833_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3501_cast_fp16)[name = tensor("op_35833_cast_fp16")]; + tensor var_35834_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3503_cast_fp16)[name = tensor("op_35834_cast_fp16")]; + tensor var_35835_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3505_cast_fp16)[name = tensor("op_35835_cast_fp16")]; + tensor var_35836_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3507_cast_fp16)[name = tensor("op_35836_cast_fp16")]; + tensor var_35837_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3509_cast_fp16)[name = tensor("op_35837_cast_fp16")]; + tensor var_35838_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3511_cast_fp16)[name = tensor("op_35838_cast_fp16")]; + tensor var_35839_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3513_cast_fp16)[name = tensor("op_35839_cast_fp16")]; + tensor var_35840_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3515_cast_fp16)[name = tensor("op_35840_cast_fp16")]; + tensor var_35841_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3517_cast_fp16)[name = tensor("op_35841_cast_fp16")]; + tensor var_35842_cast_fp16 = softmax(axis = var_34534, x = aw_chunk_3519_cast_fp16)[name = tensor("op_35842_cast_fp16")]; + tensor var_35844_equation_0 = const()[name = tensor("op_35844_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35844_cast_fp16 = einsum(equation = var_35844_equation_0, values = (var_35364_cast_fp16, var_35763_cast_fp16))[name = tensor("op_35844_cast_fp16")]; + tensor var_35846_equation_0 = const()[name = tensor("op_35846_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35846_cast_fp16 = einsum(equation = var_35846_equation_0, values = (var_35364_cast_fp16, var_35764_cast_fp16))[name = tensor("op_35846_cast_fp16")]; + tensor var_35848_equation_0 = const()[name = tensor("op_35848_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35848_cast_fp16 = einsum(equation = var_35848_equation_0, values = (var_35364_cast_fp16, var_35765_cast_fp16))[name = tensor("op_35848_cast_fp16")]; + tensor var_35850_equation_0 = const()[name = tensor("op_35850_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35850_cast_fp16 = einsum(equation = var_35850_equation_0, values = (var_35364_cast_fp16, var_35766_cast_fp16))[name = tensor("op_35850_cast_fp16")]; + tensor var_35852_equation_0 = const()[name = tensor("op_35852_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35852_cast_fp16 = einsum(equation = var_35852_equation_0, values = (var_35368_cast_fp16, var_35767_cast_fp16))[name = tensor("op_35852_cast_fp16")]; + tensor var_35854_equation_0 = const()[name = tensor("op_35854_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35854_cast_fp16 = einsum(equation = var_35854_equation_0, values = (var_35368_cast_fp16, var_35768_cast_fp16))[name = tensor("op_35854_cast_fp16")]; + tensor var_35856_equation_0 = const()[name = tensor("op_35856_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35856_cast_fp16 = einsum(equation = var_35856_equation_0, values = (var_35368_cast_fp16, var_35769_cast_fp16))[name = tensor("op_35856_cast_fp16")]; + tensor var_35858_equation_0 = const()[name = tensor("op_35858_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35858_cast_fp16 = einsum(equation = var_35858_equation_0, values = (var_35368_cast_fp16, var_35770_cast_fp16))[name = tensor("op_35858_cast_fp16")]; + tensor var_35860_equation_0 = const()[name = tensor("op_35860_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35860_cast_fp16 = einsum(equation = var_35860_equation_0, values = (var_35372_cast_fp16, var_35771_cast_fp16))[name = tensor("op_35860_cast_fp16")]; + tensor var_35862_equation_0 = const()[name = tensor("op_35862_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35862_cast_fp16 = einsum(equation = var_35862_equation_0, values = (var_35372_cast_fp16, var_35772_cast_fp16))[name = tensor("op_35862_cast_fp16")]; + tensor var_35864_equation_0 = const()[name = tensor("op_35864_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35864_cast_fp16 = einsum(equation = var_35864_equation_0, values = (var_35372_cast_fp16, var_35773_cast_fp16))[name = tensor("op_35864_cast_fp16")]; + tensor var_35866_equation_0 = const()[name = tensor("op_35866_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35866_cast_fp16 = einsum(equation = var_35866_equation_0, values = (var_35372_cast_fp16, var_35774_cast_fp16))[name = tensor("op_35866_cast_fp16")]; + tensor var_35868_equation_0 = const()[name = tensor("op_35868_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35868_cast_fp16 = einsum(equation = var_35868_equation_0, values = (var_35376_cast_fp16, var_35775_cast_fp16))[name = tensor("op_35868_cast_fp16")]; + tensor var_35870_equation_0 = const()[name = tensor("op_35870_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35870_cast_fp16 = einsum(equation = var_35870_equation_0, values = (var_35376_cast_fp16, var_35776_cast_fp16))[name = tensor("op_35870_cast_fp16")]; + tensor var_35872_equation_0 = const()[name = tensor("op_35872_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35872_cast_fp16 = einsum(equation = var_35872_equation_0, values = (var_35376_cast_fp16, var_35777_cast_fp16))[name = tensor("op_35872_cast_fp16")]; + tensor var_35874_equation_0 = const()[name = tensor("op_35874_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35874_cast_fp16 = einsum(equation = var_35874_equation_0, values = (var_35376_cast_fp16, var_35778_cast_fp16))[name = tensor("op_35874_cast_fp16")]; + tensor var_35876_equation_0 = const()[name = tensor("op_35876_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35876_cast_fp16 = einsum(equation = var_35876_equation_0, values = (var_35380_cast_fp16, var_35779_cast_fp16))[name = tensor("op_35876_cast_fp16")]; + tensor var_35878_equation_0 = const()[name = tensor("op_35878_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35878_cast_fp16 = einsum(equation = var_35878_equation_0, values = (var_35380_cast_fp16, var_35780_cast_fp16))[name = tensor("op_35878_cast_fp16")]; + tensor var_35880_equation_0 = const()[name = tensor("op_35880_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35880_cast_fp16 = einsum(equation = var_35880_equation_0, values = (var_35380_cast_fp16, var_35781_cast_fp16))[name = tensor("op_35880_cast_fp16")]; + tensor var_35882_equation_0 = const()[name = tensor("op_35882_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35882_cast_fp16 = einsum(equation = var_35882_equation_0, values = (var_35380_cast_fp16, var_35782_cast_fp16))[name = tensor("op_35882_cast_fp16")]; + tensor var_35884_equation_0 = const()[name = tensor("op_35884_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35884_cast_fp16 = einsum(equation = var_35884_equation_0, values = (var_35384_cast_fp16, var_35783_cast_fp16))[name = tensor("op_35884_cast_fp16")]; + tensor var_35886_equation_0 = const()[name = tensor("op_35886_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35886_cast_fp16 = einsum(equation = var_35886_equation_0, values = (var_35384_cast_fp16, var_35784_cast_fp16))[name = tensor("op_35886_cast_fp16")]; + tensor var_35888_equation_0 = const()[name = tensor("op_35888_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35888_cast_fp16 = einsum(equation = var_35888_equation_0, values = (var_35384_cast_fp16, var_35785_cast_fp16))[name = tensor("op_35888_cast_fp16")]; + tensor var_35890_equation_0 = const()[name = tensor("op_35890_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35890_cast_fp16 = einsum(equation = var_35890_equation_0, values = (var_35384_cast_fp16, var_35786_cast_fp16))[name = tensor("op_35890_cast_fp16")]; + tensor var_35892_equation_0 = const()[name = tensor("op_35892_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35892_cast_fp16 = einsum(equation = var_35892_equation_0, values = (var_35388_cast_fp16, var_35787_cast_fp16))[name = tensor("op_35892_cast_fp16")]; + tensor var_35894_equation_0 = const()[name = tensor("op_35894_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35894_cast_fp16 = einsum(equation = var_35894_equation_0, values = (var_35388_cast_fp16, var_35788_cast_fp16))[name = tensor("op_35894_cast_fp16")]; + tensor var_35896_equation_0 = const()[name = tensor("op_35896_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35896_cast_fp16 = einsum(equation = var_35896_equation_0, values = (var_35388_cast_fp16, var_35789_cast_fp16))[name = tensor("op_35896_cast_fp16")]; + tensor var_35898_equation_0 = const()[name = tensor("op_35898_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35898_cast_fp16 = einsum(equation = var_35898_equation_0, values = (var_35388_cast_fp16, var_35790_cast_fp16))[name = tensor("op_35898_cast_fp16")]; + tensor var_35900_equation_0 = const()[name = tensor("op_35900_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35900_cast_fp16 = einsum(equation = var_35900_equation_0, values = (var_35392_cast_fp16, var_35791_cast_fp16))[name = tensor("op_35900_cast_fp16")]; + tensor var_35902_equation_0 = const()[name = tensor("op_35902_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35902_cast_fp16 = einsum(equation = var_35902_equation_0, values = (var_35392_cast_fp16, var_35792_cast_fp16))[name = tensor("op_35902_cast_fp16")]; + tensor var_35904_equation_0 = const()[name = tensor("op_35904_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35904_cast_fp16 = einsum(equation = var_35904_equation_0, values = (var_35392_cast_fp16, var_35793_cast_fp16))[name = tensor("op_35904_cast_fp16")]; + tensor var_35906_equation_0 = const()[name = tensor("op_35906_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35906_cast_fp16 = einsum(equation = var_35906_equation_0, values = (var_35392_cast_fp16, var_35794_cast_fp16))[name = tensor("op_35906_cast_fp16")]; + tensor var_35908_equation_0 = const()[name = tensor("op_35908_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35908_cast_fp16 = einsum(equation = var_35908_equation_0, values = (var_35396_cast_fp16, var_35795_cast_fp16))[name = tensor("op_35908_cast_fp16")]; + tensor var_35910_equation_0 = const()[name = tensor("op_35910_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35910_cast_fp16 = einsum(equation = var_35910_equation_0, values = (var_35396_cast_fp16, var_35796_cast_fp16))[name = tensor("op_35910_cast_fp16")]; + tensor var_35912_equation_0 = const()[name = tensor("op_35912_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35912_cast_fp16 = einsum(equation = var_35912_equation_0, values = (var_35396_cast_fp16, var_35797_cast_fp16))[name = tensor("op_35912_cast_fp16")]; + tensor var_35914_equation_0 = const()[name = tensor("op_35914_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35914_cast_fp16 = einsum(equation = var_35914_equation_0, values = (var_35396_cast_fp16, var_35798_cast_fp16))[name = tensor("op_35914_cast_fp16")]; + tensor var_35916_equation_0 = const()[name = tensor("op_35916_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35916_cast_fp16 = einsum(equation = var_35916_equation_0, values = (var_35400_cast_fp16, var_35799_cast_fp16))[name = tensor("op_35916_cast_fp16")]; + tensor var_35918_equation_0 = const()[name = tensor("op_35918_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35918_cast_fp16 = einsum(equation = var_35918_equation_0, values = (var_35400_cast_fp16, var_35800_cast_fp16))[name = tensor("op_35918_cast_fp16")]; + tensor var_35920_equation_0 = const()[name = tensor("op_35920_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35920_cast_fp16 = einsum(equation = var_35920_equation_0, values = (var_35400_cast_fp16, var_35801_cast_fp16))[name = tensor("op_35920_cast_fp16")]; + tensor var_35922_equation_0 = const()[name = tensor("op_35922_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35922_cast_fp16 = einsum(equation = var_35922_equation_0, values = (var_35400_cast_fp16, var_35802_cast_fp16))[name = tensor("op_35922_cast_fp16")]; + tensor var_35924_equation_0 = const()[name = tensor("op_35924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35924_cast_fp16 = einsum(equation = var_35924_equation_0, values = (var_35404_cast_fp16, var_35803_cast_fp16))[name = tensor("op_35924_cast_fp16")]; + tensor var_35926_equation_0 = const()[name = tensor("op_35926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35926_cast_fp16 = einsum(equation = var_35926_equation_0, values = (var_35404_cast_fp16, var_35804_cast_fp16))[name = tensor("op_35926_cast_fp16")]; + tensor var_35928_equation_0 = const()[name = tensor("op_35928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35928_cast_fp16 = einsum(equation = var_35928_equation_0, values = (var_35404_cast_fp16, var_35805_cast_fp16))[name = tensor("op_35928_cast_fp16")]; + tensor var_35930_equation_0 = const()[name = tensor("op_35930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35930_cast_fp16 = einsum(equation = var_35930_equation_0, values = (var_35404_cast_fp16, var_35806_cast_fp16))[name = tensor("op_35930_cast_fp16")]; + tensor var_35932_equation_0 = const()[name = tensor("op_35932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35932_cast_fp16 = einsum(equation = var_35932_equation_0, values = (var_35408_cast_fp16, var_35807_cast_fp16))[name = tensor("op_35932_cast_fp16")]; + tensor var_35934_equation_0 = const()[name = tensor("op_35934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35934_cast_fp16 = einsum(equation = var_35934_equation_0, values = (var_35408_cast_fp16, var_35808_cast_fp16))[name = tensor("op_35934_cast_fp16")]; + tensor var_35936_equation_0 = const()[name = tensor("op_35936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35936_cast_fp16 = einsum(equation = var_35936_equation_0, values = (var_35408_cast_fp16, var_35809_cast_fp16))[name = tensor("op_35936_cast_fp16")]; + tensor var_35938_equation_0 = const()[name = tensor("op_35938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35938_cast_fp16 = einsum(equation = var_35938_equation_0, values = (var_35408_cast_fp16, var_35810_cast_fp16))[name = tensor("op_35938_cast_fp16")]; + tensor var_35940_equation_0 = const()[name = tensor("op_35940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35940_cast_fp16 = einsum(equation = var_35940_equation_0, values = (var_35412_cast_fp16, var_35811_cast_fp16))[name = tensor("op_35940_cast_fp16")]; + tensor var_35942_equation_0 = const()[name = tensor("op_35942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35942_cast_fp16 = einsum(equation = var_35942_equation_0, values = (var_35412_cast_fp16, var_35812_cast_fp16))[name = tensor("op_35942_cast_fp16")]; + tensor var_35944_equation_0 = const()[name = tensor("op_35944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35944_cast_fp16 = einsum(equation = var_35944_equation_0, values = (var_35412_cast_fp16, var_35813_cast_fp16))[name = tensor("op_35944_cast_fp16")]; + tensor var_35946_equation_0 = const()[name = tensor("op_35946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35946_cast_fp16 = einsum(equation = var_35946_equation_0, values = (var_35412_cast_fp16, var_35814_cast_fp16))[name = tensor("op_35946_cast_fp16")]; + tensor var_35948_equation_0 = const()[name = tensor("op_35948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35948_cast_fp16 = einsum(equation = var_35948_equation_0, values = (var_35416_cast_fp16, var_35815_cast_fp16))[name = tensor("op_35948_cast_fp16")]; + tensor var_35950_equation_0 = const()[name = tensor("op_35950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35950_cast_fp16 = einsum(equation = var_35950_equation_0, values = (var_35416_cast_fp16, var_35816_cast_fp16))[name = tensor("op_35950_cast_fp16")]; + tensor var_35952_equation_0 = const()[name = tensor("op_35952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35952_cast_fp16 = einsum(equation = var_35952_equation_0, values = (var_35416_cast_fp16, var_35817_cast_fp16))[name = tensor("op_35952_cast_fp16")]; + tensor var_35954_equation_0 = const()[name = tensor("op_35954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35954_cast_fp16 = einsum(equation = var_35954_equation_0, values = (var_35416_cast_fp16, var_35818_cast_fp16))[name = tensor("op_35954_cast_fp16")]; + tensor var_35956_equation_0 = const()[name = tensor("op_35956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35956_cast_fp16 = einsum(equation = var_35956_equation_0, values = (var_35420_cast_fp16, var_35819_cast_fp16))[name = tensor("op_35956_cast_fp16")]; + tensor var_35958_equation_0 = const()[name = tensor("op_35958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35958_cast_fp16 = einsum(equation = var_35958_equation_0, values = (var_35420_cast_fp16, var_35820_cast_fp16))[name = tensor("op_35958_cast_fp16")]; + tensor var_35960_equation_0 = const()[name = tensor("op_35960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35960_cast_fp16 = einsum(equation = var_35960_equation_0, values = (var_35420_cast_fp16, var_35821_cast_fp16))[name = tensor("op_35960_cast_fp16")]; + tensor var_35962_equation_0 = const()[name = tensor("op_35962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35962_cast_fp16 = einsum(equation = var_35962_equation_0, values = (var_35420_cast_fp16, var_35822_cast_fp16))[name = tensor("op_35962_cast_fp16")]; + tensor var_35964_equation_0 = const()[name = tensor("op_35964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35964_cast_fp16 = einsum(equation = var_35964_equation_0, values = (var_35424_cast_fp16, var_35823_cast_fp16))[name = tensor("op_35964_cast_fp16")]; + tensor var_35966_equation_0 = const()[name = tensor("op_35966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35966_cast_fp16 = einsum(equation = var_35966_equation_0, values = (var_35424_cast_fp16, var_35824_cast_fp16))[name = tensor("op_35966_cast_fp16")]; + tensor var_35968_equation_0 = const()[name = tensor("op_35968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35968_cast_fp16 = einsum(equation = var_35968_equation_0, values = (var_35424_cast_fp16, var_35825_cast_fp16))[name = tensor("op_35968_cast_fp16")]; + tensor var_35970_equation_0 = const()[name = tensor("op_35970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35970_cast_fp16 = einsum(equation = var_35970_equation_0, values = (var_35424_cast_fp16, var_35826_cast_fp16))[name = tensor("op_35970_cast_fp16")]; + tensor var_35972_equation_0 = const()[name = tensor("op_35972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35972_cast_fp16 = einsum(equation = var_35972_equation_0, values = (var_35428_cast_fp16, var_35827_cast_fp16))[name = tensor("op_35972_cast_fp16")]; + tensor var_35974_equation_0 = const()[name = tensor("op_35974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35974_cast_fp16 = einsum(equation = var_35974_equation_0, values = (var_35428_cast_fp16, var_35828_cast_fp16))[name = tensor("op_35974_cast_fp16")]; + tensor var_35976_equation_0 = const()[name = tensor("op_35976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35976_cast_fp16 = einsum(equation = var_35976_equation_0, values = (var_35428_cast_fp16, var_35829_cast_fp16))[name = tensor("op_35976_cast_fp16")]; + tensor var_35978_equation_0 = const()[name = tensor("op_35978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35978_cast_fp16 = einsum(equation = var_35978_equation_0, values = (var_35428_cast_fp16, var_35830_cast_fp16))[name = tensor("op_35978_cast_fp16")]; + tensor var_35980_equation_0 = const()[name = tensor("op_35980_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35980_cast_fp16 = einsum(equation = var_35980_equation_0, values = (var_35432_cast_fp16, var_35831_cast_fp16))[name = tensor("op_35980_cast_fp16")]; + tensor var_35982_equation_0 = const()[name = tensor("op_35982_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35982_cast_fp16 = einsum(equation = var_35982_equation_0, values = (var_35432_cast_fp16, var_35832_cast_fp16))[name = tensor("op_35982_cast_fp16")]; + tensor var_35984_equation_0 = const()[name = tensor("op_35984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35984_cast_fp16 = einsum(equation = var_35984_equation_0, values = (var_35432_cast_fp16, var_35833_cast_fp16))[name = tensor("op_35984_cast_fp16")]; + tensor var_35986_equation_0 = const()[name = tensor("op_35986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35986_cast_fp16 = einsum(equation = var_35986_equation_0, values = (var_35432_cast_fp16, var_35834_cast_fp16))[name = tensor("op_35986_cast_fp16")]; + tensor var_35988_equation_0 = const()[name = tensor("op_35988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35988_cast_fp16 = einsum(equation = var_35988_equation_0, values = (var_35436_cast_fp16, var_35835_cast_fp16))[name = tensor("op_35988_cast_fp16")]; + tensor var_35990_equation_0 = const()[name = tensor("op_35990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35990_cast_fp16 = einsum(equation = var_35990_equation_0, values = (var_35436_cast_fp16, var_35836_cast_fp16))[name = tensor("op_35990_cast_fp16")]; + tensor var_35992_equation_0 = const()[name = tensor("op_35992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35992_cast_fp16 = einsum(equation = var_35992_equation_0, values = (var_35436_cast_fp16, var_35837_cast_fp16))[name = tensor("op_35992_cast_fp16")]; + tensor var_35994_equation_0 = const()[name = tensor("op_35994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35994_cast_fp16 = einsum(equation = var_35994_equation_0, values = (var_35436_cast_fp16, var_35838_cast_fp16))[name = tensor("op_35994_cast_fp16")]; + tensor var_35996_equation_0 = const()[name = tensor("op_35996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35996_cast_fp16 = einsum(equation = var_35996_equation_0, values = (var_35440_cast_fp16, var_35839_cast_fp16))[name = tensor("op_35996_cast_fp16")]; + tensor var_35998_equation_0 = const()[name = tensor("op_35998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_35998_cast_fp16 = einsum(equation = var_35998_equation_0, values = (var_35440_cast_fp16, var_35840_cast_fp16))[name = tensor("op_35998_cast_fp16")]; + tensor var_36000_equation_0 = const()[name = tensor("op_36000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36000_cast_fp16 = einsum(equation = var_36000_equation_0, values = (var_35440_cast_fp16, var_35841_cast_fp16))[name = tensor("op_36000_cast_fp16")]; + tensor var_36002_equation_0 = const()[name = tensor("op_36002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_36002_cast_fp16 = einsum(equation = var_36002_equation_0, values = (var_35440_cast_fp16, var_35842_cast_fp16))[name = tensor("op_36002_cast_fp16")]; + tensor var_36004_interleave_0 = const()[name = tensor("op_36004_interleave_0"), val = tensor(false)]; + tensor var_36004_cast_fp16 = concat(axis = var_34509, interleave = var_36004_interleave_0, values = (var_35844_cast_fp16, var_35846_cast_fp16, var_35848_cast_fp16, var_35850_cast_fp16))[name = tensor("op_36004_cast_fp16")]; + tensor var_36006_interleave_0 = const()[name = tensor("op_36006_interleave_0"), val = tensor(false)]; + tensor var_36006_cast_fp16 = concat(axis = var_34509, interleave = var_36006_interleave_0, values = (var_35852_cast_fp16, var_35854_cast_fp16, var_35856_cast_fp16, var_35858_cast_fp16))[name = tensor("op_36006_cast_fp16")]; + tensor var_36008_interleave_0 = const()[name = tensor("op_36008_interleave_0"), val = tensor(false)]; + tensor var_36008_cast_fp16 = concat(axis = var_34509, interleave = var_36008_interleave_0, values = (var_35860_cast_fp16, var_35862_cast_fp16, var_35864_cast_fp16, var_35866_cast_fp16))[name = tensor("op_36008_cast_fp16")]; + tensor var_36010_interleave_0 = const()[name = tensor("op_36010_interleave_0"), val = tensor(false)]; + tensor var_36010_cast_fp16 = concat(axis = var_34509, interleave = var_36010_interleave_0, values = (var_35868_cast_fp16, var_35870_cast_fp16, var_35872_cast_fp16, var_35874_cast_fp16))[name = tensor("op_36010_cast_fp16")]; + tensor var_36012_interleave_0 = const()[name = tensor("op_36012_interleave_0"), val = tensor(false)]; + tensor var_36012_cast_fp16 = concat(axis = var_34509, interleave = var_36012_interleave_0, values = (var_35876_cast_fp16, var_35878_cast_fp16, var_35880_cast_fp16, var_35882_cast_fp16))[name = tensor("op_36012_cast_fp16")]; + tensor var_36014_interleave_0 = const()[name = tensor("op_36014_interleave_0"), val = tensor(false)]; + tensor var_36014_cast_fp16 = concat(axis = var_34509, interleave = var_36014_interleave_0, values = (var_35884_cast_fp16, var_35886_cast_fp16, var_35888_cast_fp16, var_35890_cast_fp16))[name = tensor("op_36014_cast_fp16")]; + tensor var_36016_interleave_0 = const()[name = tensor("op_36016_interleave_0"), val = tensor(false)]; + tensor var_36016_cast_fp16 = concat(axis = var_34509, interleave = var_36016_interleave_0, values = (var_35892_cast_fp16, var_35894_cast_fp16, var_35896_cast_fp16, var_35898_cast_fp16))[name = tensor("op_36016_cast_fp16")]; + tensor var_36018_interleave_0 = const()[name = tensor("op_36018_interleave_0"), val = tensor(false)]; + tensor var_36018_cast_fp16 = concat(axis = var_34509, interleave = var_36018_interleave_0, values = (var_35900_cast_fp16, var_35902_cast_fp16, var_35904_cast_fp16, var_35906_cast_fp16))[name = tensor("op_36018_cast_fp16")]; + tensor var_36020_interleave_0 = const()[name = tensor("op_36020_interleave_0"), val = tensor(false)]; + tensor var_36020_cast_fp16 = concat(axis = var_34509, interleave = var_36020_interleave_0, values = (var_35908_cast_fp16, var_35910_cast_fp16, var_35912_cast_fp16, var_35914_cast_fp16))[name = tensor("op_36020_cast_fp16")]; + tensor var_36022_interleave_0 = const()[name = tensor("op_36022_interleave_0"), val = tensor(false)]; + tensor var_36022_cast_fp16 = concat(axis = var_34509, interleave = var_36022_interleave_0, values = (var_35916_cast_fp16, var_35918_cast_fp16, var_35920_cast_fp16, var_35922_cast_fp16))[name = tensor("op_36022_cast_fp16")]; + tensor var_36024_interleave_0 = const()[name = tensor("op_36024_interleave_0"), val = tensor(false)]; + tensor var_36024_cast_fp16 = concat(axis = var_34509, interleave = var_36024_interleave_0, values = (var_35924_cast_fp16, var_35926_cast_fp16, var_35928_cast_fp16, var_35930_cast_fp16))[name = tensor("op_36024_cast_fp16")]; + tensor var_36026_interleave_0 = const()[name = tensor("op_36026_interleave_0"), val = tensor(false)]; + tensor var_36026_cast_fp16 = concat(axis = var_34509, interleave = var_36026_interleave_0, values = (var_35932_cast_fp16, var_35934_cast_fp16, var_35936_cast_fp16, var_35938_cast_fp16))[name = tensor("op_36026_cast_fp16")]; + tensor var_36028_interleave_0 = const()[name = tensor("op_36028_interleave_0"), val = tensor(false)]; + tensor var_36028_cast_fp16 = concat(axis = var_34509, interleave = var_36028_interleave_0, values = (var_35940_cast_fp16, var_35942_cast_fp16, var_35944_cast_fp16, var_35946_cast_fp16))[name = tensor("op_36028_cast_fp16")]; + tensor var_36030_interleave_0 = const()[name = tensor("op_36030_interleave_0"), val = tensor(false)]; + tensor var_36030_cast_fp16 = concat(axis = var_34509, interleave = var_36030_interleave_0, values = (var_35948_cast_fp16, var_35950_cast_fp16, var_35952_cast_fp16, var_35954_cast_fp16))[name = tensor("op_36030_cast_fp16")]; + tensor var_36032_interleave_0 = const()[name = tensor("op_36032_interleave_0"), val = tensor(false)]; + tensor var_36032_cast_fp16 = concat(axis = var_34509, interleave = var_36032_interleave_0, values = (var_35956_cast_fp16, var_35958_cast_fp16, var_35960_cast_fp16, var_35962_cast_fp16))[name = tensor("op_36032_cast_fp16")]; + tensor var_36034_interleave_0 = const()[name = tensor("op_36034_interleave_0"), val = tensor(false)]; + tensor var_36034_cast_fp16 = concat(axis = var_34509, interleave = var_36034_interleave_0, values = (var_35964_cast_fp16, var_35966_cast_fp16, var_35968_cast_fp16, var_35970_cast_fp16))[name = tensor("op_36034_cast_fp16")]; + tensor var_36036_interleave_0 = const()[name = tensor("op_36036_interleave_0"), val = tensor(false)]; + tensor var_36036_cast_fp16 = concat(axis = var_34509, interleave = var_36036_interleave_0, values = (var_35972_cast_fp16, var_35974_cast_fp16, var_35976_cast_fp16, var_35978_cast_fp16))[name = tensor("op_36036_cast_fp16")]; + tensor var_36038_interleave_0 = const()[name = tensor("op_36038_interleave_0"), val = tensor(false)]; + tensor var_36038_cast_fp16 = concat(axis = var_34509, interleave = var_36038_interleave_0, values = (var_35980_cast_fp16, var_35982_cast_fp16, var_35984_cast_fp16, var_35986_cast_fp16))[name = tensor("op_36038_cast_fp16")]; + tensor var_36040_interleave_0 = const()[name = tensor("op_36040_interleave_0"), val = tensor(false)]; + tensor var_36040_cast_fp16 = concat(axis = var_34509, interleave = var_36040_interleave_0, values = (var_35988_cast_fp16, var_35990_cast_fp16, var_35992_cast_fp16, var_35994_cast_fp16))[name = tensor("op_36040_cast_fp16")]; + tensor var_36042_interleave_0 = const()[name = tensor("op_36042_interleave_0"), val = tensor(false)]; + tensor var_36042_cast_fp16 = concat(axis = var_34509, interleave = var_36042_interleave_0, values = (var_35996_cast_fp16, var_35998_cast_fp16, var_36000_cast_fp16, var_36002_cast_fp16))[name = tensor("op_36042_cast_fp16")]; + tensor input_427_interleave_0 = const()[name = tensor("input_427_interleave_0"), val = tensor(false)]; + tensor input_427_cast_fp16 = concat(axis = var_34534, interleave = input_427_interleave_0, values = (var_36004_cast_fp16, var_36006_cast_fp16, var_36008_cast_fp16, var_36010_cast_fp16, var_36012_cast_fp16, var_36014_cast_fp16, var_36016_cast_fp16, var_36018_cast_fp16, var_36020_cast_fp16, var_36022_cast_fp16, var_36024_cast_fp16, var_36026_cast_fp16, var_36028_cast_fp16, var_36030_cast_fp16, var_36032_cast_fp16, var_36034_cast_fp16, var_36036_cast_fp16, var_36038_cast_fp16, var_36040_cast_fp16, var_36042_cast_fp16))[name = tensor("input_427_cast_fp16")]; + tensor var_36050 = const()[name = tensor("op_36050"), val = tensor([1, 1])]; + tensor var_36052 = const()[name = tensor("op_36052"), val = tensor([1, 1])]; + tensor pretrained_out_259_pad_type_0 = const()[name = tensor("pretrained_out_259_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_259_pad_0 = const()[name = tensor("pretrained_out_259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239639488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240458752))), name = tensor("layers_21_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_21_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_21_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240458880)))]; + tensor pretrained_out_259_cast_fp16 = conv(bias = layers_21_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_36052, groups = var_34534, pad = pretrained_out_259_pad_0, pad_type = pretrained_out_259_pad_type_0, strides = var_36050, weight = layers_21_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = tensor("pretrained_out_259_cast_fp16")]; + tensor var_36056 = const()[name = tensor("op_36056"), val = tensor([1, 1])]; + tensor var_36058 = const()[name = tensor("op_36058"), val = tensor([1, 1])]; + tensor input_429_pad_type_0 = const()[name = tensor("input_429_pad_type_0"), val = tensor("custom")]; + tensor input_429_pad_0 = const()[name = tensor("input_429_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_21_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240461504)))]; + tensor input_429_cast_fp16 = conv(dilations = var_36058, groups = var_34534, pad = input_429_pad_0, pad_type = input_429_pad_type_0, strides = var_36056, weight = layers_21_self_attn_o_proj_loraA_weight_to_fp16, x = input_427_cast_fp16)[name = tensor("input_429_cast_fp16")]; + tensor var_36062 = const()[name = tensor("op_36062"), val = tensor([1, 1])]; + tensor var_36064 = const()[name = tensor("op_36064"), val = tensor([1, 1])]; + tensor lora_out_517_pad_type_0 = const()[name = tensor("lora_out_517_pad_type_0"), val = tensor("custom")]; + tensor lora_out_517_pad_0 = const()[name = tensor("lora_out_517_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_519_weight_0_to_fp16 = const()[name = tensor("lora_out_519_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240502528)))]; + tensor lora_out_519_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_36064, groups = var_34534, pad = lora_out_517_pad_0, pad_type = lora_out_517_pad_type_0, strides = var_36062, weight = lora_out_519_weight_0_to_fp16, x = input_429_cast_fp16)[name = tensor("lora_out_519_cast_fp16")]; + tensor obj_87_cast_fp16 = add(x = pretrained_out_259_cast_fp16, y = lora_out_519_cast_fp16)[name = tensor("obj_87_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = obj_87_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; + tensor var_36073 = const()[name = tensor("op_36073"), val = tensor([1])]; + tensor channels_mean_87_cast_fp16 = reduce_mean(axes = var_36073, keep_dims = var_34535, x = inputs_87_cast_fp16)[name = tensor("channels_mean_87_cast_fp16")]; + tensor zero_mean_87_cast_fp16 = sub(x = inputs_87_cast_fp16, y = channels_mean_87_cast_fp16)[name = tensor("zero_mean_87_cast_fp16")]; + tensor zero_mean_sq_87_cast_fp16 = mul(x = zero_mean_87_cast_fp16, y = zero_mean_87_cast_fp16)[name = tensor("zero_mean_sq_87_cast_fp16")]; + tensor var_36077 = const()[name = tensor("op_36077"), val = tensor([1])]; + tensor var_36078_cast_fp16 = reduce_mean(axes = var_36077, keep_dims = var_34535, x = zero_mean_sq_87_cast_fp16)[name = tensor("op_36078_cast_fp16")]; + tensor var_36079_to_fp16 = const()[name = tensor("op_36079_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_36080_cast_fp16 = add(x = var_36078_cast_fp16, y = var_36079_to_fp16)[name = tensor("op_36080_cast_fp16")]; + tensor denom_87_epsilon_0 = const()[name = tensor("denom_87_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_87_cast_fp16 = rsqrt(epsilon = denom_87_epsilon_0, x = var_36080_cast_fp16)[name = tensor("denom_87_cast_fp16")]; + tensor out_87_cast_fp16 = mul(x = zero_mean_87_cast_fp16, y = denom_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; + tensor input_431_gamma_0_to_fp16 = const()[name = tensor("input_431_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240543552)))]; + tensor input_431_beta_0_to_fp16 = const()[name = tensor("input_431_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240546176)))]; + tensor input_431_epsilon_0_to_fp16 = const()[name = tensor("input_431_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_431_cast_fp16 = batch_norm(beta = input_431_beta_0_to_fp16, epsilon = input_431_epsilon_0_to_fp16, gamma = input_431_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor("input_431_cast_fp16")]; + tensor var_36094 = const()[name = tensor("op_36094"), val = tensor([1, 1])]; + tensor var_36096 = const()[name = tensor("op_36096"), val = tensor([1, 1])]; + tensor pretrained_out_261_pad_type_0 = const()[name = tensor("pretrained_out_261_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_261_pad_0 = const()[name = tensor("pretrained_out_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240548800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243825664))), name = tensor("layers_21_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_21_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_21_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243825792)))]; + tensor pretrained_out_261_cast_fp16 = conv(bias = layers_21_fc1_pretrained_bias_to_fp16, dilations = var_36096, groups = var_34534, pad = pretrained_out_261_pad_0, pad_type = pretrained_out_261_pad_type_0, strides = var_36094, weight = layers_21_fc1_pretrained_weight_to_fp16_palettized, x = input_431_cast_fp16)[name = tensor("pretrained_out_261_cast_fp16")]; + tensor var_36100 = const()[name = tensor("op_36100"), val = tensor([1, 1])]; + tensor var_36102 = const()[name = tensor("op_36102"), val = tensor([1, 1])]; + tensor input_433_pad_type_0 = const()[name = tensor("input_433_pad_type_0"), val = tensor("custom")]; + tensor input_433_pad_0 = const()[name = tensor("input_433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_21_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243836096)))]; + tensor input_433_cast_fp16 = conv(dilations = var_36102, groups = var_34534, pad = input_433_pad_0, pad_type = input_433_pad_type_0, strides = var_36100, weight = layers_21_fc1_loraA_weight_to_fp16, x = input_431_cast_fp16)[name = tensor("input_433_cast_fp16")]; + tensor var_36106 = const()[name = tensor("op_36106"), val = tensor([1, 1])]; + tensor var_36108 = const()[name = tensor("op_36108"), val = tensor([1, 1])]; + tensor lora_out_521_pad_type_0 = const()[name = tensor("lora_out_521_pad_type_0"), val = tensor("custom")]; + tensor lora_out_521_pad_0 = const()[name = tensor("lora_out_521_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_523_weight_0_to_fp16 = const()[name = tensor("lora_out_523_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243877120)))]; + tensor lora_out_523_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_36108, groups = var_34534, pad = lora_out_521_pad_0, pad_type = lora_out_521_pad_type_0, strides = var_36106, weight = lora_out_523_weight_0_to_fp16, x = input_433_cast_fp16)[name = tensor("lora_out_523_cast_fp16")]; + tensor input_435_cast_fp16 = add(x = pretrained_out_261_cast_fp16, y = lora_out_523_cast_fp16)[name = tensor("input_435_cast_fp16")]; + tensor input_437_mode_0 = const()[name = tensor("input_437_mode_0"), val = tensor("EXACT")]; + tensor input_437_cast_fp16 = gelu(mode = input_437_mode_0, x = input_435_cast_fp16)[name = tensor("input_437_cast_fp16")]; + tensor var_36120 = const()[name = tensor("op_36120"), val = tensor([1, 1])]; + tensor var_36122 = const()[name = tensor("op_36122"), val = tensor([1, 1])]; + tensor pretrained_out_263_pad_type_0 = const()[name = tensor("pretrained_out_263_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_263_pad_0 = const()[name = tensor("pretrained_out_263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244041024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247317888))), name = tensor("layers_21_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_21_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_21_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247318016)))]; + tensor pretrained_out_263_cast_fp16 = conv(bias = layers_21_fc2_pretrained_bias_to_fp16, dilations = var_36122, groups = var_34534, pad = pretrained_out_263_pad_0, pad_type = pretrained_out_263_pad_type_0, strides = var_36120, weight = layers_21_fc2_pretrained_weight_to_fp16_palettized, x = input_437_cast_fp16)[name = tensor("pretrained_out_263_cast_fp16")]; + tensor var_36126 = const()[name = tensor("op_36126"), val = tensor([1, 1])]; + tensor var_36128 = const()[name = tensor("op_36128"), val = tensor([1, 1])]; + tensor input_439_pad_type_0 = const()[name = tensor("input_439_pad_type_0"), val = tensor("custom")]; + tensor input_439_pad_0 = const()[name = tensor("input_439_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_21_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_21_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247320640)))]; + tensor input_439_cast_fp16 = conv(dilations = var_36128, groups = var_34534, pad = input_439_pad_0, pad_type = input_439_pad_type_0, strides = var_36126, weight = layers_21_fc2_loraA_weight_to_fp16, x = input_437_cast_fp16)[name = tensor("input_439_cast_fp16")]; + tensor var_36132 = const()[name = tensor("op_36132"), val = tensor([1, 1])]; + tensor var_36134 = const()[name = tensor("op_36134"), val = tensor([1, 1])]; + tensor lora_out_525_pad_type_0 = const()[name = tensor("lora_out_525_pad_type_0"), val = tensor("custom")]; + tensor lora_out_525_pad_0 = const()[name = tensor("lora_out_525_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_527_weight_0_to_fp16 = const()[name = tensor("lora_out_527_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247484544)))]; + tensor lora_out_527_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_36134, groups = var_34534, pad = lora_out_525_pad_0, pad_type = lora_out_525_pad_type_0, strides = var_36132, weight = lora_out_527_weight_0_to_fp16, x = input_439_cast_fp16)[name = tensor("lora_out_527_cast_fp16")]; + tensor hidden_states_47_cast_fp16 = add(x = pretrained_out_263_cast_fp16, y = lora_out_527_cast_fp16)[name = tensor("hidden_states_47_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = hidden_states_47_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; + tensor var_36144 = const()[name = tensor("op_36144"), val = tensor(3)]; + tensor var_36169 = const()[name = tensor("op_36169"), val = tensor(1)]; + tensor var_36170 = const()[name = tensor("op_36170"), val = tensor(true)]; + tensor var_36180 = const()[name = tensor("op_36180"), val = tensor([1])]; + tensor channels_mean_89_cast_fp16 = reduce_mean(axes = var_36180, keep_dims = var_36170, x = inputs_89_cast_fp16)[name = tensor("channels_mean_89_cast_fp16")]; + tensor zero_mean_89_cast_fp16 = sub(x = inputs_89_cast_fp16, y = channels_mean_89_cast_fp16)[name = tensor("zero_mean_89_cast_fp16")]; + tensor zero_mean_sq_89_cast_fp16 = mul(x = zero_mean_89_cast_fp16, y = zero_mean_89_cast_fp16)[name = tensor("zero_mean_sq_89_cast_fp16")]; + tensor var_36184 = const()[name = tensor("op_36184"), val = tensor([1])]; + tensor var_36185_cast_fp16 = reduce_mean(axes = var_36184, keep_dims = var_36170, x = zero_mean_sq_89_cast_fp16)[name = tensor("op_36185_cast_fp16")]; + tensor var_36186_to_fp16 = const()[name = tensor("op_36186_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_36187_cast_fp16 = add(x = var_36185_cast_fp16, y = var_36186_to_fp16)[name = tensor("op_36187_cast_fp16")]; + tensor denom_89_epsilon_0 = const()[name = tensor("denom_89_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_89_cast_fp16 = rsqrt(epsilon = denom_89_epsilon_0, x = var_36187_cast_fp16)[name = tensor("denom_89_cast_fp16")]; + tensor out_89_cast_fp16 = mul(x = zero_mean_89_cast_fp16, y = denom_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; + tensor obj_89_gamma_0_to_fp16 = const()[name = tensor("obj_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247525568)))]; + tensor obj_89_beta_0_to_fp16 = const()[name = tensor("obj_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247528192)))]; + tensor obj_89_epsilon_0_to_fp16 = const()[name = tensor("obj_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_89_cast_fp16 = batch_norm(beta = obj_89_beta_0_to_fp16, epsilon = obj_89_epsilon_0_to_fp16, gamma = obj_89_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor("obj_89_cast_fp16")]; + tensor var_36205 = const()[name = tensor("op_36205"), val = tensor([1, 1])]; + tensor var_36207 = const()[name = tensor("op_36207"), val = tensor([1, 1])]; + tensor pretrained_out_265_pad_type_0 = const()[name = tensor("pretrained_out_265_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_265_pad_0 = const()[name = tensor("pretrained_out_265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247530816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248350080))), name = tensor("layers_22_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_22_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248350208)))]; + tensor pretrained_out_265_cast_fp16 = conv(bias = layers_22_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_36207, groups = var_36169, pad = pretrained_out_265_pad_0, pad_type = pretrained_out_265_pad_type_0, strides = var_36205, weight = layers_22_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_89_cast_fp16)[name = tensor("pretrained_out_265_cast_fp16")]; + tensor var_36211 = const()[name = tensor("op_36211"), val = tensor([1, 1])]; + tensor var_36213 = const()[name = tensor("op_36213"), val = tensor([1, 1])]; + tensor input_441_pad_type_0 = const()[name = tensor("input_441_pad_type_0"), val = tensor("custom")]; + tensor input_441_pad_0 = const()[name = tensor("input_441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248352832)))]; + tensor input_441_cast_fp16 = conv(dilations = var_36213, groups = var_36169, pad = input_441_pad_0, pad_type = input_441_pad_type_0, strides = var_36211, weight = layers_22_self_attn_q_proj_loraA_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor("input_441_cast_fp16")]; + tensor var_36217 = const()[name = tensor("op_36217"), val = tensor([1, 1])]; + tensor var_36219 = const()[name = tensor("op_36219"), val = tensor([1, 1])]; + tensor lora_out_529_pad_type_0 = const()[name = tensor("lora_out_529_pad_type_0"), val = tensor("custom")]; + tensor lora_out_529_pad_0 = const()[name = tensor("lora_out_529_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_531_weight_0_to_fp16 = const()[name = tensor("lora_out_531_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248393856)))]; + tensor lora_out_531_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_36219, groups = var_36169, pad = lora_out_529_pad_0, pad_type = lora_out_529_pad_type_0, strides = var_36217, weight = lora_out_531_weight_0_to_fp16, x = input_441_cast_fp16)[name = tensor("lora_out_531_cast_fp16")]; + tensor query_45_cast_fp16 = add(x = pretrained_out_265_cast_fp16, y = lora_out_531_cast_fp16)[name = tensor("query_45_cast_fp16")]; + tensor var_36229 = const()[name = tensor("op_36229"), val = tensor([1, 1])]; + tensor var_36231 = const()[name = tensor("op_36231"), val = tensor([1, 1])]; + tensor pretrained_out_267_pad_type_0 = const()[name = tensor("pretrained_out_267_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_267_pad_0 = const()[name = tensor("pretrained_out_267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(248434880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249254144))), name = tensor("layers_22_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_267_cast_fp16 = conv(dilations = var_36231, groups = var_36169, pad = pretrained_out_267_pad_0, pad_type = pretrained_out_267_pad_type_0, strides = var_36229, weight = layers_22_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_89_cast_fp16)[name = tensor("pretrained_out_267_cast_fp16")]; + tensor var_36235 = const()[name = tensor("op_36235"), val = tensor([1, 1])]; + tensor var_36237 = const()[name = tensor("op_36237"), val = tensor([1, 1])]; + tensor input_443_pad_type_0 = const()[name = tensor("input_443_pad_type_0"), val = tensor("custom")]; + tensor input_443_pad_0 = const()[name = tensor("input_443_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249254272)))]; + tensor input_443_cast_fp16 = conv(dilations = var_36237, groups = var_36169, pad = input_443_pad_0, pad_type = input_443_pad_type_0, strides = var_36235, weight = layers_22_self_attn_k_proj_loraA_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor("input_443_cast_fp16")]; + tensor var_36241 = const()[name = tensor("op_36241"), val = tensor([1, 1])]; + tensor var_36243 = const()[name = tensor("op_36243"), val = tensor([1, 1])]; + tensor lora_out_533_pad_type_0 = const()[name = tensor("lora_out_533_pad_type_0"), val = tensor("custom")]; + tensor lora_out_533_pad_0 = const()[name = tensor("lora_out_533_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_535_weight_0_to_fp16 = const()[name = tensor("lora_out_535_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249295296)))]; + tensor lora_out_535_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_36243, groups = var_36169, pad = lora_out_533_pad_0, pad_type = lora_out_533_pad_type_0, strides = var_36241, weight = lora_out_535_weight_0_to_fp16, x = input_443_cast_fp16)[name = tensor("lora_out_535_cast_fp16")]; + tensor key_45_cast_fp16 = add(x = pretrained_out_267_cast_fp16, y = lora_out_535_cast_fp16)[name = tensor("key_45_cast_fp16")]; + tensor var_36254 = const()[name = tensor("op_36254"), val = tensor([1, 1])]; + tensor var_36256 = const()[name = tensor("op_36256"), val = tensor([1, 1])]; + tensor pretrained_out_269_pad_type_0 = const()[name = tensor("pretrained_out_269_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_269_pad_0 = const()[name = tensor("pretrained_out_269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249336320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250155584))), name = tensor("layers_22_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_22_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250155712)))]; + tensor pretrained_out_269_cast_fp16 = conv(bias = layers_22_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_36256, groups = var_36169, pad = pretrained_out_269_pad_0, pad_type = pretrained_out_269_pad_type_0, strides = var_36254, weight = layers_22_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_89_cast_fp16)[name = tensor("pretrained_out_269_cast_fp16")]; + tensor var_36260 = const()[name = tensor("op_36260"), val = tensor([1, 1])]; + tensor var_36262 = const()[name = tensor("op_36262"), val = tensor([1, 1])]; + tensor input_445_pad_type_0 = const()[name = tensor("input_445_pad_type_0"), val = tensor("custom")]; + tensor input_445_pad_0 = const()[name = tensor("input_445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250158336)))]; + tensor input_445_cast_fp16 = conv(dilations = var_36262, groups = var_36169, pad = input_445_pad_0, pad_type = input_445_pad_type_0, strides = var_36260, weight = layers_22_self_attn_v_proj_loraA_weight_to_fp16, x = obj_89_cast_fp16)[name = tensor("input_445_cast_fp16")]; + tensor var_36266 = const()[name = tensor("op_36266"), val = tensor([1, 1])]; + tensor var_36268 = const()[name = tensor("op_36268"), val = tensor([1, 1])]; + tensor lora_out_537_pad_type_0 = const()[name = tensor("lora_out_537_pad_type_0"), val = tensor("custom")]; + tensor lora_out_537_pad_0 = const()[name = tensor("lora_out_537_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_539_weight_0_to_fp16 = const()[name = tensor("lora_out_539_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250199360)))]; + tensor lora_out_539_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_36268, groups = var_36169, pad = lora_out_537_pad_0, pad_type = lora_out_537_pad_type_0, strides = var_36266, weight = lora_out_539_weight_0_to_fp16, x = input_445_cast_fp16)[name = tensor("lora_out_539_cast_fp16")]; + tensor value_45_cast_fp16 = add(x = pretrained_out_269_cast_fp16, y = lora_out_539_cast_fp16)[name = tensor("value_45_cast_fp16")]; + tensor var_36278_begin_0 = const()[name = tensor("op_36278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36278_end_0 = const()[name = tensor("op_36278_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36278_end_mask_0 = const()[name = tensor("op_36278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36278_cast_fp16 = slice_by_index(begin = var_36278_begin_0, end = var_36278_end_0, end_mask = var_36278_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36278_cast_fp16")]; + tensor var_36282_begin_0 = const()[name = tensor("op_36282_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_36282_end_0 = const()[name = tensor("op_36282_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_36282_end_mask_0 = const()[name = tensor("op_36282_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36282_cast_fp16 = slice_by_index(begin = var_36282_begin_0, end = var_36282_end_0, end_mask = var_36282_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36282_cast_fp16")]; + tensor var_36286_begin_0 = const()[name = tensor("op_36286_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_36286_end_0 = const()[name = tensor("op_36286_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_36286_end_mask_0 = const()[name = tensor("op_36286_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36286_cast_fp16 = slice_by_index(begin = var_36286_begin_0, end = var_36286_end_0, end_mask = var_36286_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36286_cast_fp16")]; + tensor var_36290_begin_0 = const()[name = tensor("op_36290_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_36290_end_0 = const()[name = tensor("op_36290_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_36290_end_mask_0 = const()[name = tensor("op_36290_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36290_cast_fp16 = slice_by_index(begin = var_36290_begin_0, end = var_36290_end_0, end_mask = var_36290_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36290_cast_fp16")]; + tensor var_36294_begin_0 = const()[name = tensor("op_36294_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_36294_end_0 = const()[name = tensor("op_36294_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_36294_end_mask_0 = const()[name = tensor("op_36294_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36294_cast_fp16 = slice_by_index(begin = var_36294_begin_0, end = var_36294_end_0, end_mask = var_36294_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36294_cast_fp16")]; + tensor var_36298_begin_0 = const()[name = tensor("op_36298_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_36298_end_0 = const()[name = tensor("op_36298_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_36298_end_mask_0 = const()[name = tensor("op_36298_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36298_cast_fp16 = slice_by_index(begin = var_36298_begin_0, end = var_36298_end_0, end_mask = var_36298_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36298_cast_fp16")]; + tensor var_36302_begin_0 = const()[name = tensor("op_36302_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_36302_end_0 = const()[name = tensor("op_36302_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_36302_end_mask_0 = const()[name = tensor("op_36302_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36302_cast_fp16 = slice_by_index(begin = var_36302_begin_0, end = var_36302_end_0, end_mask = var_36302_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36302_cast_fp16")]; + tensor var_36306_begin_0 = const()[name = tensor("op_36306_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_36306_end_0 = const()[name = tensor("op_36306_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_36306_end_mask_0 = const()[name = tensor("op_36306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36306_cast_fp16 = slice_by_index(begin = var_36306_begin_0, end = var_36306_end_0, end_mask = var_36306_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36306_cast_fp16")]; + tensor var_36310_begin_0 = const()[name = tensor("op_36310_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_36310_end_0 = const()[name = tensor("op_36310_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_36310_end_mask_0 = const()[name = tensor("op_36310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36310_cast_fp16 = slice_by_index(begin = var_36310_begin_0, end = var_36310_end_0, end_mask = var_36310_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36310_cast_fp16")]; + tensor var_36314_begin_0 = const()[name = tensor("op_36314_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_36314_end_0 = const()[name = tensor("op_36314_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_36314_end_mask_0 = const()[name = tensor("op_36314_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36314_cast_fp16 = slice_by_index(begin = var_36314_begin_0, end = var_36314_end_0, end_mask = var_36314_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36314_cast_fp16")]; + tensor var_36318_begin_0 = const()[name = tensor("op_36318_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_36318_end_0 = const()[name = tensor("op_36318_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_36318_end_mask_0 = const()[name = tensor("op_36318_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36318_cast_fp16 = slice_by_index(begin = var_36318_begin_0, end = var_36318_end_0, end_mask = var_36318_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36318_cast_fp16")]; + tensor var_36322_begin_0 = const()[name = tensor("op_36322_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_36322_end_0 = const()[name = tensor("op_36322_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_36322_end_mask_0 = const()[name = tensor("op_36322_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36322_cast_fp16 = slice_by_index(begin = var_36322_begin_0, end = var_36322_end_0, end_mask = var_36322_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36322_cast_fp16")]; + tensor var_36326_begin_0 = const()[name = tensor("op_36326_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_36326_end_0 = const()[name = tensor("op_36326_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_36326_end_mask_0 = const()[name = tensor("op_36326_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36326_cast_fp16 = slice_by_index(begin = var_36326_begin_0, end = var_36326_end_0, end_mask = var_36326_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36326_cast_fp16")]; + tensor var_36330_begin_0 = const()[name = tensor("op_36330_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_36330_end_0 = const()[name = tensor("op_36330_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_36330_end_mask_0 = const()[name = tensor("op_36330_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36330_cast_fp16 = slice_by_index(begin = var_36330_begin_0, end = var_36330_end_0, end_mask = var_36330_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36330_cast_fp16")]; + tensor var_36334_begin_0 = const()[name = tensor("op_36334_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_36334_end_0 = const()[name = tensor("op_36334_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_36334_end_mask_0 = const()[name = tensor("op_36334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36334_cast_fp16 = slice_by_index(begin = var_36334_begin_0, end = var_36334_end_0, end_mask = var_36334_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36334_cast_fp16")]; + tensor var_36338_begin_0 = const()[name = tensor("op_36338_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_36338_end_0 = const()[name = tensor("op_36338_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_36338_end_mask_0 = const()[name = tensor("op_36338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36338_cast_fp16 = slice_by_index(begin = var_36338_begin_0, end = var_36338_end_0, end_mask = var_36338_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36338_cast_fp16")]; + tensor var_36342_begin_0 = const()[name = tensor("op_36342_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_36342_end_0 = const()[name = tensor("op_36342_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_36342_end_mask_0 = const()[name = tensor("op_36342_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36342_cast_fp16 = slice_by_index(begin = var_36342_begin_0, end = var_36342_end_0, end_mask = var_36342_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36342_cast_fp16")]; + tensor var_36346_begin_0 = const()[name = tensor("op_36346_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_36346_end_0 = const()[name = tensor("op_36346_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_36346_end_mask_0 = const()[name = tensor("op_36346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36346_cast_fp16 = slice_by_index(begin = var_36346_begin_0, end = var_36346_end_0, end_mask = var_36346_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36346_cast_fp16")]; + tensor var_36350_begin_0 = const()[name = tensor("op_36350_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_36350_end_0 = const()[name = tensor("op_36350_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_36350_end_mask_0 = const()[name = tensor("op_36350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36350_cast_fp16 = slice_by_index(begin = var_36350_begin_0, end = var_36350_end_0, end_mask = var_36350_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36350_cast_fp16")]; + tensor var_36354_begin_0 = const()[name = tensor("op_36354_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_36354_end_0 = const()[name = tensor("op_36354_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_36354_end_mask_0 = const()[name = tensor("op_36354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36354_cast_fp16 = slice_by_index(begin = var_36354_begin_0, end = var_36354_end_0, end_mask = var_36354_end_mask_0, x = query_45_cast_fp16)[name = tensor("op_36354_cast_fp16")]; + tensor var_36363_begin_0 = const()[name = tensor("op_36363_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36363_end_0 = const()[name = tensor("op_36363_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36363_end_mask_0 = const()[name = tensor("op_36363_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36363_cast_fp16 = slice_by_index(begin = var_36363_begin_0, end = var_36363_end_0, end_mask = var_36363_end_mask_0, x = var_36278_cast_fp16)[name = tensor("op_36363_cast_fp16")]; + tensor var_36370_begin_0 = const()[name = tensor("op_36370_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36370_end_0 = const()[name = tensor("op_36370_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36370_end_mask_0 = const()[name = tensor("op_36370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36370_cast_fp16 = slice_by_index(begin = var_36370_begin_0, end = var_36370_end_0, end_mask = var_36370_end_mask_0, x = var_36278_cast_fp16)[name = tensor("op_36370_cast_fp16")]; + tensor var_36377_begin_0 = const()[name = tensor("op_36377_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36377_end_0 = const()[name = tensor("op_36377_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36377_end_mask_0 = const()[name = tensor("op_36377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36377_cast_fp16 = slice_by_index(begin = var_36377_begin_0, end = var_36377_end_0, end_mask = var_36377_end_mask_0, x = var_36278_cast_fp16)[name = tensor("op_36377_cast_fp16")]; + tensor var_36384_begin_0 = const()[name = tensor("op_36384_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36384_end_0 = const()[name = tensor("op_36384_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36384_end_mask_0 = const()[name = tensor("op_36384_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36384_cast_fp16 = slice_by_index(begin = var_36384_begin_0, end = var_36384_end_0, end_mask = var_36384_end_mask_0, x = var_36278_cast_fp16)[name = tensor("op_36384_cast_fp16")]; + tensor var_36391_begin_0 = const()[name = tensor("op_36391_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36391_end_0 = const()[name = tensor("op_36391_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36391_end_mask_0 = const()[name = tensor("op_36391_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36391_cast_fp16 = slice_by_index(begin = var_36391_begin_0, end = var_36391_end_0, end_mask = var_36391_end_mask_0, x = var_36282_cast_fp16)[name = tensor("op_36391_cast_fp16")]; + tensor var_36398_begin_0 = const()[name = tensor("op_36398_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36398_end_0 = const()[name = tensor("op_36398_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36398_end_mask_0 = const()[name = tensor("op_36398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36398_cast_fp16 = slice_by_index(begin = var_36398_begin_0, end = var_36398_end_0, end_mask = var_36398_end_mask_0, x = var_36282_cast_fp16)[name = tensor("op_36398_cast_fp16")]; + tensor var_36405_begin_0 = const()[name = tensor("op_36405_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36405_end_0 = const()[name = tensor("op_36405_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36405_end_mask_0 = const()[name = tensor("op_36405_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36405_cast_fp16 = slice_by_index(begin = var_36405_begin_0, end = var_36405_end_0, end_mask = var_36405_end_mask_0, x = var_36282_cast_fp16)[name = tensor("op_36405_cast_fp16")]; + tensor var_36412_begin_0 = const()[name = tensor("op_36412_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36412_end_0 = const()[name = tensor("op_36412_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36412_end_mask_0 = const()[name = tensor("op_36412_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36412_cast_fp16 = slice_by_index(begin = var_36412_begin_0, end = var_36412_end_0, end_mask = var_36412_end_mask_0, x = var_36282_cast_fp16)[name = tensor("op_36412_cast_fp16")]; + tensor var_36419_begin_0 = const()[name = tensor("op_36419_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36419_end_0 = const()[name = tensor("op_36419_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36419_end_mask_0 = const()[name = tensor("op_36419_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36419_cast_fp16 = slice_by_index(begin = var_36419_begin_0, end = var_36419_end_0, end_mask = var_36419_end_mask_0, x = var_36286_cast_fp16)[name = tensor("op_36419_cast_fp16")]; + tensor var_36426_begin_0 = const()[name = tensor("op_36426_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36426_end_0 = const()[name = tensor("op_36426_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36426_end_mask_0 = const()[name = tensor("op_36426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36426_cast_fp16 = slice_by_index(begin = var_36426_begin_0, end = var_36426_end_0, end_mask = var_36426_end_mask_0, x = var_36286_cast_fp16)[name = tensor("op_36426_cast_fp16")]; + tensor var_36433_begin_0 = const()[name = tensor("op_36433_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36433_end_0 = const()[name = tensor("op_36433_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36433_end_mask_0 = const()[name = tensor("op_36433_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36433_cast_fp16 = slice_by_index(begin = var_36433_begin_0, end = var_36433_end_0, end_mask = var_36433_end_mask_0, x = var_36286_cast_fp16)[name = tensor("op_36433_cast_fp16")]; + tensor var_36440_begin_0 = const()[name = tensor("op_36440_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36440_end_0 = const()[name = tensor("op_36440_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36440_end_mask_0 = const()[name = tensor("op_36440_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36440_cast_fp16 = slice_by_index(begin = var_36440_begin_0, end = var_36440_end_0, end_mask = var_36440_end_mask_0, x = var_36286_cast_fp16)[name = tensor("op_36440_cast_fp16")]; + tensor var_36447_begin_0 = const()[name = tensor("op_36447_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36447_end_0 = const()[name = tensor("op_36447_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36447_end_mask_0 = const()[name = tensor("op_36447_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36447_cast_fp16 = slice_by_index(begin = var_36447_begin_0, end = var_36447_end_0, end_mask = var_36447_end_mask_0, x = var_36290_cast_fp16)[name = tensor("op_36447_cast_fp16")]; + tensor var_36454_begin_0 = const()[name = tensor("op_36454_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36454_end_0 = const()[name = tensor("op_36454_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36454_end_mask_0 = const()[name = tensor("op_36454_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36454_cast_fp16 = slice_by_index(begin = var_36454_begin_0, end = var_36454_end_0, end_mask = var_36454_end_mask_0, x = var_36290_cast_fp16)[name = tensor("op_36454_cast_fp16")]; + tensor var_36461_begin_0 = const()[name = tensor("op_36461_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36461_end_0 = const()[name = tensor("op_36461_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36461_end_mask_0 = const()[name = tensor("op_36461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36461_cast_fp16 = slice_by_index(begin = var_36461_begin_0, end = var_36461_end_0, end_mask = var_36461_end_mask_0, x = var_36290_cast_fp16)[name = tensor("op_36461_cast_fp16")]; + tensor var_36468_begin_0 = const()[name = tensor("op_36468_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36468_end_0 = const()[name = tensor("op_36468_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36468_end_mask_0 = const()[name = tensor("op_36468_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36468_cast_fp16 = slice_by_index(begin = var_36468_begin_0, end = var_36468_end_0, end_mask = var_36468_end_mask_0, x = var_36290_cast_fp16)[name = tensor("op_36468_cast_fp16")]; + tensor var_36475_begin_0 = const()[name = tensor("op_36475_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36475_end_0 = const()[name = tensor("op_36475_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36475_end_mask_0 = const()[name = tensor("op_36475_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36475_cast_fp16 = slice_by_index(begin = var_36475_begin_0, end = var_36475_end_0, end_mask = var_36475_end_mask_0, x = var_36294_cast_fp16)[name = tensor("op_36475_cast_fp16")]; + tensor var_36482_begin_0 = const()[name = tensor("op_36482_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36482_end_0 = const()[name = tensor("op_36482_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36482_end_mask_0 = const()[name = tensor("op_36482_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36482_cast_fp16 = slice_by_index(begin = var_36482_begin_0, end = var_36482_end_0, end_mask = var_36482_end_mask_0, x = var_36294_cast_fp16)[name = tensor("op_36482_cast_fp16")]; + tensor var_36489_begin_0 = const()[name = tensor("op_36489_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36489_end_0 = const()[name = tensor("op_36489_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36489_end_mask_0 = const()[name = tensor("op_36489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36489_cast_fp16 = slice_by_index(begin = var_36489_begin_0, end = var_36489_end_0, end_mask = var_36489_end_mask_0, x = var_36294_cast_fp16)[name = tensor("op_36489_cast_fp16")]; + tensor var_36496_begin_0 = const()[name = tensor("op_36496_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36496_end_0 = const()[name = tensor("op_36496_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36496_end_mask_0 = const()[name = tensor("op_36496_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36496_cast_fp16 = slice_by_index(begin = var_36496_begin_0, end = var_36496_end_0, end_mask = var_36496_end_mask_0, x = var_36294_cast_fp16)[name = tensor("op_36496_cast_fp16")]; + tensor var_36503_begin_0 = const()[name = tensor("op_36503_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36503_end_0 = const()[name = tensor("op_36503_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36503_end_mask_0 = const()[name = tensor("op_36503_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36503_cast_fp16 = slice_by_index(begin = var_36503_begin_0, end = var_36503_end_0, end_mask = var_36503_end_mask_0, x = var_36298_cast_fp16)[name = tensor("op_36503_cast_fp16")]; + tensor var_36510_begin_0 = const()[name = tensor("op_36510_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36510_end_0 = const()[name = tensor("op_36510_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36510_end_mask_0 = const()[name = tensor("op_36510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36510_cast_fp16 = slice_by_index(begin = var_36510_begin_0, end = var_36510_end_0, end_mask = var_36510_end_mask_0, x = var_36298_cast_fp16)[name = tensor("op_36510_cast_fp16")]; + tensor var_36517_begin_0 = const()[name = tensor("op_36517_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36517_end_0 = const()[name = tensor("op_36517_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36517_end_mask_0 = const()[name = tensor("op_36517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36517_cast_fp16 = slice_by_index(begin = var_36517_begin_0, end = var_36517_end_0, end_mask = var_36517_end_mask_0, x = var_36298_cast_fp16)[name = tensor("op_36517_cast_fp16")]; + tensor var_36524_begin_0 = const()[name = tensor("op_36524_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36524_end_0 = const()[name = tensor("op_36524_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36524_end_mask_0 = const()[name = tensor("op_36524_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36524_cast_fp16 = slice_by_index(begin = var_36524_begin_0, end = var_36524_end_0, end_mask = var_36524_end_mask_0, x = var_36298_cast_fp16)[name = tensor("op_36524_cast_fp16")]; + tensor var_36531_begin_0 = const()[name = tensor("op_36531_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36531_end_0 = const()[name = tensor("op_36531_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36531_end_mask_0 = const()[name = tensor("op_36531_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36531_cast_fp16 = slice_by_index(begin = var_36531_begin_0, end = var_36531_end_0, end_mask = var_36531_end_mask_0, x = var_36302_cast_fp16)[name = tensor("op_36531_cast_fp16")]; + tensor var_36538_begin_0 = const()[name = tensor("op_36538_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36538_end_0 = const()[name = tensor("op_36538_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36538_end_mask_0 = const()[name = tensor("op_36538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36538_cast_fp16 = slice_by_index(begin = var_36538_begin_0, end = var_36538_end_0, end_mask = var_36538_end_mask_0, x = var_36302_cast_fp16)[name = tensor("op_36538_cast_fp16")]; + tensor var_36545_begin_0 = const()[name = tensor("op_36545_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36545_end_0 = const()[name = tensor("op_36545_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36545_end_mask_0 = const()[name = tensor("op_36545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36545_cast_fp16 = slice_by_index(begin = var_36545_begin_0, end = var_36545_end_0, end_mask = var_36545_end_mask_0, x = var_36302_cast_fp16)[name = tensor("op_36545_cast_fp16")]; + tensor var_36552_begin_0 = const()[name = tensor("op_36552_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36552_end_0 = const()[name = tensor("op_36552_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36552_end_mask_0 = const()[name = tensor("op_36552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36552_cast_fp16 = slice_by_index(begin = var_36552_begin_0, end = var_36552_end_0, end_mask = var_36552_end_mask_0, x = var_36302_cast_fp16)[name = tensor("op_36552_cast_fp16")]; + tensor var_36559_begin_0 = const()[name = tensor("op_36559_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36559_end_0 = const()[name = tensor("op_36559_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36559_end_mask_0 = const()[name = tensor("op_36559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36559_cast_fp16 = slice_by_index(begin = var_36559_begin_0, end = var_36559_end_0, end_mask = var_36559_end_mask_0, x = var_36306_cast_fp16)[name = tensor("op_36559_cast_fp16")]; + tensor var_36566_begin_0 = const()[name = tensor("op_36566_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36566_end_0 = const()[name = tensor("op_36566_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36566_end_mask_0 = const()[name = tensor("op_36566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36566_cast_fp16 = slice_by_index(begin = var_36566_begin_0, end = var_36566_end_0, end_mask = var_36566_end_mask_0, x = var_36306_cast_fp16)[name = tensor("op_36566_cast_fp16")]; + tensor var_36573_begin_0 = const()[name = tensor("op_36573_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36573_end_0 = const()[name = tensor("op_36573_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36573_end_mask_0 = const()[name = tensor("op_36573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36573_cast_fp16 = slice_by_index(begin = var_36573_begin_0, end = var_36573_end_0, end_mask = var_36573_end_mask_0, x = var_36306_cast_fp16)[name = tensor("op_36573_cast_fp16")]; + tensor var_36580_begin_0 = const()[name = tensor("op_36580_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36580_end_0 = const()[name = tensor("op_36580_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36580_end_mask_0 = const()[name = tensor("op_36580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36580_cast_fp16 = slice_by_index(begin = var_36580_begin_0, end = var_36580_end_0, end_mask = var_36580_end_mask_0, x = var_36306_cast_fp16)[name = tensor("op_36580_cast_fp16")]; + tensor var_36587_begin_0 = const()[name = tensor("op_36587_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36587_end_0 = const()[name = tensor("op_36587_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36587_end_mask_0 = const()[name = tensor("op_36587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36587_cast_fp16 = slice_by_index(begin = var_36587_begin_0, end = var_36587_end_0, end_mask = var_36587_end_mask_0, x = var_36310_cast_fp16)[name = tensor("op_36587_cast_fp16")]; + tensor var_36594_begin_0 = const()[name = tensor("op_36594_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36594_end_0 = const()[name = tensor("op_36594_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36594_end_mask_0 = const()[name = tensor("op_36594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36594_cast_fp16 = slice_by_index(begin = var_36594_begin_0, end = var_36594_end_0, end_mask = var_36594_end_mask_0, x = var_36310_cast_fp16)[name = tensor("op_36594_cast_fp16")]; + tensor var_36601_begin_0 = const()[name = tensor("op_36601_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36601_end_0 = const()[name = tensor("op_36601_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36601_end_mask_0 = const()[name = tensor("op_36601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36601_cast_fp16 = slice_by_index(begin = var_36601_begin_0, end = var_36601_end_0, end_mask = var_36601_end_mask_0, x = var_36310_cast_fp16)[name = tensor("op_36601_cast_fp16")]; + tensor var_36608_begin_0 = const()[name = tensor("op_36608_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36608_end_0 = const()[name = tensor("op_36608_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36608_end_mask_0 = const()[name = tensor("op_36608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36608_cast_fp16 = slice_by_index(begin = var_36608_begin_0, end = var_36608_end_0, end_mask = var_36608_end_mask_0, x = var_36310_cast_fp16)[name = tensor("op_36608_cast_fp16")]; + tensor var_36615_begin_0 = const()[name = tensor("op_36615_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36615_end_0 = const()[name = tensor("op_36615_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36615_end_mask_0 = const()[name = tensor("op_36615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36615_cast_fp16 = slice_by_index(begin = var_36615_begin_0, end = var_36615_end_0, end_mask = var_36615_end_mask_0, x = var_36314_cast_fp16)[name = tensor("op_36615_cast_fp16")]; + tensor var_36622_begin_0 = const()[name = tensor("op_36622_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36622_end_0 = const()[name = tensor("op_36622_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36622_end_mask_0 = const()[name = tensor("op_36622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36622_cast_fp16 = slice_by_index(begin = var_36622_begin_0, end = var_36622_end_0, end_mask = var_36622_end_mask_0, x = var_36314_cast_fp16)[name = tensor("op_36622_cast_fp16")]; + tensor var_36629_begin_0 = const()[name = tensor("op_36629_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36629_end_0 = const()[name = tensor("op_36629_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36629_end_mask_0 = const()[name = tensor("op_36629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36629_cast_fp16 = slice_by_index(begin = var_36629_begin_0, end = var_36629_end_0, end_mask = var_36629_end_mask_0, x = var_36314_cast_fp16)[name = tensor("op_36629_cast_fp16")]; + tensor var_36636_begin_0 = const()[name = tensor("op_36636_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36636_end_0 = const()[name = tensor("op_36636_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36636_end_mask_0 = const()[name = tensor("op_36636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36636_cast_fp16 = slice_by_index(begin = var_36636_begin_0, end = var_36636_end_0, end_mask = var_36636_end_mask_0, x = var_36314_cast_fp16)[name = tensor("op_36636_cast_fp16")]; + tensor var_36643_begin_0 = const()[name = tensor("op_36643_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36643_end_0 = const()[name = tensor("op_36643_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36643_end_mask_0 = const()[name = tensor("op_36643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36643_cast_fp16 = slice_by_index(begin = var_36643_begin_0, end = var_36643_end_0, end_mask = var_36643_end_mask_0, x = var_36318_cast_fp16)[name = tensor("op_36643_cast_fp16")]; + tensor var_36650_begin_0 = const()[name = tensor("op_36650_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36650_end_0 = const()[name = tensor("op_36650_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36650_end_mask_0 = const()[name = tensor("op_36650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36650_cast_fp16 = slice_by_index(begin = var_36650_begin_0, end = var_36650_end_0, end_mask = var_36650_end_mask_0, x = var_36318_cast_fp16)[name = tensor("op_36650_cast_fp16")]; + tensor var_36657_begin_0 = const()[name = tensor("op_36657_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36657_end_0 = const()[name = tensor("op_36657_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36657_end_mask_0 = const()[name = tensor("op_36657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36657_cast_fp16 = slice_by_index(begin = var_36657_begin_0, end = var_36657_end_0, end_mask = var_36657_end_mask_0, x = var_36318_cast_fp16)[name = tensor("op_36657_cast_fp16")]; + tensor var_36664_begin_0 = const()[name = tensor("op_36664_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36664_end_0 = const()[name = tensor("op_36664_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36664_end_mask_0 = const()[name = tensor("op_36664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36664_cast_fp16 = slice_by_index(begin = var_36664_begin_0, end = var_36664_end_0, end_mask = var_36664_end_mask_0, x = var_36318_cast_fp16)[name = tensor("op_36664_cast_fp16")]; + tensor var_36671_begin_0 = const()[name = tensor("op_36671_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36671_end_0 = const()[name = tensor("op_36671_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36671_end_mask_0 = const()[name = tensor("op_36671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36671_cast_fp16 = slice_by_index(begin = var_36671_begin_0, end = var_36671_end_0, end_mask = var_36671_end_mask_0, x = var_36322_cast_fp16)[name = tensor("op_36671_cast_fp16")]; + tensor var_36678_begin_0 = const()[name = tensor("op_36678_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36678_end_0 = const()[name = tensor("op_36678_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36678_end_mask_0 = const()[name = tensor("op_36678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36678_cast_fp16 = slice_by_index(begin = var_36678_begin_0, end = var_36678_end_0, end_mask = var_36678_end_mask_0, x = var_36322_cast_fp16)[name = tensor("op_36678_cast_fp16")]; + tensor var_36685_begin_0 = const()[name = tensor("op_36685_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36685_end_0 = const()[name = tensor("op_36685_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36685_end_mask_0 = const()[name = tensor("op_36685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36685_cast_fp16 = slice_by_index(begin = var_36685_begin_0, end = var_36685_end_0, end_mask = var_36685_end_mask_0, x = var_36322_cast_fp16)[name = tensor("op_36685_cast_fp16")]; + tensor var_36692_begin_0 = const()[name = tensor("op_36692_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36692_end_0 = const()[name = tensor("op_36692_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36692_end_mask_0 = const()[name = tensor("op_36692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36692_cast_fp16 = slice_by_index(begin = var_36692_begin_0, end = var_36692_end_0, end_mask = var_36692_end_mask_0, x = var_36322_cast_fp16)[name = tensor("op_36692_cast_fp16")]; + tensor var_36699_begin_0 = const()[name = tensor("op_36699_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36699_end_0 = const()[name = tensor("op_36699_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36699_end_mask_0 = const()[name = tensor("op_36699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36699_cast_fp16 = slice_by_index(begin = var_36699_begin_0, end = var_36699_end_0, end_mask = var_36699_end_mask_0, x = var_36326_cast_fp16)[name = tensor("op_36699_cast_fp16")]; + tensor var_36706_begin_0 = const()[name = tensor("op_36706_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36706_end_0 = const()[name = tensor("op_36706_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36706_end_mask_0 = const()[name = tensor("op_36706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36706_cast_fp16 = slice_by_index(begin = var_36706_begin_0, end = var_36706_end_0, end_mask = var_36706_end_mask_0, x = var_36326_cast_fp16)[name = tensor("op_36706_cast_fp16")]; + tensor var_36713_begin_0 = const()[name = tensor("op_36713_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36713_end_0 = const()[name = tensor("op_36713_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36713_end_mask_0 = const()[name = tensor("op_36713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36713_cast_fp16 = slice_by_index(begin = var_36713_begin_0, end = var_36713_end_0, end_mask = var_36713_end_mask_0, x = var_36326_cast_fp16)[name = tensor("op_36713_cast_fp16")]; + tensor var_36720_begin_0 = const()[name = tensor("op_36720_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36720_end_0 = const()[name = tensor("op_36720_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36720_end_mask_0 = const()[name = tensor("op_36720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36720_cast_fp16 = slice_by_index(begin = var_36720_begin_0, end = var_36720_end_0, end_mask = var_36720_end_mask_0, x = var_36326_cast_fp16)[name = tensor("op_36720_cast_fp16")]; + tensor var_36727_begin_0 = const()[name = tensor("op_36727_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36727_end_0 = const()[name = tensor("op_36727_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36727_end_mask_0 = const()[name = tensor("op_36727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36727_cast_fp16 = slice_by_index(begin = var_36727_begin_0, end = var_36727_end_0, end_mask = var_36727_end_mask_0, x = var_36330_cast_fp16)[name = tensor("op_36727_cast_fp16")]; + tensor var_36734_begin_0 = const()[name = tensor("op_36734_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36734_end_0 = const()[name = tensor("op_36734_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36734_end_mask_0 = const()[name = tensor("op_36734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36734_cast_fp16 = slice_by_index(begin = var_36734_begin_0, end = var_36734_end_0, end_mask = var_36734_end_mask_0, x = var_36330_cast_fp16)[name = tensor("op_36734_cast_fp16")]; + tensor var_36741_begin_0 = const()[name = tensor("op_36741_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36741_end_0 = const()[name = tensor("op_36741_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36741_end_mask_0 = const()[name = tensor("op_36741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36741_cast_fp16 = slice_by_index(begin = var_36741_begin_0, end = var_36741_end_0, end_mask = var_36741_end_mask_0, x = var_36330_cast_fp16)[name = tensor("op_36741_cast_fp16")]; + tensor var_36748_begin_0 = const()[name = tensor("op_36748_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36748_end_0 = const()[name = tensor("op_36748_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36748_end_mask_0 = const()[name = tensor("op_36748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36748_cast_fp16 = slice_by_index(begin = var_36748_begin_0, end = var_36748_end_0, end_mask = var_36748_end_mask_0, x = var_36330_cast_fp16)[name = tensor("op_36748_cast_fp16")]; + tensor var_36755_begin_0 = const()[name = tensor("op_36755_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36755_end_0 = const()[name = tensor("op_36755_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36755_end_mask_0 = const()[name = tensor("op_36755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36755_cast_fp16 = slice_by_index(begin = var_36755_begin_0, end = var_36755_end_0, end_mask = var_36755_end_mask_0, x = var_36334_cast_fp16)[name = tensor("op_36755_cast_fp16")]; + tensor var_36762_begin_0 = const()[name = tensor("op_36762_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36762_end_0 = const()[name = tensor("op_36762_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36762_end_mask_0 = const()[name = tensor("op_36762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36762_cast_fp16 = slice_by_index(begin = var_36762_begin_0, end = var_36762_end_0, end_mask = var_36762_end_mask_0, x = var_36334_cast_fp16)[name = tensor("op_36762_cast_fp16")]; + tensor var_36769_begin_0 = const()[name = tensor("op_36769_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36769_end_0 = const()[name = tensor("op_36769_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36769_end_mask_0 = const()[name = tensor("op_36769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36769_cast_fp16 = slice_by_index(begin = var_36769_begin_0, end = var_36769_end_0, end_mask = var_36769_end_mask_0, x = var_36334_cast_fp16)[name = tensor("op_36769_cast_fp16")]; + tensor var_36776_begin_0 = const()[name = tensor("op_36776_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36776_end_0 = const()[name = tensor("op_36776_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36776_end_mask_0 = const()[name = tensor("op_36776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36776_cast_fp16 = slice_by_index(begin = var_36776_begin_0, end = var_36776_end_0, end_mask = var_36776_end_mask_0, x = var_36334_cast_fp16)[name = tensor("op_36776_cast_fp16")]; + tensor var_36783_begin_0 = const()[name = tensor("op_36783_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36783_end_0 = const()[name = tensor("op_36783_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36783_end_mask_0 = const()[name = tensor("op_36783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36783_cast_fp16 = slice_by_index(begin = var_36783_begin_0, end = var_36783_end_0, end_mask = var_36783_end_mask_0, x = var_36338_cast_fp16)[name = tensor("op_36783_cast_fp16")]; + tensor var_36790_begin_0 = const()[name = tensor("op_36790_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36790_end_0 = const()[name = tensor("op_36790_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36790_end_mask_0 = const()[name = tensor("op_36790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36790_cast_fp16 = slice_by_index(begin = var_36790_begin_0, end = var_36790_end_0, end_mask = var_36790_end_mask_0, x = var_36338_cast_fp16)[name = tensor("op_36790_cast_fp16")]; + tensor var_36797_begin_0 = const()[name = tensor("op_36797_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36797_end_0 = const()[name = tensor("op_36797_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36797_end_mask_0 = const()[name = tensor("op_36797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36797_cast_fp16 = slice_by_index(begin = var_36797_begin_0, end = var_36797_end_0, end_mask = var_36797_end_mask_0, x = var_36338_cast_fp16)[name = tensor("op_36797_cast_fp16")]; + tensor var_36804_begin_0 = const()[name = tensor("op_36804_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36804_end_0 = const()[name = tensor("op_36804_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36804_end_mask_0 = const()[name = tensor("op_36804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36804_cast_fp16 = slice_by_index(begin = var_36804_begin_0, end = var_36804_end_0, end_mask = var_36804_end_mask_0, x = var_36338_cast_fp16)[name = tensor("op_36804_cast_fp16")]; + tensor var_36811_begin_0 = const()[name = tensor("op_36811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36811_end_0 = const()[name = tensor("op_36811_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36811_end_mask_0 = const()[name = tensor("op_36811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36811_cast_fp16 = slice_by_index(begin = var_36811_begin_0, end = var_36811_end_0, end_mask = var_36811_end_mask_0, x = var_36342_cast_fp16)[name = tensor("op_36811_cast_fp16")]; + tensor var_36818_begin_0 = const()[name = tensor("op_36818_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36818_end_0 = const()[name = tensor("op_36818_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36818_end_mask_0 = const()[name = tensor("op_36818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36818_cast_fp16 = slice_by_index(begin = var_36818_begin_0, end = var_36818_end_0, end_mask = var_36818_end_mask_0, x = var_36342_cast_fp16)[name = tensor("op_36818_cast_fp16")]; + tensor var_36825_begin_0 = const()[name = tensor("op_36825_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36825_end_0 = const()[name = tensor("op_36825_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36825_end_mask_0 = const()[name = tensor("op_36825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36825_cast_fp16 = slice_by_index(begin = var_36825_begin_0, end = var_36825_end_0, end_mask = var_36825_end_mask_0, x = var_36342_cast_fp16)[name = tensor("op_36825_cast_fp16")]; + tensor var_36832_begin_0 = const()[name = tensor("op_36832_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36832_end_0 = const()[name = tensor("op_36832_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36832_end_mask_0 = const()[name = tensor("op_36832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36832_cast_fp16 = slice_by_index(begin = var_36832_begin_0, end = var_36832_end_0, end_mask = var_36832_end_mask_0, x = var_36342_cast_fp16)[name = tensor("op_36832_cast_fp16")]; + tensor var_36839_begin_0 = const()[name = tensor("op_36839_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36839_end_0 = const()[name = tensor("op_36839_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36839_end_mask_0 = const()[name = tensor("op_36839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36839_cast_fp16 = slice_by_index(begin = var_36839_begin_0, end = var_36839_end_0, end_mask = var_36839_end_mask_0, x = var_36346_cast_fp16)[name = tensor("op_36839_cast_fp16")]; + tensor var_36846_begin_0 = const()[name = tensor("op_36846_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36846_end_0 = const()[name = tensor("op_36846_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36846_end_mask_0 = const()[name = tensor("op_36846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36846_cast_fp16 = slice_by_index(begin = var_36846_begin_0, end = var_36846_end_0, end_mask = var_36846_end_mask_0, x = var_36346_cast_fp16)[name = tensor("op_36846_cast_fp16")]; + tensor var_36853_begin_0 = const()[name = tensor("op_36853_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36853_end_0 = const()[name = tensor("op_36853_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36853_end_mask_0 = const()[name = tensor("op_36853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36853_cast_fp16 = slice_by_index(begin = var_36853_begin_0, end = var_36853_end_0, end_mask = var_36853_end_mask_0, x = var_36346_cast_fp16)[name = tensor("op_36853_cast_fp16")]; + tensor var_36860_begin_0 = const()[name = tensor("op_36860_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36860_end_0 = const()[name = tensor("op_36860_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36860_end_mask_0 = const()[name = tensor("op_36860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36860_cast_fp16 = slice_by_index(begin = var_36860_begin_0, end = var_36860_end_0, end_mask = var_36860_end_mask_0, x = var_36346_cast_fp16)[name = tensor("op_36860_cast_fp16")]; + tensor var_36867_begin_0 = const()[name = tensor("op_36867_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36867_end_0 = const()[name = tensor("op_36867_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36867_end_mask_0 = const()[name = tensor("op_36867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36867_cast_fp16 = slice_by_index(begin = var_36867_begin_0, end = var_36867_end_0, end_mask = var_36867_end_mask_0, x = var_36350_cast_fp16)[name = tensor("op_36867_cast_fp16")]; + tensor var_36874_begin_0 = const()[name = tensor("op_36874_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36874_end_0 = const()[name = tensor("op_36874_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36874_end_mask_0 = const()[name = tensor("op_36874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36874_cast_fp16 = slice_by_index(begin = var_36874_begin_0, end = var_36874_end_0, end_mask = var_36874_end_mask_0, x = var_36350_cast_fp16)[name = tensor("op_36874_cast_fp16")]; + tensor var_36881_begin_0 = const()[name = tensor("op_36881_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36881_end_0 = const()[name = tensor("op_36881_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36881_end_mask_0 = const()[name = tensor("op_36881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36881_cast_fp16 = slice_by_index(begin = var_36881_begin_0, end = var_36881_end_0, end_mask = var_36881_end_mask_0, x = var_36350_cast_fp16)[name = tensor("op_36881_cast_fp16")]; + tensor var_36888_begin_0 = const()[name = tensor("op_36888_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36888_end_0 = const()[name = tensor("op_36888_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36888_end_mask_0 = const()[name = tensor("op_36888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36888_cast_fp16 = slice_by_index(begin = var_36888_begin_0, end = var_36888_end_0, end_mask = var_36888_end_mask_0, x = var_36350_cast_fp16)[name = tensor("op_36888_cast_fp16")]; + tensor var_36895_begin_0 = const()[name = tensor("op_36895_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36895_end_0 = const()[name = tensor("op_36895_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_36895_end_mask_0 = const()[name = tensor("op_36895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36895_cast_fp16 = slice_by_index(begin = var_36895_begin_0, end = var_36895_end_0, end_mask = var_36895_end_mask_0, x = var_36354_cast_fp16)[name = tensor("op_36895_cast_fp16")]; + tensor var_36902_begin_0 = const()[name = tensor("op_36902_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_36902_end_0 = const()[name = tensor("op_36902_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_36902_end_mask_0 = const()[name = tensor("op_36902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36902_cast_fp16 = slice_by_index(begin = var_36902_begin_0, end = var_36902_end_0, end_mask = var_36902_end_mask_0, x = var_36354_cast_fp16)[name = tensor("op_36902_cast_fp16")]; + tensor var_36909_begin_0 = const()[name = tensor("op_36909_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_36909_end_0 = const()[name = tensor("op_36909_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_36909_end_mask_0 = const()[name = tensor("op_36909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36909_cast_fp16 = slice_by_index(begin = var_36909_begin_0, end = var_36909_end_0, end_mask = var_36909_end_mask_0, x = var_36354_cast_fp16)[name = tensor("op_36909_cast_fp16")]; + tensor var_36916_begin_0 = const()[name = tensor("op_36916_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_36916_end_0 = const()[name = tensor("op_36916_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36916_end_mask_0 = const()[name = tensor("op_36916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36916_cast_fp16 = slice_by_index(begin = var_36916_begin_0, end = var_36916_end_0, end_mask = var_36916_end_mask_0, x = var_36354_cast_fp16)[name = tensor("op_36916_cast_fp16")]; + tensor k_45_perm_0 = const()[name = tensor("k_45_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_36921_begin_0 = const()[name = tensor("op_36921_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36921_end_0 = const()[name = tensor("op_36921_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_36921_end_mask_0 = const()[name = tensor("op_36921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_9 = transpose(perm = k_45_perm_0, x = key_45_cast_fp16)[name = tensor("transpose_9")]; + tensor var_36921_cast_fp16 = slice_by_index(begin = var_36921_begin_0, end = var_36921_end_0, end_mask = var_36921_end_mask_0, x = transpose_9)[name = tensor("op_36921_cast_fp16")]; + tensor var_36925_begin_0 = const()[name = tensor("op_36925_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_36925_end_0 = const()[name = tensor("op_36925_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_36925_end_mask_0 = const()[name = tensor("op_36925_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36925_cast_fp16 = slice_by_index(begin = var_36925_begin_0, end = var_36925_end_0, end_mask = var_36925_end_mask_0, x = transpose_9)[name = tensor("op_36925_cast_fp16")]; + tensor var_36929_begin_0 = const()[name = tensor("op_36929_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_36929_end_0 = const()[name = tensor("op_36929_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_36929_end_mask_0 = const()[name = tensor("op_36929_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36929_cast_fp16 = slice_by_index(begin = var_36929_begin_0, end = var_36929_end_0, end_mask = var_36929_end_mask_0, x = transpose_9)[name = tensor("op_36929_cast_fp16")]; + tensor var_36933_begin_0 = const()[name = tensor("op_36933_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_36933_end_0 = const()[name = tensor("op_36933_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_36933_end_mask_0 = const()[name = tensor("op_36933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36933_cast_fp16 = slice_by_index(begin = var_36933_begin_0, end = var_36933_end_0, end_mask = var_36933_end_mask_0, x = transpose_9)[name = tensor("op_36933_cast_fp16")]; + tensor var_36937_begin_0 = const()[name = tensor("op_36937_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_36937_end_0 = const()[name = tensor("op_36937_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_36937_end_mask_0 = const()[name = tensor("op_36937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36937_cast_fp16 = slice_by_index(begin = var_36937_begin_0, end = var_36937_end_0, end_mask = var_36937_end_mask_0, x = transpose_9)[name = tensor("op_36937_cast_fp16")]; + tensor var_36941_begin_0 = const()[name = tensor("op_36941_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_36941_end_0 = const()[name = tensor("op_36941_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_36941_end_mask_0 = const()[name = tensor("op_36941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36941_cast_fp16 = slice_by_index(begin = var_36941_begin_0, end = var_36941_end_0, end_mask = var_36941_end_mask_0, x = transpose_9)[name = tensor("op_36941_cast_fp16")]; + tensor var_36945_begin_0 = const()[name = tensor("op_36945_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_36945_end_0 = const()[name = tensor("op_36945_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_36945_end_mask_0 = const()[name = tensor("op_36945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36945_cast_fp16 = slice_by_index(begin = var_36945_begin_0, end = var_36945_end_0, end_mask = var_36945_end_mask_0, x = transpose_9)[name = tensor("op_36945_cast_fp16")]; + tensor var_36949_begin_0 = const()[name = tensor("op_36949_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_36949_end_0 = const()[name = tensor("op_36949_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_36949_end_mask_0 = const()[name = tensor("op_36949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36949_cast_fp16 = slice_by_index(begin = var_36949_begin_0, end = var_36949_end_0, end_mask = var_36949_end_mask_0, x = transpose_9)[name = tensor("op_36949_cast_fp16")]; + tensor var_36953_begin_0 = const()[name = tensor("op_36953_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_36953_end_0 = const()[name = tensor("op_36953_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_36953_end_mask_0 = const()[name = tensor("op_36953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36953_cast_fp16 = slice_by_index(begin = var_36953_begin_0, end = var_36953_end_0, end_mask = var_36953_end_mask_0, x = transpose_9)[name = tensor("op_36953_cast_fp16")]; + tensor var_36957_begin_0 = const()[name = tensor("op_36957_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_36957_end_0 = const()[name = tensor("op_36957_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_36957_end_mask_0 = const()[name = tensor("op_36957_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36957_cast_fp16 = slice_by_index(begin = var_36957_begin_0, end = var_36957_end_0, end_mask = var_36957_end_mask_0, x = transpose_9)[name = tensor("op_36957_cast_fp16")]; + tensor var_36961_begin_0 = const()[name = tensor("op_36961_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_36961_end_0 = const()[name = tensor("op_36961_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_36961_end_mask_0 = const()[name = tensor("op_36961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36961_cast_fp16 = slice_by_index(begin = var_36961_begin_0, end = var_36961_end_0, end_mask = var_36961_end_mask_0, x = transpose_9)[name = tensor("op_36961_cast_fp16")]; + tensor var_36965_begin_0 = const()[name = tensor("op_36965_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_36965_end_0 = const()[name = tensor("op_36965_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_36965_end_mask_0 = const()[name = tensor("op_36965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36965_cast_fp16 = slice_by_index(begin = var_36965_begin_0, end = var_36965_end_0, end_mask = var_36965_end_mask_0, x = transpose_9)[name = tensor("op_36965_cast_fp16")]; + tensor var_36969_begin_0 = const()[name = tensor("op_36969_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_36969_end_0 = const()[name = tensor("op_36969_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_36969_end_mask_0 = const()[name = tensor("op_36969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36969_cast_fp16 = slice_by_index(begin = var_36969_begin_0, end = var_36969_end_0, end_mask = var_36969_end_mask_0, x = transpose_9)[name = tensor("op_36969_cast_fp16")]; + tensor var_36973_begin_0 = const()[name = tensor("op_36973_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_36973_end_0 = const()[name = tensor("op_36973_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_36973_end_mask_0 = const()[name = tensor("op_36973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36973_cast_fp16 = slice_by_index(begin = var_36973_begin_0, end = var_36973_end_0, end_mask = var_36973_end_mask_0, x = transpose_9)[name = tensor("op_36973_cast_fp16")]; + tensor var_36977_begin_0 = const()[name = tensor("op_36977_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_36977_end_0 = const()[name = tensor("op_36977_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_36977_end_mask_0 = const()[name = tensor("op_36977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36977_cast_fp16 = slice_by_index(begin = var_36977_begin_0, end = var_36977_end_0, end_mask = var_36977_end_mask_0, x = transpose_9)[name = tensor("op_36977_cast_fp16")]; + tensor var_36981_begin_0 = const()[name = tensor("op_36981_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_36981_end_0 = const()[name = tensor("op_36981_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_36981_end_mask_0 = const()[name = tensor("op_36981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36981_cast_fp16 = slice_by_index(begin = var_36981_begin_0, end = var_36981_end_0, end_mask = var_36981_end_mask_0, x = transpose_9)[name = tensor("op_36981_cast_fp16")]; + tensor var_36985_begin_0 = const()[name = tensor("op_36985_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_36985_end_0 = const()[name = tensor("op_36985_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_36985_end_mask_0 = const()[name = tensor("op_36985_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36985_cast_fp16 = slice_by_index(begin = var_36985_begin_0, end = var_36985_end_0, end_mask = var_36985_end_mask_0, x = transpose_9)[name = tensor("op_36985_cast_fp16")]; + tensor var_36989_begin_0 = const()[name = tensor("op_36989_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_36989_end_0 = const()[name = tensor("op_36989_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_36989_end_mask_0 = const()[name = tensor("op_36989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36989_cast_fp16 = slice_by_index(begin = var_36989_begin_0, end = var_36989_end_0, end_mask = var_36989_end_mask_0, x = transpose_9)[name = tensor("op_36989_cast_fp16")]; + tensor var_36993_begin_0 = const()[name = tensor("op_36993_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_36993_end_0 = const()[name = tensor("op_36993_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_36993_end_mask_0 = const()[name = tensor("op_36993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36993_cast_fp16 = slice_by_index(begin = var_36993_begin_0, end = var_36993_end_0, end_mask = var_36993_end_mask_0, x = transpose_9)[name = tensor("op_36993_cast_fp16")]; + tensor var_36997_begin_0 = const()[name = tensor("op_36997_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_36997_end_0 = const()[name = tensor("op_36997_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_36997_end_mask_0 = const()[name = tensor("op_36997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_36997_cast_fp16 = slice_by_index(begin = var_36997_begin_0, end = var_36997_end_0, end_mask = var_36997_end_mask_0, x = transpose_9)[name = tensor("op_36997_cast_fp16")]; + tensor var_36999_begin_0 = const()[name = tensor("op_36999_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_36999_end_0 = const()[name = tensor("op_36999_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_36999_end_mask_0 = const()[name = tensor("op_36999_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_36999_cast_fp16 = slice_by_index(begin = var_36999_begin_0, end = var_36999_end_0, end_mask = var_36999_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_36999_cast_fp16")]; + tensor var_37003_begin_0 = const()[name = tensor("op_37003_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37003_end_0 = const()[name = tensor("op_37003_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_37003_end_mask_0 = const()[name = tensor("op_37003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37003_cast_fp16 = slice_by_index(begin = var_37003_begin_0, end = var_37003_end_0, end_mask = var_37003_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37003_cast_fp16")]; + tensor var_37007_begin_0 = const()[name = tensor("op_37007_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37007_end_0 = const()[name = tensor("op_37007_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_37007_end_mask_0 = const()[name = tensor("op_37007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37007_cast_fp16 = slice_by_index(begin = var_37007_begin_0, end = var_37007_end_0, end_mask = var_37007_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37007_cast_fp16")]; + tensor var_37011_begin_0 = const()[name = tensor("op_37011_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37011_end_0 = const()[name = tensor("op_37011_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_37011_end_mask_0 = const()[name = tensor("op_37011_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37011_cast_fp16 = slice_by_index(begin = var_37011_begin_0, end = var_37011_end_0, end_mask = var_37011_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37011_cast_fp16")]; + tensor var_37015_begin_0 = const()[name = tensor("op_37015_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37015_end_0 = const()[name = tensor("op_37015_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_37015_end_mask_0 = const()[name = tensor("op_37015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37015_cast_fp16 = slice_by_index(begin = var_37015_begin_0, end = var_37015_end_0, end_mask = var_37015_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37015_cast_fp16")]; + tensor var_37019_begin_0 = const()[name = tensor("op_37019_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37019_end_0 = const()[name = tensor("op_37019_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_37019_end_mask_0 = const()[name = tensor("op_37019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37019_cast_fp16 = slice_by_index(begin = var_37019_begin_0, end = var_37019_end_0, end_mask = var_37019_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37019_cast_fp16")]; + tensor var_37023_begin_0 = const()[name = tensor("op_37023_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37023_end_0 = const()[name = tensor("op_37023_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_37023_end_mask_0 = const()[name = tensor("op_37023_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37023_cast_fp16 = slice_by_index(begin = var_37023_begin_0, end = var_37023_end_0, end_mask = var_37023_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37023_cast_fp16")]; + tensor var_37027_begin_0 = const()[name = tensor("op_37027_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37027_end_0 = const()[name = tensor("op_37027_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_37027_end_mask_0 = const()[name = tensor("op_37027_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37027_cast_fp16 = slice_by_index(begin = var_37027_begin_0, end = var_37027_end_0, end_mask = var_37027_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37027_cast_fp16")]; + tensor var_37031_begin_0 = const()[name = tensor("op_37031_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37031_end_0 = const()[name = tensor("op_37031_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_37031_end_mask_0 = const()[name = tensor("op_37031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37031_cast_fp16 = slice_by_index(begin = var_37031_begin_0, end = var_37031_end_0, end_mask = var_37031_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37031_cast_fp16")]; + tensor var_37035_begin_0 = const()[name = tensor("op_37035_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37035_end_0 = const()[name = tensor("op_37035_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_37035_end_mask_0 = const()[name = tensor("op_37035_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37035_cast_fp16 = slice_by_index(begin = var_37035_begin_0, end = var_37035_end_0, end_mask = var_37035_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37035_cast_fp16")]; + tensor var_37039_begin_0 = const()[name = tensor("op_37039_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37039_end_0 = const()[name = tensor("op_37039_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_37039_end_mask_0 = const()[name = tensor("op_37039_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37039_cast_fp16 = slice_by_index(begin = var_37039_begin_0, end = var_37039_end_0, end_mask = var_37039_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37039_cast_fp16")]; + tensor var_37043_begin_0 = const()[name = tensor("op_37043_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37043_end_0 = const()[name = tensor("op_37043_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_37043_end_mask_0 = const()[name = tensor("op_37043_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37043_cast_fp16 = slice_by_index(begin = var_37043_begin_0, end = var_37043_end_0, end_mask = var_37043_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37043_cast_fp16")]; + tensor var_37047_begin_0 = const()[name = tensor("op_37047_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37047_end_0 = const()[name = tensor("op_37047_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_37047_end_mask_0 = const()[name = tensor("op_37047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37047_cast_fp16 = slice_by_index(begin = var_37047_begin_0, end = var_37047_end_0, end_mask = var_37047_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37047_cast_fp16")]; + tensor var_37051_begin_0 = const()[name = tensor("op_37051_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37051_end_0 = const()[name = tensor("op_37051_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_37051_end_mask_0 = const()[name = tensor("op_37051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37051_cast_fp16 = slice_by_index(begin = var_37051_begin_0, end = var_37051_end_0, end_mask = var_37051_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37051_cast_fp16")]; + tensor var_37055_begin_0 = const()[name = tensor("op_37055_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37055_end_0 = const()[name = tensor("op_37055_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_37055_end_mask_0 = const()[name = tensor("op_37055_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37055_cast_fp16 = slice_by_index(begin = var_37055_begin_0, end = var_37055_end_0, end_mask = var_37055_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37055_cast_fp16")]; + tensor var_37059_begin_0 = const()[name = tensor("op_37059_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37059_end_0 = const()[name = tensor("op_37059_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_37059_end_mask_0 = const()[name = tensor("op_37059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37059_cast_fp16 = slice_by_index(begin = var_37059_begin_0, end = var_37059_end_0, end_mask = var_37059_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37059_cast_fp16")]; + tensor var_37063_begin_0 = const()[name = tensor("op_37063_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37063_end_0 = const()[name = tensor("op_37063_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_37063_end_mask_0 = const()[name = tensor("op_37063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37063_cast_fp16 = slice_by_index(begin = var_37063_begin_0, end = var_37063_end_0, end_mask = var_37063_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37063_cast_fp16")]; + tensor var_37067_begin_0 = const()[name = tensor("op_37067_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37067_end_0 = const()[name = tensor("op_37067_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_37067_end_mask_0 = const()[name = tensor("op_37067_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37067_cast_fp16 = slice_by_index(begin = var_37067_begin_0, end = var_37067_end_0, end_mask = var_37067_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37067_cast_fp16")]; + tensor var_37071_begin_0 = const()[name = tensor("op_37071_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37071_end_0 = const()[name = tensor("op_37071_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_37071_end_mask_0 = const()[name = tensor("op_37071_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37071_cast_fp16 = slice_by_index(begin = var_37071_begin_0, end = var_37071_end_0, end_mask = var_37071_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37071_cast_fp16")]; + tensor var_37075_begin_0 = const()[name = tensor("op_37075_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37075_end_0 = const()[name = tensor("op_37075_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_37075_end_mask_0 = const()[name = tensor("op_37075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37075_cast_fp16 = slice_by_index(begin = var_37075_begin_0, end = var_37075_end_0, end_mask = var_37075_end_mask_0, x = value_45_cast_fp16)[name = tensor("op_37075_cast_fp16")]; + tensor var_37079_equation_0 = const()[name = tensor("op_37079_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37079_cast_fp16 = einsum(equation = var_37079_equation_0, values = (var_36921_cast_fp16, var_36363_cast_fp16))[name = tensor("op_37079_cast_fp16")]; + tensor var_37080_to_fp16 = const()[name = tensor("op_37080_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3521_cast_fp16 = mul(x = var_37079_cast_fp16, y = var_37080_to_fp16)[name = tensor("aw_chunk_3521_cast_fp16")]; + tensor var_37083_equation_0 = const()[name = tensor("op_37083_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37083_cast_fp16 = einsum(equation = var_37083_equation_0, values = (var_36921_cast_fp16, var_36370_cast_fp16))[name = tensor("op_37083_cast_fp16")]; + tensor var_37084_to_fp16 = const()[name = tensor("op_37084_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3523_cast_fp16 = mul(x = var_37083_cast_fp16, y = var_37084_to_fp16)[name = tensor("aw_chunk_3523_cast_fp16")]; + tensor var_37087_equation_0 = const()[name = tensor("op_37087_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37087_cast_fp16 = einsum(equation = var_37087_equation_0, values = (var_36921_cast_fp16, var_36377_cast_fp16))[name = tensor("op_37087_cast_fp16")]; + tensor var_37088_to_fp16 = const()[name = tensor("op_37088_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3525_cast_fp16 = mul(x = var_37087_cast_fp16, y = var_37088_to_fp16)[name = tensor("aw_chunk_3525_cast_fp16")]; + tensor var_37091_equation_0 = const()[name = tensor("op_37091_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37091_cast_fp16 = einsum(equation = var_37091_equation_0, values = (var_36921_cast_fp16, var_36384_cast_fp16))[name = tensor("op_37091_cast_fp16")]; + tensor var_37092_to_fp16 = const()[name = tensor("op_37092_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3527_cast_fp16 = mul(x = var_37091_cast_fp16, y = var_37092_to_fp16)[name = tensor("aw_chunk_3527_cast_fp16")]; + tensor var_37095_equation_0 = const()[name = tensor("op_37095_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37095_cast_fp16 = einsum(equation = var_37095_equation_0, values = (var_36925_cast_fp16, var_36391_cast_fp16))[name = tensor("op_37095_cast_fp16")]; + tensor var_37096_to_fp16 = const()[name = tensor("op_37096_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3529_cast_fp16 = mul(x = var_37095_cast_fp16, y = var_37096_to_fp16)[name = tensor("aw_chunk_3529_cast_fp16")]; + tensor var_37099_equation_0 = const()[name = tensor("op_37099_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37099_cast_fp16 = einsum(equation = var_37099_equation_0, values = (var_36925_cast_fp16, var_36398_cast_fp16))[name = tensor("op_37099_cast_fp16")]; + tensor var_37100_to_fp16 = const()[name = tensor("op_37100_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3531_cast_fp16 = mul(x = var_37099_cast_fp16, y = var_37100_to_fp16)[name = tensor("aw_chunk_3531_cast_fp16")]; + tensor var_37103_equation_0 = const()[name = tensor("op_37103_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37103_cast_fp16 = einsum(equation = var_37103_equation_0, values = (var_36925_cast_fp16, var_36405_cast_fp16))[name = tensor("op_37103_cast_fp16")]; + tensor var_37104_to_fp16 = const()[name = tensor("op_37104_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3533_cast_fp16 = mul(x = var_37103_cast_fp16, y = var_37104_to_fp16)[name = tensor("aw_chunk_3533_cast_fp16")]; + tensor var_37107_equation_0 = const()[name = tensor("op_37107_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37107_cast_fp16 = einsum(equation = var_37107_equation_0, values = (var_36925_cast_fp16, var_36412_cast_fp16))[name = tensor("op_37107_cast_fp16")]; + tensor var_37108_to_fp16 = const()[name = tensor("op_37108_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3535_cast_fp16 = mul(x = var_37107_cast_fp16, y = var_37108_to_fp16)[name = tensor("aw_chunk_3535_cast_fp16")]; + tensor var_37111_equation_0 = const()[name = tensor("op_37111_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37111_cast_fp16 = einsum(equation = var_37111_equation_0, values = (var_36929_cast_fp16, var_36419_cast_fp16))[name = tensor("op_37111_cast_fp16")]; + tensor var_37112_to_fp16 = const()[name = tensor("op_37112_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3537_cast_fp16 = mul(x = var_37111_cast_fp16, y = var_37112_to_fp16)[name = tensor("aw_chunk_3537_cast_fp16")]; + tensor var_37115_equation_0 = const()[name = tensor("op_37115_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37115_cast_fp16 = einsum(equation = var_37115_equation_0, values = (var_36929_cast_fp16, var_36426_cast_fp16))[name = tensor("op_37115_cast_fp16")]; + tensor var_37116_to_fp16 = const()[name = tensor("op_37116_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3539_cast_fp16 = mul(x = var_37115_cast_fp16, y = var_37116_to_fp16)[name = tensor("aw_chunk_3539_cast_fp16")]; + tensor var_37119_equation_0 = const()[name = tensor("op_37119_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37119_cast_fp16 = einsum(equation = var_37119_equation_0, values = (var_36929_cast_fp16, var_36433_cast_fp16))[name = tensor("op_37119_cast_fp16")]; + tensor var_37120_to_fp16 = const()[name = tensor("op_37120_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3541_cast_fp16 = mul(x = var_37119_cast_fp16, y = var_37120_to_fp16)[name = tensor("aw_chunk_3541_cast_fp16")]; + tensor var_37123_equation_0 = const()[name = tensor("op_37123_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37123_cast_fp16 = einsum(equation = var_37123_equation_0, values = (var_36929_cast_fp16, var_36440_cast_fp16))[name = tensor("op_37123_cast_fp16")]; + tensor var_37124_to_fp16 = const()[name = tensor("op_37124_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3543_cast_fp16 = mul(x = var_37123_cast_fp16, y = var_37124_to_fp16)[name = tensor("aw_chunk_3543_cast_fp16")]; + tensor var_37127_equation_0 = const()[name = tensor("op_37127_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37127_cast_fp16 = einsum(equation = var_37127_equation_0, values = (var_36933_cast_fp16, var_36447_cast_fp16))[name = tensor("op_37127_cast_fp16")]; + tensor var_37128_to_fp16 = const()[name = tensor("op_37128_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3545_cast_fp16 = mul(x = var_37127_cast_fp16, y = var_37128_to_fp16)[name = tensor("aw_chunk_3545_cast_fp16")]; + tensor var_37131_equation_0 = const()[name = tensor("op_37131_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37131_cast_fp16 = einsum(equation = var_37131_equation_0, values = (var_36933_cast_fp16, var_36454_cast_fp16))[name = tensor("op_37131_cast_fp16")]; + tensor var_37132_to_fp16 = const()[name = tensor("op_37132_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3547_cast_fp16 = mul(x = var_37131_cast_fp16, y = var_37132_to_fp16)[name = tensor("aw_chunk_3547_cast_fp16")]; + tensor var_37135_equation_0 = const()[name = tensor("op_37135_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37135_cast_fp16 = einsum(equation = var_37135_equation_0, values = (var_36933_cast_fp16, var_36461_cast_fp16))[name = tensor("op_37135_cast_fp16")]; + tensor var_37136_to_fp16 = const()[name = tensor("op_37136_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3549_cast_fp16 = mul(x = var_37135_cast_fp16, y = var_37136_to_fp16)[name = tensor("aw_chunk_3549_cast_fp16")]; + tensor var_37139_equation_0 = const()[name = tensor("op_37139_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37139_cast_fp16 = einsum(equation = var_37139_equation_0, values = (var_36933_cast_fp16, var_36468_cast_fp16))[name = tensor("op_37139_cast_fp16")]; + tensor var_37140_to_fp16 = const()[name = tensor("op_37140_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3551_cast_fp16 = mul(x = var_37139_cast_fp16, y = var_37140_to_fp16)[name = tensor("aw_chunk_3551_cast_fp16")]; + tensor var_37143_equation_0 = const()[name = tensor("op_37143_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37143_cast_fp16 = einsum(equation = var_37143_equation_0, values = (var_36937_cast_fp16, var_36475_cast_fp16))[name = tensor("op_37143_cast_fp16")]; + tensor var_37144_to_fp16 = const()[name = tensor("op_37144_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3553_cast_fp16 = mul(x = var_37143_cast_fp16, y = var_37144_to_fp16)[name = tensor("aw_chunk_3553_cast_fp16")]; + tensor var_37147_equation_0 = const()[name = tensor("op_37147_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37147_cast_fp16 = einsum(equation = var_37147_equation_0, values = (var_36937_cast_fp16, var_36482_cast_fp16))[name = tensor("op_37147_cast_fp16")]; + tensor var_37148_to_fp16 = const()[name = tensor("op_37148_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3555_cast_fp16 = mul(x = var_37147_cast_fp16, y = var_37148_to_fp16)[name = tensor("aw_chunk_3555_cast_fp16")]; + tensor var_37151_equation_0 = const()[name = tensor("op_37151_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37151_cast_fp16 = einsum(equation = var_37151_equation_0, values = (var_36937_cast_fp16, var_36489_cast_fp16))[name = tensor("op_37151_cast_fp16")]; + tensor var_37152_to_fp16 = const()[name = tensor("op_37152_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3557_cast_fp16 = mul(x = var_37151_cast_fp16, y = var_37152_to_fp16)[name = tensor("aw_chunk_3557_cast_fp16")]; + tensor var_37155_equation_0 = const()[name = tensor("op_37155_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37155_cast_fp16 = einsum(equation = var_37155_equation_0, values = (var_36937_cast_fp16, var_36496_cast_fp16))[name = tensor("op_37155_cast_fp16")]; + tensor var_37156_to_fp16 = const()[name = tensor("op_37156_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3559_cast_fp16 = mul(x = var_37155_cast_fp16, y = var_37156_to_fp16)[name = tensor("aw_chunk_3559_cast_fp16")]; + tensor var_37159_equation_0 = const()[name = tensor("op_37159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37159_cast_fp16 = einsum(equation = var_37159_equation_0, values = (var_36941_cast_fp16, var_36503_cast_fp16))[name = tensor("op_37159_cast_fp16")]; + tensor var_37160_to_fp16 = const()[name = tensor("op_37160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3561_cast_fp16 = mul(x = var_37159_cast_fp16, y = var_37160_to_fp16)[name = tensor("aw_chunk_3561_cast_fp16")]; + tensor var_37163_equation_0 = const()[name = tensor("op_37163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37163_cast_fp16 = einsum(equation = var_37163_equation_0, values = (var_36941_cast_fp16, var_36510_cast_fp16))[name = tensor("op_37163_cast_fp16")]; + tensor var_37164_to_fp16 = const()[name = tensor("op_37164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3563_cast_fp16 = mul(x = var_37163_cast_fp16, y = var_37164_to_fp16)[name = tensor("aw_chunk_3563_cast_fp16")]; + tensor var_37167_equation_0 = const()[name = tensor("op_37167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37167_cast_fp16 = einsum(equation = var_37167_equation_0, values = (var_36941_cast_fp16, var_36517_cast_fp16))[name = tensor("op_37167_cast_fp16")]; + tensor var_37168_to_fp16 = const()[name = tensor("op_37168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3565_cast_fp16 = mul(x = var_37167_cast_fp16, y = var_37168_to_fp16)[name = tensor("aw_chunk_3565_cast_fp16")]; + tensor var_37171_equation_0 = const()[name = tensor("op_37171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37171_cast_fp16 = einsum(equation = var_37171_equation_0, values = (var_36941_cast_fp16, var_36524_cast_fp16))[name = tensor("op_37171_cast_fp16")]; + tensor var_37172_to_fp16 = const()[name = tensor("op_37172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3567_cast_fp16 = mul(x = var_37171_cast_fp16, y = var_37172_to_fp16)[name = tensor("aw_chunk_3567_cast_fp16")]; + tensor var_37175_equation_0 = const()[name = tensor("op_37175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37175_cast_fp16 = einsum(equation = var_37175_equation_0, values = (var_36945_cast_fp16, var_36531_cast_fp16))[name = tensor("op_37175_cast_fp16")]; + tensor var_37176_to_fp16 = const()[name = tensor("op_37176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3569_cast_fp16 = mul(x = var_37175_cast_fp16, y = var_37176_to_fp16)[name = tensor("aw_chunk_3569_cast_fp16")]; + tensor var_37179_equation_0 = const()[name = tensor("op_37179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37179_cast_fp16 = einsum(equation = var_37179_equation_0, values = (var_36945_cast_fp16, var_36538_cast_fp16))[name = tensor("op_37179_cast_fp16")]; + tensor var_37180_to_fp16 = const()[name = tensor("op_37180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3571_cast_fp16 = mul(x = var_37179_cast_fp16, y = var_37180_to_fp16)[name = tensor("aw_chunk_3571_cast_fp16")]; + tensor var_37183_equation_0 = const()[name = tensor("op_37183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37183_cast_fp16 = einsum(equation = var_37183_equation_0, values = (var_36945_cast_fp16, var_36545_cast_fp16))[name = tensor("op_37183_cast_fp16")]; + tensor var_37184_to_fp16 = const()[name = tensor("op_37184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3573_cast_fp16 = mul(x = var_37183_cast_fp16, y = var_37184_to_fp16)[name = tensor("aw_chunk_3573_cast_fp16")]; + tensor var_37187_equation_0 = const()[name = tensor("op_37187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37187_cast_fp16 = einsum(equation = var_37187_equation_0, values = (var_36945_cast_fp16, var_36552_cast_fp16))[name = tensor("op_37187_cast_fp16")]; + tensor var_37188_to_fp16 = const()[name = tensor("op_37188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3575_cast_fp16 = mul(x = var_37187_cast_fp16, y = var_37188_to_fp16)[name = tensor("aw_chunk_3575_cast_fp16")]; + tensor var_37191_equation_0 = const()[name = tensor("op_37191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37191_cast_fp16 = einsum(equation = var_37191_equation_0, values = (var_36949_cast_fp16, var_36559_cast_fp16))[name = tensor("op_37191_cast_fp16")]; + tensor var_37192_to_fp16 = const()[name = tensor("op_37192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3577_cast_fp16 = mul(x = var_37191_cast_fp16, y = var_37192_to_fp16)[name = tensor("aw_chunk_3577_cast_fp16")]; + tensor var_37195_equation_0 = const()[name = tensor("op_37195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37195_cast_fp16 = einsum(equation = var_37195_equation_0, values = (var_36949_cast_fp16, var_36566_cast_fp16))[name = tensor("op_37195_cast_fp16")]; + tensor var_37196_to_fp16 = const()[name = tensor("op_37196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3579_cast_fp16 = mul(x = var_37195_cast_fp16, y = var_37196_to_fp16)[name = tensor("aw_chunk_3579_cast_fp16")]; + tensor var_37199_equation_0 = const()[name = tensor("op_37199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37199_cast_fp16 = einsum(equation = var_37199_equation_0, values = (var_36949_cast_fp16, var_36573_cast_fp16))[name = tensor("op_37199_cast_fp16")]; + tensor var_37200_to_fp16 = const()[name = tensor("op_37200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3581_cast_fp16 = mul(x = var_37199_cast_fp16, y = var_37200_to_fp16)[name = tensor("aw_chunk_3581_cast_fp16")]; + tensor var_37203_equation_0 = const()[name = tensor("op_37203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37203_cast_fp16 = einsum(equation = var_37203_equation_0, values = (var_36949_cast_fp16, var_36580_cast_fp16))[name = tensor("op_37203_cast_fp16")]; + tensor var_37204_to_fp16 = const()[name = tensor("op_37204_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3583_cast_fp16 = mul(x = var_37203_cast_fp16, y = var_37204_to_fp16)[name = tensor("aw_chunk_3583_cast_fp16")]; + tensor var_37207_equation_0 = const()[name = tensor("op_37207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37207_cast_fp16 = einsum(equation = var_37207_equation_0, values = (var_36953_cast_fp16, var_36587_cast_fp16))[name = tensor("op_37207_cast_fp16")]; + tensor var_37208_to_fp16 = const()[name = tensor("op_37208_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3585_cast_fp16 = mul(x = var_37207_cast_fp16, y = var_37208_to_fp16)[name = tensor("aw_chunk_3585_cast_fp16")]; + tensor var_37211_equation_0 = const()[name = tensor("op_37211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37211_cast_fp16 = einsum(equation = var_37211_equation_0, values = (var_36953_cast_fp16, var_36594_cast_fp16))[name = tensor("op_37211_cast_fp16")]; + tensor var_37212_to_fp16 = const()[name = tensor("op_37212_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3587_cast_fp16 = mul(x = var_37211_cast_fp16, y = var_37212_to_fp16)[name = tensor("aw_chunk_3587_cast_fp16")]; + tensor var_37215_equation_0 = const()[name = tensor("op_37215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37215_cast_fp16 = einsum(equation = var_37215_equation_0, values = (var_36953_cast_fp16, var_36601_cast_fp16))[name = tensor("op_37215_cast_fp16")]; + tensor var_37216_to_fp16 = const()[name = tensor("op_37216_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3589_cast_fp16 = mul(x = var_37215_cast_fp16, y = var_37216_to_fp16)[name = tensor("aw_chunk_3589_cast_fp16")]; + tensor var_37219_equation_0 = const()[name = tensor("op_37219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37219_cast_fp16 = einsum(equation = var_37219_equation_0, values = (var_36953_cast_fp16, var_36608_cast_fp16))[name = tensor("op_37219_cast_fp16")]; + tensor var_37220_to_fp16 = const()[name = tensor("op_37220_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3591_cast_fp16 = mul(x = var_37219_cast_fp16, y = var_37220_to_fp16)[name = tensor("aw_chunk_3591_cast_fp16")]; + tensor var_37223_equation_0 = const()[name = tensor("op_37223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37223_cast_fp16 = einsum(equation = var_37223_equation_0, values = (var_36957_cast_fp16, var_36615_cast_fp16))[name = tensor("op_37223_cast_fp16")]; + tensor var_37224_to_fp16 = const()[name = tensor("op_37224_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3593_cast_fp16 = mul(x = var_37223_cast_fp16, y = var_37224_to_fp16)[name = tensor("aw_chunk_3593_cast_fp16")]; + tensor var_37227_equation_0 = const()[name = tensor("op_37227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37227_cast_fp16 = einsum(equation = var_37227_equation_0, values = (var_36957_cast_fp16, var_36622_cast_fp16))[name = tensor("op_37227_cast_fp16")]; + tensor var_37228_to_fp16 = const()[name = tensor("op_37228_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3595_cast_fp16 = mul(x = var_37227_cast_fp16, y = var_37228_to_fp16)[name = tensor("aw_chunk_3595_cast_fp16")]; + tensor var_37231_equation_0 = const()[name = tensor("op_37231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37231_cast_fp16 = einsum(equation = var_37231_equation_0, values = (var_36957_cast_fp16, var_36629_cast_fp16))[name = tensor("op_37231_cast_fp16")]; + tensor var_37232_to_fp16 = const()[name = tensor("op_37232_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3597_cast_fp16 = mul(x = var_37231_cast_fp16, y = var_37232_to_fp16)[name = tensor("aw_chunk_3597_cast_fp16")]; + tensor var_37235_equation_0 = const()[name = tensor("op_37235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37235_cast_fp16 = einsum(equation = var_37235_equation_0, values = (var_36957_cast_fp16, var_36636_cast_fp16))[name = tensor("op_37235_cast_fp16")]; + tensor var_37236_to_fp16 = const()[name = tensor("op_37236_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3599_cast_fp16 = mul(x = var_37235_cast_fp16, y = var_37236_to_fp16)[name = tensor("aw_chunk_3599_cast_fp16")]; + tensor var_37239_equation_0 = const()[name = tensor("op_37239_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37239_cast_fp16 = einsum(equation = var_37239_equation_0, values = (var_36961_cast_fp16, var_36643_cast_fp16))[name = tensor("op_37239_cast_fp16")]; + tensor var_37240_to_fp16 = const()[name = tensor("op_37240_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3601_cast_fp16 = mul(x = var_37239_cast_fp16, y = var_37240_to_fp16)[name = tensor("aw_chunk_3601_cast_fp16")]; + tensor var_37243_equation_0 = const()[name = tensor("op_37243_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37243_cast_fp16 = einsum(equation = var_37243_equation_0, values = (var_36961_cast_fp16, var_36650_cast_fp16))[name = tensor("op_37243_cast_fp16")]; + tensor var_37244_to_fp16 = const()[name = tensor("op_37244_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3603_cast_fp16 = mul(x = var_37243_cast_fp16, y = var_37244_to_fp16)[name = tensor("aw_chunk_3603_cast_fp16")]; + tensor var_37247_equation_0 = const()[name = tensor("op_37247_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37247_cast_fp16 = einsum(equation = var_37247_equation_0, values = (var_36961_cast_fp16, var_36657_cast_fp16))[name = tensor("op_37247_cast_fp16")]; + tensor var_37248_to_fp16 = const()[name = tensor("op_37248_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3605_cast_fp16 = mul(x = var_37247_cast_fp16, y = var_37248_to_fp16)[name = tensor("aw_chunk_3605_cast_fp16")]; + tensor var_37251_equation_0 = const()[name = tensor("op_37251_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37251_cast_fp16 = einsum(equation = var_37251_equation_0, values = (var_36961_cast_fp16, var_36664_cast_fp16))[name = tensor("op_37251_cast_fp16")]; + tensor var_37252_to_fp16 = const()[name = tensor("op_37252_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3607_cast_fp16 = mul(x = var_37251_cast_fp16, y = var_37252_to_fp16)[name = tensor("aw_chunk_3607_cast_fp16")]; + tensor var_37255_equation_0 = const()[name = tensor("op_37255_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37255_cast_fp16 = einsum(equation = var_37255_equation_0, values = (var_36965_cast_fp16, var_36671_cast_fp16))[name = tensor("op_37255_cast_fp16")]; + tensor var_37256_to_fp16 = const()[name = tensor("op_37256_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3609_cast_fp16 = mul(x = var_37255_cast_fp16, y = var_37256_to_fp16)[name = tensor("aw_chunk_3609_cast_fp16")]; + tensor var_37259_equation_0 = const()[name = tensor("op_37259_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37259_cast_fp16 = einsum(equation = var_37259_equation_0, values = (var_36965_cast_fp16, var_36678_cast_fp16))[name = tensor("op_37259_cast_fp16")]; + tensor var_37260_to_fp16 = const()[name = tensor("op_37260_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3611_cast_fp16 = mul(x = var_37259_cast_fp16, y = var_37260_to_fp16)[name = tensor("aw_chunk_3611_cast_fp16")]; + tensor var_37263_equation_0 = const()[name = tensor("op_37263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37263_cast_fp16 = einsum(equation = var_37263_equation_0, values = (var_36965_cast_fp16, var_36685_cast_fp16))[name = tensor("op_37263_cast_fp16")]; + tensor var_37264_to_fp16 = const()[name = tensor("op_37264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3613_cast_fp16 = mul(x = var_37263_cast_fp16, y = var_37264_to_fp16)[name = tensor("aw_chunk_3613_cast_fp16")]; + tensor var_37267_equation_0 = const()[name = tensor("op_37267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37267_cast_fp16 = einsum(equation = var_37267_equation_0, values = (var_36965_cast_fp16, var_36692_cast_fp16))[name = tensor("op_37267_cast_fp16")]; + tensor var_37268_to_fp16 = const()[name = tensor("op_37268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3615_cast_fp16 = mul(x = var_37267_cast_fp16, y = var_37268_to_fp16)[name = tensor("aw_chunk_3615_cast_fp16")]; + tensor var_37271_equation_0 = const()[name = tensor("op_37271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37271_cast_fp16 = einsum(equation = var_37271_equation_0, values = (var_36969_cast_fp16, var_36699_cast_fp16))[name = tensor("op_37271_cast_fp16")]; + tensor var_37272_to_fp16 = const()[name = tensor("op_37272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3617_cast_fp16 = mul(x = var_37271_cast_fp16, y = var_37272_to_fp16)[name = tensor("aw_chunk_3617_cast_fp16")]; + tensor var_37275_equation_0 = const()[name = tensor("op_37275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37275_cast_fp16 = einsum(equation = var_37275_equation_0, values = (var_36969_cast_fp16, var_36706_cast_fp16))[name = tensor("op_37275_cast_fp16")]; + tensor var_37276_to_fp16 = const()[name = tensor("op_37276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3619_cast_fp16 = mul(x = var_37275_cast_fp16, y = var_37276_to_fp16)[name = tensor("aw_chunk_3619_cast_fp16")]; + tensor var_37279_equation_0 = const()[name = tensor("op_37279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37279_cast_fp16 = einsum(equation = var_37279_equation_0, values = (var_36969_cast_fp16, var_36713_cast_fp16))[name = tensor("op_37279_cast_fp16")]; + tensor var_37280_to_fp16 = const()[name = tensor("op_37280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3621_cast_fp16 = mul(x = var_37279_cast_fp16, y = var_37280_to_fp16)[name = tensor("aw_chunk_3621_cast_fp16")]; + tensor var_37283_equation_0 = const()[name = tensor("op_37283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37283_cast_fp16 = einsum(equation = var_37283_equation_0, values = (var_36969_cast_fp16, var_36720_cast_fp16))[name = tensor("op_37283_cast_fp16")]; + tensor var_37284_to_fp16 = const()[name = tensor("op_37284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3623_cast_fp16 = mul(x = var_37283_cast_fp16, y = var_37284_to_fp16)[name = tensor("aw_chunk_3623_cast_fp16")]; + tensor var_37287_equation_0 = const()[name = tensor("op_37287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37287_cast_fp16 = einsum(equation = var_37287_equation_0, values = (var_36973_cast_fp16, var_36727_cast_fp16))[name = tensor("op_37287_cast_fp16")]; + tensor var_37288_to_fp16 = const()[name = tensor("op_37288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3625_cast_fp16 = mul(x = var_37287_cast_fp16, y = var_37288_to_fp16)[name = tensor("aw_chunk_3625_cast_fp16")]; + tensor var_37291_equation_0 = const()[name = tensor("op_37291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37291_cast_fp16 = einsum(equation = var_37291_equation_0, values = (var_36973_cast_fp16, var_36734_cast_fp16))[name = tensor("op_37291_cast_fp16")]; + tensor var_37292_to_fp16 = const()[name = tensor("op_37292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3627_cast_fp16 = mul(x = var_37291_cast_fp16, y = var_37292_to_fp16)[name = tensor("aw_chunk_3627_cast_fp16")]; + tensor var_37295_equation_0 = const()[name = tensor("op_37295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37295_cast_fp16 = einsum(equation = var_37295_equation_0, values = (var_36973_cast_fp16, var_36741_cast_fp16))[name = tensor("op_37295_cast_fp16")]; + tensor var_37296_to_fp16 = const()[name = tensor("op_37296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3629_cast_fp16 = mul(x = var_37295_cast_fp16, y = var_37296_to_fp16)[name = tensor("aw_chunk_3629_cast_fp16")]; + tensor var_37299_equation_0 = const()[name = tensor("op_37299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37299_cast_fp16 = einsum(equation = var_37299_equation_0, values = (var_36973_cast_fp16, var_36748_cast_fp16))[name = tensor("op_37299_cast_fp16")]; + tensor var_37300_to_fp16 = const()[name = tensor("op_37300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3631_cast_fp16 = mul(x = var_37299_cast_fp16, y = var_37300_to_fp16)[name = tensor("aw_chunk_3631_cast_fp16")]; + tensor var_37303_equation_0 = const()[name = tensor("op_37303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37303_cast_fp16 = einsum(equation = var_37303_equation_0, values = (var_36977_cast_fp16, var_36755_cast_fp16))[name = tensor("op_37303_cast_fp16")]; + tensor var_37304_to_fp16 = const()[name = tensor("op_37304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3633_cast_fp16 = mul(x = var_37303_cast_fp16, y = var_37304_to_fp16)[name = tensor("aw_chunk_3633_cast_fp16")]; + tensor var_37307_equation_0 = const()[name = tensor("op_37307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37307_cast_fp16 = einsum(equation = var_37307_equation_0, values = (var_36977_cast_fp16, var_36762_cast_fp16))[name = tensor("op_37307_cast_fp16")]; + tensor var_37308_to_fp16 = const()[name = tensor("op_37308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3635_cast_fp16 = mul(x = var_37307_cast_fp16, y = var_37308_to_fp16)[name = tensor("aw_chunk_3635_cast_fp16")]; + tensor var_37311_equation_0 = const()[name = tensor("op_37311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37311_cast_fp16 = einsum(equation = var_37311_equation_0, values = (var_36977_cast_fp16, var_36769_cast_fp16))[name = tensor("op_37311_cast_fp16")]; + tensor var_37312_to_fp16 = const()[name = tensor("op_37312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3637_cast_fp16 = mul(x = var_37311_cast_fp16, y = var_37312_to_fp16)[name = tensor("aw_chunk_3637_cast_fp16")]; + tensor var_37315_equation_0 = const()[name = tensor("op_37315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37315_cast_fp16 = einsum(equation = var_37315_equation_0, values = (var_36977_cast_fp16, var_36776_cast_fp16))[name = tensor("op_37315_cast_fp16")]; + tensor var_37316_to_fp16 = const()[name = tensor("op_37316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3639_cast_fp16 = mul(x = var_37315_cast_fp16, y = var_37316_to_fp16)[name = tensor("aw_chunk_3639_cast_fp16")]; + tensor var_37319_equation_0 = const()[name = tensor("op_37319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37319_cast_fp16 = einsum(equation = var_37319_equation_0, values = (var_36981_cast_fp16, var_36783_cast_fp16))[name = tensor("op_37319_cast_fp16")]; + tensor var_37320_to_fp16 = const()[name = tensor("op_37320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3641_cast_fp16 = mul(x = var_37319_cast_fp16, y = var_37320_to_fp16)[name = tensor("aw_chunk_3641_cast_fp16")]; + tensor var_37323_equation_0 = const()[name = tensor("op_37323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37323_cast_fp16 = einsum(equation = var_37323_equation_0, values = (var_36981_cast_fp16, var_36790_cast_fp16))[name = tensor("op_37323_cast_fp16")]; + tensor var_37324_to_fp16 = const()[name = tensor("op_37324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3643_cast_fp16 = mul(x = var_37323_cast_fp16, y = var_37324_to_fp16)[name = tensor("aw_chunk_3643_cast_fp16")]; + tensor var_37327_equation_0 = const()[name = tensor("op_37327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37327_cast_fp16 = einsum(equation = var_37327_equation_0, values = (var_36981_cast_fp16, var_36797_cast_fp16))[name = tensor("op_37327_cast_fp16")]; + tensor var_37328_to_fp16 = const()[name = tensor("op_37328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3645_cast_fp16 = mul(x = var_37327_cast_fp16, y = var_37328_to_fp16)[name = tensor("aw_chunk_3645_cast_fp16")]; + tensor var_37331_equation_0 = const()[name = tensor("op_37331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37331_cast_fp16 = einsum(equation = var_37331_equation_0, values = (var_36981_cast_fp16, var_36804_cast_fp16))[name = tensor("op_37331_cast_fp16")]; + tensor var_37332_to_fp16 = const()[name = tensor("op_37332_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3647_cast_fp16 = mul(x = var_37331_cast_fp16, y = var_37332_to_fp16)[name = tensor("aw_chunk_3647_cast_fp16")]; + tensor var_37335_equation_0 = const()[name = tensor("op_37335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37335_cast_fp16 = einsum(equation = var_37335_equation_0, values = (var_36985_cast_fp16, var_36811_cast_fp16))[name = tensor("op_37335_cast_fp16")]; + tensor var_37336_to_fp16 = const()[name = tensor("op_37336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3649_cast_fp16 = mul(x = var_37335_cast_fp16, y = var_37336_to_fp16)[name = tensor("aw_chunk_3649_cast_fp16")]; + tensor var_37339_equation_0 = const()[name = tensor("op_37339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37339_cast_fp16 = einsum(equation = var_37339_equation_0, values = (var_36985_cast_fp16, var_36818_cast_fp16))[name = tensor("op_37339_cast_fp16")]; + tensor var_37340_to_fp16 = const()[name = tensor("op_37340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3651_cast_fp16 = mul(x = var_37339_cast_fp16, y = var_37340_to_fp16)[name = tensor("aw_chunk_3651_cast_fp16")]; + tensor var_37343_equation_0 = const()[name = tensor("op_37343_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37343_cast_fp16 = einsum(equation = var_37343_equation_0, values = (var_36985_cast_fp16, var_36825_cast_fp16))[name = tensor("op_37343_cast_fp16")]; + tensor var_37344_to_fp16 = const()[name = tensor("op_37344_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3653_cast_fp16 = mul(x = var_37343_cast_fp16, y = var_37344_to_fp16)[name = tensor("aw_chunk_3653_cast_fp16")]; + tensor var_37347_equation_0 = const()[name = tensor("op_37347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37347_cast_fp16 = einsum(equation = var_37347_equation_0, values = (var_36985_cast_fp16, var_36832_cast_fp16))[name = tensor("op_37347_cast_fp16")]; + tensor var_37348_to_fp16 = const()[name = tensor("op_37348_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3655_cast_fp16 = mul(x = var_37347_cast_fp16, y = var_37348_to_fp16)[name = tensor("aw_chunk_3655_cast_fp16")]; + tensor var_37351_equation_0 = const()[name = tensor("op_37351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37351_cast_fp16 = einsum(equation = var_37351_equation_0, values = (var_36989_cast_fp16, var_36839_cast_fp16))[name = tensor("op_37351_cast_fp16")]; + tensor var_37352_to_fp16 = const()[name = tensor("op_37352_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3657_cast_fp16 = mul(x = var_37351_cast_fp16, y = var_37352_to_fp16)[name = tensor("aw_chunk_3657_cast_fp16")]; + tensor var_37355_equation_0 = const()[name = tensor("op_37355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37355_cast_fp16 = einsum(equation = var_37355_equation_0, values = (var_36989_cast_fp16, var_36846_cast_fp16))[name = tensor("op_37355_cast_fp16")]; + tensor var_37356_to_fp16 = const()[name = tensor("op_37356_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3659_cast_fp16 = mul(x = var_37355_cast_fp16, y = var_37356_to_fp16)[name = tensor("aw_chunk_3659_cast_fp16")]; + tensor var_37359_equation_0 = const()[name = tensor("op_37359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37359_cast_fp16 = einsum(equation = var_37359_equation_0, values = (var_36989_cast_fp16, var_36853_cast_fp16))[name = tensor("op_37359_cast_fp16")]; + tensor var_37360_to_fp16 = const()[name = tensor("op_37360_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3661_cast_fp16 = mul(x = var_37359_cast_fp16, y = var_37360_to_fp16)[name = tensor("aw_chunk_3661_cast_fp16")]; + tensor var_37363_equation_0 = const()[name = tensor("op_37363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37363_cast_fp16 = einsum(equation = var_37363_equation_0, values = (var_36989_cast_fp16, var_36860_cast_fp16))[name = tensor("op_37363_cast_fp16")]; + tensor var_37364_to_fp16 = const()[name = tensor("op_37364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3663_cast_fp16 = mul(x = var_37363_cast_fp16, y = var_37364_to_fp16)[name = tensor("aw_chunk_3663_cast_fp16")]; + tensor var_37367_equation_0 = const()[name = tensor("op_37367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37367_cast_fp16 = einsum(equation = var_37367_equation_0, values = (var_36993_cast_fp16, var_36867_cast_fp16))[name = tensor("op_37367_cast_fp16")]; + tensor var_37368_to_fp16 = const()[name = tensor("op_37368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3665_cast_fp16 = mul(x = var_37367_cast_fp16, y = var_37368_to_fp16)[name = tensor("aw_chunk_3665_cast_fp16")]; + tensor var_37371_equation_0 = const()[name = tensor("op_37371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37371_cast_fp16 = einsum(equation = var_37371_equation_0, values = (var_36993_cast_fp16, var_36874_cast_fp16))[name = tensor("op_37371_cast_fp16")]; + tensor var_37372_to_fp16 = const()[name = tensor("op_37372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3667_cast_fp16 = mul(x = var_37371_cast_fp16, y = var_37372_to_fp16)[name = tensor("aw_chunk_3667_cast_fp16")]; + tensor var_37375_equation_0 = const()[name = tensor("op_37375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37375_cast_fp16 = einsum(equation = var_37375_equation_0, values = (var_36993_cast_fp16, var_36881_cast_fp16))[name = tensor("op_37375_cast_fp16")]; + tensor var_37376_to_fp16 = const()[name = tensor("op_37376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3669_cast_fp16 = mul(x = var_37375_cast_fp16, y = var_37376_to_fp16)[name = tensor("aw_chunk_3669_cast_fp16")]; + tensor var_37379_equation_0 = const()[name = tensor("op_37379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37379_cast_fp16 = einsum(equation = var_37379_equation_0, values = (var_36993_cast_fp16, var_36888_cast_fp16))[name = tensor("op_37379_cast_fp16")]; + tensor var_37380_to_fp16 = const()[name = tensor("op_37380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3671_cast_fp16 = mul(x = var_37379_cast_fp16, y = var_37380_to_fp16)[name = tensor("aw_chunk_3671_cast_fp16")]; + tensor var_37383_equation_0 = const()[name = tensor("op_37383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37383_cast_fp16 = einsum(equation = var_37383_equation_0, values = (var_36997_cast_fp16, var_36895_cast_fp16))[name = tensor("op_37383_cast_fp16")]; + tensor var_37384_to_fp16 = const()[name = tensor("op_37384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3673_cast_fp16 = mul(x = var_37383_cast_fp16, y = var_37384_to_fp16)[name = tensor("aw_chunk_3673_cast_fp16")]; + tensor var_37387_equation_0 = const()[name = tensor("op_37387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37387_cast_fp16 = einsum(equation = var_37387_equation_0, values = (var_36997_cast_fp16, var_36902_cast_fp16))[name = tensor("op_37387_cast_fp16")]; + tensor var_37388_to_fp16 = const()[name = tensor("op_37388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3675_cast_fp16 = mul(x = var_37387_cast_fp16, y = var_37388_to_fp16)[name = tensor("aw_chunk_3675_cast_fp16")]; + tensor var_37391_equation_0 = const()[name = tensor("op_37391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37391_cast_fp16 = einsum(equation = var_37391_equation_0, values = (var_36997_cast_fp16, var_36909_cast_fp16))[name = tensor("op_37391_cast_fp16")]; + tensor var_37392_to_fp16 = const()[name = tensor("op_37392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3677_cast_fp16 = mul(x = var_37391_cast_fp16, y = var_37392_to_fp16)[name = tensor("aw_chunk_3677_cast_fp16")]; + tensor var_37395_equation_0 = const()[name = tensor("op_37395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_37395_cast_fp16 = einsum(equation = var_37395_equation_0, values = (var_36997_cast_fp16, var_36916_cast_fp16))[name = tensor("op_37395_cast_fp16")]; + tensor var_37396_to_fp16 = const()[name = tensor("op_37396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3679_cast_fp16 = mul(x = var_37395_cast_fp16, y = var_37396_to_fp16)[name = tensor("aw_chunk_3679_cast_fp16")]; + tensor var_37398_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3521_cast_fp16)[name = tensor("op_37398_cast_fp16")]; + tensor var_37399_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3523_cast_fp16)[name = tensor("op_37399_cast_fp16")]; + tensor var_37400_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3525_cast_fp16)[name = tensor("op_37400_cast_fp16")]; + tensor var_37401_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3527_cast_fp16)[name = tensor("op_37401_cast_fp16")]; + tensor var_37402_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3529_cast_fp16)[name = tensor("op_37402_cast_fp16")]; + tensor var_37403_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3531_cast_fp16)[name = tensor("op_37403_cast_fp16")]; + tensor var_37404_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3533_cast_fp16)[name = tensor("op_37404_cast_fp16")]; + tensor var_37405_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3535_cast_fp16)[name = tensor("op_37405_cast_fp16")]; + tensor var_37406_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3537_cast_fp16)[name = tensor("op_37406_cast_fp16")]; + tensor var_37407_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3539_cast_fp16)[name = tensor("op_37407_cast_fp16")]; + tensor var_37408_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3541_cast_fp16)[name = tensor("op_37408_cast_fp16")]; + tensor var_37409_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3543_cast_fp16)[name = tensor("op_37409_cast_fp16")]; + tensor var_37410_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3545_cast_fp16)[name = tensor("op_37410_cast_fp16")]; + tensor var_37411_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3547_cast_fp16)[name = tensor("op_37411_cast_fp16")]; + tensor var_37412_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3549_cast_fp16)[name = tensor("op_37412_cast_fp16")]; + tensor var_37413_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3551_cast_fp16)[name = tensor("op_37413_cast_fp16")]; + tensor var_37414_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3553_cast_fp16)[name = tensor("op_37414_cast_fp16")]; + tensor var_37415_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3555_cast_fp16)[name = tensor("op_37415_cast_fp16")]; + tensor var_37416_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3557_cast_fp16)[name = tensor("op_37416_cast_fp16")]; + tensor var_37417_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3559_cast_fp16)[name = tensor("op_37417_cast_fp16")]; + tensor var_37418_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3561_cast_fp16)[name = tensor("op_37418_cast_fp16")]; + tensor var_37419_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3563_cast_fp16)[name = tensor("op_37419_cast_fp16")]; + tensor var_37420_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3565_cast_fp16)[name = tensor("op_37420_cast_fp16")]; + tensor var_37421_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3567_cast_fp16)[name = tensor("op_37421_cast_fp16")]; + tensor var_37422_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3569_cast_fp16)[name = tensor("op_37422_cast_fp16")]; + tensor var_37423_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3571_cast_fp16)[name = tensor("op_37423_cast_fp16")]; + tensor var_37424_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3573_cast_fp16)[name = tensor("op_37424_cast_fp16")]; + tensor var_37425_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3575_cast_fp16)[name = tensor("op_37425_cast_fp16")]; + tensor var_37426_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3577_cast_fp16)[name = tensor("op_37426_cast_fp16")]; + tensor var_37427_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3579_cast_fp16)[name = tensor("op_37427_cast_fp16")]; + tensor var_37428_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3581_cast_fp16)[name = tensor("op_37428_cast_fp16")]; + tensor var_37429_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3583_cast_fp16)[name = tensor("op_37429_cast_fp16")]; + tensor var_37430_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3585_cast_fp16)[name = tensor("op_37430_cast_fp16")]; + tensor var_37431_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3587_cast_fp16)[name = tensor("op_37431_cast_fp16")]; + tensor var_37432_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3589_cast_fp16)[name = tensor("op_37432_cast_fp16")]; + tensor var_37433_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3591_cast_fp16)[name = tensor("op_37433_cast_fp16")]; + tensor var_37434_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3593_cast_fp16)[name = tensor("op_37434_cast_fp16")]; + tensor var_37435_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3595_cast_fp16)[name = tensor("op_37435_cast_fp16")]; + tensor var_37436_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3597_cast_fp16)[name = tensor("op_37436_cast_fp16")]; + tensor var_37437_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3599_cast_fp16)[name = tensor("op_37437_cast_fp16")]; + tensor var_37438_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3601_cast_fp16)[name = tensor("op_37438_cast_fp16")]; + tensor var_37439_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3603_cast_fp16)[name = tensor("op_37439_cast_fp16")]; + tensor var_37440_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3605_cast_fp16)[name = tensor("op_37440_cast_fp16")]; + tensor var_37441_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3607_cast_fp16)[name = tensor("op_37441_cast_fp16")]; + tensor var_37442_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3609_cast_fp16)[name = tensor("op_37442_cast_fp16")]; + tensor var_37443_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3611_cast_fp16)[name = tensor("op_37443_cast_fp16")]; + tensor var_37444_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3613_cast_fp16)[name = tensor("op_37444_cast_fp16")]; + tensor var_37445_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3615_cast_fp16)[name = tensor("op_37445_cast_fp16")]; + tensor var_37446_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3617_cast_fp16)[name = tensor("op_37446_cast_fp16")]; + tensor var_37447_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3619_cast_fp16)[name = tensor("op_37447_cast_fp16")]; + tensor var_37448_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3621_cast_fp16)[name = tensor("op_37448_cast_fp16")]; + tensor var_37449_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3623_cast_fp16)[name = tensor("op_37449_cast_fp16")]; + tensor var_37450_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3625_cast_fp16)[name = tensor("op_37450_cast_fp16")]; + tensor var_37451_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3627_cast_fp16)[name = tensor("op_37451_cast_fp16")]; + tensor var_37452_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3629_cast_fp16)[name = tensor("op_37452_cast_fp16")]; + tensor var_37453_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3631_cast_fp16)[name = tensor("op_37453_cast_fp16")]; + tensor var_37454_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3633_cast_fp16)[name = tensor("op_37454_cast_fp16")]; + tensor var_37455_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3635_cast_fp16)[name = tensor("op_37455_cast_fp16")]; + tensor var_37456_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3637_cast_fp16)[name = tensor("op_37456_cast_fp16")]; + tensor var_37457_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3639_cast_fp16)[name = tensor("op_37457_cast_fp16")]; + tensor var_37458_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3641_cast_fp16)[name = tensor("op_37458_cast_fp16")]; + tensor var_37459_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3643_cast_fp16)[name = tensor("op_37459_cast_fp16")]; + tensor var_37460_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3645_cast_fp16)[name = tensor("op_37460_cast_fp16")]; + tensor var_37461_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3647_cast_fp16)[name = tensor("op_37461_cast_fp16")]; + tensor var_37462_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3649_cast_fp16)[name = tensor("op_37462_cast_fp16")]; + tensor var_37463_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3651_cast_fp16)[name = tensor("op_37463_cast_fp16")]; + tensor var_37464_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3653_cast_fp16)[name = tensor("op_37464_cast_fp16")]; + tensor var_37465_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3655_cast_fp16)[name = tensor("op_37465_cast_fp16")]; + tensor var_37466_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3657_cast_fp16)[name = tensor("op_37466_cast_fp16")]; + tensor var_37467_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3659_cast_fp16)[name = tensor("op_37467_cast_fp16")]; + tensor var_37468_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3661_cast_fp16)[name = tensor("op_37468_cast_fp16")]; + tensor var_37469_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3663_cast_fp16)[name = tensor("op_37469_cast_fp16")]; + tensor var_37470_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3665_cast_fp16)[name = tensor("op_37470_cast_fp16")]; + tensor var_37471_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3667_cast_fp16)[name = tensor("op_37471_cast_fp16")]; + tensor var_37472_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3669_cast_fp16)[name = tensor("op_37472_cast_fp16")]; + tensor var_37473_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3671_cast_fp16)[name = tensor("op_37473_cast_fp16")]; + tensor var_37474_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3673_cast_fp16)[name = tensor("op_37474_cast_fp16")]; + tensor var_37475_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3675_cast_fp16)[name = tensor("op_37475_cast_fp16")]; + tensor var_37476_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3677_cast_fp16)[name = tensor("op_37476_cast_fp16")]; + tensor var_37477_cast_fp16 = softmax(axis = var_36169, x = aw_chunk_3679_cast_fp16)[name = tensor("op_37477_cast_fp16")]; + tensor var_37479_equation_0 = const()[name = tensor("op_37479_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37479_cast_fp16 = einsum(equation = var_37479_equation_0, values = (var_36999_cast_fp16, var_37398_cast_fp16))[name = tensor("op_37479_cast_fp16")]; + tensor var_37481_equation_0 = const()[name = tensor("op_37481_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37481_cast_fp16 = einsum(equation = var_37481_equation_0, values = (var_36999_cast_fp16, var_37399_cast_fp16))[name = tensor("op_37481_cast_fp16")]; + tensor var_37483_equation_0 = const()[name = tensor("op_37483_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37483_cast_fp16 = einsum(equation = var_37483_equation_0, values = (var_36999_cast_fp16, var_37400_cast_fp16))[name = tensor("op_37483_cast_fp16")]; + tensor var_37485_equation_0 = const()[name = tensor("op_37485_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37485_cast_fp16 = einsum(equation = var_37485_equation_0, values = (var_36999_cast_fp16, var_37401_cast_fp16))[name = tensor("op_37485_cast_fp16")]; + tensor var_37487_equation_0 = const()[name = tensor("op_37487_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37487_cast_fp16 = einsum(equation = var_37487_equation_0, values = (var_37003_cast_fp16, var_37402_cast_fp16))[name = tensor("op_37487_cast_fp16")]; + tensor var_37489_equation_0 = const()[name = tensor("op_37489_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37489_cast_fp16 = einsum(equation = var_37489_equation_0, values = (var_37003_cast_fp16, var_37403_cast_fp16))[name = tensor("op_37489_cast_fp16")]; + tensor var_37491_equation_0 = const()[name = tensor("op_37491_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37491_cast_fp16 = einsum(equation = var_37491_equation_0, values = (var_37003_cast_fp16, var_37404_cast_fp16))[name = tensor("op_37491_cast_fp16")]; + tensor var_37493_equation_0 = const()[name = tensor("op_37493_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37493_cast_fp16 = einsum(equation = var_37493_equation_0, values = (var_37003_cast_fp16, var_37405_cast_fp16))[name = tensor("op_37493_cast_fp16")]; + tensor var_37495_equation_0 = const()[name = tensor("op_37495_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37495_cast_fp16 = einsum(equation = var_37495_equation_0, values = (var_37007_cast_fp16, var_37406_cast_fp16))[name = tensor("op_37495_cast_fp16")]; + tensor var_37497_equation_0 = const()[name = tensor("op_37497_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37497_cast_fp16 = einsum(equation = var_37497_equation_0, values = (var_37007_cast_fp16, var_37407_cast_fp16))[name = tensor("op_37497_cast_fp16")]; + tensor var_37499_equation_0 = const()[name = tensor("op_37499_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37499_cast_fp16 = einsum(equation = var_37499_equation_0, values = (var_37007_cast_fp16, var_37408_cast_fp16))[name = tensor("op_37499_cast_fp16")]; + tensor var_37501_equation_0 = const()[name = tensor("op_37501_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37501_cast_fp16 = einsum(equation = var_37501_equation_0, values = (var_37007_cast_fp16, var_37409_cast_fp16))[name = tensor("op_37501_cast_fp16")]; + tensor var_37503_equation_0 = const()[name = tensor("op_37503_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37503_cast_fp16 = einsum(equation = var_37503_equation_0, values = (var_37011_cast_fp16, var_37410_cast_fp16))[name = tensor("op_37503_cast_fp16")]; + tensor var_37505_equation_0 = const()[name = tensor("op_37505_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37505_cast_fp16 = einsum(equation = var_37505_equation_0, values = (var_37011_cast_fp16, var_37411_cast_fp16))[name = tensor("op_37505_cast_fp16")]; + tensor var_37507_equation_0 = const()[name = tensor("op_37507_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37507_cast_fp16 = einsum(equation = var_37507_equation_0, values = (var_37011_cast_fp16, var_37412_cast_fp16))[name = tensor("op_37507_cast_fp16")]; + tensor var_37509_equation_0 = const()[name = tensor("op_37509_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37509_cast_fp16 = einsum(equation = var_37509_equation_0, values = (var_37011_cast_fp16, var_37413_cast_fp16))[name = tensor("op_37509_cast_fp16")]; + tensor var_37511_equation_0 = const()[name = tensor("op_37511_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37511_cast_fp16 = einsum(equation = var_37511_equation_0, values = (var_37015_cast_fp16, var_37414_cast_fp16))[name = tensor("op_37511_cast_fp16")]; + tensor var_37513_equation_0 = const()[name = tensor("op_37513_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37513_cast_fp16 = einsum(equation = var_37513_equation_0, values = (var_37015_cast_fp16, var_37415_cast_fp16))[name = tensor("op_37513_cast_fp16")]; + tensor var_37515_equation_0 = const()[name = tensor("op_37515_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37515_cast_fp16 = einsum(equation = var_37515_equation_0, values = (var_37015_cast_fp16, var_37416_cast_fp16))[name = tensor("op_37515_cast_fp16")]; + tensor var_37517_equation_0 = const()[name = tensor("op_37517_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37517_cast_fp16 = einsum(equation = var_37517_equation_0, values = (var_37015_cast_fp16, var_37417_cast_fp16))[name = tensor("op_37517_cast_fp16")]; + tensor var_37519_equation_0 = const()[name = tensor("op_37519_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37519_cast_fp16 = einsum(equation = var_37519_equation_0, values = (var_37019_cast_fp16, var_37418_cast_fp16))[name = tensor("op_37519_cast_fp16")]; + tensor var_37521_equation_0 = const()[name = tensor("op_37521_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37521_cast_fp16 = einsum(equation = var_37521_equation_0, values = (var_37019_cast_fp16, var_37419_cast_fp16))[name = tensor("op_37521_cast_fp16")]; + tensor var_37523_equation_0 = const()[name = tensor("op_37523_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37523_cast_fp16 = einsum(equation = var_37523_equation_0, values = (var_37019_cast_fp16, var_37420_cast_fp16))[name = tensor("op_37523_cast_fp16")]; + tensor var_37525_equation_0 = const()[name = tensor("op_37525_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37525_cast_fp16 = einsum(equation = var_37525_equation_0, values = (var_37019_cast_fp16, var_37421_cast_fp16))[name = tensor("op_37525_cast_fp16")]; + tensor var_37527_equation_0 = const()[name = tensor("op_37527_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37527_cast_fp16 = einsum(equation = var_37527_equation_0, values = (var_37023_cast_fp16, var_37422_cast_fp16))[name = tensor("op_37527_cast_fp16")]; + tensor var_37529_equation_0 = const()[name = tensor("op_37529_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37529_cast_fp16 = einsum(equation = var_37529_equation_0, values = (var_37023_cast_fp16, var_37423_cast_fp16))[name = tensor("op_37529_cast_fp16")]; + tensor var_37531_equation_0 = const()[name = tensor("op_37531_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37531_cast_fp16 = einsum(equation = var_37531_equation_0, values = (var_37023_cast_fp16, var_37424_cast_fp16))[name = tensor("op_37531_cast_fp16")]; + tensor var_37533_equation_0 = const()[name = tensor("op_37533_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37533_cast_fp16 = einsum(equation = var_37533_equation_0, values = (var_37023_cast_fp16, var_37425_cast_fp16))[name = tensor("op_37533_cast_fp16")]; + tensor var_37535_equation_0 = const()[name = tensor("op_37535_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37535_cast_fp16 = einsum(equation = var_37535_equation_0, values = (var_37027_cast_fp16, var_37426_cast_fp16))[name = tensor("op_37535_cast_fp16")]; + tensor var_37537_equation_0 = const()[name = tensor("op_37537_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37537_cast_fp16 = einsum(equation = var_37537_equation_0, values = (var_37027_cast_fp16, var_37427_cast_fp16))[name = tensor("op_37537_cast_fp16")]; + tensor var_37539_equation_0 = const()[name = tensor("op_37539_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37539_cast_fp16 = einsum(equation = var_37539_equation_0, values = (var_37027_cast_fp16, var_37428_cast_fp16))[name = tensor("op_37539_cast_fp16")]; + tensor var_37541_equation_0 = const()[name = tensor("op_37541_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37541_cast_fp16 = einsum(equation = var_37541_equation_0, values = (var_37027_cast_fp16, var_37429_cast_fp16))[name = tensor("op_37541_cast_fp16")]; + tensor var_37543_equation_0 = const()[name = tensor("op_37543_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37543_cast_fp16 = einsum(equation = var_37543_equation_0, values = (var_37031_cast_fp16, var_37430_cast_fp16))[name = tensor("op_37543_cast_fp16")]; + tensor var_37545_equation_0 = const()[name = tensor("op_37545_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37545_cast_fp16 = einsum(equation = var_37545_equation_0, values = (var_37031_cast_fp16, var_37431_cast_fp16))[name = tensor("op_37545_cast_fp16")]; + tensor var_37547_equation_0 = const()[name = tensor("op_37547_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37547_cast_fp16 = einsum(equation = var_37547_equation_0, values = (var_37031_cast_fp16, var_37432_cast_fp16))[name = tensor("op_37547_cast_fp16")]; + tensor var_37549_equation_0 = const()[name = tensor("op_37549_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37549_cast_fp16 = einsum(equation = var_37549_equation_0, values = (var_37031_cast_fp16, var_37433_cast_fp16))[name = tensor("op_37549_cast_fp16")]; + tensor var_37551_equation_0 = const()[name = tensor("op_37551_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37551_cast_fp16 = einsum(equation = var_37551_equation_0, values = (var_37035_cast_fp16, var_37434_cast_fp16))[name = tensor("op_37551_cast_fp16")]; + tensor var_37553_equation_0 = const()[name = tensor("op_37553_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37553_cast_fp16 = einsum(equation = var_37553_equation_0, values = (var_37035_cast_fp16, var_37435_cast_fp16))[name = tensor("op_37553_cast_fp16")]; + tensor var_37555_equation_0 = const()[name = tensor("op_37555_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37555_cast_fp16 = einsum(equation = var_37555_equation_0, values = (var_37035_cast_fp16, var_37436_cast_fp16))[name = tensor("op_37555_cast_fp16")]; + tensor var_37557_equation_0 = const()[name = tensor("op_37557_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37557_cast_fp16 = einsum(equation = var_37557_equation_0, values = (var_37035_cast_fp16, var_37437_cast_fp16))[name = tensor("op_37557_cast_fp16")]; + tensor var_37559_equation_0 = const()[name = tensor("op_37559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37559_cast_fp16 = einsum(equation = var_37559_equation_0, values = (var_37039_cast_fp16, var_37438_cast_fp16))[name = tensor("op_37559_cast_fp16")]; + tensor var_37561_equation_0 = const()[name = tensor("op_37561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37561_cast_fp16 = einsum(equation = var_37561_equation_0, values = (var_37039_cast_fp16, var_37439_cast_fp16))[name = tensor("op_37561_cast_fp16")]; + tensor var_37563_equation_0 = const()[name = tensor("op_37563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37563_cast_fp16 = einsum(equation = var_37563_equation_0, values = (var_37039_cast_fp16, var_37440_cast_fp16))[name = tensor("op_37563_cast_fp16")]; + tensor var_37565_equation_0 = const()[name = tensor("op_37565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37565_cast_fp16 = einsum(equation = var_37565_equation_0, values = (var_37039_cast_fp16, var_37441_cast_fp16))[name = tensor("op_37565_cast_fp16")]; + tensor var_37567_equation_0 = const()[name = tensor("op_37567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37567_cast_fp16 = einsum(equation = var_37567_equation_0, values = (var_37043_cast_fp16, var_37442_cast_fp16))[name = tensor("op_37567_cast_fp16")]; + tensor var_37569_equation_0 = const()[name = tensor("op_37569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37569_cast_fp16 = einsum(equation = var_37569_equation_0, values = (var_37043_cast_fp16, var_37443_cast_fp16))[name = tensor("op_37569_cast_fp16")]; + tensor var_37571_equation_0 = const()[name = tensor("op_37571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37571_cast_fp16 = einsum(equation = var_37571_equation_0, values = (var_37043_cast_fp16, var_37444_cast_fp16))[name = tensor("op_37571_cast_fp16")]; + tensor var_37573_equation_0 = const()[name = tensor("op_37573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37573_cast_fp16 = einsum(equation = var_37573_equation_0, values = (var_37043_cast_fp16, var_37445_cast_fp16))[name = tensor("op_37573_cast_fp16")]; + tensor var_37575_equation_0 = const()[name = tensor("op_37575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37575_cast_fp16 = einsum(equation = var_37575_equation_0, values = (var_37047_cast_fp16, var_37446_cast_fp16))[name = tensor("op_37575_cast_fp16")]; + tensor var_37577_equation_0 = const()[name = tensor("op_37577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37577_cast_fp16 = einsum(equation = var_37577_equation_0, values = (var_37047_cast_fp16, var_37447_cast_fp16))[name = tensor("op_37577_cast_fp16")]; + tensor var_37579_equation_0 = const()[name = tensor("op_37579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37579_cast_fp16 = einsum(equation = var_37579_equation_0, values = (var_37047_cast_fp16, var_37448_cast_fp16))[name = tensor("op_37579_cast_fp16")]; + tensor var_37581_equation_0 = const()[name = tensor("op_37581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37581_cast_fp16 = einsum(equation = var_37581_equation_0, values = (var_37047_cast_fp16, var_37449_cast_fp16))[name = tensor("op_37581_cast_fp16")]; + tensor var_37583_equation_0 = const()[name = tensor("op_37583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37583_cast_fp16 = einsum(equation = var_37583_equation_0, values = (var_37051_cast_fp16, var_37450_cast_fp16))[name = tensor("op_37583_cast_fp16")]; + tensor var_37585_equation_0 = const()[name = tensor("op_37585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37585_cast_fp16 = einsum(equation = var_37585_equation_0, values = (var_37051_cast_fp16, var_37451_cast_fp16))[name = tensor("op_37585_cast_fp16")]; + tensor var_37587_equation_0 = const()[name = tensor("op_37587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37587_cast_fp16 = einsum(equation = var_37587_equation_0, values = (var_37051_cast_fp16, var_37452_cast_fp16))[name = tensor("op_37587_cast_fp16")]; + tensor var_37589_equation_0 = const()[name = tensor("op_37589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37589_cast_fp16 = einsum(equation = var_37589_equation_0, values = (var_37051_cast_fp16, var_37453_cast_fp16))[name = tensor("op_37589_cast_fp16")]; + tensor var_37591_equation_0 = const()[name = tensor("op_37591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37591_cast_fp16 = einsum(equation = var_37591_equation_0, values = (var_37055_cast_fp16, var_37454_cast_fp16))[name = tensor("op_37591_cast_fp16")]; + tensor var_37593_equation_0 = const()[name = tensor("op_37593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37593_cast_fp16 = einsum(equation = var_37593_equation_0, values = (var_37055_cast_fp16, var_37455_cast_fp16))[name = tensor("op_37593_cast_fp16")]; + tensor var_37595_equation_0 = const()[name = tensor("op_37595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37595_cast_fp16 = einsum(equation = var_37595_equation_0, values = (var_37055_cast_fp16, var_37456_cast_fp16))[name = tensor("op_37595_cast_fp16")]; + tensor var_37597_equation_0 = const()[name = tensor("op_37597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37597_cast_fp16 = einsum(equation = var_37597_equation_0, values = (var_37055_cast_fp16, var_37457_cast_fp16))[name = tensor("op_37597_cast_fp16")]; + tensor var_37599_equation_0 = const()[name = tensor("op_37599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37599_cast_fp16 = einsum(equation = var_37599_equation_0, values = (var_37059_cast_fp16, var_37458_cast_fp16))[name = tensor("op_37599_cast_fp16")]; + tensor var_37601_equation_0 = const()[name = tensor("op_37601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37601_cast_fp16 = einsum(equation = var_37601_equation_0, values = (var_37059_cast_fp16, var_37459_cast_fp16))[name = tensor("op_37601_cast_fp16")]; + tensor var_37603_equation_0 = const()[name = tensor("op_37603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37603_cast_fp16 = einsum(equation = var_37603_equation_0, values = (var_37059_cast_fp16, var_37460_cast_fp16))[name = tensor("op_37603_cast_fp16")]; + tensor var_37605_equation_0 = const()[name = tensor("op_37605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37605_cast_fp16 = einsum(equation = var_37605_equation_0, values = (var_37059_cast_fp16, var_37461_cast_fp16))[name = tensor("op_37605_cast_fp16")]; + tensor var_37607_equation_0 = const()[name = tensor("op_37607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37607_cast_fp16 = einsum(equation = var_37607_equation_0, values = (var_37063_cast_fp16, var_37462_cast_fp16))[name = tensor("op_37607_cast_fp16")]; + tensor var_37609_equation_0 = const()[name = tensor("op_37609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37609_cast_fp16 = einsum(equation = var_37609_equation_0, values = (var_37063_cast_fp16, var_37463_cast_fp16))[name = tensor("op_37609_cast_fp16")]; + tensor var_37611_equation_0 = const()[name = tensor("op_37611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37611_cast_fp16 = einsum(equation = var_37611_equation_0, values = (var_37063_cast_fp16, var_37464_cast_fp16))[name = tensor("op_37611_cast_fp16")]; + tensor var_37613_equation_0 = const()[name = tensor("op_37613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37613_cast_fp16 = einsum(equation = var_37613_equation_0, values = (var_37063_cast_fp16, var_37465_cast_fp16))[name = tensor("op_37613_cast_fp16")]; + tensor var_37615_equation_0 = const()[name = tensor("op_37615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37615_cast_fp16 = einsum(equation = var_37615_equation_0, values = (var_37067_cast_fp16, var_37466_cast_fp16))[name = tensor("op_37615_cast_fp16")]; + tensor var_37617_equation_0 = const()[name = tensor("op_37617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37617_cast_fp16 = einsum(equation = var_37617_equation_0, values = (var_37067_cast_fp16, var_37467_cast_fp16))[name = tensor("op_37617_cast_fp16")]; + tensor var_37619_equation_0 = const()[name = tensor("op_37619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37619_cast_fp16 = einsum(equation = var_37619_equation_0, values = (var_37067_cast_fp16, var_37468_cast_fp16))[name = tensor("op_37619_cast_fp16")]; + tensor var_37621_equation_0 = const()[name = tensor("op_37621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37621_cast_fp16 = einsum(equation = var_37621_equation_0, values = (var_37067_cast_fp16, var_37469_cast_fp16))[name = tensor("op_37621_cast_fp16")]; + tensor var_37623_equation_0 = const()[name = tensor("op_37623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37623_cast_fp16 = einsum(equation = var_37623_equation_0, values = (var_37071_cast_fp16, var_37470_cast_fp16))[name = tensor("op_37623_cast_fp16")]; + tensor var_37625_equation_0 = const()[name = tensor("op_37625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37625_cast_fp16 = einsum(equation = var_37625_equation_0, values = (var_37071_cast_fp16, var_37471_cast_fp16))[name = tensor("op_37625_cast_fp16")]; + tensor var_37627_equation_0 = const()[name = tensor("op_37627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37627_cast_fp16 = einsum(equation = var_37627_equation_0, values = (var_37071_cast_fp16, var_37472_cast_fp16))[name = tensor("op_37627_cast_fp16")]; + tensor var_37629_equation_0 = const()[name = tensor("op_37629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37629_cast_fp16 = einsum(equation = var_37629_equation_0, values = (var_37071_cast_fp16, var_37473_cast_fp16))[name = tensor("op_37629_cast_fp16")]; + tensor var_37631_equation_0 = const()[name = tensor("op_37631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37631_cast_fp16 = einsum(equation = var_37631_equation_0, values = (var_37075_cast_fp16, var_37474_cast_fp16))[name = tensor("op_37631_cast_fp16")]; + tensor var_37633_equation_0 = const()[name = tensor("op_37633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37633_cast_fp16 = einsum(equation = var_37633_equation_0, values = (var_37075_cast_fp16, var_37475_cast_fp16))[name = tensor("op_37633_cast_fp16")]; + tensor var_37635_equation_0 = const()[name = tensor("op_37635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37635_cast_fp16 = einsum(equation = var_37635_equation_0, values = (var_37075_cast_fp16, var_37476_cast_fp16))[name = tensor("op_37635_cast_fp16")]; + tensor var_37637_equation_0 = const()[name = tensor("op_37637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_37637_cast_fp16 = einsum(equation = var_37637_equation_0, values = (var_37075_cast_fp16, var_37477_cast_fp16))[name = tensor("op_37637_cast_fp16")]; + tensor var_37639_interleave_0 = const()[name = tensor("op_37639_interleave_0"), val = tensor(false)]; + tensor var_37639_cast_fp16 = concat(axis = var_36144, interleave = var_37639_interleave_0, values = (var_37479_cast_fp16, var_37481_cast_fp16, var_37483_cast_fp16, var_37485_cast_fp16))[name = tensor("op_37639_cast_fp16")]; + tensor var_37641_interleave_0 = const()[name = tensor("op_37641_interleave_0"), val = tensor(false)]; + tensor var_37641_cast_fp16 = concat(axis = var_36144, interleave = var_37641_interleave_0, values = (var_37487_cast_fp16, var_37489_cast_fp16, var_37491_cast_fp16, var_37493_cast_fp16))[name = tensor("op_37641_cast_fp16")]; + tensor var_37643_interleave_0 = const()[name = tensor("op_37643_interleave_0"), val = tensor(false)]; + tensor var_37643_cast_fp16 = concat(axis = var_36144, interleave = var_37643_interleave_0, values = (var_37495_cast_fp16, var_37497_cast_fp16, var_37499_cast_fp16, var_37501_cast_fp16))[name = tensor("op_37643_cast_fp16")]; + tensor var_37645_interleave_0 = const()[name = tensor("op_37645_interleave_0"), val = tensor(false)]; + tensor var_37645_cast_fp16 = concat(axis = var_36144, interleave = var_37645_interleave_0, values = (var_37503_cast_fp16, var_37505_cast_fp16, var_37507_cast_fp16, var_37509_cast_fp16))[name = tensor("op_37645_cast_fp16")]; + tensor var_37647_interleave_0 = const()[name = tensor("op_37647_interleave_0"), val = tensor(false)]; + tensor var_37647_cast_fp16 = concat(axis = var_36144, interleave = var_37647_interleave_0, values = (var_37511_cast_fp16, var_37513_cast_fp16, var_37515_cast_fp16, var_37517_cast_fp16))[name = tensor("op_37647_cast_fp16")]; + tensor var_37649_interleave_0 = const()[name = tensor("op_37649_interleave_0"), val = tensor(false)]; + tensor var_37649_cast_fp16 = concat(axis = var_36144, interleave = var_37649_interleave_0, values = (var_37519_cast_fp16, var_37521_cast_fp16, var_37523_cast_fp16, var_37525_cast_fp16))[name = tensor("op_37649_cast_fp16")]; + tensor var_37651_interleave_0 = const()[name = tensor("op_37651_interleave_0"), val = tensor(false)]; + tensor var_37651_cast_fp16 = concat(axis = var_36144, interleave = var_37651_interleave_0, values = (var_37527_cast_fp16, var_37529_cast_fp16, var_37531_cast_fp16, var_37533_cast_fp16))[name = tensor("op_37651_cast_fp16")]; + tensor var_37653_interleave_0 = const()[name = tensor("op_37653_interleave_0"), val = tensor(false)]; + tensor var_37653_cast_fp16 = concat(axis = var_36144, interleave = var_37653_interleave_0, values = (var_37535_cast_fp16, var_37537_cast_fp16, var_37539_cast_fp16, var_37541_cast_fp16))[name = tensor("op_37653_cast_fp16")]; + tensor var_37655_interleave_0 = const()[name = tensor("op_37655_interleave_0"), val = tensor(false)]; + tensor var_37655_cast_fp16 = concat(axis = var_36144, interleave = var_37655_interleave_0, values = (var_37543_cast_fp16, var_37545_cast_fp16, var_37547_cast_fp16, var_37549_cast_fp16))[name = tensor("op_37655_cast_fp16")]; + tensor var_37657_interleave_0 = const()[name = tensor("op_37657_interleave_0"), val = tensor(false)]; + tensor var_37657_cast_fp16 = concat(axis = var_36144, interleave = var_37657_interleave_0, values = (var_37551_cast_fp16, var_37553_cast_fp16, var_37555_cast_fp16, var_37557_cast_fp16))[name = tensor("op_37657_cast_fp16")]; + tensor var_37659_interleave_0 = const()[name = tensor("op_37659_interleave_0"), val = tensor(false)]; + tensor var_37659_cast_fp16 = concat(axis = var_36144, interleave = var_37659_interleave_0, values = (var_37559_cast_fp16, var_37561_cast_fp16, var_37563_cast_fp16, var_37565_cast_fp16))[name = tensor("op_37659_cast_fp16")]; + tensor var_37661_interleave_0 = const()[name = tensor("op_37661_interleave_0"), val = tensor(false)]; + tensor var_37661_cast_fp16 = concat(axis = var_36144, interleave = var_37661_interleave_0, values = (var_37567_cast_fp16, var_37569_cast_fp16, var_37571_cast_fp16, var_37573_cast_fp16))[name = tensor("op_37661_cast_fp16")]; + tensor var_37663_interleave_0 = const()[name = tensor("op_37663_interleave_0"), val = tensor(false)]; + tensor var_37663_cast_fp16 = concat(axis = var_36144, interleave = var_37663_interleave_0, values = (var_37575_cast_fp16, var_37577_cast_fp16, var_37579_cast_fp16, var_37581_cast_fp16))[name = tensor("op_37663_cast_fp16")]; + tensor var_37665_interleave_0 = const()[name = tensor("op_37665_interleave_0"), val = tensor(false)]; + tensor var_37665_cast_fp16 = concat(axis = var_36144, interleave = var_37665_interleave_0, values = (var_37583_cast_fp16, var_37585_cast_fp16, var_37587_cast_fp16, var_37589_cast_fp16))[name = tensor("op_37665_cast_fp16")]; + tensor var_37667_interleave_0 = const()[name = tensor("op_37667_interleave_0"), val = tensor(false)]; + tensor var_37667_cast_fp16 = concat(axis = var_36144, interleave = var_37667_interleave_0, values = (var_37591_cast_fp16, var_37593_cast_fp16, var_37595_cast_fp16, var_37597_cast_fp16))[name = tensor("op_37667_cast_fp16")]; + tensor var_37669_interleave_0 = const()[name = tensor("op_37669_interleave_0"), val = tensor(false)]; + tensor var_37669_cast_fp16 = concat(axis = var_36144, interleave = var_37669_interleave_0, values = (var_37599_cast_fp16, var_37601_cast_fp16, var_37603_cast_fp16, var_37605_cast_fp16))[name = tensor("op_37669_cast_fp16")]; + tensor var_37671_interleave_0 = const()[name = tensor("op_37671_interleave_0"), val = tensor(false)]; + tensor var_37671_cast_fp16 = concat(axis = var_36144, interleave = var_37671_interleave_0, values = (var_37607_cast_fp16, var_37609_cast_fp16, var_37611_cast_fp16, var_37613_cast_fp16))[name = tensor("op_37671_cast_fp16")]; + tensor var_37673_interleave_0 = const()[name = tensor("op_37673_interleave_0"), val = tensor(false)]; + tensor var_37673_cast_fp16 = concat(axis = var_36144, interleave = var_37673_interleave_0, values = (var_37615_cast_fp16, var_37617_cast_fp16, var_37619_cast_fp16, var_37621_cast_fp16))[name = tensor("op_37673_cast_fp16")]; + tensor var_37675_interleave_0 = const()[name = tensor("op_37675_interleave_0"), val = tensor(false)]; + tensor var_37675_cast_fp16 = concat(axis = var_36144, interleave = var_37675_interleave_0, values = (var_37623_cast_fp16, var_37625_cast_fp16, var_37627_cast_fp16, var_37629_cast_fp16))[name = tensor("op_37675_cast_fp16")]; + tensor var_37677_interleave_0 = const()[name = tensor("op_37677_interleave_0"), val = tensor(false)]; + tensor var_37677_cast_fp16 = concat(axis = var_36144, interleave = var_37677_interleave_0, values = (var_37631_cast_fp16, var_37633_cast_fp16, var_37635_cast_fp16, var_37637_cast_fp16))[name = tensor("op_37677_cast_fp16")]; + tensor input_447_interleave_0 = const()[name = tensor("input_447_interleave_0"), val = tensor(false)]; + tensor input_447_cast_fp16 = concat(axis = var_36169, interleave = input_447_interleave_0, values = (var_37639_cast_fp16, var_37641_cast_fp16, var_37643_cast_fp16, var_37645_cast_fp16, var_37647_cast_fp16, var_37649_cast_fp16, var_37651_cast_fp16, var_37653_cast_fp16, var_37655_cast_fp16, var_37657_cast_fp16, var_37659_cast_fp16, var_37661_cast_fp16, var_37663_cast_fp16, var_37665_cast_fp16, var_37667_cast_fp16, var_37669_cast_fp16, var_37671_cast_fp16, var_37673_cast_fp16, var_37675_cast_fp16, var_37677_cast_fp16))[name = tensor("input_447_cast_fp16")]; + tensor var_37685 = const()[name = tensor("op_37685"), val = tensor([1, 1])]; + tensor var_37687 = const()[name = tensor("op_37687"), val = tensor([1, 1])]; + tensor pretrained_out_271_pad_type_0 = const()[name = tensor("pretrained_out_271_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_271_pad_0 = const()[name = tensor("pretrained_out_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250240384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251059648))), name = tensor("layers_22_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_22_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_22_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251059776)))]; + tensor pretrained_out_271_cast_fp16 = conv(bias = layers_22_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_37687, groups = var_36169, pad = pretrained_out_271_pad_0, pad_type = pretrained_out_271_pad_type_0, strides = var_37685, weight = layers_22_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_447_cast_fp16)[name = tensor("pretrained_out_271_cast_fp16")]; + tensor var_37691 = const()[name = tensor("op_37691"), val = tensor([1, 1])]; + tensor var_37693 = const()[name = tensor("op_37693"), val = tensor([1, 1])]; + tensor input_449_pad_type_0 = const()[name = tensor("input_449_pad_type_0"), val = tensor("custom")]; + tensor input_449_pad_0 = const()[name = tensor("input_449_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_22_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251062400)))]; + tensor input_449_cast_fp16 = conv(dilations = var_37693, groups = var_36169, pad = input_449_pad_0, pad_type = input_449_pad_type_0, strides = var_37691, weight = layers_22_self_attn_o_proj_loraA_weight_to_fp16, x = input_447_cast_fp16)[name = tensor("input_449_cast_fp16")]; + tensor var_37697 = const()[name = tensor("op_37697"), val = tensor([1, 1])]; + tensor var_37699 = const()[name = tensor("op_37699"), val = tensor([1, 1])]; + tensor lora_out_541_pad_type_0 = const()[name = tensor("lora_out_541_pad_type_0"), val = tensor("custom")]; + tensor lora_out_541_pad_0 = const()[name = tensor("lora_out_541_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_543_weight_0_to_fp16 = const()[name = tensor("lora_out_543_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251103424)))]; + tensor lora_out_543_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_37699, groups = var_36169, pad = lora_out_541_pad_0, pad_type = lora_out_541_pad_type_0, strides = var_37697, weight = lora_out_543_weight_0_to_fp16, x = input_449_cast_fp16)[name = tensor("lora_out_543_cast_fp16")]; + tensor obj_91_cast_fp16 = add(x = pretrained_out_271_cast_fp16, y = lora_out_543_cast_fp16)[name = tensor("obj_91_cast_fp16")]; + tensor inputs_91_cast_fp16 = add(x = inputs_89_cast_fp16, y = obj_91_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; + tensor var_37708 = const()[name = tensor("op_37708"), val = tensor([1])]; + tensor channels_mean_91_cast_fp16 = reduce_mean(axes = var_37708, keep_dims = var_36170, x = inputs_91_cast_fp16)[name = tensor("channels_mean_91_cast_fp16")]; + tensor zero_mean_91_cast_fp16 = sub(x = inputs_91_cast_fp16, y = channels_mean_91_cast_fp16)[name = tensor("zero_mean_91_cast_fp16")]; + tensor zero_mean_sq_91_cast_fp16 = mul(x = zero_mean_91_cast_fp16, y = zero_mean_91_cast_fp16)[name = tensor("zero_mean_sq_91_cast_fp16")]; + tensor var_37712 = const()[name = tensor("op_37712"), val = tensor([1])]; + tensor var_37713_cast_fp16 = reduce_mean(axes = var_37712, keep_dims = var_36170, x = zero_mean_sq_91_cast_fp16)[name = tensor("op_37713_cast_fp16")]; + tensor var_37714_to_fp16 = const()[name = tensor("op_37714_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_37715_cast_fp16 = add(x = var_37713_cast_fp16, y = var_37714_to_fp16)[name = tensor("op_37715_cast_fp16")]; + tensor denom_91_epsilon_0 = const()[name = tensor("denom_91_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_91_cast_fp16 = rsqrt(epsilon = denom_91_epsilon_0, x = var_37715_cast_fp16)[name = tensor("denom_91_cast_fp16")]; + tensor out_91_cast_fp16 = mul(x = zero_mean_91_cast_fp16, y = denom_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; + tensor input_451_gamma_0_to_fp16 = const()[name = tensor("input_451_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251144448)))]; + tensor input_451_beta_0_to_fp16 = const()[name = tensor("input_451_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251147072)))]; + tensor input_451_epsilon_0_to_fp16 = const()[name = tensor("input_451_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_451_cast_fp16 = batch_norm(beta = input_451_beta_0_to_fp16, epsilon = input_451_epsilon_0_to_fp16, gamma = input_451_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor("input_451_cast_fp16")]; + tensor var_37729 = const()[name = tensor("op_37729"), val = tensor([1, 1])]; + tensor var_37731 = const()[name = tensor("op_37731"), val = tensor([1, 1])]; + tensor pretrained_out_273_pad_type_0 = const()[name = tensor("pretrained_out_273_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_273_pad_0 = const()[name = tensor("pretrained_out_273_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(251149696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254426560))), name = tensor("layers_22_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_22_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_22_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254426688)))]; + tensor pretrained_out_273_cast_fp16 = conv(bias = layers_22_fc1_pretrained_bias_to_fp16, dilations = var_37731, groups = var_36169, pad = pretrained_out_273_pad_0, pad_type = pretrained_out_273_pad_type_0, strides = var_37729, weight = layers_22_fc1_pretrained_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = tensor("pretrained_out_273_cast_fp16")]; + tensor var_37735 = const()[name = tensor("op_37735"), val = tensor([1, 1])]; + tensor var_37737 = const()[name = tensor("op_37737"), val = tensor([1, 1])]; + tensor input_453_pad_type_0 = const()[name = tensor("input_453_pad_type_0"), val = tensor("custom")]; + tensor input_453_pad_0 = const()[name = tensor("input_453_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_22_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254436992)))]; + tensor input_453_cast_fp16 = conv(dilations = var_37737, groups = var_36169, pad = input_453_pad_0, pad_type = input_453_pad_type_0, strides = var_37735, weight = layers_22_fc1_loraA_weight_to_fp16, x = input_451_cast_fp16)[name = tensor("input_453_cast_fp16")]; + tensor var_37741 = const()[name = tensor("op_37741"), val = tensor([1, 1])]; + tensor var_37743 = const()[name = tensor("op_37743"), val = tensor([1, 1])]; + tensor lora_out_545_pad_type_0 = const()[name = tensor("lora_out_545_pad_type_0"), val = tensor("custom")]; + tensor lora_out_545_pad_0 = const()[name = tensor("lora_out_545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_547_weight_0_to_fp16 = const()[name = tensor("lora_out_547_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254478016)))]; + tensor lora_out_547_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_37743, groups = var_36169, pad = lora_out_545_pad_0, pad_type = lora_out_545_pad_type_0, strides = var_37741, weight = lora_out_547_weight_0_to_fp16, x = input_453_cast_fp16)[name = tensor("lora_out_547_cast_fp16")]; + tensor input_455_cast_fp16 = add(x = pretrained_out_273_cast_fp16, y = lora_out_547_cast_fp16)[name = tensor("input_455_cast_fp16")]; + tensor input_457_mode_0 = const()[name = tensor("input_457_mode_0"), val = tensor("EXACT")]; + tensor input_457_cast_fp16 = gelu(mode = input_457_mode_0, x = input_455_cast_fp16)[name = tensor("input_457_cast_fp16")]; + tensor var_37755 = const()[name = tensor("op_37755"), val = tensor([1, 1])]; + tensor var_37757 = const()[name = tensor("op_37757"), val = tensor([1, 1])]; + tensor pretrained_out_275_pad_type_0 = const()[name = tensor("pretrained_out_275_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_275_pad_0 = const()[name = tensor("pretrained_out_275_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254641920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257918784))), name = tensor("layers_22_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_22_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_22_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257918912)))]; + tensor pretrained_out_275_cast_fp16 = conv(bias = layers_22_fc2_pretrained_bias_to_fp16, dilations = var_37757, groups = var_36169, pad = pretrained_out_275_pad_0, pad_type = pretrained_out_275_pad_type_0, strides = var_37755, weight = layers_22_fc2_pretrained_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = tensor("pretrained_out_275_cast_fp16")]; + tensor var_37761 = const()[name = tensor("op_37761"), val = tensor([1, 1])]; + tensor var_37763 = const()[name = tensor("op_37763"), val = tensor([1, 1])]; + tensor input_459_pad_type_0 = const()[name = tensor("input_459_pad_type_0"), val = tensor("custom")]; + tensor input_459_pad_0 = const()[name = tensor("input_459_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_22_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_22_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257921536)))]; + tensor input_459_cast_fp16 = conv(dilations = var_37763, groups = var_36169, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = var_37761, weight = layers_22_fc2_loraA_weight_to_fp16, x = input_457_cast_fp16)[name = tensor("input_459_cast_fp16")]; + tensor var_37767 = const()[name = tensor("op_37767"), val = tensor([1, 1])]; + tensor var_37769 = const()[name = tensor("op_37769"), val = tensor([1, 1])]; + tensor lora_out_549_pad_type_0 = const()[name = tensor("lora_out_549_pad_type_0"), val = tensor("custom")]; + tensor lora_out_549_pad_0 = const()[name = tensor("lora_out_549_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_551_weight_0_to_fp16 = const()[name = tensor("lora_out_551_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258085440)))]; + tensor lora_out_551_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_37769, groups = var_36169, pad = lora_out_549_pad_0, pad_type = lora_out_549_pad_type_0, strides = var_37767, weight = lora_out_551_weight_0_to_fp16, x = input_459_cast_fp16)[name = tensor("lora_out_551_cast_fp16")]; + tensor hidden_states_49_cast_fp16 = add(x = pretrained_out_275_cast_fp16, y = lora_out_551_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = hidden_states_49_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; + tensor var_37779 = const()[name = tensor("op_37779"), val = tensor(3)]; + tensor var_37804 = const()[name = tensor("op_37804"), val = tensor(1)]; + tensor var_37805 = const()[name = tensor("op_37805"), val = tensor(true)]; + tensor var_37815 = const()[name = tensor("op_37815"), val = tensor([1])]; + tensor channels_mean_93_cast_fp16 = reduce_mean(axes = var_37815, keep_dims = var_37805, x = inputs_93_cast_fp16)[name = tensor("channels_mean_93_cast_fp16")]; + tensor zero_mean_93_cast_fp16 = sub(x = inputs_93_cast_fp16, y = channels_mean_93_cast_fp16)[name = tensor("zero_mean_93_cast_fp16")]; + tensor zero_mean_sq_93_cast_fp16 = mul(x = zero_mean_93_cast_fp16, y = zero_mean_93_cast_fp16)[name = tensor("zero_mean_sq_93_cast_fp16")]; + tensor var_37819 = const()[name = tensor("op_37819"), val = tensor([1])]; + tensor var_37820_cast_fp16 = reduce_mean(axes = var_37819, keep_dims = var_37805, x = zero_mean_sq_93_cast_fp16)[name = tensor("op_37820_cast_fp16")]; + tensor var_37821_to_fp16 = const()[name = tensor("op_37821_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_37822_cast_fp16 = add(x = var_37820_cast_fp16, y = var_37821_to_fp16)[name = tensor("op_37822_cast_fp16")]; + tensor denom_93_epsilon_0 = const()[name = tensor("denom_93_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_93_cast_fp16 = rsqrt(epsilon = denom_93_epsilon_0, x = var_37822_cast_fp16)[name = tensor("denom_93_cast_fp16")]; + tensor out_93_cast_fp16 = mul(x = zero_mean_93_cast_fp16, y = denom_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; + tensor obj_93_gamma_0_to_fp16 = const()[name = tensor("obj_93_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258126464)))]; + tensor obj_93_beta_0_to_fp16 = const()[name = tensor("obj_93_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258129088)))]; + tensor obj_93_epsilon_0_to_fp16 = const()[name = tensor("obj_93_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_93_cast_fp16 = batch_norm(beta = obj_93_beta_0_to_fp16, epsilon = obj_93_epsilon_0_to_fp16, gamma = obj_93_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor("obj_93_cast_fp16")]; + tensor var_37840 = const()[name = tensor("op_37840"), val = tensor([1, 1])]; + tensor var_37842 = const()[name = tensor("op_37842"), val = tensor([1, 1])]; + tensor pretrained_out_277_pad_type_0 = const()[name = tensor("pretrained_out_277_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_277_pad_0 = const()[name = tensor("pretrained_out_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258131712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258950976))), name = tensor("layers_23_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_23_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258951104)))]; + tensor pretrained_out_277_cast_fp16 = conv(bias = layers_23_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_37842, groups = var_37804, pad = pretrained_out_277_pad_0, pad_type = pretrained_out_277_pad_type_0, strides = var_37840, weight = layers_23_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_93_cast_fp16)[name = tensor("pretrained_out_277_cast_fp16")]; + tensor var_37846 = const()[name = tensor("op_37846"), val = tensor([1, 1])]; + tensor var_37848 = const()[name = tensor("op_37848"), val = tensor([1, 1])]; + tensor input_461_pad_type_0 = const()[name = tensor("input_461_pad_type_0"), val = tensor("custom")]; + tensor input_461_pad_0 = const()[name = tensor("input_461_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258953728)))]; + tensor input_461_cast_fp16 = conv(dilations = var_37848, groups = var_37804, pad = input_461_pad_0, pad_type = input_461_pad_type_0, strides = var_37846, weight = layers_23_self_attn_q_proj_loraA_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor("input_461_cast_fp16")]; + tensor var_37852 = const()[name = tensor("op_37852"), val = tensor([1, 1])]; + tensor var_37854 = const()[name = tensor("op_37854"), val = tensor([1, 1])]; + tensor lora_out_553_pad_type_0 = const()[name = tensor("lora_out_553_pad_type_0"), val = tensor("custom")]; + tensor lora_out_553_pad_0 = const()[name = tensor("lora_out_553_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_555_weight_0_to_fp16 = const()[name = tensor("lora_out_555_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258994752)))]; + tensor lora_out_555_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_37854, groups = var_37804, pad = lora_out_553_pad_0, pad_type = lora_out_553_pad_type_0, strides = var_37852, weight = lora_out_555_weight_0_to_fp16, x = input_461_cast_fp16)[name = tensor("lora_out_555_cast_fp16")]; + tensor query_47_cast_fp16 = add(x = pretrained_out_277_cast_fp16, y = lora_out_555_cast_fp16)[name = tensor("query_47_cast_fp16")]; + tensor var_37864 = const()[name = tensor("op_37864"), val = tensor([1, 1])]; + tensor var_37866 = const()[name = tensor("op_37866"), val = tensor([1, 1])]; + tensor pretrained_out_279_pad_type_0 = const()[name = tensor("pretrained_out_279_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_279_pad_0 = const()[name = tensor("pretrained_out_279_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259035776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259855040))), name = tensor("layers_23_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_279_cast_fp16 = conv(dilations = var_37866, groups = var_37804, pad = pretrained_out_279_pad_0, pad_type = pretrained_out_279_pad_type_0, strides = var_37864, weight = layers_23_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_93_cast_fp16)[name = tensor("pretrained_out_279_cast_fp16")]; + tensor var_37870 = const()[name = tensor("op_37870"), val = tensor([1, 1])]; + tensor var_37872 = const()[name = tensor("op_37872"), val = tensor([1, 1])]; + tensor input_463_pad_type_0 = const()[name = tensor("input_463_pad_type_0"), val = tensor("custom")]; + tensor input_463_pad_0 = const()[name = tensor("input_463_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259855168)))]; + tensor input_463_cast_fp16 = conv(dilations = var_37872, groups = var_37804, pad = input_463_pad_0, pad_type = input_463_pad_type_0, strides = var_37870, weight = layers_23_self_attn_k_proj_loraA_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor("input_463_cast_fp16")]; + tensor var_37876 = const()[name = tensor("op_37876"), val = tensor([1, 1])]; + tensor var_37878 = const()[name = tensor("op_37878"), val = tensor([1, 1])]; + tensor lora_out_557_pad_type_0 = const()[name = tensor("lora_out_557_pad_type_0"), val = tensor("custom")]; + tensor lora_out_557_pad_0 = const()[name = tensor("lora_out_557_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_559_weight_0_to_fp16 = const()[name = tensor("lora_out_559_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259896192)))]; + tensor lora_out_559_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_37878, groups = var_37804, pad = lora_out_557_pad_0, pad_type = lora_out_557_pad_type_0, strides = var_37876, weight = lora_out_559_weight_0_to_fp16, x = input_463_cast_fp16)[name = tensor("lora_out_559_cast_fp16")]; + tensor key_47_cast_fp16 = add(x = pretrained_out_279_cast_fp16, y = lora_out_559_cast_fp16)[name = tensor("key_47_cast_fp16")]; + tensor var_37889 = const()[name = tensor("op_37889"), val = tensor([1, 1])]; + tensor var_37891 = const()[name = tensor("op_37891"), val = tensor([1, 1])]; + tensor pretrained_out_281_pad_type_0 = const()[name = tensor("pretrained_out_281_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_281_pad_0 = const()[name = tensor("pretrained_out_281_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(259937216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260756480))), name = tensor("layers_23_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_23_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260756608)))]; + tensor pretrained_out_281_cast_fp16 = conv(bias = layers_23_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_37891, groups = var_37804, pad = pretrained_out_281_pad_0, pad_type = pretrained_out_281_pad_type_0, strides = var_37889, weight = layers_23_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_93_cast_fp16)[name = tensor("pretrained_out_281_cast_fp16")]; + tensor var_37895 = const()[name = tensor("op_37895"), val = tensor([1, 1])]; + tensor var_37897 = const()[name = tensor("op_37897"), val = tensor([1, 1])]; + tensor input_465_pad_type_0 = const()[name = tensor("input_465_pad_type_0"), val = tensor("custom")]; + tensor input_465_pad_0 = const()[name = tensor("input_465_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260759232)))]; + tensor input_465_cast_fp16 = conv(dilations = var_37897, groups = var_37804, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = var_37895, weight = layers_23_self_attn_v_proj_loraA_weight_to_fp16, x = obj_93_cast_fp16)[name = tensor("input_465_cast_fp16")]; + tensor var_37901 = const()[name = tensor("op_37901"), val = tensor([1, 1])]; + tensor var_37903 = const()[name = tensor("op_37903"), val = tensor([1, 1])]; + tensor lora_out_561_pad_type_0 = const()[name = tensor("lora_out_561_pad_type_0"), val = tensor("custom")]; + tensor lora_out_561_pad_0 = const()[name = tensor("lora_out_561_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_563_weight_0_to_fp16 = const()[name = tensor("lora_out_563_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260800256)))]; + tensor lora_out_563_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_37903, groups = var_37804, pad = lora_out_561_pad_0, pad_type = lora_out_561_pad_type_0, strides = var_37901, weight = lora_out_563_weight_0_to_fp16, x = input_465_cast_fp16)[name = tensor("lora_out_563_cast_fp16")]; + tensor value_47_cast_fp16 = add(x = pretrained_out_281_cast_fp16, y = lora_out_563_cast_fp16)[name = tensor("value_47_cast_fp16")]; + tensor var_37913_begin_0 = const()[name = tensor("op_37913_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37913_end_0 = const()[name = tensor("op_37913_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_37913_end_mask_0 = const()[name = tensor("op_37913_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37913_cast_fp16 = slice_by_index(begin = var_37913_begin_0, end = var_37913_end_0, end_mask = var_37913_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37913_cast_fp16")]; + tensor var_37917_begin_0 = const()[name = tensor("op_37917_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_37917_end_0 = const()[name = tensor("op_37917_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_37917_end_mask_0 = const()[name = tensor("op_37917_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37917_cast_fp16 = slice_by_index(begin = var_37917_begin_0, end = var_37917_end_0, end_mask = var_37917_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37917_cast_fp16")]; + tensor var_37921_begin_0 = const()[name = tensor("op_37921_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_37921_end_0 = const()[name = tensor("op_37921_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_37921_end_mask_0 = const()[name = tensor("op_37921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37921_cast_fp16 = slice_by_index(begin = var_37921_begin_0, end = var_37921_end_0, end_mask = var_37921_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37921_cast_fp16")]; + tensor var_37925_begin_0 = const()[name = tensor("op_37925_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_37925_end_0 = const()[name = tensor("op_37925_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_37925_end_mask_0 = const()[name = tensor("op_37925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37925_cast_fp16 = slice_by_index(begin = var_37925_begin_0, end = var_37925_end_0, end_mask = var_37925_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37925_cast_fp16")]; + tensor var_37929_begin_0 = const()[name = tensor("op_37929_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_37929_end_0 = const()[name = tensor("op_37929_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_37929_end_mask_0 = const()[name = tensor("op_37929_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37929_cast_fp16 = slice_by_index(begin = var_37929_begin_0, end = var_37929_end_0, end_mask = var_37929_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37929_cast_fp16")]; + tensor var_37933_begin_0 = const()[name = tensor("op_37933_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_37933_end_0 = const()[name = tensor("op_37933_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_37933_end_mask_0 = const()[name = tensor("op_37933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37933_cast_fp16 = slice_by_index(begin = var_37933_begin_0, end = var_37933_end_0, end_mask = var_37933_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37933_cast_fp16")]; + tensor var_37937_begin_0 = const()[name = tensor("op_37937_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_37937_end_0 = const()[name = tensor("op_37937_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_37937_end_mask_0 = const()[name = tensor("op_37937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37937_cast_fp16 = slice_by_index(begin = var_37937_begin_0, end = var_37937_end_0, end_mask = var_37937_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37937_cast_fp16")]; + tensor var_37941_begin_0 = const()[name = tensor("op_37941_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_37941_end_0 = const()[name = tensor("op_37941_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_37941_end_mask_0 = const()[name = tensor("op_37941_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37941_cast_fp16 = slice_by_index(begin = var_37941_begin_0, end = var_37941_end_0, end_mask = var_37941_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37941_cast_fp16")]; + tensor var_37945_begin_0 = const()[name = tensor("op_37945_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_37945_end_0 = const()[name = tensor("op_37945_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_37945_end_mask_0 = const()[name = tensor("op_37945_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37945_cast_fp16 = slice_by_index(begin = var_37945_begin_0, end = var_37945_end_0, end_mask = var_37945_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37945_cast_fp16")]; + tensor var_37949_begin_0 = const()[name = tensor("op_37949_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_37949_end_0 = const()[name = tensor("op_37949_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_37949_end_mask_0 = const()[name = tensor("op_37949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37949_cast_fp16 = slice_by_index(begin = var_37949_begin_0, end = var_37949_end_0, end_mask = var_37949_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37949_cast_fp16")]; + tensor var_37953_begin_0 = const()[name = tensor("op_37953_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_37953_end_0 = const()[name = tensor("op_37953_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_37953_end_mask_0 = const()[name = tensor("op_37953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37953_cast_fp16 = slice_by_index(begin = var_37953_begin_0, end = var_37953_end_0, end_mask = var_37953_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37953_cast_fp16")]; + tensor var_37957_begin_0 = const()[name = tensor("op_37957_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_37957_end_0 = const()[name = tensor("op_37957_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_37957_end_mask_0 = const()[name = tensor("op_37957_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37957_cast_fp16 = slice_by_index(begin = var_37957_begin_0, end = var_37957_end_0, end_mask = var_37957_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37957_cast_fp16")]; + tensor var_37961_begin_0 = const()[name = tensor("op_37961_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_37961_end_0 = const()[name = tensor("op_37961_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_37961_end_mask_0 = const()[name = tensor("op_37961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37961_cast_fp16 = slice_by_index(begin = var_37961_begin_0, end = var_37961_end_0, end_mask = var_37961_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37961_cast_fp16")]; + tensor var_37965_begin_0 = const()[name = tensor("op_37965_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_37965_end_0 = const()[name = tensor("op_37965_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_37965_end_mask_0 = const()[name = tensor("op_37965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37965_cast_fp16 = slice_by_index(begin = var_37965_begin_0, end = var_37965_end_0, end_mask = var_37965_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37965_cast_fp16")]; + tensor var_37969_begin_0 = const()[name = tensor("op_37969_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_37969_end_0 = const()[name = tensor("op_37969_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_37969_end_mask_0 = const()[name = tensor("op_37969_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37969_cast_fp16 = slice_by_index(begin = var_37969_begin_0, end = var_37969_end_0, end_mask = var_37969_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37969_cast_fp16")]; + tensor var_37973_begin_0 = const()[name = tensor("op_37973_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_37973_end_0 = const()[name = tensor("op_37973_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_37973_end_mask_0 = const()[name = tensor("op_37973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37973_cast_fp16 = slice_by_index(begin = var_37973_begin_0, end = var_37973_end_0, end_mask = var_37973_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37973_cast_fp16")]; + tensor var_37977_begin_0 = const()[name = tensor("op_37977_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_37977_end_0 = const()[name = tensor("op_37977_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_37977_end_mask_0 = const()[name = tensor("op_37977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37977_cast_fp16 = slice_by_index(begin = var_37977_begin_0, end = var_37977_end_0, end_mask = var_37977_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37977_cast_fp16")]; + tensor var_37981_begin_0 = const()[name = tensor("op_37981_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_37981_end_0 = const()[name = tensor("op_37981_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_37981_end_mask_0 = const()[name = tensor("op_37981_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37981_cast_fp16 = slice_by_index(begin = var_37981_begin_0, end = var_37981_end_0, end_mask = var_37981_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37981_cast_fp16")]; + tensor var_37985_begin_0 = const()[name = tensor("op_37985_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_37985_end_0 = const()[name = tensor("op_37985_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_37985_end_mask_0 = const()[name = tensor("op_37985_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37985_cast_fp16 = slice_by_index(begin = var_37985_begin_0, end = var_37985_end_0, end_mask = var_37985_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37985_cast_fp16")]; + tensor var_37989_begin_0 = const()[name = tensor("op_37989_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_37989_end_0 = const()[name = tensor("op_37989_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_37989_end_mask_0 = const()[name = tensor("op_37989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_37989_cast_fp16 = slice_by_index(begin = var_37989_begin_0, end = var_37989_end_0, end_mask = var_37989_end_mask_0, x = query_47_cast_fp16)[name = tensor("op_37989_cast_fp16")]; + tensor var_37998_begin_0 = const()[name = tensor("op_37998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_37998_end_0 = const()[name = tensor("op_37998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_37998_end_mask_0 = const()[name = tensor("op_37998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_37998_cast_fp16 = slice_by_index(begin = var_37998_begin_0, end = var_37998_end_0, end_mask = var_37998_end_mask_0, x = var_37913_cast_fp16)[name = tensor("op_37998_cast_fp16")]; + tensor var_38005_begin_0 = const()[name = tensor("op_38005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38005_end_0 = const()[name = tensor("op_38005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38005_end_mask_0 = const()[name = tensor("op_38005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38005_cast_fp16 = slice_by_index(begin = var_38005_begin_0, end = var_38005_end_0, end_mask = var_38005_end_mask_0, x = var_37913_cast_fp16)[name = tensor("op_38005_cast_fp16")]; + tensor var_38012_begin_0 = const()[name = tensor("op_38012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38012_end_0 = const()[name = tensor("op_38012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38012_end_mask_0 = const()[name = tensor("op_38012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38012_cast_fp16 = slice_by_index(begin = var_38012_begin_0, end = var_38012_end_0, end_mask = var_38012_end_mask_0, x = var_37913_cast_fp16)[name = tensor("op_38012_cast_fp16")]; + tensor var_38019_begin_0 = const()[name = tensor("op_38019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38019_end_0 = const()[name = tensor("op_38019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38019_end_mask_0 = const()[name = tensor("op_38019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38019_cast_fp16 = slice_by_index(begin = var_38019_begin_0, end = var_38019_end_0, end_mask = var_38019_end_mask_0, x = var_37913_cast_fp16)[name = tensor("op_38019_cast_fp16")]; + tensor var_38026_begin_0 = const()[name = tensor("op_38026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38026_end_0 = const()[name = tensor("op_38026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38026_end_mask_0 = const()[name = tensor("op_38026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38026_cast_fp16 = slice_by_index(begin = var_38026_begin_0, end = var_38026_end_0, end_mask = var_38026_end_mask_0, x = var_37917_cast_fp16)[name = tensor("op_38026_cast_fp16")]; + tensor var_38033_begin_0 = const()[name = tensor("op_38033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38033_end_0 = const()[name = tensor("op_38033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38033_end_mask_0 = const()[name = tensor("op_38033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38033_cast_fp16 = slice_by_index(begin = var_38033_begin_0, end = var_38033_end_0, end_mask = var_38033_end_mask_0, x = var_37917_cast_fp16)[name = tensor("op_38033_cast_fp16")]; + tensor var_38040_begin_0 = const()[name = tensor("op_38040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38040_end_0 = const()[name = tensor("op_38040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38040_end_mask_0 = const()[name = tensor("op_38040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38040_cast_fp16 = slice_by_index(begin = var_38040_begin_0, end = var_38040_end_0, end_mask = var_38040_end_mask_0, x = var_37917_cast_fp16)[name = tensor("op_38040_cast_fp16")]; + tensor var_38047_begin_0 = const()[name = tensor("op_38047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38047_end_0 = const()[name = tensor("op_38047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38047_end_mask_0 = const()[name = tensor("op_38047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38047_cast_fp16 = slice_by_index(begin = var_38047_begin_0, end = var_38047_end_0, end_mask = var_38047_end_mask_0, x = var_37917_cast_fp16)[name = tensor("op_38047_cast_fp16")]; + tensor var_38054_begin_0 = const()[name = tensor("op_38054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38054_end_0 = const()[name = tensor("op_38054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38054_end_mask_0 = const()[name = tensor("op_38054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38054_cast_fp16 = slice_by_index(begin = var_38054_begin_0, end = var_38054_end_0, end_mask = var_38054_end_mask_0, x = var_37921_cast_fp16)[name = tensor("op_38054_cast_fp16")]; + tensor var_38061_begin_0 = const()[name = tensor("op_38061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38061_end_0 = const()[name = tensor("op_38061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38061_end_mask_0 = const()[name = tensor("op_38061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38061_cast_fp16 = slice_by_index(begin = var_38061_begin_0, end = var_38061_end_0, end_mask = var_38061_end_mask_0, x = var_37921_cast_fp16)[name = tensor("op_38061_cast_fp16")]; + tensor var_38068_begin_0 = const()[name = tensor("op_38068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38068_end_0 = const()[name = tensor("op_38068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38068_end_mask_0 = const()[name = tensor("op_38068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38068_cast_fp16 = slice_by_index(begin = var_38068_begin_0, end = var_38068_end_0, end_mask = var_38068_end_mask_0, x = var_37921_cast_fp16)[name = tensor("op_38068_cast_fp16")]; + tensor var_38075_begin_0 = const()[name = tensor("op_38075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38075_end_0 = const()[name = tensor("op_38075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38075_end_mask_0 = const()[name = tensor("op_38075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38075_cast_fp16 = slice_by_index(begin = var_38075_begin_0, end = var_38075_end_0, end_mask = var_38075_end_mask_0, x = var_37921_cast_fp16)[name = tensor("op_38075_cast_fp16")]; + tensor var_38082_begin_0 = const()[name = tensor("op_38082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38082_end_0 = const()[name = tensor("op_38082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38082_end_mask_0 = const()[name = tensor("op_38082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38082_cast_fp16 = slice_by_index(begin = var_38082_begin_0, end = var_38082_end_0, end_mask = var_38082_end_mask_0, x = var_37925_cast_fp16)[name = tensor("op_38082_cast_fp16")]; + tensor var_38089_begin_0 = const()[name = tensor("op_38089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38089_end_0 = const()[name = tensor("op_38089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38089_end_mask_0 = const()[name = tensor("op_38089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38089_cast_fp16 = slice_by_index(begin = var_38089_begin_0, end = var_38089_end_0, end_mask = var_38089_end_mask_0, x = var_37925_cast_fp16)[name = tensor("op_38089_cast_fp16")]; + tensor var_38096_begin_0 = const()[name = tensor("op_38096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38096_end_0 = const()[name = tensor("op_38096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38096_end_mask_0 = const()[name = tensor("op_38096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38096_cast_fp16 = slice_by_index(begin = var_38096_begin_0, end = var_38096_end_0, end_mask = var_38096_end_mask_0, x = var_37925_cast_fp16)[name = tensor("op_38096_cast_fp16")]; + tensor var_38103_begin_0 = const()[name = tensor("op_38103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38103_end_0 = const()[name = tensor("op_38103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38103_end_mask_0 = const()[name = tensor("op_38103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38103_cast_fp16 = slice_by_index(begin = var_38103_begin_0, end = var_38103_end_0, end_mask = var_38103_end_mask_0, x = var_37925_cast_fp16)[name = tensor("op_38103_cast_fp16")]; + tensor var_38110_begin_0 = const()[name = tensor("op_38110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38110_end_0 = const()[name = tensor("op_38110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38110_end_mask_0 = const()[name = tensor("op_38110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38110_cast_fp16 = slice_by_index(begin = var_38110_begin_0, end = var_38110_end_0, end_mask = var_38110_end_mask_0, x = var_37929_cast_fp16)[name = tensor("op_38110_cast_fp16")]; + tensor var_38117_begin_0 = const()[name = tensor("op_38117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38117_end_0 = const()[name = tensor("op_38117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38117_end_mask_0 = const()[name = tensor("op_38117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38117_cast_fp16 = slice_by_index(begin = var_38117_begin_0, end = var_38117_end_0, end_mask = var_38117_end_mask_0, x = var_37929_cast_fp16)[name = tensor("op_38117_cast_fp16")]; + tensor var_38124_begin_0 = const()[name = tensor("op_38124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38124_end_0 = const()[name = tensor("op_38124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38124_end_mask_0 = const()[name = tensor("op_38124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38124_cast_fp16 = slice_by_index(begin = var_38124_begin_0, end = var_38124_end_0, end_mask = var_38124_end_mask_0, x = var_37929_cast_fp16)[name = tensor("op_38124_cast_fp16")]; + tensor var_38131_begin_0 = const()[name = tensor("op_38131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38131_end_0 = const()[name = tensor("op_38131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38131_end_mask_0 = const()[name = tensor("op_38131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38131_cast_fp16 = slice_by_index(begin = var_38131_begin_0, end = var_38131_end_0, end_mask = var_38131_end_mask_0, x = var_37929_cast_fp16)[name = tensor("op_38131_cast_fp16")]; + tensor var_38138_begin_0 = const()[name = tensor("op_38138_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38138_end_0 = const()[name = tensor("op_38138_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38138_end_mask_0 = const()[name = tensor("op_38138_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38138_cast_fp16 = slice_by_index(begin = var_38138_begin_0, end = var_38138_end_0, end_mask = var_38138_end_mask_0, x = var_37933_cast_fp16)[name = tensor("op_38138_cast_fp16")]; + tensor var_38145_begin_0 = const()[name = tensor("op_38145_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38145_end_0 = const()[name = tensor("op_38145_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38145_end_mask_0 = const()[name = tensor("op_38145_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38145_cast_fp16 = slice_by_index(begin = var_38145_begin_0, end = var_38145_end_0, end_mask = var_38145_end_mask_0, x = var_37933_cast_fp16)[name = tensor("op_38145_cast_fp16")]; + tensor var_38152_begin_0 = const()[name = tensor("op_38152_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38152_end_0 = const()[name = tensor("op_38152_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38152_end_mask_0 = const()[name = tensor("op_38152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38152_cast_fp16 = slice_by_index(begin = var_38152_begin_0, end = var_38152_end_0, end_mask = var_38152_end_mask_0, x = var_37933_cast_fp16)[name = tensor("op_38152_cast_fp16")]; + tensor var_38159_begin_0 = const()[name = tensor("op_38159_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38159_end_0 = const()[name = tensor("op_38159_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38159_end_mask_0 = const()[name = tensor("op_38159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38159_cast_fp16 = slice_by_index(begin = var_38159_begin_0, end = var_38159_end_0, end_mask = var_38159_end_mask_0, x = var_37933_cast_fp16)[name = tensor("op_38159_cast_fp16")]; + tensor var_38166_begin_0 = const()[name = tensor("op_38166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38166_end_0 = const()[name = tensor("op_38166_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38166_end_mask_0 = const()[name = tensor("op_38166_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38166_cast_fp16 = slice_by_index(begin = var_38166_begin_0, end = var_38166_end_0, end_mask = var_38166_end_mask_0, x = var_37937_cast_fp16)[name = tensor("op_38166_cast_fp16")]; + tensor var_38173_begin_0 = const()[name = tensor("op_38173_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38173_end_0 = const()[name = tensor("op_38173_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38173_end_mask_0 = const()[name = tensor("op_38173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38173_cast_fp16 = slice_by_index(begin = var_38173_begin_0, end = var_38173_end_0, end_mask = var_38173_end_mask_0, x = var_37937_cast_fp16)[name = tensor("op_38173_cast_fp16")]; + tensor var_38180_begin_0 = const()[name = tensor("op_38180_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38180_end_0 = const()[name = tensor("op_38180_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38180_end_mask_0 = const()[name = tensor("op_38180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38180_cast_fp16 = slice_by_index(begin = var_38180_begin_0, end = var_38180_end_0, end_mask = var_38180_end_mask_0, x = var_37937_cast_fp16)[name = tensor("op_38180_cast_fp16")]; + tensor var_38187_begin_0 = const()[name = tensor("op_38187_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38187_end_0 = const()[name = tensor("op_38187_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38187_end_mask_0 = const()[name = tensor("op_38187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38187_cast_fp16 = slice_by_index(begin = var_38187_begin_0, end = var_38187_end_0, end_mask = var_38187_end_mask_0, x = var_37937_cast_fp16)[name = tensor("op_38187_cast_fp16")]; + tensor var_38194_begin_0 = const()[name = tensor("op_38194_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38194_end_0 = const()[name = tensor("op_38194_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38194_end_mask_0 = const()[name = tensor("op_38194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38194_cast_fp16 = slice_by_index(begin = var_38194_begin_0, end = var_38194_end_0, end_mask = var_38194_end_mask_0, x = var_37941_cast_fp16)[name = tensor("op_38194_cast_fp16")]; + tensor var_38201_begin_0 = const()[name = tensor("op_38201_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38201_end_0 = const()[name = tensor("op_38201_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38201_end_mask_0 = const()[name = tensor("op_38201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38201_cast_fp16 = slice_by_index(begin = var_38201_begin_0, end = var_38201_end_0, end_mask = var_38201_end_mask_0, x = var_37941_cast_fp16)[name = tensor("op_38201_cast_fp16")]; + tensor var_38208_begin_0 = const()[name = tensor("op_38208_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38208_end_0 = const()[name = tensor("op_38208_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38208_end_mask_0 = const()[name = tensor("op_38208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38208_cast_fp16 = slice_by_index(begin = var_38208_begin_0, end = var_38208_end_0, end_mask = var_38208_end_mask_0, x = var_37941_cast_fp16)[name = tensor("op_38208_cast_fp16")]; + tensor var_38215_begin_0 = const()[name = tensor("op_38215_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38215_end_0 = const()[name = tensor("op_38215_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38215_end_mask_0 = const()[name = tensor("op_38215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38215_cast_fp16 = slice_by_index(begin = var_38215_begin_0, end = var_38215_end_0, end_mask = var_38215_end_mask_0, x = var_37941_cast_fp16)[name = tensor("op_38215_cast_fp16")]; + tensor var_38222_begin_0 = const()[name = tensor("op_38222_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38222_end_0 = const()[name = tensor("op_38222_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38222_end_mask_0 = const()[name = tensor("op_38222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38222_cast_fp16 = slice_by_index(begin = var_38222_begin_0, end = var_38222_end_0, end_mask = var_38222_end_mask_0, x = var_37945_cast_fp16)[name = tensor("op_38222_cast_fp16")]; + tensor var_38229_begin_0 = const()[name = tensor("op_38229_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38229_end_0 = const()[name = tensor("op_38229_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38229_end_mask_0 = const()[name = tensor("op_38229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38229_cast_fp16 = slice_by_index(begin = var_38229_begin_0, end = var_38229_end_0, end_mask = var_38229_end_mask_0, x = var_37945_cast_fp16)[name = tensor("op_38229_cast_fp16")]; + tensor var_38236_begin_0 = const()[name = tensor("op_38236_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38236_end_0 = const()[name = tensor("op_38236_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38236_end_mask_0 = const()[name = tensor("op_38236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38236_cast_fp16 = slice_by_index(begin = var_38236_begin_0, end = var_38236_end_0, end_mask = var_38236_end_mask_0, x = var_37945_cast_fp16)[name = tensor("op_38236_cast_fp16")]; + tensor var_38243_begin_0 = const()[name = tensor("op_38243_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38243_end_0 = const()[name = tensor("op_38243_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38243_end_mask_0 = const()[name = tensor("op_38243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38243_cast_fp16 = slice_by_index(begin = var_38243_begin_0, end = var_38243_end_0, end_mask = var_38243_end_mask_0, x = var_37945_cast_fp16)[name = tensor("op_38243_cast_fp16")]; + tensor var_38250_begin_0 = const()[name = tensor("op_38250_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38250_end_0 = const()[name = tensor("op_38250_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38250_end_mask_0 = const()[name = tensor("op_38250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38250_cast_fp16 = slice_by_index(begin = var_38250_begin_0, end = var_38250_end_0, end_mask = var_38250_end_mask_0, x = var_37949_cast_fp16)[name = tensor("op_38250_cast_fp16")]; + tensor var_38257_begin_0 = const()[name = tensor("op_38257_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38257_end_0 = const()[name = tensor("op_38257_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38257_end_mask_0 = const()[name = tensor("op_38257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38257_cast_fp16 = slice_by_index(begin = var_38257_begin_0, end = var_38257_end_0, end_mask = var_38257_end_mask_0, x = var_37949_cast_fp16)[name = tensor("op_38257_cast_fp16")]; + tensor var_38264_begin_0 = const()[name = tensor("op_38264_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38264_end_0 = const()[name = tensor("op_38264_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38264_end_mask_0 = const()[name = tensor("op_38264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38264_cast_fp16 = slice_by_index(begin = var_38264_begin_0, end = var_38264_end_0, end_mask = var_38264_end_mask_0, x = var_37949_cast_fp16)[name = tensor("op_38264_cast_fp16")]; + tensor var_38271_begin_0 = const()[name = tensor("op_38271_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38271_end_0 = const()[name = tensor("op_38271_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38271_end_mask_0 = const()[name = tensor("op_38271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38271_cast_fp16 = slice_by_index(begin = var_38271_begin_0, end = var_38271_end_0, end_mask = var_38271_end_mask_0, x = var_37949_cast_fp16)[name = tensor("op_38271_cast_fp16")]; + tensor var_38278_begin_0 = const()[name = tensor("op_38278_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38278_end_0 = const()[name = tensor("op_38278_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38278_end_mask_0 = const()[name = tensor("op_38278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38278_cast_fp16 = slice_by_index(begin = var_38278_begin_0, end = var_38278_end_0, end_mask = var_38278_end_mask_0, x = var_37953_cast_fp16)[name = tensor("op_38278_cast_fp16")]; + tensor var_38285_begin_0 = const()[name = tensor("op_38285_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38285_end_0 = const()[name = tensor("op_38285_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38285_end_mask_0 = const()[name = tensor("op_38285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38285_cast_fp16 = slice_by_index(begin = var_38285_begin_0, end = var_38285_end_0, end_mask = var_38285_end_mask_0, x = var_37953_cast_fp16)[name = tensor("op_38285_cast_fp16")]; + tensor var_38292_begin_0 = const()[name = tensor("op_38292_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38292_end_0 = const()[name = tensor("op_38292_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38292_end_mask_0 = const()[name = tensor("op_38292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38292_cast_fp16 = slice_by_index(begin = var_38292_begin_0, end = var_38292_end_0, end_mask = var_38292_end_mask_0, x = var_37953_cast_fp16)[name = tensor("op_38292_cast_fp16")]; + tensor var_38299_begin_0 = const()[name = tensor("op_38299_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38299_end_0 = const()[name = tensor("op_38299_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38299_end_mask_0 = const()[name = tensor("op_38299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38299_cast_fp16 = slice_by_index(begin = var_38299_begin_0, end = var_38299_end_0, end_mask = var_38299_end_mask_0, x = var_37953_cast_fp16)[name = tensor("op_38299_cast_fp16")]; + tensor var_38306_begin_0 = const()[name = tensor("op_38306_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38306_end_0 = const()[name = tensor("op_38306_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38306_end_mask_0 = const()[name = tensor("op_38306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38306_cast_fp16 = slice_by_index(begin = var_38306_begin_0, end = var_38306_end_0, end_mask = var_38306_end_mask_0, x = var_37957_cast_fp16)[name = tensor("op_38306_cast_fp16")]; + tensor var_38313_begin_0 = const()[name = tensor("op_38313_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38313_end_0 = const()[name = tensor("op_38313_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38313_end_mask_0 = const()[name = tensor("op_38313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38313_cast_fp16 = slice_by_index(begin = var_38313_begin_0, end = var_38313_end_0, end_mask = var_38313_end_mask_0, x = var_37957_cast_fp16)[name = tensor("op_38313_cast_fp16")]; + tensor var_38320_begin_0 = const()[name = tensor("op_38320_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38320_end_0 = const()[name = tensor("op_38320_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38320_end_mask_0 = const()[name = tensor("op_38320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38320_cast_fp16 = slice_by_index(begin = var_38320_begin_0, end = var_38320_end_0, end_mask = var_38320_end_mask_0, x = var_37957_cast_fp16)[name = tensor("op_38320_cast_fp16")]; + tensor var_38327_begin_0 = const()[name = tensor("op_38327_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38327_end_0 = const()[name = tensor("op_38327_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38327_end_mask_0 = const()[name = tensor("op_38327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38327_cast_fp16 = slice_by_index(begin = var_38327_begin_0, end = var_38327_end_0, end_mask = var_38327_end_mask_0, x = var_37957_cast_fp16)[name = tensor("op_38327_cast_fp16")]; + tensor var_38334_begin_0 = const()[name = tensor("op_38334_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38334_end_0 = const()[name = tensor("op_38334_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38334_end_mask_0 = const()[name = tensor("op_38334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38334_cast_fp16 = slice_by_index(begin = var_38334_begin_0, end = var_38334_end_0, end_mask = var_38334_end_mask_0, x = var_37961_cast_fp16)[name = tensor("op_38334_cast_fp16")]; + tensor var_38341_begin_0 = const()[name = tensor("op_38341_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38341_end_0 = const()[name = tensor("op_38341_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38341_end_mask_0 = const()[name = tensor("op_38341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38341_cast_fp16 = slice_by_index(begin = var_38341_begin_0, end = var_38341_end_0, end_mask = var_38341_end_mask_0, x = var_37961_cast_fp16)[name = tensor("op_38341_cast_fp16")]; + tensor var_38348_begin_0 = const()[name = tensor("op_38348_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38348_end_0 = const()[name = tensor("op_38348_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38348_end_mask_0 = const()[name = tensor("op_38348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38348_cast_fp16 = slice_by_index(begin = var_38348_begin_0, end = var_38348_end_0, end_mask = var_38348_end_mask_0, x = var_37961_cast_fp16)[name = tensor("op_38348_cast_fp16")]; + tensor var_38355_begin_0 = const()[name = tensor("op_38355_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38355_end_0 = const()[name = tensor("op_38355_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38355_end_mask_0 = const()[name = tensor("op_38355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38355_cast_fp16 = slice_by_index(begin = var_38355_begin_0, end = var_38355_end_0, end_mask = var_38355_end_mask_0, x = var_37961_cast_fp16)[name = tensor("op_38355_cast_fp16")]; + tensor var_38362_begin_0 = const()[name = tensor("op_38362_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38362_end_0 = const()[name = tensor("op_38362_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38362_end_mask_0 = const()[name = tensor("op_38362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38362_cast_fp16 = slice_by_index(begin = var_38362_begin_0, end = var_38362_end_0, end_mask = var_38362_end_mask_0, x = var_37965_cast_fp16)[name = tensor("op_38362_cast_fp16")]; + tensor var_38369_begin_0 = const()[name = tensor("op_38369_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38369_end_0 = const()[name = tensor("op_38369_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38369_end_mask_0 = const()[name = tensor("op_38369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38369_cast_fp16 = slice_by_index(begin = var_38369_begin_0, end = var_38369_end_0, end_mask = var_38369_end_mask_0, x = var_37965_cast_fp16)[name = tensor("op_38369_cast_fp16")]; + tensor var_38376_begin_0 = const()[name = tensor("op_38376_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38376_end_0 = const()[name = tensor("op_38376_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38376_end_mask_0 = const()[name = tensor("op_38376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38376_cast_fp16 = slice_by_index(begin = var_38376_begin_0, end = var_38376_end_0, end_mask = var_38376_end_mask_0, x = var_37965_cast_fp16)[name = tensor("op_38376_cast_fp16")]; + tensor var_38383_begin_0 = const()[name = tensor("op_38383_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38383_end_0 = const()[name = tensor("op_38383_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38383_end_mask_0 = const()[name = tensor("op_38383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38383_cast_fp16 = slice_by_index(begin = var_38383_begin_0, end = var_38383_end_0, end_mask = var_38383_end_mask_0, x = var_37965_cast_fp16)[name = tensor("op_38383_cast_fp16")]; + tensor var_38390_begin_0 = const()[name = tensor("op_38390_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38390_end_0 = const()[name = tensor("op_38390_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38390_end_mask_0 = const()[name = tensor("op_38390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38390_cast_fp16 = slice_by_index(begin = var_38390_begin_0, end = var_38390_end_0, end_mask = var_38390_end_mask_0, x = var_37969_cast_fp16)[name = tensor("op_38390_cast_fp16")]; + tensor var_38397_begin_0 = const()[name = tensor("op_38397_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38397_end_0 = const()[name = tensor("op_38397_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38397_end_mask_0 = const()[name = tensor("op_38397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38397_cast_fp16 = slice_by_index(begin = var_38397_begin_0, end = var_38397_end_0, end_mask = var_38397_end_mask_0, x = var_37969_cast_fp16)[name = tensor("op_38397_cast_fp16")]; + tensor var_38404_begin_0 = const()[name = tensor("op_38404_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38404_end_0 = const()[name = tensor("op_38404_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38404_end_mask_0 = const()[name = tensor("op_38404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38404_cast_fp16 = slice_by_index(begin = var_38404_begin_0, end = var_38404_end_0, end_mask = var_38404_end_mask_0, x = var_37969_cast_fp16)[name = tensor("op_38404_cast_fp16")]; + tensor var_38411_begin_0 = const()[name = tensor("op_38411_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38411_end_0 = const()[name = tensor("op_38411_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38411_end_mask_0 = const()[name = tensor("op_38411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38411_cast_fp16 = slice_by_index(begin = var_38411_begin_0, end = var_38411_end_0, end_mask = var_38411_end_mask_0, x = var_37969_cast_fp16)[name = tensor("op_38411_cast_fp16")]; + tensor var_38418_begin_0 = const()[name = tensor("op_38418_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38418_end_0 = const()[name = tensor("op_38418_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38418_end_mask_0 = const()[name = tensor("op_38418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38418_cast_fp16 = slice_by_index(begin = var_38418_begin_0, end = var_38418_end_0, end_mask = var_38418_end_mask_0, x = var_37973_cast_fp16)[name = tensor("op_38418_cast_fp16")]; + tensor var_38425_begin_0 = const()[name = tensor("op_38425_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38425_end_0 = const()[name = tensor("op_38425_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38425_end_mask_0 = const()[name = tensor("op_38425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38425_cast_fp16 = slice_by_index(begin = var_38425_begin_0, end = var_38425_end_0, end_mask = var_38425_end_mask_0, x = var_37973_cast_fp16)[name = tensor("op_38425_cast_fp16")]; + tensor var_38432_begin_0 = const()[name = tensor("op_38432_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38432_end_0 = const()[name = tensor("op_38432_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38432_end_mask_0 = const()[name = tensor("op_38432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38432_cast_fp16 = slice_by_index(begin = var_38432_begin_0, end = var_38432_end_0, end_mask = var_38432_end_mask_0, x = var_37973_cast_fp16)[name = tensor("op_38432_cast_fp16")]; + tensor var_38439_begin_0 = const()[name = tensor("op_38439_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38439_end_0 = const()[name = tensor("op_38439_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38439_end_mask_0 = const()[name = tensor("op_38439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38439_cast_fp16 = slice_by_index(begin = var_38439_begin_0, end = var_38439_end_0, end_mask = var_38439_end_mask_0, x = var_37973_cast_fp16)[name = tensor("op_38439_cast_fp16")]; + tensor var_38446_begin_0 = const()[name = tensor("op_38446_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38446_end_0 = const()[name = tensor("op_38446_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38446_end_mask_0 = const()[name = tensor("op_38446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38446_cast_fp16 = slice_by_index(begin = var_38446_begin_0, end = var_38446_end_0, end_mask = var_38446_end_mask_0, x = var_37977_cast_fp16)[name = tensor("op_38446_cast_fp16")]; + tensor var_38453_begin_0 = const()[name = tensor("op_38453_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38453_end_0 = const()[name = tensor("op_38453_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38453_end_mask_0 = const()[name = tensor("op_38453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38453_cast_fp16 = slice_by_index(begin = var_38453_begin_0, end = var_38453_end_0, end_mask = var_38453_end_mask_0, x = var_37977_cast_fp16)[name = tensor("op_38453_cast_fp16")]; + tensor var_38460_begin_0 = const()[name = tensor("op_38460_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38460_end_0 = const()[name = tensor("op_38460_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38460_end_mask_0 = const()[name = tensor("op_38460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38460_cast_fp16 = slice_by_index(begin = var_38460_begin_0, end = var_38460_end_0, end_mask = var_38460_end_mask_0, x = var_37977_cast_fp16)[name = tensor("op_38460_cast_fp16")]; + tensor var_38467_begin_0 = const()[name = tensor("op_38467_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38467_end_0 = const()[name = tensor("op_38467_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38467_end_mask_0 = const()[name = tensor("op_38467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38467_cast_fp16 = slice_by_index(begin = var_38467_begin_0, end = var_38467_end_0, end_mask = var_38467_end_mask_0, x = var_37977_cast_fp16)[name = tensor("op_38467_cast_fp16")]; + tensor var_38474_begin_0 = const()[name = tensor("op_38474_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38474_end_0 = const()[name = tensor("op_38474_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38474_end_mask_0 = const()[name = tensor("op_38474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38474_cast_fp16 = slice_by_index(begin = var_38474_begin_0, end = var_38474_end_0, end_mask = var_38474_end_mask_0, x = var_37981_cast_fp16)[name = tensor("op_38474_cast_fp16")]; + tensor var_38481_begin_0 = const()[name = tensor("op_38481_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38481_end_0 = const()[name = tensor("op_38481_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38481_end_mask_0 = const()[name = tensor("op_38481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38481_cast_fp16 = slice_by_index(begin = var_38481_begin_0, end = var_38481_end_0, end_mask = var_38481_end_mask_0, x = var_37981_cast_fp16)[name = tensor("op_38481_cast_fp16")]; + tensor var_38488_begin_0 = const()[name = tensor("op_38488_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38488_end_0 = const()[name = tensor("op_38488_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38488_end_mask_0 = const()[name = tensor("op_38488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38488_cast_fp16 = slice_by_index(begin = var_38488_begin_0, end = var_38488_end_0, end_mask = var_38488_end_mask_0, x = var_37981_cast_fp16)[name = tensor("op_38488_cast_fp16")]; + tensor var_38495_begin_0 = const()[name = tensor("op_38495_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38495_end_0 = const()[name = tensor("op_38495_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38495_end_mask_0 = const()[name = tensor("op_38495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38495_cast_fp16 = slice_by_index(begin = var_38495_begin_0, end = var_38495_end_0, end_mask = var_38495_end_mask_0, x = var_37981_cast_fp16)[name = tensor("op_38495_cast_fp16")]; + tensor var_38502_begin_0 = const()[name = tensor("op_38502_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38502_end_0 = const()[name = tensor("op_38502_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38502_end_mask_0 = const()[name = tensor("op_38502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38502_cast_fp16 = slice_by_index(begin = var_38502_begin_0, end = var_38502_end_0, end_mask = var_38502_end_mask_0, x = var_37985_cast_fp16)[name = tensor("op_38502_cast_fp16")]; + tensor var_38509_begin_0 = const()[name = tensor("op_38509_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38509_end_0 = const()[name = tensor("op_38509_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38509_end_mask_0 = const()[name = tensor("op_38509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38509_cast_fp16 = slice_by_index(begin = var_38509_begin_0, end = var_38509_end_0, end_mask = var_38509_end_mask_0, x = var_37985_cast_fp16)[name = tensor("op_38509_cast_fp16")]; + tensor var_38516_begin_0 = const()[name = tensor("op_38516_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38516_end_0 = const()[name = tensor("op_38516_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38516_end_mask_0 = const()[name = tensor("op_38516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38516_cast_fp16 = slice_by_index(begin = var_38516_begin_0, end = var_38516_end_0, end_mask = var_38516_end_mask_0, x = var_37985_cast_fp16)[name = tensor("op_38516_cast_fp16")]; + tensor var_38523_begin_0 = const()[name = tensor("op_38523_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38523_end_0 = const()[name = tensor("op_38523_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38523_end_mask_0 = const()[name = tensor("op_38523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38523_cast_fp16 = slice_by_index(begin = var_38523_begin_0, end = var_38523_end_0, end_mask = var_38523_end_mask_0, x = var_37985_cast_fp16)[name = tensor("op_38523_cast_fp16")]; + tensor var_38530_begin_0 = const()[name = tensor("op_38530_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38530_end_0 = const()[name = tensor("op_38530_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_38530_end_mask_0 = const()[name = tensor("op_38530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38530_cast_fp16 = slice_by_index(begin = var_38530_begin_0, end = var_38530_end_0, end_mask = var_38530_end_mask_0, x = var_37989_cast_fp16)[name = tensor("op_38530_cast_fp16")]; + tensor var_38537_begin_0 = const()[name = tensor("op_38537_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_38537_end_0 = const()[name = tensor("op_38537_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_38537_end_mask_0 = const()[name = tensor("op_38537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38537_cast_fp16 = slice_by_index(begin = var_38537_begin_0, end = var_38537_end_0, end_mask = var_38537_end_mask_0, x = var_37989_cast_fp16)[name = tensor("op_38537_cast_fp16")]; + tensor var_38544_begin_0 = const()[name = tensor("op_38544_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_38544_end_0 = const()[name = tensor("op_38544_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_38544_end_mask_0 = const()[name = tensor("op_38544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38544_cast_fp16 = slice_by_index(begin = var_38544_begin_0, end = var_38544_end_0, end_mask = var_38544_end_mask_0, x = var_37989_cast_fp16)[name = tensor("op_38544_cast_fp16")]; + tensor var_38551_begin_0 = const()[name = tensor("op_38551_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_38551_end_0 = const()[name = tensor("op_38551_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38551_end_mask_0 = const()[name = tensor("op_38551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38551_cast_fp16 = slice_by_index(begin = var_38551_begin_0, end = var_38551_end_0, end_mask = var_38551_end_mask_0, x = var_37989_cast_fp16)[name = tensor("op_38551_cast_fp16")]; + tensor k_47_perm_0 = const()[name = tensor("k_47_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_38556_begin_0 = const()[name = tensor("op_38556_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38556_end_0 = const()[name = tensor("op_38556_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_38556_end_mask_0 = const()[name = tensor("op_38556_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_8 = transpose(perm = k_47_perm_0, x = key_47_cast_fp16)[name = tensor("transpose_8")]; + tensor var_38556_cast_fp16 = slice_by_index(begin = var_38556_begin_0, end = var_38556_end_0, end_mask = var_38556_end_mask_0, x = transpose_8)[name = tensor("op_38556_cast_fp16")]; + tensor var_38560_begin_0 = const()[name = tensor("op_38560_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_38560_end_0 = const()[name = tensor("op_38560_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_38560_end_mask_0 = const()[name = tensor("op_38560_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38560_cast_fp16 = slice_by_index(begin = var_38560_begin_0, end = var_38560_end_0, end_mask = var_38560_end_mask_0, x = transpose_8)[name = tensor("op_38560_cast_fp16")]; + tensor var_38564_begin_0 = const()[name = tensor("op_38564_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_38564_end_0 = const()[name = tensor("op_38564_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_38564_end_mask_0 = const()[name = tensor("op_38564_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38564_cast_fp16 = slice_by_index(begin = var_38564_begin_0, end = var_38564_end_0, end_mask = var_38564_end_mask_0, x = transpose_8)[name = tensor("op_38564_cast_fp16")]; + tensor var_38568_begin_0 = const()[name = tensor("op_38568_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_38568_end_0 = const()[name = tensor("op_38568_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_38568_end_mask_0 = const()[name = tensor("op_38568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38568_cast_fp16 = slice_by_index(begin = var_38568_begin_0, end = var_38568_end_0, end_mask = var_38568_end_mask_0, x = transpose_8)[name = tensor("op_38568_cast_fp16")]; + tensor var_38572_begin_0 = const()[name = tensor("op_38572_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_38572_end_0 = const()[name = tensor("op_38572_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_38572_end_mask_0 = const()[name = tensor("op_38572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38572_cast_fp16 = slice_by_index(begin = var_38572_begin_0, end = var_38572_end_0, end_mask = var_38572_end_mask_0, x = transpose_8)[name = tensor("op_38572_cast_fp16")]; + tensor var_38576_begin_0 = const()[name = tensor("op_38576_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_38576_end_0 = const()[name = tensor("op_38576_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_38576_end_mask_0 = const()[name = tensor("op_38576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38576_cast_fp16 = slice_by_index(begin = var_38576_begin_0, end = var_38576_end_0, end_mask = var_38576_end_mask_0, x = transpose_8)[name = tensor("op_38576_cast_fp16")]; + tensor var_38580_begin_0 = const()[name = tensor("op_38580_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_38580_end_0 = const()[name = tensor("op_38580_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_38580_end_mask_0 = const()[name = tensor("op_38580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38580_cast_fp16 = slice_by_index(begin = var_38580_begin_0, end = var_38580_end_0, end_mask = var_38580_end_mask_0, x = transpose_8)[name = tensor("op_38580_cast_fp16")]; + tensor var_38584_begin_0 = const()[name = tensor("op_38584_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_38584_end_0 = const()[name = tensor("op_38584_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_38584_end_mask_0 = const()[name = tensor("op_38584_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38584_cast_fp16 = slice_by_index(begin = var_38584_begin_0, end = var_38584_end_0, end_mask = var_38584_end_mask_0, x = transpose_8)[name = tensor("op_38584_cast_fp16")]; + tensor var_38588_begin_0 = const()[name = tensor("op_38588_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_38588_end_0 = const()[name = tensor("op_38588_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_38588_end_mask_0 = const()[name = tensor("op_38588_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38588_cast_fp16 = slice_by_index(begin = var_38588_begin_0, end = var_38588_end_0, end_mask = var_38588_end_mask_0, x = transpose_8)[name = tensor("op_38588_cast_fp16")]; + tensor var_38592_begin_0 = const()[name = tensor("op_38592_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_38592_end_0 = const()[name = tensor("op_38592_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_38592_end_mask_0 = const()[name = tensor("op_38592_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38592_cast_fp16 = slice_by_index(begin = var_38592_begin_0, end = var_38592_end_0, end_mask = var_38592_end_mask_0, x = transpose_8)[name = tensor("op_38592_cast_fp16")]; + tensor var_38596_begin_0 = const()[name = tensor("op_38596_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_38596_end_0 = const()[name = tensor("op_38596_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_38596_end_mask_0 = const()[name = tensor("op_38596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38596_cast_fp16 = slice_by_index(begin = var_38596_begin_0, end = var_38596_end_0, end_mask = var_38596_end_mask_0, x = transpose_8)[name = tensor("op_38596_cast_fp16")]; + tensor var_38600_begin_0 = const()[name = tensor("op_38600_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_38600_end_0 = const()[name = tensor("op_38600_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_38600_end_mask_0 = const()[name = tensor("op_38600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38600_cast_fp16 = slice_by_index(begin = var_38600_begin_0, end = var_38600_end_0, end_mask = var_38600_end_mask_0, x = transpose_8)[name = tensor("op_38600_cast_fp16")]; + tensor var_38604_begin_0 = const()[name = tensor("op_38604_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_38604_end_0 = const()[name = tensor("op_38604_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_38604_end_mask_0 = const()[name = tensor("op_38604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38604_cast_fp16 = slice_by_index(begin = var_38604_begin_0, end = var_38604_end_0, end_mask = var_38604_end_mask_0, x = transpose_8)[name = tensor("op_38604_cast_fp16")]; + tensor var_38608_begin_0 = const()[name = tensor("op_38608_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_38608_end_0 = const()[name = tensor("op_38608_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_38608_end_mask_0 = const()[name = tensor("op_38608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38608_cast_fp16 = slice_by_index(begin = var_38608_begin_0, end = var_38608_end_0, end_mask = var_38608_end_mask_0, x = transpose_8)[name = tensor("op_38608_cast_fp16")]; + tensor var_38612_begin_0 = const()[name = tensor("op_38612_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_38612_end_0 = const()[name = tensor("op_38612_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_38612_end_mask_0 = const()[name = tensor("op_38612_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38612_cast_fp16 = slice_by_index(begin = var_38612_begin_0, end = var_38612_end_0, end_mask = var_38612_end_mask_0, x = transpose_8)[name = tensor("op_38612_cast_fp16")]; + tensor var_38616_begin_0 = const()[name = tensor("op_38616_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_38616_end_0 = const()[name = tensor("op_38616_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_38616_end_mask_0 = const()[name = tensor("op_38616_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38616_cast_fp16 = slice_by_index(begin = var_38616_begin_0, end = var_38616_end_0, end_mask = var_38616_end_mask_0, x = transpose_8)[name = tensor("op_38616_cast_fp16")]; + tensor var_38620_begin_0 = const()[name = tensor("op_38620_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_38620_end_0 = const()[name = tensor("op_38620_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_38620_end_mask_0 = const()[name = tensor("op_38620_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38620_cast_fp16 = slice_by_index(begin = var_38620_begin_0, end = var_38620_end_0, end_mask = var_38620_end_mask_0, x = transpose_8)[name = tensor("op_38620_cast_fp16")]; + tensor var_38624_begin_0 = const()[name = tensor("op_38624_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_38624_end_0 = const()[name = tensor("op_38624_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_38624_end_mask_0 = const()[name = tensor("op_38624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38624_cast_fp16 = slice_by_index(begin = var_38624_begin_0, end = var_38624_end_0, end_mask = var_38624_end_mask_0, x = transpose_8)[name = tensor("op_38624_cast_fp16")]; + tensor var_38628_begin_0 = const()[name = tensor("op_38628_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_38628_end_0 = const()[name = tensor("op_38628_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_38628_end_mask_0 = const()[name = tensor("op_38628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38628_cast_fp16 = slice_by_index(begin = var_38628_begin_0, end = var_38628_end_0, end_mask = var_38628_end_mask_0, x = transpose_8)[name = tensor("op_38628_cast_fp16")]; + tensor var_38632_begin_0 = const()[name = tensor("op_38632_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_38632_end_0 = const()[name = tensor("op_38632_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_38632_end_mask_0 = const()[name = tensor("op_38632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_38632_cast_fp16 = slice_by_index(begin = var_38632_begin_0, end = var_38632_end_0, end_mask = var_38632_end_mask_0, x = transpose_8)[name = tensor("op_38632_cast_fp16")]; + tensor var_38634_begin_0 = const()[name = tensor("op_38634_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_38634_end_0 = const()[name = tensor("op_38634_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_38634_end_mask_0 = const()[name = tensor("op_38634_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38634_cast_fp16 = slice_by_index(begin = var_38634_begin_0, end = var_38634_end_0, end_mask = var_38634_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38634_cast_fp16")]; + tensor var_38638_begin_0 = const()[name = tensor("op_38638_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_38638_end_0 = const()[name = tensor("op_38638_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_38638_end_mask_0 = const()[name = tensor("op_38638_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38638_cast_fp16 = slice_by_index(begin = var_38638_begin_0, end = var_38638_end_0, end_mask = var_38638_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38638_cast_fp16")]; + tensor var_38642_begin_0 = const()[name = tensor("op_38642_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_38642_end_0 = const()[name = tensor("op_38642_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_38642_end_mask_0 = const()[name = tensor("op_38642_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38642_cast_fp16 = slice_by_index(begin = var_38642_begin_0, end = var_38642_end_0, end_mask = var_38642_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38642_cast_fp16")]; + tensor var_38646_begin_0 = const()[name = tensor("op_38646_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_38646_end_0 = const()[name = tensor("op_38646_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_38646_end_mask_0 = const()[name = tensor("op_38646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38646_cast_fp16 = slice_by_index(begin = var_38646_begin_0, end = var_38646_end_0, end_mask = var_38646_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38646_cast_fp16")]; + tensor var_38650_begin_0 = const()[name = tensor("op_38650_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_38650_end_0 = const()[name = tensor("op_38650_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_38650_end_mask_0 = const()[name = tensor("op_38650_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38650_cast_fp16 = slice_by_index(begin = var_38650_begin_0, end = var_38650_end_0, end_mask = var_38650_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38650_cast_fp16")]; + tensor var_38654_begin_0 = const()[name = tensor("op_38654_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_38654_end_0 = const()[name = tensor("op_38654_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_38654_end_mask_0 = const()[name = tensor("op_38654_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38654_cast_fp16 = slice_by_index(begin = var_38654_begin_0, end = var_38654_end_0, end_mask = var_38654_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38654_cast_fp16")]; + tensor var_38658_begin_0 = const()[name = tensor("op_38658_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_38658_end_0 = const()[name = tensor("op_38658_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_38658_end_mask_0 = const()[name = tensor("op_38658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38658_cast_fp16 = slice_by_index(begin = var_38658_begin_0, end = var_38658_end_0, end_mask = var_38658_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38658_cast_fp16")]; + tensor var_38662_begin_0 = const()[name = tensor("op_38662_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_38662_end_0 = const()[name = tensor("op_38662_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_38662_end_mask_0 = const()[name = tensor("op_38662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38662_cast_fp16 = slice_by_index(begin = var_38662_begin_0, end = var_38662_end_0, end_mask = var_38662_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38662_cast_fp16")]; + tensor var_38666_begin_0 = const()[name = tensor("op_38666_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_38666_end_0 = const()[name = tensor("op_38666_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_38666_end_mask_0 = const()[name = tensor("op_38666_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38666_cast_fp16 = slice_by_index(begin = var_38666_begin_0, end = var_38666_end_0, end_mask = var_38666_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38666_cast_fp16")]; + tensor var_38670_begin_0 = const()[name = tensor("op_38670_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_38670_end_0 = const()[name = tensor("op_38670_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_38670_end_mask_0 = const()[name = tensor("op_38670_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38670_cast_fp16 = slice_by_index(begin = var_38670_begin_0, end = var_38670_end_0, end_mask = var_38670_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38670_cast_fp16")]; + tensor var_38674_begin_0 = const()[name = tensor("op_38674_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_38674_end_0 = const()[name = tensor("op_38674_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_38674_end_mask_0 = const()[name = tensor("op_38674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38674_cast_fp16 = slice_by_index(begin = var_38674_begin_0, end = var_38674_end_0, end_mask = var_38674_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38674_cast_fp16")]; + tensor var_38678_begin_0 = const()[name = tensor("op_38678_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_38678_end_0 = const()[name = tensor("op_38678_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_38678_end_mask_0 = const()[name = tensor("op_38678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38678_cast_fp16 = slice_by_index(begin = var_38678_begin_0, end = var_38678_end_0, end_mask = var_38678_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38678_cast_fp16")]; + tensor var_38682_begin_0 = const()[name = tensor("op_38682_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_38682_end_0 = const()[name = tensor("op_38682_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_38682_end_mask_0 = const()[name = tensor("op_38682_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38682_cast_fp16 = slice_by_index(begin = var_38682_begin_0, end = var_38682_end_0, end_mask = var_38682_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38682_cast_fp16")]; + tensor var_38686_begin_0 = const()[name = tensor("op_38686_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_38686_end_0 = const()[name = tensor("op_38686_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_38686_end_mask_0 = const()[name = tensor("op_38686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38686_cast_fp16 = slice_by_index(begin = var_38686_begin_0, end = var_38686_end_0, end_mask = var_38686_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38686_cast_fp16")]; + tensor var_38690_begin_0 = const()[name = tensor("op_38690_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_38690_end_0 = const()[name = tensor("op_38690_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_38690_end_mask_0 = const()[name = tensor("op_38690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38690_cast_fp16 = slice_by_index(begin = var_38690_begin_0, end = var_38690_end_0, end_mask = var_38690_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38690_cast_fp16")]; + tensor var_38694_begin_0 = const()[name = tensor("op_38694_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_38694_end_0 = const()[name = tensor("op_38694_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_38694_end_mask_0 = const()[name = tensor("op_38694_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38694_cast_fp16 = slice_by_index(begin = var_38694_begin_0, end = var_38694_end_0, end_mask = var_38694_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38694_cast_fp16")]; + tensor var_38698_begin_0 = const()[name = tensor("op_38698_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_38698_end_0 = const()[name = tensor("op_38698_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_38698_end_mask_0 = const()[name = tensor("op_38698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38698_cast_fp16 = slice_by_index(begin = var_38698_begin_0, end = var_38698_end_0, end_mask = var_38698_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38698_cast_fp16")]; + tensor var_38702_begin_0 = const()[name = tensor("op_38702_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_38702_end_0 = const()[name = tensor("op_38702_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_38702_end_mask_0 = const()[name = tensor("op_38702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38702_cast_fp16 = slice_by_index(begin = var_38702_begin_0, end = var_38702_end_0, end_mask = var_38702_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38702_cast_fp16")]; + tensor var_38706_begin_0 = const()[name = tensor("op_38706_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_38706_end_0 = const()[name = tensor("op_38706_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_38706_end_mask_0 = const()[name = tensor("op_38706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38706_cast_fp16 = slice_by_index(begin = var_38706_begin_0, end = var_38706_end_0, end_mask = var_38706_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38706_cast_fp16")]; + tensor var_38710_begin_0 = const()[name = tensor("op_38710_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_38710_end_0 = const()[name = tensor("op_38710_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_38710_end_mask_0 = const()[name = tensor("op_38710_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_38710_cast_fp16 = slice_by_index(begin = var_38710_begin_0, end = var_38710_end_0, end_mask = var_38710_end_mask_0, x = value_47_cast_fp16)[name = tensor("op_38710_cast_fp16")]; + tensor var_38714_equation_0 = const()[name = tensor("op_38714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38714_cast_fp16 = einsum(equation = var_38714_equation_0, values = (var_38556_cast_fp16, var_37998_cast_fp16))[name = tensor("op_38714_cast_fp16")]; + tensor var_38715_to_fp16 = const()[name = tensor("op_38715_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3681_cast_fp16 = mul(x = var_38714_cast_fp16, y = var_38715_to_fp16)[name = tensor("aw_chunk_3681_cast_fp16")]; + tensor var_38718_equation_0 = const()[name = tensor("op_38718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38718_cast_fp16 = einsum(equation = var_38718_equation_0, values = (var_38556_cast_fp16, var_38005_cast_fp16))[name = tensor("op_38718_cast_fp16")]; + tensor var_38719_to_fp16 = const()[name = tensor("op_38719_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3683_cast_fp16 = mul(x = var_38718_cast_fp16, y = var_38719_to_fp16)[name = tensor("aw_chunk_3683_cast_fp16")]; + tensor var_38722_equation_0 = const()[name = tensor("op_38722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38722_cast_fp16 = einsum(equation = var_38722_equation_0, values = (var_38556_cast_fp16, var_38012_cast_fp16))[name = tensor("op_38722_cast_fp16")]; + tensor var_38723_to_fp16 = const()[name = tensor("op_38723_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3685_cast_fp16 = mul(x = var_38722_cast_fp16, y = var_38723_to_fp16)[name = tensor("aw_chunk_3685_cast_fp16")]; + tensor var_38726_equation_0 = const()[name = tensor("op_38726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38726_cast_fp16 = einsum(equation = var_38726_equation_0, values = (var_38556_cast_fp16, var_38019_cast_fp16))[name = tensor("op_38726_cast_fp16")]; + tensor var_38727_to_fp16 = const()[name = tensor("op_38727_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3687_cast_fp16 = mul(x = var_38726_cast_fp16, y = var_38727_to_fp16)[name = tensor("aw_chunk_3687_cast_fp16")]; + tensor var_38730_equation_0 = const()[name = tensor("op_38730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38730_cast_fp16 = einsum(equation = var_38730_equation_0, values = (var_38560_cast_fp16, var_38026_cast_fp16))[name = tensor("op_38730_cast_fp16")]; + tensor var_38731_to_fp16 = const()[name = tensor("op_38731_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3689_cast_fp16 = mul(x = var_38730_cast_fp16, y = var_38731_to_fp16)[name = tensor("aw_chunk_3689_cast_fp16")]; + tensor var_38734_equation_0 = const()[name = tensor("op_38734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38734_cast_fp16 = einsum(equation = var_38734_equation_0, values = (var_38560_cast_fp16, var_38033_cast_fp16))[name = tensor("op_38734_cast_fp16")]; + tensor var_38735_to_fp16 = const()[name = tensor("op_38735_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3691_cast_fp16 = mul(x = var_38734_cast_fp16, y = var_38735_to_fp16)[name = tensor("aw_chunk_3691_cast_fp16")]; + tensor var_38738_equation_0 = const()[name = tensor("op_38738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38738_cast_fp16 = einsum(equation = var_38738_equation_0, values = (var_38560_cast_fp16, var_38040_cast_fp16))[name = tensor("op_38738_cast_fp16")]; + tensor var_38739_to_fp16 = const()[name = tensor("op_38739_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3693_cast_fp16 = mul(x = var_38738_cast_fp16, y = var_38739_to_fp16)[name = tensor("aw_chunk_3693_cast_fp16")]; + tensor var_38742_equation_0 = const()[name = tensor("op_38742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38742_cast_fp16 = einsum(equation = var_38742_equation_0, values = (var_38560_cast_fp16, var_38047_cast_fp16))[name = tensor("op_38742_cast_fp16")]; + tensor var_38743_to_fp16 = const()[name = tensor("op_38743_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3695_cast_fp16 = mul(x = var_38742_cast_fp16, y = var_38743_to_fp16)[name = tensor("aw_chunk_3695_cast_fp16")]; + tensor var_38746_equation_0 = const()[name = tensor("op_38746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38746_cast_fp16 = einsum(equation = var_38746_equation_0, values = (var_38564_cast_fp16, var_38054_cast_fp16))[name = tensor("op_38746_cast_fp16")]; + tensor var_38747_to_fp16 = const()[name = tensor("op_38747_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3697_cast_fp16 = mul(x = var_38746_cast_fp16, y = var_38747_to_fp16)[name = tensor("aw_chunk_3697_cast_fp16")]; + tensor var_38750_equation_0 = const()[name = tensor("op_38750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38750_cast_fp16 = einsum(equation = var_38750_equation_0, values = (var_38564_cast_fp16, var_38061_cast_fp16))[name = tensor("op_38750_cast_fp16")]; + tensor var_38751_to_fp16 = const()[name = tensor("op_38751_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3699_cast_fp16 = mul(x = var_38750_cast_fp16, y = var_38751_to_fp16)[name = tensor("aw_chunk_3699_cast_fp16")]; + tensor var_38754_equation_0 = const()[name = tensor("op_38754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38754_cast_fp16 = einsum(equation = var_38754_equation_0, values = (var_38564_cast_fp16, var_38068_cast_fp16))[name = tensor("op_38754_cast_fp16")]; + tensor var_38755_to_fp16 = const()[name = tensor("op_38755_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3701_cast_fp16 = mul(x = var_38754_cast_fp16, y = var_38755_to_fp16)[name = tensor("aw_chunk_3701_cast_fp16")]; + tensor var_38758_equation_0 = const()[name = tensor("op_38758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38758_cast_fp16 = einsum(equation = var_38758_equation_0, values = (var_38564_cast_fp16, var_38075_cast_fp16))[name = tensor("op_38758_cast_fp16")]; + tensor var_38759_to_fp16 = const()[name = tensor("op_38759_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3703_cast_fp16 = mul(x = var_38758_cast_fp16, y = var_38759_to_fp16)[name = tensor("aw_chunk_3703_cast_fp16")]; + tensor var_38762_equation_0 = const()[name = tensor("op_38762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38762_cast_fp16 = einsum(equation = var_38762_equation_0, values = (var_38568_cast_fp16, var_38082_cast_fp16))[name = tensor("op_38762_cast_fp16")]; + tensor var_38763_to_fp16 = const()[name = tensor("op_38763_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3705_cast_fp16 = mul(x = var_38762_cast_fp16, y = var_38763_to_fp16)[name = tensor("aw_chunk_3705_cast_fp16")]; + tensor var_38766_equation_0 = const()[name = tensor("op_38766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38766_cast_fp16 = einsum(equation = var_38766_equation_0, values = (var_38568_cast_fp16, var_38089_cast_fp16))[name = tensor("op_38766_cast_fp16")]; + tensor var_38767_to_fp16 = const()[name = tensor("op_38767_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3707_cast_fp16 = mul(x = var_38766_cast_fp16, y = var_38767_to_fp16)[name = tensor("aw_chunk_3707_cast_fp16")]; + tensor var_38770_equation_0 = const()[name = tensor("op_38770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38770_cast_fp16 = einsum(equation = var_38770_equation_0, values = (var_38568_cast_fp16, var_38096_cast_fp16))[name = tensor("op_38770_cast_fp16")]; + tensor var_38771_to_fp16 = const()[name = tensor("op_38771_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3709_cast_fp16 = mul(x = var_38770_cast_fp16, y = var_38771_to_fp16)[name = tensor("aw_chunk_3709_cast_fp16")]; + tensor var_38774_equation_0 = const()[name = tensor("op_38774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38774_cast_fp16 = einsum(equation = var_38774_equation_0, values = (var_38568_cast_fp16, var_38103_cast_fp16))[name = tensor("op_38774_cast_fp16")]; + tensor var_38775_to_fp16 = const()[name = tensor("op_38775_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3711_cast_fp16 = mul(x = var_38774_cast_fp16, y = var_38775_to_fp16)[name = tensor("aw_chunk_3711_cast_fp16")]; + tensor var_38778_equation_0 = const()[name = tensor("op_38778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38778_cast_fp16 = einsum(equation = var_38778_equation_0, values = (var_38572_cast_fp16, var_38110_cast_fp16))[name = tensor("op_38778_cast_fp16")]; + tensor var_38779_to_fp16 = const()[name = tensor("op_38779_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3713_cast_fp16 = mul(x = var_38778_cast_fp16, y = var_38779_to_fp16)[name = tensor("aw_chunk_3713_cast_fp16")]; + tensor var_38782_equation_0 = const()[name = tensor("op_38782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38782_cast_fp16 = einsum(equation = var_38782_equation_0, values = (var_38572_cast_fp16, var_38117_cast_fp16))[name = tensor("op_38782_cast_fp16")]; + tensor var_38783_to_fp16 = const()[name = tensor("op_38783_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3715_cast_fp16 = mul(x = var_38782_cast_fp16, y = var_38783_to_fp16)[name = tensor("aw_chunk_3715_cast_fp16")]; + tensor var_38786_equation_0 = const()[name = tensor("op_38786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38786_cast_fp16 = einsum(equation = var_38786_equation_0, values = (var_38572_cast_fp16, var_38124_cast_fp16))[name = tensor("op_38786_cast_fp16")]; + tensor var_38787_to_fp16 = const()[name = tensor("op_38787_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3717_cast_fp16 = mul(x = var_38786_cast_fp16, y = var_38787_to_fp16)[name = tensor("aw_chunk_3717_cast_fp16")]; + tensor var_38790_equation_0 = const()[name = tensor("op_38790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38790_cast_fp16 = einsum(equation = var_38790_equation_0, values = (var_38572_cast_fp16, var_38131_cast_fp16))[name = tensor("op_38790_cast_fp16")]; + tensor var_38791_to_fp16 = const()[name = tensor("op_38791_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3719_cast_fp16 = mul(x = var_38790_cast_fp16, y = var_38791_to_fp16)[name = tensor("aw_chunk_3719_cast_fp16")]; + tensor var_38794_equation_0 = const()[name = tensor("op_38794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38794_cast_fp16 = einsum(equation = var_38794_equation_0, values = (var_38576_cast_fp16, var_38138_cast_fp16))[name = tensor("op_38794_cast_fp16")]; + tensor var_38795_to_fp16 = const()[name = tensor("op_38795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3721_cast_fp16 = mul(x = var_38794_cast_fp16, y = var_38795_to_fp16)[name = tensor("aw_chunk_3721_cast_fp16")]; + tensor var_38798_equation_0 = const()[name = tensor("op_38798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38798_cast_fp16 = einsum(equation = var_38798_equation_0, values = (var_38576_cast_fp16, var_38145_cast_fp16))[name = tensor("op_38798_cast_fp16")]; + tensor var_38799_to_fp16 = const()[name = tensor("op_38799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3723_cast_fp16 = mul(x = var_38798_cast_fp16, y = var_38799_to_fp16)[name = tensor("aw_chunk_3723_cast_fp16")]; + tensor var_38802_equation_0 = const()[name = tensor("op_38802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38802_cast_fp16 = einsum(equation = var_38802_equation_0, values = (var_38576_cast_fp16, var_38152_cast_fp16))[name = tensor("op_38802_cast_fp16")]; + tensor var_38803_to_fp16 = const()[name = tensor("op_38803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3725_cast_fp16 = mul(x = var_38802_cast_fp16, y = var_38803_to_fp16)[name = tensor("aw_chunk_3725_cast_fp16")]; + tensor var_38806_equation_0 = const()[name = tensor("op_38806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38806_cast_fp16 = einsum(equation = var_38806_equation_0, values = (var_38576_cast_fp16, var_38159_cast_fp16))[name = tensor("op_38806_cast_fp16")]; + tensor var_38807_to_fp16 = const()[name = tensor("op_38807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3727_cast_fp16 = mul(x = var_38806_cast_fp16, y = var_38807_to_fp16)[name = tensor("aw_chunk_3727_cast_fp16")]; + tensor var_38810_equation_0 = const()[name = tensor("op_38810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38810_cast_fp16 = einsum(equation = var_38810_equation_0, values = (var_38580_cast_fp16, var_38166_cast_fp16))[name = tensor("op_38810_cast_fp16")]; + tensor var_38811_to_fp16 = const()[name = tensor("op_38811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3729_cast_fp16 = mul(x = var_38810_cast_fp16, y = var_38811_to_fp16)[name = tensor("aw_chunk_3729_cast_fp16")]; + tensor var_38814_equation_0 = const()[name = tensor("op_38814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38814_cast_fp16 = einsum(equation = var_38814_equation_0, values = (var_38580_cast_fp16, var_38173_cast_fp16))[name = tensor("op_38814_cast_fp16")]; + tensor var_38815_to_fp16 = const()[name = tensor("op_38815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3731_cast_fp16 = mul(x = var_38814_cast_fp16, y = var_38815_to_fp16)[name = tensor("aw_chunk_3731_cast_fp16")]; + tensor var_38818_equation_0 = const()[name = tensor("op_38818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38818_cast_fp16 = einsum(equation = var_38818_equation_0, values = (var_38580_cast_fp16, var_38180_cast_fp16))[name = tensor("op_38818_cast_fp16")]; + tensor var_38819_to_fp16 = const()[name = tensor("op_38819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3733_cast_fp16 = mul(x = var_38818_cast_fp16, y = var_38819_to_fp16)[name = tensor("aw_chunk_3733_cast_fp16")]; + tensor var_38822_equation_0 = const()[name = tensor("op_38822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38822_cast_fp16 = einsum(equation = var_38822_equation_0, values = (var_38580_cast_fp16, var_38187_cast_fp16))[name = tensor("op_38822_cast_fp16")]; + tensor var_38823_to_fp16 = const()[name = tensor("op_38823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3735_cast_fp16 = mul(x = var_38822_cast_fp16, y = var_38823_to_fp16)[name = tensor("aw_chunk_3735_cast_fp16")]; + tensor var_38826_equation_0 = const()[name = tensor("op_38826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38826_cast_fp16 = einsum(equation = var_38826_equation_0, values = (var_38584_cast_fp16, var_38194_cast_fp16))[name = tensor("op_38826_cast_fp16")]; + tensor var_38827_to_fp16 = const()[name = tensor("op_38827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3737_cast_fp16 = mul(x = var_38826_cast_fp16, y = var_38827_to_fp16)[name = tensor("aw_chunk_3737_cast_fp16")]; + tensor var_38830_equation_0 = const()[name = tensor("op_38830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38830_cast_fp16 = einsum(equation = var_38830_equation_0, values = (var_38584_cast_fp16, var_38201_cast_fp16))[name = tensor("op_38830_cast_fp16")]; + tensor var_38831_to_fp16 = const()[name = tensor("op_38831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3739_cast_fp16 = mul(x = var_38830_cast_fp16, y = var_38831_to_fp16)[name = tensor("aw_chunk_3739_cast_fp16")]; + tensor var_38834_equation_0 = const()[name = tensor("op_38834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38834_cast_fp16 = einsum(equation = var_38834_equation_0, values = (var_38584_cast_fp16, var_38208_cast_fp16))[name = tensor("op_38834_cast_fp16")]; + tensor var_38835_to_fp16 = const()[name = tensor("op_38835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3741_cast_fp16 = mul(x = var_38834_cast_fp16, y = var_38835_to_fp16)[name = tensor("aw_chunk_3741_cast_fp16")]; + tensor var_38838_equation_0 = const()[name = tensor("op_38838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38838_cast_fp16 = einsum(equation = var_38838_equation_0, values = (var_38584_cast_fp16, var_38215_cast_fp16))[name = tensor("op_38838_cast_fp16")]; + tensor var_38839_to_fp16 = const()[name = tensor("op_38839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3743_cast_fp16 = mul(x = var_38838_cast_fp16, y = var_38839_to_fp16)[name = tensor("aw_chunk_3743_cast_fp16")]; + tensor var_38842_equation_0 = const()[name = tensor("op_38842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38842_cast_fp16 = einsum(equation = var_38842_equation_0, values = (var_38588_cast_fp16, var_38222_cast_fp16))[name = tensor("op_38842_cast_fp16")]; + tensor var_38843_to_fp16 = const()[name = tensor("op_38843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3745_cast_fp16 = mul(x = var_38842_cast_fp16, y = var_38843_to_fp16)[name = tensor("aw_chunk_3745_cast_fp16")]; + tensor var_38846_equation_0 = const()[name = tensor("op_38846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38846_cast_fp16 = einsum(equation = var_38846_equation_0, values = (var_38588_cast_fp16, var_38229_cast_fp16))[name = tensor("op_38846_cast_fp16")]; + tensor var_38847_to_fp16 = const()[name = tensor("op_38847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3747_cast_fp16 = mul(x = var_38846_cast_fp16, y = var_38847_to_fp16)[name = tensor("aw_chunk_3747_cast_fp16")]; + tensor var_38850_equation_0 = const()[name = tensor("op_38850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38850_cast_fp16 = einsum(equation = var_38850_equation_0, values = (var_38588_cast_fp16, var_38236_cast_fp16))[name = tensor("op_38850_cast_fp16")]; + tensor var_38851_to_fp16 = const()[name = tensor("op_38851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3749_cast_fp16 = mul(x = var_38850_cast_fp16, y = var_38851_to_fp16)[name = tensor("aw_chunk_3749_cast_fp16")]; + tensor var_38854_equation_0 = const()[name = tensor("op_38854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38854_cast_fp16 = einsum(equation = var_38854_equation_0, values = (var_38588_cast_fp16, var_38243_cast_fp16))[name = tensor("op_38854_cast_fp16")]; + tensor var_38855_to_fp16 = const()[name = tensor("op_38855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3751_cast_fp16 = mul(x = var_38854_cast_fp16, y = var_38855_to_fp16)[name = tensor("aw_chunk_3751_cast_fp16")]; + tensor var_38858_equation_0 = const()[name = tensor("op_38858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38858_cast_fp16 = einsum(equation = var_38858_equation_0, values = (var_38592_cast_fp16, var_38250_cast_fp16))[name = tensor("op_38858_cast_fp16")]; + tensor var_38859_to_fp16 = const()[name = tensor("op_38859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3753_cast_fp16 = mul(x = var_38858_cast_fp16, y = var_38859_to_fp16)[name = tensor("aw_chunk_3753_cast_fp16")]; + tensor var_38862_equation_0 = const()[name = tensor("op_38862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38862_cast_fp16 = einsum(equation = var_38862_equation_0, values = (var_38592_cast_fp16, var_38257_cast_fp16))[name = tensor("op_38862_cast_fp16")]; + tensor var_38863_to_fp16 = const()[name = tensor("op_38863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3755_cast_fp16 = mul(x = var_38862_cast_fp16, y = var_38863_to_fp16)[name = tensor("aw_chunk_3755_cast_fp16")]; + tensor var_38866_equation_0 = const()[name = tensor("op_38866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38866_cast_fp16 = einsum(equation = var_38866_equation_0, values = (var_38592_cast_fp16, var_38264_cast_fp16))[name = tensor("op_38866_cast_fp16")]; + tensor var_38867_to_fp16 = const()[name = tensor("op_38867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3757_cast_fp16 = mul(x = var_38866_cast_fp16, y = var_38867_to_fp16)[name = tensor("aw_chunk_3757_cast_fp16")]; + tensor var_38870_equation_0 = const()[name = tensor("op_38870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38870_cast_fp16 = einsum(equation = var_38870_equation_0, values = (var_38592_cast_fp16, var_38271_cast_fp16))[name = tensor("op_38870_cast_fp16")]; + tensor var_38871_to_fp16 = const()[name = tensor("op_38871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3759_cast_fp16 = mul(x = var_38870_cast_fp16, y = var_38871_to_fp16)[name = tensor("aw_chunk_3759_cast_fp16")]; + tensor var_38874_equation_0 = const()[name = tensor("op_38874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38874_cast_fp16 = einsum(equation = var_38874_equation_0, values = (var_38596_cast_fp16, var_38278_cast_fp16))[name = tensor("op_38874_cast_fp16")]; + tensor var_38875_to_fp16 = const()[name = tensor("op_38875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3761_cast_fp16 = mul(x = var_38874_cast_fp16, y = var_38875_to_fp16)[name = tensor("aw_chunk_3761_cast_fp16")]; + tensor var_38878_equation_0 = const()[name = tensor("op_38878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38878_cast_fp16 = einsum(equation = var_38878_equation_0, values = (var_38596_cast_fp16, var_38285_cast_fp16))[name = tensor("op_38878_cast_fp16")]; + tensor var_38879_to_fp16 = const()[name = tensor("op_38879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3763_cast_fp16 = mul(x = var_38878_cast_fp16, y = var_38879_to_fp16)[name = tensor("aw_chunk_3763_cast_fp16")]; + tensor var_38882_equation_0 = const()[name = tensor("op_38882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38882_cast_fp16 = einsum(equation = var_38882_equation_0, values = (var_38596_cast_fp16, var_38292_cast_fp16))[name = tensor("op_38882_cast_fp16")]; + tensor var_38883_to_fp16 = const()[name = tensor("op_38883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3765_cast_fp16 = mul(x = var_38882_cast_fp16, y = var_38883_to_fp16)[name = tensor("aw_chunk_3765_cast_fp16")]; + tensor var_38886_equation_0 = const()[name = tensor("op_38886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38886_cast_fp16 = einsum(equation = var_38886_equation_0, values = (var_38596_cast_fp16, var_38299_cast_fp16))[name = tensor("op_38886_cast_fp16")]; + tensor var_38887_to_fp16 = const()[name = tensor("op_38887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3767_cast_fp16 = mul(x = var_38886_cast_fp16, y = var_38887_to_fp16)[name = tensor("aw_chunk_3767_cast_fp16")]; + tensor var_38890_equation_0 = const()[name = tensor("op_38890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38890_cast_fp16 = einsum(equation = var_38890_equation_0, values = (var_38600_cast_fp16, var_38306_cast_fp16))[name = tensor("op_38890_cast_fp16")]; + tensor var_38891_to_fp16 = const()[name = tensor("op_38891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3769_cast_fp16 = mul(x = var_38890_cast_fp16, y = var_38891_to_fp16)[name = tensor("aw_chunk_3769_cast_fp16")]; + tensor var_38894_equation_0 = const()[name = tensor("op_38894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38894_cast_fp16 = einsum(equation = var_38894_equation_0, values = (var_38600_cast_fp16, var_38313_cast_fp16))[name = tensor("op_38894_cast_fp16")]; + tensor var_38895_to_fp16 = const()[name = tensor("op_38895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3771_cast_fp16 = mul(x = var_38894_cast_fp16, y = var_38895_to_fp16)[name = tensor("aw_chunk_3771_cast_fp16")]; + tensor var_38898_equation_0 = const()[name = tensor("op_38898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38898_cast_fp16 = einsum(equation = var_38898_equation_0, values = (var_38600_cast_fp16, var_38320_cast_fp16))[name = tensor("op_38898_cast_fp16")]; + tensor var_38899_to_fp16 = const()[name = tensor("op_38899_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3773_cast_fp16 = mul(x = var_38898_cast_fp16, y = var_38899_to_fp16)[name = tensor("aw_chunk_3773_cast_fp16")]; + tensor var_38902_equation_0 = const()[name = tensor("op_38902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38902_cast_fp16 = einsum(equation = var_38902_equation_0, values = (var_38600_cast_fp16, var_38327_cast_fp16))[name = tensor("op_38902_cast_fp16")]; + tensor var_38903_to_fp16 = const()[name = tensor("op_38903_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3775_cast_fp16 = mul(x = var_38902_cast_fp16, y = var_38903_to_fp16)[name = tensor("aw_chunk_3775_cast_fp16")]; + tensor var_38906_equation_0 = const()[name = tensor("op_38906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38906_cast_fp16 = einsum(equation = var_38906_equation_0, values = (var_38604_cast_fp16, var_38334_cast_fp16))[name = tensor("op_38906_cast_fp16")]; + tensor var_38907_to_fp16 = const()[name = tensor("op_38907_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3777_cast_fp16 = mul(x = var_38906_cast_fp16, y = var_38907_to_fp16)[name = tensor("aw_chunk_3777_cast_fp16")]; + tensor var_38910_equation_0 = const()[name = tensor("op_38910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38910_cast_fp16 = einsum(equation = var_38910_equation_0, values = (var_38604_cast_fp16, var_38341_cast_fp16))[name = tensor("op_38910_cast_fp16")]; + tensor var_38911_to_fp16 = const()[name = tensor("op_38911_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3779_cast_fp16 = mul(x = var_38910_cast_fp16, y = var_38911_to_fp16)[name = tensor("aw_chunk_3779_cast_fp16")]; + tensor var_38914_equation_0 = const()[name = tensor("op_38914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38914_cast_fp16 = einsum(equation = var_38914_equation_0, values = (var_38604_cast_fp16, var_38348_cast_fp16))[name = tensor("op_38914_cast_fp16")]; + tensor var_38915_to_fp16 = const()[name = tensor("op_38915_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3781_cast_fp16 = mul(x = var_38914_cast_fp16, y = var_38915_to_fp16)[name = tensor("aw_chunk_3781_cast_fp16")]; + tensor var_38918_equation_0 = const()[name = tensor("op_38918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38918_cast_fp16 = einsum(equation = var_38918_equation_0, values = (var_38604_cast_fp16, var_38355_cast_fp16))[name = tensor("op_38918_cast_fp16")]; + tensor var_38919_to_fp16 = const()[name = tensor("op_38919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3783_cast_fp16 = mul(x = var_38918_cast_fp16, y = var_38919_to_fp16)[name = tensor("aw_chunk_3783_cast_fp16")]; + tensor var_38922_equation_0 = const()[name = tensor("op_38922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38922_cast_fp16 = einsum(equation = var_38922_equation_0, values = (var_38608_cast_fp16, var_38362_cast_fp16))[name = tensor("op_38922_cast_fp16")]; + tensor var_38923_to_fp16 = const()[name = tensor("op_38923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3785_cast_fp16 = mul(x = var_38922_cast_fp16, y = var_38923_to_fp16)[name = tensor("aw_chunk_3785_cast_fp16")]; + tensor var_38926_equation_0 = const()[name = tensor("op_38926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38926_cast_fp16 = einsum(equation = var_38926_equation_0, values = (var_38608_cast_fp16, var_38369_cast_fp16))[name = tensor("op_38926_cast_fp16")]; + tensor var_38927_to_fp16 = const()[name = tensor("op_38927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3787_cast_fp16 = mul(x = var_38926_cast_fp16, y = var_38927_to_fp16)[name = tensor("aw_chunk_3787_cast_fp16")]; + tensor var_38930_equation_0 = const()[name = tensor("op_38930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38930_cast_fp16 = einsum(equation = var_38930_equation_0, values = (var_38608_cast_fp16, var_38376_cast_fp16))[name = tensor("op_38930_cast_fp16")]; + tensor var_38931_to_fp16 = const()[name = tensor("op_38931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3789_cast_fp16 = mul(x = var_38930_cast_fp16, y = var_38931_to_fp16)[name = tensor("aw_chunk_3789_cast_fp16")]; + tensor var_38934_equation_0 = const()[name = tensor("op_38934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38934_cast_fp16 = einsum(equation = var_38934_equation_0, values = (var_38608_cast_fp16, var_38383_cast_fp16))[name = tensor("op_38934_cast_fp16")]; + tensor var_38935_to_fp16 = const()[name = tensor("op_38935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3791_cast_fp16 = mul(x = var_38934_cast_fp16, y = var_38935_to_fp16)[name = tensor("aw_chunk_3791_cast_fp16")]; + tensor var_38938_equation_0 = const()[name = tensor("op_38938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38938_cast_fp16 = einsum(equation = var_38938_equation_0, values = (var_38612_cast_fp16, var_38390_cast_fp16))[name = tensor("op_38938_cast_fp16")]; + tensor var_38939_to_fp16 = const()[name = tensor("op_38939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3793_cast_fp16 = mul(x = var_38938_cast_fp16, y = var_38939_to_fp16)[name = tensor("aw_chunk_3793_cast_fp16")]; + tensor var_38942_equation_0 = const()[name = tensor("op_38942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38942_cast_fp16 = einsum(equation = var_38942_equation_0, values = (var_38612_cast_fp16, var_38397_cast_fp16))[name = tensor("op_38942_cast_fp16")]; + tensor var_38943_to_fp16 = const()[name = tensor("op_38943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3795_cast_fp16 = mul(x = var_38942_cast_fp16, y = var_38943_to_fp16)[name = tensor("aw_chunk_3795_cast_fp16")]; + tensor var_38946_equation_0 = const()[name = tensor("op_38946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38946_cast_fp16 = einsum(equation = var_38946_equation_0, values = (var_38612_cast_fp16, var_38404_cast_fp16))[name = tensor("op_38946_cast_fp16")]; + tensor var_38947_to_fp16 = const()[name = tensor("op_38947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3797_cast_fp16 = mul(x = var_38946_cast_fp16, y = var_38947_to_fp16)[name = tensor("aw_chunk_3797_cast_fp16")]; + tensor var_38950_equation_0 = const()[name = tensor("op_38950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38950_cast_fp16 = einsum(equation = var_38950_equation_0, values = (var_38612_cast_fp16, var_38411_cast_fp16))[name = tensor("op_38950_cast_fp16")]; + tensor var_38951_to_fp16 = const()[name = tensor("op_38951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3799_cast_fp16 = mul(x = var_38950_cast_fp16, y = var_38951_to_fp16)[name = tensor("aw_chunk_3799_cast_fp16")]; + tensor var_38954_equation_0 = const()[name = tensor("op_38954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38954_cast_fp16 = einsum(equation = var_38954_equation_0, values = (var_38616_cast_fp16, var_38418_cast_fp16))[name = tensor("op_38954_cast_fp16")]; + tensor var_38955_to_fp16 = const()[name = tensor("op_38955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3801_cast_fp16 = mul(x = var_38954_cast_fp16, y = var_38955_to_fp16)[name = tensor("aw_chunk_3801_cast_fp16")]; + tensor var_38958_equation_0 = const()[name = tensor("op_38958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38958_cast_fp16 = einsum(equation = var_38958_equation_0, values = (var_38616_cast_fp16, var_38425_cast_fp16))[name = tensor("op_38958_cast_fp16")]; + tensor var_38959_to_fp16 = const()[name = tensor("op_38959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3803_cast_fp16 = mul(x = var_38958_cast_fp16, y = var_38959_to_fp16)[name = tensor("aw_chunk_3803_cast_fp16")]; + tensor var_38962_equation_0 = const()[name = tensor("op_38962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38962_cast_fp16 = einsum(equation = var_38962_equation_0, values = (var_38616_cast_fp16, var_38432_cast_fp16))[name = tensor("op_38962_cast_fp16")]; + tensor var_38963_to_fp16 = const()[name = tensor("op_38963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3805_cast_fp16 = mul(x = var_38962_cast_fp16, y = var_38963_to_fp16)[name = tensor("aw_chunk_3805_cast_fp16")]; + tensor var_38966_equation_0 = const()[name = tensor("op_38966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38966_cast_fp16 = einsum(equation = var_38966_equation_0, values = (var_38616_cast_fp16, var_38439_cast_fp16))[name = tensor("op_38966_cast_fp16")]; + tensor var_38967_to_fp16 = const()[name = tensor("op_38967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3807_cast_fp16 = mul(x = var_38966_cast_fp16, y = var_38967_to_fp16)[name = tensor("aw_chunk_3807_cast_fp16")]; + tensor var_38970_equation_0 = const()[name = tensor("op_38970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38970_cast_fp16 = einsum(equation = var_38970_equation_0, values = (var_38620_cast_fp16, var_38446_cast_fp16))[name = tensor("op_38970_cast_fp16")]; + tensor var_38971_to_fp16 = const()[name = tensor("op_38971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3809_cast_fp16 = mul(x = var_38970_cast_fp16, y = var_38971_to_fp16)[name = tensor("aw_chunk_3809_cast_fp16")]; + tensor var_38974_equation_0 = const()[name = tensor("op_38974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38974_cast_fp16 = einsum(equation = var_38974_equation_0, values = (var_38620_cast_fp16, var_38453_cast_fp16))[name = tensor("op_38974_cast_fp16")]; + tensor var_38975_to_fp16 = const()[name = tensor("op_38975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3811_cast_fp16 = mul(x = var_38974_cast_fp16, y = var_38975_to_fp16)[name = tensor("aw_chunk_3811_cast_fp16")]; + tensor var_38978_equation_0 = const()[name = tensor("op_38978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38978_cast_fp16 = einsum(equation = var_38978_equation_0, values = (var_38620_cast_fp16, var_38460_cast_fp16))[name = tensor("op_38978_cast_fp16")]; + tensor var_38979_to_fp16 = const()[name = tensor("op_38979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3813_cast_fp16 = mul(x = var_38978_cast_fp16, y = var_38979_to_fp16)[name = tensor("aw_chunk_3813_cast_fp16")]; + tensor var_38982_equation_0 = const()[name = tensor("op_38982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38982_cast_fp16 = einsum(equation = var_38982_equation_0, values = (var_38620_cast_fp16, var_38467_cast_fp16))[name = tensor("op_38982_cast_fp16")]; + tensor var_38983_to_fp16 = const()[name = tensor("op_38983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3815_cast_fp16 = mul(x = var_38982_cast_fp16, y = var_38983_to_fp16)[name = tensor("aw_chunk_3815_cast_fp16")]; + tensor var_38986_equation_0 = const()[name = tensor("op_38986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38986_cast_fp16 = einsum(equation = var_38986_equation_0, values = (var_38624_cast_fp16, var_38474_cast_fp16))[name = tensor("op_38986_cast_fp16")]; + tensor var_38987_to_fp16 = const()[name = tensor("op_38987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3817_cast_fp16 = mul(x = var_38986_cast_fp16, y = var_38987_to_fp16)[name = tensor("aw_chunk_3817_cast_fp16")]; + tensor var_38990_equation_0 = const()[name = tensor("op_38990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38990_cast_fp16 = einsum(equation = var_38990_equation_0, values = (var_38624_cast_fp16, var_38481_cast_fp16))[name = tensor("op_38990_cast_fp16")]; + tensor var_38991_to_fp16 = const()[name = tensor("op_38991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3819_cast_fp16 = mul(x = var_38990_cast_fp16, y = var_38991_to_fp16)[name = tensor("aw_chunk_3819_cast_fp16")]; + tensor var_38994_equation_0 = const()[name = tensor("op_38994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38994_cast_fp16 = einsum(equation = var_38994_equation_0, values = (var_38624_cast_fp16, var_38488_cast_fp16))[name = tensor("op_38994_cast_fp16")]; + tensor var_38995_to_fp16 = const()[name = tensor("op_38995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3821_cast_fp16 = mul(x = var_38994_cast_fp16, y = var_38995_to_fp16)[name = tensor("aw_chunk_3821_cast_fp16")]; + tensor var_38998_equation_0 = const()[name = tensor("op_38998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_38998_cast_fp16 = einsum(equation = var_38998_equation_0, values = (var_38624_cast_fp16, var_38495_cast_fp16))[name = tensor("op_38998_cast_fp16")]; + tensor var_38999_to_fp16 = const()[name = tensor("op_38999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3823_cast_fp16 = mul(x = var_38998_cast_fp16, y = var_38999_to_fp16)[name = tensor("aw_chunk_3823_cast_fp16")]; + tensor var_39002_equation_0 = const()[name = tensor("op_39002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39002_cast_fp16 = einsum(equation = var_39002_equation_0, values = (var_38628_cast_fp16, var_38502_cast_fp16))[name = tensor("op_39002_cast_fp16")]; + tensor var_39003_to_fp16 = const()[name = tensor("op_39003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3825_cast_fp16 = mul(x = var_39002_cast_fp16, y = var_39003_to_fp16)[name = tensor("aw_chunk_3825_cast_fp16")]; + tensor var_39006_equation_0 = const()[name = tensor("op_39006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39006_cast_fp16 = einsum(equation = var_39006_equation_0, values = (var_38628_cast_fp16, var_38509_cast_fp16))[name = tensor("op_39006_cast_fp16")]; + tensor var_39007_to_fp16 = const()[name = tensor("op_39007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3827_cast_fp16 = mul(x = var_39006_cast_fp16, y = var_39007_to_fp16)[name = tensor("aw_chunk_3827_cast_fp16")]; + tensor var_39010_equation_0 = const()[name = tensor("op_39010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39010_cast_fp16 = einsum(equation = var_39010_equation_0, values = (var_38628_cast_fp16, var_38516_cast_fp16))[name = tensor("op_39010_cast_fp16")]; + tensor var_39011_to_fp16 = const()[name = tensor("op_39011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3829_cast_fp16 = mul(x = var_39010_cast_fp16, y = var_39011_to_fp16)[name = tensor("aw_chunk_3829_cast_fp16")]; + tensor var_39014_equation_0 = const()[name = tensor("op_39014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39014_cast_fp16 = einsum(equation = var_39014_equation_0, values = (var_38628_cast_fp16, var_38523_cast_fp16))[name = tensor("op_39014_cast_fp16")]; + tensor var_39015_to_fp16 = const()[name = tensor("op_39015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3831_cast_fp16 = mul(x = var_39014_cast_fp16, y = var_39015_to_fp16)[name = tensor("aw_chunk_3831_cast_fp16")]; + tensor var_39018_equation_0 = const()[name = tensor("op_39018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39018_cast_fp16 = einsum(equation = var_39018_equation_0, values = (var_38632_cast_fp16, var_38530_cast_fp16))[name = tensor("op_39018_cast_fp16")]; + tensor var_39019_to_fp16 = const()[name = tensor("op_39019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3833_cast_fp16 = mul(x = var_39018_cast_fp16, y = var_39019_to_fp16)[name = tensor("aw_chunk_3833_cast_fp16")]; + tensor var_39022_equation_0 = const()[name = tensor("op_39022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39022_cast_fp16 = einsum(equation = var_39022_equation_0, values = (var_38632_cast_fp16, var_38537_cast_fp16))[name = tensor("op_39022_cast_fp16")]; + tensor var_39023_to_fp16 = const()[name = tensor("op_39023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3835_cast_fp16 = mul(x = var_39022_cast_fp16, y = var_39023_to_fp16)[name = tensor("aw_chunk_3835_cast_fp16")]; + tensor var_39026_equation_0 = const()[name = tensor("op_39026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39026_cast_fp16 = einsum(equation = var_39026_equation_0, values = (var_38632_cast_fp16, var_38544_cast_fp16))[name = tensor("op_39026_cast_fp16")]; + tensor var_39027_to_fp16 = const()[name = tensor("op_39027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3837_cast_fp16 = mul(x = var_39026_cast_fp16, y = var_39027_to_fp16)[name = tensor("aw_chunk_3837_cast_fp16")]; + tensor var_39030_equation_0 = const()[name = tensor("op_39030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_39030_cast_fp16 = einsum(equation = var_39030_equation_0, values = (var_38632_cast_fp16, var_38551_cast_fp16))[name = tensor("op_39030_cast_fp16")]; + tensor var_39031_to_fp16 = const()[name = tensor("op_39031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3839_cast_fp16 = mul(x = var_39030_cast_fp16, y = var_39031_to_fp16)[name = tensor("aw_chunk_3839_cast_fp16")]; + tensor var_39033_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3681_cast_fp16)[name = tensor("op_39033_cast_fp16")]; + tensor var_39034_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3683_cast_fp16)[name = tensor("op_39034_cast_fp16")]; + tensor var_39035_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3685_cast_fp16)[name = tensor("op_39035_cast_fp16")]; + tensor var_39036_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3687_cast_fp16)[name = tensor("op_39036_cast_fp16")]; + tensor var_39037_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3689_cast_fp16)[name = tensor("op_39037_cast_fp16")]; + tensor var_39038_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3691_cast_fp16)[name = tensor("op_39038_cast_fp16")]; + tensor var_39039_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3693_cast_fp16)[name = tensor("op_39039_cast_fp16")]; + tensor var_39040_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3695_cast_fp16)[name = tensor("op_39040_cast_fp16")]; + tensor var_39041_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3697_cast_fp16)[name = tensor("op_39041_cast_fp16")]; + tensor var_39042_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3699_cast_fp16)[name = tensor("op_39042_cast_fp16")]; + tensor var_39043_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3701_cast_fp16)[name = tensor("op_39043_cast_fp16")]; + tensor var_39044_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3703_cast_fp16)[name = tensor("op_39044_cast_fp16")]; + tensor var_39045_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3705_cast_fp16)[name = tensor("op_39045_cast_fp16")]; + tensor var_39046_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3707_cast_fp16)[name = tensor("op_39046_cast_fp16")]; + tensor var_39047_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3709_cast_fp16)[name = tensor("op_39047_cast_fp16")]; + tensor var_39048_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3711_cast_fp16)[name = tensor("op_39048_cast_fp16")]; + tensor var_39049_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3713_cast_fp16)[name = tensor("op_39049_cast_fp16")]; + tensor var_39050_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3715_cast_fp16)[name = tensor("op_39050_cast_fp16")]; + tensor var_39051_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3717_cast_fp16)[name = tensor("op_39051_cast_fp16")]; + tensor var_39052_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3719_cast_fp16)[name = tensor("op_39052_cast_fp16")]; + tensor var_39053_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3721_cast_fp16)[name = tensor("op_39053_cast_fp16")]; + tensor var_39054_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3723_cast_fp16)[name = tensor("op_39054_cast_fp16")]; + tensor var_39055_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3725_cast_fp16)[name = tensor("op_39055_cast_fp16")]; + tensor var_39056_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3727_cast_fp16)[name = tensor("op_39056_cast_fp16")]; + tensor var_39057_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3729_cast_fp16)[name = tensor("op_39057_cast_fp16")]; + tensor var_39058_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3731_cast_fp16)[name = tensor("op_39058_cast_fp16")]; + tensor var_39059_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3733_cast_fp16)[name = tensor("op_39059_cast_fp16")]; + tensor var_39060_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3735_cast_fp16)[name = tensor("op_39060_cast_fp16")]; + tensor var_39061_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3737_cast_fp16)[name = tensor("op_39061_cast_fp16")]; + tensor var_39062_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3739_cast_fp16)[name = tensor("op_39062_cast_fp16")]; + tensor var_39063_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3741_cast_fp16)[name = tensor("op_39063_cast_fp16")]; + tensor var_39064_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3743_cast_fp16)[name = tensor("op_39064_cast_fp16")]; + tensor var_39065_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3745_cast_fp16)[name = tensor("op_39065_cast_fp16")]; + tensor var_39066_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3747_cast_fp16)[name = tensor("op_39066_cast_fp16")]; + tensor var_39067_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3749_cast_fp16)[name = tensor("op_39067_cast_fp16")]; + tensor var_39068_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3751_cast_fp16)[name = tensor("op_39068_cast_fp16")]; + tensor var_39069_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3753_cast_fp16)[name = tensor("op_39069_cast_fp16")]; + tensor var_39070_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3755_cast_fp16)[name = tensor("op_39070_cast_fp16")]; + tensor var_39071_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3757_cast_fp16)[name = tensor("op_39071_cast_fp16")]; + tensor var_39072_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3759_cast_fp16)[name = tensor("op_39072_cast_fp16")]; + tensor var_39073_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3761_cast_fp16)[name = tensor("op_39073_cast_fp16")]; + tensor var_39074_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3763_cast_fp16)[name = tensor("op_39074_cast_fp16")]; + tensor var_39075_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3765_cast_fp16)[name = tensor("op_39075_cast_fp16")]; + tensor var_39076_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3767_cast_fp16)[name = tensor("op_39076_cast_fp16")]; + tensor var_39077_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3769_cast_fp16)[name = tensor("op_39077_cast_fp16")]; + tensor var_39078_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3771_cast_fp16)[name = tensor("op_39078_cast_fp16")]; + tensor var_39079_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3773_cast_fp16)[name = tensor("op_39079_cast_fp16")]; + tensor var_39080_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3775_cast_fp16)[name = tensor("op_39080_cast_fp16")]; + tensor var_39081_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3777_cast_fp16)[name = tensor("op_39081_cast_fp16")]; + tensor var_39082_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3779_cast_fp16)[name = tensor("op_39082_cast_fp16")]; + tensor var_39083_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3781_cast_fp16)[name = tensor("op_39083_cast_fp16")]; + tensor var_39084_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3783_cast_fp16)[name = tensor("op_39084_cast_fp16")]; + tensor var_39085_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3785_cast_fp16)[name = tensor("op_39085_cast_fp16")]; + tensor var_39086_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3787_cast_fp16)[name = tensor("op_39086_cast_fp16")]; + tensor var_39087_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3789_cast_fp16)[name = tensor("op_39087_cast_fp16")]; + tensor var_39088_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3791_cast_fp16)[name = tensor("op_39088_cast_fp16")]; + tensor var_39089_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3793_cast_fp16)[name = tensor("op_39089_cast_fp16")]; + tensor var_39090_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3795_cast_fp16)[name = tensor("op_39090_cast_fp16")]; + tensor var_39091_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3797_cast_fp16)[name = tensor("op_39091_cast_fp16")]; + tensor var_39092_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3799_cast_fp16)[name = tensor("op_39092_cast_fp16")]; + tensor var_39093_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3801_cast_fp16)[name = tensor("op_39093_cast_fp16")]; + tensor var_39094_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3803_cast_fp16)[name = tensor("op_39094_cast_fp16")]; + tensor var_39095_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3805_cast_fp16)[name = tensor("op_39095_cast_fp16")]; + tensor var_39096_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3807_cast_fp16)[name = tensor("op_39096_cast_fp16")]; + tensor var_39097_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3809_cast_fp16)[name = tensor("op_39097_cast_fp16")]; + tensor var_39098_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3811_cast_fp16)[name = tensor("op_39098_cast_fp16")]; + tensor var_39099_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3813_cast_fp16)[name = tensor("op_39099_cast_fp16")]; + tensor var_39100_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3815_cast_fp16)[name = tensor("op_39100_cast_fp16")]; + tensor var_39101_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3817_cast_fp16)[name = tensor("op_39101_cast_fp16")]; + tensor var_39102_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3819_cast_fp16)[name = tensor("op_39102_cast_fp16")]; + tensor var_39103_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3821_cast_fp16)[name = tensor("op_39103_cast_fp16")]; + tensor var_39104_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3823_cast_fp16)[name = tensor("op_39104_cast_fp16")]; + tensor var_39105_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3825_cast_fp16)[name = tensor("op_39105_cast_fp16")]; + tensor var_39106_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3827_cast_fp16)[name = tensor("op_39106_cast_fp16")]; + tensor var_39107_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3829_cast_fp16)[name = tensor("op_39107_cast_fp16")]; + tensor var_39108_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3831_cast_fp16)[name = tensor("op_39108_cast_fp16")]; + tensor var_39109_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3833_cast_fp16)[name = tensor("op_39109_cast_fp16")]; + tensor var_39110_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3835_cast_fp16)[name = tensor("op_39110_cast_fp16")]; + tensor var_39111_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3837_cast_fp16)[name = tensor("op_39111_cast_fp16")]; + tensor var_39112_cast_fp16 = softmax(axis = var_37804, x = aw_chunk_3839_cast_fp16)[name = tensor("op_39112_cast_fp16")]; + tensor var_39114_equation_0 = const()[name = tensor("op_39114_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39114_cast_fp16 = einsum(equation = var_39114_equation_0, values = (var_38634_cast_fp16, var_39033_cast_fp16))[name = tensor("op_39114_cast_fp16")]; + tensor var_39116_equation_0 = const()[name = tensor("op_39116_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39116_cast_fp16 = einsum(equation = var_39116_equation_0, values = (var_38634_cast_fp16, var_39034_cast_fp16))[name = tensor("op_39116_cast_fp16")]; + tensor var_39118_equation_0 = const()[name = tensor("op_39118_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39118_cast_fp16 = einsum(equation = var_39118_equation_0, values = (var_38634_cast_fp16, var_39035_cast_fp16))[name = tensor("op_39118_cast_fp16")]; + tensor var_39120_equation_0 = const()[name = tensor("op_39120_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39120_cast_fp16 = einsum(equation = var_39120_equation_0, values = (var_38634_cast_fp16, var_39036_cast_fp16))[name = tensor("op_39120_cast_fp16")]; + tensor var_39122_equation_0 = const()[name = tensor("op_39122_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39122_cast_fp16 = einsum(equation = var_39122_equation_0, values = (var_38638_cast_fp16, var_39037_cast_fp16))[name = tensor("op_39122_cast_fp16")]; + tensor var_39124_equation_0 = const()[name = tensor("op_39124_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39124_cast_fp16 = einsum(equation = var_39124_equation_0, values = (var_38638_cast_fp16, var_39038_cast_fp16))[name = tensor("op_39124_cast_fp16")]; + tensor var_39126_equation_0 = const()[name = tensor("op_39126_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39126_cast_fp16 = einsum(equation = var_39126_equation_0, values = (var_38638_cast_fp16, var_39039_cast_fp16))[name = tensor("op_39126_cast_fp16")]; + tensor var_39128_equation_0 = const()[name = tensor("op_39128_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39128_cast_fp16 = einsum(equation = var_39128_equation_0, values = (var_38638_cast_fp16, var_39040_cast_fp16))[name = tensor("op_39128_cast_fp16")]; + tensor var_39130_equation_0 = const()[name = tensor("op_39130_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39130_cast_fp16 = einsum(equation = var_39130_equation_0, values = (var_38642_cast_fp16, var_39041_cast_fp16))[name = tensor("op_39130_cast_fp16")]; + tensor var_39132_equation_0 = const()[name = tensor("op_39132_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39132_cast_fp16 = einsum(equation = var_39132_equation_0, values = (var_38642_cast_fp16, var_39042_cast_fp16))[name = tensor("op_39132_cast_fp16")]; + tensor var_39134_equation_0 = const()[name = tensor("op_39134_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39134_cast_fp16 = einsum(equation = var_39134_equation_0, values = (var_38642_cast_fp16, var_39043_cast_fp16))[name = tensor("op_39134_cast_fp16")]; + tensor var_39136_equation_0 = const()[name = tensor("op_39136_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39136_cast_fp16 = einsum(equation = var_39136_equation_0, values = (var_38642_cast_fp16, var_39044_cast_fp16))[name = tensor("op_39136_cast_fp16")]; + tensor var_39138_equation_0 = const()[name = tensor("op_39138_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39138_cast_fp16 = einsum(equation = var_39138_equation_0, values = (var_38646_cast_fp16, var_39045_cast_fp16))[name = tensor("op_39138_cast_fp16")]; + tensor var_39140_equation_0 = const()[name = tensor("op_39140_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39140_cast_fp16 = einsum(equation = var_39140_equation_0, values = (var_38646_cast_fp16, var_39046_cast_fp16))[name = tensor("op_39140_cast_fp16")]; + tensor var_39142_equation_0 = const()[name = tensor("op_39142_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39142_cast_fp16 = einsum(equation = var_39142_equation_0, values = (var_38646_cast_fp16, var_39047_cast_fp16))[name = tensor("op_39142_cast_fp16")]; + tensor var_39144_equation_0 = const()[name = tensor("op_39144_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39144_cast_fp16 = einsum(equation = var_39144_equation_0, values = (var_38646_cast_fp16, var_39048_cast_fp16))[name = tensor("op_39144_cast_fp16")]; + tensor var_39146_equation_0 = const()[name = tensor("op_39146_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39146_cast_fp16 = einsum(equation = var_39146_equation_0, values = (var_38650_cast_fp16, var_39049_cast_fp16))[name = tensor("op_39146_cast_fp16")]; + tensor var_39148_equation_0 = const()[name = tensor("op_39148_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39148_cast_fp16 = einsum(equation = var_39148_equation_0, values = (var_38650_cast_fp16, var_39050_cast_fp16))[name = tensor("op_39148_cast_fp16")]; + tensor var_39150_equation_0 = const()[name = tensor("op_39150_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39150_cast_fp16 = einsum(equation = var_39150_equation_0, values = (var_38650_cast_fp16, var_39051_cast_fp16))[name = tensor("op_39150_cast_fp16")]; + tensor var_39152_equation_0 = const()[name = tensor("op_39152_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39152_cast_fp16 = einsum(equation = var_39152_equation_0, values = (var_38650_cast_fp16, var_39052_cast_fp16))[name = tensor("op_39152_cast_fp16")]; + tensor var_39154_equation_0 = const()[name = tensor("op_39154_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39154_cast_fp16 = einsum(equation = var_39154_equation_0, values = (var_38654_cast_fp16, var_39053_cast_fp16))[name = tensor("op_39154_cast_fp16")]; + tensor var_39156_equation_0 = const()[name = tensor("op_39156_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39156_cast_fp16 = einsum(equation = var_39156_equation_0, values = (var_38654_cast_fp16, var_39054_cast_fp16))[name = tensor("op_39156_cast_fp16")]; + tensor var_39158_equation_0 = const()[name = tensor("op_39158_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39158_cast_fp16 = einsum(equation = var_39158_equation_0, values = (var_38654_cast_fp16, var_39055_cast_fp16))[name = tensor("op_39158_cast_fp16")]; + tensor var_39160_equation_0 = const()[name = tensor("op_39160_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39160_cast_fp16 = einsum(equation = var_39160_equation_0, values = (var_38654_cast_fp16, var_39056_cast_fp16))[name = tensor("op_39160_cast_fp16")]; + tensor var_39162_equation_0 = const()[name = tensor("op_39162_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39162_cast_fp16 = einsum(equation = var_39162_equation_0, values = (var_38658_cast_fp16, var_39057_cast_fp16))[name = tensor("op_39162_cast_fp16")]; + tensor var_39164_equation_0 = const()[name = tensor("op_39164_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39164_cast_fp16 = einsum(equation = var_39164_equation_0, values = (var_38658_cast_fp16, var_39058_cast_fp16))[name = tensor("op_39164_cast_fp16")]; + tensor var_39166_equation_0 = const()[name = tensor("op_39166_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39166_cast_fp16 = einsum(equation = var_39166_equation_0, values = (var_38658_cast_fp16, var_39059_cast_fp16))[name = tensor("op_39166_cast_fp16")]; + tensor var_39168_equation_0 = const()[name = tensor("op_39168_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39168_cast_fp16 = einsum(equation = var_39168_equation_0, values = (var_38658_cast_fp16, var_39060_cast_fp16))[name = tensor("op_39168_cast_fp16")]; + tensor var_39170_equation_0 = const()[name = tensor("op_39170_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39170_cast_fp16 = einsum(equation = var_39170_equation_0, values = (var_38662_cast_fp16, var_39061_cast_fp16))[name = tensor("op_39170_cast_fp16")]; + tensor var_39172_equation_0 = const()[name = tensor("op_39172_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39172_cast_fp16 = einsum(equation = var_39172_equation_0, values = (var_38662_cast_fp16, var_39062_cast_fp16))[name = tensor("op_39172_cast_fp16")]; + tensor var_39174_equation_0 = const()[name = tensor("op_39174_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39174_cast_fp16 = einsum(equation = var_39174_equation_0, values = (var_38662_cast_fp16, var_39063_cast_fp16))[name = tensor("op_39174_cast_fp16")]; + tensor var_39176_equation_0 = const()[name = tensor("op_39176_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39176_cast_fp16 = einsum(equation = var_39176_equation_0, values = (var_38662_cast_fp16, var_39064_cast_fp16))[name = tensor("op_39176_cast_fp16")]; + tensor var_39178_equation_0 = const()[name = tensor("op_39178_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39178_cast_fp16 = einsum(equation = var_39178_equation_0, values = (var_38666_cast_fp16, var_39065_cast_fp16))[name = tensor("op_39178_cast_fp16")]; + tensor var_39180_equation_0 = const()[name = tensor("op_39180_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39180_cast_fp16 = einsum(equation = var_39180_equation_0, values = (var_38666_cast_fp16, var_39066_cast_fp16))[name = tensor("op_39180_cast_fp16")]; + tensor var_39182_equation_0 = const()[name = tensor("op_39182_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39182_cast_fp16 = einsum(equation = var_39182_equation_0, values = (var_38666_cast_fp16, var_39067_cast_fp16))[name = tensor("op_39182_cast_fp16")]; + tensor var_39184_equation_0 = const()[name = tensor("op_39184_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39184_cast_fp16 = einsum(equation = var_39184_equation_0, values = (var_38666_cast_fp16, var_39068_cast_fp16))[name = tensor("op_39184_cast_fp16")]; + tensor var_39186_equation_0 = const()[name = tensor("op_39186_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39186_cast_fp16 = einsum(equation = var_39186_equation_0, values = (var_38670_cast_fp16, var_39069_cast_fp16))[name = tensor("op_39186_cast_fp16")]; + tensor var_39188_equation_0 = const()[name = tensor("op_39188_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39188_cast_fp16 = einsum(equation = var_39188_equation_0, values = (var_38670_cast_fp16, var_39070_cast_fp16))[name = tensor("op_39188_cast_fp16")]; + tensor var_39190_equation_0 = const()[name = tensor("op_39190_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39190_cast_fp16 = einsum(equation = var_39190_equation_0, values = (var_38670_cast_fp16, var_39071_cast_fp16))[name = tensor("op_39190_cast_fp16")]; + tensor var_39192_equation_0 = const()[name = tensor("op_39192_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39192_cast_fp16 = einsum(equation = var_39192_equation_0, values = (var_38670_cast_fp16, var_39072_cast_fp16))[name = tensor("op_39192_cast_fp16")]; + tensor var_39194_equation_0 = const()[name = tensor("op_39194_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39194_cast_fp16 = einsum(equation = var_39194_equation_0, values = (var_38674_cast_fp16, var_39073_cast_fp16))[name = tensor("op_39194_cast_fp16")]; + tensor var_39196_equation_0 = const()[name = tensor("op_39196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39196_cast_fp16 = einsum(equation = var_39196_equation_0, values = (var_38674_cast_fp16, var_39074_cast_fp16))[name = tensor("op_39196_cast_fp16")]; + tensor var_39198_equation_0 = const()[name = tensor("op_39198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39198_cast_fp16 = einsum(equation = var_39198_equation_0, values = (var_38674_cast_fp16, var_39075_cast_fp16))[name = tensor("op_39198_cast_fp16")]; + tensor var_39200_equation_0 = const()[name = tensor("op_39200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39200_cast_fp16 = einsum(equation = var_39200_equation_0, values = (var_38674_cast_fp16, var_39076_cast_fp16))[name = tensor("op_39200_cast_fp16")]; + tensor var_39202_equation_0 = const()[name = tensor("op_39202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39202_cast_fp16 = einsum(equation = var_39202_equation_0, values = (var_38678_cast_fp16, var_39077_cast_fp16))[name = tensor("op_39202_cast_fp16")]; + tensor var_39204_equation_0 = const()[name = tensor("op_39204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39204_cast_fp16 = einsum(equation = var_39204_equation_0, values = (var_38678_cast_fp16, var_39078_cast_fp16))[name = tensor("op_39204_cast_fp16")]; + tensor var_39206_equation_0 = const()[name = tensor("op_39206_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39206_cast_fp16 = einsum(equation = var_39206_equation_0, values = (var_38678_cast_fp16, var_39079_cast_fp16))[name = tensor("op_39206_cast_fp16")]; + tensor var_39208_equation_0 = const()[name = tensor("op_39208_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39208_cast_fp16 = einsum(equation = var_39208_equation_0, values = (var_38678_cast_fp16, var_39080_cast_fp16))[name = tensor("op_39208_cast_fp16")]; + tensor var_39210_equation_0 = const()[name = tensor("op_39210_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39210_cast_fp16 = einsum(equation = var_39210_equation_0, values = (var_38682_cast_fp16, var_39081_cast_fp16))[name = tensor("op_39210_cast_fp16")]; + tensor var_39212_equation_0 = const()[name = tensor("op_39212_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39212_cast_fp16 = einsum(equation = var_39212_equation_0, values = (var_38682_cast_fp16, var_39082_cast_fp16))[name = tensor("op_39212_cast_fp16")]; + tensor var_39214_equation_0 = const()[name = tensor("op_39214_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39214_cast_fp16 = einsum(equation = var_39214_equation_0, values = (var_38682_cast_fp16, var_39083_cast_fp16))[name = tensor("op_39214_cast_fp16")]; + tensor var_39216_equation_0 = const()[name = tensor("op_39216_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39216_cast_fp16 = einsum(equation = var_39216_equation_0, values = (var_38682_cast_fp16, var_39084_cast_fp16))[name = tensor("op_39216_cast_fp16")]; + tensor var_39218_equation_0 = const()[name = tensor("op_39218_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39218_cast_fp16 = einsum(equation = var_39218_equation_0, values = (var_38686_cast_fp16, var_39085_cast_fp16))[name = tensor("op_39218_cast_fp16")]; + tensor var_39220_equation_0 = const()[name = tensor("op_39220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39220_cast_fp16 = einsum(equation = var_39220_equation_0, values = (var_38686_cast_fp16, var_39086_cast_fp16))[name = tensor("op_39220_cast_fp16")]; + tensor var_39222_equation_0 = const()[name = tensor("op_39222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39222_cast_fp16 = einsum(equation = var_39222_equation_0, values = (var_38686_cast_fp16, var_39087_cast_fp16))[name = tensor("op_39222_cast_fp16")]; + tensor var_39224_equation_0 = const()[name = tensor("op_39224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39224_cast_fp16 = einsum(equation = var_39224_equation_0, values = (var_38686_cast_fp16, var_39088_cast_fp16))[name = tensor("op_39224_cast_fp16")]; + tensor var_39226_equation_0 = const()[name = tensor("op_39226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39226_cast_fp16 = einsum(equation = var_39226_equation_0, values = (var_38690_cast_fp16, var_39089_cast_fp16))[name = tensor("op_39226_cast_fp16")]; + tensor var_39228_equation_0 = const()[name = tensor("op_39228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39228_cast_fp16 = einsum(equation = var_39228_equation_0, values = (var_38690_cast_fp16, var_39090_cast_fp16))[name = tensor("op_39228_cast_fp16")]; + tensor var_39230_equation_0 = const()[name = tensor("op_39230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39230_cast_fp16 = einsum(equation = var_39230_equation_0, values = (var_38690_cast_fp16, var_39091_cast_fp16))[name = tensor("op_39230_cast_fp16")]; + tensor var_39232_equation_0 = const()[name = tensor("op_39232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39232_cast_fp16 = einsum(equation = var_39232_equation_0, values = (var_38690_cast_fp16, var_39092_cast_fp16))[name = tensor("op_39232_cast_fp16")]; + tensor var_39234_equation_0 = const()[name = tensor("op_39234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39234_cast_fp16 = einsum(equation = var_39234_equation_0, values = (var_38694_cast_fp16, var_39093_cast_fp16))[name = tensor("op_39234_cast_fp16")]; + tensor var_39236_equation_0 = const()[name = tensor("op_39236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39236_cast_fp16 = einsum(equation = var_39236_equation_0, values = (var_38694_cast_fp16, var_39094_cast_fp16))[name = tensor("op_39236_cast_fp16")]; + tensor var_39238_equation_0 = const()[name = tensor("op_39238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39238_cast_fp16 = einsum(equation = var_39238_equation_0, values = (var_38694_cast_fp16, var_39095_cast_fp16))[name = tensor("op_39238_cast_fp16")]; + tensor var_39240_equation_0 = const()[name = tensor("op_39240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39240_cast_fp16 = einsum(equation = var_39240_equation_0, values = (var_38694_cast_fp16, var_39096_cast_fp16))[name = tensor("op_39240_cast_fp16")]; + tensor var_39242_equation_0 = const()[name = tensor("op_39242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39242_cast_fp16 = einsum(equation = var_39242_equation_0, values = (var_38698_cast_fp16, var_39097_cast_fp16))[name = tensor("op_39242_cast_fp16")]; + tensor var_39244_equation_0 = const()[name = tensor("op_39244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39244_cast_fp16 = einsum(equation = var_39244_equation_0, values = (var_38698_cast_fp16, var_39098_cast_fp16))[name = tensor("op_39244_cast_fp16")]; + tensor var_39246_equation_0 = const()[name = tensor("op_39246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39246_cast_fp16 = einsum(equation = var_39246_equation_0, values = (var_38698_cast_fp16, var_39099_cast_fp16))[name = tensor("op_39246_cast_fp16")]; + tensor var_39248_equation_0 = const()[name = tensor("op_39248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39248_cast_fp16 = einsum(equation = var_39248_equation_0, values = (var_38698_cast_fp16, var_39100_cast_fp16))[name = tensor("op_39248_cast_fp16")]; + tensor var_39250_equation_0 = const()[name = tensor("op_39250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39250_cast_fp16 = einsum(equation = var_39250_equation_0, values = (var_38702_cast_fp16, var_39101_cast_fp16))[name = tensor("op_39250_cast_fp16")]; + tensor var_39252_equation_0 = const()[name = tensor("op_39252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39252_cast_fp16 = einsum(equation = var_39252_equation_0, values = (var_38702_cast_fp16, var_39102_cast_fp16))[name = tensor("op_39252_cast_fp16")]; + tensor var_39254_equation_0 = const()[name = tensor("op_39254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39254_cast_fp16 = einsum(equation = var_39254_equation_0, values = (var_38702_cast_fp16, var_39103_cast_fp16))[name = tensor("op_39254_cast_fp16")]; + tensor var_39256_equation_0 = const()[name = tensor("op_39256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39256_cast_fp16 = einsum(equation = var_39256_equation_0, values = (var_38702_cast_fp16, var_39104_cast_fp16))[name = tensor("op_39256_cast_fp16")]; + tensor var_39258_equation_0 = const()[name = tensor("op_39258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39258_cast_fp16 = einsum(equation = var_39258_equation_0, values = (var_38706_cast_fp16, var_39105_cast_fp16))[name = tensor("op_39258_cast_fp16")]; + tensor var_39260_equation_0 = const()[name = tensor("op_39260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39260_cast_fp16 = einsum(equation = var_39260_equation_0, values = (var_38706_cast_fp16, var_39106_cast_fp16))[name = tensor("op_39260_cast_fp16")]; + tensor var_39262_equation_0 = const()[name = tensor("op_39262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39262_cast_fp16 = einsum(equation = var_39262_equation_0, values = (var_38706_cast_fp16, var_39107_cast_fp16))[name = tensor("op_39262_cast_fp16")]; + tensor var_39264_equation_0 = const()[name = tensor("op_39264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39264_cast_fp16 = einsum(equation = var_39264_equation_0, values = (var_38706_cast_fp16, var_39108_cast_fp16))[name = tensor("op_39264_cast_fp16")]; + tensor var_39266_equation_0 = const()[name = tensor("op_39266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39266_cast_fp16 = einsum(equation = var_39266_equation_0, values = (var_38710_cast_fp16, var_39109_cast_fp16))[name = tensor("op_39266_cast_fp16")]; + tensor var_39268_equation_0 = const()[name = tensor("op_39268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39268_cast_fp16 = einsum(equation = var_39268_equation_0, values = (var_38710_cast_fp16, var_39110_cast_fp16))[name = tensor("op_39268_cast_fp16")]; + tensor var_39270_equation_0 = const()[name = tensor("op_39270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39270_cast_fp16 = einsum(equation = var_39270_equation_0, values = (var_38710_cast_fp16, var_39111_cast_fp16))[name = tensor("op_39270_cast_fp16")]; + tensor var_39272_equation_0 = const()[name = tensor("op_39272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_39272_cast_fp16 = einsum(equation = var_39272_equation_0, values = (var_38710_cast_fp16, var_39112_cast_fp16))[name = tensor("op_39272_cast_fp16")]; + tensor var_39274_interleave_0 = const()[name = tensor("op_39274_interleave_0"), val = tensor(false)]; + tensor var_39274_cast_fp16 = concat(axis = var_37779, interleave = var_39274_interleave_0, values = (var_39114_cast_fp16, var_39116_cast_fp16, var_39118_cast_fp16, var_39120_cast_fp16))[name = tensor("op_39274_cast_fp16")]; + tensor var_39276_interleave_0 = const()[name = tensor("op_39276_interleave_0"), val = tensor(false)]; + tensor var_39276_cast_fp16 = concat(axis = var_37779, interleave = var_39276_interleave_0, values = (var_39122_cast_fp16, var_39124_cast_fp16, var_39126_cast_fp16, var_39128_cast_fp16))[name = tensor("op_39276_cast_fp16")]; + tensor var_39278_interleave_0 = const()[name = tensor("op_39278_interleave_0"), val = tensor(false)]; + tensor var_39278_cast_fp16 = concat(axis = var_37779, interleave = var_39278_interleave_0, values = (var_39130_cast_fp16, var_39132_cast_fp16, var_39134_cast_fp16, var_39136_cast_fp16))[name = tensor("op_39278_cast_fp16")]; + tensor var_39280_interleave_0 = const()[name = tensor("op_39280_interleave_0"), val = tensor(false)]; + tensor var_39280_cast_fp16 = concat(axis = var_37779, interleave = var_39280_interleave_0, values = (var_39138_cast_fp16, var_39140_cast_fp16, var_39142_cast_fp16, var_39144_cast_fp16))[name = tensor("op_39280_cast_fp16")]; + tensor var_39282_interleave_0 = const()[name = tensor("op_39282_interleave_0"), val = tensor(false)]; + tensor var_39282_cast_fp16 = concat(axis = var_37779, interleave = var_39282_interleave_0, values = (var_39146_cast_fp16, var_39148_cast_fp16, var_39150_cast_fp16, var_39152_cast_fp16))[name = tensor("op_39282_cast_fp16")]; + tensor var_39284_interleave_0 = const()[name = tensor("op_39284_interleave_0"), val = tensor(false)]; + tensor var_39284_cast_fp16 = concat(axis = var_37779, interleave = var_39284_interleave_0, values = (var_39154_cast_fp16, var_39156_cast_fp16, var_39158_cast_fp16, var_39160_cast_fp16))[name = tensor("op_39284_cast_fp16")]; + tensor var_39286_interleave_0 = const()[name = tensor("op_39286_interleave_0"), val = tensor(false)]; + tensor var_39286_cast_fp16 = concat(axis = var_37779, interleave = var_39286_interleave_0, values = (var_39162_cast_fp16, var_39164_cast_fp16, var_39166_cast_fp16, var_39168_cast_fp16))[name = tensor("op_39286_cast_fp16")]; + tensor var_39288_interleave_0 = const()[name = tensor("op_39288_interleave_0"), val = tensor(false)]; + tensor var_39288_cast_fp16 = concat(axis = var_37779, interleave = var_39288_interleave_0, values = (var_39170_cast_fp16, var_39172_cast_fp16, var_39174_cast_fp16, var_39176_cast_fp16))[name = tensor("op_39288_cast_fp16")]; + tensor var_39290_interleave_0 = const()[name = tensor("op_39290_interleave_0"), val = tensor(false)]; + tensor var_39290_cast_fp16 = concat(axis = var_37779, interleave = var_39290_interleave_0, values = (var_39178_cast_fp16, var_39180_cast_fp16, var_39182_cast_fp16, var_39184_cast_fp16))[name = tensor("op_39290_cast_fp16")]; + tensor var_39292_interleave_0 = const()[name = tensor("op_39292_interleave_0"), val = tensor(false)]; + tensor var_39292_cast_fp16 = concat(axis = var_37779, interleave = var_39292_interleave_0, values = (var_39186_cast_fp16, var_39188_cast_fp16, var_39190_cast_fp16, var_39192_cast_fp16))[name = tensor("op_39292_cast_fp16")]; + tensor var_39294_interleave_0 = const()[name = tensor("op_39294_interleave_0"), val = tensor(false)]; + tensor var_39294_cast_fp16 = concat(axis = var_37779, interleave = var_39294_interleave_0, values = (var_39194_cast_fp16, var_39196_cast_fp16, var_39198_cast_fp16, var_39200_cast_fp16))[name = tensor("op_39294_cast_fp16")]; + tensor var_39296_interleave_0 = const()[name = tensor("op_39296_interleave_0"), val = tensor(false)]; + tensor var_39296_cast_fp16 = concat(axis = var_37779, interleave = var_39296_interleave_0, values = (var_39202_cast_fp16, var_39204_cast_fp16, var_39206_cast_fp16, var_39208_cast_fp16))[name = tensor("op_39296_cast_fp16")]; + tensor var_39298_interleave_0 = const()[name = tensor("op_39298_interleave_0"), val = tensor(false)]; + tensor var_39298_cast_fp16 = concat(axis = var_37779, interleave = var_39298_interleave_0, values = (var_39210_cast_fp16, var_39212_cast_fp16, var_39214_cast_fp16, var_39216_cast_fp16))[name = tensor("op_39298_cast_fp16")]; + tensor var_39300_interleave_0 = const()[name = tensor("op_39300_interleave_0"), val = tensor(false)]; + tensor var_39300_cast_fp16 = concat(axis = var_37779, interleave = var_39300_interleave_0, values = (var_39218_cast_fp16, var_39220_cast_fp16, var_39222_cast_fp16, var_39224_cast_fp16))[name = tensor("op_39300_cast_fp16")]; + tensor var_39302_interleave_0 = const()[name = tensor("op_39302_interleave_0"), val = tensor(false)]; + tensor var_39302_cast_fp16 = concat(axis = var_37779, interleave = var_39302_interleave_0, values = (var_39226_cast_fp16, var_39228_cast_fp16, var_39230_cast_fp16, var_39232_cast_fp16))[name = tensor("op_39302_cast_fp16")]; + tensor var_39304_interleave_0 = const()[name = tensor("op_39304_interleave_0"), val = tensor(false)]; + tensor var_39304_cast_fp16 = concat(axis = var_37779, interleave = var_39304_interleave_0, values = (var_39234_cast_fp16, var_39236_cast_fp16, var_39238_cast_fp16, var_39240_cast_fp16))[name = tensor("op_39304_cast_fp16")]; + tensor var_39306_interleave_0 = const()[name = tensor("op_39306_interleave_0"), val = tensor(false)]; + tensor var_39306_cast_fp16 = concat(axis = var_37779, interleave = var_39306_interleave_0, values = (var_39242_cast_fp16, var_39244_cast_fp16, var_39246_cast_fp16, var_39248_cast_fp16))[name = tensor("op_39306_cast_fp16")]; + tensor var_39308_interleave_0 = const()[name = tensor("op_39308_interleave_0"), val = tensor(false)]; + tensor var_39308_cast_fp16 = concat(axis = var_37779, interleave = var_39308_interleave_0, values = (var_39250_cast_fp16, var_39252_cast_fp16, var_39254_cast_fp16, var_39256_cast_fp16))[name = tensor("op_39308_cast_fp16")]; + tensor var_39310_interleave_0 = const()[name = tensor("op_39310_interleave_0"), val = tensor(false)]; + tensor var_39310_cast_fp16 = concat(axis = var_37779, interleave = var_39310_interleave_0, values = (var_39258_cast_fp16, var_39260_cast_fp16, var_39262_cast_fp16, var_39264_cast_fp16))[name = tensor("op_39310_cast_fp16")]; + tensor var_39312_interleave_0 = const()[name = tensor("op_39312_interleave_0"), val = tensor(false)]; + tensor var_39312_cast_fp16 = concat(axis = var_37779, interleave = var_39312_interleave_0, values = (var_39266_cast_fp16, var_39268_cast_fp16, var_39270_cast_fp16, var_39272_cast_fp16))[name = tensor("op_39312_cast_fp16")]; + tensor input_467_interleave_0 = const()[name = tensor("input_467_interleave_0"), val = tensor(false)]; + tensor input_467_cast_fp16 = concat(axis = var_37804, interleave = input_467_interleave_0, values = (var_39274_cast_fp16, var_39276_cast_fp16, var_39278_cast_fp16, var_39280_cast_fp16, var_39282_cast_fp16, var_39284_cast_fp16, var_39286_cast_fp16, var_39288_cast_fp16, var_39290_cast_fp16, var_39292_cast_fp16, var_39294_cast_fp16, var_39296_cast_fp16, var_39298_cast_fp16, var_39300_cast_fp16, var_39302_cast_fp16, var_39304_cast_fp16, var_39306_cast_fp16, var_39308_cast_fp16, var_39310_cast_fp16, var_39312_cast_fp16))[name = tensor("input_467_cast_fp16")]; + tensor var_39320 = const()[name = tensor("op_39320"), val = tensor([1, 1])]; + tensor var_39322 = const()[name = tensor("op_39322"), val = tensor([1, 1])]; + tensor pretrained_out_283_pad_type_0 = const()[name = tensor("pretrained_out_283_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_283_pad_0 = const()[name = tensor("pretrained_out_283_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260841280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261660544))), name = tensor("layers_23_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_23_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_23_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261660672)))]; + tensor pretrained_out_283_cast_fp16 = conv(bias = layers_23_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_39322, groups = var_37804, pad = pretrained_out_283_pad_0, pad_type = pretrained_out_283_pad_type_0, strides = var_39320, weight = layers_23_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_467_cast_fp16)[name = tensor("pretrained_out_283_cast_fp16")]; + tensor var_39326 = const()[name = tensor("op_39326"), val = tensor([1, 1])]; + tensor var_39328 = const()[name = tensor("op_39328"), val = tensor([1, 1])]; + tensor input_469_pad_type_0 = const()[name = tensor("input_469_pad_type_0"), val = tensor("custom")]; + tensor input_469_pad_0 = const()[name = tensor("input_469_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_23_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261663296)))]; + tensor input_469_cast_fp16 = conv(dilations = var_39328, groups = var_37804, pad = input_469_pad_0, pad_type = input_469_pad_type_0, strides = var_39326, weight = layers_23_self_attn_o_proj_loraA_weight_to_fp16, x = input_467_cast_fp16)[name = tensor("input_469_cast_fp16")]; + tensor var_39332 = const()[name = tensor("op_39332"), val = tensor([1, 1])]; + tensor var_39334 = const()[name = tensor("op_39334"), val = tensor([1, 1])]; + tensor lora_out_565_pad_type_0 = const()[name = tensor("lora_out_565_pad_type_0"), val = tensor("custom")]; + tensor lora_out_565_pad_0 = const()[name = tensor("lora_out_565_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_567_weight_0_to_fp16 = const()[name = tensor("lora_out_567_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261704320)))]; + tensor lora_out_567_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_39334, groups = var_37804, pad = lora_out_565_pad_0, pad_type = lora_out_565_pad_type_0, strides = var_39332, weight = lora_out_567_weight_0_to_fp16, x = input_469_cast_fp16)[name = tensor("lora_out_567_cast_fp16")]; + tensor obj_95_cast_fp16 = add(x = pretrained_out_283_cast_fp16, y = lora_out_567_cast_fp16)[name = tensor("obj_95_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_95_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; + tensor var_39343 = const()[name = tensor("op_39343"), val = tensor([1])]; + tensor channels_mean_95_cast_fp16 = reduce_mean(axes = var_39343, keep_dims = var_37805, x = inputs_95_cast_fp16)[name = tensor("channels_mean_95_cast_fp16")]; + tensor zero_mean_95_cast_fp16 = sub(x = inputs_95_cast_fp16, y = channels_mean_95_cast_fp16)[name = tensor("zero_mean_95_cast_fp16")]; + tensor zero_mean_sq_95_cast_fp16 = mul(x = zero_mean_95_cast_fp16, y = zero_mean_95_cast_fp16)[name = tensor("zero_mean_sq_95_cast_fp16")]; + tensor var_39347 = const()[name = tensor("op_39347"), val = tensor([1])]; + tensor var_39348_cast_fp16 = reduce_mean(axes = var_39347, keep_dims = var_37805, x = zero_mean_sq_95_cast_fp16)[name = tensor("op_39348_cast_fp16")]; + tensor var_39349_to_fp16 = const()[name = tensor("op_39349_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_39350_cast_fp16 = add(x = var_39348_cast_fp16, y = var_39349_to_fp16)[name = tensor("op_39350_cast_fp16")]; + tensor denom_95_epsilon_0 = const()[name = tensor("denom_95_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_95_cast_fp16 = rsqrt(epsilon = denom_95_epsilon_0, x = var_39350_cast_fp16)[name = tensor("denom_95_cast_fp16")]; + tensor out_95_cast_fp16 = mul(x = zero_mean_95_cast_fp16, y = denom_95_cast_fp16)[name = tensor("out_95_cast_fp16")]; + tensor input_471_gamma_0_to_fp16 = const()[name = tensor("input_471_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261745344)))]; + tensor input_471_beta_0_to_fp16 = const()[name = tensor("input_471_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261747968)))]; + tensor input_471_epsilon_0_to_fp16 = const()[name = tensor("input_471_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_471_cast_fp16 = batch_norm(beta = input_471_beta_0_to_fp16, epsilon = input_471_epsilon_0_to_fp16, gamma = input_471_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor("input_471_cast_fp16")]; + tensor var_39364 = const()[name = tensor("op_39364"), val = tensor([1, 1])]; + tensor var_39366 = const()[name = tensor("op_39366"), val = tensor([1, 1])]; + tensor pretrained_out_285_pad_type_0 = const()[name = tensor("pretrained_out_285_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_285_pad_0 = const()[name = tensor("pretrained_out_285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261750592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265027456))), name = tensor("layers_23_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_23_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_23_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265027584)))]; + tensor pretrained_out_285_cast_fp16 = conv(bias = layers_23_fc1_pretrained_bias_to_fp16, dilations = var_39366, groups = var_37804, pad = pretrained_out_285_pad_0, pad_type = pretrained_out_285_pad_type_0, strides = var_39364, weight = layers_23_fc1_pretrained_weight_to_fp16_palettized, x = input_471_cast_fp16)[name = tensor("pretrained_out_285_cast_fp16")]; + tensor var_39370 = const()[name = tensor("op_39370"), val = tensor([1, 1])]; + tensor var_39372 = const()[name = tensor("op_39372"), val = tensor([1, 1])]; + tensor input_473_pad_type_0 = const()[name = tensor("input_473_pad_type_0"), val = tensor("custom")]; + tensor input_473_pad_0 = const()[name = tensor("input_473_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_23_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265037888)))]; + tensor input_473_cast_fp16 = conv(dilations = var_39372, groups = var_37804, pad = input_473_pad_0, pad_type = input_473_pad_type_0, strides = var_39370, weight = layers_23_fc1_loraA_weight_to_fp16, x = input_471_cast_fp16)[name = tensor("input_473_cast_fp16")]; + tensor var_39376 = const()[name = tensor("op_39376"), val = tensor([1, 1])]; + tensor var_39378 = const()[name = tensor("op_39378"), val = tensor([1, 1])]; + tensor lora_out_569_pad_type_0 = const()[name = tensor("lora_out_569_pad_type_0"), val = tensor("custom")]; + tensor lora_out_569_pad_0 = const()[name = tensor("lora_out_569_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_571_weight_0_to_fp16 = const()[name = tensor("lora_out_571_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265078912)))]; + tensor lora_out_571_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_39378, groups = var_37804, pad = lora_out_569_pad_0, pad_type = lora_out_569_pad_type_0, strides = var_39376, weight = lora_out_571_weight_0_to_fp16, x = input_473_cast_fp16)[name = tensor("lora_out_571_cast_fp16")]; + tensor input_475_cast_fp16 = add(x = pretrained_out_285_cast_fp16, y = lora_out_571_cast_fp16)[name = tensor("input_475_cast_fp16")]; + tensor input_477_mode_0 = const()[name = tensor("input_477_mode_0"), val = tensor("EXACT")]; + tensor input_477_cast_fp16 = gelu(mode = input_477_mode_0, x = input_475_cast_fp16)[name = tensor("input_477_cast_fp16")]; + tensor var_39390 = const()[name = tensor("op_39390"), val = tensor([1, 1])]; + tensor var_39392 = const()[name = tensor("op_39392"), val = tensor([1, 1])]; + tensor pretrained_out_287_pad_type_0 = const()[name = tensor("pretrained_out_287_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_287_pad_0 = const()[name = tensor("pretrained_out_287_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265242816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268519680))), name = tensor("layers_23_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_23_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_23_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268519808)))]; + tensor pretrained_out_287_cast_fp16 = conv(bias = layers_23_fc2_pretrained_bias_to_fp16, dilations = var_39392, groups = var_37804, pad = pretrained_out_287_pad_0, pad_type = pretrained_out_287_pad_type_0, strides = var_39390, weight = layers_23_fc2_pretrained_weight_to_fp16_palettized, x = input_477_cast_fp16)[name = tensor("pretrained_out_287_cast_fp16")]; + tensor var_39396 = const()[name = tensor("op_39396"), val = tensor([1, 1])]; + tensor var_39398 = const()[name = tensor("op_39398"), val = tensor([1, 1])]; + tensor input_479_pad_type_0 = const()[name = tensor("input_479_pad_type_0"), val = tensor("custom")]; + tensor input_479_pad_0 = const()[name = tensor("input_479_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_23_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_23_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268522432)))]; + tensor input_479_cast_fp16 = conv(dilations = var_39398, groups = var_37804, pad = input_479_pad_0, pad_type = input_479_pad_type_0, strides = var_39396, weight = layers_23_fc2_loraA_weight_to_fp16, x = input_477_cast_fp16)[name = tensor("input_479_cast_fp16")]; + tensor var_39402 = const()[name = tensor("op_39402"), val = tensor([1, 1])]; + tensor var_39404 = const()[name = tensor("op_39404"), val = tensor([1, 1])]; + tensor lora_out_573_pad_type_0 = const()[name = tensor("lora_out_573_pad_type_0"), val = tensor("custom")]; + tensor lora_out_573_pad_0 = const()[name = tensor("lora_out_573_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_575_weight_0_to_fp16 = const()[name = tensor("lora_out_575_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268686336)))]; + tensor lora_out_575_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_39404, groups = var_37804, pad = lora_out_573_pad_0, pad_type = lora_out_573_pad_type_0, strides = var_39402, weight = lora_out_575_weight_0_to_fp16, x = input_479_cast_fp16)[name = tensor("lora_out_575_cast_fp16")]; + tensor hidden_states_51_cast_fp16 = add(x = pretrained_out_287_cast_fp16, y = lora_out_575_cast_fp16)[name = tensor("hidden_states_51_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = hidden_states_51_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; + tensor var_39414 = const()[name = tensor("op_39414"), val = tensor(3)]; + tensor var_39439 = const()[name = tensor("op_39439"), val = tensor(1)]; + tensor var_39440 = const()[name = tensor("op_39440"), val = tensor(true)]; + tensor var_39450 = const()[name = tensor("op_39450"), val = tensor([1])]; + tensor channels_mean_97_cast_fp16 = reduce_mean(axes = var_39450, keep_dims = var_39440, x = inputs_97_cast_fp16)[name = tensor("channels_mean_97_cast_fp16")]; + tensor zero_mean_97_cast_fp16 = sub(x = inputs_97_cast_fp16, y = channels_mean_97_cast_fp16)[name = tensor("zero_mean_97_cast_fp16")]; + tensor zero_mean_sq_97_cast_fp16 = mul(x = zero_mean_97_cast_fp16, y = zero_mean_97_cast_fp16)[name = tensor("zero_mean_sq_97_cast_fp16")]; + tensor var_39454 = const()[name = tensor("op_39454"), val = tensor([1])]; + tensor var_39455_cast_fp16 = reduce_mean(axes = var_39454, keep_dims = var_39440, x = zero_mean_sq_97_cast_fp16)[name = tensor("op_39455_cast_fp16")]; + tensor var_39456_to_fp16 = const()[name = tensor("op_39456_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_39457_cast_fp16 = add(x = var_39455_cast_fp16, y = var_39456_to_fp16)[name = tensor("op_39457_cast_fp16")]; + tensor denom_97_epsilon_0 = const()[name = tensor("denom_97_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_97_cast_fp16 = rsqrt(epsilon = denom_97_epsilon_0, x = var_39457_cast_fp16)[name = tensor("denom_97_cast_fp16")]; + tensor out_97_cast_fp16 = mul(x = zero_mean_97_cast_fp16, y = denom_97_cast_fp16)[name = tensor("out_97_cast_fp16")]; + tensor obj_97_gamma_0_to_fp16 = const()[name = tensor("obj_97_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268727360)))]; + tensor obj_97_beta_0_to_fp16 = const()[name = tensor("obj_97_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268729984)))]; + tensor obj_97_epsilon_0_to_fp16 = const()[name = tensor("obj_97_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_97_cast_fp16 = batch_norm(beta = obj_97_beta_0_to_fp16, epsilon = obj_97_epsilon_0_to_fp16, gamma = obj_97_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_97_cast_fp16)[name = tensor("obj_97_cast_fp16")]; + tensor var_39475 = const()[name = tensor("op_39475"), val = tensor([1, 1])]; + tensor var_39477 = const()[name = tensor("op_39477"), val = tensor([1, 1])]; + tensor pretrained_out_289_pad_type_0 = const()[name = tensor("pretrained_out_289_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_289_pad_0 = const()[name = tensor("pretrained_out_289_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268732608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269551872))), name = tensor("layers_24_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_24_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269552000)))]; + tensor pretrained_out_289_cast_fp16 = conv(bias = layers_24_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_39477, groups = var_39439, pad = pretrained_out_289_pad_0, pad_type = pretrained_out_289_pad_type_0, strides = var_39475, weight = layers_24_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_97_cast_fp16)[name = tensor("pretrained_out_289_cast_fp16")]; + tensor var_39481 = const()[name = tensor("op_39481"), val = tensor([1, 1])]; + tensor var_39483 = const()[name = tensor("op_39483"), val = tensor([1, 1])]; + tensor input_481_pad_type_0 = const()[name = tensor("input_481_pad_type_0"), val = tensor("custom")]; + tensor input_481_pad_0 = const()[name = tensor("input_481_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269554624)))]; + tensor input_481_cast_fp16 = conv(dilations = var_39483, groups = var_39439, pad = input_481_pad_0, pad_type = input_481_pad_type_0, strides = var_39481, weight = layers_24_self_attn_q_proj_loraA_weight_to_fp16, x = obj_97_cast_fp16)[name = tensor("input_481_cast_fp16")]; + tensor var_39487 = const()[name = tensor("op_39487"), val = tensor([1, 1])]; + tensor var_39489 = const()[name = tensor("op_39489"), val = tensor([1, 1])]; + tensor lora_out_577_pad_type_0 = const()[name = tensor("lora_out_577_pad_type_0"), val = tensor("custom")]; + tensor lora_out_577_pad_0 = const()[name = tensor("lora_out_577_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_579_weight_0_to_fp16 = const()[name = tensor("lora_out_579_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269595648)))]; + tensor lora_out_579_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_39489, groups = var_39439, pad = lora_out_577_pad_0, pad_type = lora_out_577_pad_type_0, strides = var_39487, weight = lora_out_579_weight_0_to_fp16, x = input_481_cast_fp16)[name = tensor("lora_out_579_cast_fp16")]; + tensor query_49_cast_fp16 = add(x = pretrained_out_289_cast_fp16, y = lora_out_579_cast_fp16)[name = tensor("query_49_cast_fp16")]; + tensor var_39499 = const()[name = tensor("op_39499"), val = tensor([1, 1])]; + tensor var_39501 = const()[name = tensor("op_39501"), val = tensor([1, 1])]; + tensor pretrained_out_291_pad_type_0 = const()[name = tensor("pretrained_out_291_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_291_pad_0 = const()[name = tensor("pretrained_out_291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269636672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270455936))), name = tensor("layers_24_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_291_cast_fp16 = conv(dilations = var_39501, groups = var_39439, pad = pretrained_out_291_pad_0, pad_type = pretrained_out_291_pad_type_0, strides = var_39499, weight = layers_24_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_97_cast_fp16)[name = tensor("pretrained_out_291_cast_fp16")]; + tensor var_39505 = const()[name = tensor("op_39505"), val = tensor([1, 1])]; + tensor var_39507 = const()[name = tensor("op_39507"), val = tensor([1, 1])]; + tensor input_483_pad_type_0 = const()[name = tensor("input_483_pad_type_0"), val = tensor("custom")]; + tensor input_483_pad_0 = const()[name = tensor("input_483_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270456064)))]; + tensor input_483_cast_fp16 = conv(dilations = var_39507, groups = var_39439, pad = input_483_pad_0, pad_type = input_483_pad_type_0, strides = var_39505, weight = layers_24_self_attn_k_proj_loraA_weight_to_fp16, x = obj_97_cast_fp16)[name = tensor("input_483_cast_fp16")]; + tensor var_39511 = const()[name = tensor("op_39511"), val = tensor([1, 1])]; + tensor var_39513 = const()[name = tensor("op_39513"), val = tensor([1, 1])]; + tensor lora_out_581_pad_type_0 = const()[name = tensor("lora_out_581_pad_type_0"), val = tensor("custom")]; + tensor lora_out_581_pad_0 = const()[name = tensor("lora_out_581_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_583_weight_0_to_fp16 = const()[name = tensor("lora_out_583_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270497088)))]; + tensor lora_out_583_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_39513, groups = var_39439, pad = lora_out_581_pad_0, pad_type = lora_out_581_pad_type_0, strides = var_39511, weight = lora_out_583_weight_0_to_fp16, x = input_483_cast_fp16)[name = tensor("lora_out_583_cast_fp16")]; + tensor key_49_cast_fp16 = add(x = pretrained_out_291_cast_fp16, y = lora_out_583_cast_fp16)[name = tensor("key_49_cast_fp16")]; + tensor var_39524 = const()[name = tensor("op_39524"), val = tensor([1, 1])]; + tensor var_39526 = const()[name = tensor("op_39526"), val = tensor([1, 1])]; + tensor pretrained_out_293_pad_type_0 = const()[name = tensor("pretrained_out_293_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_293_pad_0 = const()[name = tensor("pretrained_out_293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270538112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(271357376))), name = tensor("layers_24_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_24_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(271357504)))]; + tensor pretrained_out_293_cast_fp16 = conv(bias = layers_24_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_39526, groups = var_39439, pad = pretrained_out_293_pad_0, pad_type = pretrained_out_293_pad_type_0, strides = var_39524, weight = layers_24_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_97_cast_fp16)[name = tensor("pretrained_out_293_cast_fp16")]; + tensor var_39530 = const()[name = tensor("op_39530"), val = tensor([1, 1])]; + tensor var_39532 = const()[name = tensor("op_39532"), val = tensor([1, 1])]; + tensor input_485_pad_type_0 = const()[name = tensor("input_485_pad_type_0"), val = tensor("custom")]; + tensor input_485_pad_0 = const()[name = tensor("input_485_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(271360128)))]; + tensor input_485_cast_fp16 = conv(dilations = var_39532, groups = var_39439, pad = input_485_pad_0, pad_type = input_485_pad_type_0, strides = var_39530, weight = layers_24_self_attn_v_proj_loraA_weight_to_fp16, x = obj_97_cast_fp16)[name = tensor("input_485_cast_fp16")]; + tensor var_39536 = const()[name = tensor("op_39536"), val = tensor([1, 1])]; + tensor var_39538 = const()[name = tensor("op_39538"), val = tensor([1, 1])]; + tensor lora_out_585_pad_type_0 = const()[name = tensor("lora_out_585_pad_type_0"), val = tensor("custom")]; + tensor lora_out_585_pad_0 = const()[name = tensor("lora_out_585_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_587_weight_0_to_fp16 = const()[name = tensor("lora_out_587_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(271401152)))]; + tensor lora_out_587_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_39538, groups = var_39439, pad = lora_out_585_pad_0, pad_type = lora_out_585_pad_type_0, strides = var_39536, weight = lora_out_587_weight_0_to_fp16, x = input_485_cast_fp16)[name = tensor("lora_out_587_cast_fp16")]; + tensor value_49_cast_fp16 = add(x = pretrained_out_293_cast_fp16, y = lora_out_587_cast_fp16)[name = tensor("value_49_cast_fp16")]; + tensor var_39548_begin_0 = const()[name = tensor("op_39548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39548_end_0 = const()[name = tensor("op_39548_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39548_end_mask_0 = const()[name = tensor("op_39548_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39548_cast_fp16 = slice_by_index(begin = var_39548_begin_0, end = var_39548_end_0, end_mask = var_39548_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39548_cast_fp16")]; + tensor var_39552_begin_0 = const()[name = tensor("op_39552_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_39552_end_0 = const()[name = tensor("op_39552_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_39552_end_mask_0 = const()[name = tensor("op_39552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39552_cast_fp16 = slice_by_index(begin = var_39552_begin_0, end = var_39552_end_0, end_mask = var_39552_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39552_cast_fp16")]; + tensor var_39556_begin_0 = const()[name = tensor("op_39556_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_39556_end_0 = const()[name = tensor("op_39556_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_39556_end_mask_0 = const()[name = tensor("op_39556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39556_cast_fp16 = slice_by_index(begin = var_39556_begin_0, end = var_39556_end_0, end_mask = var_39556_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39556_cast_fp16")]; + tensor var_39560_begin_0 = const()[name = tensor("op_39560_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_39560_end_0 = const()[name = tensor("op_39560_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_39560_end_mask_0 = const()[name = tensor("op_39560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39560_cast_fp16 = slice_by_index(begin = var_39560_begin_0, end = var_39560_end_0, end_mask = var_39560_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39560_cast_fp16")]; + tensor var_39564_begin_0 = const()[name = tensor("op_39564_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_39564_end_0 = const()[name = tensor("op_39564_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_39564_end_mask_0 = const()[name = tensor("op_39564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39564_cast_fp16 = slice_by_index(begin = var_39564_begin_0, end = var_39564_end_0, end_mask = var_39564_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39564_cast_fp16")]; + tensor var_39568_begin_0 = const()[name = tensor("op_39568_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_39568_end_0 = const()[name = tensor("op_39568_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_39568_end_mask_0 = const()[name = tensor("op_39568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39568_cast_fp16 = slice_by_index(begin = var_39568_begin_0, end = var_39568_end_0, end_mask = var_39568_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39568_cast_fp16")]; + tensor var_39572_begin_0 = const()[name = tensor("op_39572_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_39572_end_0 = const()[name = tensor("op_39572_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_39572_end_mask_0 = const()[name = tensor("op_39572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39572_cast_fp16 = slice_by_index(begin = var_39572_begin_0, end = var_39572_end_0, end_mask = var_39572_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39572_cast_fp16")]; + tensor var_39576_begin_0 = const()[name = tensor("op_39576_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_39576_end_0 = const()[name = tensor("op_39576_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_39576_end_mask_0 = const()[name = tensor("op_39576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39576_cast_fp16 = slice_by_index(begin = var_39576_begin_0, end = var_39576_end_0, end_mask = var_39576_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39576_cast_fp16")]; + tensor var_39580_begin_0 = const()[name = tensor("op_39580_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_39580_end_0 = const()[name = tensor("op_39580_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_39580_end_mask_0 = const()[name = tensor("op_39580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39580_cast_fp16 = slice_by_index(begin = var_39580_begin_0, end = var_39580_end_0, end_mask = var_39580_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39580_cast_fp16")]; + tensor var_39584_begin_0 = const()[name = tensor("op_39584_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_39584_end_0 = const()[name = tensor("op_39584_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_39584_end_mask_0 = const()[name = tensor("op_39584_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39584_cast_fp16 = slice_by_index(begin = var_39584_begin_0, end = var_39584_end_0, end_mask = var_39584_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39584_cast_fp16")]; + tensor var_39588_begin_0 = const()[name = tensor("op_39588_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_39588_end_0 = const()[name = tensor("op_39588_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_39588_end_mask_0 = const()[name = tensor("op_39588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39588_cast_fp16 = slice_by_index(begin = var_39588_begin_0, end = var_39588_end_0, end_mask = var_39588_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39588_cast_fp16")]; + tensor var_39592_begin_0 = const()[name = tensor("op_39592_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_39592_end_0 = const()[name = tensor("op_39592_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_39592_end_mask_0 = const()[name = tensor("op_39592_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39592_cast_fp16 = slice_by_index(begin = var_39592_begin_0, end = var_39592_end_0, end_mask = var_39592_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39592_cast_fp16")]; + tensor var_39596_begin_0 = const()[name = tensor("op_39596_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_39596_end_0 = const()[name = tensor("op_39596_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_39596_end_mask_0 = const()[name = tensor("op_39596_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39596_cast_fp16 = slice_by_index(begin = var_39596_begin_0, end = var_39596_end_0, end_mask = var_39596_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39596_cast_fp16")]; + tensor var_39600_begin_0 = const()[name = tensor("op_39600_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_39600_end_0 = const()[name = tensor("op_39600_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_39600_end_mask_0 = const()[name = tensor("op_39600_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39600_cast_fp16 = slice_by_index(begin = var_39600_begin_0, end = var_39600_end_0, end_mask = var_39600_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39600_cast_fp16")]; + tensor var_39604_begin_0 = const()[name = tensor("op_39604_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_39604_end_0 = const()[name = tensor("op_39604_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_39604_end_mask_0 = const()[name = tensor("op_39604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39604_cast_fp16 = slice_by_index(begin = var_39604_begin_0, end = var_39604_end_0, end_mask = var_39604_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39604_cast_fp16")]; + tensor var_39608_begin_0 = const()[name = tensor("op_39608_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_39608_end_0 = const()[name = tensor("op_39608_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_39608_end_mask_0 = const()[name = tensor("op_39608_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39608_cast_fp16 = slice_by_index(begin = var_39608_begin_0, end = var_39608_end_0, end_mask = var_39608_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39608_cast_fp16")]; + tensor var_39612_begin_0 = const()[name = tensor("op_39612_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_39612_end_0 = const()[name = tensor("op_39612_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_39612_end_mask_0 = const()[name = tensor("op_39612_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39612_cast_fp16 = slice_by_index(begin = var_39612_begin_0, end = var_39612_end_0, end_mask = var_39612_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39612_cast_fp16")]; + tensor var_39616_begin_0 = const()[name = tensor("op_39616_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_39616_end_0 = const()[name = tensor("op_39616_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_39616_end_mask_0 = const()[name = tensor("op_39616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39616_cast_fp16 = slice_by_index(begin = var_39616_begin_0, end = var_39616_end_0, end_mask = var_39616_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39616_cast_fp16")]; + tensor var_39620_begin_0 = const()[name = tensor("op_39620_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_39620_end_0 = const()[name = tensor("op_39620_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_39620_end_mask_0 = const()[name = tensor("op_39620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39620_cast_fp16 = slice_by_index(begin = var_39620_begin_0, end = var_39620_end_0, end_mask = var_39620_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39620_cast_fp16")]; + tensor var_39624_begin_0 = const()[name = tensor("op_39624_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_39624_end_0 = const()[name = tensor("op_39624_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_39624_end_mask_0 = const()[name = tensor("op_39624_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_39624_cast_fp16 = slice_by_index(begin = var_39624_begin_0, end = var_39624_end_0, end_mask = var_39624_end_mask_0, x = query_49_cast_fp16)[name = tensor("op_39624_cast_fp16")]; + tensor var_39633_begin_0 = const()[name = tensor("op_39633_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39633_end_0 = const()[name = tensor("op_39633_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39633_end_mask_0 = const()[name = tensor("op_39633_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39633_cast_fp16 = slice_by_index(begin = var_39633_begin_0, end = var_39633_end_0, end_mask = var_39633_end_mask_0, x = var_39548_cast_fp16)[name = tensor("op_39633_cast_fp16")]; + tensor var_39640_begin_0 = const()[name = tensor("op_39640_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39640_end_0 = const()[name = tensor("op_39640_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39640_end_mask_0 = const()[name = tensor("op_39640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39640_cast_fp16 = slice_by_index(begin = var_39640_begin_0, end = var_39640_end_0, end_mask = var_39640_end_mask_0, x = var_39548_cast_fp16)[name = tensor("op_39640_cast_fp16")]; + tensor var_39647_begin_0 = const()[name = tensor("op_39647_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39647_end_0 = const()[name = tensor("op_39647_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39647_end_mask_0 = const()[name = tensor("op_39647_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39647_cast_fp16 = slice_by_index(begin = var_39647_begin_0, end = var_39647_end_0, end_mask = var_39647_end_mask_0, x = var_39548_cast_fp16)[name = tensor("op_39647_cast_fp16")]; + tensor var_39654_begin_0 = const()[name = tensor("op_39654_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39654_end_0 = const()[name = tensor("op_39654_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39654_end_mask_0 = const()[name = tensor("op_39654_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39654_cast_fp16 = slice_by_index(begin = var_39654_begin_0, end = var_39654_end_0, end_mask = var_39654_end_mask_0, x = var_39548_cast_fp16)[name = tensor("op_39654_cast_fp16")]; + tensor var_39661_begin_0 = const()[name = tensor("op_39661_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39661_end_0 = const()[name = tensor("op_39661_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39661_end_mask_0 = const()[name = tensor("op_39661_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39661_cast_fp16 = slice_by_index(begin = var_39661_begin_0, end = var_39661_end_0, end_mask = var_39661_end_mask_0, x = var_39552_cast_fp16)[name = tensor("op_39661_cast_fp16")]; + tensor var_39668_begin_0 = const()[name = tensor("op_39668_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39668_end_0 = const()[name = tensor("op_39668_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39668_end_mask_0 = const()[name = tensor("op_39668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39668_cast_fp16 = slice_by_index(begin = var_39668_begin_0, end = var_39668_end_0, end_mask = var_39668_end_mask_0, x = var_39552_cast_fp16)[name = tensor("op_39668_cast_fp16")]; + tensor var_39675_begin_0 = const()[name = tensor("op_39675_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39675_end_0 = const()[name = tensor("op_39675_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39675_end_mask_0 = const()[name = tensor("op_39675_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39675_cast_fp16 = slice_by_index(begin = var_39675_begin_0, end = var_39675_end_0, end_mask = var_39675_end_mask_0, x = var_39552_cast_fp16)[name = tensor("op_39675_cast_fp16")]; + tensor var_39682_begin_0 = const()[name = tensor("op_39682_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39682_end_0 = const()[name = tensor("op_39682_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39682_end_mask_0 = const()[name = tensor("op_39682_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39682_cast_fp16 = slice_by_index(begin = var_39682_begin_0, end = var_39682_end_0, end_mask = var_39682_end_mask_0, x = var_39552_cast_fp16)[name = tensor("op_39682_cast_fp16")]; + tensor var_39689_begin_0 = const()[name = tensor("op_39689_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39689_end_0 = const()[name = tensor("op_39689_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39689_end_mask_0 = const()[name = tensor("op_39689_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39689_cast_fp16 = slice_by_index(begin = var_39689_begin_0, end = var_39689_end_0, end_mask = var_39689_end_mask_0, x = var_39556_cast_fp16)[name = tensor("op_39689_cast_fp16")]; + tensor var_39696_begin_0 = const()[name = tensor("op_39696_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39696_end_0 = const()[name = tensor("op_39696_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39696_end_mask_0 = const()[name = tensor("op_39696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39696_cast_fp16 = slice_by_index(begin = var_39696_begin_0, end = var_39696_end_0, end_mask = var_39696_end_mask_0, x = var_39556_cast_fp16)[name = tensor("op_39696_cast_fp16")]; + tensor var_39703_begin_0 = const()[name = tensor("op_39703_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39703_end_0 = const()[name = tensor("op_39703_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39703_end_mask_0 = const()[name = tensor("op_39703_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39703_cast_fp16 = slice_by_index(begin = var_39703_begin_0, end = var_39703_end_0, end_mask = var_39703_end_mask_0, x = var_39556_cast_fp16)[name = tensor("op_39703_cast_fp16")]; + tensor var_39710_begin_0 = const()[name = tensor("op_39710_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39710_end_0 = const()[name = tensor("op_39710_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39710_end_mask_0 = const()[name = tensor("op_39710_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39710_cast_fp16 = slice_by_index(begin = var_39710_begin_0, end = var_39710_end_0, end_mask = var_39710_end_mask_0, x = var_39556_cast_fp16)[name = tensor("op_39710_cast_fp16")]; + tensor var_39717_begin_0 = const()[name = tensor("op_39717_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39717_end_0 = const()[name = tensor("op_39717_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39717_end_mask_0 = const()[name = tensor("op_39717_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39717_cast_fp16 = slice_by_index(begin = var_39717_begin_0, end = var_39717_end_0, end_mask = var_39717_end_mask_0, x = var_39560_cast_fp16)[name = tensor("op_39717_cast_fp16")]; + tensor var_39724_begin_0 = const()[name = tensor("op_39724_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39724_end_0 = const()[name = tensor("op_39724_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39724_end_mask_0 = const()[name = tensor("op_39724_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39724_cast_fp16 = slice_by_index(begin = var_39724_begin_0, end = var_39724_end_0, end_mask = var_39724_end_mask_0, x = var_39560_cast_fp16)[name = tensor("op_39724_cast_fp16")]; + tensor var_39731_begin_0 = const()[name = tensor("op_39731_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39731_end_0 = const()[name = tensor("op_39731_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39731_end_mask_0 = const()[name = tensor("op_39731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39731_cast_fp16 = slice_by_index(begin = var_39731_begin_0, end = var_39731_end_0, end_mask = var_39731_end_mask_0, x = var_39560_cast_fp16)[name = tensor("op_39731_cast_fp16")]; + tensor var_39738_begin_0 = const()[name = tensor("op_39738_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39738_end_0 = const()[name = tensor("op_39738_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39738_end_mask_0 = const()[name = tensor("op_39738_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39738_cast_fp16 = slice_by_index(begin = var_39738_begin_0, end = var_39738_end_0, end_mask = var_39738_end_mask_0, x = var_39560_cast_fp16)[name = tensor("op_39738_cast_fp16")]; + tensor var_39745_begin_0 = const()[name = tensor("op_39745_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39745_end_0 = const()[name = tensor("op_39745_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39745_end_mask_0 = const()[name = tensor("op_39745_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39745_cast_fp16 = slice_by_index(begin = var_39745_begin_0, end = var_39745_end_0, end_mask = var_39745_end_mask_0, x = var_39564_cast_fp16)[name = tensor("op_39745_cast_fp16")]; + tensor var_39752_begin_0 = const()[name = tensor("op_39752_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39752_end_0 = const()[name = tensor("op_39752_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39752_end_mask_0 = const()[name = tensor("op_39752_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39752_cast_fp16 = slice_by_index(begin = var_39752_begin_0, end = var_39752_end_0, end_mask = var_39752_end_mask_0, x = var_39564_cast_fp16)[name = tensor("op_39752_cast_fp16")]; + tensor var_39759_begin_0 = const()[name = tensor("op_39759_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39759_end_0 = const()[name = tensor("op_39759_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39759_end_mask_0 = const()[name = tensor("op_39759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39759_cast_fp16 = slice_by_index(begin = var_39759_begin_0, end = var_39759_end_0, end_mask = var_39759_end_mask_0, x = var_39564_cast_fp16)[name = tensor("op_39759_cast_fp16")]; + tensor var_39766_begin_0 = const()[name = tensor("op_39766_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39766_end_0 = const()[name = tensor("op_39766_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39766_end_mask_0 = const()[name = tensor("op_39766_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39766_cast_fp16 = slice_by_index(begin = var_39766_begin_0, end = var_39766_end_0, end_mask = var_39766_end_mask_0, x = var_39564_cast_fp16)[name = tensor("op_39766_cast_fp16")]; + tensor var_39773_begin_0 = const()[name = tensor("op_39773_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39773_end_0 = const()[name = tensor("op_39773_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39773_end_mask_0 = const()[name = tensor("op_39773_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39773_cast_fp16 = slice_by_index(begin = var_39773_begin_0, end = var_39773_end_0, end_mask = var_39773_end_mask_0, x = var_39568_cast_fp16)[name = tensor("op_39773_cast_fp16")]; + tensor var_39780_begin_0 = const()[name = tensor("op_39780_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39780_end_0 = const()[name = tensor("op_39780_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39780_end_mask_0 = const()[name = tensor("op_39780_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39780_cast_fp16 = slice_by_index(begin = var_39780_begin_0, end = var_39780_end_0, end_mask = var_39780_end_mask_0, x = var_39568_cast_fp16)[name = tensor("op_39780_cast_fp16")]; + tensor var_39787_begin_0 = const()[name = tensor("op_39787_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39787_end_0 = const()[name = tensor("op_39787_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39787_end_mask_0 = const()[name = tensor("op_39787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39787_cast_fp16 = slice_by_index(begin = var_39787_begin_0, end = var_39787_end_0, end_mask = var_39787_end_mask_0, x = var_39568_cast_fp16)[name = tensor("op_39787_cast_fp16")]; + tensor var_39794_begin_0 = const()[name = tensor("op_39794_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39794_end_0 = const()[name = tensor("op_39794_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39794_end_mask_0 = const()[name = tensor("op_39794_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39794_cast_fp16 = slice_by_index(begin = var_39794_begin_0, end = var_39794_end_0, end_mask = var_39794_end_mask_0, x = var_39568_cast_fp16)[name = tensor("op_39794_cast_fp16")]; + tensor var_39801_begin_0 = const()[name = tensor("op_39801_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39801_end_0 = const()[name = tensor("op_39801_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39801_end_mask_0 = const()[name = tensor("op_39801_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39801_cast_fp16 = slice_by_index(begin = var_39801_begin_0, end = var_39801_end_0, end_mask = var_39801_end_mask_0, x = var_39572_cast_fp16)[name = tensor("op_39801_cast_fp16")]; + tensor var_39808_begin_0 = const()[name = tensor("op_39808_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39808_end_0 = const()[name = tensor("op_39808_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39808_end_mask_0 = const()[name = tensor("op_39808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39808_cast_fp16 = slice_by_index(begin = var_39808_begin_0, end = var_39808_end_0, end_mask = var_39808_end_mask_0, x = var_39572_cast_fp16)[name = tensor("op_39808_cast_fp16")]; + tensor var_39815_begin_0 = const()[name = tensor("op_39815_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39815_end_0 = const()[name = tensor("op_39815_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39815_end_mask_0 = const()[name = tensor("op_39815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39815_cast_fp16 = slice_by_index(begin = var_39815_begin_0, end = var_39815_end_0, end_mask = var_39815_end_mask_0, x = var_39572_cast_fp16)[name = tensor("op_39815_cast_fp16")]; + tensor var_39822_begin_0 = const()[name = tensor("op_39822_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39822_end_0 = const()[name = tensor("op_39822_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39822_end_mask_0 = const()[name = tensor("op_39822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39822_cast_fp16 = slice_by_index(begin = var_39822_begin_0, end = var_39822_end_0, end_mask = var_39822_end_mask_0, x = var_39572_cast_fp16)[name = tensor("op_39822_cast_fp16")]; + tensor var_39829_begin_0 = const()[name = tensor("op_39829_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39829_end_0 = const()[name = tensor("op_39829_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39829_end_mask_0 = const()[name = tensor("op_39829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39829_cast_fp16 = slice_by_index(begin = var_39829_begin_0, end = var_39829_end_0, end_mask = var_39829_end_mask_0, x = var_39576_cast_fp16)[name = tensor("op_39829_cast_fp16")]; + tensor var_39836_begin_0 = const()[name = tensor("op_39836_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39836_end_0 = const()[name = tensor("op_39836_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39836_end_mask_0 = const()[name = tensor("op_39836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39836_cast_fp16 = slice_by_index(begin = var_39836_begin_0, end = var_39836_end_0, end_mask = var_39836_end_mask_0, x = var_39576_cast_fp16)[name = tensor("op_39836_cast_fp16")]; + tensor var_39843_begin_0 = const()[name = tensor("op_39843_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39843_end_0 = const()[name = tensor("op_39843_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39843_end_mask_0 = const()[name = tensor("op_39843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39843_cast_fp16 = slice_by_index(begin = var_39843_begin_0, end = var_39843_end_0, end_mask = var_39843_end_mask_0, x = var_39576_cast_fp16)[name = tensor("op_39843_cast_fp16")]; + tensor var_39850_begin_0 = const()[name = tensor("op_39850_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39850_end_0 = const()[name = tensor("op_39850_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39850_end_mask_0 = const()[name = tensor("op_39850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39850_cast_fp16 = slice_by_index(begin = var_39850_begin_0, end = var_39850_end_0, end_mask = var_39850_end_mask_0, x = var_39576_cast_fp16)[name = tensor("op_39850_cast_fp16")]; + tensor var_39857_begin_0 = const()[name = tensor("op_39857_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39857_end_0 = const()[name = tensor("op_39857_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39857_end_mask_0 = const()[name = tensor("op_39857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39857_cast_fp16 = slice_by_index(begin = var_39857_begin_0, end = var_39857_end_0, end_mask = var_39857_end_mask_0, x = var_39580_cast_fp16)[name = tensor("op_39857_cast_fp16")]; + tensor var_39864_begin_0 = const()[name = tensor("op_39864_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39864_end_0 = const()[name = tensor("op_39864_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39864_end_mask_0 = const()[name = tensor("op_39864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39864_cast_fp16 = slice_by_index(begin = var_39864_begin_0, end = var_39864_end_0, end_mask = var_39864_end_mask_0, x = var_39580_cast_fp16)[name = tensor("op_39864_cast_fp16")]; + tensor var_39871_begin_0 = const()[name = tensor("op_39871_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39871_end_0 = const()[name = tensor("op_39871_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39871_end_mask_0 = const()[name = tensor("op_39871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39871_cast_fp16 = slice_by_index(begin = var_39871_begin_0, end = var_39871_end_0, end_mask = var_39871_end_mask_0, x = var_39580_cast_fp16)[name = tensor("op_39871_cast_fp16")]; + tensor var_39878_begin_0 = const()[name = tensor("op_39878_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39878_end_0 = const()[name = tensor("op_39878_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39878_end_mask_0 = const()[name = tensor("op_39878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39878_cast_fp16 = slice_by_index(begin = var_39878_begin_0, end = var_39878_end_0, end_mask = var_39878_end_mask_0, x = var_39580_cast_fp16)[name = tensor("op_39878_cast_fp16")]; + tensor var_39885_begin_0 = const()[name = tensor("op_39885_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39885_end_0 = const()[name = tensor("op_39885_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39885_end_mask_0 = const()[name = tensor("op_39885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39885_cast_fp16 = slice_by_index(begin = var_39885_begin_0, end = var_39885_end_0, end_mask = var_39885_end_mask_0, x = var_39584_cast_fp16)[name = tensor("op_39885_cast_fp16")]; + tensor var_39892_begin_0 = const()[name = tensor("op_39892_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39892_end_0 = const()[name = tensor("op_39892_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39892_end_mask_0 = const()[name = tensor("op_39892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39892_cast_fp16 = slice_by_index(begin = var_39892_begin_0, end = var_39892_end_0, end_mask = var_39892_end_mask_0, x = var_39584_cast_fp16)[name = tensor("op_39892_cast_fp16")]; + tensor var_39899_begin_0 = const()[name = tensor("op_39899_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39899_end_0 = const()[name = tensor("op_39899_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39899_end_mask_0 = const()[name = tensor("op_39899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39899_cast_fp16 = slice_by_index(begin = var_39899_begin_0, end = var_39899_end_0, end_mask = var_39899_end_mask_0, x = var_39584_cast_fp16)[name = tensor("op_39899_cast_fp16")]; + tensor var_39906_begin_0 = const()[name = tensor("op_39906_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39906_end_0 = const()[name = tensor("op_39906_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39906_end_mask_0 = const()[name = tensor("op_39906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39906_cast_fp16 = slice_by_index(begin = var_39906_begin_0, end = var_39906_end_0, end_mask = var_39906_end_mask_0, x = var_39584_cast_fp16)[name = tensor("op_39906_cast_fp16")]; + tensor var_39913_begin_0 = const()[name = tensor("op_39913_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39913_end_0 = const()[name = tensor("op_39913_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39913_end_mask_0 = const()[name = tensor("op_39913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39913_cast_fp16 = slice_by_index(begin = var_39913_begin_0, end = var_39913_end_0, end_mask = var_39913_end_mask_0, x = var_39588_cast_fp16)[name = tensor("op_39913_cast_fp16")]; + tensor var_39920_begin_0 = const()[name = tensor("op_39920_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39920_end_0 = const()[name = tensor("op_39920_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39920_end_mask_0 = const()[name = tensor("op_39920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39920_cast_fp16 = slice_by_index(begin = var_39920_begin_0, end = var_39920_end_0, end_mask = var_39920_end_mask_0, x = var_39588_cast_fp16)[name = tensor("op_39920_cast_fp16")]; + tensor var_39927_begin_0 = const()[name = tensor("op_39927_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39927_end_0 = const()[name = tensor("op_39927_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39927_end_mask_0 = const()[name = tensor("op_39927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39927_cast_fp16 = slice_by_index(begin = var_39927_begin_0, end = var_39927_end_0, end_mask = var_39927_end_mask_0, x = var_39588_cast_fp16)[name = tensor("op_39927_cast_fp16")]; + tensor var_39934_begin_0 = const()[name = tensor("op_39934_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39934_end_0 = const()[name = tensor("op_39934_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39934_end_mask_0 = const()[name = tensor("op_39934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39934_cast_fp16 = slice_by_index(begin = var_39934_begin_0, end = var_39934_end_0, end_mask = var_39934_end_mask_0, x = var_39588_cast_fp16)[name = tensor("op_39934_cast_fp16")]; + tensor var_39941_begin_0 = const()[name = tensor("op_39941_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39941_end_0 = const()[name = tensor("op_39941_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39941_end_mask_0 = const()[name = tensor("op_39941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39941_cast_fp16 = slice_by_index(begin = var_39941_begin_0, end = var_39941_end_0, end_mask = var_39941_end_mask_0, x = var_39592_cast_fp16)[name = tensor("op_39941_cast_fp16")]; + tensor var_39948_begin_0 = const()[name = tensor("op_39948_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39948_end_0 = const()[name = tensor("op_39948_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39948_end_mask_0 = const()[name = tensor("op_39948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39948_cast_fp16 = slice_by_index(begin = var_39948_begin_0, end = var_39948_end_0, end_mask = var_39948_end_mask_0, x = var_39592_cast_fp16)[name = tensor("op_39948_cast_fp16")]; + tensor var_39955_begin_0 = const()[name = tensor("op_39955_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39955_end_0 = const()[name = tensor("op_39955_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39955_end_mask_0 = const()[name = tensor("op_39955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39955_cast_fp16 = slice_by_index(begin = var_39955_begin_0, end = var_39955_end_0, end_mask = var_39955_end_mask_0, x = var_39592_cast_fp16)[name = tensor("op_39955_cast_fp16")]; + tensor var_39962_begin_0 = const()[name = tensor("op_39962_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39962_end_0 = const()[name = tensor("op_39962_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39962_end_mask_0 = const()[name = tensor("op_39962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39962_cast_fp16 = slice_by_index(begin = var_39962_begin_0, end = var_39962_end_0, end_mask = var_39962_end_mask_0, x = var_39592_cast_fp16)[name = tensor("op_39962_cast_fp16")]; + tensor var_39969_begin_0 = const()[name = tensor("op_39969_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39969_end_0 = const()[name = tensor("op_39969_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39969_end_mask_0 = const()[name = tensor("op_39969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39969_cast_fp16 = slice_by_index(begin = var_39969_begin_0, end = var_39969_end_0, end_mask = var_39969_end_mask_0, x = var_39596_cast_fp16)[name = tensor("op_39969_cast_fp16")]; + tensor var_39976_begin_0 = const()[name = tensor("op_39976_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_39976_end_0 = const()[name = tensor("op_39976_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_39976_end_mask_0 = const()[name = tensor("op_39976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39976_cast_fp16 = slice_by_index(begin = var_39976_begin_0, end = var_39976_end_0, end_mask = var_39976_end_mask_0, x = var_39596_cast_fp16)[name = tensor("op_39976_cast_fp16")]; + tensor var_39983_begin_0 = const()[name = tensor("op_39983_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_39983_end_0 = const()[name = tensor("op_39983_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_39983_end_mask_0 = const()[name = tensor("op_39983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39983_cast_fp16 = slice_by_index(begin = var_39983_begin_0, end = var_39983_end_0, end_mask = var_39983_end_mask_0, x = var_39596_cast_fp16)[name = tensor("op_39983_cast_fp16")]; + tensor var_39990_begin_0 = const()[name = tensor("op_39990_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_39990_end_0 = const()[name = tensor("op_39990_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_39990_end_mask_0 = const()[name = tensor("op_39990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39990_cast_fp16 = slice_by_index(begin = var_39990_begin_0, end = var_39990_end_0, end_mask = var_39990_end_mask_0, x = var_39596_cast_fp16)[name = tensor("op_39990_cast_fp16")]; + tensor var_39997_begin_0 = const()[name = tensor("op_39997_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_39997_end_0 = const()[name = tensor("op_39997_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_39997_end_mask_0 = const()[name = tensor("op_39997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_39997_cast_fp16 = slice_by_index(begin = var_39997_begin_0, end = var_39997_end_0, end_mask = var_39997_end_mask_0, x = var_39600_cast_fp16)[name = tensor("op_39997_cast_fp16")]; + tensor var_40004_begin_0 = const()[name = tensor("op_40004_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40004_end_0 = const()[name = tensor("op_40004_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40004_end_mask_0 = const()[name = tensor("op_40004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40004_cast_fp16 = slice_by_index(begin = var_40004_begin_0, end = var_40004_end_0, end_mask = var_40004_end_mask_0, x = var_39600_cast_fp16)[name = tensor("op_40004_cast_fp16")]; + tensor var_40011_begin_0 = const()[name = tensor("op_40011_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40011_end_0 = const()[name = tensor("op_40011_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40011_end_mask_0 = const()[name = tensor("op_40011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40011_cast_fp16 = slice_by_index(begin = var_40011_begin_0, end = var_40011_end_0, end_mask = var_40011_end_mask_0, x = var_39600_cast_fp16)[name = tensor("op_40011_cast_fp16")]; + tensor var_40018_begin_0 = const()[name = tensor("op_40018_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40018_end_0 = const()[name = tensor("op_40018_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40018_end_mask_0 = const()[name = tensor("op_40018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40018_cast_fp16 = slice_by_index(begin = var_40018_begin_0, end = var_40018_end_0, end_mask = var_40018_end_mask_0, x = var_39600_cast_fp16)[name = tensor("op_40018_cast_fp16")]; + tensor var_40025_begin_0 = const()[name = tensor("op_40025_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40025_end_0 = const()[name = tensor("op_40025_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40025_end_mask_0 = const()[name = tensor("op_40025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40025_cast_fp16 = slice_by_index(begin = var_40025_begin_0, end = var_40025_end_0, end_mask = var_40025_end_mask_0, x = var_39604_cast_fp16)[name = tensor("op_40025_cast_fp16")]; + tensor var_40032_begin_0 = const()[name = tensor("op_40032_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40032_end_0 = const()[name = tensor("op_40032_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40032_end_mask_0 = const()[name = tensor("op_40032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40032_cast_fp16 = slice_by_index(begin = var_40032_begin_0, end = var_40032_end_0, end_mask = var_40032_end_mask_0, x = var_39604_cast_fp16)[name = tensor("op_40032_cast_fp16")]; + tensor var_40039_begin_0 = const()[name = tensor("op_40039_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40039_end_0 = const()[name = tensor("op_40039_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40039_end_mask_0 = const()[name = tensor("op_40039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40039_cast_fp16 = slice_by_index(begin = var_40039_begin_0, end = var_40039_end_0, end_mask = var_40039_end_mask_0, x = var_39604_cast_fp16)[name = tensor("op_40039_cast_fp16")]; + tensor var_40046_begin_0 = const()[name = tensor("op_40046_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40046_end_0 = const()[name = tensor("op_40046_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40046_end_mask_0 = const()[name = tensor("op_40046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40046_cast_fp16 = slice_by_index(begin = var_40046_begin_0, end = var_40046_end_0, end_mask = var_40046_end_mask_0, x = var_39604_cast_fp16)[name = tensor("op_40046_cast_fp16")]; + tensor var_40053_begin_0 = const()[name = tensor("op_40053_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40053_end_0 = const()[name = tensor("op_40053_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40053_end_mask_0 = const()[name = tensor("op_40053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40053_cast_fp16 = slice_by_index(begin = var_40053_begin_0, end = var_40053_end_0, end_mask = var_40053_end_mask_0, x = var_39608_cast_fp16)[name = tensor("op_40053_cast_fp16")]; + tensor var_40060_begin_0 = const()[name = tensor("op_40060_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40060_end_0 = const()[name = tensor("op_40060_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40060_end_mask_0 = const()[name = tensor("op_40060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40060_cast_fp16 = slice_by_index(begin = var_40060_begin_0, end = var_40060_end_0, end_mask = var_40060_end_mask_0, x = var_39608_cast_fp16)[name = tensor("op_40060_cast_fp16")]; + tensor var_40067_begin_0 = const()[name = tensor("op_40067_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40067_end_0 = const()[name = tensor("op_40067_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40067_end_mask_0 = const()[name = tensor("op_40067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40067_cast_fp16 = slice_by_index(begin = var_40067_begin_0, end = var_40067_end_0, end_mask = var_40067_end_mask_0, x = var_39608_cast_fp16)[name = tensor("op_40067_cast_fp16")]; + tensor var_40074_begin_0 = const()[name = tensor("op_40074_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40074_end_0 = const()[name = tensor("op_40074_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40074_end_mask_0 = const()[name = tensor("op_40074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40074_cast_fp16 = slice_by_index(begin = var_40074_begin_0, end = var_40074_end_0, end_mask = var_40074_end_mask_0, x = var_39608_cast_fp16)[name = tensor("op_40074_cast_fp16")]; + tensor var_40081_begin_0 = const()[name = tensor("op_40081_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40081_end_0 = const()[name = tensor("op_40081_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40081_end_mask_0 = const()[name = tensor("op_40081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40081_cast_fp16 = slice_by_index(begin = var_40081_begin_0, end = var_40081_end_0, end_mask = var_40081_end_mask_0, x = var_39612_cast_fp16)[name = tensor("op_40081_cast_fp16")]; + tensor var_40088_begin_0 = const()[name = tensor("op_40088_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40088_end_0 = const()[name = tensor("op_40088_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40088_end_mask_0 = const()[name = tensor("op_40088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40088_cast_fp16 = slice_by_index(begin = var_40088_begin_0, end = var_40088_end_0, end_mask = var_40088_end_mask_0, x = var_39612_cast_fp16)[name = tensor("op_40088_cast_fp16")]; + tensor var_40095_begin_0 = const()[name = tensor("op_40095_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40095_end_0 = const()[name = tensor("op_40095_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40095_end_mask_0 = const()[name = tensor("op_40095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40095_cast_fp16 = slice_by_index(begin = var_40095_begin_0, end = var_40095_end_0, end_mask = var_40095_end_mask_0, x = var_39612_cast_fp16)[name = tensor("op_40095_cast_fp16")]; + tensor var_40102_begin_0 = const()[name = tensor("op_40102_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40102_end_0 = const()[name = tensor("op_40102_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40102_end_mask_0 = const()[name = tensor("op_40102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40102_cast_fp16 = slice_by_index(begin = var_40102_begin_0, end = var_40102_end_0, end_mask = var_40102_end_mask_0, x = var_39612_cast_fp16)[name = tensor("op_40102_cast_fp16")]; + tensor var_40109_begin_0 = const()[name = tensor("op_40109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40109_end_0 = const()[name = tensor("op_40109_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40109_end_mask_0 = const()[name = tensor("op_40109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40109_cast_fp16 = slice_by_index(begin = var_40109_begin_0, end = var_40109_end_0, end_mask = var_40109_end_mask_0, x = var_39616_cast_fp16)[name = tensor("op_40109_cast_fp16")]; + tensor var_40116_begin_0 = const()[name = tensor("op_40116_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40116_end_0 = const()[name = tensor("op_40116_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40116_end_mask_0 = const()[name = tensor("op_40116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40116_cast_fp16 = slice_by_index(begin = var_40116_begin_0, end = var_40116_end_0, end_mask = var_40116_end_mask_0, x = var_39616_cast_fp16)[name = tensor("op_40116_cast_fp16")]; + tensor var_40123_begin_0 = const()[name = tensor("op_40123_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40123_end_0 = const()[name = tensor("op_40123_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40123_end_mask_0 = const()[name = tensor("op_40123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40123_cast_fp16 = slice_by_index(begin = var_40123_begin_0, end = var_40123_end_0, end_mask = var_40123_end_mask_0, x = var_39616_cast_fp16)[name = tensor("op_40123_cast_fp16")]; + tensor var_40130_begin_0 = const()[name = tensor("op_40130_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40130_end_0 = const()[name = tensor("op_40130_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40130_end_mask_0 = const()[name = tensor("op_40130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40130_cast_fp16 = slice_by_index(begin = var_40130_begin_0, end = var_40130_end_0, end_mask = var_40130_end_mask_0, x = var_39616_cast_fp16)[name = tensor("op_40130_cast_fp16")]; + tensor var_40137_begin_0 = const()[name = tensor("op_40137_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40137_end_0 = const()[name = tensor("op_40137_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40137_end_mask_0 = const()[name = tensor("op_40137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40137_cast_fp16 = slice_by_index(begin = var_40137_begin_0, end = var_40137_end_0, end_mask = var_40137_end_mask_0, x = var_39620_cast_fp16)[name = tensor("op_40137_cast_fp16")]; + tensor var_40144_begin_0 = const()[name = tensor("op_40144_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40144_end_0 = const()[name = tensor("op_40144_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40144_end_mask_0 = const()[name = tensor("op_40144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40144_cast_fp16 = slice_by_index(begin = var_40144_begin_0, end = var_40144_end_0, end_mask = var_40144_end_mask_0, x = var_39620_cast_fp16)[name = tensor("op_40144_cast_fp16")]; + tensor var_40151_begin_0 = const()[name = tensor("op_40151_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40151_end_0 = const()[name = tensor("op_40151_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40151_end_mask_0 = const()[name = tensor("op_40151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40151_cast_fp16 = slice_by_index(begin = var_40151_begin_0, end = var_40151_end_0, end_mask = var_40151_end_mask_0, x = var_39620_cast_fp16)[name = tensor("op_40151_cast_fp16")]; + tensor var_40158_begin_0 = const()[name = tensor("op_40158_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40158_end_0 = const()[name = tensor("op_40158_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40158_end_mask_0 = const()[name = tensor("op_40158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40158_cast_fp16 = slice_by_index(begin = var_40158_begin_0, end = var_40158_end_0, end_mask = var_40158_end_mask_0, x = var_39620_cast_fp16)[name = tensor("op_40158_cast_fp16")]; + tensor var_40165_begin_0 = const()[name = tensor("op_40165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40165_end_0 = const()[name = tensor("op_40165_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_40165_end_mask_0 = const()[name = tensor("op_40165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40165_cast_fp16 = slice_by_index(begin = var_40165_begin_0, end = var_40165_end_0, end_mask = var_40165_end_mask_0, x = var_39624_cast_fp16)[name = tensor("op_40165_cast_fp16")]; + tensor var_40172_begin_0 = const()[name = tensor("op_40172_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_40172_end_0 = const()[name = tensor("op_40172_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_40172_end_mask_0 = const()[name = tensor("op_40172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40172_cast_fp16 = slice_by_index(begin = var_40172_begin_0, end = var_40172_end_0, end_mask = var_40172_end_mask_0, x = var_39624_cast_fp16)[name = tensor("op_40172_cast_fp16")]; + tensor var_40179_begin_0 = const()[name = tensor("op_40179_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_40179_end_0 = const()[name = tensor("op_40179_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_40179_end_mask_0 = const()[name = tensor("op_40179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40179_cast_fp16 = slice_by_index(begin = var_40179_begin_0, end = var_40179_end_0, end_mask = var_40179_end_mask_0, x = var_39624_cast_fp16)[name = tensor("op_40179_cast_fp16")]; + tensor var_40186_begin_0 = const()[name = tensor("op_40186_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_40186_end_0 = const()[name = tensor("op_40186_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40186_end_mask_0 = const()[name = tensor("op_40186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40186_cast_fp16 = slice_by_index(begin = var_40186_begin_0, end = var_40186_end_0, end_mask = var_40186_end_mask_0, x = var_39624_cast_fp16)[name = tensor("op_40186_cast_fp16")]; + tensor k_49_perm_0 = const()[name = tensor("k_49_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_40191_begin_0 = const()[name = tensor("op_40191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40191_end_0 = const()[name = tensor("op_40191_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_40191_end_mask_0 = const()[name = tensor("op_40191_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_7 = transpose(perm = k_49_perm_0, x = key_49_cast_fp16)[name = tensor("transpose_7")]; + tensor var_40191_cast_fp16 = slice_by_index(begin = var_40191_begin_0, end = var_40191_end_0, end_mask = var_40191_end_mask_0, x = transpose_7)[name = tensor("op_40191_cast_fp16")]; + tensor var_40195_begin_0 = const()[name = tensor("op_40195_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_40195_end_0 = const()[name = tensor("op_40195_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_40195_end_mask_0 = const()[name = tensor("op_40195_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40195_cast_fp16 = slice_by_index(begin = var_40195_begin_0, end = var_40195_end_0, end_mask = var_40195_end_mask_0, x = transpose_7)[name = tensor("op_40195_cast_fp16")]; + tensor var_40199_begin_0 = const()[name = tensor("op_40199_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_40199_end_0 = const()[name = tensor("op_40199_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_40199_end_mask_0 = const()[name = tensor("op_40199_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40199_cast_fp16 = slice_by_index(begin = var_40199_begin_0, end = var_40199_end_0, end_mask = var_40199_end_mask_0, x = transpose_7)[name = tensor("op_40199_cast_fp16")]; + tensor var_40203_begin_0 = const()[name = tensor("op_40203_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_40203_end_0 = const()[name = tensor("op_40203_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_40203_end_mask_0 = const()[name = tensor("op_40203_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40203_cast_fp16 = slice_by_index(begin = var_40203_begin_0, end = var_40203_end_0, end_mask = var_40203_end_mask_0, x = transpose_7)[name = tensor("op_40203_cast_fp16")]; + tensor var_40207_begin_0 = const()[name = tensor("op_40207_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_40207_end_0 = const()[name = tensor("op_40207_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_40207_end_mask_0 = const()[name = tensor("op_40207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40207_cast_fp16 = slice_by_index(begin = var_40207_begin_0, end = var_40207_end_0, end_mask = var_40207_end_mask_0, x = transpose_7)[name = tensor("op_40207_cast_fp16")]; + tensor var_40211_begin_0 = const()[name = tensor("op_40211_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_40211_end_0 = const()[name = tensor("op_40211_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_40211_end_mask_0 = const()[name = tensor("op_40211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40211_cast_fp16 = slice_by_index(begin = var_40211_begin_0, end = var_40211_end_0, end_mask = var_40211_end_mask_0, x = transpose_7)[name = tensor("op_40211_cast_fp16")]; + tensor var_40215_begin_0 = const()[name = tensor("op_40215_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_40215_end_0 = const()[name = tensor("op_40215_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_40215_end_mask_0 = const()[name = tensor("op_40215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40215_cast_fp16 = slice_by_index(begin = var_40215_begin_0, end = var_40215_end_0, end_mask = var_40215_end_mask_0, x = transpose_7)[name = tensor("op_40215_cast_fp16")]; + tensor var_40219_begin_0 = const()[name = tensor("op_40219_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_40219_end_0 = const()[name = tensor("op_40219_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_40219_end_mask_0 = const()[name = tensor("op_40219_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40219_cast_fp16 = slice_by_index(begin = var_40219_begin_0, end = var_40219_end_0, end_mask = var_40219_end_mask_0, x = transpose_7)[name = tensor("op_40219_cast_fp16")]; + tensor var_40223_begin_0 = const()[name = tensor("op_40223_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_40223_end_0 = const()[name = tensor("op_40223_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_40223_end_mask_0 = const()[name = tensor("op_40223_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40223_cast_fp16 = slice_by_index(begin = var_40223_begin_0, end = var_40223_end_0, end_mask = var_40223_end_mask_0, x = transpose_7)[name = tensor("op_40223_cast_fp16")]; + tensor var_40227_begin_0 = const()[name = tensor("op_40227_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_40227_end_0 = const()[name = tensor("op_40227_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_40227_end_mask_0 = const()[name = tensor("op_40227_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40227_cast_fp16 = slice_by_index(begin = var_40227_begin_0, end = var_40227_end_0, end_mask = var_40227_end_mask_0, x = transpose_7)[name = tensor("op_40227_cast_fp16")]; + tensor var_40231_begin_0 = const()[name = tensor("op_40231_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_40231_end_0 = const()[name = tensor("op_40231_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_40231_end_mask_0 = const()[name = tensor("op_40231_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40231_cast_fp16 = slice_by_index(begin = var_40231_begin_0, end = var_40231_end_0, end_mask = var_40231_end_mask_0, x = transpose_7)[name = tensor("op_40231_cast_fp16")]; + tensor var_40235_begin_0 = const()[name = tensor("op_40235_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_40235_end_0 = const()[name = tensor("op_40235_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_40235_end_mask_0 = const()[name = tensor("op_40235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40235_cast_fp16 = slice_by_index(begin = var_40235_begin_0, end = var_40235_end_0, end_mask = var_40235_end_mask_0, x = transpose_7)[name = tensor("op_40235_cast_fp16")]; + tensor var_40239_begin_0 = const()[name = tensor("op_40239_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_40239_end_0 = const()[name = tensor("op_40239_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_40239_end_mask_0 = const()[name = tensor("op_40239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40239_cast_fp16 = slice_by_index(begin = var_40239_begin_0, end = var_40239_end_0, end_mask = var_40239_end_mask_0, x = transpose_7)[name = tensor("op_40239_cast_fp16")]; + tensor var_40243_begin_0 = const()[name = tensor("op_40243_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_40243_end_0 = const()[name = tensor("op_40243_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_40243_end_mask_0 = const()[name = tensor("op_40243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40243_cast_fp16 = slice_by_index(begin = var_40243_begin_0, end = var_40243_end_0, end_mask = var_40243_end_mask_0, x = transpose_7)[name = tensor("op_40243_cast_fp16")]; + tensor var_40247_begin_0 = const()[name = tensor("op_40247_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_40247_end_0 = const()[name = tensor("op_40247_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_40247_end_mask_0 = const()[name = tensor("op_40247_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40247_cast_fp16 = slice_by_index(begin = var_40247_begin_0, end = var_40247_end_0, end_mask = var_40247_end_mask_0, x = transpose_7)[name = tensor("op_40247_cast_fp16")]; + tensor var_40251_begin_0 = const()[name = tensor("op_40251_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_40251_end_0 = const()[name = tensor("op_40251_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_40251_end_mask_0 = const()[name = tensor("op_40251_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40251_cast_fp16 = slice_by_index(begin = var_40251_begin_0, end = var_40251_end_0, end_mask = var_40251_end_mask_0, x = transpose_7)[name = tensor("op_40251_cast_fp16")]; + tensor var_40255_begin_0 = const()[name = tensor("op_40255_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_40255_end_0 = const()[name = tensor("op_40255_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_40255_end_mask_0 = const()[name = tensor("op_40255_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40255_cast_fp16 = slice_by_index(begin = var_40255_begin_0, end = var_40255_end_0, end_mask = var_40255_end_mask_0, x = transpose_7)[name = tensor("op_40255_cast_fp16")]; + tensor var_40259_begin_0 = const()[name = tensor("op_40259_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_40259_end_0 = const()[name = tensor("op_40259_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_40259_end_mask_0 = const()[name = tensor("op_40259_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40259_cast_fp16 = slice_by_index(begin = var_40259_begin_0, end = var_40259_end_0, end_mask = var_40259_end_mask_0, x = transpose_7)[name = tensor("op_40259_cast_fp16")]; + tensor var_40263_begin_0 = const()[name = tensor("op_40263_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_40263_end_0 = const()[name = tensor("op_40263_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_40263_end_mask_0 = const()[name = tensor("op_40263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40263_cast_fp16 = slice_by_index(begin = var_40263_begin_0, end = var_40263_end_0, end_mask = var_40263_end_mask_0, x = transpose_7)[name = tensor("op_40263_cast_fp16")]; + tensor var_40267_begin_0 = const()[name = tensor("op_40267_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_40267_end_0 = const()[name = tensor("op_40267_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_40267_end_mask_0 = const()[name = tensor("op_40267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_40267_cast_fp16 = slice_by_index(begin = var_40267_begin_0, end = var_40267_end_0, end_mask = var_40267_end_mask_0, x = transpose_7)[name = tensor("op_40267_cast_fp16")]; + tensor var_40269_begin_0 = const()[name = tensor("op_40269_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_40269_end_0 = const()[name = tensor("op_40269_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_40269_end_mask_0 = const()[name = tensor("op_40269_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40269_cast_fp16 = slice_by_index(begin = var_40269_begin_0, end = var_40269_end_0, end_mask = var_40269_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40269_cast_fp16")]; + tensor var_40273_begin_0 = const()[name = tensor("op_40273_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_40273_end_0 = const()[name = tensor("op_40273_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_40273_end_mask_0 = const()[name = tensor("op_40273_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40273_cast_fp16 = slice_by_index(begin = var_40273_begin_0, end = var_40273_end_0, end_mask = var_40273_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40273_cast_fp16")]; + tensor var_40277_begin_0 = const()[name = tensor("op_40277_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_40277_end_0 = const()[name = tensor("op_40277_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_40277_end_mask_0 = const()[name = tensor("op_40277_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40277_cast_fp16 = slice_by_index(begin = var_40277_begin_0, end = var_40277_end_0, end_mask = var_40277_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40277_cast_fp16")]; + tensor var_40281_begin_0 = const()[name = tensor("op_40281_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_40281_end_0 = const()[name = tensor("op_40281_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_40281_end_mask_0 = const()[name = tensor("op_40281_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40281_cast_fp16 = slice_by_index(begin = var_40281_begin_0, end = var_40281_end_0, end_mask = var_40281_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40281_cast_fp16")]; + tensor var_40285_begin_0 = const()[name = tensor("op_40285_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_40285_end_0 = const()[name = tensor("op_40285_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_40285_end_mask_0 = const()[name = tensor("op_40285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40285_cast_fp16 = slice_by_index(begin = var_40285_begin_0, end = var_40285_end_0, end_mask = var_40285_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40285_cast_fp16")]; + tensor var_40289_begin_0 = const()[name = tensor("op_40289_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_40289_end_0 = const()[name = tensor("op_40289_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_40289_end_mask_0 = const()[name = tensor("op_40289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40289_cast_fp16 = slice_by_index(begin = var_40289_begin_0, end = var_40289_end_0, end_mask = var_40289_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40289_cast_fp16")]; + tensor var_40293_begin_0 = const()[name = tensor("op_40293_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_40293_end_0 = const()[name = tensor("op_40293_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_40293_end_mask_0 = const()[name = tensor("op_40293_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40293_cast_fp16 = slice_by_index(begin = var_40293_begin_0, end = var_40293_end_0, end_mask = var_40293_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40293_cast_fp16")]; + tensor var_40297_begin_0 = const()[name = tensor("op_40297_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_40297_end_0 = const()[name = tensor("op_40297_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_40297_end_mask_0 = const()[name = tensor("op_40297_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40297_cast_fp16 = slice_by_index(begin = var_40297_begin_0, end = var_40297_end_0, end_mask = var_40297_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40297_cast_fp16")]; + tensor var_40301_begin_0 = const()[name = tensor("op_40301_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_40301_end_0 = const()[name = tensor("op_40301_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_40301_end_mask_0 = const()[name = tensor("op_40301_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40301_cast_fp16 = slice_by_index(begin = var_40301_begin_0, end = var_40301_end_0, end_mask = var_40301_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40301_cast_fp16")]; + tensor var_40305_begin_0 = const()[name = tensor("op_40305_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_40305_end_0 = const()[name = tensor("op_40305_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_40305_end_mask_0 = const()[name = tensor("op_40305_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40305_cast_fp16 = slice_by_index(begin = var_40305_begin_0, end = var_40305_end_0, end_mask = var_40305_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40305_cast_fp16")]; + tensor var_40309_begin_0 = const()[name = tensor("op_40309_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_40309_end_0 = const()[name = tensor("op_40309_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_40309_end_mask_0 = const()[name = tensor("op_40309_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40309_cast_fp16 = slice_by_index(begin = var_40309_begin_0, end = var_40309_end_0, end_mask = var_40309_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40309_cast_fp16")]; + tensor var_40313_begin_0 = const()[name = tensor("op_40313_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_40313_end_0 = const()[name = tensor("op_40313_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_40313_end_mask_0 = const()[name = tensor("op_40313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40313_cast_fp16 = slice_by_index(begin = var_40313_begin_0, end = var_40313_end_0, end_mask = var_40313_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40313_cast_fp16")]; + tensor var_40317_begin_0 = const()[name = tensor("op_40317_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_40317_end_0 = const()[name = tensor("op_40317_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_40317_end_mask_0 = const()[name = tensor("op_40317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40317_cast_fp16 = slice_by_index(begin = var_40317_begin_0, end = var_40317_end_0, end_mask = var_40317_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40317_cast_fp16")]; + tensor var_40321_begin_0 = const()[name = tensor("op_40321_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_40321_end_0 = const()[name = tensor("op_40321_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_40321_end_mask_0 = const()[name = tensor("op_40321_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40321_cast_fp16 = slice_by_index(begin = var_40321_begin_0, end = var_40321_end_0, end_mask = var_40321_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40321_cast_fp16")]; + tensor var_40325_begin_0 = const()[name = tensor("op_40325_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_40325_end_0 = const()[name = tensor("op_40325_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_40325_end_mask_0 = const()[name = tensor("op_40325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40325_cast_fp16 = slice_by_index(begin = var_40325_begin_0, end = var_40325_end_0, end_mask = var_40325_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40325_cast_fp16")]; + tensor var_40329_begin_0 = const()[name = tensor("op_40329_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_40329_end_0 = const()[name = tensor("op_40329_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_40329_end_mask_0 = const()[name = tensor("op_40329_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40329_cast_fp16 = slice_by_index(begin = var_40329_begin_0, end = var_40329_end_0, end_mask = var_40329_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40329_cast_fp16")]; + tensor var_40333_begin_0 = const()[name = tensor("op_40333_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_40333_end_0 = const()[name = tensor("op_40333_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_40333_end_mask_0 = const()[name = tensor("op_40333_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40333_cast_fp16 = slice_by_index(begin = var_40333_begin_0, end = var_40333_end_0, end_mask = var_40333_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40333_cast_fp16")]; + tensor var_40337_begin_0 = const()[name = tensor("op_40337_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_40337_end_0 = const()[name = tensor("op_40337_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_40337_end_mask_0 = const()[name = tensor("op_40337_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40337_cast_fp16 = slice_by_index(begin = var_40337_begin_0, end = var_40337_end_0, end_mask = var_40337_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40337_cast_fp16")]; + tensor var_40341_begin_0 = const()[name = tensor("op_40341_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_40341_end_0 = const()[name = tensor("op_40341_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_40341_end_mask_0 = const()[name = tensor("op_40341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40341_cast_fp16 = slice_by_index(begin = var_40341_begin_0, end = var_40341_end_0, end_mask = var_40341_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40341_cast_fp16")]; + tensor var_40345_begin_0 = const()[name = tensor("op_40345_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_40345_end_0 = const()[name = tensor("op_40345_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_40345_end_mask_0 = const()[name = tensor("op_40345_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_40345_cast_fp16 = slice_by_index(begin = var_40345_begin_0, end = var_40345_end_0, end_mask = var_40345_end_mask_0, x = value_49_cast_fp16)[name = tensor("op_40345_cast_fp16")]; + tensor var_40349_equation_0 = const()[name = tensor("op_40349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40349_cast_fp16 = einsum(equation = var_40349_equation_0, values = (var_40191_cast_fp16, var_39633_cast_fp16))[name = tensor("op_40349_cast_fp16")]; + tensor var_40350_to_fp16 = const()[name = tensor("op_40350_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3841_cast_fp16 = mul(x = var_40349_cast_fp16, y = var_40350_to_fp16)[name = tensor("aw_chunk_3841_cast_fp16")]; + tensor var_40353_equation_0 = const()[name = tensor("op_40353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40353_cast_fp16 = einsum(equation = var_40353_equation_0, values = (var_40191_cast_fp16, var_39640_cast_fp16))[name = tensor("op_40353_cast_fp16")]; + tensor var_40354_to_fp16 = const()[name = tensor("op_40354_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3843_cast_fp16 = mul(x = var_40353_cast_fp16, y = var_40354_to_fp16)[name = tensor("aw_chunk_3843_cast_fp16")]; + tensor var_40357_equation_0 = const()[name = tensor("op_40357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40357_cast_fp16 = einsum(equation = var_40357_equation_0, values = (var_40191_cast_fp16, var_39647_cast_fp16))[name = tensor("op_40357_cast_fp16")]; + tensor var_40358_to_fp16 = const()[name = tensor("op_40358_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3845_cast_fp16 = mul(x = var_40357_cast_fp16, y = var_40358_to_fp16)[name = tensor("aw_chunk_3845_cast_fp16")]; + tensor var_40361_equation_0 = const()[name = tensor("op_40361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40361_cast_fp16 = einsum(equation = var_40361_equation_0, values = (var_40191_cast_fp16, var_39654_cast_fp16))[name = tensor("op_40361_cast_fp16")]; + tensor var_40362_to_fp16 = const()[name = tensor("op_40362_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3847_cast_fp16 = mul(x = var_40361_cast_fp16, y = var_40362_to_fp16)[name = tensor("aw_chunk_3847_cast_fp16")]; + tensor var_40365_equation_0 = const()[name = tensor("op_40365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40365_cast_fp16 = einsum(equation = var_40365_equation_0, values = (var_40195_cast_fp16, var_39661_cast_fp16))[name = tensor("op_40365_cast_fp16")]; + tensor var_40366_to_fp16 = const()[name = tensor("op_40366_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3849_cast_fp16 = mul(x = var_40365_cast_fp16, y = var_40366_to_fp16)[name = tensor("aw_chunk_3849_cast_fp16")]; + tensor var_40369_equation_0 = const()[name = tensor("op_40369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40369_cast_fp16 = einsum(equation = var_40369_equation_0, values = (var_40195_cast_fp16, var_39668_cast_fp16))[name = tensor("op_40369_cast_fp16")]; + tensor var_40370_to_fp16 = const()[name = tensor("op_40370_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3851_cast_fp16 = mul(x = var_40369_cast_fp16, y = var_40370_to_fp16)[name = tensor("aw_chunk_3851_cast_fp16")]; + tensor var_40373_equation_0 = const()[name = tensor("op_40373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40373_cast_fp16 = einsum(equation = var_40373_equation_0, values = (var_40195_cast_fp16, var_39675_cast_fp16))[name = tensor("op_40373_cast_fp16")]; + tensor var_40374_to_fp16 = const()[name = tensor("op_40374_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3853_cast_fp16 = mul(x = var_40373_cast_fp16, y = var_40374_to_fp16)[name = tensor("aw_chunk_3853_cast_fp16")]; + tensor var_40377_equation_0 = const()[name = tensor("op_40377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40377_cast_fp16 = einsum(equation = var_40377_equation_0, values = (var_40195_cast_fp16, var_39682_cast_fp16))[name = tensor("op_40377_cast_fp16")]; + tensor var_40378_to_fp16 = const()[name = tensor("op_40378_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3855_cast_fp16 = mul(x = var_40377_cast_fp16, y = var_40378_to_fp16)[name = tensor("aw_chunk_3855_cast_fp16")]; + tensor var_40381_equation_0 = const()[name = tensor("op_40381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40381_cast_fp16 = einsum(equation = var_40381_equation_0, values = (var_40199_cast_fp16, var_39689_cast_fp16))[name = tensor("op_40381_cast_fp16")]; + tensor var_40382_to_fp16 = const()[name = tensor("op_40382_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3857_cast_fp16 = mul(x = var_40381_cast_fp16, y = var_40382_to_fp16)[name = tensor("aw_chunk_3857_cast_fp16")]; + tensor var_40385_equation_0 = const()[name = tensor("op_40385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40385_cast_fp16 = einsum(equation = var_40385_equation_0, values = (var_40199_cast_fp16, var_39696_cast_fp16))[name = tensor("op_40385_cast_fp16")]; + tensor var_40386_to_fp16 = const()[name = tensor("op_40386_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3859_cast_fp16 = mul(x = var_40385_cast_fp16, y = var_40386_to_fp16)[name = tensor("aw_chunk_3859_cast_fp16")]; + tensor var_40389_equation_0 = const()[name = tensor("op_40389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40389_cast_fp16 = einsum(equation = var_40389_equation_0, values = (var_40199_cast_fp16, var_39703_cast_fp16))[name = tensor("op_40389_cast_fp16")]; + tensor var_40390_to_fp16 = const()[name = tensor("op_40390_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3861_cast_fp16 = mul(x = var_40389_cast_fp16, y = var_40390_to_fp16)[name = tensor("aw_chunk_3861_cast_fp16")]; + tensor var_40393_equation_0 = const()[name = tensor("op_40393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40393_cast_fp16 = einsum(equation = var_40393_equation_0, values = (var_40199_cast_fp16, var_39710_cast_fp16))[name = tensor("op_40393_cast_fp16")]; + tensor var_40394_to_fp16 = const()[name = tensor("op_40394_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3863_cast_fp16 = mul(x = var_40393_cast_fp16, y = var_40394_to_fp16)[name = tensor("aw_chunk_3863_cast_fp16")]; + tensor var_40397_equation_0 = const()[name = tensor("op_40397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40397_cast_fp16 = einsum(equation = var_40397_equation_0, values = (var_40203_cast_fp16, var_39717_cast_fp16))[name = tensor("op_40397_cast_fp16")]; + tensor var_40398_to_fp16 = const()[name = tensor("op_40398_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3865_cast_fp16 = mul(x = var_40397_cast_fp16, y = var_40398_to_fp16)[name = tensor("aw_chunk_3865_cast_fp16")]; + tensor var_40401_equation_0 = const()[name = tensor("op_40401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40401_cast_fp16 = einsum(equation = var_40401_equation_0, values = (var_40203_cast_fp16, var_39724_cast_fp16))[name = tensor("op_40401_cast_fp16")]; + tensor var_40402_to_fp16 = const()[name = tensor("op_40402_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3867_cast_fp16 = mul(x = var_40401_cast_fp16, y = var_40402_to_fp16)[name = tensor("aw_chunk_3867_cast_fp16")]; + tensor var_40405_equation_0 = const()[name = tensor("op_40405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40405_cast_fp16 = einsum(equation = var_40405_equation_0, values = (var_40203_cast_fp16, var_39731_cast_fp16))[name = tensor("op_40405_cast_fp16")]; + tensor var_40406_to_fp16 = const()[name = tensor("op_40406_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3869_cast_fp16 = mul(x = var_40405_cast_fp16, y = var_40406_to_fp16)[name = tensor("aw_chunk_3869_cast_fp16")]; + tensor var_40409_equation_0 = const()[name = tensor("op_40409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40409_cast_fp16 = einsum(equation = var_40409_equation_0, values = (var_40203_cast_fp16, var_39738_cast_fp16))[name = tensor("op_40409_cast_fp16")]; + tensor var_40410_to_fp16 = const()[name = tensor("op_40410_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3871_cast_fp16 = mul(x = var_40409_cast_fp16, y = var_40410_to_fp16)[name = tensor("aw_chunk_3871_cast_fp16")]; + tensor var_40413_equation_0 = const()[name = tensor("op_40413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40413_cast_fp16 = einsum(equation = var_40413_equation_0, values = (var_40207_cast_fp16, var_39745_cast_fp16))[name = tensor("op_40413_cast_fp16")]; + tensor var_40414_to_fp16 = const()[name = tensor("op_40414_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3873_cast_fp16 = mul(x = var_40413_cast_fp16, y = var_40414_to_fp16)[name = tensor("aw_chunk_3873_cast_fp16")]; + tensor var_40417_equation_0 = const()[name = tensor("op_40417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40417_cast_fp16 = einsum(equation = var_40417_equation_0, values = (var_40207_cast_fp16, var_39752_cast_fp16))[name = tensor("op_40417_cast_fp16")]; + tensor var_40418_to_fp16 = const()[name = tensor("op_40418_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3875_cast_fp16 = mul(x = var_40417_cast_fp16, y = var_40418_to_fp16)[name = tensor("aw_chunk_3875_cast_fp16")]; + tensor var_40421_equation_0 = const()[name = tensor("op_40421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40421_cast_fp16 = einsum(equation = var_40421_equation_0, values = (var_40207_cast_fp16, var_39759_cast_fp16))[name = tensor("op_40421_cast_fp16")]; + tensor var_40422_to_fp16 = const()[name = tensor("op_40422_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3877_cast_fp16 = mul(x = var_40421_cast_fp16, y = var_40422_to_fp16)[name = tensor("aw_chunk_3877_cast_fp16")]; + tensor var_40425_equation_0 = const()[name = tensor("op_40425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40425_cast_fp16 = einsum(equation = var_40425_equation_0, values = (var_40207_cast_fp16, var_39766_cast_fp16))[name = tensor("op_40425_cast_fp16")]; + tensor var_40426_to_fp16 = const()[name = tensor("op_40426_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3879_cast_fp16 = mul(x = var_40425_cast_fp16, y = var_40426_to_fp16)[name = tensor("aw_chunk_3879_cast_fp16")]; + tensor var_40429_equation_0 = const()[name = tensor("op_40429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40429_cast_fp16 = einsum(equation = var_40429_equation_0, values = (var_40211_cast_fp16, var_39773_cast_fp16))[name = tensor("op_40429_cast_fp16")]; + tensor var_40430_to_fp16 = const()[name = tensor("op_40430_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3881_cast_fp16 = mul(x = var_40429_cast_fp16, y = var_40430_to_fp16)[name = tensor("aw_chunk_3881_cast_fp16")]; + tensor var_40433_equation_0 = const()[name = tensor("op_40433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40433_cast_fp16 = einsum(equation = var_40433_equation_0, values = (var_40211_cast_fp16, var_39780_cast_fp16))[name = tensor("op_40433_cast_fp16")]; + tensor var_40434_to_fp16 = const()[name = tensor("op_40434_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3883_cast_fp16 = mul(x = var_40433_cast_fp16, y = var_40434_to_fp16)[name = tensor("aw_chunk_3883_cast_fp16")]; + tensor var_40437_equation_0 = const()[name = tensor("op_40437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40437_cast_fp16 = einsum(equation = var_40437_equation_0, values = (var_40211_cast_fp16, var_39787_cast_fp16))[name = tensor("op_40437_cast_fp16")]; + tensor var_40438_to_fp16 = const()[name = tensor("op_40438_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3885_cast_fp16 = mul(x = var_40437_cast_fp16, y = var_40438_to_fp16)[name = tensor("aw_chunk_3885_cast_fp16")]; + tensor var_40441_equation_0 = const()[name = tensor("op_40441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40441_cast_fp16 = einsum(equation = var_40441_equation_0, values = (var_40211_cast_fp16, var_39794_cast_fp16))[name = tensor("op_40441_cast_fp16")]; + tensor var_40442_to_fp16 = const()[name = tensor("op_40442_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3887_cast_fp16 = mul(x = var_40441_cast_fp16, y = var_40442_to_fp16)[name = tensor("aw_chunk_3887_cast_fp16")]; + tensor var_40445_equation_0 = const()[name = tensor("op_40445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40445_cast_fp16 = einsum(equation = var_40445_equation_0, values = (var_40215_cast_fp16, var_39801_cast_fp16))[name = tensor("op_40445_cast_fp16")]; + tensor var_40446_to_fp16 = const()[name = tensor("op_40446_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3889_cast_fp16 = mul(x = var_40445_cast_fp16, y = var_40446_to_fp16)[name = tensor("aw_chunk_3889_cast_fp16")]; + tensor var_40449_equation_0 = const()[name = tensor("op_40449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40449_cast_fp16 = einsum(equation = var_40449_equation_0, values = (var_40215_cast_fp16, var_39808_cast_fp16))[name = tensor("op_40449_cast_fp16")]; + tensor var_40450_to_fp16 = const()[name = tensor("op_40450_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3891_cast_fp16 = mul(x = var_40449_cast_fp16, y = var_40450_to_fp16)[name = tensor("aw_chunk_3891_cast_fp16")]; + tensor var_40453_equation_0 = const()[name = tensor("op_40453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40453_cast_fp16 = einsum(equation = var_40453_equation_0, values = (var_40215_cast_fp16, var_39815_cast_fp16))[name = tensor("op_40453_cast_fp16")]; + tensor var_40454_to_fp16 = const()[name = tensor("op_40454_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3893_cast_fp16 = mul(x = var_40453_cast_fp16, y = var_40454_to_fp16)[name = tensor("aw_chunk_3893_cast_fp16")]; + tensor var_40457_equation_0 = const()[name = tensor("op_40457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40457_cast_fp16 = einsum(equation = var_40457_equation_0, values = (var_40215_cast_fp16, var_39822_cast_fp16))[name = tensor("op_40457_cast_fp16")]; + tensor var_40458_to_fp16 = const()[name = tensor("op_40458_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3895_cast_fp16 = mul(x = var_40457_cast_fp16, y = var_40458_to_fp16)[name = tensor("aw_chunk_3895_cast_fp16")]; + tensor var_40461_equation_0 = const()[name = tensor("op_40461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40461_cast_fp16 = einsum(equation = var_40461_equation_0, values = (var_40219_cast_fp16, var_39829_cast_fp16))[name = tensor("op_40461_cast_fp16")]; + tensor var_40462_to_fp16 = const()[name = tensor("op_40462_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3897_cast_fp16 = mul(x = var_40461_cast_fp16, y = var_40462_to_fp16)[name = tensor("aw_chunk_3897_cast_fp16")]; + tensor var_40465_equation_0 = const()[name = tensor("op_40465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40465_cast_fp16 = einsum(equation = var_40465_equation_0, values = (var_40219_cast_fp16, var_39836_cast_fp16))[name = tensor("op_40465_cast_fp16")]; + tensor var_40466_to_fp16 = const()[name = tensor("op_40466_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3899_cast_fp16 = mul(x = var_40465_cast_fp16, y = var_40466_to_fp16)[name = tensor("aw_chunk_3899_cast_fp16")]; + tensor var_40469_equation_0 = const()[name = tensor("op_40469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40469_cast_fp16 = einsum(equation = var_40469_equation_0, values = (var_40219_cast_fp16, var_39843_cast_fp16))[name = tensor("op_40469_cast_fp16")]; + tensor var_40470_to_fp16 = const()[name = tensor("op_40470_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3901_cast_fp16 = mul(x = var_40469_cast_fp16, y = var_40470_to_fp16)[name = tensor("aw_chunk_3901_cast_fp16")]; + tensor var_40473_equation_0 = const()[name = tensor("op_40473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40473_cast_fp16 = einsum(equation = var_40473_equation_0, values = (var_40219_cast_fp16, var_39850_cast_fp16))[name = tensor("op_40473_cast_fp16")]; + tensor var_40474_to_fp16 = const()[name = tensor("op_40474_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3903_cast_fp16 = mul(x = var_40473_cast_fp16, y = var_40474_to_fp16)[name = tensor("aw_chunk_3903_cast_fp16")]; + tensor var_40477_equation_0 = const()[name = tensor("op_40477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40477_cast_fp16 = einsum(equation = var_40477_equation_0, values = (var_40223_cast_fp16, var_39857_cast_fp16))[name = tensor("op_40477_cast_fp16")]; + tensor var_40478_to_fp16 = const()[name = tensor("op_40478_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3905_cast_fp16 = mul(x = var_40477_cast_fp16, y = var_40478_to_fp16)[name = tensor("aw_chunk_3905_cast_fp16")]; + tensor var_40481_equation_0 = const()[name = tensor("op_40481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40481_cast_fp16 = einsum(equation = var_40481_equation_0, values = (var_40223_cast_fp16, var_39864_cast_fp16))[name = tensor("op_40481_cast_fp16")]; + tensor var_40482_to_fp16 = const()[name = tensor("op_40482_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3907_cast_fp16 = mul(x = var_40481_cast_fp16, y = var_40482_to_fp16)[name = tensor("aw_chunk_3907_cast_fp16")]; + tensor var_40485_equation_0 = const()[name = tensor("op_40485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40485_cast_fp16 = einsum(equation = var_40485_equation_0, values = (var_40223_cast_fp16, var_39871_cast_fp16))[name = tensor("op_40485_cast_fp16")]; + tensor var_40486_to_fp16 = const()[name = tensor("op_40486_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3909_cast_fp16 = mul(x = var_40485_cast_fp16, y = var_40486_to_fp16)[name = tensor("aw_chunk_3909_cast_fp16")]; + tensor var_40489_equation_0 = const()[name = tensor("op_40489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40489_cast_fp16 = einsum(equation = var_40489_equation_0, values = (var_40223_cast_fp16, var_39878_cast_fp16))[name = tensor("op_40489_cast_fp16")]; + tensor var_40490_to_fp16 = const()[name = tensor("op_40490_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3911_cast_fp16 = mul(x = var_40489_cast_fp16, y = var_40490_to_fp16)[name = tensor("aw_chunk_3911_cast_fp16")]; + tensor var_40493_equation_0 = const()[name = tensor("op_40493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40493_cast_fp16 = einsum(equation = var_40493_equation_0, values = (var_40227_cast_fp16, var_39885_cast_fp16))[name = tensor("op_40493_cast_fp16")]; + tensor var_40494_to_fp16 = const()[name = tensor("op_40494_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3913_cast_fp16 = mul(x = var_40493_cast_fp16, y = var_40494_to_fp16)[name = tensor("aw_chunk_3913_cast_fp16")]; + tensor var_40497_equation_0 = const()[name = tensor("op_40497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40497_cast_fp16 = einsum(equation = var_40497_equation_0, values = (var_40227_cast_fp16, var_39892_cast_fp16))[name = tensor("op_40497_cast_fp16")]; + tensor var_40498_to_fp16 = const()[name = tensor("op_40498_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3915_cast_fp16 = mul(x = var_40497_cast_fp16, y = var_40498_to_fp16)[name = tensor("aw_chunk_3915_cast_fp16")]; + tensor var_40501_equation_0 = const()[name = tensor("op_40501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40501_cast_fp16 = einsum(equation = var_40501_equation_0, values = (var_40227_cast_fp16, var_39899_cast_fp16))[name = tensor("op_40501_cast_fp16")]; + tensor var_40502_to_fp16 = const()[name = tensor("op_40502_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3917_cast_fp16 = mul(x = var_40501_cast_fp16, y = var_40502_to_fp16)[name = tensor("aw_chunk_3917_cast_fp16")]; + tensor var_40505_equation_0 = const()[name = tensor("op_40505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40505_cast_fp16 = einsum(equation = var_40505_equation_0, values = (var_40227_cast_fp16, var_39906_cast_fp16))[name = tensor("op_40505_cast_fp16")]; + tensor var_40506_to_fp16 = const()[name = tensor("op_40506_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3919_cast_fp16 = mul(x = var_40505_cast_fp16, y = var_40506_to_fp16)[name = tensor("aw_chunk_3919_cast_fp16")]; + tensor var_40509_equation_0 = const()[name = tensor("op_40509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40509_cast_fp16 = einsum(equation = var_40509_equation_0, values = (var_40231_cast_fp16, var_39913_cast_fp16))[name = tensor("op_40509_cast_fp16")]; + tensor var_40510_to_fp16 = const()[name = tensor("op_40510_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3921_cast_fp16 = mul(x = var_40509_cast_fp16, y = var_40510_to_fp16)[name = tensor("aw_chunk_3921_cast_fp16")]; + tensor var_40513_equation_0 = const()[name = tensor("op_40513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40513_cast_fp16 = einsum(equation = var_40513_equation_0, values = (var_40231_cast_fp16, var_39920_cast_fp16))[name = tensor("op_40513_cast_fp16")]; + tensor var_40514_to_fp16 = const()[name = tensor("op_40514_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3923_cast_fp16 = mul(x = var_40513_cast_fp16, y = var_40514_to_fp16)[name = tensor("aw_chunk_3923_cast_fp16")]; + tensor var_40517_equation_0 = const()[name = tensor("op_40517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40517_cast_fp16 = einsum(equation = var_40517_equation_0, values = (var_40231_cast_fp16, var_39927_cast_fp16))[name = tensor("op_40517_cast_fp16")]; + tensor var_40518_to_fp16 = const()[name = tensor("op_40518_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3925_cast_fp16 = mul(x = var_40517_cast_fp16, y = var_40518_to_fp16)[name = tensor("aw_chunk_3925_cast_fp16")]; + tensor var_40521_equation_0 = const()[name = tensor("op_40521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40521_cast_fp16 = einsum(equation = var_40521_equation_0, values = (var_40231_cast_fp16, var_39934_cast_fp16))[name = tensor("op_40521_cast_fp16")]; + tensor var_40522_to_fp16 = const()[name = tensor("op_40522_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3927_cast_fp16 = mul(x = var_40521_cast_fp16, y = var_40522_to_fp16)[name = tensor("aw_chunk_3927_cast_fp16")]; + tensor var_40525_equation_0 = const()[name = tensor("op_40525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40525_cast_fp16 = einsum(equation = var_40525_equation_0, values = (var_40235_cast_fp16, var_39941_cast_fp16))[name = tensor("op_40525_cast_fp16")]; + tensor var_40526_to_fp16 = const()[name = tensor("op_40526_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3929_cast_fp16 = mul(x = var_40525_cast_fp16, y = var_40526_to_fp16)[name = tensor("aw_chunk_3929_cast_fp16")]; + tensor var_40529_equation_0 = const()[name = tensor("op_40529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40529_cast_fp16 = einsum(equation = var_40529_equation_0, values = (var_40235_cast_fp16, var_39948_cast_fp16))[name = tensor("op_40529_cast_fp16")]; + tensor var_40530_to_fp16 = const()[name = tensor("op_40530_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3931_cast_fp16 = mul(x = var_40529_cast_fp16, y = var_40530_to_fp16)[name = tensor("aw_chunk_3931_cast_fp16")]; + tensor var_40533_equation_0 = const()[name = tensor("op_40533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40533_cast_fp16 = einsum(equation = var_40533_equation_0, values = (var_40235_cast_fp16, var_39955_cast_fp16))[name = tensor("op_40533_cast_fp16")]; + tensor var_40534_to_fp16 = const()[name = tensor("op_40534_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3933_cast_fp16 = mul(x = var_40533_cast_fp16, y = var_40534_to_fp16)[name = tensor("aw_chunk_3933_cast_fp16")]; + tensor var_40537_equation_0 = const()[name = tensor("op_40537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40537_cast_fp16 = einsum(equation = var_40537_equation_0, values = (var_40235_cast_fp16, var_39962_cast_fp16))[name = tensor("op_40537_cast_fp16")]; + tensor var_40538_to_fp16 = const()[name = tensor("op_40538_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3935_cast_fp16 = mul(x = var_40537_cast_fp16, y = var_40538_to_fp16)[name = tensor("aw_chunk_3935_cast_fp16")]; + tensor var_40541_equation_0 = const()[name = tensor("op_40541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40541_cast_fp16 = einsum(equation = var_40541_equation_0, values = (var_40239_cast_fp16, var_39969_cast_fp16))[name = tensor("op_40541_cast_fp16")]; + tensor var_40542_to_fp16 = const()[name = tensor("op_40542_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3937_cast_fp16 = mul(x = var_40541_cast_fp16, y = var_40542_to_fp16)[name = tensor("aw_chunk_3937_cast_fp16")]; + tensor var_40545_equation_0 = const()[name = tensor("op_40545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40545_cast_fp16 = einsum(equation = var_40545_equation_0, values = (var_40239_cast_fp16, var_39976_cast_fp16))[name = tensor("op_40545_cast_fp16")]; + tensor var_40546_to_fp16 = const()[name = tensor("op_40546_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3939_cast_fp16 = mul(x = var_40545_cast_fp16, y = var_40546_to_fp16)[name = tensor("aw_chunk_3939_cast_fp16")]; + tensor var_40549_equation_0 = const()[name = tensor("op_40549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40549_cast_fp16 = einsum(equation = var_40549_equation_0, values = (var_40239_cast_fp16, var_39983_cast_fp16))[name = tensor("op_40549_cast_fp16")]; + tensor var_40550_to_fp16 = const()[name = tensor("op_40550_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3941_cast_fp16 = mul(x = var_40549_cast_fp16, y = var_40550_to_fp16)[name = tensor("aw_chunk_3941_cast_fp16")]; + tensor var_40553_equation_0 = const()[name = tensor("op_40553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40553_cast_fp16 = einsum(equation = var_40553_equation_0, values = (var_40239_cast_fp16, var_39990_cast_fp16))[name = tensor("op_40553_cast_fp16")]; + tensor var_40554_to_fp16 = const()[name = tensor("op_40554_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3943_cast_fp16 = mul(x = var_40553_cast_fp16, y = var_40554_to_fp16)[name = tensor("aw_chunk_3943_cast_fp16")]; + tensor var_40557_equation_0 = const()[name = tensor("op_40557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40557_cast_fp16 = einsum(equation = var_40557_equation_0, values = (var_40243_cast_fp16, var_39997_cast_fp16))[name = tensor("op_40557_cast_fp16")]; + tensor var_40558_to_fp16 = const()[name = tensor("op_40558_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3945_cast_fp16 = mul(x = var_40557_cast_fp16, y = var_40558_to_fp16)[name = tensor("aw_chunk_3945_cast_fp16")]; + tensor var_40561_equation_0 = const()[name = tensor("op_40561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40561_cast_fp16 = einsum(equation = var_40561_equation_0, values = (var_40243_cast_fp16, var_40004_cast_fp16))[name = tensor("op_40561_cast_fp16")]; + tensor var_40562_to_fp16 = const()[name = tensor("op_40562_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3947_cast_fp16 = mul(x = var_40561_cast_fp16, y = var_40562_to_fp16)[name = tensor("aw_chunk_3947_cast_fp16")]; + tensor var_40565_equation_0 = const()[name = tensor("op_40565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40565_cast_fp16 = einsum(equation = var_40565_equation_0, values = (var_40243_cast_fp16, var_40011_cast_fp16))[name = tensor("op_40565_cast_fp16")]; + tensor var_40566_to_fp16 = const()[name = tensor("op_40566_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3949_cast_fp16 = mul(x = var_40565_cast_fp16, y = var_40566_to_fp16)[name = tensor("aw_chunk_3949_cast_fp16")]; + tensor var_40569_equation_0 = const()[name = tensor("op_40569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40569_cast_fp16 = einsum(equation = var_40569_equation_0, values = (var_40243_cast_fp16, var_40018_cast_fp16))[name = tensor("op_40569_cast_fp16")]; + tensor var_40570_to_fp16 = const()[name = tensor("op_40570_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3951_cast_fp16 = mul(x = var_40569_cast_fp16, y = var_40570_to_fp16)[name = tensor("aw_chunk_3951_cast_fp16")]; + tensor var_40573_equation_0 = const()[name = tensor("op_40573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40573_cast_fp16 = einsum(equation = var_40573_equation_0, values = (var_40247_cast_fp16, var_40025_cast_fp16))[name = tensor("op_40573_cast_fp16")]; + tensor var_40574_to_fp16 = const()[name = tensor("op_40574_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3953_cast_fp16 = mul(x = var_40573_cast_fp16, y = var_40574_to_fp16)[name = tensor("aw_chunk_3953_cast_fp16")]; + tensor var_40577_equation_0 = const()[name = tensor("op_40577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40577_cast_fp16 = einsum(equation = var_40577_equation_0, values = (var_40247_cast_fp16, var_40032_cast_fp16))[name = tensor("op_40577_cast_fp16")]; + tensor var_40578_to_fp16 = const()[name = tensor("op_40578_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3955_cast_fp16 = mul(x = var_40577_cast_fp16, y = var_40578_to_fp16)[name = tensor("aw_chunk_3955_cast_fp16")]; + tensor var_40581_equation_0 = const()[name = tensor("op_40581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40581_cast_fp16 = einsum(equation = var_40581_equation_0, values = (var_40247_cast_fp16, var_40039_cast_fp16))[name = tensor("op_40581_cast_fp16")]; + tensor var_40582_to_fp16 = const()[name = tensor("op_40582_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3957_cast_fp16 = mul(x = var_40581_cast_fp16, y = var_40582_to_fp16)[name = tensor("aw_chunk_3957_cast_fp16")]; + tensor var_40585_equation_0 = const()[name = tensor("op_40585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40585_cast_fp16 = einsum(equation = var_40585_equation_0, values = (var_40247_cast_fp16, var_40046_cast_fp16))[name = tensor("op_40585_cast_fp16")]; + tensor var_40586_to_fp16 = const()[name = tensor("op_40586_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3959_cast_fp16 = mul(x = var_40585_cast_fp16, y = var_40586_to_fp16)[name = tensor("aw_chunk_3959_cast_fp16")]; + tensor var_40589_equation_0 = const()[name = tensor("op_40589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40589_cast_fp16 = einsum(equation = var_40589_equation_0, values = (var_40251_cast_fp16, var_40053_cast_fp16))[name = tensor("op_40589_cast_fp16")]; + tensor var_40590_to_fp16 = const()[name = tensor("op_40590_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3961_cast_fp16 = mul(x = var_40589_cast_fp16, y = var_40590_to_fp16)[name = tensor("aw_chunk_3961_cast_fp16")]; + tensor var_40593_equation_0 = const()[name = tensor("op_40593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40593_cast_fp16 = einsum(equation = var_40593_equation_0, values = (var_40251_cast_fp16, var_40060_cast_fp16))[name = tensor("op_40593_cast_fp16")]; + tensor var_40594_to_fp16 = const()[name = tensor("op_40594_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3963_cast_fp16 = mul(x = var_40593_cast_fp16, y = var_40594_to_fp16)[name = tensor("aw_chunk_3963_cast_fp16")]; + tensor var_40597_equation_0 = const()[name = tensor("op_40597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40597_cast_fp16 = einsum(equation = var_40597_equation_0, values = (var_40251_cast_fp16, var_40067_cast_fp16))[name = tensor("op_40597_cast_fp16")]; + tensor var_40598_to_fp16 = const()[name = tensor("op_40598_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3965_cast_fp16 = mul(x = var_40597_cast_fp16, y = var_40598_to_fp16)[name = tensor("aw_chunk_3965_cast_fp16")]; + tensor var_40601_equation_0 = const()[name = tensor("op_40601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40601_cast_fp16 = einsum(equation = var_40601_equation_0, values = (var_40251_cast_fp16, var_40074_cast_fp16))[name = tensor("op_40601_cast_fp16")]; + tensor var_40602_to_fp16 = const()[name = tensor("op_40602_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3967_cast_fp16 = mul(x = var_40601_cast_fp16, y = var_40602_to_fp16)[name = tensor("aw_chunk_3967_cast_fp16")]; + tensor var_40605_equation_0 = const()[name = tensor("op_40605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40605_cast_fp16 = einsum(equation = var_40605_equation_0, values = (var_40255_cast_fp16, var_40081_cast_fp16))[name = tensor("op_40605_cast_fp16")]; + tensor var_40606_to_fp16 = const()[name = tensor("op_40606_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3969_cast_fp16 = mul(x = var_40605_cast_fp16, y = var_40606_to_fp16)[name = tensor("aw_chunk_3969_cast_fp16")]; + tensor var_40609_equation_0 = const()[name = tensor("op_40609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40609_cast_fp16 = einsum(equation = var_40609_equation_0, values = (var_40255_cast_fp16, var_40088_cast_fp16))[name = tensor("op_40609_cast_fp16")]; + tensor var_40610_to_fp16 = const()[name = tensor("op_40610_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3971_cast_fp16 = mul(x = var_40609_cast_fp16, y = var_40610_to_fp16)[name = tensor("aw_chunk_3971_cast_fp16")]; + tensor var_40613_equation_0 = const()[name = tensor("op_40613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40613_cast_fp16 = einsum(equation = var_40613_equation_0, values = (var_40255_cast_fp16, var_40095_cast_fp16))[name = tensor("op_40613_cast_fp16")]; + tensor var_40614_to_fp16 = const()[name = tensor("op_40614_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3973_cast_fp16 = mul(x = var_40613_cast_fp16, y = var_40614_to_fp16)[name = tensor("aw_chunk_3973_cast_fp16")]; + tensor var_40617_equation_0 = const()[name = tensor("op_40617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40617_cast_fp16 = einsum(equation = var_40617_equation_0, values = (var_40255_cast_fp16, var_40102_cast_fp16))[name = tensor("op_40617_cast_fp16")]; + tensor var_40618_to_fp16 = const()[name = tensor("op_40618_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3975_cast_fp16 = mul(x = var_40617_cast_fp16, y = var_40618_to_fp16)[name = tensor("aw_chunk_3975_cast_fp16")]; + tensor var_40621_equation_0 = const()[name = tensor("op_40621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40621_cast_fp16 = einsum(equation = var_40621_equation_0, values = (var_40259_cast_fp16, var_40109_cast_fp16))[name = tensor("op_40621_cast_fp16")]; + tensor var_40622_to_fp16 = const()[name = tensor("op_40622_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3977_cast_fp16 = mul(x = var_40621_cast_fp16, y = var_40622_to_fp16)[name = tensor("aw_chunk_3977_cast_fp16")]; + tensor var_40625_equation_0 = const()[name = tensor("op_40625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40625_cast_fp16 = einsum(equation = var_40625_equation_0, values = (var_40259_cast_fp16, var_40116_cast_fp16))[name = tensor("op_40625_cast_fp16")]; + tensor var_40626_to_fp16 = const()[name = tensor("op_40626_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3979_cast_fp16 = mul(x = var_40625_cast_fp16, y = var_40626_to_fp16)[name = tensor("aw_chunk_3979_cast_fp16")]; + tensor var_40629_equation_0 = const()[name = tensor("op_40629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40629_cast_fp16 = einsum(equation = var_40629_equation_0, values = (var_40259_cast_fp16, var_40123_cast_fp16))[name = tensor("op_40629_cast_fp16")]; + tensor var_40630_to_fp16 = const()[name = tensor("op_40630_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3981_cast_fp16 = mul(x = var_40629_cast_fp16, y = var_40630_to_fp16)[name = tensor("aw_chunk_3981_cast_fp16")]; + tensor var_40633_equation_0 = const()[name = tensor("op_40633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40633_cast_fp16 = einsum(equation = var_40633_equation_0, values = (var_40259_cast_fp16, var_40130_cast_fp16))[name = tensor("op_40633_cast_fp16")]; + tensor var_40634_to_fp16 = const()[name = tensor("op_40634_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3983_cast_fp16 = mul(x = var_40633_cast_fp16, y = var_40634_to_fp16)[name = tensor("aw_chunk_3983_cast_fp16")]; + tensor var_40637_equation_0 = const()[name = tensor("op_40637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40637_cast_fp16 = einsum(equation = var_40637_equation_0, values = (var_40263_cast_fp16, var_40137_cast_fp16))[name = tensor("op_40637_cast_fp16")]; + tensor var_40638_to_fp16 = const()[name = tensor("op_40638_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3985_cast_fp16 = mul(x = var_40637_cast_fp16, y = var_40638_to_fp16)[name = tensor("aw_chunk_3985_cast_fp16")]; + tensor var_40641_equation_0 = const()[name = tensor("op_40641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40641_cast_fp16 = einsum(equation = var_40641_equation_0, values = (var_40263_cast_fp16, var_40144_cast_fp16))[name = tensor("op_40641_cast_fp16")]; + tensor var_40642_to_fp16 = const()[name = tensor("op_40642_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3987_cast_fp16 = mul(x = var_40641_cast_fp16, y = var_40642_to_fp16)[name = tensor("aw_chunk_3987_cast_fp16")]; + tensor var_40645_equation_0 = const()[name = tensor("op_40645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40645_cast_fp16 = einsum(equation = var_40645_equation_0, values = (var_40263_cast_fp16, var_40151_cast_fp16))[name = tensor("op_40645_cast_fp16")]; + tensor var_40646_to_fp16 = const()[name = tensor("op_40646_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3989_cast_fp16 = mul(x = var_40645_cast_fp16, y = var_40646_to_fp16)[name = tensor("aw_chunk_3989_cast_fp16")]; + tensor var_40649_equation_0 = const()[name = tensor("op_40649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40649_cast_fp16 = einsum(equation = var_40649_equation_0, values = (var_40263_cast_fp16, var_40158_cast_fp16))[name = tensor("op_40649_cast_fp16")]; + tensor var_40650_to_fp16 = const()[name = tensor("op_40650_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3991_cast_fp16 = mul(x = var_40649_cast_fp16, y = var_40650_to_fp16)[name = tensor("aw_chunk_3991_cast_fp16")]; + tensor var_40653_equation_0 = const()[name = tensor("op_40653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40653_cast_fp16 = einsum(equation = var_40653_equation_0, values = (var_40267_cast_fp16, var_40165_cast_fp16))[name = tensor("op_40653_cast_fp16")]; + tensor var_40654_to_fp16 = const()[name = tensor("op_40654_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3993_cast_fp16 = mul(x = var_40653_cast_fp16, y = var_40654_to_fp16)[name = tensor("aw_chunk_3993_cast_fp16")]; + tensor var_40657_equation_0 = const()[name = tensor("op_40657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40657_cast_fp16 = einsum(equation = var_40657_equation_0, values = (var_40267_cast_fp16, var_40172_cast_fp16))[name = tensor("op_40657_cast_fp16")]; + tensor var_40658_to_fp16 = const()[name = tensor("op_40658_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3995_cast_fp16 = mul(x = var_40657_cast_fp16, y = var_40658_to_fp16)[name = tensor("aw_chunk_3995_cast_fp16")]; + tensor var_40661_equation_0 = const()[name = tensor("op_40661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40661_cast_fp16 = einsum(equation = var_40661_equation_0, values = (var_40267_cast_fp16, var_40179_cast_fp16))[name = tensor("op_40661_cast_fp16")]; + tensor var_40662_to_fp16 = const()[name = tensor("op_40662_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3997_cast_fp16 = mul(x = var_40661_cast_fp16, y = var_40662_to_fp16)[name = tensor("aw_chunk_3997_cast_fp16")]; + tensor var_40665_equation_0 = const()[name = tensor("op_40665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_40665_cast_fp16 = einsum(equation = var_40665_equation_0, values = (var_40267_cast_fp16, var_40186_cast_fp16))[name = tensor("op_40665_cast_fp16")]; + tensor var_40666_to_fp16 = const()[name = tensor("op_40666_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_3999_cast_fp16 = mul(x = var_40665_cast_fp16, y = var_40666_to_fp16)[name = tensor("aw_chunk_3999_cast_fp16")]; + tensor var_40668_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3841_cast_fp16)[name = tensor("op_40668_cast_fp16")]; + tensor var_40669_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3843_cast_fp16)[name = tensor("op_40669_cast_fp16")]; + tensor var_40670_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3845_cast_fp16)[name = tensor("op_40670_cast_fp16")]; + tensor var_40671_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3847_cast_fp16)[name = tensor("op_40671_cast_fp16")]; + tensor var_40672_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3849_cast_fp16)[name = tensor("op_40672_cast_fp16")]; + tensor var_40673_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3851_cast_fp16)[name = tensor("op_40673_cast_fp16")]; + tensor var_40674_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3853_cast_fp16)[name = tensor("op_40674_cast_fp16")]; + tensor var_40675_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3855_cast_fp16)[name = tensor("op_40675_cast_fp16")]; + tensor var_40676_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3857_cast_fp16)[name = tensor("op_40676_cast_fp16")]; + tensor var_40677_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3859_cast_fp16)[name = tensor("op_40677_cast_fp16")]; + tensor var_40678_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3861_cast_fp16)[name = tensor("op_40678_cast_fp16")]; + tensor var_40679_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3863_cast_fp16)[name = tensor("op_40679_cast_fp16")]; + tensor var_40680_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3865_cast_fp16)[name = tensor("op_40680_cast_fp16")]; + tensor var_40681_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3867_cast_fp16)[name = tensor("op_40681_cast_fp16")]; + tensor var_40682_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3869_cast_fp16)[name = tensor("op_40682_cast_fp16")]; + tensor var_40683_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3871_cast_fp16)[name = tensor("op_40683_cast_fp16")]; + tensor var_40684_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3873_cast_fp16)[name = tensor("op_40684_cast_fp16")]; + tensor var_40685_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3875_cast_fp16)[name = tensor("op_40685_cast_fp16")]; + tensor var_40686_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3877_cast_fp16)[name = tensor("op_40686_cast_fp16")]; + tensor var_40687_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3879_cast_fp16)[name = tensor("op_40687_cast_fp16")]; + tensor var_40688_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3881_cast_fp16)[name = tensor("op_40688_cast_fp16")]; + tensor var_40689_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3883_cast_fp16)[name = tensor("op_40689_cast_fp16")]; + tensor var_40690_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3885_cast_fp16)[name = tensor("op_40690_cast_fp16")]; + tensor var_40691_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3887_cast_fp16)[name = tensor("op_40691_cast_fp16")]; + tensor var_40692_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3889_cast_fp16)[name = tensor("op_40692_cast_fp16")]; + tensor var_40693_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3891_cast_fp16)[name = tensor("op_40693_cast_fp16")]; + tensor var_40694_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3893_cast_fp16)[name = tensor("op_40694_cast_fp16")]; + tensor var_40695_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3895_cast_fp16)[name = tensor("op_40695_cast_fp16")]; + tensor var_40696_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3897_cast_fp16)[name = tensor("op_40696_cast_fp16")]; + tensor var_40697_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3899_cast_fp16)[name = tensor("op_40697_cast_fp16")]; + tensor var_40698_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3901_cast_fp16)[name = tensor("op_40698_cast_fp16")]; + tensor var_40699_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3903_cast_fp16)[name = tensor("op_40699_cast_fp16")]; + tensor var_40700_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3905_cast_fp16)[name = tensor("op_40700_cast_fp16")]; + tensor var_40701_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3907_cast_fp16)[name = tensor("op_40701_cast_fp16")]; + tensor var_40702_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3909_cast_fp16)[name = tensor("op_40702_cast_fp16")]; + tensor var_40703_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3911_cast_fp16)[name = tensor("op_40703_cast_fp16")]; + tensor var_40704_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3913_cast_fp16)[name = tensor("op_40704_cast_fp16")]; + tensor var_40705_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3915_cast_fp16)[name = tensor("op_40705_cast_fp16")]; + tensor var_40706_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3917_cast_fp16)[name = tensor("op_40706_cast_fp16")]; + tensor var_40707_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3919_cast_fp16)[name = tensor("op_40707_cast_fp16")]; + tensor var_40708_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3921_cast_fp16)[name = tensor("op_40708_cast_fp16")]; + tensor var_40709_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3923_cast_fp16)[name = tensor("op_40709_cast_fp16")]; + tensor var_40710_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3925_cast_fp16)[name = tensor("op_40710_cast_fp16")]; + tensor var_40711_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3927_cast_fp16)[name = tensor("op_40711_cast_fp16")]; + tensor var_40712_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3929_cast_fp16)[name = tensor("op_40712_cast_fp16")]; + tensor var_40713_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3931_cast_fp16)[name = tensor("op_40713_cast_fp16")]; + tensor var_40714_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3933_cast_fp16)[name = tensor("op_40714_cast_fp16")]; + tensor var_40715_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3935_cast_fp16)[name = tensor("op_40715_cast_fp16")]; + tensor var_40716_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3937_cast_fp16)[name = tensor("op_40716_cast_fp16")]; + tensor var_40717_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3939_cast_fp16)[name = tensor("op_40717_cast_fp16")]; + tensor var_40718_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3941_cast_fp16)[name = tensor("op_40718_cast_fp16")]; + tensor var_40719_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3943_cast_fp16)[name = tensor("op_40719_cast_fp16")]; + tensor var_40720_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3945_cast_fp16)[name = tensor("op_40720_cast_fp16")]; + tensor var_40721_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3947_cast_fp16)[name = tensor("op_40721_cast_fp16")]; + tensor var_40722_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3949_cast_fp16)[name = tensor("op_40722_cast_fp16")]; + tensor var_40723_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3951_cast_fp16)[name = tensor("op_40723_cast_fp16")]; + tensor var_40724_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3953_cast_fp16)[name = tensor("op_40724_cast_fp16")]; + tensor var_40725_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3955_cast_fp16)[name = tensor("op_40725_cast_fp16")]; + tensor var_40726_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3957_cast_fp16)[name = tensor("op_40726_cast_fp16")]; + tensor var_40727_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3959_cast_fp16)[name = tensor("op_40727_cast_fp16")]; + tensor var_40728_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3961_cast_fp16)[name = tensor("op_40728_cast_fp16")]; + tensor var_40729_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3963_cast_fp16)[name = tensor("op_40729_cast_fp16")]; + tensor var_40730_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3965_cast_fp16)[name = tensor("op_40730_cast_fp16")]; + tensor var_40731_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3967_cast_fp16)[name = tensor("op_40731_cast_fp16")]; + tensor var_40732_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3969_cast_fp16)[name = tensor("op_40732_cast_fp16")]; + tensor var_40733_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3971_cast_fp16)[name = tensor("op_40733_cast_fp16")]; + tensor var_40734_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3973_cast_fp16)[name = tensor("op_40734_cast_fp16")]; + tensor var_40735_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3975_cast_fp16)[name = tensor("op_40735_cast_fp16")]; + tensor var_40736_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3977_cast_fp16)[name = tensor("op_40736_cast_fp16")]; + tensor var_40737_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3979_cast_fp16)[name = tensor("op_40737_cast_fp16")]; + tensor var_40738_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3981_cast_fp16)[name = tensor("op_40738_cast_fp16")]; + tensor var_40739_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3983_cast_fp16)[name = tensor("op_40739_cast_fp16")]; + tensor var_40740_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3985_cast_fp16)[name = tensor("op_40740_cast_fp16")]; + tensor var_40741_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3987_cast_fp16)[name = tensor("op_40741_cast_fp16")]; + tensor var_40742_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3989_cast_fp16)[name = tensor("op_40742_cast_fp16")]; + tensor var_40743_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3991_cast_fp16)[name = tensor("op_40743_cast_fp16")]; + tensor var_40744_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3993_cast_fp16)[name = tensor("op_40744_cast_fp16")]; + tensor var_40745_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3995_cast_fp16)[name = tensor("op_40745_cast_fp16")]; + tensor var_40746_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3997_cast_fp16)[name = tensor("op_40746_cast_fp16")]; + tensor var_40747_cast_fp16 = softmax(axis = var_39439, x = aw_chunk_3999_cast_fp16)[name = tensor("op_40747_cast_fp16")]; + tensor var_40749_equation_0 = const()[name = tensor("op_40749_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40749_cast_fp16 = einsum(equation = var_40749_equation_0, values = (var_40269_cast_fp16, var_40668_cast_fp16))[name = tensor("op_40749_cast_fp16")]; + tensor var_40751_equation_0 = const()[name = tensor("op_40751_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40751_cast_fp16 = einsum(equation = var_40751_equation_0, values = (var_40269_cast_fp16, var_40669_cast_fp16))[name = tensor("op_40751_cast_fp16")]; + tensor var_40753_equation_0 = const()[name = tensor("op_40753_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40753_cast_fp16 = einsum(equation = var_40753_equation_0, values = (var_40269_cast_fp16, var_40670_cast_fp16))[name = tensor("op_40753_cast_fp16")]; + tensor var_40755_equation_0 = const()[name = tensor("op_40755_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40755_cast_fp16 = einsum(equation = var_40755_equation_0, values = (var_40269_cast_fp16, var_40671_cast_fp16))[name = tensor("op_40755_cast_fp16")]; + tensor var_40757_equation_0 = const()[name = tensor("op_40757_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40757_cast_fp16 = einsum(equation = var_40757_equation_0, values = (var_40273_cast_fp16, var_40672_cast_fp16))[name = tensor("op_40757_cast_fp16")]; + tensor var_40759_equation_0 = const()[name = tensor("op_40759_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40759_cast_fp16 = einsum(equation = var_40759_equation_0, values = (var_40273_cast_fp16, var_40673_cast_fp16))[name = tensor("op_40759_cast_fp16")]; + tensor var_40761_equation_0 = const()[name = tensor("op_40761_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40761_cast_fp16 = einsum(equation = var_40761_equation_0, values = (var_40273_cast_fp16, var_40674_cast_fp16))[name = tensor("op_40761_cast_fp16")]; + tensor var_40763_equation_0 = const()[name = tensor("op_40763_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40763_cast_fp16 = einsum(equation = var_40763_equation_0, values = (var_40273_cast_fp16, var_40675_cast_fp16))[name = tensor("op_40763_cast_fp16")]; + tensor var_40765_equation_0 = const()[name = tensor("op_40765_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40765_cast_fp16 = einsum(equation = var_40765_equation_0, values = (var_40277_cast_fp16, var_40676_cast_fp16))[name = tensor("op_40765_cast_fp16")]; + tensor var_40767_equation_0 = const()[name = tensor("op_40767_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40767_cast_fp16 = einsum(equation = var_40767_equation_0, values = (var_40277_cast_fp16, var_40677_cast_fp16))[name = tensor("op_40767_cast_fp16")]; + tensor var_40769_equation_0 = const()[name = tensor("op_40769_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40769_cast_fp16 = einsum(equation = var_40769_equation_0, values = (var_40277_cast_fp16, var_40678_cast_fp16))[name = tensor("op_40769_cast_fp16")]; + tensor var_40771_equation_0 = const()[name = tensor("op_40771_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40771_cast_fp16 = einsum(equation = var_40771_equation_0, values = (var_40277_cast_fp16, var_40679_cast_fp16))[name = tensor("op_40771_cast_fp16")]; + tensor var_40773_equation_0 = const()[name = tensor("op_40773_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40773_cast_fp16 = einsum(equation = var_40773_equation_0, values = (var_40281_cast_fp16, var_40680_cast_fp16))[name = tensor("op_40773_cast_fp16")]; + tensor var_40775_equation_0 = const()[name = tensor("op_40775_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40775_cast_fp16 = einsum(equation = var_40775_equation_0, values = (var_40281_cast_fp16, var_40681_cast_fp16))[name = tensor("op_40775_cast_fp16")]; + tensor var_40777_equation_0 = const()[name = tensor("op_40777_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40777_cast_fp16 = einsum(equation = var_40777_equation_0, values = (var_40281_cast_fp16, var_40682_cast_fp16))[name = tensor("op_40777_cast_fp16")]; + tensor var_40779_equation_0 = const()[name = tensor("op_40779_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40779_cast_fp16 = einsum(equation = var_40779_equation_0, values = (var_40281_cast_fp16, var_40683_cast_fp16))[name = tensor("op_40779_cast_fp16")]; + tensor var_40781_equation_0 = const()[name = tensor("op_40781_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40781_cast_fp16 = einsum(equation = var_40781_equation_0, values = (var_40285_cast_fp16, var_40684_cast_fp16))[name = tensor("op_40781_cast_fp16")]; + tensor var_40783_equation_0 = const()[name = tensor("op_40783_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40783_cast_fp16 = einsum(equation = var_40783_equation_0, values = (var_40285_cast_fp16, var_40685_cast_fp16))[name = tensor("op_40783_cast_fp16")]; + tensor var_40785_equation_0 = const()[name = tensor("op_40785_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40785_cast_fp16 = einsum(equation = var_40785_equation_0, values = (var_40285_cast_fp16, var_40686_cast_fp16))[name = tensor("op_40785_cast_fp16")]; + tensor var_40787_equation_0 = const()[name = tensor("op_40787_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40787_cast_fp16 = einsum(equation = var_40787_equation_0, values = (var_40285_cast_fp16, var_40687_cast_fp16))[name = tensor("op_40787_cast_fp16")]; + tensor var_40789_equation_0 = const()[name = tensor("op_40789_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40789_cast_fp16 = einsum(equation = var_40789_equation_0, values = (var_40289_cast_fp16, var_40688_cast_fp16))[name = tensor("op_40789_cast_fp16")]; + tensor var_40791_equation_0 = const()[name = tensor("op_40791_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40791_cast_fp16 = einsum(equation = var_40791_equation_0, values = (var_40289_cast_fp16, var_40689_cast_fp16))[name = tensor("op_40791_cast_fp16")]; + tensor var_40793_equation_0 = const()[name = tensor("op_40793_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40793_cast_fp16 = einsum(equation = var_40793_equation_0, values = (var_40289_cast_fp16, var_40690_cast_fp16))[name = tensor("op_40793_cast_fp16")]; + tensor var_40795_equation_0 = const()[name = tensor("op_40795_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40795_cast_fp16 = einsum(equation = var_40795_equation_0, values = (var_40289_cast_fp16, var_40691_cast_fp16))[name = tensor("op_40795_cast_fp16")]; + tensor var_40797_equation_0 = const()[name = tensor("op_40797_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40797_cast_fp16 = einsum(equation = var_40797_equation_0, values = (var_40293_cast_fp16, var_40692_cast_fp16))[name = tensor("op_40797_cast_fp16")]; + tensor var_40799_equation_0 = const()[name = tensor("op_40799_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40799_cast_fp16 = einsum(equation = var_40799_equation_0, values = (var_40293_cast_fp16, var_40693_cast_fp16))[name = tensor("op_40799_cast_fp16")]; + tensor var_40801_equation_0 = const()[name = tensor("op_40801_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40801_cast_fp16 = einsum(equation = var_40801_equation_0, values = (var_40293_cast_fp16, var_40694_cast_fp16))[name = tensor("op_40801_cast_fp16")]; + tensor var_40803_equation_0 = const()[name = tensor("op_40803_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40803_cast_fp16 = einsum(equation = var_40803_equation_0, values = (var_40293_cast_fp16, var_40695_cast_fp16))[name = tensor("op_40803_cast_fp16")]; + tensor var_40805_equation_0 = const()[name = tensor("op_40805_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40805_cast_fp16 = einsum(equation = var_40805_equation_0, values = (var_40297_cast_fp16, var_40696_cast_fp16))[name = tensor("op_40805_cast_fp16")]; + tensor var_40807_equation_0 = const()[name = tensor("op_40807_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40807_cast_fp16 = einsum(equation = var_40807_equation_0, values = (var_40297_cast_fp16, var_40697_cast_fp16))[name = tensor("op_40807_cast_fp16")]; + tensor var_40809_equation_0 = const()[name = tensor("op_40809_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40809_cast_fp16 = einsum(equation = var_40809_equation_0, values = (var_40297_cast_fp16, var_40698_cast_fp16))[name = tensor("op_40809_cast_fp16")]; + tensor var_40811_equation_0 = const()[name = tensor("op_40811_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40811_cast_fp16 = einsum(equation = var_40811_equation_0, values = (var_40297_cast_fp16, var_40699_cast_fp16))[name = tensor("op_40811_cast_fp16")]; + tensor var_40813_equation_0 = const()[name = tensor("op_40813_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40813_cast_fp16 = einsum(equation = var_40813_equation_0, values = (var_40301_cast_fp16, var_40700_cast_fp16))[name = tensor("op_40813_cast_fp16")]; + tensor var_40815_equation_0 = const()[name = tensor("op_40815_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40815_cast_fp16 = einsum(equation = var_40815_equation_0, values = (var_40301_cast_fp16, var_40701_cast_fp16))[name = tensor("op_40815_cast_fp16")]; + tensor var_40817_equation_0 = const()[name = tensor("op_40817_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40817_cast_fp16 = einsum(equation = var_40817_equation_0, values = (var_40301_cast_fp16, var_40702_cast_fp16))[name = tensor("op_40817_cast_fp16")]; + tensor var_40819_equation_0 = const()[name = tensor("op_40819_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40819_cast_fp16 = einsum(equation = var_40819_equation_0, values = (var_40301_cast_fp16, var_40703_cast_fp16))[name = tensor("op_40819_cast_fp16")]; + tensor var_40821_equation_0 = const()[name = tensor("op_40821_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40821_cast_fp16 = einsum(equation = var_40821_equation_0, values = (var_40305_cast_fp16, var_40704_cast_fp16))[name = tensor("op_40821_cast_fp16")]; + tensor var_40823_equation_0 = const()[name = tensor("op_40823_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40823_cast_fp16 = einsum(equation = var_40823_equation_0, values = (var_40305_cast_fp16, var_40705_cast_fp16))[name = tensor("op_40823_cast_fp16")]; + tensor var_40825_equation_0 = const()[name = tensor("op_40825_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40825_cast_fp16 = einsum(equation = var_40825_equation_0, values = (var_40305_cast_fp16, var_40706_cast_fp16))[name = tensor("op_40825_cast_fp16")]; + tensor var_40827_equation_0 = const()[name = tensor("op_40827_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40827_cast_fp16 = einsum(equation = var_40827_equation_0, values = (var_40305_cast_fp16, var_40707_cast_fp16))[name = tensor("op_40827_cast_fp16")]; + tensor var_40829_equation_0 = const()[name = tensor("op_40829_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40829_cast_fp16 = einsum(equation = var_40829_equation_0, values = (var_40309_cast_fp16, var_40708_cast_fp16))[name = tensor("op_40829_cast_fp16")]; + tensor var_40831_equation_0 = const()[name = tensor("op_40831_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40831_cast_fp16 = einsum(equation = var_40831_equation_0, values = (var_40309_cast_fp16, var_40709_cast_fp16))[name = tensor("op_40831_cast_fp16")]; + tensor var_40833_equation_0 = const()[name = tensor("op_40833_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40833_cast_fp16 = einsum(equation = var_40833_equation_0, values = (var_40309_cast_fp16, var_40710_cast_fp16))[name = tensor("op_40833_cast_fp16")]; + tensor var_40835_equation_0 = const()[name = tensor("op_40835_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40835_cast_fp16 = einsum(equation = var_40835_equation_0, values = (var_40309_cast_fp16, var_40711_cast_fp16))[name = tensor("op_40835_cast_fp16")]; + tensor var_40837_equation_0 = const()[name = tensor("op_40837_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40837_cast_fp16 = einsum(equation = var_40837_equation_0, values = (var_40313_cast_fp16, var_40712_cast_fp16))[name = tensor("op_40837_cast_fp16")]; + tensor var_40839_equation_0 = const()[name = tensor("op_40839_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40839_cast_fp16 = einsum(equation = var_40839_equation_0, values = (var_40313_cast_fp16, var_40713_cast_fp16))[name = tensor("op_40839_cast_fp16")]; + tensor var_40841_equation_0 = const()[name = tensor("op_40841_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40841_cast_fp16 = einsum(equation = var_40841_equation_0, values = (var_40313_cast_fp16, var_40714_cast_fp16))[name = tensor("op_40841_cast_fp16")]; + tensor var_40843_equation_0 = const()[name = tensor("op_40843_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40843_cast_fp16 = einsum(equation = var_40843_equation_0, values = (var_40313_cast_fp16, var_40715_cast_fp16))[name = tensor("op_40843_cast_fp16")]; + tensor var_40845_equation_0 = const()[name = tensor("op_40845_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40845_cast_fp16 = einsum(equation = var_40845_equation_0, values = (var_40317_cast_fp16, var_40716_cast_fp16))[name = tensor("op_40845_cast_fp16")]; + tensor var_40847_equation_0 = const()[name = tensor("op_40847_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40847_cast_fp16 = einsum(equation = var_40847_equation_0, values = (var_40317_cast_fp16, var_40717_cast_fp16))[name = tensor("op_40847_cast_fp16")]; + tensor var_40849_equation_0 = const()[name = tensor("op_40849_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40849_cast_fp16 = einsum(equation = var_40849_equation_0, values = (var_40317_cast_fp16, var_40718_cast_fp16))[name = tensor("op_40849_cast_fp16")]; + tensor var_40851_equation_0 = const()[name = tensor("op_40851_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40851_cast_fp16 = einsum(equation = var_40851_equation_0, values = (var_40317_cast_fp16, var_40719_cast_fp16))[name = tensor("op_40851_cast_fp16")]; + tensor var_40853_equation_0 = const()[name = tensor("op_40853_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40853_cast_fp16 = einsum(equation = var_40853_equation_0, values = (var_40321_cast_fp16, var_40720_cast_fp16))[name = tensor("op_40853_cast_fp16")]; + tensor var_40855_equation_0 = const()[name = tensor("op_40855_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40855_cast_fp16 = einsum(equation = var_40855_equation_0, values = (var_40321_cast_fp16, var_40721_cast_fp16))[name = tensor("op_40855_cast_fp16")]; + tensor var_40857_equation_0 = const()[name = tensor("op_40857_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40857_cast_fp16 = einsum(equation = var_40857_equation_0, values = (var_40321_cast_fp16, var_40722_cast_fp16))[name = tensor("op_40857_cast_fp16")]; + tensor var_40859_equation_0 = const()[name = tensor("op_40859_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40859_cast_fp16 = einsum(equation = var_40859_equation_0, values = (var_40321_cast_fp16, var_40723_cast_fp16))[name = tensor("op_40859_cast_fp16")]; + tensor var_40861_equation_0 = const()[name = tensor("op_40861_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40861_cast_fp16 = einsum(equation = var_40861_equation_0, values = (var_40325_cast_fp16, var_40724_cast_fp16))[name = tensor("op_40861_cast_fp16")]; + tensor var_40863_equation_0 = const()[name = tensor("op_40863_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40863_cast_fp16 = einsum(equation = var_40863_equation_0, values = (var_40325_cast_fp16, var_40725_cast_fp16))[name = tensor("op_40863_cast_fp16")]; + tensor var_40865_equation_0 = const()[name = tensor("op_40865_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40865_cast_fp16 = einsum(equation = var_40865_equation_0, values = (var_40325_cast_fp16, var_40726_cast_fp16))[name = tensor("op_40865_cast_fp16")]; + tensor var_40867_equation_0 = const()[name = tensor("op_40867_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40867_cast_fp16 = einsum(equation = var_40867_equation_0, values = (var_40325_cast_fp16, var_40727_cast_fp16))[name = tensor("op_40867_cast_fp16")]; + tensor var_40869_equation_0 = const()[name = tensor("op_40869_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40869_cast_fp16 = einsum(equation = var_40869_equation_0, values = (var_40329_cast_fp16, var_40728_cast_fp16))[name = tensor("op_40869_cast_fp16")]; + tensor var_40871_equation_0 = const()[name = tensor("op_40871_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40871_cast_fp16 = einsum(equation = var_40871_equation_0, values = (var_40329_cast_fp16, var_40729_cast_fp16))[name = tensor("op_40871_cast_fp16")]; + tensor var_40873_equation_0 = const()[name = tensor("op_40873_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40873_cast_fp16 = einsum(equation = var_40873_equation_0, values = (var_40329_cast_fp16, var_40730_cast_fp16))[name = tensor("op_40873_cast_fp16")]; + tensor var_40875_equation_0 = const()[name = tensor("op_40875_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40875_cast_fp16 = einsum(equation = var_40875_equation_0, values = (var_40329_cast_fp16, var_40731_cast_fp16))[name = tensor("op_40875_cast_fp16")]; + tensor var_40877_equation_0 = const()[name = tensor("op_40877_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40877_cast_fp16 = einsum(equation = var_40877_equation_0, values = (var_40333_cast_fp16, var_40732_cast_fp16))[name = tensor("op_40877_cast_fp16")]; + tensor var_40879_equation_0 = const()[name = tensor("op_40879_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40879_cast_fp16 = einsum(equation = var_40879_equation_0, values = (var_40333_cast_fp16, var_40733_cast_fp16))[name = tensor("op_40879_cast_fp16")]; + tensor var_40881_equation_0 = const()[name = tensor("op_40881_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40881_cast_fp16 = einsum(equation = var_40881_equation_0, values = (var_40333_cast_fp16, var_40734_cast_fp16))[name = tensor("op_40881_cast_fp16")]; + tensor var_40883_equation_0 = const()[name = tensor("op_40883_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40883_cast_fp16 = einsum(equation = var_40883_equation_0, values = (var_40333_cast_fp16, var_40735_cast_fp16))[name = tensor("op_40883_cast_fp16")]; + tensor var_40885_equation_0 = const()[name = tensor("op_40885_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40885_cast_fp16 = einsum(equation = var_40885_equation_0, values = (var_40337_cast_fp16, var_40736_cast_fp16))[name = tensor("op_40885_cast_fp16")]; + tensor var_40887_equation_0 = const()[name = tensor("op_40887_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40887_cast_fp16 = einsum(equation = var_40887_equation_0, values = (var_40337_cast_fp16, var_40737_cast_fp16))[name = tensor("op_40887_cast_fp16")]; + tensor var_40889_equation_0 = const()[name = tensor("op_40889_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40889_cast_fp16 = einsum(equation = var_40889_equation_0, values = (var_40337_cast_fp16, var_40738_cast_fp16))[name = tensor("op_40889_cast_fp16")]; + tensor var_40891_equation_0 = const()[name = tensor("op_40891_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40891_cast_fp16 = einsum(equation = var_40891_equation_0, values = (var_40337_cast_fp16, var_40739_cast_fp16))[name = tensor("op_40891_cast_fp16")]; + tensor var_40893_equation_0 = const()[name = tensor("op_40893_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40893_cast_fp16 = einsum(equation = var_40893_equation_0, values = (var_40341_cast_fp16, var_40740_cast_fp16))[name = tensor("op_40893_cast_fp16")]; + tensor var_40895_equation_0 = const()[name = tensor("op_40895_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40895_cast_fp16 = einsum(equation = var_40895_equation_0, values = (var_40341_cast_fp16, var_40741_cast_fp16))[name = tensor("op_40895_cast_fp16")]; + tensor var_40897_equation_0 = const()[name = tensor("op_40897_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40897_cast_fp16 = einsum(equation = var_40897_equation_0, values = (var_40341_cast_fp16, var_40742_cast_fp16))[name = tensor("op_40897_cast_fp16")]; + tensor var_40899_equation_0 = const()[name = tensor("op_40899_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40899_cast_fp16 = einsum(equation = var_40899_equation_0, values = (var_40341_cast_fp16, var_40743_cast_fp16))[name = tensor("op_40899_cast_fp16")]; + tensor var_40901_equation_0 = const()[name = tensor("op_40901_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40901_cast_fp16 = einsum(equation = var_40901_equation_0, values = (var_40345_cast_fp16, var_40744_cast_fp16))[name = tensor("op_40901_cast_fp16")]; + tensor var_40903_equation_0 = const()[name = tensor("op_40903_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40903_cast_fp16 = einsum(equation = var_40903_equation_0, values = (var_40345_cast_fp16, var_40745_cast_fp16))[name = tensor("op_40903_cast_fp16")]; + tensor var_40905_equation_0 = const()[name = tensor("op_40905_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40905_cast_fp16 = einsum(equation = var_40905_equation_0, values = (var_40345_cast_fp16, var_40746_cast_fp16))[name = tensor("op_40905_cast_fp16")]; + tensor var_40907_equation_0 = const()[name = tensor("op_40907_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_40907_cast_fp16 = einsum(equation = var_40907_equation_0, values = (var_40345_cast_fp16, var_40747_cast_fp16))[name = tensor("op_40907_cast_fp16")]; + tensor var_40909_interleave_0 = const()[name = tensor("op_40909_interleave_0"), val = tensor(false)]; + tensor var_40909_cast_fp16 = concat(axis = var_39414, interleave = var_40909_interleave_0, values = (var_40749_cast_fp16, var_40751_cast_fp16, var_40753_cast_fp16, var_40755_cast_fp16))[name = tensor("op_40909_cast_fp16")]; + tensor var_40911_interleave_0 = const()[name = tensor("op_40911_interleave_0"), val = tensor(false)]; + tensor var_40911_cast_fp16 = concat(axis = var_39414, interleave = var_40911_interleave_0, values = (var_40757_cast_fp16, var_40759_cast_fp16, var_40761_cast_fp16, var_40763_cast_fp16))[name = tensor("op_40911_cast_fp16")]; + tensor var_40913_interleave_0 = const()[name = tensor("op_40913_interleave_0"), val = tensor(false)]; + tensor var_40913_cast_fp16 = concat(axis = var_39414, interleave = var_40913_interleave_0, values = (var_40765_cast_fp16, var_40767_cast_fp16, var_40769_cast_fp16, var_40771_cast_fp16))[name = tensor("op_40913_cast_fp16")]; + tensor var_40915_interleave_0 = const()[name = tensor("op_40915_interleave_0"), val = tensor(false)]; + tensor var_40915_cast_fp16 = concat(axis = var_39414, interleave = var_40915_interleave_0, values = (var_40773_cast_fp16, var_40775_cast_fp16, var_40777_cast_fp16, var_40779_cast_fp16))[name = tensor("op_40915_cast_fp16")]; + tensor var_40917_interleave_0 = const()[name = tensor("op_40917_interleave_0"), val = tensor(false)]; + tensor var_40917_cast_fp16 = concat(axis = var_39414, interleave = var_40917_interleave_0, values = (var_40781_cast_fp16, var_40783_cast_fp16, var_40785_cast_fp16, var_40787_cast_fp16))[name = tensor("op_40917_cast_fp16")]; + tensor var_40919_interleave_0 = const()[name = tensor("op_40919_interleave_0"), val = tensor(false)]; + tensor var_40919_cast_fp16 = concat(axis = var_39414, interleave = var_40919_interleave_0, values = (var_40789_cast_fp16, var_40791_cast_fp16, var_40793_cast_fp16, var_40795_cast_fp16))[name = tensor("op_40919_cast_fp16")]; + tensor var_40921_interleave_0 = const()[name = tensor("op_40921_interleave_0"), val = tensor(false)]; + tensor var_40921_cast_fp16 = concat(axis = var_39414, interleave = var_40921_interleave_0, values = (var_40797_cast_fp16, var_40799_cast_fp16, var_40801_cast_fp16, var_40803_cast_fp16))[name = tensor("op_40921_cast_fp16")]; + tensor var_40923_interleave_0 = const()[name = tensor("op_40923_interleave_0"), val = tensor(false)]; + tensor var_40923_cast_fp16 = concat(axis = var_39414, interleave = var_40923_interleave_0, values = (var_40805_cast_fp16, var_40807_cast_fp16, var_40809_cast_fp16, var_40811_cast_fp16))[name = tensor("op_40923_cast_fp16")]; + tensor var_40925_interleave_0 = const()[name = tensor("op_40925_interleave_0"), val = tensor(false)]; + tensor var_40925_cast_fp16 = concat(axis = var_39414, interleave = var_40925_interleave_0, values = (var_40813_cast_fp16, var_40815_cast_fp16, var_40817_cast_fp16, var_40819_cast_fp16))[name = tensor("op_40925_cast_fp16")]; + tensor var_40927_interleave_0 = const()[name = tensor("op_40927_interleave_0"), val = tensor(false)]; + tensor var_40927_cast_fp16 = concat(axis = var_39414, interleave = var_40927_interleave_0, values = (var_40821_cast_fp16, var_40823_cast_fp16, var_40825_cast_fp16, var_40827_cast_fp16))[name = tensor("op_40927_cast_fp16")]; + tensor var_40929_interleave_0 = const()[name = tensor("op_40929_interleave_0"), val = tensor(false)]; + tensor var_40929_cast_fp16 = concat(axis = var_39414, interleave = var_40929_interleave_0, values = (var_40829_cast_fp16, var_40831_cast_fp16, var_40833_cast_fp16, var_40835_cast_fp16))[name = tensor("op_40929_cast_fp16")]; + tensor var_40931_interleave_0 = const()[name = tensor("op_40931_interleave_0"), val = tensor(false)]; + tensor var_40931_cast_fp16 = concat(axis = var_39414, interleave = var_40931_interleave_0, values = (var_40837_cast_fp16, var_40839_cast_fp16, var_40841_cast_fp16, var_40843_cast_fp16))[name = tensor("op_40931_cast_fp16")]; + tensor var_40933_interleave_0 = const()[name = tensor("op_40933_interleave_0"), val = tensor(false)]; + tensor var_40933_cast_fp16 = concat(axis = var_39414, interleave = var_40933_interleave_0, values = (var_40845_cast_fp16, var_40847_cast_fp16, var_40849_cast_fp16, var_40851_cast_fp16))[name = tensor("op_40933_cast_fp16")]; + tensor var_40935_interleave_0 = const()[name = tensor("op_40935_interleave_0"), val = tensor(false)]; + tensor var_40935_cast_fp16 = concat(axis = var_39414, interleave = var_40935_interleave_0, values = (var_40853_cast_fp16, var_40855_cast_fp16, var_40857_cast_fp16, var_40859_cast_fp16))[name = tensor("op_40935_cast_fp16")]; + tensor var_40937_interleave_0 = const()[name = tensor("op_40937_interleave_0"), val = tensor(false)]; + tensor var_40937_cast_fp16 = concat(axis = var_39414, interleave = var_40937_interleave_0, values = (var_40861_cast_fp16, var_40863_cast_fp16, var_40865_cast_fp16, var_40867_cast_fp16))[name = tensor("op_40937_cast_fp16")]; + tensor var_40939_interleave_0 = const()[name = tensor("op_40939_interleave_0"), val = tensor(false)]; + tensor var_40939_cast_fp16 = concat(axis = var_39414, interleave = var_40939_interleave_0, values = (var_40869_cast_fp16, var_40871_cast_fp16, var_40873_cast_fp16, var_40875_cast_fp16))[name = tensor("op_40939_cast_fp16")]; + tensor var_40941_interleave_0 = const()[name = tensor("op_40941_interleave_0"), val = tensor(false)]; + tensor var_40941_cast_fp16 = concat(axis = var_39414, interleave = var_40941_interleave_0, values = (var_40877_cast_fp16, var_40879_cast_fp16, var_40881_cast_fp16, var_40883_cast_fp16))[name = tensor("op_40941_cast_fp16")]; + tensor var_40943_interleave_0 = const()[name = tensor("op_40943_interleave_0"), val = tensor(false)]; + tensor var_40943_cast_fp16 = concat(axis = var_39414, interleave = var_40943_interleave_0, values = (var_40885_cast_fp16, var_40887_cast_fp16, var_40889_cast_fp16, var_40891_cast_fp16))[name = tensor("op_40943_cast_fp16")]; + tensor var_40945_interleave_0 = const()[name = tensor("op_40945_interleave_0"), val = tensor(false)]; + tensor var_40945_cast_fp16 = concat(axis = var_39414, interleave = var_40945_interleave_0, values = (var_40893_cast_fp16, var_40895_cast_fp16, var_40897_cast_fp16, var_40899_cast_fp16))[name = tensor("op_40945_cast_fp16")]; + tensor var_40947_interleave_0 = const()[name = tensor("op_40947_interleave_0"), val = tensor(false)]; + tensor var_40947_cast_fp16 = concat(axis = var_39414, interleave = var_40947_interleave_0, values = (var_40901_cast_fp16, var_40903_cast_fp16, var_40905_cast_fp16, var_40907_cast_fp16))[name = tensor("op_40947_cast_fp16")]; + tensor input_487_interleave_0 = const()[name = tensor("input_487_interleave_0"), val = tensor(false)]; + tensor input_487_cast_fp16 = concat(axis = var_39439, interleave = input_487_interleave_0, values = (var_40909_cast_fp16, var_40911_cast_fp16, var_40913_cast_fp16, var_40915_cast_fp16, var_40917_cast_fp16, var_40919_cast_fp16, var_40921_cast_fp16, var_40923_cast_fp16, var_40925_cast_fp16, var_40927_cast_fp16, var_40929_cast_fp16, var_40931_cast_fp16, var_40933_cast_fp16, var_40935_cast_fp16, var_40937_cast_fp16, var_40939_cast_fp16, var_40941_cast_fp16, var_40943_cast_fp16, var_40945_cast_fp16, var_40947_cast_fp16))[name = tensor("input_487_cast_fp16")]; + tensor var_40955 = const()[name = tensor("op_40955"), val = tensor([1, 1])]; + tensor var_40957 = const()[name = tensor("op_40957"), val = tensor([1, 1])]; + tensor pretrained_out_295_pad_type_0 = const()[name = tensor("pretrained_out_295_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_295_pad_0 = const()[name = tensor("pretrained_out_295_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(271442176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272261440))), name = tensor("layers_24_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_24_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_24_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272261568)))]; + tensor pretrained_out_295_cast_fp16 = conv(bias = layers_24_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_40957, groups = var_39439, pad = pretrained_out_295_pad_0, pad_type = pretrained_out_295_pad_type_0, strides = var_40955, weight = layers_24_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_487_cast_fp16)[name = tensor("pretrained_out_295_cast_fp16")]; + tensor var_40961 = const()[name = tensor("op_40961"), val = tensor([1, 1])]; + tensor var_40963 = const()[name = tensor("op_40963"), val = tensor([1, 1])]; + tensor input_489_pad_type_0 = const()[name = tensor("input_489_pad_type_0"), val = tensor("custom")]; + tensor input_489_pad_0 = const()[name = tensor("input_489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_24_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272264192)))]; + tensor input_489_cast_fp16 = conv(dilations = var_40963, groups = var_39439, pad = input_489_pad_0, pad_type = input_489_pad_type_0, strides = var_40961, weight = layers_24_self_attn_o_proj_loraA_weight_to_fp16, x = input_487_cast_fp16)[name = tensor("input_489_cast_fp16")]; + tensor var_40967 = const()[name = tensor("op_40967"), val = tensor([1, 1])]; + tensor var_40969 = const()[name = tensor("op_40969"), val = tensor([1, 1])]; + tensor lora_out_589_pad_type_0 = const()[name = tensor("lora_out_589_pad_type_0"), val = tensor("custom")]; + tensor lora_out_589_pad_0 = const()[name = tensor("lora_out_589_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_591_weight_0_to_fp16 = const()[name = tensor("lora_out_591_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272305216)))]; + tensor lora_out_591_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_40969, groups = var_39439, pad = lora_out_589_pad_0, pad_type = lora_out_589_pad_type_0, strides = var_40967, weight = lora_out_591_weight_0_to_fp16, x = input_489_cast_fp16)[name = tensor("lora_out_591_cast_fp16")]; + tensor obj_99_cast_fp16 = add(x = pretrained_out_295_cast_fp16, y = lora_out_591_cast_fp16)[name = tensor("obj_99_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = obj_99_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; + tensor var_40978 = const()[name = tensor("op_40978"), val = tensor([1])]; + tensor channels_mean_99_cast_fp16 = reduce_mean(axes = var_40978, keep_dims = var_39440, x = inputs_99_cast_fp16)[name = tensor("channels_mean_99_cast_fp16")]; + tensor zero_mean_99_cast_fp16 = sub(x = inputs_99_cast_fp16, y = channels_mean_99_cast_fp16)[name = tensor("zero_mean_99_cast_fp16")]; + tensor zero_mean_sq_99_cast_fp16 = mul(x = zero_mean_99_cast_fp16, y = zero_mean_99_cast_fp16)[name = tensor("zero_mean_sq_99_cast_fp16")]; + tensor var_40982 = const()[name = tensor("op_40982"), val = tensor([1])]; + tensor var_40983_cast_fp16 = reduce_mean(axes = var_40982, keep_dims = var_39440, x = zero_mean_sq_99_cast_fp16)[name = tensor("op_40983_cast_fp16")]; + tensor var_40984_to_fp16 = const()[name = tensor("op_40984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_40985_cast_fp16 = add(x = var_40983_cast_fp16, y = var_40984_to_fp16)[name = tensor("op_40985_cast_fp16")]; + tensor denom_99_epsilon_0 = const()[name = tensor("denom_99_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_99_cast_fp16 = rsqrt(epsilon = denom_99_epsilon_0, x = var_40985_cast_fp16)[name = tensor("denom_99_cast_fp16")]; + tensor out_99_cast_fp16 = mul(x = zero_mean_99_cast_fp16, y = denom_99_cast_fp16)[name = tensor("out_99_cast_fp16")]; + tensor input_491_gamma_0_to_fp16 = const()[name = tensor("input_491_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272346240)))]; + tensor input_491_beta_0_to_fp16 = const()[name = tensor("input_491_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272348864)))]; + tensor input_491_epsilon_0_to_fp16 = const()[name = tensor("input_491_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_491_cast_fp16 = batch_norm(beta = input_491_beta_0_to_fp16, epsilon = input_491_epsilon_0_to_fp16, gamma = input_491_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_99_cast_fp16)[name = tensor("input_491_cast_fp16")]; + tensor var_40999 = const()[name = tensor("op_40999"), val = tensor([1, 1])]; + tensor var_41001 = const()[name = tensor("op_41001"), val = tensor([1, 1])]; + tensor pretrained_out_297_pad_type_0 = const()[name = tensor("pretrained_out_297_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_297_pad_0 = const()[name = tensor("pretrained_out_297_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272351488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275628352))), name = tensor("layers_24_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_24_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_24_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275628480)))]; + tensor pretrained_out_297_cast_fp16 = conv(bias = layers_24_fc1_pretrained_bias_to_fp16, dilations = var_41001, groups = var_39439, pad = pretrained_out_297_pad_0, pad_type = pretrained_out_297_pad_type_0, strides = var_40999, weight = layers_24_fc1_pretrained_weight_to_fp16_palettized, x = input_491_cast_fp16)[name = tensor("pretrained_out_297_cast_fp16")]; + tensor var_41005 = const()[name = tensor("op_41005"), val = tensor([1, 1])]; + tensor var_41007 = const()[name = tensor("op_41007"), val = tensor([1, 1])]; + tensor input_493_pad_type_0 = const()[name = tensor("input_493_pad_type_0"), val = tensor("custom")]; + tensor input_493_pad_0 = const()[name = tensor("input_493_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_24_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275638784)))]; + tensor input_493_cast_fp16 = conv(dilations = var_41007, groups = var_39439, pad = input_493_pad_0, pad_type = input_493_pad_type_0, strides = var_41005, weight = layers_24_fc1_loraA_weight_to_fp16, x = input_491_cast_fp16)[name = tensor("input_493_cast_fp16")]; + tensor var_41011 = const()[name = tensor("op_41011"), val = tensor([1, 1])]; + tensor var_41013 = const()[name = tensor("op_41013"), val = tensor([1, 1])]; + tensor lora_out_593_pad_type_0 = const()[name = tensor("lora_out_593_pad_type_0"), val = tensor("custom")]; + tensor lora_out_593_pad_0 = const()[name = tensor("lora_out_593_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_595_weight_0_to_fp16 = const()[name = tensor("lora_out_595_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275679808)))]; + tensor lora_out_595_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_41013, groups = var_39439, pad = lora_out_593_pad_0, pad_type = lora_out_593_pad_type_0, strides = var_41011, weight = lora_out_595_weight_0_to_fp16, x = input_493_cast_fp16)[name = tensor("lora_out_595_cast_fp16")]; + tensor input_495_cast_fp16 = add(x = pretrained_out_297_cast_fp16, y = lora_out_595_cast_fp16)[name = tensor("input_495_cast_fp16")]; + tensor input_497_mode_0 = const()[name = tensor("input_497_mode_0"), val = tensor("EXACT")]; + tensor input_497_cast_fp16 = gelu(mode = input_497_mode_0, x = input_495_cast_fp16)[name = tensor("input_497_cast_fp16")]; + tensor var_41025 = const()[name = tensor("op_41025"), val = tensor([1, 1])]; + tensor var_41027 = const()[name = tensor("op_41027"), val = tensor([1, 1])]; + tensor pretrained_out_299_pad_type_0 = const()[name = tensor("pretrained_out_299_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_299_pad_0 = const()[name = tensor("pretrained_out_299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(275843712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279120576))), name = tensor("layers_24_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_24_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_24_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279120704)))]; + tensor pretrained_out_299_cast_fp16 = conv(bias = layers_24_fc2_pretrained_bias_to_fp16, dilations = var_41027, groups = var_39439, pad = pretrained_out_299_pad_0, pad_type = pretrained_out_299_pad_type_0, strides = var_41025, weight = layers_24_fc2_pretrained_weight_to_fp16_palettized, x = input_497_cast_fp16)[name = tensor("pretrained_out_299_cast_fp16")]; + tensor var_41031 = const()[name = tensor("op_41031"), val = tensor([1, 1])]; + tensor var_41033 = const()[name = tensor("op_41033"), val = tensor([1, 1])]; + tensor input_499_pad_type_0 = const()[name = tensor("input_499_pad_type_0"), val = tensor("custom")]; + tensor input_499_pad_0 = const()[name = tensor("input_499_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_24_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_24_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279123328)))]; + tensor input_499_cast_fp16 = conv(dilations = var_41033, groups = var_39439, pad = input_499_pad_0, pad_type = input_499_pad_type_0, strides = var_41031, weight = layers_24_fc2_loraA_weight_to_fp16, x = input_497_cast_fp16)[name = tensor("input_499_cast_fp16")]; + tensor var_41037 = const()[name = tensor("op_41037"), val = tensor([1, 1])]; + tensor var_41039 = const()[name = tensor("op_41039"), val = tensor([1, 1])]; + tensor lora_out_597_pad_type_0 = const()[name = tensor("lora_out_597_pad_type_0"), val = tensor("custom")]; + tensor lora_out_597_pad_0 = const()[name = tensor("lora_out_597_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_599_weight_0_to_fp16 = const()[name = tensor("lora_out_599_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279287232)))]; + tensor lora_out_599_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_41039, groups = var_39439, pad = lora_out_597_pad_0, pad_type = lora_out_597_pad_type_0, strides = var_41037, weight = lora_out_599_weight_0_to_fp16, x = input_499_cast_fp16)[name = tensor("lora_out_599_cast_fp16")]; + tensor hidden_states_53_cast_fp16 = add(x = pretrained_out_299_cast_fp16, y = lora_out_599_cast_fp16)[name = tensor("hidden_states_53_cast_fp16")]; + tensor inputs_101_cast_fp16 = add(x = inputs_99_cast_fp16, y = hidden_states_53_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; + tensor var_41049 = const()[name = tensor("op_41049"), val = tensor(3)]; + tensor var_41074 = const()[name = tensor("op_41074"), val = tensor(1)]; + tensor var_41075 = const()[name = tensor("op_41075"), val = tensor(true)]; + tensor var_41085 = const()[name = tensor("op_41085"), val = tensor([1])]; + tensor channels_mean_101_cast_fp16 = reduce_mean(axes = var_41085, keep_dims = var_41075, x = inputs_101_cast_fp16)[name = tensor("channels_mean_101_cast_fp16")]; + tensor zero_mean_101_cast_fp16 = sub(x = inputs_101_cast_fp16, y = channels_mean_101_cast_fp16)[name = tensor("zero_mean_101_cast_fp16")]; + tensor zero_mean_sq_101_cast_fp16 = mul(x = zero_mean_101_cast_fp16, y = zero_mean_101_cast_fp16)[name = tensor("zero_mean_sq_101_cast_fp16")]; + tensor var_41089 = const()[name = tensor("op_41089"), val = tensor([1])]; + tensor var_41090_cast_fp16 = reduce_mean(axes = var_41089, keep_dims = var_41075, x = zero_mean_sq_101_cast_fp16)[name = tensor("op_41090_cast_fp16")]; + tensor var_41091_to_fp16 = const()[name = tensor("op_41091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_41092_cast_fp16 = add(x = var_41090_cast_fp16, y = var_41091_to_fp16)[name = tensor("op_41092_cast_fp16")]; + tensor denom_101_epsilon_0 = const()[name = tensor("denom_101_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_101_cast_fp16 = rsqrt(epsilon = denom_101_epsilon_0, x = var_41092_cast_fp16)[name = tensor("denom_101_cast_fp16")]; + tensor out_101_cast_fp16 = mul(x = zero_mean_101_cast_fp16, y = denom_101_cast_fp16)[name = tensor("out_101_cast_fp16")]; + tensor obj_101_gamma_0_to_fp16 = const()[name = tensor("obj_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279328256)))]; + tensor obj_101_beta_0_to_fp16 = const()[name = tensor("obj_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279330880)))]; + tensor obj_101_epsilon_0_to_fp16 = const()[name = tensor("obj_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_101_cast_fp16 = batch_norm(beta = obj_101_beta_0_to_fp16, epsilon = obj_101_epsilon_0_to_fp16, gamma = obj_101_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_101_cast_fp16)[name = tensor("obj_101_cast_fp16")]; + tensor var_41110 = const()[name = tensor("op_41110"), val = tensor([1, 1])]; + tensor var_41112 = const()[name = tensor("op_41112"), val = tensor([1, 1])]; + tensor pretrained_out_301_pad_type_0 = const()[name = tensor("pretrained_out_301_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_301_pad_0 = const()[name = tensor("pretrained_out_301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(279333504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280152768))), name = tensor("layers_25_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_25_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280152896)))]; + tensor pretrained_out_301_cast_fp16 = conv(bias = layers_25_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_41112, groups = var_41074, pad = pretrained_out_301_pad_0, pad_type = pretrained_out_301_pad_type_0, strides = var_41110, weight = layers_25_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_101_cast_fp16)[name = tensor("pretrained_out_301_cast_fp16")]; + tensor var_41116 = const()[name = tensor("op_41116"), val = tensor([1, 1])]; + tensor var_41118 = const()[name = tensor("op_41118"), val = tensor([1, 1])]; + tensor input_501_pad_type_0 = const()[name = tensor("input_501_pad_type_0"), val = tensor("custom")]; + tensor input_501_pad_0 = const()[name = tensor("input_501_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280155520)))]; + tensor input_501_cast_fp16 = conv(dilations = var_41118, groups = var_41074, pad = input_501_pad_0, pad_type = input_501_pad_type_0, strides = var_41116, weight = layers_25_self_attn_q_proj_loraA_weight_to_fp16, x = obj_101_cast_fp16)[name = tensor("input_501_cast_fp16")]; + tensor var_41122 = const()[name = tensor("op_41122"), val = tensor([1, 1])]; + tensor var_41124 = const()[name = tensor("op_41124"), val = tensor([1, 1])]; + tensor lora_out_601_pad_type_0 = const()[name = tensor("lora_out_601_pad_type_0"), val = tensor("custom")]; + tensor lora_out_601_pad_0 = const()[name = tensor("lora_out_601_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_603_weight_0_to_fp16 = const()[name = tensor("lora_out_603_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280196544)))]; + tensor lora_out_603_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_41124, groups = var_41074, pad = lora_out_601_pad_0, pad_type = lora_out_601_pad_type_0, strides = var_41122, weight = lora_out_603_weight_0_to_fp16, x = input_501_cast_fp16)[name = tensor("lora_out_603_cast_fp16")]; + tensor query_51_cast_fp16 = add(x = pretrained_out_301_cast_fp16, y = lora_out_603_cast_fp16)[name = tensor("query_51_cast_fp16")]; + tensor var_41134 = const()[name = tensor("op_41134"), val = tensor([1, 1])]; + tensor var_41136 = const()[name = tensor("op_41136"), val = tensor([1, 1])]; + tensor pretrained_out_303_pad_type_0 = const()[name = tensor("pretrained_out_303_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_303_pad_0 = const()[name = tensor("pretrained_out_303_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280237568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281056832))), name = tensor("layers_25_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_303_cast_fp16 = conv(dilations = var_41136, groups = var_41074, pad = pretrained_out_303_pad_0, pad_type = pretrained_out_303_pad_type_0, strides = var_41134, weight = layers_25_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_101_cast_fp16)[name = tensor("pretrained_out_303_cast_fp16")]; + tensor var_41140 = const()[name = tensor("op_41140"), val = tensor([1, 1])]; + tensor var_41142 = const()[name = tensor("op_41142"), val = tensor([1, 1])]; + tensor input_503_pad_type_0 = const()[name = tensor("input_503_pad_type_0"), val = tensor("custom")]; + tensor input_503_pad_0 = const()[name = tensor("input_503_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281056960)))]; + tensor input_503_cast_fp16 = conv(dilations = var_41142, groups = var_41074, pad = input_503_pad_0, pad_type = input_503_pad_type_0, strides = var_41140, weight = layers_25_self_attn_k_proj_loraA_weight_to_fp16, x = obj_101_cast_fp16)[name = tensor("input_503_cast_fp16")]; + tensor var_41146 = const()[name = tensor("op_41146"), val = tensor([1, 1])]; + tensor var_41148 = const()[name = tensor("op_41148"), val = tensor([1, 1])]; + tensor lora_out_605_pad_type_0 = const()[name = tensor("lora_out_605_pad_type_0"), val = tensor("custom")]; + tensor lora_out_605_pad_0 = const()[name = tensor("lora_out_605_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_607_weight_0_to_fp16 = const()[name = tensor("lora_out_607_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281097984)))]; + tensor lora_out_607_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_41148, groups = var_41074, pad = lora_out_605_pad_0, pad_type = lora_out_605_pad_type_0, strides = var_41146, weight = lora_out_607_weight_0_to_fp16, x = input_503_cast_fp16)[name = tensor("lora_out_607_cast_fp16")]; + tensor key_51_cast_fp16 = add(x = pretrained_out_303_cast_fp16, y = lora_out_607_cast_fp16)[name = tensor("key_51_cast_fp16")]; + tensor var_41159 = const()[name = tensor("op_41159"), val = tensor([1, 1])]; + tensor var_41161 = const()[name = tensor("op_41161"), val = tensor([1, 1])]; + tensor pretrained_out_305_pad_type_0 = const()[name = tensor("pretrained_out_305_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_305_pad_0 = const()[name = tensor("pretrained_out_305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281139008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281958272))), name = tensor("layers_25_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_25_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281958400)))]; + tensor pretrained_out_305_cast_fp16 = conv(bias = layers_25_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_41161, groups = var_41074, pad = pretrained_out_305_pad_0, pad_type = pretrained_out_305_pad_type_0, strides = var_41159, weight = layers_25_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_101_cast_fp16)[name = tensor("pretrained_out_305_cast_fp16")]; + tensor var_41165 = const()[name = tensor("op_41165"), val = tensor([1, 1])]; + tensor var_41167 = const()[name = tensor("op_41167"), val = tensor([1, 1])]; + tensor input_505_pad_type_0 = const()[name = tensor("input_505_pad_type_0"), val = tensor("custom")]; + tensor input_505_pad_0 = const()[name = tensor("input_505_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281961024)))]; + tensor input_505_cast_fp16 = conv(dilations = var_41167, groups = var_41074, pad = input_505_pad_0, pad_type = input_505_pad_type_0, strides = var_41165, weight = layers_25_self_attn_v_proj_loraA_weight_to_fp16, x = obj_101_cast_fp16)[name = tensor("input_505_cast_fp16")]; + tensor var_41171 = const()[name = tensor("op_41171"), val = tensor([1, 1])]; + tensor var_41173 = const()[name = tensor("op_41173"), val = tensor([1, 1])]; + tensor lora_out_609_pad_type_0 = const()[name = tensor("lora_out_609_pad_type_0"), val = tensor("custom")]; + tensor lora_out_609_pad_0 = const()[name = tensor("lora_out_609_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_611_weight_0_to_fp16 = const()[name = tensor("lora_out_611_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282002048)))]; + tensor lora_out_611_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_41173, groups = var_41074, pad = lora_out_609_pad_0, pad_type = lora_out_609_pad_type_0, strides = var_41171, weight = lora_out_611_weight_0_to_fp16, x = input_505_cast_fp16)[name = tensor("lora_out_611_cast_fp16")]; + tensor value_51_cast_fp16 = add(x = pretrained_out_305_cast_fp16, y = lora_out_611_cast_fp16)[name = tensor("value_51_cast_fp16")]; + tensor var_41183_begin_0 = const()[name = tensor("op_41183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41183_end_0 = const()[name = tensor("op_41183_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41183_end_mask_0 = const()[name = tensor("op_41183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41183_cast_fp16 = slice_by_index(begin = var_41183_begin_0, end = var_41183_end_0, end_mask = var_41183_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41183_cast_fp16")]; + tensor var_41187_begin_0 = const()[name = tensor("op_41187_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41187_end_0 = const()[name = tensor("op_41187_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_41187_end_mask_0 = const()[name = tensor("op_41187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41187_cast_fp16 = slice_by_index(begin = var_41187_begin_0, end = var_41187_end_0, end_mask = var_41187_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41187_cast_fp16")]; + tensor var_41191_begin_0 = const()[name = tensor("op_41191_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41191_end_0 = const()[name = tensor("op_41191_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_41191_end_mask_0 = const()[name = tensor("op_41191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41191_cast_fp16 = slice_by_index(begin = var_41191_begin_0, end = var_41191_end_0, end_mask = var_41191_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41191_cast_fp16")]; + tensor var_41195_begin_0 = const()[name = tensor("op_41195_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41195_end_0 = const()[name = tensor("op_41195_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_41195_end_mask_0 = const()[name = tensor("op_41195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41195_cast_fp16 = slice_by_index(begin = var_41195_begin_0, end = var_41195_end_0, end_mask = var_41195_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41195_cast_fp16")]; + tensor var_41199_begin_0 = const()[name = tensor("op_41199_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41199_end_0 = const()[name = tensor("op_41199_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_41199_end_mask_0 = const()[name = tensor("op_41199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41199_cast_fp16 = slice_by_index(begin = var_41199_begin_0, end = var_41199_end_0, end_mask = var_41199_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41199_cast_fp16")]; + tensor var_41203_begin_0 = const()[name = tensor("op_41203_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41203_end_0 = const()[name = tensor("op_41203_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_41203_end_mask_0 = const()[name = tensor("op_41203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41203_cast_fp16 = slice_by_index(begin = var_41203_begin_0, end = var_41203_end_0, end_mask = var_41203_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41203_cast_fp16")]; + tensor var_41207_begin_0 = const()[name = tensor("op_41207_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41207_end_0 = const()[name = tensor("op_41207_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_41207_end_mask_0 = const()[name = tensor("op_41207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41207_cast_fp16 = slice_by_index(begin = var_41207_begin_0, end = var_41207_end_0, end_mask = var_41207_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41207_cast_fp16")]; + tensor var_41211_begin_0 = const()[name = tensor("op_41211_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41211_end_0 = const()[name = tensor("op_41211_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_41211_end_mask_0 = const()[name = tensor("op_41211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41211_cast_fp16 = slice_by_index(begin = var_41211_begin_0, end = var_41211_end_0, end_mask = var_41211_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41211_cast_fp16")]; + tensor var_41215_begin_0 = const()[name = tensor("op_41215_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41215_end_0 = const()[name = tensor("op_41215_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_41215_end_mask_0 = const()[name = tensor("op_41215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41215_cast_fp16 = slice_by_index(begin = var_41215_begin_0, end = var_41215_end_0, end_mask = var_41215_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41215_cast_fp16")]; + tensor var_41219_begin_0 = const()[name = tensor("op_41219_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41219_end_0 = const()[name = tensor("op_41219_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_41219_end_mask_0 = const()[name = tensor("op_41219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41219_cast_fp16 = slice_by_index(begin = var_41219_begin_0, end = var_41219_end_0, end_mask = var_41219_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41219_cast_fp16")]; + tensor var_41223_begin_0 = const()[name = tensor("op_41223_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41223_end_0 = const()[name = tensor("op_41223_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_41223_end_mask_0 = const()[name = tensor("op_41223_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41223_cast_fp16 = slice_by_index(begin = var_41223_begin_0, end = var_41223_end_0, end_mask = var_41223_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41223_cast_fp16")]; + tensor var_41227_begin_0 = const()[name = tensor("op_41227_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41227_end_0 = const()[name = tensor("op_41227_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_41227_end_mask_0 = const()[name = tensor("op_41227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41227_cast_fp16 = slice_by_index(begin = var_41227_begin_0, end = var_41227_end_0, end_mask = var_41227_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41227_cast_fp16")]; + tensor var_41231_begin_0 = const()[name = tensor("op_41231_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41231_end_0 = const()[name = tensor("op_41231_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_41231_end_mask_0 = const()[name = tensor("op_41231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41231_cast_fp16 = slice_by_index(begin = var_41231_begin_0, end = var_41231_end_0, end_mask = var_41231_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41231_cast_fp16")]; + tensor var_41235_begin_0 = const()[name = tensor("op_41235_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41235_end_0 = const()[name = tensor("op_41235_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_41235_end_mask_0 = const()[name = tensor("op_41235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41235_cast_fp16 = slice_by_index(begin = var_41235_begin_0, end = var_41235_end_0, end_mask = var_41235_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41235_cast_fp16")]; + tensor var_41239_begin_0 = const()[name = tensor("op_41239_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41239_end_0 = const()[name = tensor("op_41239_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_41239_end_mask_0 = const()[name = tensor("op_41239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41239_cast_fp16 = slice_by_index(begin = var_41239_begin_0, end = var_41239_end_0, end_mask = var_41239_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41239_cast_fp16")]; + tensor var_41243_begin_0 = const()[name = tensor("op_41243_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41243_end_0 = const()[name = tensor("op_41243_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_41243_end_mask_0 = const()[name = tensor("op_41243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41243_cast_fp16 = slice_by_index(begin = var_41243_begin_0, end = var_41243_end_0, end_mask = var_41243_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41243_cast_fp16")]; + tensor var_41247_begin_0 = const()[name = tensor("op_41247_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41247_end_0 = const()[name = tensor("op_41247_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_41247_end_mask_0 = const()[name = tensor("op_41247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41247_cast_fp16 = slice_by_index(begin = var_41247_begin_0, end = var_41247_end_0, end_mask = var_41247_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41247_cast_fp16")]; + tensor var_41251_begin_0 = const()[name = tensor("op_41251_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41251_end_0 = const()[name = tensor("op_41251_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_41251_end_mask_0 = const()[name = tensor("op_41251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41251_cast_fp16 = slice_by_index(begin = var_41251_begin_0, end = var_41251_end_0, end_mask = var_41251_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41251_cast_fp16")]; + tensor var_41255_begin_0 = const()[name = tensor("op_41255_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41255_end_0 = const()[name = tensor("op_41255_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_41255_end_mask_0 = const()[name = tensor("op_41255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41255_cast_fp16 = slice_by_index(begin = var_41255_begin_0, end = var_41255_end_0, end_mask = var_41255_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41255_cast_fp16")]; + tensor var_41259_begin_0 = const()[name = tensor("op_41259_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41259_end_0 = const()[name = tensor("op_41259_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_41259_end_mask_0 = const()[name = tensor("op_41259_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41259_cast_fp16 = slice_by_index(begin = var_41259_begin_0, end = var_41259_end_0, end_mask = var_41259_end_mask_0, x = query_51_cast_fp16)[name = tensor("op_41259_cast_fp16")]; + tensor var_41268_begin_0 = const()[name = tensor("op_41268_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41268_end_0 = const()[name = tensor("op_41268_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41268_end_mask_0 = const()[name = tensor("op_41268_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41268_cast_fp16 = slice_by_index(begin = var_41268_begin_0, end = var_41268_end_0, end_mask = var_41268_end_mask_0, x = var_41183_cast_fp16)[name = tensor("op_41268_cast_fp16")]; + tensor var_41275_begin_0 = const()[name = tensor("op_41275_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41275_end_0 = const()[name = tensor("op_41275_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41275_end_mask_0 = const()[name = tensor("op_41275_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41275_cast_fp16 = slice_by_index(begin = var_41275_begin_0, end = var_41275_end_0, end_mask = var_41275_end_mask_0, x = var_41183_cast_fp16)[name = tensor("op_41275_cast_fp16")]; + tensor var_41282_begin_0 = const()[name = tensor("op_41282_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41282_end_0 = const()[name = tensor("op_41282_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41282_end_mask_0 = const()[name = tensor("op_41282_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41282_cast_fp16 = slice_by_index(begin = var_41282_begin_0, end = var_41282_end_0, end_mask = var_41282_end_mask_0, x = var_41183_cast_fp16)[name = tensor("op_41282_cast_fp16")]; + tensor var_41289_begin_0 = const()[name = tensor("op_41289_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41289_end_0 = const()[name = tensor("op_41289_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41289_end_mask_0 = const()[name = tensor("op_41289_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41289_cast_fp16 = slice_by_index(begin = var_41289_begin_0, end = var_41289_end_0, end_mask = var_41289_end_mask_0, x = var_41183_cast_fp16)[name = tensor("op_41289_cast_fp16")]; + tensor var_41296_begin_0 = const()[name = tensor("op_41296_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41296_end_0 = const()[name = tensor("op_41296_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41296_end_mask_0 = const()[name = tensor("op_41296_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41296_cast_fp16 = slice_by_index(begin = var_41296_begin_0, end = var_41296_end_0, end_mask = var_41296_end_mask_0, x = var_41187_cast_fp16)[name = tensor("op_41296_cast_fp16")]; + tensor var_41303_begin_0 = const()[name = tensor("op_41303_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41303_end_0 = const()[name = tensor("op_41303_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41303_end_mask_0 = const()[name = tensor("op_41303_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41303_cast_fp16 = slice_by_index(begin = var_41303_begin_0, end = var_41303_end_0, end_mask = var_41303_end_mask_0, x = var_41187_cast_fp16)[name = tensor("op_41303_cast_fp16")]; + tensor var_41310_begin_0 = const()[name = tensor("op_41310_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41310_end_0 = const()[name = tensor("op_41310_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41310_end_mask_0 = const()[name = tensor("op_41310_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41310_cast_fp16 = slice_by_index(begin = var_41310_begin_0, end = var_41310_end_0, end_mask = var_41310_end_mask_0, x = var_41187_cast_fp16)[name = tensor("op_41310_cast_fp16")]; + tensor var_41317_begin_0 = const()[name = tensor("op_41317_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41317_end_0 = const()[name = tensor("op_41317_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41317_end_mask_0 = const()[name = tensor("op_41317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41317_cast_fp16 = slice_by_index(begin = var_41317_begin_0, end = var_41317_end_0, end_mask = var_41317_end_mask_0, x = var_41187_cast_fp16)[name = tensor("op_41317_cast_fp16")]; + tensor var_41324_begin_0 = const()[name = tensor("op_41324_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41324_end_0 = const()[name = tensor("op_41324_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41324_end_mask_0 = const()[name = tensor("op_41324_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41324_cast_fp16 = slice_by_index(begin = var_41324_begin_0, end = var_41324_end_0, end_mask = var_41324_end_mask_0, x = var_41191_cast_fp16)[name = tensor("op_41324_cast_fp16")]; + tensor var_41331_begin_0 = const()[name = tensor("op_41331_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41331_end_0 = const()[name = tensor("op_41331_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41331_end_mask_0 = const()[name = tensor("op_41331_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41331_cast_fp16 = slice_by_index(begin = var_41331_begin_0, end = var_41331_end_0, end_mask = var_41331_end_mask_0, x = var_41191_cast_fp16)[name = tensor("op_41331_cast_fp16")]; + tensor var_41338_begin_0 = const()[name = tensor("op_41338_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41338_end_0 = const()[name = tensor("op_41338_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41338_end_mask_0 = const()[name = tensor("op_41338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41338_cast_fp16 = slice_by_index(begin = var_41338_begin_0, end = var_41338_end_0, end_mask = var_41338_end_mask_0, x = var_41191_cast_fp16)[name = tensor("op_41338_cast_fp16")]; + tensor var_41345_begin_0 = const()[name = tensor("op_41345_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41345_end_0 = const()[name = tensor("op_41345_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41345_end_mask_0 = const()[name = tensor("op_41345_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41345_cast_fp16 = slice_by_index(begin = var_41345_begin_0, end = var_41345_end_0, end_mask = var_41345_end_mask_0, x = var_41191_cast_fp16)[name = tensor("op_41345_cast_fp16")]; + tensor var_41352_begin_0 = const()[name = tensor("op_41352_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41352_end_0 = const()[name = tensor("op_41352_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41352_end_mask_0 = const()[name = tensor("op_41352_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41352_cast_fp16 = slice_by_index(begin = var_41352_begin_0, end = var_41352_end_0, end_mask = var_41352_end_mask_0, x = var_41195_cast_fp16)[name = tensor("op_41352_cast_fp16")]; + tensor var_41359_begin_0 = const()[name = tensor("op_41359_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41359_end_0 = const()[name = tensor("op_41359_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41359_end_mask_0 = const()[name = tensor("op_41359_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41359_cast_fp16 = slice_by_index(begin = var_41359_begin_0, end = var_41359_end_0, end_mask = var_41359_end_mask_0, x = var_41195_cast_fp16)[name = tensor("op_41359_cast_fp16")]; + tensor var_41366_begin_0 = const()[name = tensor("op_41366_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41366_end_0 = const()[name = tensor("op_41366_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41366_end_mask_0 = const()[name = tensor("op_41366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41366_cast_fp16 = slice_by_index(begin = var_41366_begin_0, end = var_41366_end_0, end_mask = var_41366_end_mask_0, x = var_41195_cast_fp16)[name = tensor("op_41366_cast_fp16")]; + tensor var_41373_begin_0 = const()[name = tensor("op_41373_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41373_end_0 = const()[name = tensor("op_41373_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41373_end_mask_0 = const()[name = tensor("op_41373_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41373_cast_fp16 = slice_by_index(begin = var_41373_begin_0, end = var_41373_end_0, end_mask = var_41373_end_mask_0, x = var_41195_cast_fp16)[name = tensor("op_41373_cast_fp16")]; + tensor var_41380_begin_0 = const()[name = tensor("op_41380_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41380_end_0 = const()[name = tensor("op_41380_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41380_end_mask_0 = const()[name = tensor("op_41380_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41380_cast_fp16 = slice_by_index(begin = var_41380_begin_0, end = var_41380_end_0, end_mask = var_41380_end_mask_0, x = var_41199_cast_fp16)[name = tensor("op_41380_cast_fp16")]; + tensor var_41387_begin_0 = const()[name = tensor("op_41387_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41387_end_0 = const()[name = tensor("op_41387_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41387_end_mask_0 = const()[name = tensor("op_41387_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41387_cast_fp16 = slice_by_index(begin = var_41387_begin_0, end = var_41387_end_0, end_mask = var_41387_end_mask_0, x = var_41199_cast_fp16)[name = tensor("op_41387_cast_fp16")]; + tensor var_41394_begin_0 = const()[name = tensor("op_41394_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41394_end_0 = const()[name = tensor("op_41394_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41394_end_mask_0 = const()[name = tensor("op_41394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41394_cast_fp16 = slice_by_index(begin = var_41394_begin_0, end = var_41394_end_0, end_mask = var_41394_end_mask_0, x = var_41199_cast_fp16)[name = tensor("op_41394_cast_fp16")]; + tensor var_41401_begin_0 = const()[name = tensor("op_41401_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41401_end_0 = const()[name = tensor("op_41401_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41401_end_mask_0 = const()[name = tensor("op_41401_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41401_cast_fp16 = slice_by_index(begin = var_41401_begin_0, end = var_41401_end_0, end_mask = var_41401_end_mask_0, x = var_41199_cast_fp16)[name = tensor("op_41401_cast_fp16")]; + tensor var_41408_begin_0 = const()[name = tensor("op_41408_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41408_end_0 = const()[name = tensor("op_41408_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41408_end_mask_0 = const()[name = tensor("op_41408_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41408_cast_fp16 = slice_by_index(begin = var_41408_begin_0, end = var_41408_end_0, end_mask = var_41408_end_mask_0, x = var_41203_cast_fp16)[name = tensor("op_41408_cast_fp16")]; + tensor var_41415_begin_0 = const()[name = tensor("op_41415_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41415_end_0 = const()[name = tensor("op_41415_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41415_end_mask_0 = const()[name = tensor("op_41415_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41415_cast_fp16 = slice_by_index(begin = var_41415_begin_0, end = var_41415_end_0, end_mask = var_41415_end_mask_0, x = var_41203_cast_fp16)[name = tensor("op_41415_cast_fp16")]; + tensor var_41422_begin_0 = const()[name = tensor("op_41422_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41422_end_0 = const()[name = tensor("op_41422_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41422_end_mask_0 = const()[name = tensor("op_41422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41422_cast_fp16 = slice_by_index(begin = var_41422_begin_0, end = var_41422_end_0, end_mask = var_41422_end_mask_0, x = var_41203_cast_fp16)[name = tensor("op_41422_cast_fp16")]; + tensor var_41429_begin_0 = const()[name = tensor("op_41429_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41429_end_0 = const()[name = tensor("op_41429_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41429_end_mask_0 = const()[name = tensor("op_41429_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41429_cast_fp16 = slice_by_index(begin = var_41429_begin_0, end = var_41429_end_0, end_mask = var_41429_end_mask_0, x = var_41203_cast_fp16)[name = tensor("op_41429_cast_fp16")]; + tensor var_41436_begin_0 = const()[name = tensor("op_41436_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41436_end_0 = const()[name = tensor("op_41436_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41436_end_mask_0 = const()[name = tensor("op_41436_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41436_cast_fp16 = slice_by_index(begin = var_41436_begin_0, end = var_41436_end_0, end_mask = var_41436_end_mask_0, x = var_41207_cast_fp16)[name = tensor("op_41436_cast_fp16")]; + tensor var_41443_begin_0 = const()[name = tensor("op_41443_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41443_end_0 = const()[name = tensor("op_41443_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41443_end_mask_0 = const()[name = tensor("op_41443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41443_cast_fp16 = slice_by_index(begin = var_41443_begin_0, end = var_41443_end_0, end_mask = var_41443_end_mask_0, x = var_41207_cast_fp16)[name = tensor("op_41443_cast_fp16")]; + tensor var_41450_begin_0 = const()[name = tensor("op_41450_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41450_end_0 = const()[name = tensor("op_41450_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41450_end_mask_0 = const()[name = tensor("op_41450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41450_cast_fp16 = slice_by_index(begin = var_41450_begin_0, end = var_41450_end_0, end_mask = var_41450_end_mask_0, x = var_41207_cast_fp16)[name = tensor("op_41450_cast_fp16")]; + tensor var_41457_begin_0 = const()[name = tensor("op_41457_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41457_end_0 = const()[name = tensor("op_41457_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41457_end_mask_0 = const()[name = tensor("op_41457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41457_cast_fp16 = slice_by_index(begin = var_41457_begin_0, end = var_41457_end_0, end_mask = var_41457_end_mask_0, x = var_41207_cast_fp16)[name = tensor("op_41457_cast_fp16")]; + tensor var_41464_begin_0 = const()[name = tensor("op_41464_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41464_end_0 = const()[name = tensor("op_41464_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41464_end_mask_0 = const()[name = tensor("op_41464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41464_cast_fp16 = slice_by_index(begin = var_41464_begin_0, end = var_41464_end_0, end_mask = var_41464_end_mask_0, x = var_41211_cast_fp16)[name = tensor("op_41464_cast_fp16")]; + tensor var_41471_begin_0 = const()[name = tensor("op_41471_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41471_end_0 = const()[name = tensor("op_41471_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41471_end_mask_0 = const()[name = tensor("op_41471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41471_cast_fp16 = slice_by_index(begin = var_41471_begin_0, end = var_41471_end_0, end_mask = var_41471_end_mask_0, x = var_41211_cast_fp16)[name = tensor("op_41471_cast_fp16")]; + tensor var_41478_begin_0 = const()[name = tensor("op_41478_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41478_end_0 = const()[name = tensor("op_41478_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41478_end_mask_0 = const()[name = tensor("op_41478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41478_cast_fp16 = slice_by_index(begin = var_41478_begin_0, end = var_41478_end_0, end_mask = var_41478_end_mask_0, x = var_41211_cast_fp16)[name = tensor("op_41478_cast_fp16")]; + tensor var_41485_begin_0 = const()[name = tensor("op_41485_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41485_end_0 = const()[name = tensor("op_41485_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41485_end_mask_0 = const()[name = tensor("op_41485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41485_cast_fp16 = slice_by_index(begin = var_41485_begin_0, end = var_41485_end_0, end_mask = var_41485_end_mask_0, x = var_41211_cast_fp16)[name = tensor("op_41485_cast_fp16")]; + tensor var_41492_begin_0 = const()[name = tensor("op_41492_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41492_end_0 = const()[name = tensor("op_41492_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41492_end_mask_0 = const()[name = tensor("op_41492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41492_cast_fp16 = slice_by_index(begin = var_41492_begin_0, end = var_41492_end_0, end_mask = var_41492_end_mask_0, x = var_41215_cast_fp16)[name = tensor("op_41492_cast_fp16")]; + tensor var_41499_begin_0 = const()[name = tensor("op_41499_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41499_end_0 = const()[name = tensor("op_41499_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41499_end_mask_0 = const()[name = tensor("op_41499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41499_cast_fp16 = slice_by_index(begin = var_41499_begin_0, end = var_41499_end_0, end_mask = var_41499_end_mask_0, x = var_41215_cast_fp16)[name = tensor("op_41499_cast_fp16")]; + tensor var_41506_begin_0 = const()[name = tensor("op_41506_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41506_end_0 = const()[name = tensor("op_41506_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41506_end_mask_0 = const()[name = tensor("op_41506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41506_cast_fp16 = slice_by_index(begin = var_41506_begin_0, end = var_41506_end_0, end_mask = var_41506_end_mask_0, x = var_41215_cast_fp16)[name = tensor("op_41506_cast_fp16")]; + tensor var_41513_begin_0 = const()[name = tensor("op_41513_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41513_end_0 = const()[name = tensor("op_41513_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41513_end_mask_0 = const()[name = tensor("op_41513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41513_cast_fp16 = slice_by_index(begin = var_41513_begin_0, end = var_41513_end_0, end_mask = var_41513_end_mask_0, x = var_41215_cast_fp16)[name = tensor("op_41513_cast_fp16")]; + tensor var_41520_begin_0 = const()[name = tensor("op_41520_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41520_end_0 = const()[name = tensor("op_41520_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41520_end_mask_0 = const()[name = tensor("op_41520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41520_cast_fp16 = slice_by_index(begin = var_41520_begin_0, end = var_41520_end_0, end_mask = var_41520_end_mask_0, x = var_41219_cast_fp16)[name = tensor("op_41520_cast_fp16")]; + tensor var_41527_begin_0 = const()[name = tensor("op_41527_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41527_end_0 = const()[name = tensor("op_41527_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41527_end_mask_0 = const()[name = tensor("op_41527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41527_cast_fp16 = slice_by_index(begin = var_41527_begin_0, end = var_41527_end_0, end_mask = var_41527_end_mask_0, x = var_41219_cast_fp16)[name = tensor("op_41527_cast_fp16")]; + tensor var_41534_begin_0 = const()[name = tensor("op_41534_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41534_end_0 = const()[name = tensor("op_41534_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41534_end_mask_0 = const()[name = tensor("op_41534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41534_cast_fp16 = slice_by_index(begin = var_41534_begin_0, end = var_41534_end_0, end_mask = var_41534_end_mask_0, x = var_41219_cast_fp16)[name = tensor("op_41534_cast_fp16")]; + tensor var_41541_begin_0 = const()[name = tensor("op_41541_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41541_end_0 = const()[name = tensor("op_41541_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41541_end_mask_0 = const()[name = tensor("op_41541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41541_cast_fp16 = slice_by_index(begin = var_41541_begin_0, end = var_41541_end_0, end_mask = var_41541_end_mask_0, x = var_41219_cast_fp16)[name = tensor("op_41541_cast_fp16")]; + tensor var_41548_begin_0 = const()[name = tensor("op_41548_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41548_end_0 = const()[name = tensor("op_41548_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41548_end_mask_0 = const()[name = tensor("op_41548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41548_cast_fp16 = slice_by_index(begin = var_41548_begin_0, end = var_41548_end_0, end_mask = var_41548_end_mask_0, x = var_41223_cast_fp16)[name = tensor("op_41548_cast_fp16")]; + tensor var_41555_begin_0 = const()[name = tensor("op_41555_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41555_end_0 = const()[name = tensor("op_41555_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41555_end_mask_0 = const()[name = tensor("op_41555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41555_cast_fp16 = slice_by_index(begin = var_41555_begin_0, end = var_41555_end_0, end_mask = var_41555_end_mask_0, x = var_41223_cast_fp16)[name = tensor("op_41555_cast_fp16")]; + tensor var_41562_begin_0 = const()[name = tensor("op_41562_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41562_end_0 = const()[name = tensor("op_41562_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41562_end_mask_0 = const()[name = tensor("op_41562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41562_cast_fp16 = slice_by_index(begin = var_41562_begin_0, end = var_41562_end_0, end_mask = var_41562_end_mask_0, x = var_41223_cast_fp16)[name = tensor("op_41562_cast_fp16")]; + tensor var_41569_begin_0 = const()[name = tensor("op_41569_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41569_end_0 = const()[name = tensor("op_41569_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41569_end_mask_0 = const()[name = tensor("op_41569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41569_cast_fp16 = slice_by_index(begin = var_41569_begin_0, end = var_41569_end_0, end_mask = var_41569_end_mask_0, x = var_41223_cast_fp16)[name = tensor("op_41569_cast_fp16")]; + tensor var_41576_begin_0 = const()[name = tensor("op_41576_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41576_end_0 = const()[name = tensor("op_41576_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41576_end_mask_0 = const()[name = tensor("op_41576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41576_cast_fp16 = slice_by_index(begin = var_41576_begin_0, end = var_41576_end_0, end_mask = var_41576_end_mask_0, x = var_41227_cast_fp16)[name = tensor("op_41576_cast_fp16")]; + tensor var_41583_begin_0 = const()[name = tensor("op_41583_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41583_end_0 = const()[name = tensor("op_41583_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41583_end_mask_0 = const()[name = tensor("op_41583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41583_cast_fp16 = slice_by_index(begin = var_41583_begin_0, end = var_41583_end_0, end_mask = var_41583_end_mask_0, x = var_41227_cast_fp16)[name = tensor("op_41583_cast_fp16")]; + tensor var_41590_begin_0 = const()[name = tensor("op_41590_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41590_end_0 = const()[name = tensor("op_41590_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41590_end_mask_0 = const()[name = tensor("op_41590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41590_cast_fp16 = slice_by_index(begin = var_41590_begin_0, end = var_41590_end_0, end_mask = var_41590_end_mask_0, x = var_41227_cast_fp16)[name = tensor("op_41590_cast_fp16")]; + tensor var_41597_begin_0 = const()[name = tensor("op_41597_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41597_end_0 = const()[name = tensor("op_41597_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41597_end_mask_0 = const()[name = tensor("op_41597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41597_cast_fp16 = slice_by_index(begin = var_41597_begin_0, end = var_41597_end_0, end_mask = var_41597_end_mask_0, x = var_41227_cast_fp16)[name = tensor("op_41597_cast_fp16")]; + tensor var_41604_begin_0 = const()[name = tensor("op_41604_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41604_end_0 = const()[name = tensor("op_41604_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41604_end_mask_0 = const()[name = tensor("op_41604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41604_cast_fp16 = slice_by_index(begin = var_41604_begin_0, end = var_41604_end_0, end_mask = var_41604_end_mask_0, x = var_41231_cast_fp16)[name = tensor("op_41604_cast_fp16")]; + tensor var_41611_begin_0 = const()[name = tensor("op_41611_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41611_end_0 = const()[name = tensor("op_41611_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41611_end_mask_0 = const()[name = tensor("op_41611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41611_cast_fp16 = slice_by_index(begin = var_41611_begin_0, end = var_41611_end_0, end_mask = var_41611_end_mask_0, x = var_41231_cast_fp16)[name = tensor("op_41611_cast_fp16")]; + tensor var_41618_begin_0 = const()[name = tensor("op_41618_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41618_end_0 = const()[name = tensor("op_41618_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41618_end_mask_0 = const()[name = tensor("op_41618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41618_cast_fp16 = slice_by_index(begin = var_41618_begin_0, end = var_41618_end_0, end_mask = var_41618_end_mask_0, x = var_41231_cast_fp16)[name = tensor("op_41618_cast_fp16")]; + tensor var_41625_begin_0 = const()[name = tensor("op_41625_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41625_end_0 = const()[name = tensor("op_41625_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41625_end_mask_0 = const()[name = tensor("op_41625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41625_cast_fp16 = slice_by_index(begin = var_41625_begin_0, end = var_41625_end_0, end_mask = var_41625_end_mask_0, x = var_41231_cast_fp16)[name = tensor("op_41625_cast_fp16")]; + tensor var_41632_begin_0 = const()[name = tensor("op_41632_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41632_end_0 = const()[name = tensor("op_41632_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41632_end_mask_0 = const()[name = tensor("op_41632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41632_cast_fp16 = slice_by_index(begin = var_41632_begin_0, end = var_41632_end_0, end_mask = var_41632_end_mask_0, x = var_41235_cast_fp16)[name = tensor("op_41632_cast_fp16")]; + tensor var_41639_begin_0 = const()[name = tensor("op_41639_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41639_end_0 = const()[name = tensor("op_41639_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41639_end_mask_0 = const()[name = tensor("op_41639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41639_cast_fp16 = slice_by_index(begin = var_41639_begin_0, end = var_41639_end_0, end_mask = var_41639_end_mask_0, x = var_41235_cast_fp16)[name = tensor("op_41639_cast_fp16")]; + tensor var_41646_begin_0 = const()[name = tensor("op_41646_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41646_end_0 = const()[name = tensor("op_41646_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41646_end_mask_0 = const()[name = tensor("op_41646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41646_cast_fp16 = slice_by_index(begin = var_41646_begin_0, end = var_41646_end_0, end_mask = var_41646_end_mask_0, x = var_41235_cast_fp16)[name = tensor("op_41646_cast_fp16")]; + tensor var_41653_begin_0 = const()[name = tensor("op_41653_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41653_end_0 = const()[name = tensor("op_41653_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41653_end_mask_0 = const()[name = tensor("op_41653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41653_cast_fp16 = slice_by_index(begin = var_41653_begin_0, end = var_41653_end_0, end_mask = var_41653_end_mask_0, x = var_41235_cast_fp16)[name = tensor("op_41653_cast_fp16")]; + tensor var_41660_begin_0 = const()[name = tensor("op_41660_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41660_end_0 = const()[name = tensor("op_41660_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41660_end_mask_0 = const()[name = tensor("op_41660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41660_cast_fp16 = slice_by_index(begin = var_41660_begin_0, end = var_41660_end_0, end_mask = var_41660_end_mask_0, x = var_41239_cast_fp16)[name = tensor("op_41660_cast_fp16")]; + tensor var_41667_begin_0 = const()[name = tensor("op_41667_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41667_end_0 = const()[name = tensor("op_41667_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41667_end_mask_0 = const()[name = tensor("op_41667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41667_cast_fp16 = slice_by_index(begin = var_41667_begin_0, end = var_41667_end_0, end_mask = var_41667_end_mask_0, x = var_41239_cast_fp16)[name = tensor("op_41667_cast_fp16")]; + tensor var_41674_begin_0 = const()[name = tensor("op_41674_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41674_end_0 = const()[name = tensor("op_41674_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41674_end_mask_0 = const()[name = tensor("op_41674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41674_cast_fp16 = slice_by_index(begin = var_41674_begin_0, end = var_41674_end_0, end_mask = var_41674_end_mask_0, x = var_41239_cast_fp16)[name = tensor("op_41674_cast_fp16")]; + tensor var_41681_begin_0 = const()[name = tensor("op_41681_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41681_end_0 = const()[name = tensor("op_41681_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41681_end_mask_0 = const()[name = tensor("op_41681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41681_cast_fp16 = slice_by_index(begin = var_41681_begin_0, end = var_41681_end_0, end_mask = var_41681_end_mask_0, x = var_41239_cast_fp16)[name = tensor("op_41681_cast_fp16")]; + tensor var_41688_begin_0 = const()[name = tensor("op_41688_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41688_end_0 = const()[name = tensor("op_41688_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41688_end_mask_0 = const()[name = tensor("op_41688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41688_cast_fp16 = slice_by_index(begin = var_41688_begin_0, end = var_41688_end_0, end_mask = var_41688_end_mask_0, x = var_41243_cast_fp16)[name = tensor("op_41688_cast_fp16")]; + tensor var_41695_begin_0 = const()[name = tensor("op_41695_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41695_end_0 = const()[name = tensor("op_41695_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41695_end_mask_0 = const()[name = tensor("op_41695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41695_cast_fp16 = slice_by_index(begin = var_41695_begin_0, end = var_41695_end_0, end_mask = var_41695_end_mask_0, x = var_41243_cast_fp16)[name = tensor("op_41695_cast_fp16")]; + tensor var_41702_begin_0 = const()[name = tensor("op_41702_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41702_end_0 = const()[name = tensor("op_41702_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41702_end_mask_0 = const()[name = tensor("op_41702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41702_cast_fp16 = slice_by_index(begin = var_41702_begin_0, end = var_41702_end_0, end_mask = var_41702_end_mask_0, x = var_41243_cast_fp16)[name = tensor("op_41702_cast_fp16")]; + tensor var_41709_begin_0 = const()[name = tensor("op_41709_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41709_end_0 = const()[name = tensor("op_41709_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41709_end_mask_0 = const()[name = tensor("op_41709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41709_cast_fp16 = slice_by_index(begin = var_41709_begin_0, end = var_41709_end_0, end_mask = var_41709_end_mask_0, x = var_41243_cast_fp16)[name = tensor("op_41709_cast_fp16")]; + tensor var_41716_begin_0 = const()[name = tensor("op_41716_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41716_end_0 = const()[name = tensor("op_41716_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41716_end_mask_0 = const()[name = tensor("op_41716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41716_cast_fp16 = slice_by_index(begin = var_41716_begin_0, end = var_41716_end_0, end_mask = var_41716_end_mask_0, x = var_41247_cast_fp16)[name = tensor("op_41716_cast_fp16")]; + tensor var_41723_begin_0 = const()[name = tensor("op_41723_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41723_end_0 = const()[name = tensor("op_41723_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41723_end_mask_0 = const()[name = tensor("op_41723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41723_cast_fp16 = slice_by_index(begin = var_41723_begin_0, end = var_41723_end_0, end_mask = var_41723_end_mask_0, x = var_41247_cast_fp16)[name = tensor("op_41723_cast_fp16")]; + tensor var_41730_begin_0 = const()[name = tensor("op_41730_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41730_end_0 = const()[name = tensor("op_41730_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41730_end_mask_0 = const()[name = tensor("op_41730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41730_cast_fp16 = slice_by_index(begin = var_41730_begin_0, end = var_41730_end_0, end_mask = var_41730_end_mask_0, x = var_41247_cast_fp16)[name = tensor("op_41730_cast_fp16")]; + tensor var_41737_begin_0 = const()[name = tensor("op_41737_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41737_end_0 = const()[name = tensor("op_41737_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41737_end_mask_0 = const()[name = tensor("op_41737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41737_cast_fp16 = slice_by_index(begin = var_41737_begin_0, end = var_41737_end_0, end_mask = var_41737_end_mask_0, x = var_41247_cast_fp16)[name = tensor("op_41737_cast_fp16")]; + tensor var_41744_begin_0 = const()[name = tensor("op_41744_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41744_end_0 = const()[name = tensor("op_41744_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41744_end_mask_0 = const()[name = tensor("op_41744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41744_cast_fp16 = slice_by_index(begin = var_41744_begin_0, end = var_41744_end_0, end_mask = var_41744_end_mask_0, x = var_41251_cast_fp16)[name = tensor("op_41744_cast_fp16")]; + tensor var_41751_begin_0 = const()[name = tensor("op_41751_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41751_end_0 = const()[name = tensor("op_41751_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41751_end_mask_0 = const()[name = tensor("op_41751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41751_cast_fp16 = slice_by_index(begin = var_41751_begin_0, end = var_41751_end_0, end_mask = var_41751_end_mask_0, x = var_41251_cast_fp16)[name = tensor("op_41751_cast_fp16")]; + tensor var_41758_begin_0 = const()[name = tensor("op_41758_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41758_end_0 = const()[name = tensor("op_41758_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41758_end_mask_0 = const()[name = tensor("op_41758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41758_cast_fp16 = slice_by_index(begin = var_41758_begin_0, end = var_41758_end_0, end_mask = var_41758_end_mask_0, x = var_41251_cast_fp16)[name = tensor("op_41758_cast_fp16")]; + tensor var_41765_begin_0 = const()[name = tensor("op_41765_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41765_end_0 = const()[name = tensor("op_41765_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41765_end_mask_0 = const()[name = tensor("op_41765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41765_cast_fp16 = slice_by_index(begin = var_41765_begin_0, end = var_41765_end_0, end_mask = var_41765_end_mask_0, x = var_41251_cast_fp16)[name = tensor("op_41765_cast_fp16")]; + tensor var_41772_begin_0 = const()[name = tensor("op_41772_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41772_end_0 = const()[name = tensor("op_41772_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41772_end_mask_0 = const()[name = tensor("op_41772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41772_cast_fp16 = slice_by_index(begin = var_41772_begin_0, end = var_41772_end_0, end_mask = var_41772_end_mask_0, x = var_41255_cast_fp16)[name = tensor("op_41772_cast_fp16")]; + tensor var_41779_begin_0 = const()[name = tensor("op_41779_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41779_end_0 = const()[name = tensor("op_41779_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41779_end_mask_0 = const()[name = tensor("op_41779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41779_cast_fp16 = slice_by_index(begin = var_41779_begin_0, end = var_41779_end_0, end_mask = var_41779_end_mask_0, x = var_41255_cast_fp16)[name = tensor("op_41779_cast_fp16")]; + tensor var_41786_begin_0 = const()[name = tensor("op_41786_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41786_end_0 = const()[name = tensor("op_41786_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41786_end_mask_0 = const()[name = tensor("op_41786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41786_cast_fp16 = slice_by_index(begin = var_41786_begin_0, end = var_41786_end_0, end_mask = var_41786_end_mask_0, x = var_41255_cast_fp16)[name = tensor("op_41786_cast_fp16")]; + tensor var_41793_begin_0 = const()[name = tensor("op_41793_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41793_end_0 = const()[name = tensor("op_41793_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41793_end_mask_0 = const()[name = tensor("op_41793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41793_cast_fp16 = slice_by_index(begin = var_41793_begin_0, end = var_41793_end_0, end_mask = var_41793_end_mask_0, x = var_41255_cast_fp16)[name = tensor("op_41793_cast_fp16")]; + tensor var_41800_begin_0 = const()[name = tensor("op_41800_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41800_end_0 = const()[name = tensor("op_41800_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_41800_end_mask_0 = const()[name = tensor("op_41800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41800_cast_fp16 = slice_by_index(begin = var_41800_begin_0, end = var_41800_end_0, end_mask = var_41800_end_mask_0, x = var_41259_cast_fp16)[name = tensor("op_41800_cast_fp16")]; + tensor var_41807_begin_0 = const()[name = tensor("op_41807_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_41807_end_0 = const()[name = tensor("op_41807_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_41807_end_mask_0 = const()[name = tensor("op_41807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41807_cast_fp16 = slice_by_index(begin = var_41807_begin_0, end = var_41807_end_0, end_mask = var_41807_end_mask_0, x = var_41259_cast_fp16)[name = tensor("op_41807_cast_fp16")]; + tensor var_41814_begin_0 = const()[name = tensor("op_41814_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_41814_end_0 = const()[name = tensor("op_41814_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_41814_end_mask_0 = const()[name = tensor("op_41814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41814_cast_fp16 = slice_by_index(begin = var_41814_begin_0, end = var_41814_end_0, end_mask = var_41814_end_mask_0, x = var_41259_cast_fp16)[name = tensor("op_41814_cast_fp16")]; + tensor var_41821_begin_0 = const()[name = tensor("op_41821_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_41821_end_0 = const()[name = tensor("op_41821_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41821_end_mask_0 = const()[name = tensor("op_41821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41821_cast_fp16 = slice_by_index(begin = var_41821_begin_0, end = var_41821_end_0, end_mask = var_41821_end_mask_0, x = var_41259_cast_fp16)[name = tensor("op_41821_cast_fp16")]; + tensor k_51_perm_0 = const()[name = tensor("k_51_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_41826_begin_0 = const()[name = tensor("op_41826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41826_end_0 = const()[name = tensor("op_41826_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_41826_end_mask_0 = const()[name = tensor("op_41826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_6 = transpose(perm = k_51_perm_0, x = key_51_cast_fp16)[name = tensor("transpose_6")]; + tensor var_41826_cast_fp16 = slice_by_index(begin = var_41826_begin_0, end = var_41826_end_0, end_mask = var_41826_end_mask_0, x = transpose_6)[name = tensor("op_41826_cast_fp16")]; + tensor var_41830_begin_0 = const()[name = tensor("op_41830_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_41830_end_0 = const()[name = tensor("op_41830_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_41830_end_mask_0 = const()[name = tensor("op_41830_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41830_cast_fp16 = slice_by_index(begin = var_41830_begin_0, end = var_41830_end_0, end_mask = var_41830_end_mask_0, x = transpose_6)[name = tensor("op_41830_cast_fp16")]; + tensor var_41834_begin_0 = const()[name = tensor("op_41834_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_41834_end_0 = const()[name = tensor("op_41834_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_41834_end_mask_0 = const()[name = tensor("op_41834_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41834_cast_fp16 = slice_by_index(begin = var_41834_begin_0, end = var_41834_end_0, end_mask = var_41834_end_mask_0, x = transpose_6)[name = tensor("op_41834_cast_fp16")]; + tensor var_41838_begin_0 = const()[name = tensor("op_41838_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_41838_end_0 = const()[name = tensor("op_41838_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_41838_end_mask_0 = const()[name = tensor("op_41838_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41838_cast_fp16 = slice_by_index(begin = var_41838_begin_0, end = var_41838_end_0, end_mask = var_41838_end_mask_0, x = transpose_6)[name = tensor("op_41838_cast_fp16")]; + tensor var_41842_begin_0 = const()[name = tensor("op_41842_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_41842_end_0 = const()[name = tensor("op_41842_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_41842_end_mask_0 = const()[name = tensor("op_41842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41842_cast_fp16 = slice_by_index(begin = var_41842_begin_0, end = var_41842_end_0, end_mask = var_41842_end_mask_0, x = transpose_6)[name = tensor("op_41842_cast_fp16")]; + tensor var_41846_begin_0 = const()[name = tensor("op_41846_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_41846_end_0 = const()[name = tensor("op_41846_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_41846_end_mask_0 = const()[name = tensor("op_41846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41846_cast_fp16 = slice_by_index(begin = var_41846_begin_0, end = var_41846_end_0, end_mask = var_41846_end_mask_0, x = transpose_6)[name = tensor("op_41846_cast_fp16")]; + tensor var_41850_begin_0 = const()[name = tensor("op_41850_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_41850_end_0 = const()[name = tensor("op_41850_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_41850_end_mask_0 = const()[name = tensor("op_41850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41850_cast_fp16 = slice_by_index(begin = var_41850_begin_0, end = var_41850_end_0, end_mask = var_41850_end_mask_0, x = transpose_6)[name = tensor("op_41850_cast_fp16")]; + tensor var_41854_begin_0 = const()[name = tensor("op_41854_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_41854_end_0 = const()[name = tensor("op_41854_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_41854_end_mask_0 = const()[name = tensor("op_41854_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41854_cast_fp16 = slice_by_index(begin = var_41854_begin_0, end = var_41854_end_0, end_mask = var_41854_end_mask_0, x = transpose_6)[name = tensor("op_41854_cast_fp16")]; + tensor var_41858_begin_0 = const()[name = tensor("op_41858_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_41858_end_0 = const()[name = tensor("op_41858_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_41858_end_mask_0 = const()[name = tensor("op_41858_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41858_cast_fp16 = slice_by_index(begin = var_41858_begin_0, end = var_41858_end_0, end_mask = var_41858_end_mask_0, x = transpose_6)[name = tensor("op_41858_cast_fp16")]; + tensor var_41862_begin_0 = const()[name = tensor("op_41862_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_41862_end_0 = const()[name = tensor("op_41862_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_41862_end_mask_0 = const()[name = tensor("op_41862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41862_cast_fp16 = slice_by_index(begin = var_41862_begin_0, end = var_41862_end_0, end_mask = var_41862_end_mask_0, x = transpose_6)[name = tensor("op_41862_cast_fp16")]; + tensor var_41866_begin_0 = const()[name = tensor("op_41866_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_41866_end_0 = const()[name = tensor("op_41866_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_41866_end_mask_0 = const()[name = tensor("op_41866_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41866_cast_fp16 = slice_by_index(begin = var_41866_begin_0, end = var_41866_end_0, end_mask = var_41866_end_mask_0, x = transpose_6)[name = tensor("op_41866_cast_fp16")]; + tensor var_41870_begin_0 = const()[name = tensor("op_41870_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_41870_end_0 = const()[name = tensor("op_41870_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_41870_end_mask_0 = const()[name = tensor("op_41870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41870_cast_fp16 = slice_by_index(begin = var_41870_begin_0, end = var_41870_end_0, end_mask = var_41870_end_mask_0, x = transpose_6)[name = tensor("op_41870_cast_fp16")]; + tensor var_41874_begin_0 = const()[name = tensor("op_41874_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_41874_end_0 = const()[name = tensor("op_41874_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_41874_end_mask_0 = const()[name = tensor("op_41874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41874_cast_fp16 = slice_by_index(begin = var_41874_begin_0, end = var_41874_end_0, end_mask = var_41874_end_mask_0, x = transpose_6)[name = tensor("op_41874_cast_fp16")]; + tensor var_41878_begin_0 = const()[name = tensor("op_41878_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_41878_end_0 = const()[name = tensor("op_41878_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_41878_end_mask_0 = const()[name = tensor("op_41878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41878_cast_fp16 = slice_by_index(begin = var_41878_begin_0, end = var_41878_end_0, end_mask = var_41878_end_mask_0, x = transpose_6)[name = tensor("op_41878_cast_fp16")]; + tensor var_41882_begin_0 = const()[name = tensor("op_41882_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_41882_end_0 = const()[name = tensor("op_41882_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_41882_end_mask_0 = const()[name = tensor("op_41882_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41882_cast_fp16 = slice_by_index(begin = var_41882_begin_0, end = var_41882_end_0, end_mask = var_41882_end_mask_0, x = transpose_6)[name = tensor("op_41882_cast_fp16")]; + tensor var_41886_begin_0 = const()[name = tensor("op_41886_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_41886_end_0 = const()[name = tensor("op_41886_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_41886_end_mask_0 = const()[name = tensor("op_41886_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41886_cast_fp16 = slice_by_index(begin = var_41886_begin_0, end = var_41886_end_0, end_mask = var_41886_end_mask_0, x = transpose_6)[name = tensor("op_41886_cast_fp16")]; + tensor var_41890_begin_0 = const()[name = tensor("op_41890_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_41890_end_0 = const()[name = tensor("op_41890_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_41890_end_mask_0 = const()[name = tensor("op_41890_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41890_cast_fp16 = slice_by_index(begin = var_41890_begin_0, end = var_41890_end_0, end_mask = var_41890_end_mask_0, x = transpose_6)[name = tensor("op_41890_cast_fp16")]; + tensor var_41894_begin_0 = const()[name = tensor("op_41894_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_41894_end_0 = const()[name = tensor("op_41894_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_41894_end_mask_0 = const()[name = tensor("op_41894_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41894_cast_fp16 = slice_by_index(begin = var_41894_begin_0, end = var_41894_end_0, end_mask = var_41894_end_mask_0, x = transpose_6)[name = tensor("op_41894_cast_fp16")]; + tensor var_41898_begin_0 = const()[name = tensor("op_41898_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_41898_end_0 = const()[name = tensor("op_41898_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_41898_end_mask_0 = const()[name = tensor("op_41898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41898_cast_fp16 = slice_by_index(begin = var_41898_begin_0, end = var_41898_end_0, end_mask = var_41898_end_mask_0, x = transpose_6)[name = tensor("op_41898_cast_fp16")]; + tensor var_41902_begin_0 = const()[name = tensor("op_41902_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_41902_end_0 = const()[name = tensor("op_41902_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_41902_end_mask_0 = const()[name = tensor("op_41902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_41902_cast_fp16 = slice_by_index(begin = var_41902_begin_0, end = var_41902_end_0, end_mask = var_41902_end_mask_0, x = transpose_6)[name = tensor("op_41902_cast_fp16")]; + tensor var_41904_begin_0 = const()[name = tensor("op_41904_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_41904_end_0 = const()[name = tensor("op_41904_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_41904_end_mask_0 = const()[name = tensor("op_41904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41904_cast_fp16 = slice_by_index(begin = var_41904_begin_0, end = var_41904_end_0, end_mask = var_41904_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41904_cast_fp16")]; + tensor var_41908_begin_0 = const()[name = tensor("op_41908_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_41908_end_0 = const()[name = tensor("op_41908_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_41908_end_mask_0 = const()[name = tensor("op_41908_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41908_cast_fp16 = slice_by_index(begin = var_41908_begin_0, end = var_41908_end_0, end_mask = var_41908_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41908_cast_fp16")]; + tensor var_41912_begin_0 = const()[name = tensor("op_41912_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_41912_end_0 = const()[name = tensor("op_41912_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_41912_end_mask_0 = const()[name = tensor("op_41912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41912_cast_fp16 = slice_by_index(begin = var_41912_begin_0, end = var_41912_end_0, end_mask = var_41912_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41912_cast_fp16")]; + tensor var_41916_begin_0 = const()[name = tensor("op_41916_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_41916_end_0 = const()[name = tensor("op_41916_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_41916_end_mask_0 = const()[name = tensor("op_41916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41916_cast_fp16 = slice_by_index(begin = var_41916_begin_0, end = var_41916_end_0, end_mask = var_41916_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41916_cast_fp16")]; + tensor var_41920_begin_0 = const()[name = tensor("op_41920_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_41920_end_0 = const()[name = tensor("op_41920_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_41920_end_mask_0 = const()[name = tensor("op_41920_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41920_cast_fp16 = slice_by_index(begin = var_41920_begin_0, end = var_41920_end_0, end_mask = var_41920_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41920_cast_fp16")]; + tensor var_41924_begin_0 = const()[name = tensor("op_41924_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_41924_end_0 = const()[name = tensor("op_41924_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_41924_end_mask_0 = const()[name = tensor("op_41924_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41924_cast_fp16 = slice_by_index(begin = var_41924_begin_0, end = var_41924_end_0, end_mask = var_41924_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41924_cast_fp16")]; + tensor var_41928_begin_0 = const()[name = tensor("op_41928_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_41928_end_0 = const()[name = tensor("op_41928_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_41928_end_mask_0 = const()[name = tensor("op_41928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41928_cast_fp16 = slice_by_index(begin = var_41928_begin_0, end = var_41928_end_0, end_mask = var_41928_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41928_cast_fp16")]; + tensor var_41932_begin_0 = const()[name = tensor("op_41932_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_41932_end_0 = const()[name = tensor("op_41932_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_41932_end_mask_0 = const()[name = tensor("op_41932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41932_cast_fp16 = slice_by_index(begin = var_41932_begin_0, end = var_41932_end_0, end_mask = var_41932_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41932_cast_fp16")]; + tensor var_41936_begin_0 = const()[name = tensor("op_41936_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_41936_end_0 = const()[name = tensor("op_41936_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_41936_end_mask_0 = const()[name = tensor("op_41936_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41936_cast_fp16 = slice_by_index(begin = var_41936_begin_0, end = var_41936_end_0, end_mask = var_41936_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41936_cast_fp16")]; + tensor var_41940_begin_0 = const()[name = tensor("op_41940_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_41940_end_0 = const()[name = tensor("op_41940_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_41940_end_mask_0 = const()[name = tensor("op_41940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41940_cast_fp16 = slice_by_index(begin = var_41940_begin_0, end = var_41940_end_0, end_mask = var_41940_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41940_cast_fp16")]; + tensor var_41944_begin_0 = const()[name = tensor("op_41944_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_41944_end_0 = const()[name = tensor("op_41944_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_41944_end_mask_0 = const()[name = tensor("op_41944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41944_cast_fp16 = slice_by_index(begin = var_41944_begin_0, end = var_41944_end_0, end_mask = var_41944_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41944_cast_fp16")]; + tensor var_41948_begin_0 = const()[name = tensor("op_41948_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_41948_end_0 = const()[name = tensor("op_41948_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_41948_end_mask_0 = const()[name = tensor("op_41948_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41948_cast_fp16 = slice_by_index(begin = var_41948_begin_0, end = var_41948_end_0, end_mask = var_41948_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41948_cast_fp16")]; + tensor var_41952_begin_0 = const()[name = tensor("op_41952_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_41952_end_0 = const()[name = tensor("op_41952_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_41952_end_mask_0 = const()[name = tensor("op_41952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41952_cast_fp16 = slice_by_index(begin = var_41952_begin_0, end = var_41952_end_0, end_mask = var_41952_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41952_cast_fp16")]; + tensor var_41956_begin_0 = const()[name = tensor("op_41956_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_41956_end_0 = const()[name = tensor("op_41956_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_41956_end_mask_0 = const()[name = tensor("op_41956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41956_cast_fp16 = slice_by_index(begin = var_41956_begin_0, end = var_41956_end_0, end_mask = var_41956_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41956_cast_fp16")]; + tensor var_41960_begin_0 = const()[name = tensor("op_41960_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_41960_end_0 = const()[name = tensor("op_41960_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_41960_end_mask_0 = const()[name = tensor("op_41960_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41960_cast_fp16 = slice_by_index(begin = var_41960_begin_0, end = var_41960_end_0, end_mask = var_41960_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41960_cast_fp16")]; + tensor var_41964_begin_0 = const()[name = tensor("op_41964_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_41964_end_0 = const()[name = tensor("op_41964_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_41964_end_mask_0 = const()[name = tensor("op_41964_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41964_cast_fp16 = slice_by_index(begin = var_41964_begin_0, end = var_41964_end_0, end_mask = var_41964_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41964_cast_fp16")]; + tensor var_41968_begin_0 = const()[name = tensor("op_41968_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_41968_end_0 = const()[name = tensor("op_41968_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_41968_end_mask_0 = const()[name = tensor("op_41968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41968_cast_fp16 = slice_by_index(begin = var_41968_begin_0, end = var_41968_end_0, end_mask = var_41968_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41968_cast_fp16")]; + tensor var_41972_begin_0 = const()[name = tensor("op_41972_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_41972_end_0 = const()[name = tensor("op_41972_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_41972_end_mask_0 = const()[name = tensor("op_41972_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41972_cast_fp16 = slice_by_index(begin = var_41972_begin_0, end = var_41972_end_0, end_mask = var_41972_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41972_cast_fp16")]; + tensor var_41976_begin_0 = const()[name = tensor("op_41976_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_41976_end_0 = const()[name = tensor("op_41976_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_41976_end_mask_0 = const()[name = tensor("op_41976_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41976_cast_fp16 = slice_by_index(begin = var_41976_begin_0, end = var_41976_end_0, end_mask = var_41976_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41976_cast_fp16")]; + tensor var_41980_begin_0 = const()[name = tensor("op_41980_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_41980_end_0 = const()[name = tensor("op_41980_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_41980_end_mask_0 = const()[name = tensor("op_41980_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_41980_cast_fp16 = slice_by_index(begin = var_41980_begin_0, end = var_41980_end_0, end_mask = var_41980_end_mask_0, x = value_51_cast_fp16)[name = tensor("op_41980_cast_fp16")]; + tensor var_41984_equation_0 = const()[name = tensor("op_41984_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41984_cast_fp16 = einsum(equation = var_41984_equation_0, values = (var_41826_cast_fp16, var_41268_cast_fp16))[name = tensor("op_41984_cast_fp16")]; + tensor var_41985_to_fp16 = const()[name = tensor("op_41985_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4001_cast_fp16 = mul(x = var_41984_cast_fp16, y = var_41985_to_fp16)[name = tensor("aw_chunk_4001_cast_fp16")]; + tensor var_41988_equation_0 = const()[name = tensor("op_41988_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41988_cast_fp16 = einsum(equation = var_41988_equation_0, values = (var_41826_cast_fp16, var_41275_cast_fp16))[name = tensor("op_41988_cast_fp16")]; + tensor var_41989_to_fp16 = const()[name = tensor("op_41989_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4003_cast_fp16 = mul(x = var_41988_cast_fp16, y = var_41989_to_fp16)[name = tensor("aw_chunk_4003_cast_fp16")]; + tensor var_41992_equation_0 = const()[name = tensor("op_41992_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41992_cast_fp16 = einsum(equation = var_41992_equation_0, values = (var_41826_cast_fp16, var_41282_cast_fp16))[name = tensor("op_41992_cast_fp16")]; + tensor var_41993_to_fp16 = const()[name = tensor("op_41993_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4005_cast_fp16 = mul(x = var_41992_cast_fp16, y = var_41993_to_fp16)[name = tensor("aw_chunk_4005_cast_fp16")]; + tensor var_41996_equation_0 = const()[name = tensor("op_41996_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_41996_cast_fp16 = einsum(equation = var_41996_equation_0, values = (var_41826_cast_fp16, var_41289_cast_fp16))[name = tensor("op_41996_cast_fp16")]; + tensor var_41997_to_fp16 = const()[name = tensor("op_41997_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4007_cast_fp16 = mul(x = var_41996_cast_fp16, y = var_41997_to_fp16)[name = tensor("aw_chunk_4007_cast_fp16")]; + tensor var_42000_equation_0 = const()[name = tensor("op_42000_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42000_cast_fp16 = einsum(equation = var_42000_equation_0, values = (var_41830_cast_fp16, var_41296_cast_fp16))[name = tensor("op_42000_cast_fp16")]; + tensor var_42001_to_fp16 = const()[name = tensor("op_42001_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4009_cast_fp16 = mul(x = var_42000_cast_fp16, y = var_42001_to_fp16)[name = tensor("aw_chunk_4009_cast_fp16")]; + tensor var_42004_equation_0 = const()[name = tensor("op_42004_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42004_cast_fp16 = einsum(equation = var_42004_equation_0, values = (var_41830_cast_fp16, var_41303_cast_fp16))[name = tensor("op_42004_cast_fp16")]; + tensor var_42005_to_fp16 = const()[name = tensor("op_42005_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4011_cast_fp16 = mul(x = var_42004_cast_fp16, y = var_42005_to_fp16)[name = tensor("aw_chunk_4011_cast_fp16")]; + tensor var_42008_equation_0 = const()[name = tensor("op_42008_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42008_cast_fp16 = einsum(equation = var_42008_equation_0, values = (var_41830_cast_fp16, var_41310_cast_fp16))[name = tensor("op_42008_cast_fp16")]; + tensor var_42009_to_fp16 = const()[name = tensor("op_42009_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4013_cast_fp16 = mul(x = var_42008_cast_fp16, y = var_42009_to_fp16)[name = tensor("aw_chunk_4013_cast_fp16")]; + tensor var_42012_equation_0 = const()[name = tensor("op_42012_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42012_cast_fp16 = einsum(equation = var_42012_equation_0, values = (var_41830_cast_fp16, var_41317_cast_fp16))[name = tensor("op_42012_cast_fp16")]; + tensor var_42013_to_fp16 = const()[name = tensor("op_42013_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4015_cast_fp16 = mul(x = var_42012_cast_fp16, y = var_42013_to_fp16)[name = tensor("aw_chunk_4015_cast_fp16")]; + tensor var_42016_equation_0 = const()[name = tensor("op_42016_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42016_cast_fp16 = einsum(equation = var_42016_equation_0, values = (var_41834_cast_fp16, var_41324_cast_fp16))[name = tensor("op_42016_cast_fp16")]; + tensor var_42017_to_fp16 = const()[name = tensor("op_42017_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4017_cast_fp16 = mul(x = var_42016_cast_fp16, y = var_42017_to_fp16)[name = tensor("aw_chunk_4017_cast_fp16")]; + tensor var_42020_equation_0 = const()[name = tensor("op_42020_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42020_cast_fp16 = einsum(equation = var_42020_equation_0, values = (var_41834_cast_fp16, var_41331_cast_fp16))[name = tensor("op_42020_cast_fp16")]; + tensor var_42021_to_fp16 = const()[name = tensor("op_42021_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4019_cast_fp16 = mul(x = var_42020_cast_fp16, y = var_42021_to_fp16)[name = tensor("aw_chunk_4019_cast_fp16")]; + tensor var_42024_equation_0 = const()[name = tensor("op_42024_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42024_cast_fp16 = einsum(equation = var_42024_equation_0, values = (var_41834_cast_fp16, var_41338_cast_fp16))[name = tensor("op_42024_cast_fp16")]; + tensor var_42025_to_fp16 = const()[name = tensor("op_42025_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4021_cast_fp16 = mul(x = var_42024_cast_fp16, y = var_42025_to_fp16)[name = tensor("aw_chunk_4021_cast_fp16")]; + tensor var_42028_equation_0 = const()[name = tensor("op_42028_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42028_cast_fp16 = einsum(equation = var_42028_equation_0, values = (var_41834_cast_fp16, var_41345_cast_fp16))[name = tensor("op_42028_cast_fp16")]; + tensor var_42029_to_fp16 = const()[name = tensor("op_42029_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4023_cast_fp16 = mul(x = var_42028_cast_fp16, y = var_42029_to_fp16)[name = tensor("aw_chunk_4023_cast_fp16")]; + tensor var_42032_equation_0 = const()[name = tensor("op_42032_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42032_cast_fp16 = einsum(equation = var_42032_equation_0, values = (var_41838_cast_fp16, var_41352_cast_fp16))[name = tensor("op_42032_cast_fp16")]; + tensor var_42033_to_fp16 = const()[name = tensor("op_42033_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4025_cast_fp16 = mul(x = var_42032_cast_fp16, y = var_42033_to_fp16)[name = tensor("aw_chunk_4025_cast_fp16")]; + tensor var_42036_equation_0 = const()[name = tensor("op_42036_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42036_cast_fp16 = einsum(equation = var_42036_equation_0, values = (var_41838_cast_fp16, var_41359_cast_fp16))[name = tensor("op_42036_cast_fp16")]; + tensor var_42037_to_fp16 = const()[name = tensor("op_42037_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4027_cast_fp16 = mul(x = var_42036_cast_fp16, y = var_42037_to_fp16)[name = tensor("aw_chunk_4027_cast_fp16")]; + tensor var_42040_equation_0 = const()[name = tensor("op_42040_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42040_cast_fp16 = einsum(equation = var_42040_equation_0, values = (var_41838_cast_fp16, var_41366_cast_fp16))[name = tensor("op_42040_cast_fp16")]; + tensor var_42041_to_fp16 = const()[name = tensor("op_42041_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4029_cast_fp16 = mul(x = var_42040_cast_fp16, y = var_42041_to_fp16)[name = tensor("aw_chunk_4029_cast_fp16")]; + tensor var_42044_equation_0 = const()[name = tensor("op_42044_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42044_cast_fp16 = einsum(equation = var_42044_equation_0, values = (var_41838_cast_fp16, var_41373_cast_fp16))[name = tensor("op_42044_cast_fp16")]; + tensor var_42045_to_fp16 = const()[name = tensor("op_42045_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4031_cast_fp16 = mul(x = var_42044_cast_fp16, y = var_42045_to_fp16)[name = tensor("aw_chunk_4031_cast_fp16")]; + tensor var_42048_equation_0 = const()[name = tensor("op_42048_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42048_cast_fp16 = einsum(equation = var_42048_equation_0, values = (var_41842_cast_fp16, var_41380_cast_fp16))[name = tensor("op_42048_cast_fp16")]; + tensor var_42049_to_fp16 = const()[name = tensor("op_42049_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4033_cast_fp16 = mul(x = var_42048_cast_fp16, y = var_42049_to_fp16)[name = tensor("aw_chunk_4033_cast_fp16")]; + tensor var_42052_equation_0 = const()[name = tensor("op_42052_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42052_cast_fp16 = einsum(equation = var_42052_equation_0, values = (var_41842_cast_fp16, var_41387_cast_fp16))[name = tensor("op_42052_cast_fp16")]; + tensor var_42053_to_fp16 = const()[name = tensor("op_42053_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4035_cast_fp16 = mul(x = var_42052_cast_fp16, y = var_42053_to_fp16)[name = tensor("aw_chunk_4035_cast_fp16")]; + tensor var_42056_equation_0 = const()[name = tensor("op_42056_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42056_cast_fp16 = einsum(equation = var_42056_equation_0, values = (var_41842_cast_fp16, var_41394_cast_fp16))[name = tensor("op_42056_cast_fp16")]; + tensor var_42057_to_fp16 = const()[name = tensor("op_42057_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4037_cast_fp16 = mul(x = var_42056_cast_fp16, y = var_42057_to_fp16)[name = tensor("aw_chunk_4037_cast_fp16")]; + tensor var_42060_equation_0 = const()[name = tensor("op_42060_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42060_cast_fp16 = einsum(equation = var_42060_equation_0, values = (var_41842_cast_fp16, var_41401_cast_fp16))[name = tensor("op_42060_cast_fp16")]; + tensor var_42061_to_fp16 = const()[name = tensor("op_42061_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4039_cast_fp16 = mul(x = var_42060_cast_fp16, y = var_42061_to_fp16)[name = tensor("aw_chunk_4039_cast_fp16")]; + tensor var_42064_equation_0 = const()[name = tensor("op_42064_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42064_cast_fp16 = einsum(equation = var_42064_equation_0, values = (var_41846_cast_fp16, var_41408_cast_fp16))[name = tensor("op_42064_cast_fp16")]; + tensor var_42065_to_fp16 = const()[name = tensor("op_42065_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4041_cast_fp16 = mul(x = var_42064_cast_fp16, y = var_42065_to_fp16)[name = tensor("aw_chunk_4041_cast_fp16")]; + tensor var_42068_equation_0 = const()[name = tensor("op_42068_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42068_cast_fp16 = einsum(equation = var_42068_equation_0, values = (var_41846_cast_fp16, var_41415_cast_fp16))[name = tensor("op_42068_cast_fp16")]; + tensor var_42069_to_fp16 = const()[name = tensor("op_42069_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4043_cast_fp16 = mul(x = var_42068_cast_fp16, y = var_42069_to_fp16)[name = tensor("aw_chunk_4043_cast_fp16")]; + tensor var_42072_equation_0 = const()[name = tensor("op_42072_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42072_cast_fp16 = einsum(equation = var_42072_equation_0, values = (var_41846_cast_fp16, var_41422_cast_fp16))[name = tensor("op_42072_cast_fp16")]; + tensor var_42073_to_fp16 = const()[name = tensor("op_42073_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4045_cast_fp16 = mul(x = var_42072_cast_fp16, y = var_42073_to_fp16)[name = tensor("aw_chunk_4045_cast_fp16")]; + tensor var_42076_equation_0 = const()[name = tensor("op_42076_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42076_cast_fp16 = einsum(equation = var_42076_equation_0, values = (var_41846_cast_fp16, var_41429_cast_fp16))[name = tensor("op_42076_cast_fp16")]; + tensor var_42077_to_fp16 = const()[name = tensor("op_42077_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4047_cast_fp16 = mul(x = var_42076_cast_fp16, y = var_42077_to_fp16)[name = tensor("aw_chunk_4047_cast_fp16")]; + tensor var_42080_equation_0 = const()[name = tensor("op_42080_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42080_cast_fp16 = einsum(equation = var_42080_equation_0, values = (var_41850_cast_fp16, var_41436_cast_fp16))[name = tensor("op_42080_cast_fp16")]; + tensor var_42081_to_fp16 = const()[name = tensor("op_42081_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4049_cast_fp16 = mul(x = var_42080_cast_fp16, y = var_42081_to_fp16)[name = tensor("aw_chunk_4049_cast_fp16")]; + tensor var_42084_equation_0 = const()[name = tensor("op_42084_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42084_cast_fp16 = einsum(equation = var_42084_equation_0, values = (var_41850_cast_fp16, var_41443_cast_fp16))[name = tensor("op_42084_cast_fp16")]; + tensor var_42085_to_fp16 = const()[name = tensor("op_42085_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4051_cast_fp16 = mul(x = var_42084_cast_fp16, y = var_42085_to_fp16)[name = tensor("aw_chunk_4051_cast_fp16")]; + tensor var_42088_equation_0 = const()[name = tensor("op_42088_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42088_cast_fp16 = einsum(equation = var_42088_equation_0, values = (var_41850_cast_fp16, var_41450_cast_fp16))[name = tensor("op_42088_cast_fp16")]; + tensor var_42089_to_fp16 = const()[name = tensor("op_42089_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4053_cast_fp16 = mul(x = var_42088_cast_fp16, y = var_42089_to_fp16)[name = tensor("aw_chunk_4053_cast_fp16")]; + tensor var_42092_equation_0 = const()[name = tensor("op_42092_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42092_cast_fp16 = einsum(equation = var_42092_equation_0, values = (var_41850_cast_fp16, var_41457_cast_fp16))[name = tensor("op_42092_cast_fp16")]; + tensor var_42093_to_fp16 = const()[name = tensor("op_42093_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4055_cast_fp16 = mul(x = var_42092_cast_fp16, y = var_42093_to_fp16)[name = tensor("aw_chunk_4055_cast_fp16")]; + tensor var_42096_equation_0 = const()[name = tensor("op_42096_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42096_cast_fp16 = einsum(equation = var_42096_equation_0, values = (var_41854_cast_fp16, var_41464_cast_fp16))[name = tensor("op_42096_cast_fp16")]; + tensor var_42097_to_fp16 = const()[name = tensor("op_42097_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4057_cast_fp16 = mul(x = var_42096_cast_fp16, y = var_42097_to_fp16)[name = tensor("aw_chunk_4057_cast_fp16")]; + tensor var_42100_equation_0 = const()[name = tensor("op_42100_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42100_cast_fp16 = einsum(equation = var_42100_equation_0, values = (var_41854_cast_fp16, var_41471_cast_fp16))[name = tensor("op_42100_cast_fp16")]; + tensor var_42101_to_fp16 = const()[name = tensor("op_42101_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4059_cast_fp16 = mul(x = var_42100_cast_fp16, y = var_42101_to_fp16)[name = tensor("aw_chunk_4059_cast_fp16")]; + tensor var_42104_equation_0 = const()[name = tensor("op_42104_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42104_cast_fp16 = einsum(equation = var_42104_equation_0, values = (var_41854_cast_fp16, var_41478_cast_fp16))[name = tensor("op_42104_cast_fp16")]; + tensor var_42105_to_fp16 = const()[name = tensor("op_42105_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4061_cast_fp16 = mul(x = var_42104_cast_fp16, y = var_42105_to_fp16)[name = tensor("aw_chunk_4061_cast_fp16")]; + tensor var_42108_equation_0 = const()[name = tensor("op_42108_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42108_cast_fp16 = einsum(equation = var_42108_equation_0, values = (var_41854_cast_fp16, var_41485_cast_fp16))[name = tensor("op_42108_cast_fp16")]; + tensor var_42109_to_fp16 = const()[name = tensor("op_42109_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4063_cast_fp16 = mul(x = var_42108_cast_fp16, y = var_42109_to_fp16)[name = tensor("aw_chunk_4063_cast_fp16")]; + tensor var_42112_equation_0 = const()[name = tensor("op_42112_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42112_cast_fp16 = einsum(equation = var_42112_equation_0, values = (var_41858_cast_fp16, var_41492_cast_fp16))[name = tensor("op_42112_cast_fp16")]; + tensor var_42113_to_fp16 = const()[name = tensor("op_42113_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4065_cast_fp16 = mul(x = var_42112_cast_fp16, y = var_42113_to_fp16)[name = tensor("aw_chunk_4065_cast_fp16")]; + tensor var_42116_equation_0 = const()[name = tensor("op_42116_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42116_cast_fp16 = einsum(equation = var_42116_equation_0, values = (var_41858_cast_fp16, var_41499_cast_fp16))[name = tensor("op_42116_cast_fp16")]; + tensor var_42117_to_fp16 = const()[name = tensor("op_42117_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4067_cast_fp16 = mul(x = var_42116_cast_fp16, y = var_42117_to_fp16)[name = tensor("aw_chunk_4067_cast_fp16")]; + tensor var_42120_equation_0 = const()[name = tensor("op_42120_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42120_cast_fp16 = einsum(equation = var_42120_equation_0, values = (var_41858_cast_fp16, var_41506_cast_fp16))[name = tensor("op_42120_cast_fp16")]; + tensor var_42121_to_fp16 = const()[name = tensor("op_42121_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4069_cast_fp16 = mul(x = var_42120_cast_fp16, y = var_42121_to_fp16)[name = tensor("aw_chunk_4069_cast_fp16")]; + tensor var_42124_equation_0 = const()[name = tensor("op_42124_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42124_cast_fp16 = einsum(equation = var_42124_equation_0, values = (var_41858_cast_fp16, var_41513_cast_fp16))[name = tensor("op_42124_cast_fp16")]; + tensor var_42125_to_fp16 = const()[name = tensor("op_42125_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4071_cast_fp16 = mul(x = var_42124_cast_fp16, y = var_42125_to_fp16)[name = tensor("aw_chunk_4071_cast_fp16")]; + tensor var_42128_equation_0 = const()[name = tensor("op_42128_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42128_cast_fp16 = einsum(equation = var_42128_equation_0, values = (var_41862_cast_fp16, var_41520_cast_fp16))[name = tensor("op_42128_cast_fp16")]; + tensor var_42129_to_fp16 = const()[name = tensor("op_42129_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4073_cast_fp16 = mul(x = var_42128_cast_fp16, y = var_42129_to_fp16)[name = tensor("aw_chunk_4073_cast_fp16")]; + tensor var_42132_equation_0 = const()[name = tensor("op_42132_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42132_cast_fp16 = einsum(equation = var_42132_equation_0, values = (var_41862_cast_fp16, var_41527_cast_fp16))[name = tensor("op_42132_cast_fp16")]; + tensor var_42133_to_fp16 = const()[name = tensor("op_42133_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4075_cast_fp16 = mul(x = var_42132_cast_fp16, y = var_42133_to_fp16)[name = tensor("aw_chunk_4075_cast_fp16")]; + tensor var_42136_equation_0 = const()[name = tensor("op_42136_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42136_cast_fp16 = einsum(equation = var_42136_equation_0, values = (var_41862_cast_fp16, var_41534_cast_fp16))[name = tensor("op_42136_cast_fp16")]; + tensor var_42137_to_fp16 = const()[name = tensor("op_42137_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4077_cast_fp16 = mul(x = var_42136_cast_fp16, y = var_42137_to_fp16)[name = tensor("aw_chunk_4077_cast_fp16")]; + tensor var_42140_equation_0 = const()[name = tensor("op_42140_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42140_cast_fp16 = einsum(equation = var_42140_equation_0, values = (var_41862_cast_fp16, var_41541_cast_fp16))[name = tensor("op_42140_cast_fp16")]; + tensor var_42141_to_fp16 = const()[name = tensor("op_42141_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4079_cast_fp16 = mul(x = var_42140_cast_fp16, y = var_42141_to_fp16)[name = tensor("aw_chunk_4079_cast_fp16")]; + tensor var_42144_equation_0 = const()[name = tensor("op_42144_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42144_cast_fp16 = einsum(equation = var_42144_equation_0, values = (var_41866_cast_fp16, var_41548_cast_fp16))[name = tensor("op_42144_cast_fp16")]; + tensor var_42145_to_fp16 = const()[name = tensor("op_42145_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4081_cast_fp16 = mul(x = var_42144_cast_fp16, y = var_42145_to_fp16)[name = tensor("aw_chunk_4081_cast_fp16")]; + tensor var_42148_equation_0 = const()[name = tensor("op_42148_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42148_cast_fp16 = einsum(equation = var_42148_equation_0, values = (var_41866_cast_fp16, var_41555_cast_fp16))[name = tensor("op_42148_cast_fp16")]; + tensor var_42149_to_fp16 = const()[name = tensor("op_42149_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4083_cast_fp16 = mul(x = var_42148_cast_fp16, y = var_42149_to_fp16)[name = tensor("aw_chunk_4083_cast_fp16")]; + tensor var_42152_equation_0 = const()[name = tensor("op_42152_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42152_cast_fp16 = einsum(equation = var_42152_equation_0, values = (var_41866_cast_fp16, var_41562_cast_fp16))[name = tensor("op_42152_cast_fp16")]; + tensor var_42153_to_fp16 = const()[name = tensor("op_42153_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4085_cast_fp16 = mul(x = var_42152_cast_fp16, y = var_42153_to_fp16)[name = tensor("aw_chunk_4085_cast_fp16")]; + tensor var_42156_equation_0 = const()[name = tensor("op_42156_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42156_cast_fp16 = einsum(equation = var_42156_equation_0, values = (var_41866_cast_fp16, var_41569_cast_fp16))[name = tensor("op_42156_cast_fp16")]; + tensor var_42157_to_fp16 = const()[name = tensor("op_42157_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4087_cast_fp16 = mul(x = var_42156_cast_fp16, y = var_42157_to_fp16)[name = tensor("aw_chunk_4087_cast_fp16")]; + tensor var_42160_equation_0 = const()[name = tensor("op_42160_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42160_cast_fp16 = einsum(equation = var_42160_equation_0, values = (var_41870_cast_fp16, var_41576_cast_fp16))[name = tensor("op_42160_cast_fp16")]; + tensor var_42161_to_fp16 = const()[name = tensor("op_42161_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4089_cast_fp16 = mul(x = var_42160_cast_fp16, y = var_42161_to_fp16)[name = tensor("aw_chunk_4089_cast_fp16")]; + tensor var_42164_equation_0 = const()[name = tensor("op_42164_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42164_cast_fp16 = einsum(equation = var_42164_equation_0, values = (var_41870_cast_fp16, var_41583_cast_fp16))[name = tensor("op_42164_cast_fp16")]; + tensor var_42165_to_fp16 = const()[name = tensor("op_42165_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4091_cast_fp16 = mul(x = var_42164_cast_fp16, y = var_42165_to_fp16)[name = tensor("aw_chunk_4091_cast_fp16")]; + tensor var_42168_equation_0 = const()[name = tensor("op_42168_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42168_cast_fp16 = einsum(equation = var_42168_equation_0, values = (var_41870_cast_fp16, var_41590_cast_fp16))[name = tensor("op_42168_cast_fp16")]; + tensor var_42169_to_fp16 = const()[name = tensor("op_42169_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4093_cast_fp16 = mul(x = var_42168_cast_fp16, y = var_42169_to_fp16)[name = tensor("aw_chunk_4093_cast_fp16")]; + tensor var_42172_equation_0 = const()[name = tensor("op_42172_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42172_cast_fp16 = einsum(equation = var_42172_equation_0, values = (var_41870_cast_fp16, var_41597_cast_fp16))[name = tensor("op_42172_cast_fp16")]; + tensor var_42173_to_fp16 = const()[name = tensor("op_42173_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4095_cast_fp16 = mul(x = var_42172_cast_fp16, y = var_42173_to_fp16)[name = tensor("aw_chunk_4095_cast_fp16")]; + tensor var_42176_equation_0 = const()[name = tensor("op_42176_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42176_cast_fp16 = einsum(equation = var_42176_equation_0, values = (var_41874_cast_fp16, var_41604_cast_fp16))[name = tensor("op_42176_cast_fp16")]; + tensor var_42177_to_fp16 = const()[name = tensor("op_42177_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4097_cast_fp16 = mul(x = var_42176_cast_fp16, y = var_42177_to_fp16)[name = tensor("aw_chunk_4097_cast_fp16")]; + tensor var_42180_equation_0 = const()[name = tensor("op_42180_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42180_cast_fp16 = einsum(equation = var_42180_equation_0, values = (var_41874_cast_fp16, var_41611_cast_fp16))[name = tensor("op_42180_cast_fp16")]; + tensor var_42181_to_fp16 = const()[name = tensor("op_42181_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4099_cast_fp16 = mul(x = var_42180_cast_fp16, y = var_42181_to_fp16)[name = tensor("aw_chunk_4099_cast_fp16")]; + tensor var_42184_equation_0 = const()[name = tensor("op_42184_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42184_cast_fp16 = einsum(equation = var_42184_equation_0, values = (var_41874_cast_fp16, var_41618_cast_fp16))[name = tensor("op_42184_cast_fp16")]; + tensor var_42185_to_fp16 = const()[name = tensor("op_42185_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4101_cast_fp16 = mul(x = var_42184_cast_fp16, y = var_42185_to_fp16)[name = tensor("aw_chunk_4101_cast_fp16")]; + tensor var_42188_equation_0 = const()[name = tensor("op_42188_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42188_cast_fp16 = einsum(equation = var_42188_equation_0, values = (var_41874_cast_fp16, var_41625_cast_fp16))[name = tensor("op_42188_cast_fp16")]; + tensor var_42189_to_fp16 = const()[name = tensor("op_42189_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4103_cast_fp16 = mul(x = var_42188_cast_fp16, y = var_42189_to_fp16)[name = tensor("aw_chunk_4103_cast_fp16")]; + tensor var_42192_equation_0 = const()[name = tensor("op_42192_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42192_cast_fp16 = einsum(equation = var_42192_equation_0, values = (var_41878_cast_fp16, var_41632_cast_fp16))[name = tensor("op_42192_cast_fp16")]; + tensor var_42193_to_fp16 = const()[name = tensor("op_42193_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4105_cast_fp16 = mul(x = var_42192_cast_fp16, y = var_42193_to_fp16)[name = tensor("aw_chunk_4105_cast_fp16")]; + tensor var_42196_equation_0 = const()[name = tensor("op_42196_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42196_cast_fp16 = einsum(equation = var_42196_equation_0, values = (var_41878_cast_fp16, var_41639_cast_fp16))[name = tensor("op_42196_cast_fp16")]; + tensor var_42197_to_fp16 = const()[name = tensor("op_42197_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4107_cast_fp16 = mul(x = var_42196_cast_fp16, y = var_42197_to_fp16)[name = tensor("aw_chunk_4107_cast_fp16")]; + tensor var_42200_equation_0 = const()[name = tensor("op_42200_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42200_cast_fp16 = einsum(equation = var_42200_equation_0, values = (var_41878_cast_fp16, var_41646_cast_fp16))[name = tensor("op_42200_cast_fp16")]; + tensor var_42201_to_fp16 = const()[name = tensor("op_42201_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4109_cast_fp16 = mul(x = var_42200_cast_fp16, y = var_42201_to_fp16)[name = tensor("aw_chunk_4109_cast_fp16")]; + tensor var_42204_equation_0 = const()[name = tensor("op_42204_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42204_cast_fp16 = einsum(equation = var_42204_equation_0, values = (var_41878_cast_fp16, var_41653_cast_fp16))[name = tensor("op_42204_cast_fp16")]; + tensor var_42205_to_fp16 = const()[name = tensor("op_42205_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4111_cast_fp16 = mul(x = var_42204_cast_fp16, y = var_42205_to_fp16)[name = tensor("aw_chunk_4111_cast_fp16")]; + tensor var_42208_equation_0 = const()[name = tensor("op_42208_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42208_cast_fp16 = einsum(equation = var_42208_equation_0, values = (var_41882_cast_fp16, var_41660_cast_fp16))[name = tensor("op_42208_cast_fp16")]; + tensor var_42209_to_fp16 = const()[name = tensor("op_42209_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4113_cast_fp16 = mul(x = var_42208_cast_fp16, y = var_42209_to_fp16)[name = tensor("aw_chunk_4113_cast_fp16")]; + tensor var_42212_equation_0 = const()[name = tensor("op_42212_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42212_cast_fp16 = einsum(equation = var_42212_equation_0, values = (var_41882_cast_fp16, var_41667_cast_fp16))[name = tensor("op_42212_cast_fp16")]; + tensor var_42213_to_fp16 = const()[name = tensor("op_42213_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4115_cast_fp16 = mul(x = var_42212_cast_fp16, y = var_42213_to_fp16)[name = tensor("aw_chunk_4115_cast_fp16")]; + tensor var_42216_equation_0 = const()[name = tensor("op_42216_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42216_cast_fp16 = einsum(equation = var_42216_equation_0, values = (var_41882_cast_fp16, var_41674_cast_fp16))[name = tensor("op_42216_cast_fp16")]; + tensor var_42217_to_fp16 = const()[name = tensor("op_42217_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4117_cast_fp16 = mul(x = var_42216_cast_fp16, y = var_42217_to_fp16)[name = tensor("aw_chunk_4117_cast_fp16")]; + tensor var_42220_equation_0 = const()[name = tensor("op_42220_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42220_cast_fp16 = einsum(equation = var_42220_equation_0, values = (var_41882_cast_fp16, var_41681_cast_fp16))[name = tensor("op_42220_cast_fp16")]; + tensor var_42221_to_fp16 = const()[name = tensor("op_42221_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4119_cast_fp16 = mul(x = var_42220_cast_fp16, y = var_42221_to_fp16)[name = tensor("aw_chunk_4119_cast_fp16")]; + tensor var_42224_equation_0 = const()[name = tensor("op_42224_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42224_cast_fp16 = einsum(equation = var_42224_equation_0, values = (var_41886_cast_fp16, var_41688_cast_fp16))[name = tensor("op_42224_cast_fp16")]; + tensor var_42225_to_fp16 = const()[name = tensor("op_42225_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4121_cast_fp16 = mul(x = var_42224_cast_fp16, y = var_42225_to_fp16)[name = tensor("aw_chunk_4121_cast_fp16")]; + tensor var_42228_equation_0 = const()[name = tensor("op_42228_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42228_cast_fp16 = einsum(equation = var_42228_equation_0, values = (var_41886_cast_fp16, var_41695_cast_fp16))[name = tensor("op_42228_cast_fp16")]; + tensor var_42229_to_fp16 = const()[name = tensor("op_42229_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4123_cast_fp16 = mul(x = var_42228_cast_fp16, y = var_42229_to_fp16)[name = tensor("aw_chunk_4123_cast_fp16")]; + tensor var_42232_equation_0 = const()[name = tensor("op_42232_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42232_cast_fp16 = einsum(equation = var_42232_equation_0, values = (var_41886_cast_fp16, var_41702_cast_fp16))[name = tensor("op_42232_cast_fp16")]; + tensor var_42233_to_fp16 = const()[name = tensor("op_42233_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4125_cast_fp16 = mul(x = var_42232_cast_fp16, y = var_42233_to_fp16)[name = tensor("aw_chunk_4125_cast_fp16")]; + tensor var_42236_equation_0 = const()[name = tensor("op_42236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42236_cast_fp16 = einsum(equation = var_42236_equation_0, values = (var_41886_cast_fp16, var_41709_cast_fp16))[name = tensor("op_42236_cast_fp16")]; + tensor var_42237_to_fp16 = const()[name = tensor("op_42237_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4127_cast_fp16 = mul(x = var_42236_cast_fp16, y = var_42237_to_fp16)[name = tensor("aw_chunk_4127_cast_fp16")]; + tensor var_42240_equation_0 = const()[name = tensor("op_42240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42240_cast_fp16 = einsum(equation = var_42240_equation_0, values = (var_41890_cast_fp16, var_41716_cast_fp16))[name = tensor("op_42240_cast_fp16")]; + tensor var_42241_to_fp16 = const()[name = tensor("op_42241_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4129_cast_fp16 = mul(x = var_42240_cast_fp16, y = var_42241_to_fp16)[name = tensor("aw_chunk_4129_cast_fp16")]; + tensor var_42244_equation_0 = const()[name = tensor("op_42244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42244_cast_fp16 = einsum(equation = var_42244_equation_0, values = (var_41890_cast_fp16, var_41723_cast_fp16))[name = tensor("op_42244_cast_fp16")]; + tensor var_42245_to_fp16 = const()[name = tensor("op_42245_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4131_cast_fp16 = mul(x = var_42244_cast_fp16, y = var_42245_to_fp16)[name = tensor("aw_chunk_4131_cast_fp16")]; + tensor var_42248_equation_0 = const()[name = tensor("op_42248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42248_cast_fp16 = einsum(equation = var_42248_equation_0, values = (var_41890_cast_fp16, var_41730_cast_fp16))[name = tensor("op_42248_cast_fp16")]; + tensor var_42249_to_fp16 = const()[name = tensor("op_42249_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4133_cast_fp16 = mul(x = var_42248_cast_fp16, y = var_42249_to_fp16)[name = tensor("aw_chunk_4133_cast_fp16")]; + tensor var_42252_equation_0 = const()[name = tensor("op_42252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42252_cast_fp16 = einsum(equation = var_42252_equation_0, values = (var_41890_cast_fp16, var_41737_cast_fp16))[name = tensor("op_42252_cast_fp16")]; + tensor var_42253_to_fp16 = const()[name = tensor("op_42253_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4135_cast_fp16 = mul(x = var_42252_cast_fp16, y = var_42253_to_fp16)[name = tensor("aw_chunk_4135_cast_fp16")]; + tensor var_42256_equation_0 = const()[name = tensor("op_42256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42256_cast_fp16 = einsum(equation = var_42256_equation_0, values = (var_41894_cast_fp16, var_41744_cast_fp16))[name = tensor("op_42256_cast_fp16")]; + tensor var_42257_to_fp16 = const()[name = tensor("op_42257_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4137_cast_fp16 = mul(x = var_42256_cast_fp16, y = var_42257_to_fp16)[name = tensor("aw_chunk_4137_cast_fp16")]; + tensor var_42260_equation_0 = const()[name = tensor("op_42260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42260_cast_fp16 = einsum(equation = var_42260_equation_0, values = (var_41894_cast_fp16, var_41751_cast_fp16))[name = tensor("op_42260_cast_fp16")]; + tensor var_42261_to_fp16 = const()[name = tensor("op_42261_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4139_cast_fp16 = mul(x = var_42260_cast_fp16, y = var_42261_to_fp16)[name = tensor("aw_chunk_4139_cast_fp16")]; + tensor var_42264_equation_0 = const()[name = tensor("op_42264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42264_cast_fp16 = einsum(equation = var_42264_equation_0, values = (var_41894_cast_fp16, var_41758_cast_fp16))[name = tensor("op_42264_cast_fp16")]; + tensor var_42265_to_fp16 = const()[name = tensor("op_42265_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4141_cast_fp16 = mul(x = var_42264_cast_fp16, y = var_42265_to_fp16)[name = tensor("aw_chunk_4141_cast_fp16")]; + tensor var_42268_equation_0 = const()[name = tensor("op_42268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42268_cast_fp16 = einsum(equation = var_42268_equation_0, values = (var_41894_cast_fp16, var_41765_cast_fp16))[name = tensor("op_42268_cast_fp16")]; + tensor var_42269_to_fp16 = const()[name = tensor("op_42269_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4143_cast_fp16 = mul(x = var_42268_cast_fp16, y = var_42269_to_fp16)[name = tensor("aw_chunk_4143_cast_fp16")]; + tensor var_42272_equation_0 = const()[name = tensor("op_42272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42272_cast_fp16 = einsum(equation = var_42272_equation_0, values = (var_41898_cast_fp16, var_41772_cast_fp16))[name = tensor("op_42272_cast_fp16")]; + tensor var_42273_to_fp16 = const()[name = tensor("op_42273_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4145_cast_fp16 = mul(x = var_42272_cast_fp16, y = var_42273_to_fp16)[name = tensor("aw_chunk_4145_cast_fp16")]; + tensor var_42276_equation_0 = const()[name = tensor("op_42276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42276_cast_fp16 = einsum(equation = var_42276_equation_0, values = (var_41898_cast_fp16, var_41779_cast_fp16))[name = tensor("op_42276_cast_fp16")]; + tensor var_42277_to_fp16 = const()[name = tensor("op_42277_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4147_cast_fp16 = mul(x = var_42276_cast_fp16, y = var_42277_to_fp16)[name = tensor("aw_chunk_4147_cast_fp16")]; + tensor var_42280_equation_0 = const()[name = tensor("op_42280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42280_cast_fp16 = einsum(equation = var_42280_equation_0, values = (var_41898_cast_fp16, var_41786_cast_fp16))[name = tensor("op_42280_cast_fp16")]; + tensor var_42281_to_fp16 = const()[name = tensor("op_42281_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4149_cast_fp16 = mul(x = var_42280_cast_fp16, y = var_42281_to_fp16)[name = tensor("aw_chunk_4149_cast_fp16")]; + tensor var_42284_equation_0 = const()[name = tensor("op_42284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42284_cast_fp16 = einsum(equation = var_42284_equation_0, values = (var_41898_cast_fp16, var_41793_cast_fp16))[name = tensor("op_42284_cast_fp16")]; + tensor var_42285_to_fp16 = const()[name = tensor("op_42285_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4151_cast_fp16 = mul(x = var_42284_cast_fp16, y = var_42285_to_fp16)[name = tensor("aw_chunk_4151_cast_fp16")]; + tensor var_42288_equation_0 = const()[name = tensor("op_42288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42288_cast_fp16 = einsum(equation = var_42288_equation_0, values = (var_41902_cast_fp16, var_41800_cast_fp16))[name = tensor("op_42288_cast_fp16")]; + tensor var_42289_to_fp16 = const()[name = tensor("op_42289_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4153_cast_fp16 = mul(x = var_42288_cast_fp16, y = var_42289_to_fp16)[name = tensor("aw_chunk_4153_cast_fp16")]; + tensor var_42292_equation_0 = const()[name = tensor("op_42292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42292_cast_fp16 = einsum(equation = var_42292_equation_0, values = (var_41902_cast_fp16, var_41807_cast_fp16))[name = tensor("op_42292_cast_fp16")]; + tensor var_42293_to_fp16 = const()[name = tensor("op_42293_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4155_cast_fp16 = mul(x = var_42292_cast_fp16, y = var_42293_to_fp16)[name = tensor("aw_chunk_4155_cast_fp16")]; + tensor var_42296_equation_0 = const()[name = tensor("op_42296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42296_cast_fp16 = einsum(equation = var_42296_equation_0, values = (var_41902_cast_fp16, var_41814_cast_fp16))[name = tensor("op_42296_cast_fp16")]; + tensor var_42297_to_fp16 = const()[name = tensor("op_42297_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4157_cast_fp16 = mul(x = var_42296_cast_fp16, y = var_42297_to_fp16)[name = tensor("aw_chunk_4157_cast_fp16")]; + tensor var_42300_equation_0 = const()[name = tensor("op_42300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_42300_cast_fp16 = einsum(equation = var_42300_equation_0, values = (var_41902_cast_fp16, var_41821_cast_fp16))[name = tensor("op_42300_cast_fp16")]; + tensor var_42301_to_fp16 = const()[name = tensor("op_42301_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4159_cast_fp16 = mul(x = var_42300_cast_fp16, y = var_42301_to_fp16)[name = tensor("aw_chunk_4159_cast_fp16")]; + tensor var_42303_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4001_cast_fp16)[name = tensor("op_42303_cast_fp16")]; + tensor var_42304_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4003_cast_fp16)[name = tensor("op_42304_cast_fp16")]; + tensor var_42305_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4005_cast_fp16)[name = tensor("op_42305_cast_fp16")]; + tensor var_42306_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4007_cast_fp16)[name = tensor("op_42306_cast_fp16")]; + tensor var_42307_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4009_cast_fp16)[name = tensor("op_42307_cast_fp16")]; + tensor var_42308_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4011_cast_fp16)[name = tensor("op_42308_cast_fp16")]; + tensor var_42309_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4013_cast_fp16)[name = tensor("op_42309_cast_fp16")]; + tensor var_42310_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4015_cast_fp16)[name = tensor("op_42310_cast_fp16")]; + tensor var_42311_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4017_cast_fp16)[name = tensor("op_42311_cast_fp16")]; + tensor var_42312_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4019_cast_fp16)[name = tensor("op_42312_cast_fp16")]; + tensor var_42313_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4021_cast_fp16)[name = tensor("op_42313_cast_fp16")]; + tensor var_42314_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4023_cast_fp16)[name = tensor("op_42314_cast_fp16")]; + tensor var_42315_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4025_cast_fp16)[name = tensor("op_42315_cast_fp16")]; + tensor var_42316_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4027_cast_fp16)[name = tensor("op_42316_cast_fp16")]; + tensor var_42317_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4029_cast_fp16)[name = tensor("op_42317_cast_fp16")]; + tensor var_42318_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4031_cast_fp16)[name = tensor("op_42318_cast_fp16")]; + tensor var_42319_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4033_cast_fp16)[name = tensor("op_42319_cast_fp16")]; + tensor var_42320_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4035_cast_fp16)[name = tensor("op_42320_cast_fp16")]; + tensor var_42321_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4037_cast_fp16)[name = tensor("op_42321_cast_fp16")]; + tensor var_42322_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4039_cast_fp16)[name = tensor("op_42322_cast_fp16")]; + tensor var_42323_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4041_cast_fp16)[name = tensor("op_42323_cast_fp16")]; + tensor var_42324_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4043_cast_fp16)[name = tensor("op_42324_cast_fp16")]; + tensor var_42325_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4045_cast_fp16)[name = tensor("op_42325_cast_fp16")]; + tensor var_42326_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4047_cast_fp16)[name = tensor("op_42326_cast_fp16")]; + tensor var_42327_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4049_cast_fp16)[name = tensor("op_42327_cast_fp16")]; + tensor var_42328_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4051_cast_fp16)[name = tensor("op_42328_cast_fp16")]; + tensor var_42329_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4053_cast_fp16)[name = tensor("op_42329_cast_fp16")]; + tensor var_42330_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4055_cast_fp16)[name = tensor("op_42330_cast_fp16")]; + tensor var_42331_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4057_cast_fp16)[name = tensor("op_42331_cast_fp16")]; + tensor var_42332_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4059_cast_fp16)[name = tensor("op_42332_cast_fp16")]; + tensor var_42333_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4061_cast_fp16)[name = tensor("op_42333_cast_fp16")]; + tensor var_42334_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4063_cast_fp16)[name = tensor("op_42334_cast_fp16")]; + tensor var_42335_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4065_cast_fp16)[name = tensor("op_42335_cast_fp16")]; + tensor var_42336_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4067_cast_fp16)[name = tensor("op_42336_cast_fp16")]; + tensor var_42337_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4069_cast_fp16)[name = tensor("op_42337_cast_fp16")]; + tensor var_42338_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4071_cast_fp16)[name = tensor("op_42338_cast_fp16")]; + tensor var_42339_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4073_cast_fp16)[name = tensor("op_42339_cast_fp16")]; + tensor var_42340_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4075_cast_fp16)[name = tensor("op_42340_cast_fp16")]; + tensor var_42341_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4077_cast_fp16)[name = tensor("op_42341_cast_fp16")]; + tensor var_42342_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4079_cast_fp16)[name = tensor("op_42342_cast_fp16")]; + tensor var_42343_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4081_cast_fp16)[name = tensor("op_42343_cast_fp16")]; + tensor var_42344_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4083_cast_fp16)[name = tensor("op_42344_cast_fp16")]; + tensor var_42345_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4085_cast_fp16)[name = tensor("op_42345_cast_fp16")]; + tensor var_42346_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4087_cast_fp16)[name = tensor("op_42346_cast_fp16")]; + tensor var_42347_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4089_cast_fp16)[name = tensor("op_42347_cast_fp16")]; + tensor var_42348_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4091_cast_fp16)[name = tensor("op_42348_cast_fp16")]; + tensor var_42349_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4093_cast_fp16)[name = tensor("op_42349_cast_fp16")]; + tensor var_42350_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4095_cast_fp16)[name = tensor("op_42350_cast_fp16")]; + tensor var_42351_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4097_cast_fp16)[name = tensor("op_42351_cast_fp16")]; + tensor var_42352_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4099_cast_fp16)[name = tensor("op_42352_cast_fp16")]; + tensor var_42353_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4101_cast_fp16)[name = tensor("op_42353_cast_fp16")]; + tensor var_42354_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4103_cast_fp16)[name = tensor("op_42354_cast_fp16")]; + tensor var_42355_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4105_cast_fp16)[name = tensor("op_42355_cast_fp16")]; + tensor var_42356_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4107_cast_fp16)[name = tensor("op_42356_cast_fp16")]; + tensor var_42357_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4109_cast_fp16)[name = tensor("op_42357_cast_fp16")]; + tensor var_42358_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4111_cast_fp16)[name = tensor("op_42358_cast_fp16")]; + tensor var_42359_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4113_cast_fp16)[name = tensor("op_42359_cast_fp16")]; + tensor var_42360_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4115_cast_fp16)[name = tensor("op_42360_cast_fp16")]; + tensor var_42361_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4117_cast_fp16)[name = tensor("op_42361_cast_fp16")]; + tensor var_42362_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4119_cast_fp16)[name = tensor("op_42362_cast_fp16")]; + tensor var_42363_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4121_cast_fp16)[name = tensor("op_42363_cast_fp16")]; + tensor var_42364_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4123_cast_fp16)[name = tensor("op_42364_cast_fp16")]; + tensor var_42365_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4125_cast_fp16)[name = tensor("op_42365_cast_fp16")]; + tensor var_42366_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4127_cast_fp16)[name = tensor("op_42366_cast_fp16")]; + tensor var_42367_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4129_cast_fp16)[name = tensor("op_42367_cast_fp16")]; + tensor var_42368_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4131_cast_fp16)[name = tensor("op_42368_cast_fp16")]; + tensor var_42369_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4133_cast_fp16)[name = tensor("op_42369_cast_fp16")]; + tensor var_42370_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4135_cast_fp16)[name = tensor("op_42370_cast_fp16")]; + tensor var_42371_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4137_cast_fp16)[name = tensor("op_42371_cast_fp16")]; + tensor var_42372_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4139_cast_fp16)[name = tensor("op_42372_cast_fp16")]; + tensor var_42373_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4141_cast_fp16)[name = tensor("op_42373_cast_fp16")]; + tensor var_42374_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4143_cast_fp16)[name = tensor("op_42374_cast_fp16")]; + tensor var_42375_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4145_cast_fp16)[name = tensor("op_42375_cast_fp16")]; + tensor var_42376_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4147_cast_fp16)[name = tensor("op_42376_cast_fp16")]; + tensor var_42377_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4149_cast_fp16)[name = tensor("op_42377_cast_fp16")]; + tensor var_42378_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4151_cast_fp16)[name = tensor("op_42378_cast_fp16")]; + tensor var_42379_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4153_cast_fp16)[name = tensor("op_42379_cast_fp16")]; + tensor var_42380_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4155_cast_fp16)[name = tensor("op_42380_cast_fp16")]; + tensor var_42381_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4157_cast_fp16)[name = tensor("op_42381_cast_fp16")]; + tensor var_42382_cast_fp16 = softmax(axis = var_41074, x = aw_chunk_4159_cast_fp16)[name = tensor("op_42382_cast_fp16")]; + tensor var_42384_equation_0 = const()[name = tensor("op_42384_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42384_cast_fp16 = einsum(equation = var_42384_equation_0, values = (var_41904_cast_fp16, var_42303_cast_fp16))[name = tensor("op_42384_cast_fp16")]; + tensor var_42386_equation_0 = const()[name = tensor("op_42386_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42386_cast_fp16 = einsum(equation = var_42386_equation_0, values = (var_41904_cast_fp16, var_42304_cast_fp16))[name = tensor("op_42386_cast_fp16")]; + tensor var_42388_equation_0 = const()[name = tensor("op_42388_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42388_cast_fp16 = einsum(equation = var_42388_equation_0, values = (var_41904_cast_fp16, var_42305_cast_fp16))[name = tensor("op_42388_cast_fp16")]; + tensor var_42390_equation_0 = const()[name = tensor("op_42390_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42390_cast_fp16 = einsum(equation = var_42390_equation_0, values = (var_41904_cast_fp16, var_42306_cast_fp16))[name = tensor("op_42390_cast_fp16")]; + tensor var_42392_equation_0 = const()[name = tensor("op_42392_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42392_cast_fp16 = einsum(equation = var_42392_equation_0, values = (var_41908_cast_fp16, var_42307_cast_fp16))[name = tensor("op_42392_cast_fp16")]; + tensor var_42394_equation_0 = const()[name = tensor("op_42394_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42394_cast_fp16 = einsum(equation = var_42394_equation_0, values = (var_41908_cast_fp16, var_42308_cast_fp16))[name = tensor("op_42394_cast_fp16")]; + tensor var_42396_equation_0 = const()[name = tensor("op_42396_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42396_cast_fp16 = einsum(equation = var_42396_equation_0, values = (var_41908_cast_fp16, var_42309_cast_fp16))[name = tensor("op_42396_cast_fp16")]; + tensor var_42398_equation_0 = const()[name = tensor("op_42398_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42398_cast_fp16 = einsum(equation = var_42398_equation_0, values = (var_41908_cast_fp16, var_42310_cast_fp16))[name = tensor("op_42398_cast_fp16")]; + tensor var_42400_equation_0 = const()[name = tensor("op_42400_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42400_cast_fp16 = einsum(equation = var_42400_equation_0, values = (var_41912_cast_fp16, var_42311_cast_fp16))[name = tensor("op_42400_cast_fp16")]; + tensor var_42402_equation_0 = const()[name = tensor("op_42402_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42402_cast_fp16 = einsum(equation = var_42402_equation_0, values = (var_41912_cast_fp16, var_42312_cast_fp16))[name = tensor("op_42402_cast_fp16")]; + tensor var_42404_equation_0 = const()[name = tensor("op_42404_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42404_cast_fp16 = einsum(equation = var_42404_equation_0, values = (var_41912_cast_fp16, var_42313_cast_fp16))[name = tensor("op_42404_cast_fp16")]; + tensor var_42406_equation_0 = const()[name = tensor("op_42406_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42406_cast_fp16 = einsum(equation = var_42406_equation_0, values = (var_41912_cast_fp16, var_42314_cast_fp16))[name = tensor("op_42406_cast_fp16")]; + tensor var_42408_equation_0 = const()[name = tensor("op_42408_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42408_cast_fp16 = einsum(equation = var_42408_equation_0, values = (var_41916_cast_fp16, var_42315_cast_fp16))[name = tensor("op_42408_cast_fp16")]; + tensor var_42410_equation_0 = const()[name = tensor("op_42410_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42410_cast_fp16 = einsum(equation = var_42410_equation_0, values = (var_41916_cast_fp16, var_42316_cast_fp16))[name = tensor("op_42410_cast_fp16")]; + tensor var_42412_equation_0 = const()[name = tensor("op_42412_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42412_cast_fp16 = einsum(equation = var_42412_equation_0, values = (var_41916_cast_fp16, var_42317_cast_fp16))[name = tensor("op_42412_cast_fp16")]; + tensor var_42414_equation_0 = const()[name = tensor("op_42414_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42414_cast_fp16 = einsum(equation = var_42414_equation_0, values = (var_41916_cast_fp16, var_42318_cast_fp16))[name = tensor("op_42414_cast_fp16")]; + tensor var_42416_equation_0 = const()[name = tensor("op_42416_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42416_cast_fp16 = einsum(equation = var_42416_equation_0, values = (var_41920_cast_fp16, var_42319_cast_fp16))[name = tensor("op_42416_cast_fp16")]; + tensor var_42418_equation_0 = const()[name = tensor("op_42418_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42418_cast_fp16 = einsum(equation = var_42418_equation_0, values = (var_41920_cast_fp16, var_42320_cast_fp16))[name = tensor("op_42418_cast_fp16")]; + tensor var_42420_equation_0 = const()[name = tensor("op_42420_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42420_cast_fp16 = einsum(equation = var_42420_equation_0, values = (var_41920_cast_fp16, var_42321_cast_fp16))[name = tensor("op_42420_cast_fp16")]; + tensor var_42422_equation_0 = const()[name = tensor("op_42422_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42422_cast_fp16 = einsum(equation = var_42422_equation_0, values = (var_41920_cast_fp16, var_42322_cast_fp16))[name = tensor("op_42422_cast_fp16")]; + tensor var_42424_equation_0 = const()[name = tensor("op_42424_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42424_cast_fp16 = einsum(equation = var_42424_equation_0, values = (var_41924_cast_fp16, var_42323_cast_fp16))[name = tensor("op_42424_cast_fp16")]; + tensor var_42426_equation_0 = const()[name = tensor("op_42426_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42426_cast_fp16 = einsum(equation = var_42426_equation_0, values = (var_41924_cast_fp16, var_42324_cast_fp16))[name = tensor("op_42426_cast_fp16")]; + tensor var_42428_equation_0 = const()[name = tensor("op_42428_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42428_cast_fp16 = einsum(equation = var_42428_equation_0, values = (var_41924_cast_fp16, var_42325_cast_fp16))[name = tensor("op_42428_cast_fp16")]; + tensor var_42430_equation_0 = const()[name = tensor("op_42430_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42430_cast_fp16 = einsum(equation = var_42430_equation_0, values = (var_41924_cast_fp16, var_42326_cast_fp16))[name = tensor("op_42430_cast_fp16")]; + tensor var_42432_equation_0 = const()[name = tensor("op_42432_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42432_cast_fp16 = einsum(equation = var_42432_equation_0, values = (var_41928_cast_fp16, var_42327_cast_fp16))[name = tensor("op_42432_cast_fp16")]; + tensor var_42434_equation_0 = const()[name = tensor("op_42434_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42434_cast_fp16 = einsum(equation = var_42434_equation_0, values = (var_41928_cast_fp16, var_42328_cast_fp16))[name = tensor("op_42434_cast_fp16")]; + tensor var_42436_equation_0 = const()[name = tensor("op_42436_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42436_cast_fp16 = einsum(equation = var_42436_equation_0, values = (var_41928_cast_fp16, var_42329_cast_fp16))[name = tensor("op_42436_cast_fp16")]; + tensor var_42438_equation_0 = const()[name = tensor("op_42438_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42438_cast_fp16 = einsum(equation = var_42438_equation_0, values = (var_41928_cast_fp16, var_42330_cast_fp16))[name = tensor("op_42438_cast_fp16")]; + tensor var_42440_equation_0 = const()[name = tensor("op_42440_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42440_cast_fp16 = einsum(equation = var_42440_equation_0, values = (var_41932_cast_fp16, var_42331_cast_fp16))[name = tensor("op_42440_cast_fp16")]; + tensor var_42442_equation_0 = const()[name = tensor("op_42442_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42442_cast_fp16 = einsum(equation = var_42442_equation_0, values = (var_41932_cast_fp16, var_42332_cast_fp16))[name = tensor("op_42442_cast_fp16")]; + tensor var_42444_equation_0 = const()[name = tensor("op_42444_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42444_cast_fp16 = einsum(equation = var_42444_equation_0, values = (var_41932_cast_fp16, var_42333_cast_fp16))[name = tensor("op_42444_cast_fp16")]; + tensor var_42446_equation_0 = const()[name = tensor("op_42446_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42446_cast_fp16 = einsum(equation = var_42446_equation_0, values = (var_41932_cast_fp16, var_42334_cast_fp16))[name = tensor("op_42446_cast_fp16")]; + tensor var_42448_equation_0 = const()[name = tensor("op_42448_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42448_cast_fp16 = einsum(equation = var_42448_equation_0, values = (var_41936_cast_fp16, var_42335_cast_fp16))[name = tensor("op_42448_cast_fp16")]; + tensor var_42450_equation_0 = const()[name = tensor("op_42450_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42450_cast_fp16 = einsum(equation = var_42450_equation_0, values = (var_41936_cast_fp16, var_42336_cast_fp16))[name = tensor("op_42450_cast_fp16")]; + tensor var_42452_equation_0 = const()[name = tensor("op_42452_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42452_cast_fp16 = einsum(equation = var_42452_equation_0, values = (var_41936_cast_fp16, var_42337_cast_fp16))[name = tensor("op_42452_cast_fp16")]; + tensor var_42454_equation_0 = const()[name = tensor("op_42454_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42454_cast_fp16 = einsum(equation = var_42454_equation_0, values = (var_41936_cast_fp16, var_42338_cast_fp16))[name = tensor("op_42454_cast_fp16")]; + tensor var_42456_equation_0 = const()[name = tensor("op_42456_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42456_cast_fp16 = einsum(equation = var_42456_equation_0, values = (var_41940_cast_fp16, var_42339_cast_fp16))[name = tensor("op_42456_cast_fp16")]; + tensor var_42458_equation_0 = const()[name = tensor("op_42458_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42458_cast_fp16 = einsum(equation = var_42458_equation_0, values = (var_41940_cast_fp16, var_42340_cast_fp16))[name = tensor("op_42458_cast_fp16")]; + tensor var_42460_equation_0 = const()[name = tensor("op_42460_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42460_cast_fp16 = einsum(equation = var_42460_equation_0, values = (var_41940_cast_fp16, var_42341_cast_fp16))[name = tensor("op_42460_cast_fp16")]; + tensor var_42462_equation_0 = const()[name = tensor("op_42462_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42462_cast_fp16 = einsum(equation = var_42462_equation_0, values = (var_41940_cast_fp16, var_42342_cast_fp16))[name = tensor("op_42462_cast_fp16")]; + tensor var_42464_equation_0 = const()[name = tensor("op_42464_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42464_cast_fp16 = einsum(equation = var_42464_equation_0, values = (var_41944_cast_fp16, var_42343_cast_fp16))[name = tensor("op_42464_cast_fp16")]; + tensor var_42466_equation_0 = const()[name = tensor("op_42466_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42466_cast_fp16 = einsum(equation = var_42466_equation_0, values = (var_41944_cast_fp16, var_42344_cast_fp16))[name = tensor("op_42466_cast_fp16")]; + tensor var_42468_equation_0 = const()[name = tensor("op_42468_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42468_cast_fp16 = einsum(equation = var_42468_equation_0, values = (var_41944_cast_fp16, var_42345_cast_fp16))[name = tensor("op_42468_cast_fp16")]; + tensor var_42470_equation_0 = const()[name = tensor("op_42470_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42470_cast_fp16 = einsum(equation = var_42470_equation_0, values = (var_41944_cast_fp16, var_42346_cast_fp16))[name = tensor("op_42470_cast_fp16")]; + tensor var_42472_equation_0 = const()[name = tensor("op_42472_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42472_cast_fp16 = einsum(equation = var_42472_equation_0, values = (var_41948_cast_fp16, var_42347_cast_fp16))[name = tensor("op_42472_cast_fp16")]; + tensor var_42474_equation_0 = const()[name = tensor("op_42474_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42474_cast_fp16 = einsum(equation = var_42474_equation_0, values = (var_41948_cast_fp16, var_42348_cast_fp16))[name = tensor("op_42474_cast_fp16")]; + tensor var_42476_equation_0 = const()[name = tensor("op_42476_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42476_cast_fp16 = einsum(equation = var_42476_equation_0, values = (var_41948_cast_fp16, var_42349_cast_fp16))[name = tensor("op_42476_cast_fp16")]; + tensor var_42478_equation_0 = const()[name = tensor("op_42478_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42478_cast_fp16 = einsum(equation = var_42478_equation_0, values = (var_41948_cast_fp16, var_42350_cast_fp16))[name = tensor("op_42478_cast_fp16")]; + tensor var_42480_equation_0 = const()[name = tensor("op_42480_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42480_cast_fp16 = einsum(equation = var_42480_equation_0, values = (var_41952_cast_fp16, var_42351_cast_fp16))[name = tensor("op_42480_cast_fp16")]; + tensor var_42482_equation_0 = const()[name = tensor("op_42482_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42482_cast_fp16 = einsum(equation = var_42482_equation_0, values = (var_41952_cast_fp16, var_42352_cast_fp16))[name = tensor("op_42482_cast_fp16")]; + tensor var_42484_equation_0 = const()[name = tensor("op_42484_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42484_cast_fp16 = einsum(equation = var_42484_equation_0, values = (var_41952_cast_fp16, var_42353_cast_fp16))[name = tensor("op_42484_cast_fp16")]; + tensor var_42486_equation_0 = const()[name = tensor("op_42486_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42486_cast_fp16 = einsum(equation = var_42486_equation_0, values = (var_41952_cast_fp16, var_42354_cast_fp16))[name = tensor("op_42486_cast_fp16")]; + tensor var_42488_equation_0 = const()[name = tensor("op_42488_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42488_cast_fp16 = einsum(equation = var_42488_equation_0, values = (var_41956_cast_fp16, var_42355_cast_fp16))[name = tensor("op_42488_cast_fp16")]; + tensor var_42490_equation_0 = const()[name = tensor("op_42490_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42490_cast_fp16 = einsum(equation = var_42490_equation_0, values = (var_41956_cast_fp16, var_42356_cast_fp16))[name = tensor("op_42490_cast_fp16")]; + tensor var_42492_equation_0 = const()[name = tensor("op_42492_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42492_cast_fp16 = einsum(equation = var_42492_equation_0, values = (var_41956_cast_fp16, var_42357_cast_fp16))[name = tensor("op_42492_cast_fp16")]; + tensor var_42494_equation_0 = const()[name = tensor("op_42494_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42494_cast_fp16 = einsum(equation = var_42494_equation_0, values = (var_41956_cast_fp16, var_42358_cast_fp16))[name = tensor("op_42494_cast_fp16")]; + tensor var_42496_equation_0 = const()[name = tensor("op_42496_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42496_cast_fp16 = einsum(equation = var_42496_equation_0, values = (var_41960_cast_fp16, var_42359_cast_fp16))[name = tensor("op_42496_cast_fp16")]; + tensor var_42498_equation_0 = const()[name = tensor("op_42498_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42498_cast_fp16 = einsum(equation = var_42498_equation_0, values = (var_41960_cast_fp16, var_42360_cast_fp16))[name = tensor("op_42498_cast_fp16")]; + tensor var_42500_equation_0 = const()[name = tensor("op_42500_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42500_cast_fp16 = einsum(equation = var_42500_equation_0, values = (var_41960_cast_fp16, var_42361_cast_fp16))[name = tensor("op_42500_cast_fp16")]; + tensor var_42502_equation_0 = const()[name = tensor("op_42502_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42502_cast_fp16 = einsum(equation = var_42502_equation_0, values = (var_41960_cast_fp16, var_42362_cast_fp16))[name = tensor("op_42502_cast_fp16")]; + tensor var_42504_equation_0 = const()[name = tensor("op_42504_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42504_cast_fp16 = einsum(equation = var_42504_equation_0, values = (var_41964_cast_fp16, var_42363_cast_fp16))[name = tensor("op_42504_cast_fp16")]; + tensor var_42506_equation_0 = const()[name = tensor("op_42506_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42506_cast_fp16 = einsum(equation = var_42506_equation_0, values = (var_41964_cast_fp16, var_42364_cast_fp16))[name = tensor("op_42506_cast_fp16")]; + tensor var_42508_equation_0 = const()[name = tensor("op_42508_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42508_cast_fp16 = einsum(equation = var_42508_equation_0, values = (var_41964_cast_fp16, var_42365_cast_fp16))[name = tensor("op_42508_cast_fp16")]; + tensor var_42510_equation_0 = const()[name = tensor("op_42510_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42510_cast_fp16 = einsum(equation = var_42510_equation_0, values = (var_41964_cast_fp16, var_42366_cast_fp16))[name = tensor("op_42510_cast_fp16")]; + tensor var_42512_equation_0 = const()[name = tensor("op_42512_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42512_cast_fp16 = einsum(equation = var_42512_equation_0, values = (var_41968_cast_fp16, var_42367_cast_fp16))[name = tensor("op_42512_cast_fp16")]; + tensor var_42514_equation_0 = const()[name = tensor("op_42514_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42514_cast_fp16 = einsum(equation = var_42514_equation_0, values = (var_41968_cast_fp16, var_42368_cast_fp16))[name = tensor("op_42514_cast_fp16")]; + tensor var_42516_equation_0 = const()[name = tensor("op_42516_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42516_cast_fp16 = einsum(equation = var_42516_equation_0, values = (var_41968_cast_fp16, var_42369_cast_fp16))[name = tensor("op_42516_cast_fp16")]; + tensor var_42518_equation_0 = const()[name = tensor("op_42518_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42518_cast_fp16 = einsum(equation = var_42518_equation_0, values = (var_41968_cast_fp16, var_42370_cast_fp16))[name = tensor("op_42518_cast_fp16")]; + tensor var_42520_equation_0 = const()[name = tensor("op_42520_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42520_cast_fp16 = einsum(equation = var_42520_equation_0, values = (var_41972_cast_fp16, var_42371_cast_fp16))[name = tensor("op_42520_cast_fp16")]; + tensor var_42522_equation_0 = const()[name = tensor("op_42522_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42522_cast_fp16 = einsum(equation = var_42522_equation_0, values = (var_41972_cast_fp16, var_42372_cast_fp16))[name = tensor("op_42522_cast_fp16")]; + tensor var_42524_equation_0 = const()[name = tensor("op_42524_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42524_cast_fp16 = einsum(equation = var_42524_equation_0, values = (var_41972_cast_fp16, var_42373_cast_fp16))[name = tensor("op_42524_cast_fp16")]; + tensor var_42526_equation_0 = const()[name = tensor("op_42526_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42526_cast_fp16 = einsum(equation = var_42526_equation_0, values = (var_41972_cast_fp16, var_42374_cast_fp16))[name = tensor("op_42526_cast_fp16")]; + tensor var_42528_equation_0 = const()[name = tensor("op_42528_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42528_cast_fp16 = einsum(equation = var_42528_equation_0, values = (var_41976_cast_fp16, var_42375_cast_fp16))[name = tensor("op_42528_cast_fp16")]; + tensor var_42530_equation_0 = const()[name = tensor("op_42530_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42530_cast_fp16 = einsum(equation = var_42530_equation_0, values = (var_41976_cast_fp16, var_42376_cast_fp16))[name = tensor("op_42530_cast_fp16")]; + tensor var_42532_equation_0 = const()[name = tensor("op_42532_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42532_cast_fp16 = einsum(equation = var_42532_equation_0, values = (var_41976_cast_fp16, var_42377_cast_fp16))[name = tensor("op_42532_cast_fp16")]; + tensor var_42534_equation_0 = const()[name = tensor("op_42534_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42534_cast_fp16 = einsum(equation = var_42534_equation_0, values = (var_41976_cast_fp16, var_42378_cast_fp16))[name = tensor("op_42534_cast_fp16")]; + tensor var_42536_equation_0 = const()[name = tensor("op_42536_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42536_cast_fp16 = einsum(equation = var_42536_equation_0, values = (var_41980_cast_fp16, var_42379_cast_fp16))[name = tensor("op_42536_cast_fp16")]; + tensor var_42538_equation_0 = const()[name = tensor("op_42538_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42538_cast_fp16 = einsum(equation = var_42538_equation_0, values = (var_41980_cast_fp16, var_42380_cast_fp16))[name = tensor("op_42538_cast_fp16")]; + tensor var_42540_equation_0 = const()[name = tensor("op_42540_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42540_cast_fp16 = einsum(equation = var_42540_equation_0, values = (var_41980_cast_fp16, var_42381_cast_fp16))[name = tensor("op_42540_cast_fp16")]; + tensor var_42542_equation_0 = const()[name = tensor("op_42542_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_42542_cast_fp16 = einsum(equation = var_42542_equation_0, values = (var_41980_cast_fp16, var_42382_cast_fp16))[name = tensor("op_42542_cast_fp16")]; + tensor var_42544_interleave_0 = const()[name = tensor("op_42544_interleave_0"), val = tensor(false)]; + tensor var_42544_cast_fp16 = concat(axis = var_41049, interleave = var_42544_interleave_0, values = (var_42384_cast_fp16, var_42386_cast_fp16, var_42388_cast_fp16, var_42390_cast_fp16))[name = tensor("op_42544_cast_fp16")]; + tensor var_42546_interleave_0 = const()[name = tensor("op_42546_interleave_0"), val = tensor(false)]; + tensor var_42546_cast_fp16 = concat(axis = var_41049, interleave = var_42546_interleave_0, values = (var_42392_cast_fp16, var_42394_cast_fp16, var_42396_cast_fp16, var_42398_cast_fp16))[name = tensor("op_42546_cast_fp16")]; + tensor var_42548_interleave_0 = const()[name = tensor("op_42548_interleave_0"), val = tensor(false)]; + tensor var_42548_cast_fp16 = concat(axis = var_41049, interleave = var_42548_interleave_0, values = (var_42400_cast_fp16, var_42402_cast_fp16, var_42404_cast_fp16, var_42406_cast_fp16))[name = tensor("op_42548_cast_fp16")]; + tensor var_42550_interleave_0 = const()[name = tensor("op_42550_interleave_0"), val = tensor(false)]; + tensor var_42550_cast_fp16 = concat(axis = var_41049, interleave = var_42550_interleave_0, values = (var_42408_cast_fp16, var_42410_cast_fp16, var_42412_cast_fp16, var_42414_cast_fp16))[name = tensor("op_42550_cast_fp16")]; + tensor var_42552_interleave_0 = const()[name = tensor("op_42552_interleave_0"), val = tensor(false)]; + tensor var_42552_cast_fp16 = concat(axis = var_41049, interleave = var_42552_interleave_0, values = (var_42416_cast_fp16, var_42418_cast_fp16, var_42420_cast_fp16, var_42422_cast_fp16))[name = tensor("op_42552_cast_fp16")]; + tensor var_42554_interleave_0 = const()[name = tensor("op_42554_interleave_0"), val = tensor(false)]; + tensor var_42554_cast_fp16 = concat(axis = var_41049, interleave = var_42554_interleave_0, values = (var_42424_cast_fp16, var_42426_cast_fp16, var_42428_cast_fp16, var_42430_cast_fp16))[name = tensor("op_42554_cast_fp16")]; + tensor var_42556_interleave_0 = const()[name = tensor("op_42556_interleave_0"), val = tensor(false)]; + tensor var_42556_cast_fp16 = concat(axis = var_41049, interleave = var_42556_interleave_0, values = (var_42432_cast_fp16, var_42434_cast_fp16, var_42436_cast_fp16, var_42438_cast_fp16))[name = tensor("op_42556_cast_fp16")]; + tensor var_42558_interleave_0 = const()[name = tensor("op_42558_interleave_0"), val = tensor(false)]; + tensor var_42558_cast_fp16 = concat(axis = var_41049, interleave = var_42558_interleave_0, values = (var_42440_cast_fp16, var_42442_cast_fp16, var_42444_cast_fp16, var_42446_cast_fp16))[name = tensor("op_42558_cast_fp16")]; + tensor var_42560_interleave_0 = const()[name = tensor("op_42560_interleave_0"), val = tensor(false)]; + tensor var_42560_cast_fp16 = concat(axis = var_41049, interleave = var_42560_interleave_0, values = (var_42448_cast_fp16, var_42450_cast_fp16, var_42452_cast_fp16, var_42454_cast_fp16))[name = tensor("op_42560_cast_fp16")]; + tensor var_42562_interleave_0 = const()[name = tensor("op_42562_interleave_0"), val = tensor(false)]; + tensor var_42562_cast_fp16 = concat(axis = var_41049, interleave = var_42562_interleave_0, values = (var_42456_cast_fp16, var_42458_cast_fp16, var_42460_cast_fp16, var_42462_cast_fp16))[name = tensor("op_42562_cast_fp16")]; + tensor var_42564_interleave_0 = const()[name = tensor("op_42564_interleave_0"), val = tensor(false)]; + tensor var_42564_cast_fp16 = concat(axis = var_41049, interleave = var_42564_interleave_0, values = (var_42464_cast_fp16, var_42466_cast_fp16, var_42468_cast_fp16, var_42470_cast_fp16))[name = tensor("op_42564_cast_fp16")]; + tensor var_42566_interleave_0 = const()[name = tensor("op_42566_interleave_0"), val = tensor(false)]; + tensor var_42566_cast_fp16 = concat(axis = var_41049, interleave = var_42566_interleave_0, values = (var_42472_cast_fp16, var_42474_cast_fp16, var_42476_cast_fp16, var_42478_cast_fp16))[name = tensor("op_42566_cast_fp16")]; + tensor var_42568_interleave_0 = const()[name = tensor("op_42568_interleave_0"), val = tensor(false)]; + tensor var_42568_cast_fp16 = concat(axis = var_41049, interleave = var_42568_interleave_0, values = (var_42480_cast_fp16, var_42482_cast_fp16, var_42484_cast_fp16, var_42486_cast_fp16))[name = tensor("op_42568_cast_fp16")]; + tensor var_42570_interleave_0 = const()[name = tensor("op_42570_interleave_0"), val = tensor(false)]; + tensor var_42570_cast_fp16 = concat(axis = var_41049, interleave = var_42570_interleave_0, values = (var_42488_cast_fp16, var_42490_cast_fp16, var_42492_cast_fp16, var_42494_cast_fp16))[name = tensor("op_42570_cast_fp16")]; + tensor var_42572_interleave_0 = const()[name = tensor("op_42572_interleave_0"), val = tensor(false)]; + tensor var_42572_cast_fp16 = concat(axis = var_41049, interleave = var_42572_interleave_0, values = (var_42496_cast_fp16, var_42498_cast_fp16, var_42500_cast_fp16, var_42502_cast_fp16))[name = tensor("op_42572_cast_fp16")]; + tensor var_42574_interleave_0 = const()[name = tensor("op_42574_interleave_0"), val = tensor(false)]; + tensor var_42574_cast_fp16 = concat(axis = var_41049, interleave = var_42574_interleave_0, values = (var_42504_cast_fp16, var_42506_cast_fp16, var_42508_cast_fp16, var_42510_cast_fp16))[name = tensor("op_42574_cast_fp16")]; + tensor var_42576_interleave_0 = const()[name = tensor("op_42576_interleave_0"), val = tensor(false)]; + tensor var_42576_cast_fp16 = concat(axis = var_41049, interleave = var_42576_interleave_0, values = (var_42512_cast_fp16, var_42514_cast_fp16, var_42516_cast_fp16, var_42518_cast_fp16))[name = tensor("op_42576_cast_fp16")]; + tensor var_42578_interleave_0 = const()[name = tensor("op_42578_interleave_0"), val = tensor(false)]; + tensor var_42578_cast_fp16 = concat(axis = var_41049, interleave = var_42578_interleave_0, values = (var_42520_cast_fp16, var_42522_cast_fp16, var_42524_cast_fp16, var_42526_cast_fp16))[name = tensor("op_42578_cast_fp16")]; + tensor var_42580_interleave_0 = const()[name = tensor("op_42580_interleave_0"), val = tensor(false)]; + tensor var_42580_cast_fp16 = concat(axis = var_41049, interleave = var_42580_interleave_0, values = (var_42528_cast_fp16, var_42530_cast_fp16, var_42532_cast_fp16, var_42534_cast_fp16))[name = tensor("op_42580_cast_fp16")]; + tensor var_42582_interleave_0 = const()[name = tensor("op_42582_interleave_0"), val = tensor(false)]; + tensor var_42582_cast_fp16 = concat(axis = var_41049, interleave = var_42582_interleave_0, values = (var_42536_cast_fp16, var_42538_cast_fp16, var_42540_cast_fp16, var_42542_cast_fp16))[name = tensor("op_42582_cast_fp16")]; + tensor input_507_interleave_0 = const()[name = tensor("input_507_interleave_0"), val = tensor(false)]; + tensor input_507_cast_fp16 = concat(axis = var_41074, interleave = input_507_interleave_0, values = (var_42544_cast_fp16, var_42546_cast_fp16, var_42548_cast_fp16, var_42550_cast_fp16, var_42552_cast_fp16, var_42554_cast_fp16, var_42556_cast_fp16, var_42558_cast_fp16, var_42560_cast_fp16, var_42562_cast_fp16, var_42564_cast_fp16, var_42566_cast_fp16, var_42568_cast_fp16, var_42570_cast_fp16, var_42572_cast_fp16, var_42574_cast_fp16, var_42576_cast_fp16, var_42578_cast_fp16, var_42580_cast_fp16, var_42582_cast_fp16))[name = tensor("input_507_cast_fp16")]; + tensor var_42590 = const()[name = tensor("op_42590"), val = tensor([1, 1])]; + tensor var_42592 = const()[name = tensor("op_42592"), val = tensor([1, 1])]; + tensor pretrained_out_307_pad_type_0 = const()[name = tensor("pretrained_out_307_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_307_pad_0 = const()[name = tensor("pretrained_out_307_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282043072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282862336))), name = tensor("layers_25_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_25_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_25_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282862464)))]; + tensor pretrained_out_307_cast_fp16 = conv(bias = layers_25_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_42592, groups = var_41074, pad = pretrained_out_307_pad_0, pad_type = pretrained_out_307_pad_type_0, strides = var_42590, weight = layers_25_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_507_cast_fp16)[name = tensor("pretrained_out_307_cast_fp16")]; + tensor var_42596 = const()[name = tensor("op_42596"), val = tensor([1, 1])]; + tensor var_42598 = const()[name = tensor("op_42598"), val = tensor([1, 1])]; + tensor input_509_pad_type_0 = const()[name = tensor("input_509_pad_type_0"), val = tensor("custom")]; + tensor input_509_pad_0 = const()[name = tensor("input_509_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_25_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282865088)))]; + tensor input_509_cast_fp16 = conv(dilations = var_42598, groups = var_41074, pad = input_509_pad_0, pad_type = input_509_pad_type_0, strides = var_42596, weight = layers_25_self_attn_o_proj_loraA_weight_to_fp16, x = input_507_cast_fp16)[name = tensor("input_509_cast_fp16")]; + tensor var_42602 = const()[name = tensor("op_42602"), val = tensor([1, 1])]; + tensor var_42604 = const()[name = tensor("op_42604"), val = tensor([1, 1])]; + tensor lora_out_613_pad_type_0 = const()[name = tensor("lora_out_613_pad_type_0"), val = tensor("custom")]; + tensor lora_out_613_pad_0 = const()[name = tensor("lora_out_613_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_615_weight_0_to_fp16 = const()[name = tensor("lora_out_615_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282906112)))]; + tensor lora_out_615_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_42604, groups = var_41074, pad = lora_out_613_pad_0, pad_type = lora_out_613_pad_type_0, strides = var_42602, weight = lora_out_615_weight_0_to_fp16, x = input_509_cast_fp16)[name = tensor("lora_out_615_cast_fp16")]; + tensor obj_103_cast_fp16 = add(x = pretrained_out_307_cast_fp16, y = lora_out_615_cast_fp16)[name = tensor("obj_103_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = obj_103_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; + tensor var_42613 = const()[name = tensor("op_42613"), val = tensor([1])]; + tensor channels_mean_103_cast_fp16 = reduce_mean(axes = var_42613, keep_dims = var_41075, x = inputs_103_cast_fp16)[name = tensor("channels_mean_103_cast_fp16")]; + tensor zero_mean_103_cast_fp16 = sub(x = inputs_103_cast_fp16, y = channels_mean_103_cast_fp16)[name = tensor("zero_mean_103_cast_fp16")]; + tensor zero_mean_sq_103_cast_fp16 = mul(x = zero_mean_103_cast_fp16, y = zero_mean_103_cast_fp16)[name = tensor("zero_mean_sq_103_cast_fp16")]; + tensor var_42617 = const()[name = tensor("op_42617"), val = tensor([1])]; + tensor var_42618_cast_fp16 = reduce_mean(axes = var_42617, keep_dims = var_41075, x = zero_mean_sq_103_cast_fp16)[name = tensor("op_42618_cast_fp16")]; + tensor var_42619_to_fp16 = const()[name = tensor("op_42619_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_42620_cast_fp16 = add(x = var_42618_cast_fp16, y = var_42619_to_fp16)[name = tensor("op_42620_cast_fp16")]; + tensor denom_103_epsilon_0 = const()[name = tensor("denom_103_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_103_cast_fp16 = rsqrt(epsilon = denom_103_epsilon_0, x = var_42620_cast_fp16)[name = tensor("denom_103_cast_fp16")]; + tensor out_103_cast_fp16 = mul(x = zero_mean_103_cast_fp16, y = denom_103_cast_fp16)[name = tensor("out_103_cast_fp16")]; + tensor input_511_gamma_0_to_fp16 = const()[name = tensor("input_511_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282947136)))]; + tensor input_511_beta_0_to_fp16 = const()[name = tensor("input_511_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282949760)))]; + tensor input_511_epsilon_0_to_fp16 = const()[name = tensor("input_511_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_511_cast_fp16 = batch_norm(beta = input_511_beta_0_to_fp16, epsilon = input_511_epsilon_0_to_fp16, gamma = input_511_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_103_cast_fp16)[name = tensor("input_511_cast_fp16")]; + tensor var_42634 = const()[name = tensor("op_42634"), val = tensor([1, 1])]; + tensor var_42636 = const()[name = tensor("op_42636"), val = tensor([1, 1])]; + tensor pretrained_out_309_pad_type_0 = const()[name = tensor("pretrained_out_309_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_309_pad_0 = const()[name = tensor("pretrained_out_309_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(282952384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286229248))), name = tensor("layers_25_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_25_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_25_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286229376)))]; + tensor pretrained_out_309_cast_fp16 = conv(bias = layers_25_fc1_pretrained_bias_to_fp16, dilations = var_42636, groups = var_41074, pad = pretrained_out_309_pad_0, pad_type = pretrained_out_309_pad_type_0, strides = var_42634, weight = layers_25_fc1_pretrained_weight_to_fp16_palettized, x = input_511_cast_fp16)[name = tensor("pretrained_out_309_cast_fp16")]; + tensor var_42640 = const()[name = tensor("op_42640"), val = tensor([1, 1])]; + tensor var_42642 = const()[name = tensor("op_42642"), val = tensor([1, 1])]; + tensor input_513_pad_type_0 = const()[name = tensor("input_513_pad_type_0"), val = tensor("custom")]; + tensor input_513_pad_0 = const()[name = tensor("input_513_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_25_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286239680)))]; + tensor input_513_cast_fp16 = conv(dilations = var_42642, groups = var_41074, pad = input_513_pad_0, pad_type = input_513_pad_type_0, strides = var_42640, weight = layers_25_fc1_loraA_weight_to_fp16, x = input_511_cast_fp16)[name = tensor("input_513_cast_fp16")]; + tensor var_42646 = const()[name = tensor("op_42646"), val = tensor([1, 1])]; + tensor var_42648 = const()[name = tensor("op_42648"), val = tensor([1, 1])]; + tensor lora_out_617_pad_type_0 = const()[name = tensor("lora_out_617_pad_type_0"), val = tensor("custom")]; + tensor lora_out_617_pad_0 = const()[name = tensor("lora_out_617_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_619_weight_0_to_fp16 = const()[name = tensor("lora_out_619_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286280704)))]; + tensor lora_out_619_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_42648, groups = var_41074, pad = lora_out_617_pad_0, pad_type = lora_out_617_pad_type_0, strides = var_42646, weight = lora_out_619_weight_0_to_fp16, x = input_513_cast_fp16)[name = tensor("lora_out_619_cast_fp16")]; + tensor input_515_cast_fp16 = add(x = pretrained_out_309_cast_fp16, y = lora_out_619_cast_fp16)[name = tensor("input_515_cast_fp16")]; + tensor input_517_mode_0 = const()[name = tensor("input_517_mode_0"), val = tensor("EXACT")]; + tensor input_517_cast_fp16 = gelu(mode = input_517_mode_0, x = input_515_cast_fp16)[name = tensor("input_517_cast_fp16")]; + tensor var_42660 = const()[name = tensor("op_42660"), val = tensor([1, 1])]; + tensor var_42662 = const()[name = tensor("op_42662"), val = tensor([1, 1])]; + tensor pretrained_out_311_pad_type_0 = const()[name = tensor("pretrained_out_311_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_311_pad_0 = const()[name = tensor("pretrained_out_311_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286444608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289721472))), name = tensor("layers_25_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_25_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_25_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289721600)))]; + tensor pretrained_out_311_cast_fp16 = conv(bias = layers_25_fc2_pretrained_bias_to_fp16, dilations = var_42662, groups = var_41074, pad = pretrained_out_311_pad_0, pad_type = pretrained_out_311_pad_type_0, strides = var_42660, weight = layers_25_fc2_pretrained_weight_to_fp16_palettized, x = input_517_cast_fp16)[name = tensor("pretrained_out_311_cast_fp16")]; + tensor var_42666 = const()[name = tensor("op_42666"), val = tensor([1, 1])]; + tensor var_42668 = const()[name = tensor("op_42668"), val = tensor([1, 1])]; + tensor input_519_pad_type_0 = const()[name = tensor("input_519_pad_type_0"), val = tensor("custom")]; + tensor input_519_pad_0 = const()[name = tensor("input_519_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_25_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_25_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289724224)))]; + tensor input_519_cast_fp16 = conv(dilations = var_42668, groups = var_41074, pad = input_519_pad_0, pad_type = input_519_pad_type_0, strides = var_42666, weight = layers_25_fc2_loraA_weight_to_fp16, x = input_517_cast_fp16)[name = tensor("input_519_cast_fp16")]; + tensor var_42672 = const()[name = tensor("op_42672"), val = tensor([1, 1])]; + tensor var_42674 = const()[name = tensor("op_42674"), val = tensor([1, 1])]; + tensor lora_out_621_pad_type_0 = const()[name = tensor("lora_out_621_pad_type_0"), val = tensor("custom")]; + tensor lora_out_621_pad_0 = const()[name = tensor("lora_out_621_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_623_weight_0_to_fp16 = const()[name = tensor("lora_out_623_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289888128)))]; + tensor lora_out_623_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_42674, groups = var_41074, pad = lora_out_621_pad_0, pad_type = lora_out_621_pad_type_0, strides = var_42672, weight = lora_out_623_weight_0_to_fp16, x = input_519_cast_fp16)[name = tensor("lora_out_623_cast_fp16")]; + tensor hidden_states_55_cast_fp16 = add(x = pretrained_out_311_cast_fp16, y = lora_out_623_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = hidden_states_55_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; + tensor var_42684 = const()[name = tensor("op_42684"), val = tensor(3)]; + tensor var_42709 = const()[name = tensor("op_42709"), val = tensor(1)]; + tensor var_42710 = const()[name = tensor("op_42710"), val = tensor(true)]; + tensor var_42720 = const()[name = tensor("op_42720"), val = tensor([1])]; + tensor channels_mean_105_cast_fp16 = reduce_mean(axes = var_42720, keep_dims = var_42710, x = inputs_105_cast_fp16)[name = tensor("channels_mean_105_cast_fp16")]; + tensor zero_mean_105_cast_fp16 = sub(x = inputs_105_cast_fp16, y = channels_mean_105_cast_fp16)[name = tensor("zero_mean_105_cast_fp16")]; + tensor zero_mean_sq_105_cast_fp16 = mul(x = zero_mean_105_cast_fp16, y = zero_mean_105_cast_fp16)[name = tensor("zero_mean_sq_105_cast_fp16")]; + tensor var_42724 = const()[name = tensor("op_42724"), val = tensor([1])]; + tensor var_42725_cast_fp16 = reduce_mean(axes = var_42724, keep_dims = var_42710, x = zero_mean_sq_105_cast_fp16)[name = tensor("op_42725_cast_fp16")]; + tensor var_42726_to_fp16 = const()[name = tensor("op_42726_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_42727_cast_fp16 = add(x = var_42725_cast_fp16, y = var_42726_to_fp16)[name = tensor("op_42727_cast_fp16")]; + tensor denom_105_epsilon_0 = const()[name = tensor("denom_105_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_105_cast_fp16 = rsqrt(epsilon = denom_105_epsilon_0, x = var_42727_cast_fp16)[name = tensor("denom_105_cast_fp16")]; + tensor out_105_cast_fp16 = mul(x = zero_mean_105_cast_fp16, y = denom_105_cast_fp16)[name = tensor("out_105_cast_fp16")]; + tensor obj_105_gamma_0_to_fp16 = const()[name = tensor("obj_105_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289929152)))]; + tensor obj_105_beta_0_to_fp16 = const()[name = tensor("obj_105_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289931776)))]; + tensor obj_105_epsilon_0_to_fp16 = const()[name = tensor("obj_105_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_105_cast_fp16 = batch_norm(beta = obj_105_beta_0_to_fp16, epsilon = obj_105_epsilon_0_to_fp16, gamma = obj_105_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_105_cast_fp16)[name = tensor("obj_105_cast_fp16")]; + tensor var_42745 = const()[name = tensor("op_42745"), val = tensor([1, 1])]; + tensor var_42747 = const()[name = tensor("op_42747"), val = tensor([1, 1])]; + tensor pretrained_out_313_pad_type_0 = const()[name = tensor("pretrained_out_313_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_313_pad_0 = const()[name = tensor("pretrained_out_313_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289934400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290753664))), name = tensor("layers_26_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_26_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290753792)))]; + tensor pretrained_out_313_cast_fp16 = conv(bias = layers_26_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_42747, groups = var_42709, pad = pretrained_out_313_pad_0, pad_type = pretrained_out_313_pad_type_0, strides = var_42745, weight = layers_26_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_105_cast_fp16)[name = tensor("pretrained_out_313_cast_fp16")]; + tensor var_42751 = const()[name = tensor("op_42751"), val = tensor([1, 1])]; + tensor var_42753 = const()[name = tensor("op_42753"), val = tensor([1, 1])]; + tensor input_521_pad_type_0 = const()[name = tensor("input_521_pad_type_0"), val = tensor("custom")]; + tensor input_521_pad_0 = const()[name = tensor("input_521_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290756416)))]; + tensor input_521_cast_fp16 = conv(dilations = var_42753, groups = var_42709, pad = input_521_pad_0, pad_type = input_521_pad_type_0, strides = var_42751, weight = layers_26_self_attn_q_proj_loraA_weight_to_fp16, x = obj_105_cast_fp16)[name = tensor("input_521_cast_fp16")]; + tensor var_42757 = const()[name = tensor("op_42757"), val = tensor([1, 1])]; + tensor var_42759 = const()[name = tensor("op_42759"), val = tensor([1, 1])]; + tensor lora_out_625_pad_type_0 = const()[name = tensor("lora_out_625_pad_type_0"), val = tensor("custom")]; + tensor lora_out_625_pad_0 = const()[name = tensor("lora_out_625_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_627_weight_0_to_fp16 = const()[name = tensor("lora_out_627_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290797440)))]; + tensor lora_out_627_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_42759, groups = var_42709, pad = lora_out_625_pad_0, pad_type = lora_out_625_pad_type_0, strides = var_42757, weight = lora_out_627_weight_0_to_fp16, x = input_521_cast_fp16)[name = tensor("lora_out_627_cast_fp16")]; + tensor query_53_cast_fp16 = add(x = pretrained_out_313_cast_fp16, y = lora_out_627_cast_fp16)[name = tensor("query_53_cast_fp16")]; + tensor var_42769 = const()[name = tensor("op_42769"), val = tensor([1, 1])]; + tensor var_42771 = const()[name = tensor("op_42771"), val = tensor([1, 1])]; + tensor pretrained_out_315_pad_type_0 = const()[name = tensor("pretrained_out_315_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_315_pad_0 = const()[name = tensor("pretrained_out_315_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290838464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291657728))), name = tensor("layers_26_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_315_cast_fp16 = conv(dilations = var_42771, groups = var_42709, pad = pretrained_out_315_pad_0, pad_type = pretrained_out_315_pad_type_0, strides = var_42769, weight = layers_26_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_105_cast_fp16)[name = tensor("pretrained_out_315_cast_fp16")]; + tensor var_42775 = const()[name = tensor("op_42775"), val = tensor([1, 1])]; + tensor var_42777 = const()[name = tensor("op_42777"), val = tensor([1, 1])]; + tensor input_523_pad_type_0 = const()[name = tensor("input_523_pad_type_0"), val = tensor("custom")]; + tensor input_523_pad_0 = const()[name = tensor("input_523_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291657856)))]; + tensor input_523_cast_fp16 = conv(dilations = var_42777, groups = var_42709, pad = input_523_pad_0, pad_type = input_523_pad_type_0, strides = var_42775, weight = layers_26_self_attn_k_proj_loraA_weight_to_fp16, x = obj_105_cast_fp16)[name = tensor("input_523_cast_fp16")]; + tensor var_42781 = const()[name = tensor("op_42781"), val = tensor([1, 1])]; + tensor var_42783 = const()[name = tensor("op_42783"), val = tensor([1, 1])]; + tensor lora_out_629_pad_type_0 = const()[name = tensor("lora_out_629_pad_type_0"), val = tensor("custom")]; + tensor lora_out_629_pad_0 = const()[name = tensor("lora_out_629_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_631_weight_0_to_fp16 = const()[name = tensor("lora_out_631_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291698880)))]; + tensor lora_out_631_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_42783, groups = var_42709, pad = lora_out_629_pad_0, pad_type = lora_out_629_pad_type_0, strides = var_42781, weight = lora_out_631_weight_0_to_fp16, x = input_523_cast_fp16)[name = tensor("lora_out_631_cast_fp16")]; + tensor key_53_cast_fp16 = add(x = pretrained_out_315_cast_fp16, y = lora_out_631_cast_fp16)[name = tensor("key_53_cast_fp16")]; + tensor var_42794 = const()[name = tensor("op_42794"), val = tensor([1, 1])]; + tensor var_42796 = const()[name = tensor("op_42796"), val = tensor([1, 1])]; + tensor pretrained_out_317_pad_type_0 = const()[name = tensor("pretrained_out_317_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_317_pad_0 = const()[name = tensor("pretrained_out_317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291739904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292559168))), name = tensor("layers_26_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_26_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292559296)))]; + tensor pretrained_out_317_cast_fp16 = conv(bias = layers_26_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_42796, groups = var_42709, pad = pretrained_out_317_pad_0, pad_type = pretrained_out_317_pad_type_0, strides = var_42794, weight = layers_26_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_105_cast_fp16)[name = tensor("pretrained_out_317_cast_fp16")]; + tensor var_42800 = const()[name = tensor("op_42800"), val = tensor([1, 1])]; + tensor var_42802 = const()[name = tensor("op_42802"), val = tensor([1, 1])]; + tensor input_525_pad_type_0 = const()[name = tensor("input_525_pad_type_0"), val = tensor("custom")]; + tensor input_525_pad_0 = const()[name = tensor("input_525_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292561920)))]; + tensor input_525_cast_fp16 = conv(dilations = var_42802, groups = var_42709, pad = input_525_pad_0, pad_type = input_525_pad_type_0, strides = var_42800, weight = layers_26_self_attn_v_proj_loraA_weight_to_fp16, x = obj_105_cast_fp16)[name = tensor("input_525_cast_fp16")]; + tensor var_42806 = const()[name = tensor("op_42806"), val = tensor([1, 1])]; + tensor var_42808 = const()[name = tensor("op_42808"), val = tensor([1, 1])]; + tensor lora_out_633_pad_type_0 = const()[name = tensor("lora_out_633_pad_type_0"), val = tensor("custom")]; + tensor lora_out_633_pad_0 = const()[name = tensor("lora_out_633_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_635_weight_0_to_fp16 = const()[name = tensor("lora_out_635_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292602944)))]; + tensor lora_out_635_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_42808, groups = var_42709, pad = lora_out_633_pad_0, pad_type = lora_out_633_pad_type_0, strides = var_42806, weight = lora_out_635_weight_0_to_fp16, x = input_525_cast_fp16)[name = tensor("lora_out_635_cast_fp16")]; + tensor value_53_cast_fp16 = add(x = pretrained_out_317_cast_fp16, y = lora_out_635_cast_fp16)[name = tensor("value_53_cast_fp16")]; + tensor var_42818_begin_0 = const()[name = tensor("op_42818_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42818_end_0 = const()[name = tensor("op_42818_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42818_end_mask_0 = const()[name = tensor("op_42818_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42818_cast_fp16 = slice_by_index(begin = var_42818_begin_0, end = var_42818_end_0, end_mask = var_42818_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42818_cast_fp16")]; + tensor var_42822_begin_0 = const()[name = tensor("op_42822_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_42822_end_0 = const()[name = tensor("op_42822_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_42822_end_mask_0 = const()[name = tensor("op_42822_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42822_cast_fp16 = slice_by_index(begin = var_42822_begin_0, end = var_42822_end_0, end_mask = var_42822_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42822_cast_fp16")]; + tensor var_42826_begin_0 = const()[name = tensor("op_42826_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_42826_end_0 = const()[name = tensor("op_42826_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_42826_end_mask_0 = const()[name = tensor("op_42826_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42826_cast_fp16 = slice_by_index(begin = var_42826_begin_0, end = var_42826_end_0, end_mask = var_42826_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42826_cast_fp16")]; + tensor var_42830_begin_0 = const()[name = tensor("op_42830_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_42830_end_0 = const()[name = tensor("op_42830_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_42830_end_mask_0 = const()[name = tensor("op_42830_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42830_cast_fp16 = slice_by_index(begin = var_42830_begin_0, end = var_42830_end_0, end_mask = var_42830_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42830_cast_fp16")]; + tensor var_42834_begin_0 = const()[name = tensor("op_42834_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_42834_end_0 = const()[name = tensor("op_42834_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_42834_end_mask_0 = const()[name = tensor("op_42834_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42834_cast_fp16 = slice_by_index(begin = var_42834_begin_0, end = var_42834_end_0, end_mask = var_42834_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42834_cast_fp16")]; + tensor var_42838_begin_0 = const()[name = tensor("op_42838_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_42838_end_0 = const()[name = tensor("op_42838_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_42838_end_mask_0 = const()[name = tensor("op_42838_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42838_cast_fp16 = slice_by_index(begin = var_42838_begin_0, end = var_42838_end_0, end_mask = var_42838_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42838_cast_fp16")]; + tensor var_42842_begin_0 = const()[name = tensor("op_42842_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_42842_end_0 = const()[name = tensor("op_42842_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_42842_end_mask_0 = const()[name = tensor("op_42842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42842_cast_fp16 = slice_by_index(begin = var_42842_begin_0, end = var_42842_end_0, end_mask = var_42842_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42842_cast_fp16")]; + tensor var_42846_begin_0 = const()[name = tensor("op_42846_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_42846_end_0 = const()[name = tensor("op_42846_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_42846_end_mask_0 = const()[name = tensor("op_42846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42846_cast_fp16 = slice_by_index(begin = var_42846_begin_0, end = var_42846_end_0, end_mask = var_42846_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42846_cast_fp16")]; + tensor var_42850_begin_0 = const()[name = tensor("op_42850_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_42850_end_0 = const()[name = tensor("op_42850_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_42850_end_mask_0 = const()[name = tensor("op_42850_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42850_cast_fp16 = slice_by_index(begin = var_42850_begin_0, end = var_42850_end_0, end_mask = var_42850_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42850_cast_fp16")]; + tensor var_42854_begin_0 = const()[name = tensor("op_42854_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_42854_end_0 = const()[name = tensor("op_42854_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_42854_end_mask_0 = const()[name = tensor("op_42854_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42854_cast_fp16 = slice_by_index(begin = var_42854_begin_0, end = var_42854_end_0, end_mask = var_42854_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42854_cast_fp16")]; + tensor var_42858_begin_0 = const()[name = tensor("op_42858_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_42858_end_0 = const()[name = tensor("op_42858_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_42858_end_mask_0 = const()[name = tensor("op_42858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42858_cast_fp16 = slice_by_index(begin = var_42858_begin_0, end = var_42858_end_0, end_mask = var_42858_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42858_cast_fp16")]; + tensor var_42862_begin_0 = const()[name = tensor("op_42862_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_42862_end_0 = const()[name = tensor("op_42862_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_42862_end_mask_0 = const()[name = tensor("op_42862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42862_cast_fp16 = slice_by_index(begin = var_42862_begin_0, end = var_42862_end_0, end_mask = var_42862_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42862_cast_fp16")]; + tensor var_42866_begin_0 = const()[name = tensor("op_42866_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_42866_end_0 = const()[name = tensor("op_42866_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_42866_end_mask_0 = const()[name = tensor("op_42866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42866_cast_fp16 = slice_by_index(begin = var_42866_begin_0, end = var_42866_end_0, end_mask = var_42866_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42866_cast_fp16")]; + tensor var_42870_begin_0 = const()[name = tensor("op_42870_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_42870_end_0 = const()[name = tensor("op_42870_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_42870_end_mask_0 = const()[name = tensor("op_42870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42870_cast_fp16 = slice_by_index(begin = var_42870_begin_0, end = var_42870_end_0, end_mask = var_42870_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42870_cast_fp16")]; + tensor var_42874_begin_0 = const()[name = tensor("op_42874_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_42874_end_0 = const()[name = tensor("op_42874_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_42874_end_mask_0 = const()[name = tensor("op_42874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42874_cast_fp16 = slice_by_index(begin = var_42874_begin_0, end = var_42874_end_0, end_mask = var_42874_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42874_cast_fp16")]; + tensor var_42878_begin_0 = const()[name = tensor("op_42878_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_42878_end_0 = const()[name = tensor("op_42878_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_42878_end_mask_0 = const()[name = tensor("op_42878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42878_cast_fp16 = slice_by_index(begin = var_42878_begin_0, end = var_42878_end_0, end_mask = var_42878_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42878_cast_fp16")]; + tensor var_42882_begin_0 = const()[name = tensor("op_42882_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_42882_end_0 = const()[name = tensor("op_42882_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_42882_end_mask_0 = const()[name = tensor("op_42882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42882_cast_fp16 = slice_by_index(begin = var_42882_begin_0, end = var_42882_end_0, end_mask = var_42882_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42882_cast_fp16")]; + tensor var_42886_begin_0 = const()[name = tensor("op_42886_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_42886_end_0 = const()[name = tensor("op_42886_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_42886_end_mask_0 = const()[name = tensor("op_42886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42886_cast_fp16 = slice_by_index(begin = var_42886_begin_0, end = var_42886_end_0, end_mask = var_42886_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42886_cast_fp16")]; + tensor var_42890_begin_0 = const()[name = tensor("op_42890_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_42890_end_0 = const()[name = tensor("op_42890_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_42890_end_mask_0 = const()[name = tensor("op_42890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42890_cast_fp16 = slice_by_index(begin = var_42890_begin_0, end = var_42890_end_0, end_mask = var_42890_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42890_cast_fp16")]; + tensor var_42894_begin_0 = const()[name = tensor("op_42894_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_42894_end_0 = const()[name = tensor("op_42894_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_42894_end_mask_0 = const()[name = tensor("op_42894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_42894_cast_fp16 = slice_by_index(begin = var_42894_begin_0, end = var_42894_end_0, end_mask = var_42894_end_mask_0, x = query_53_cast_fp16)[name = tensor("op_42894_cast_fp16")]; + tensor var_42903_begin_0 = const()[name = tensor("op_42903_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42903_end_0 = const()[name = tensor("op_42903_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42903_end_mask_0 = const()[name = tensor("op_42903_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42903_cast_fp16 = slice_by_index(begin = var_42903_begin_0, end = var_42903_end_0, end_mask = var_42903_end_mask_0, x = var_42818_cast_fp16)[name = tensor("op_42903_cast_fp16")]; + tensor var_42910_begin_0 = const()[name = tensor("op_42910_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42910_end_0 = const()[name = tensor("op_42910_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42910_end_mask_0 = const()[name = tensor("op_42910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42910_cast_fp16 = slice_by_index(begin = var_42910_begin_0, end = var_42910_end_0, end_mask = var_42910_end_mask_0, x = var_42818_cast_fp16)[name = tensor("op_42910_cast_fp16")]; + tensor var_42917_begin_0 = const()[name = tensor("op_42917_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42917_end_0 = const()[name = tensor("op_42917_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42917_end_mask_0 = const()[name = tensor("op_42917_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42917_cast_fp16 = slice_by_index(begin = var_42917_begin_0, end = var_42917_end_0, end_mask = var_42917_end_mask_0, x = var_42818_cast_fp16)[name = tensor("op_42917_cast_fp16")]; + tensor var_42924_begin_0 = const()[name = tensor("op_42924_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42924_end_0 = const()[name = tensor("op_42924_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42924_end_mask_0 = const()[name = tensor("op_42924_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42924_cast_fp16 = slice_by_index(begin = var_42924_begin_0, end = var_42924_end_0, end_mask = var_42924_end_mask_0, x = var_42818_cast_fp16)[name = tensor("op_42924_cast_fp16")]; + tensor var_42931_begin_0 = const()[name = tensor("op_42931_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42931_end_0 = const()[name = tensor("op_42931_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42931_end_mask_0 = const()[name = tensor("op_42931_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42931_cast_fp16 = slice_by_index(begin = var_42931_begin_0, end = var_42931_end_0, end_mask = var_42931_end_mask_0, x = var_42822_cast_fp16)[name = tensor("op_42931_cast_fp16")]; + tensor var_42938_begin_0 = const()[name = tensor("op_42938_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42938_end_0 = const()[name = tensor("op_42938_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42938_end_mask_0 = const()[name = tensor("op_42938_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42938_cast_fp16 = slice_by_index(begin = var_42938_begin_0, end = var_42938_end_0, end_mask = var_42938_end_mask_0, x = var_42822_cast_fp16)[name = tensor("op_42938_cast_fp16")]; + tensor var_42945_begin_0 = const()[name = tensor("op_42945_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42945_end_0 = const()[name = tensor("op_42945_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42945_end_mask_0 = const()[name = tensor("op_42945_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42945_cast_fp16 = slice_by_index(begin = var_42945_begin_0, end = var_42945_end_0, end_mask = var_42945_end_mask_0, x = var_42822_cast_fp16)[name = tensor("op_42945_cast_fp16")]; + tensor var_42952_begin_0 = const()[name = tensor("op_42952_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42952_end_0 = const()[name = tensor("op_42952_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42952_end_mask_0 = const()[name = tensor("op_42952_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42952_cast_fp16 = slice_by_index(begin = var_42952_begin_0, end = var_42952_end_0, end_mask = var_42952_end_mask_0, x = var_42822_cast_fp16)[name = tensor("op_42952_cast_fp16")]; + tensor var_42959_begin_0 = const()[name = tensor("op_42959_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42959_end_0 = const()[name = tensor("op_42959_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42959_end_mask_0 = const()[name = tensor("op_42959_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42959_cast_fp16 = slice_by_index(begin = var_42959_begin_0, end = var_42959_end_0, end_mask = var_42959_end_mask_0, x = var_42826_cast_fp16)[name = tensor("op_42959_cast_fp16")]; + tensor var_42966_begin_0 = const()[name = tensor("op_42966_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42966_end_0 = const()[name = tensor("op_42966_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42966_end_mask_0 = const()[name = tensor("op_42966_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42966_cast_fp16 = slice_by_index(begin = var_42966_begin_0, end = var_42966_end_0, end_mask = var_42966_end_mask_0, x = var_42826_cast_fp16)[name = tensor("op_42966_cast_fp16")]; + tensor var_42973_begin_0 = const()[name = tensor("op_42973_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_42973_end_0 = const()[name = tensor("op_42973_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_42973_end_mask_0 = const()[name = tensor("op_42973_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42973_cast_fp16 = slice_by_index(begin = var_42973_begin_0, end = var_42973_end_0, end_mask = var_42973_end_mask_0, x = var_42826_cast_fp16)[name = tensor("op_42973_cast_fp16")]; + tensor var_42980_begin_0 = const()[name = tensor("op_42980_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_42980_end_0 = const()[name = tensor("op_42980_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_42980_end_mask_0 = const()[name = tensor("op_42980_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42980_cast_fp16 = slice_by_index(begin = var_42980_begin_0, end = var_42980_end_0, end_mask = var_42980_end_mask_0, x = var_42826_cast_fp16)[name = tensor("op_42980_cast_fp16")]; + tensor var_42987_begin_0 = const()[name = tensor("op_42987_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_42987_end_0 = const()[name = tensor("op_42987_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_42987_end_mask_0 = const()[name = tensor("op_42987_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42987_cast_fp16 = slice_by_index(begin = var_42987_begin_0, end = var_42987_end_0, end_mask = var_42987_end_mask_0, x = var_42830_cast_fp16)[name = tensor("op_42987_cast_fp16")]; + tensor var_42994_begin_0 = const()[name = tensor("op_42994_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_42994_end_0 = const()[name = tensor("op_42994_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_42994_end_mask_0 = const()[name = tensor("op_42994_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_42994_cast_fp16 = slice_by_index(begin = var_42994_begin_0, end = var_42994_end_0, end_mask = var_42994_end_mask_0, x = var_42830_cast_fp16)[name = tensor("op_42994_cast_fp16")]; + tensor var_43001_begin_0 = const()[name = tensor("op_43001_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43001_end_0 = const()[name = tensor("op_43001_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43001_end_mask_0 = const()[name = tensor("op_43001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43001_cast_fp16 = slice_by_index(begin = var_43001_begin_0, end = var_43001_end_0, end_mask = var_43001_end_mask_0, x = var_42830_cast_fp16)[name = tensor("op_43001_cast_fp16")]; + tensor var_43008_begin_0 = const()[name = tensor("op_43008_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43008_end_0 = const()[name = tensor("op_43008_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43008_end_mask_0 = const()[name = tensor("op_43008_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43008_cast_fp16 = slice_by_index(begin = var_43008_begin_0, end = var_43008_end_0, end_mask = var_43008_end_mask_0, x = var_42830_cast_fp16)[name = tensor("op_43008_cast_fp16")]; + tensor var_43015_begin_0 = const()[name = tensor("op_43015_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43015_end_0 = const()[name = tensor("op_43015_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43015_end_mask_0 = const()[name = tensor("op_43015_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43015_cast_fp16 = slice_by_index(begin = var_43015_begin_0, end = var_43015_end_0, end_mask = var_43015_end_mask_0, x = var_42834_cast_fp16)[name = tensor("op_43015_cast_fp16")]; + tensor var_43022_begin_0 = const()[name = tensor("op_43022_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43022_end_0 = const()[name = tensor("op_43022_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43022_end_mask_0 = const()[name = tensor("op_43022_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43022_cast_fp16 = slice_by_index(begin = var_43022_begin_0, end = var_43022_end_0, end_mask = var_43022_end_mask_0, x = var_42834_cast_fp16)[name = tensor("op_43022_cast_fp16")]; + tensor var_43029_begin_0 = const()[name = tensor("op_43029_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43029_end_0 = const()[name = tensor("op_43029_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43029_end_mask_0 = const()[name = tensor("op_43029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43029_cast_fp16 = slice_by_index(begin = var_43029_begin_0, end = var_43029_end_0, end_mask = var_43029_end_mask_0, x = var_42834_cast_fp16)[name = tensor("op_43029_cast_fp16")]; + tensor var_43036_begin_0 = const()[name = tensor("op_43036_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43036_end_0 = const()[name = tensor("op_43036_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43036_end_mask_0 = const()[name = tensor("op_43036_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43036_cast_fp16 = slice_by_index(begin = var_43036_begin_0, end = var_43036_end_0, end_mask = var_43036_end_mask_0, x = var_42834_cast_fp16)[name = tensor("op_43036_cast_fp16")]; + tensor var_43043_begin_0 = const()[name = tensor("op_43043_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43043_end_0 = const()[name = tensor("op_43043_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43043_end_mask_0 = const()[name = tensor("op_43043_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43043_cast_fp16 = slice_by_index(begin = var_43043_begin_0, end = var_43043_end_0, end_mask = var_43043_end_mask_0, x = var_42838_cast_fp16)[name = tensor("op_43043_cast_fp16")]; + tensor var_43050_begin_0 = const()[name = tensor("op_43050_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43050_end_0 = const()[name = tensor("op_43050_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43050_end_mask_0 = const()[name = tensor("op_43050_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43050_cast_fp16 = slice_by_index(begin = var_43050_begin_0, end = var_43050_end_0, end_mask = var_43050_end_mask_0, x = var_42838_cast_fp16)[name = tensor("op_43050_cast_fp16")]; + tensor var_43057_begin_0 = const()[name = tensor("op_43057_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43057_end_0 = const()[name = tensor("op_43057_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43057_end_mask_0 = const()[name = tensor("op_43057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43057_cast_fp16 = slice_by_index(begin = var_43057_begin_0, end = var_43057_end_0, end_mask = var_43057_end_mask_0, x = var_42838_cast_fp16)[name = tensor("op_43057_cast_fp16")]; + tensor var_43064_begin_0 = const()[name = tensor("op_43064_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43064_end_0 = const()[name = tensor("op_43064_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43064_end_mask_0 = const()[name = tensor("op_43064_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43064_cast_fp16 = slice_by_index(begin = var_43064_begin_0, end = var_43064_end_0, end_mask = var_43064_end_mask_0, x = var_42838_cast_fp16)[name = tensor("op_43064_cast_fp16")]; + tensor var_43071_begin_0 = const()[name = tensor("op_43071_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43071_end_0 = const()[name = tensor("op_43071_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43071_end_mask_0 = const()[name = tensor("op_43071_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43071_cast_fp16 = slice_by_index(begin = var_43071_begin_0, end = var_43071_end_0, end_mask = var_43071_end_mask_0, x = var_42842_cast_fp16)[name = tensor("op_43071_cast_fp16")]; + tensor var_43078_begin_0 = const()[name = tensor("op_43078_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43078_end_0 = const()[name = tensor("op_43078_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43078_end_mask_0 = const()[name = tensor("op_43078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43078_cast_fp16 = slice_by_index(begin = var_43078_begin_0, end = var_43078_end_0, end_mask = var_43078_end_mask_0, x = var_42842_cast_fp16)[name = tensor("op_43078_cast_fp16")]; + tensor var_43085_begin_0 = const()[name = tensor("op_43085_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43085_end_0 = const()[name = tensor("op_43085_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43085_end_mask_0 = const()[name = tensor("op_43085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43085_cast_fp16 = slice_by_index(begin = var_43085_begin_0, end = var_43085_end_0, end_mask = var_43085_end_mask_0, x = var_42842_cast_fp16)[name = tensor("op_43085_cast_fp16")]; + tensor var_43092_begin_0 = const()[name = tensor("op_43092_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43092_end_0 = const()[name = tensor("op_43092_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43092_end_mask_0 = const()[name = tensor("op_43092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43092_cast_fp16 = slice_by_index(begin = var_43092_begin_0, end = var_43092_end_0, end_mask = var_43092_end_mask_0, x = var_42842_cast_fp16)[name = tensor("op_43092_cast_fp16")]; + tensor var_43099_begin_0 = const()[name = tensor("op_43099_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43099_end_0 = const()[name = tensor("op_43099_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43099_end_mask_0 = const()[name = tensor("op_43099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43099_cast_fp16 = slice_by_index(begin = var_43099_begin_0, end = var_43099_end_0, end_mask = var_43099_end_mask_0, x = var_42846_cast_fp16)[name = tensor("op_43099_cast_fp16")]; + tensor var_43106_begin_0 = const()[name = tensor("op_43106_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43106_end_0 = const()[name = tensor("op_43106_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43106_end_mask_0 = const()[name = tensor("op_43106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43106_cast_fp16 = slice_by_index(begin = var_43106_begin_0, end = var_43106_end_0, end_mask = var_43106_end_mask_0, x = var_42846_cast_fp16)[name = tensor("op_43106_cast_fp16")]; + tensor var_43113_begin_0 = const()[name = tensor("op_43113_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43113_end_0 = const()[name = tensor("op_43113_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43113_end_mask_0 = const()[name = tensor("op_43113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43113_cast_fp16 = slice_by_index(begin = var_43113_begin_0, end = var_43113_end_0, end_mask = var_43113_end_mask_0, x = var_42846_cast_fp16)[name = tensor("op_43113_cast_fp16")]; + tensor var_43120_begin_0 = const()[name = tensor("op_43120_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43120_end_0 = const()[name = tensor("op_43120_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43120_end_mask_0 = const()[name = tensor("op_43120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43120_cast_fp16 = slice_by_index(begin = var_43120_begin_0, end = var_43120_end_0, end_mask = var_43120_end_mask_0, x = var_42846_cast_fp16)[name = tensor("op_43120_cast_fp16")]; + tensor var_43127_begin_0 = const()[name = tensor("op_43127_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43127_end_0 = const()[name = tensor("op_43127_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43127_end_mask_0 = const()[name = tensor("op_43127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43127_cast_fp16 = slice_by_index(begin = var_43127_begin_0, end = var_43127_end_0, end_mask = var_43127_end_mask_0, x = var_42850_cast_fp16)[name = tensor("op_43127_cast_fp16")]; + tensor var_43134_begin_0 = const()[name = tensor("op_43134_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43134_end_0 = const()[name = tensor("op_43134_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43134_end_mask_0 = const()[name = tensor("op_43134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43134_cast_fp16 = slice_by_index(begin = var_43134_begin_0, end = var_43134_end_0, end_mask = var_43134_end_mask_0, x = var_42850_cast_fp16)[name = tensor("op_43134_cast_fp16")]; + tensor var_43141_begin_0 = const()[name = tensor("op_43141_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43141_end_0 = const()[name = tensor("op_43141_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43141_end_mask_0 = const()[name = tensor("op_43141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43141_cast_fp16 = slice_by_index(begin = var_43141_begin_0, end = var_43141_end_0, end_mask = var_43141_end_mask_0, x = var_42850_cast_fp16)[name = tensor("op_43141_cast_fp16")]; + tensor var_43148_begin_0 = const()[name = tensor("op_43148_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43148_end_0 = const()[name = tensor("op_43148_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43148_end_mask_0 = const()[name = tensor("op_43148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43148_cast_fp16 = slice_by_index(begin = var_43148_begin_0, end = var_43148_end_0, end_mask = var_43148_end_mask_0, x = var_42850_cast_fp16)[name = tensor("op_43148_cast_fp16")]; + tensor var_43155_begin_0 = const()[name = tensor("op_43155_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43155_end_0 = const()[name = tensor("op_43155_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43155_end_mask_0 = const()[name = tensor("op_43155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43155_cast_fp16 = slice_by_index(begin = var_43155_begin_0, end = var_43155_end_0, end_mask = var_43155_end_mask_0, x = var_42854_cast_fp16)[name = tensor("op_43155_cast_fp16")]; + tensor var_43162_begin_0 = const()[name = tensor("op_43162_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43162_end_0 = const()[name = tensor("op_43162_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43162_end_mask_0 = const()[name = tensor("op_43162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43162_cast_fp16 = slice_by_index(begin = var_43162_begin_0, end = var_43162_end_0, end_mask = var_43162_end_mask_0, x = var_42854_cast_fp16)[name = tensor("op_43162_cast_fp16")]; + tensor var_43169_begin_0 = const()[name = tensor("op_43169_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43169_end_0 = const()[name = tensor("op_43169_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43169_end_mask_0 = const()[name = tensor("op_43169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43169_cast_fp16 = slice_by_index(begin = var_43169_begin_0, end = var_43169_end_0, end_mask = var_43169_end_mask_0, x = var_42854_cast_fp16)[name = tensor("op_43169_cast_fp16")]; + tensor var_43176_begin_0 = const()[name = tensor("op_43176_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43176_end_0 = const()[name = tensor("op_43176_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43176_end_mask_0 = const()[name = tensor("op_43176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43176_cast_fp16 = slice_by_index(begin = var_43176_begin_0, end = var_43176_end_0, end_mask = var_43176_end_mask_0, x = var_42854_cast_fp16)[name = tensor("op_43176_cast_fp16")]; + tensor var_43183_begin_0 = const()[name = tensor("op_43183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43183_end_0 = const()[name = tensor("op_43183_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43183_end_mask_0 = const()[name = tensor("op_43183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43183_cast_fp16 = slice_by_index(begin = var_43183_begin_0, end = var_43183_end_0, end_mask = var_43183_end_mask_0, x = var_42858_cast_fp16)[name = tensor("op_43183_cast_fp16")]; + tensor var_43190_begin_0 = const()[name = tensor("op_43190_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43190_end_0 = const()[name = tensor("op_43190_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43190_end_mask_0 = const()[name = tensor("op_43190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43190_cast_fp16 = slice_by_index(begin = var_43190_begin_0, end = var_43190_end_0, end_mask = var_43190_end_mask_0, x = var_42858_cast_fp16)[name = tensor("op_43190_cast_fp16")]; + tensor var_43197_begin_0 = const()[name = tensor("op_43197_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43197_end_0 = const()[name = tensor("op_43197_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43197_end_mask_0 = const()[name = tensor("op_43197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43197_cast_fp16 = slice_by_index(begin = var_43197_begin_0, end = var_43197_end_0, end_mask = var_43197_end_mask_0, x = var_42858_cast_fp16)[name = tensor("op_43197_cast_fp16")]; + tensor var_43204_begin_0 = const()[name = tensor("op_43204_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43204_end_0 = const()[name = tensor("op_43204_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43204_end_mask_0 = const()[name = tensor("op_43204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43204_cast_fp16 = slice_by_index(begin = var_43204_begin_0, end = var_43204_end_0, end_mask = var_43204_end_mask_0, x = var_42858_cast_fp16)[name = tensor("op_43204_cast_fp16")]; + tensor var_43211_begin_0 = const()[name = tensor("op_43211_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43211_end_0 = const()[name = tensor("op_43211_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43211_end_mask_0 = const()[name = tensor("op_43211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43211_cast_fp16 = slice_by_index(begin = var_43211_begin_0, end = var_43211_end_0, end_mask = var_43211_end_mask_0, x = var_42862_cast_fp16)[name = tensor("op_43211_cast_fp16")]; + tensor var_43218_begin_0 = const()[name = tensor("op_43218_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43218_end_0 = const()[name = tensor("op_43218_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43218_end_mask_0 = const()[name = tensor("op_43218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43218_cast_fp16 = slice_by_index(begin = var_43218_begin_0, end = var_43218_end_0, end_mask = var_43218_end_mask_0, x = var_42862_cast_fp16)[name = tensor("op_43218_cast_fp16")]; + tensor var_43225_begin_0 = const()[name = tensor("op_43225_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43225_end_0 = const()[name = tensor("op_43225_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43225_end_mask_0 = const()[name = tensor("op_43225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43225_cast_fp16 = slice_by_index(begin = var_43225_begin_0, end = var_43225_end_0, end_mask = var_43225_end_mask_0, x = var_42862_cast_fp16)[name = tensor("op_43225_cast_fp16")]; + tensor var_43232_begin_0 = const()[name = tensor("op_43232_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43232_end_0 = const()[name = tensor("op_43232_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43232_end_mask_0 = const()[name = tensor("op_43232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43232_cast_fp16 = slice_by_index(begin = var_43232_begin_0, end = var_43232_end_0, end_mask = var_43232_end_mask_0, x = var_42862_cast_fp16)[name = tensor("op_43232_cast_fp16")]; + tensor var_43239_begin_0 = const()[name = tensor("op_43239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43239_end_0 = const()[name = tensor("op_43239_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43239_end_mask_0 = const()[name = tensor("op_43239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43239_cast_fp16 = slice_by_index(begin = var_43239_begin_0, end = var_43239_end_0, end_mask = var_43239_end_mask_0, x = var_42866_cast_fp16)[name = tensor("op_43239_cast_fp16")]; + tensor var_43246_begin_0 = const()[name = tensor("op_43246_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43246_end_0 = const()[name = tensor("op_43246_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43246_end_mask_0 = const()[name = tensor("op_43246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43246_cast_fp16 = slice_by_index(begin = var_43246_begin_0, end = var_43246_end_0, end_mask = var_43246_end_mask_0, x = var_42866_cast_fp16)[name = tensor("op_43246_cast_fp16")]; + tensor var_43253_begin_0 = const()[name = tensor("op_43253_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43253_end_0 = const()[name = tensor("op_43253_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43253_end_mask_0 = const()[name = tensor("op_43253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43253_cast_fp16 = slice_by_index(begin = var_43253_begin_0, end = var_43253_end_0, end_mask = var_43253_end_mask_0, x = var_42866_cast_fp16)[name = tensor("op_43253_cast_fp16")]; + tensor var_43260_begin_0 = const()[name = tensor("op_43260_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43260_end_0 = const()[name = tensor("op_43260_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43260_end_mask_0 = const()[name = tensor("op_43260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43260_cast_fp16 = slice_by_index(begin = var_43260_begin_0, end = var_43260_end_0, end_mask = var_43260_end_mask_0, x = var_42866_cast_fp16)[name = tensor("op_43260_cast_fp16")]; + tensor var_43267_begin_0 = const()[name = tensor("op_43267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43267_end_0 = const()[name = tensor("op_43267_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43267_end_mask_0 = const()[name = tensor("op_43267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43267_cast_fp16 = slice_by_index(begin = var_43267_begin_0, end = var_43267_end_0, end_mask = var_43267_end_mask_0, x = var_42870_cast_fp16)[name = tensor("op_43267_cast_fp16")]; + tensor var_43274_begin_0 = const()[name = tensor("op_43274_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43274_end_0 = const()[name = tensor("op_43274_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43274_end_mask_0 = const()[name = tensor("op_43274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43274_cast_fp16 = slice_by_index(begin = var_43274_begin_0, end = var_43274_end_0, end_mask = var_43274_end_mask_0, x = var_42870_cast_fp16)[name = tensor("op_43274_cast_fp16")]; + tensor var_43281_begin_0 = const()[name = tensor("op_43281_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43281_end_0 = const()[name = tensor("op_43281_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43281_end_mask_0 = const()[name = tensor("op_43281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43281_cast_fp16 = slice_by_index(begin = var_43281_begin_0, end = var_43281_end_0, end_mask = var_43281_end_mask_0, x = var_42870_cast_fp16)[name = tensor("op_43281_cast_fp16")]; + tensor var_43288_begin_0 = const()[name = tensor("op_43288_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43288_end_0 = const()[name = tensor("op_43288_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43288_end_mask_0 = const()[name = tensor("op_43288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43288_cast_fp16 = slice_by_index(begin = var_43288_begin_0, end = var_43288_end_0, end_mask = var_43288_end_mask_0, x = var_42870_cast_fp16)[name = tensor("op_43288_cast_fp16")]; + tensor var_43295_begin_0 = const()[name = tensor("op_43295_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43295_end_0 = const()[name = tensor("op_43295_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43295_end_mask_0 = const()[name = tensor("op_43295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43295_cast_fp16 = slice_by_index(begin = var_43295_begin_0, end = var_43295_end_0, end_mask = var_43295_end_mask_0, x = var_42874_cast_fp16)[name = tensor("op_43295_cast_fp16")]; + tensor var_43302_begin_0 = const()[name = tensor("op_43302_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43302_end_0 = const()[name = tensor("op_43302_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43302_end_mask_0 = const()[name = tensor("op_43302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43302_cast_fp16 = slice_by_index(begin = var_43302_begin_0, end = var_43302_end_0, end_mask = var_43302_end_mask_0, x = var_42874_cast_fp16)[name = tensor("op_43302_cast_fp16")]; + tensor var_43309_begin_0 = const()[name = tensor("op_43309_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43309_end_0 = const()[name = tensor("op_43309_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43309_end_mask_0 = const()[name = tensor("op_43309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43309_cast_fp16 = slice_by_index(begin = var_43309_begin_0, end = var_43309_end_0, end_mask = var_43309_end_mask_0, x = var_42874_cast_fp16)[name = tensor("op_43309_cast_fp16")]; + tensor var_43316_begin_0 = const()[name = tensor("op_43316_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43316_end_0 = const()[name = tensor("op_43316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43316_end_mask_0 = const()[name = tensor("op_43316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43316_cast_fp16 = slice_by_index(begin = var_43316_begin_0, end = var_43316_end_0, end_mask = var_43316_end_mask_0, x = var_42874_cast_fp16)[name = tensor("op_43316_cast_fp16")]; + tensor var_43323_begin_0 = const()[name = tensor("op_43323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43323_end_0 = const()[name = tensor("op_43323_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43323_end_mask_0 = const()[name = tensor("op_43323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43323_cast_fp16 = slice_by_index(begin = var_43323_begin_0, end = var_43323_end_0, end_mask = var_43323_end_mask_0, x = var_42878_cast_fp16)[name = tensor("op_43323_cast_fp16")]; + tensor var_43330_begin_0 = const()[name = tensor("op_43330_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43330_end_0 = const()[name = tensor("op_43330_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43330_end_mask_0 = const()[name = tensor("op_43330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43330_cast_fp16 = slice_by_index(begin = var_43330_begin_0, end = var_43330_end_0, end_mask = var_43330_end_mask_0, x = var_42878_cast_fp16)[name = tensor("op_43330_cast_fp16")]; + tensor var_43337_begin_0 = const()[name = tensor("op_43337_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43337_end_0 = const()[name = tensor("op_43337_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43337_end_mask_0 = const()[name = tensor("op_43337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43337_cast_fp16 = slice_by_index(begin = var_43337_begin_0, end = var_43337_end_0, end_mask = var_43337_end_mask_0, x = var_42878_cast_fp16)[name = tensor("op_43337_cast_fp16")]; + tensor var_43344_begin_0 = const()[name = tensor("op_43344_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43344_end_0 = const()[name = tensor("op_43344_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43344_end_mask_0 = const()[name = tensor("op_43344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43344_cast_fp16 = slice_by_index(begin = var_43344_begin_0, end = var_43344_end_0, end_mask = var_43344_end_mask_0, x = var_42878_cast_fp16)[name = tensor("op_43344_cast_fp16")]; + tensor var_43351_begin_0 = const()[name = tensor("op_43351_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43351_end_0 = const()[name = tensor("op_43351_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43351_end_mask_0 = const()[name = tensor("op_43351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43351_cast_fp16 = slice_by_index(begin = var_43351_begin_0, end = var_43351_end_0, end_mask = var_43351_end_mask_0, x = var_42882_cast_fp16)[name = tensor("op_43351_cast_fp16")]; + tensor var_43358_begin_0 = const()[name = tensor("op_43358_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43358_end_0 = const()[name = tensor("op_43358_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43358_end_mask_0 = const()[name = tensor("op_43358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43358_cast_fp16 = slice_by_index(begin = var_43358_begin_0, end = var_43358_end_0, end_mask = var_43358_end_mask_0, x = var_42882_cast_fp16)[name = tensor("op_43358_cast_fp16")]; + tensor var_43365_begin_0 = const()[name = tensor("op_43365_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43365_end_0 = const()[name = tensor("op_43365_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43365_end_mask_0 = const()[name = tensor("op_43365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43365_cast_fp16 = slice_by_index(begin = var_43365_begin_0, end = var_43365_end_0, end_mask = var_43365_end_mask_0, x = var_42882_cast_fp16)[name = tensor("op_43365_cast_fp16")]; + tensor var_43372_begin_0 = const()[name = tensor("op_43372_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43372_end_0 = const()[name = tensor("op_43372_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43372_end_mask_0 = const()[name = tensor("op_43372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43372_cast_fp16 = slice_by_index(begin = var_43372_begin_0, end = var_43372_end_0, end_mask = var_43372_end_mask_0, x = var_42882_cast_fp16)[name = tensor("op_43372_cast_fp16")]; + tensor var_43379_begin_0 = const()[name = tensor("op_43379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43379_end_0 = const()[name = tensor("op_43379_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43379_end_mask_0 = const()[name = tensor("op_43379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43379_cast_fp16 = slice_by_index(begin = var_43379_begin_0, end = var_43379_end_0, end_mask = var_43379_end_mask_0, x = var_42886_cast_fp16)[name = tensor("op_43379_cast_fp16")]; + tensor var_43386_begin_0 = const()[name = tensor("op_43386_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43386_end_0 = const()[name = tensor("op_43386_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43386_end_mask_0 = const()[name = tensor("op_43386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43386_cast_fp16 = slice_by_index(begin = var_43386_begin_0, end = var_43386_end_0, end_mask = var_43386_end_mask_0, x = var_42886_cast_fp16)[name = tensor("op_43386_cast_fp16")]; + tensor var_43393_begin_0 = const()[name = tensor("op_43393_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43393_end_0 = const()[name = tensor("op_43393_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43393_end_mask_0 = const()[name = tensor("op_43393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43393_cast_fp16 = slice_by_index(begin = var_43393_begin_0, end = var_43393_end_0, end_mask = var_43393_end_mask_0, x = var_42886_cast_fp16)[name = tensor("op_43393_cast_fp16")]; + tensor var_43400_begin_0 = const()[name = tensor("op_43400_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43400_end_0 = const()[name = tensor("op_43400_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43400_end_mask_0 = const()[name = tensor("op_43400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43400_cast_fp16 = slice_by_index(begin = var_43400_begin_0, end = var_43400_end_0, end_mask = var_43400_end_mask_0, x = var_42886_cast_fp16)[name = tensor("op_43400_cast_fp16")]; + tensor var_43407_begin_0 = const()[name = tensor("op_43407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43407_end_0 = const()[name = tensor("op_43407_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43407_end_mask_0 = const()[name = tensor("op_43407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43407_cast_fp16 = slice_by_index(begin = var_43407_begin_0, end = var_43407_end_0, end_mask = var_43407_end_mask_0, x = var_42890_cast_fp16)[name = tensor("op_43407_cast_fp16")]; + tensor var_43414_begin_0 = const()[name = tensor("op_43414_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43414_end_0 = const()[name = tensor("op_43414_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43414_end_mask_0 = const()[name = tensor("op_43414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43414_cast_fp16 = slice_by_index(begin = var_43414_begin_0, end = var_43414_end_0, end_mask = var_43414_end_mask_0, x = var_42890_cast_fp16)[name = tensor("op_43414_cast_fp16")]; + tensor var_43421_begin_0 = const()[name = tensor("op_43421_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43421_end_0 = const()[name = tensor("op_43421_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43421_end_mask_0 = const()[name = tensor("op_43421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43421_cast_fp16 = slice_by_index(begin = var_43421_begin_0, end = var_43421_end_0, end_mask = var_43421_end_mask_0, x = var_42890_cast_fp16)[name = tensor("op_43421_cast_fp16")]; + tensor var_43428_begin_0 = const()[name = tensor("op_43428_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43428_end_0 = const()[name = tensor("op_43428_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43428_end_mask_0 = const()[name = tensor("op_43428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43428_cast_fp16 = slice_by_index(begin = var_43428_begin_0, end = var_43428_end_0, end_mask = var_43428_end_mask_0, x = var_42890_cast_fp16)[name = tensor("op_43428_cast_fp16")]; + tensor var_43435_begin_0 = const()[name = tensor("op_43435_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43435_end_0 = const()[name = tensor("op_43435_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_43435_end_mask_0 = const()[name = tensor("op_43435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43435_cast_fp16 = slice_by_index(begin = var_43435_begin_0, end = var_43435_end_0, end_mask = var_43435_end_mask_0, x = var_42894_cast_fp16)[name = tensor("op_43435_cast_fp16")]; + tensor var_43442_begin_0 = const()[name = tensor("op_43442_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_43442_end_0 = const()[name = tensor("op_43442_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_43442_end_mask_0 = const()[name = tensor("op_43442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43442_cast_fp16 = slice_by_index(begin = var_43442_begin_0, end = var_43442_end_0, end_mask = var_43442_end_mask_0, x = var_42894_cast_fp16)[name = tensor("op_43442_cast_fp16")]; + tensor var_43449_begin_0 = const()[name = tensor("op_43449_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_43449_end_0 = const()[name = tensor("op_43449_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_43449_end_mask_0 = const()[name = tensor("op_43449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43449_cast_fp16 = slice_by_index(begin = var_43449_begin_0, end = var_43449_end_0, end_mask = var_43449_end_mask_0, x = var_42894_cast_fp16)[name = tensor("op_43449_cast_fp16")]; + tensor var_43456_begin_0 = const()[name = tensor("op_43456_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_43456_end_0 = const()[name = tensor("op_43456_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43456_end_mask_0 = const()[name = tensor("op_43456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43456_cast_fp16 = slice_by_index(begin = var_43456_begin_0, end = var_43456_end_0, end_mask = var_43456_end_mask_0, x = var_42894_cast_fp16)[name = tensor("op_43456_cast_fp16")]; + tensor k_53_perm_0 = const()[name = tensor("k_53_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_43461_begin_0 = const()[name = tensor("op_43461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43461_end_0 = const()[name = tensor("op_43461_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_43461_end_mask_0 = const()[name = tensor("op_43461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_5 = transpose(perm = k_53_perm_0, x = key_53_cast_fp16)[name = tensor("transpose_5")]; + tensor var_43461_cast_fp16 = slice_by_index(begin = var_43461_begin_0, end = var_43461_end_0, end_mask = var_43461_end_mask_0, x = transpose_5)[name = tensor("op_43461_cast_fp16")]; + tensor var_43465_begin_0 = const()[name = tensor("op_43465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_43465_end_0 = const()[name = tensor("op_43465_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_43465_end_mask_0 = const()[name = tensor("op_43465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43465_cast_fp16 = slice_by_index(begin = var_43465_begin_0, end = var_43465_end_0, end_mask = var_43465_end_mask_0, x = transpose_5)[name = tensor("op_43465_cast_fp16")]; + tensor var_43469_begin_0 = const()[name = tensor("op_43469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_43469_end_0 = const()[name = tensor("op_43469_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_43469_end_mask_0 = const()[name = tensor("op_43469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43469_cast_fp16 = slice_by_index(begin = var_43469_begin_0, end = var_43469_end_0, end_mask = var_43469_end_mask_0, x = transpose_5)[name = tensor("op_43469_cast_fp16")]; + tensor var_43473_begin_0 = const()[name = tensor("op_43473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_43473_end_0 = const()[name = tensor("op_43473_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_43473_end_mask_0 = const()[name = tensor("op_43473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43473_cast_fp16 = slice_by_index(begin = var_43473_begin_0, end = var_43473_end_0, end_mask = var_43473_end_mask_0, x = transpose_5)[name = tensor("op_43473_cast_fp16")]; + tensor var_43477_begin_0 = const()[name = tensor("op_43477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_43477_end_0 = const()[name = tensor("op_43477_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_43477_end_mask_0 = const()[name = tensor("op_43477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43477_cast_fp16 = slice_by_index(begin = var_43477_begin_0, end = var_43477_end_0, end_mask = var_43477_end_mask_0, x = transpose_5)[name = tensor("op_43477_cast_fp16")]; + tensor var_43481_begin_0 = const()[name = tensor("op_43481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_43481_end_0 = const()[name = tensor("op_43481_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_43481_end_mask_0 = const()[name = tensor("op_43481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43481_cast_fp16 = slice_by_index(begin = var_43481_begin_0, end = var_43481_end_0, end_mask = var_43481_end_mask_0, x = transpose_5)[name = tensor("op_43481_cast_fp16")]; + tensor var_43485_begin_0 = const()[name = tensor("op_43485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_43485_end_0 = const()[name = tensor("op_43485_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_43485_end_mask_0 = const()[name = tensor("op_43485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43485_cast_fp16 = slice_by_index(begin = var_43485_begin_0, end = var_43485_end_0, end_mask = var_43485_end_mask_0, x = transpose_5)[name = tensor("op_43485_cast_fp16")]; + tensor var_43489_begin_0 = const()[name = tensor("op_43489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_43489_end_0 = const()[name = tensor("op_43489_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_43489_end_mask_0 = const()[name = tensor("op_43489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43489_cast_fp16 = slice_by_index(begin = var_43489_begin_0, end = var_43489_end_0, end_mask = var_43489_end_mask_0, x = transpose_5)[name = tensor("op_43489_cast_fp16")]; + tensor var_43493_begin_0 = const()[name = tensor("op_43493_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_43493_end_0 = const()[name = tensor("op_43493_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_43493_end_mask_0 = const()[name = tensor("op_43493_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43493_cast_fp16 = slice_by_index(begin = var_43493_begin_0, end = var_43493_end_0, end_mask = var_43493_end_mask_0, x = transpose_5)[name = tensor("op_43493_cast_fp16")]; + tensor var_43497_begin_0 = const()[name = tensor("op_43497_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_43497_end_0 = const()[name = tensor("op_43497_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_43497_end_mask_0 = const()[name = tensor("op_43497_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43497_cast_fp16 = slice_by_index(begin = var_43497_begin_0, end = var_43497_end_0, end_mask = var_43497_end_mask_0, x = transpose_5)[name = tensor("op_43497_cast_fp16")]; + tensor var_43501_begin_0 = const()[name = tensor("op_43501_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_43501_end_0 = const()[name = tensor("op_43501_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_43501_end_mask_0 = const()[name = tensor("op_43501_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43501_cast_fp16 = slice_by_index(begin = var_43501_begin_0, end = var_43501_end_0, end_mask = var_43501_end_mask_0, x = transpose_5)[name = tensor("op_43501_cast_fp16")]; + tensor var_43505_begin_0 = const()[name = tensor("op_43505_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_43505_end_0 = const()[name = tensor("op_43505_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_43505_end_mask_0 = const()[name = tensor("op_43505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43505_cast_fp16 = slice_by_index(begin = var_43505_begin_0, end = var_43505_end_0, end_mask = var_43505_end_mask_0, x = transpose_5)[name = tensor("op_43505_cast_fp16")]; + tensor var_43509_begin_0 = const()[name = tensor("op_43509_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_43509_end_0 = const()[name = tensor("op_43509_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_43509_end_mask_0 = const()[name = tensor("op_43509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43509_cast_fp16 = slice_by_index(begin = var_43509_begin_0, end = var_43509_end_0, end_mask = var_43509_end_mask_0, x = transpose_5)[name = tensor("op_43509_cast_fp16")]; + tensor var_43513_begin_0 = const()[name = tensor("op_43513_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_43513_end_0 = const()[name = tensor("op_43513_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_43513_end_mask_0 = const()[name = tensor("op_43513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43513_cast_fp16 = slice_by_index(begin = var_43513_begin_0, end = var_43513_end_0, end_mask = var_43513_end_mask_0, x = transpose_5)[name = tensor("op_43513_cast_fp16")]; + tensor var_43517_begin_0 = const()[name = tensor("op_43517_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_43517_end_0 = const()[name = tensor("op_43517_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_43517_end_mask_0 = const()[name = tensor("op_43517_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43517_cast_fp16 = slice_by_index(begin = var_43517_begin_0, end = var_43517_end_0, end_mask = var_43517_end_mask_0, x = transpose_5)[name = tensor("op_43517_cast_fp16")]; + tensor var_43521_begin_0 = const()[name = tensor("op_43521_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_43521_end_0 = const()[name = tensor("op_43521_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_43521_end_mask_0 = const()[name = tensor("op_43521_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43521_cast_fp16 = slice_by_index(begin = var_43521_begin_0, end = var_43521_end_0, end_mask = var_43521_end_mask_0, x = transpose_5)[name = tensor("op_43521_cast_fp16")]; + tensor var_43525_begin_0 = const()[name = tensor("op_43525_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_43525_end_0 = const()[name = tensor("op_43525_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_43525_end_mask_0 = const()[name = tensor("op_43525_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43525_cast_fp16 = slice_by_index(begin = var_43525_begin_0, end = var_43525_end_0, end_mask = var_43525_end_mask_0, x = transpose_5)[name = tensor("op_43525_cast_fp16")]; + tensor var_43529_begin_0 = const()[name = tensor("op_43529_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_43529_end_0 = const()[name = tensor("op_43529_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_43529_end_mask_0 = const()[name = tensor("op_43529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43529_cast_fp16 = slice_by_index(begin = var_43529_begin_0, end = var_43529_end_0, end_mask = var_43529_end_mask_0, x = transpose_5)[name = tensor("op_43529_cast_fp16")]; + tensor var_43533_begin_0 = const()[name = tensor("op_43533_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_43533_end_0 = const()[name = tensor("op_43533_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_43533_end_mask_0 = const()[name = tensor("op_43533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43533_cast_fp16 = slice_by_index(begin = var_43533_begin_0, end = var_43533_end_0, end_mask = var_43533_end_mask_0, x = transpose_5)[name = tensor("op_43533_cast_fp16")]; + tensor var_43537_begin_0 = const()[name = tensor("op_43537_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_43537_end_0 = const()[name = tensor("op_43537_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_43537_end_mask_0 = const()[name = tensor("op_43537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_43537_cast_fp16 = slice_by_index(begin = var_43537_begin_0, end = var_43537_end_0, end_mask = var_43537_end_mask_0, x = transpose_5)[name = tensor("op_43537_cast_fp16")]; + tensor var_43539_begin_0 = const()[name = tensor("op_43539_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_43539_end_0 = const()[name = tensor("op_43539_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_43539_end_mask_0 = const()[name = tensor("op_43539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43539_cast_fp16 = slice_by_index(begin = var_43539_begin_0, end = var_43539_end_0, end_mask = var_43539_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43539_cast_fp16")]; + tensor var_43543_begin_0 = const()[name = tensor("op_43543_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_43543_end_0 = const()[name = tensor("op_43543_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_43543_end_mask_0 = const()[name = tensor("op_43543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43543_cast_fp16 = slice_by_index(begin = var_43543_begin_0, end = var_43543_end_0, end_mask = var_43543_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43543_cast_fp16")]; + tensor var_43547_begin_0 = const()[name = tensor("op_43547_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_43547_end_0 = const()[name = tensor("op_43547_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_43547_end_mask_0 = const()[name = tensor("op_43547_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43547_cast_fp16 = slice_by_index(begin = var_43547_begin_0, end = var_43547_end_0, end_mask = var_43547_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43547_cast_fp16")]; + tensor var_43551_begin_0 = const()[name = tensor("op_43551_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_43551_end_0 = const()[name = tensor("op_43551_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_43551_end_mask_0 = const()[name = tensor("op_43551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43551_cast_fp16 = slice_by_index(begin = var_43551_begin_0, end = var_43551_end_0, end_mask = var_43551_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43551_cast_fp16")]; + tensor var_43555_begin_0 = const()[name = tensor("op_43555_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_43555_end_0 = const()[name = tensor("op_43555_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_43555_end_mask_0 = const()[name = tensor("op_43555_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43555_cast_fp16 = slice_by_index(begin = var_43555_begin_0, end = var_43555_end_0, end_mask = var_43555_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43555_cast_fp16")]; + tensor var_43559_begin_0 = const()[name = tensor("op_43559_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_43559_end_0 = const()[name = tensor("op_43559_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_43559_end_mask_0 = const()[name = tensor("op_43559_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43559_cast_fp16 = slice_by_index(begin = var_43559_begin_0, end = var_43559_end_0, end_mask = var_43559_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43559_cast_fp16")]; + tensor var_43563_begin_0 = const()[name = tensor("op_43563_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_43563_end_0 = const()[name = tensor("op_43563_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_43563_end_mask_0 = const()[name = tensor("op_43563_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43563_cast_fp16 = slice_by_index(begin = var_43563_begin_0, end = var_43563_end_0, end_mask = var_43563_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43563_cast_fp16")]; + tensor var_43567_begin_0 = const()[name = tensor("op_43567_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_43567_end_0 = const()[name = tensor("op_43567_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_43567_end_mask_0 = const()[name = tensor("op_43567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43567_cast_fp16 = slice_by_index(begin = var_43567_begin_0, end = var_43567_end_0, end_mask = var_43567_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43567_cast_fp16")]; + tensor var_43571_begin_0 = const()[name = tensor("op_43571_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_43571_end_0 = const()[name = tensor("op_43571_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_43571_end_mask_0 = const()[name = tensor("op_43571_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43571_cast_fp16 = slice_by_index(begin = var_43571_begin_0, end = var_43571_end_0, end_mask = var_43571_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43571_cast_fp16")]; + tensor var_43575_begin_0 = const()[name = tensor("op_43575_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_43575_end_0 = const()[name = tensor("op_43575_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_43575_end_mask_0 = const()[name = tensor("op_43575_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43575_cast_fp16 = slice_by_index(begin = var_43575_begin_0, end = var_43575_end_0, end_mask = var_43575_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43575_cast_fp16")]; + tensor var_43579_begin_0 = const()[name = tensor("op_43579_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_43579_end_0 = const()[name = tensor("op_43579_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_43579_end_mask_0 = const()[name = tensor("op_43579_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43579_cast_fp16 = slice_by_index(begin = var_43579_begin_0, end = var_43579_end_0, end_mask = var_43579_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43579_cast_fp16")]; + tensor var_43583_begin_0 = const()[name = tensor("op_43583_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_43583_end_0 = const()[name = tensor("op_43583_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_43583_end_mask_0 = const()[name = tensor("op_43583_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43583_cast_fp16 = slice_by_index(begin = var_43583_begin_0, end = var_43583_end_0, end_mask = var_43583_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43583_cast_fp16")]; + tensor var_43587_begin_0 = const()[name = tensor("op_43587_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_43587_end_0 = const()[name = tensor("op_43587_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_43587_end_mask_0 = const()[name = tensor("op_43587_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43587_cast_fp16 = slice_by_index(begin = var_43587_begin_0, end = var_43587_end_0, end_mask = var_43587_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43587_cast_fp16")]; + tensor var_43591_begin_0 = const()[name = tensor("op_43591_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_43591_end_0 = const()[name = tensor("op_43591_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_43591_end_mask_0 = const()[name = tensor("op_43591_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43591_cast_fp16 = slice_by_index(begin = var_43591_begin_0, end = var_43591_end_0, end_mask = var_43591_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43591_cast_fp16")]; + tensor var_43595_begin_0 = const()[name = tensor("op_43595_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_43595_end_0 = const()[name = tensor("op_43595_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_43595_end_mask_0 = const()[name = tensor("op_43595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43595_cast_fp16 = slice_by_index(begin = var_43595_begin_0, end = var_43595_end_0, end_mask = var_43595_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43595_cast_fp16")]; + tensor var_43599_begin_0 = const()[name = tensor("op_43599_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_43599_end_0 = const()[name = tensor("op_43599_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_43599_end_mask_0 = const()[name = tensor("op_43599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43599_cast_fp16 = slice_by_index(begin = var_43599_begin_0, end = var_43599_end_0, end_mask = var_43599_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43599_cast_fp16")]; + tensor var_43603_begin_0 = const()[name = tensor("op_43603_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_43603_end_0 = const()[name = tensor("op_43603_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_43603_end_mask_0 = const()[name = tensor("op_43603_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43603_cast_fp16 = slice_by_index(begin = var_43603_begin_0, end = var_43603_end_0, end_mask = var_43603_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43603_cast_fp16")]; + tensor var_43607_begin_0 = const()[name = tensor("op_43607_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_43607_end_0 = const()[name = tensor("op_43607_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_43607_end_mask_0 = const()[name = tensor("op_43607_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43607_cast_fp16 = slice_by_index(begin = var_43607_begin_0, end = var_43607_end_0, end_mask = var_43607_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43607_cast_fp16")]; + tensor var_43611_begin_0 = const()[name = tensor("op_43611_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_43611_end_0 = const()[name = tensor("op_43611_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_43611_end_mask_0 = const()[name = tensor("op_43611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43611_cast_fp16 = slice_by_index(begin = var_43611_begin_0, end = var_43611_end_0, end_mask = var_43611_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43611_cast_fp16")]; + tensor var_43615_begin_0 = const()[name = tensor("op_43615_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_43615_end_0 = const()[name = tensor("op_43615_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_43615_end_mask_0 = const()[name = tensor("op_43615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_43615_cast_fp16 = slice_by_index(begin = var_43615_begin_0, end = var_43615_end_0, end_mask = var_43615_end_mask_0, x = value_53_cast_fp16)[name = tensor("op_43615_cast_fp16")]; + tensor var_43619_equation_0 = const()[name = tensor("op_43619_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43619_cast_fp16 = einsum(equation = var_43619_equation_0, values = (var_43461_cast_fp16, var_42903_cast_fp16))[name = tensor("op_43619_cast_fp16")]; + tensor var_43620_to_fp16 = const()[name = tensor("op_43620_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4161_cast_fp16 = mul(x = var_43619_cast_fp16, y = var_43620_to_fp16)[name = tensor("aw_chunk_4161_cast_fp16")]; + tensor var_43623_equation_0 = const()[name = tensor("op_43623_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43623_cast_fp16 = einsum(equation = var_43623_equation_0, values = (var_43461_cast_fp16, var_42910_cast_fp16))[name = tensor("op_43623_cast_fp16")]; + tensor var_43624_to_fp16 = const()[name = tensor("op_43624_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4163_cast_fp16 = mul(x = var_43623_cast_fp16, y = var_43624_to_fp16)[name = tensor("aw_chunk_4163_cast_fp16")]; + tensor var_43627_equation_0 = const()[name = tensor("op_43627_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43627_cast_fp16 = einsum(equation = var_43627_equation_0, values = (var_43461_cast_fp16, var_42917_cast_fp16))[name = tensor("op_43627_cast_fp16")]; + tensor var_43628_to_fp16 = const()[name = tensor("op_43628_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4165_cast_fp16 = mul(x = var_43627_cast_fp16, y = var_43628_to_fp16)[name = tensor("aw_chunk_4165_cast_fp16")]; + tensor var_43631_equation_0 = const()[name = tensor("op_43631_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43631_cast_fp16 = einsum(equation = var_43631_equation_0, values = (var_43461_cast_fp16, var_42924_cast_fp16))[name = tensor("op_43631_cast_fp16")]; + tensor var_43632_to_fp16 = const()[name = tensor("op_43632_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4167_cast_fp16 = mul(x = var_43631_cast_fp16, y = var_43632_to_fp16)[name = tensor("aw_chunk_4167_cast_fp16")]; + tensor var_43635_equation_0 = const()[name = tensor("op_43635_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43635_cast_fp16 = einsum(equation = var_43635_equation_0, values = (var_43465_cast_fp16, var_42931_cast_fp16))[name = tensor("op_43635_cast_fp16")]; + tensor var_43636_to_fp16 = const()[name = tensor("op_43636_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4169_cast_fp16 = mul(x = var_43635_cast_fp16, y = var_43636_to_fp16)[name = tensor("aw_chunk_4169_cast_fp16")]; + tensor var_43639_equation_0 = const()[name = tensor("op_43639_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43639_cast_fp16 = einsum(equation = var_43639_equation_0, values = (var_43465_cast_fp16, var_42938_cast_fp16))[name = tensor("op_43639_cast_fp16")]; + tensor var_43640_to_fp16 = const()[name = tensor("op_43640_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4171_cast_fp16 = mul(x = var_43639_cast_fp16, y = var_43640_to_fp16)[name = tensor("aw_chunk_4171_cast_fp16")]; + tensor var_43643_equation_0 = const()[name = tensor("op_43643_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43643_cast_fp16 = einsum(equation = var_43643_equation_0, values = (var_43465_cast_fp16, var_42945_cast_fp16))[name = tensor("op_43643_cast_fp16")]; + tensor var_43644_to_fp16 = const()[name = tensor("op_43644_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4173_cast_fp16 = mul(x = var_43643_cast_fp16, y = var_43644_to_fp16)[name = tensor("aw_chunk_4173_cast_fp16")]; + tensor var_43647_equation_0 = const()[name = tensor("op_43647_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43647_cast_fp16 = einsum(equation = var_43647_equation_0, values = (var_43465_cast_fp16, var_42952_cast_fp16))[name = tensor("op_43647_cast_fp16")]; + tensor var_43648_to_fp16 = const()[name = tensor("op_43648_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4175_cast_fp16 = mul(x = var_43647_cast_fp16, y = var_43648_to_fp16)[name = tensor("aw_chunk_4175_cast_fp16")]; + tensor var_43651_equation_0 = const()[name = tensor("op_43651_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43651_cast_fp16 = einsum(equation = var_43651_equation_0, values = (var_43469_cast_fp16, var_42959_cast_fp16))[name = tensor("op_43651_cast_fp16")]; + tensor var_43652_to_fp16 = const()[name = tensor("op_43652_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4177_cast_fp16 = mul(x = var_43651_cast_fp16, y = var_43652_to_fp16)[name = tensor("aw_chunk_4177_cast_fp16")]; + tensor var_43655_equation_0 = const()[name = tensor("op_43655_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43655_cast_fp16 = einsum(equation = var_43655_equation_0, values = (var_43469_cast_fp16, var_42966_cast_fp16))[name = tensor("op_43655_cast_fp16")]; + tensor var_43656_to_fp16 = const()[name = tensor("op_43656_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4179_cast_fp16 = mul(x = var_43655_cast_fp16, y = var_43656_to_fp16)[name = tensor("aw_chunk_4179_cast_fp16")]; + tensor var_43659_equation_0 = const()[name = tensor("op_43659_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43659_cast_fp16 = einsum(equation = var_43659_equation_0, values = (var_43469_cast_fp16, var_42973_cast_fp16))[name = tensor("op_43659_cast_fp16")]; + tensor var_43660_to_fp16 = const()[name = tensor("op_43660_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4181_cast_fp16 = mul(x = var_43659_cast_fp16, y = var_43660_to_fp16)[name = tensor("aw_chunk_4181_cast_fp16")]; + tensor var_43663_equation_0 = const()[name = tensor("op_43663_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43663_cast_fp16 = einsum(equation = var_43663_equation_0, values = (var_43469_cast_fp16, var_42980_cast_fp16))[name = tensor("op_43663_cast_fp16")]; + tensor var_43664_to_fp16 = const()[name = tensor("op_43664_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4183_cast_fp16 = mul(x = var_43663_cast_fp16, y = var_43664_to_fp16)[name = tensor("aw_chunk_4183_cast_fp16")]; + tensor var_43667_equation_0 = const()[name = tensor("op_43667_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43667_cast_fp16 = einsum(equation = var_43667_equation_0, values = (var_43473_cast_fp16, var_42987_cast_fp16))[name = tensor("op_43667_cast_fp16")]; + tensor var_43668_to_fp16 = const()[name = tensor("op_43668_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4185_cast_fp16 = mul(x = var_43667_cast_fp16, y = var_43668_to_fp16)[name = tensor("aw_chunk_4185_cast_fp16")]; + tensor var_43671_equation_0 = const()[name = tensor("op_43671_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43671_cast_fp16 = einsum(equation = var_43671_equation_0, values = (var_43473_cast_fp16, var_42994_cast_fp16))[name = tensor("op_43671_cast_fp16")]; + tensor var_43672_to_fp16 = const()[name = tensor("op_43672_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4187_cast_fp16 = mul(x = var_43671_cast_fp16, y = var_43672_to_fp16)[name = tensor("aw_chunk_4187_cast_fp16")]; + tensor var_43675_equation_0 = const()[name = tensor("op_43675_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43675_cast_fp16 = einsum(equation = var_43675_equation_0, values = (var_43473_cast_fp16, var_43001_cast_fp16))[name = tensor("op_43675_cast_fp16")]; + tensor var_43676_to_fp16 = const()[name = tensor("op_43676_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4189_cast_fp16 = mul(x = var_43675_cast_fp16, y = var_43676_to_fp16)[name = tensor("aw_chunk_4189_cast_fp16")]; + tensor var_43679_equation_0 = const()[name = tensor("op_43679_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43679_cast_fp16 = einsum(equation = var_43679_equation_0, values = (var_43473_cast_fp16, var_43008_cast_fp16))[name = tensor("op_43679_cast_fp16")]; + tensor var_43680_to_fp16 = const()[name = tensor("op_43680_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4191_cast_fp16 = mul(x = var_43679_cast_fp16, y = var_43680_to_fp16)[name = tensor("aw_chunk_4191_cast_fp16")]; + tensor var_43683_equation_0 = const()[name = tensor("op_43683_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43683_cast_fp16 = einsum(equation = var_43683_equation_0, values = (var_43477_cast_fp16, var_43015_cast_fp16))[name = tensor("op_43683_cast_fp16")]; + tensor var_43684_to_fp16 = const()[name = tensor("op_43684_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4193_cast_fp16 = mul(x = var_43683_cast_fp16, y = var_43684_to_fp16)[name = tensor("aw_chunk_4193_cast_fp16")]; + tensor var_43687_equation_0 = const()[name = tensor("op_43687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43687_cast_fp16 = einsum(equation = var_43687_equation_0, values = (var_43477_cast_fp16, var_43022_cast_fp16))[name = tensor("op_43687_cast_fp16")]; + tensor var_43688_to_fp16 = const()[name = tensor("op_43688_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4195_cast_fp16 = mul(x = var_43687_cast_fp16, y = var_43688_to_fp16)[name = tensor("aw_chunk_4195_cast_fp16")]; + tensor var_43691_equation_0 = const()[name = tensor("op_43691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43691_cast_fp16 = einsum(equation = var_43691_equation_0, values = (var_43477_cast_fp16, var_43029_cast_fp16))[name = tensor("op_43691_cast_fp16")]; + tensor var_43692_to_fp16 = const()[name = tensor("op_43692_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4197_cast_fp16 = mul(x = var_43691_cast_fp16, y = var_43692_to_fp16)[name = tensor("aw_chunk_4197_cast_fp16")]; + tensor var_43695_equation_0 = const()[name = tensor("op_43695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43695_cast_fp16 = einsum(equation = var_43695_equation_0, values = (var_43477_cast_fp16, var_43036_cast_fp16))[name = tensor("op_43695_cast_fp16")]; + tensor var_43696_to_fp16 = const()[name = tensor("op_43696_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4199_cast_fp16 = mul(x = var_43695_cast_fp16, y = var_43696_to_fp16)[name = tensor("aw_chunk_4199_cast_fp16")]; + tensor var_43699_equation_0 = const()[name = tensor("op_43699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43699_cast_fp16 = einsum(equation = var_43699_equation_0, values = (var_43481_cast_fp16, var_43043_cast_fp16))[name = tensor("op_43699_cast_fp16")]; + tensor var_43700_to_fp16 = const()[name = tensor("op_43700_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4201_cast_fp16 = mul(x = var_43699_cast_fp16, y = var_43700_to_fp16)[name = tensor("aw_chunk_4201_cast_fp16")]; + tensor var_43703_equation_0 = const()[name = tensor("op_43703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43703_cast_fp16 = einsum(equation = var_43703_equation_0, values = (var_43481_cast_fp16, var_43050_cast_fp16))[name = tensor("op_43703_cast_fp16")]; + tensor var_43704_to_fp16 = const()[name = tensor("op_43704_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4203_cast_fp16 = mul(x = var_43703_cast_fp16, y = var_43704_to_fp16)[name = tensor("aw_chunk_4203_cast_fp16")]; + tensor var_43707_equation_0 = const()[name = tensor("op_43707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43707_cast_fp16 = einsum(equation = var_43707_equation_0, values = (var_43481_cast_fp16, var_43057_cast_fp16))[name = tensor("op_43707_cast_fp16")]; + tensor var_43708_to_fp16 = const()[name = tensor("op_43708_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4205_cast_fp16 = mul(x = var_43707_cast_fp16, y = var_43708_to_fp16)[name = tensor("aw_chunk_4205_cast_fp16")]; + tensor var_43711_equation_0 = const()[name = tensor("op_43711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43711_cast_fp16 = einsum(equation = var_43711_equation_0, values = (var_43481_cast_fp16, var_43064_cast_fp16))[name = tensor("op_43711_cast_fp16")]; + tensor var_43712_to_fp16 = const()[name = tensor("op_43712_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4207_cast_fp16 = mul(x = var_43711_cast_fp16, y = var_43712_to_fp16)[name = tensor("aw_chunk_4207_cast_fp16")]; + tensor var_43715_equation_0 = const()[name = tensor("op_43715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43715_cast_fp16 = einsum(equation = var_43715_equation_0, values = (var_43485_cast_fp16, var_43071_cast_fp16))[name = tensor("op_43715_cast_fp16")]; + tensor var_43716_to_fp16 = const()[name = tensor("op_43716_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4209_cast_fp16 = mul(x = var_43715_cast_fp16, y = var_43716_to_fp16)[name = tensor("aw_chunk_4209_cast_fp16")]; + tensor var_43719_equation_0 = const()[name = tensor("op_43719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43719_cast_fp16 = einsum(equation = var_43719_equation_0, values = (var_43485_cast_fp16, var_43078_cast_fp16))[name = tensor("op_43719_cast_fp16")]; + tensor var_43720_to_fp16 = const()[name = tensor("op_43720_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4211_cast_fp16 = mul(x = var_43719_cast_fp16, y = var_43720_to_fp16)[name = tensor("aw_chunk_4211_cast_fp16")]; + tensor var_43723_equation_0 = const()[name = tensor("op_43723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43723_cast_fp16 = einsum(equation = var_43723_equation_0, values = (var_43485_cast_fp16, var_43085_cast_fp16))[name = tensor("op_43723_cast_fp16")]; + tensor var_43724_to_fp16 = const()[name = tensor("op_43724_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4213_cast_fp16 = mul(x = var_43723_cast_fp16, y = var_43724_to_fp16)[name = tensor("aw_chunk_4213_cast_fp16")]; + tensor var_43727_equation_0 = const()[name = tensor("op_43727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43727_cast_fp16 = einsum(equation = var_43727_equation_0, values = (var_43485_cast_fp16, var_43092_cast_fp16))[name = tensor("op_43727_cast_fp16")]; + tensor var_43728_to_fp16 = const()[name = tensor("op_43728_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4215_cast_fp16 = mul(x = var_43727_cast_fp16, y = var_43728_to_fp16)[name = tensor("aw_chunk_4215_cast_fp16")]; + tensor var_43731_equation_0 = const()[name = tensor("op_43731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43731_cast_fp16 = einsum(equation = var_43731_equation_0, values = (var_43489_cast_fp16, var_43099_cast_fp16))[name = tensor("op_43731_cast_fp16")]; + tensor var_43732_to_fp16 = const()[name = tensor("op_43732_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4217_cast_fp16 = mul(x = var_43731_cast_fp16, y = var_43732_to_fp16)[name = tensor("aw_chunk_4217_cast_fp16")]; + tensor var_43735_equation_0 = const()[name = tensor("op_43735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43735_cast_fp16 = einsum(equation = var_43735_equation_0, values = (var_43489_cast_fp16, var_43106_cast_fp16))[name = tensor("op_43735_cast_fp16")]; + tensor var_43736_to_fp16 = const()[name = tensor("op_43736_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4219_cast_fp16 = mul(x = var_43735_cast_fp16, y = var_43736_to_fp16)[name = tensor("aw_chunk_4219_cast_fp16")]; + tensor var_43739_equation_0 = const()[name = tensor("op_43739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43739_cast_fp16 = einsum(equation = var_43739_equation_0, values = (var_43489_cast_fp16, var_43113_cast_fp16))[name = tensor("op_43739_cast_fp16")]; + tensor var_43740_to_fp16 = const()[name = tensor("op_43740_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4221_cast_fp16 = mul(x = var_43739_cast_fp16, y = var_43740_to_fp16)[name = tensor("aw_chunk_4221_cast_fp16")]; + tensor var_43743_equation_0 = const()[name = tensor("op_43743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43743_cast_fp16 = einsum(equation = var_43743_equation_0, values = (var_43489_cast_fp16, var_43120_cast_fp16))[name = tensor("op_43743_cast_fp16")]; + tensor var_43744_to_fp16 = const()[name = tensor("op_43744_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4223_cast_fp16 = mul(x = var_43743_cast_fp16, y = var_43744_to_fp16)[name = tensor("aw_chunk_4223_cast_fp16")]; + tensor var_43747_equation_0 = const()[name = tensor("op_43747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43747_cast_fp16 = einsum(equation = var_43747_equation_0, values = (var_43493_cast_fp16, var_43127_cast_fp16))[name = tensor("op_43747_cast_fp16")]; + tensor var_43748_to_fp16 = const()[name = tensor("op_43748_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4225_cast_fp16 = mul(x = var_43747_cast_fp16, y = var_43748_to_fp16)[name = tensor("aw_chunk_4225_cast_fp16")]; + tensor var_43751_equation_0 = const()[name = tensor("op_43751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43751_cast_fp16 = einsum(equation = var_43751_equation_0, values = (var_43493_cast_fp16, var_43134_cast_fp16))[name = tensor("op_43751_cast_fp16")]; + tensor var_43752_to_fp16 = const()[name = tensor("op_43752_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4227_cast_fp16 = mul(x = var_43751_cast_fp16, y = var_43752_to_fp16)[name = tensor("aw_chunk_4227_cast_fp16")]; + tensor var_43755_equation_0 = const()[name = tensor("op_43755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43755_cast_fp16 = einsum(equation = var_43755_equation_0, values = (var_43493_cast_fp16, var_43141_cast_fp16))[name = tensor("op_43755_cast_fp16")]; + tensor var_43756_to_fp16 = const()[name = tensor("op_43756_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4229_cast_fp16 = mul(x = var_43755_cast_fp16, y = var_43756_to_fp16)[name = tensor("aw_chunk_4229_cast_fp16")]; + tensor var_43759_equation_0 = const()[name = tensor("op_43759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43759_cast_fp16 = einsum(equation = var_43759_equation_0, values = (var_43493_cast_fp16, var_43148_cast_fp16))[name = tensor("op_43759_cast_fp16")]; + tensor var_43760_to_fp16 = const()[name = tensor("op_43760_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4231_cast_fp16 = mul(x = var_43759_cast_fp16, y = var_43760_to_fp16)[name = tensor("aw_chunk_4231_cast_fp16")]; + tensor var_43763_equation_0 = const()[name = tensor("op_43763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43763_cast_fp16 = einsum(equation = var_43763_equation_0, values = (var_43497_cast_fp16, var_43155_cast_fp16))[name = tensor("op_43763_cast_fp16")]; + tensor var_43764_to_fp16 = const()[name = tensor("op_43764_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4233_cast_fp16 = mul(x = var_43763_cast_fp16, y = var_43764_to_fp16)[name = tensor("aw_chunk_4233_cast_fp16")]; + tensor var_43767_equation_0 = const()[name = tensor("op_43767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43767_cast_fp16 = einsum(equation = var_43767_equation_0, values = (var_43497_cast_fp16, var_43162_cast_fp16))[name = tensor("op_43767_cast_fp16")]; + tensor var_43768_to_fp16 = const()[name = tensor("op_43768_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4235_cast_fp16 = mul(x = var_43767_cast_fp16, y = var_43768_to_fp16)[name = tensor("aw_chunk_4235_cast_fp16")]; + tensor var_43771_equation_0 = const()[name = tensor("op_43771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43771_cast_fp16 = einsum(equation = var_43771_equation_0, values = (var_43497_cast_fp16, var_43169_cast_fp16))[name = tensor("op_43771_cast_fp16")]; + tensor var_43772_to_fp16 = const()[name = tensor("op_43772_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4237_cast_fp16 = mul(x = var_43771_cast_fp16, y = var_43772_to_fp16)[name = tensor("aw_chunk_4237_cast_fp16")]; + tensor var_43775_equation_0 = const()[name = tensor("op_43775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43775_cast_fp16 = einsum(equation = var_43775_equation_0, values = (var_43497_cast_fp16, var_43176_cast_fp16))[name = tensor("op_43775_cast_fp16")]; + tensor var_43776_to_fp16 = const()[name = tensor("op_43776_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4239_cast_fp16 = mul(x = var_43775_cast_fp16, y = var_43776_to_fp16)[name = tensor("aw_chunk_4239_cast_fp16")]; + tensor var_43779_equation_0 = const()[name = tensor("op_43779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43779_cast_fp16 = einsum(equation = var_43779_equation_0, values = (var_43501_cast_fp16, var_43183_cast_fp16))[name = tensor("op_43779_cast_fp16")]; + tensor var_43780_to_fp16 = const()[name = tensor("op_43780_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4241_cast_fp16 = mul(x = var_43779_cast_fp16, y = var_43780_to_fp16)[name = tensor("aw_chunk_4241_cast_fp16")]; + tensor var_43783_equation_0 = const()[name = tensor("op_43783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43783_cast_fp16 = einsum(equation = var_43783_equation_0, values = (var_43501_cast_fp16, var_43190_cast_fp16))[name = tensor("op_43783_cast_fp16")]; + tensor var_43784_to_fp16 = const()[name = tensor("op_43784_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4243_cast_fp16 = mul(x = var_43783_cast_fp16, y = var_43784_to_fp16)[name = tensor("aw_chunk_4243_cast_fp16")]; + tensor var_43787_equation_0 = const()[name = tensor("op_43787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43787_cast_fp16 = einsum(equation = var_43787_equation_0, values = (var_43501_cast_fp16, var_43197_cast_fp16))[name = tensor("op_43787_cast_fp16")]; + tensor var_43788_to_fp16 = const()[name = tensor("op_43788_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4245_cast_fp16 = mul(x = var_43787_cast_fp16, y = var_43788_to_fp16)[name = tensor("aw_chunk_4245_cast_fp16")]; + tensor var_43791_equation_0 = const()[name = tensor("op_43791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43791_cast_fp16 = einsum(equation = var_43791_equation_0, values = (var_43501_cast_fp16, var_43204_cast_fp16))[name = tensor("op_43791_cast_fp16")]; + tensor var_43792_to_fp16 = const()[name = tensor("op_43792_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4247_cast_fp16 = mul(x = var_43791_cast_fp16, y = var_43792_to_fp16)[name = tensor("aw_chunk_4247_cast_fp16")]; + tensor var_43795_equation_0 = const()[name = tensor("op_43795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43795_cast_fp16 = einsum(equation = var_43795_equation_0, values = (var_43505_cast_fp16, var_43211_cast_fp16))[name = tensor("op_43795_cast_fp16")]; + tensor var_43796_to_fp16 = const()[name = tensor("op_43796_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4249_cast_fp16 = mul(x = var_43795_cast_fp16, y = var_43796_to_fp16)[name = tensor("aw_chunk_4249_cast_fp16")]; + tensor var_43799_equation_0 = const()[name = tensor("op_43799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43799_cast_fp16 = einsum(equation = var_43799_equation_0, values = (var_43505_cast_fp16, var_43218_cast_fp16))[name = tensor("op_43799_cast_fp16")]; + tensor var_43800_to_fp16 = const()[name = tensor("op_43800_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4251_cast_fp16 = mul(x = var_43799_cast_fp16, y = var_43800_to_fp16)[name = tensor("aw_chunk_4251_cast_fp16")]; + tensor var_43803_equation_0 = const()[name = tensor("op_43803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43803_cast_fp16 = einsum(equation = var_43803_equation_0, values = (var_43505_cast_fp16, var_43225_cast_fp16))[name = tensor("op_43803_cast_fp16")]; + tensor var_43804_to_fp16 = const()[name = tensor("op_43804_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4253_cast_fp16 = mul(x = var_43803_cast_fp16, y = var_43804_to_fp16)[name = tensor("aw_chunk_4253_cast_fp16")]; + tensor var_43807_equation_0 = const()[name = tensor("op_43807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43807_cast_fp16 = einsum(equation = var_43807_equation_0, values = (var_43505_cast_fp16, var_43232_cast_fp16))[name = tensor("op_43807_cast_fp16")]; + tensor var_43808_to_fp16 = const()[name = tensor("op_43808_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4255_cast_fp16 = mul(x = var_43807_cast_fp16, y = var_43808_to_fp16)[name = tensor("aw_chunk_4255_cast_fp16")]; + tensor var_43811_equation_0 = const()[name = tensor("op_43811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43811_cast_fp16 = einsum(equation = var_43811_equation_0, values = (var_43509_cast_fp16, var_43239_cast_fp16))[name = tensor("op_43811_cast_fp16")]; + tensor var_43812_to_fp16 = const()[name = tensor("op_43812_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4257_cast_fp16 = mul(x = var_43811_cast_fp16, y = var_43812_to_fp16)[name = tensor("aw_chunk_4257_cast_fp16")]; + tensor var_43815_equation_0 = const()[name = tensor("op_43815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43815_cast_fp16 = einsum(equation = var_43815_equation_0, values = (var_43509_cast_fp16, var_43246_cast_fp16))[name = tensor("op_43815_cast_fp16")]; + tensor var_43816_to_fp16 = const()[name = tensor("op_43816_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4259_cast_fp16 = mul(x = var_43815_cast_fp16, y = var_43816_to_fp16)[name = tensor("aw_chunk_4259_cast_fp16")]; + tensor var_43819_equation_0 = const()[name = tensor("op_43819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43819_cast_fp16 = einsum(equation = var_43819_equation_0, values = (var_43509_cast_fp16, var_43253_cast_fp16))[name = tensor("op_43819_cast_fp16")]; + tensor var_43820_to_fp16 = const()[name = tensor("op_43820_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4261_cast_fp16 = mul(x = var_43819_cast_fp16, y = var_43820_to_fp16)[name = tensor("aw_chunk_4261_cast_fp16")]; + tensor var_43823_equation_0 = const()[name = tensor("op_43823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43823_cast_fp16 = einsum(equation = var_43823_equation_0, values = (var_43509_cast_fp16, var_43260_cast_fp16))[name = tensor("op_43823_cast_fp16")]; + tensor var_43824_to_fp16 = const()[name = tensor("op_43824_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4263_cast_fp16 = mul(x = var_43823_cast_fp16, y = var_43824_to_fp16)[name = tensor("aw_chunk_4263_cast_fp16")]; + tensor var_43827_equation_0 = const()[name = tensor("op_43827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43827_cast_fp16 = einsum(equation = var_43827_equation_0, values = (var_43513_cast_fp16, var_43267_cast_fp16))[name = tensor("op_43827_cast_fp16")]; + tensor var_43828_to_fp16 = const()[name = tensor("op_43828_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4265_cast_fp16 = mul(x = var_43827_cast_fp16, y = var_43828_to_fp16)[name = tensor("aw_chunk_4265_cast_fp16")]; + tensor var_43831_equation_0 = const()[name = tensor("op_43831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43831_cast_fp16 = einsum(equation = var_43831_equation_0, values = (var_43513_cast_fp16, var_43274_cast_fp16))[name = tensor("op_43831_cast_fp16")]; + tensor var_43832_to_fp16 = const()[name = tensor("op_43832_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4267_cast_fp16 = mul(x = var_43831_cast_fp16, y = var_43832_to_fp16)[name = tensor("aw_chunk_4267_cast_fp16")]; + tensor var_43835_equation_0 = const()[name = tensor("op_43835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43835_cast_fp16 = einsum(equation = var_43835_equation_0, values = (var_43513_cast_fp16, var_43281_cast_fp16))[name = tensor("op_43835_cast_fp16")]; + tensor var_43836_to_fp16 = const()[name = tensor("op_43836_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4269_cast_fp16 = mul(x = var_43835_cast_fp16, y = var_43836_to_fp16)[name = tensor("aw_chunk_4269_cast_fp16")]; + tensor var_43839_equation_0 = const()[name = tensor("op_43839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43839_cast_fp16 = einsum(equation = var_43839_equation_0, values = (var_43513_cast_fp16, var_43288_cast_fp16))[name = tensor("op_43839_cast_fp16")]; + tensor var_43840_to_fp16 = const()[name = tensor("op_43840_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4271_cast_fp16 = mul(x = var_43839_cast_fp16, y = var_43840_to_fp16)[name = tensor("aw_chunk_4271_cast_fp16")]; + tensor var_43843_equation_0 = const()[name = tensor("op_43843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43843_cast_fp16 = einsum(equation = var_43843_equation_0, values = (var_43517_cast_fp16, var_43295_cast_fp16))[name = tensor("op_43843_cast_fp16")]; + tensor var_43844_to_fp16 = const()[name = tensor("op_43844_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4273_cast_fp16 = mul(x = var_43843_cast_fp16, y = var_43844_to_fp16)[name = tensor("aw_chunk_4273_cast_fp16")]; + tensor var_43847_equation_0 = const()[name = tensor("op_43847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43847_cast_fp16 = einsum(equation = var_43847_equation_0, values = (var_43517_cast_fp16, var_43302_cast_fp16))[name = tensor("op_43847_cast_fp16")]; + tensor var_43848_to_fp16 = const()[name = tensor("op_43848_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4275_cast_fp16 = mul(x = var_43847_cast_fp16, y = var_43848_to_fp16)[name = tensor("aw_chunk_4275_cast_fp16")]; + tensor var_43851_equation_0 = const()[name = tensor("op_43851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43851_cast_fp16 = einsum(equation = var_43851_equation_0, values = (var_43517_cast_fp16, var_43309_cast_fp16))[name = tensor("op_43851_cast_fp16")]; + tensor var_43852_to_fp16 = const()[name = tensor("op_43852_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4277_cast_fp16 = mul(x = var_43851_cast_fp16, y = var_43852_to_fp16)[name = tensor("aw_chunk_4277_cast_fp16")]; + tensor var_43855_equation_0 = const()[name = tensor("op_43855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43855_cast_fp16 = einsum(equation = var_43855_equation_0, values = (var_43517_cast_fp16, var_43316_cast_fp16))[name = tensor("op_43855_cast_fp16")]; + tensor var_43856_to_fp16 = const()[name = tensor("op_43856_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4279_cast_fp16 = mul(x = var_43855_cast_fp16, y = var_43856_to_fp16)[name = tensor("aw_chunk_4279_cast_fp16")]; + tensor var_43859_equation_0 = const()[name = tensor("op_43859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43859_cast_fp16 = einsum(equation = var_43859_equation_0, values = (var_43521_cast_fp16, var_43323_cast_fp16))[name = tensor("op_43859_cast_fp16")]; + tensor var_43860_to_fp16 = const()[name = tensor("op_43860_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4281_cast_fp16 = mul(x = var_43859_cast_fp16, y = var_43860_to_fp16)[name = tensor("aw_chunk_4281_cast_fp16")]; + tensor var_43863_equation_0 = const()[name = tensor("op_43863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43863_cast_fp16 = einsum(equation = var_43863_equation_0, values = (var_43521_cast_fp16, var_43330_cast_fp16))[name = tensor("op_43863_cast_fp16")]; + tensor var_43864_to_fp16 = const()[name = tensor("op_43864_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4283_cast_fp16 = mul(x = var_43863_cast_fp16, y = var_43864_to_fp16)[name = tensor("aw_chunk_4283_cast_fp16")]; + tensor var_43867_equation_0 = const()[name = tensor("op_43867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43867_cast_fp16 = einsum(equation = var_43867_equation_0, values = (var_43521_cast_fp16, var_43337_cast_fp16))[name = tensor("op_43867_cast_fp16")]; + tensor var_43868_to_fp16 = const()[name = tensor("op_43868_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4285_cast_fp16 = mul(x = var_43867_cast_fp16, y = var_43868_to_fp16)[name = tensor("aw_chunk_4285_cast_fp16")]; + tensor var_43871_equation_0 = const()[name = tensor("op_43871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43871_cast_fp16 = einsum(equation = var_43871_equation_0, values = (var_43521_cast_fp16, var_43344_cast_fp16))[name = tensor("op_43871_cast_fp16")]; + tensor var_43872_to_fp16 = const()[name = tensor("op_43872_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4287_cast_fp16 = mul(x = var_43871_cast_fp16, y = var_43872_to_fp16)[name = tensor("aw_chunk_4287_cast_fp16")]; + tensor var_43875_equation_0 = const()[name = tensor("op_43875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43875_cast_fp16 = einsum(equation = var_43875_equation_0, values = (var_43525_cast_fp16, var_43351_cast_fp16))[name = tensor("op_43875_cast_fp16")]; + tensor var_43876_to_fp16 = const()[name = tensor("op_43876_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4289_cast_fp16 = mul(x = var_43875_cast_fp16, y = var_43876_to_fp16)[name = tensor("aw_chunk_4289_cast_fp16")]; + tensor var_43879_equation_0 = const()[name = tensor("op_43879_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43879_cast_fp16 = einsum(equation = var_43879_equation_0, values = (var_43525_cast_fp16, var_43358_cast_fp16))[name = tensor("op_43879_cast_fp16")]; + tensor var_43880_to_fp16 = const()[name = tensor("op_43880_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4291_cast_fp16 = mul(x = var_43879_cast_fp16, y = var_43880_to_fp16)[name = tensor("aw_chunk_4291_cast_fp16")]; + tensor var_43883_equation_0 = const()[name = tensor("op_43883_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43883_cast_fp16 = einsum(equation = var_43883_equation_0, values = (var_43525_cast_fp16, var_43365_cast_fp16))[name = tensor("op_43883_cast_fp16")]; + tensor var_43884_to_fp16 = const()[name = tensor("op_43884_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4293_cast_fp16 = mul(x = var_43883_cast_fp16, y = var_43884_to_fp16)[name = tensor("aw_chunk_4293_cast_fp16")]; + tensor var_43887_equation_0 = const()[name = tensor("op_43887_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43887_cast_fp16 = einsum(equation = var_43887_equation_0, values = (var_43525_cast_fp16, var_43372_cast_fp16))[name = tensor("op_43887_cast_fp16")]; + tensor var_43888_to_fp16 = const()[name = tensor("op_43888_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4295_cast_fp16 = mul(x = var_43887_cast_fp16, y = var_43888_to_fp16)[name = tensor("aw_chunk_4295_cast_fp16")]; + tensor var_43891_equation_0 = const()[name = tensor("op_43891_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43891_cast_fp16 = einsum(equation = var_43891_equation_0, values = (var_43529_cast_fp16, var_43379_cast_fp16))[name = tensor("op_43891_cast_fp16")]; + tensor var_43892_to_fp16 = const()[name = tensor("op_43892_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4297_cast_fp16 = mul(x = var_43891_cast_fp16, y = var_43892_to_fp16)[name = tensor("aw_chunk_4297_cast_fp16")]; + tensor var_43895_equation_0 = const()[name = tensor("op_43895_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43895_cast_fp16 = einsum(equation = var_43895_equation_0, values = (var_43529_cast_fp16, var_43386_cast_fp16))[name = tensor("op_43895_cast_fp16")]; + tensor var_43896_to_fp16 = const()[name = tensor("op_43896_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4299_cast_fp16 = mul(x = var_43895_cast_fp16, y = var_43896_to_fp16)[name = tensor("aw_chunk_4299_cast_fp16")]; + tensor var_43899_equation_0 = const()[name = tensor("op_43899_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43899_cast_fp16 = einsum(equation = var_43899_equation_0, values = (var_43529_cast_fp16, var_43393_cast_fp16))[name = tensor("op_43899_cast_fp16")]; + tensor var_43900_to_fp16 = const()[name = tensor("op_43900_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4301_cast_fp16 = mul(x = var_43899_cast_fp16, y = var_43900_to_fp16)[name = tensor("aw_chunk_4301_cast_fp16")]; + tensor var_43903_equation_0 = const()[name = tensor("op_43903_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43903_cast_fp16 = einsum(equation = var_43903_equation_0, values = (var_43529_cast_fp16, var_43400_cast_fp16))[name = tensor("op_43903_cast_fp16")]; + tensor var_43904_to_fp16 = const()[name = tensor("op_43904_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4303_cast_fp16 = mul(x = var_43903_cast_fp16, y = var_43904_to_fp16)[name = tensor("aw_chunk_4303_cast_fp16")]; + tensor var_43907_equation_0 = const()[name = tensor("op_43907_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43907_cast_fp16 = einsum(equation = var_43907_equation_0, values = (var_43533_cast_fp16, var_43407_cast_fp16))[name = tensor("op_43907_cast_fp16")]; + tensor var_43908_to_fp16 = const()[name = tensor("op_43908_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4305_cast_fp16 = mul(x = var_43907_cast_fp16, y = var_43908_to_fp16)[name = tensor("aw_chunk_4305_cast_fp16")]; + tensor var_43911_equation_0 = const()[name = tensor("op_43911_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43911_cast_fp16 = einsum(equation = var_43911_equation_0, values = (var_43533_cast_fp16, var_43414_cast_fp16))[name = tensor("op_43911_cast_fp16")]; + tensor var_43912_to_fp16 = const()[name = tensor("op_43912_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4307_cast_fp16 = mul(x = var_43911_cast_fp16, y = var_43912_to_fp16)[name = tensor("aw_chunk_4307_cast_fp16")]; + tensor var_43915_equation_0 = const()[name = tensor("op_43915_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43915_cast_fp16 = einsum(equation = var_43915_equation_0, values = (var_43533_cast_fp16, var_43421_cast_fp16))[name = tensor("op_43915_cast_fp16")]; + tensor var_43916_to_fp16 = const()[name = tensor("op_43916_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4309_cast_fp16 = mul(x = var_43915_cast_fp16, y = var_43916_to_fp16)[name = tensor("aw_chunk_4309_cast_fp16")]; + tensor var_43919_equation_0 = const()[name = tensor("op_43919_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43919_cast_fp16 = einsum(equation = var_43919_equation_0, values = (var_43533_cast_fp16, var_43428_cast_fp16))[name = tensor("op_43919_cast_fp16")]; + tensor var_43920_to_fp16 = const()[name = tensor("op_43920_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4311_cast_fp16 = mul(x = var_43919_cast_fp16, y = var_43920_to_fp16)[name = tensor("aw_chunk_4311_cast_fp16")]; + tensor var_43923_equation_0 = const()[name = tensor("op_43923_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43923_cast_fp16 = einsum(equation = var_43923_equation_0, values = (var_43537_cast_fp16, var_43435_cast_fp16))[name = tensor("op_43923_cast_fp16")]; + tensor var_43924_to_fp16 = const()[name = tensor("op_43924_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4313_cast_fp16 = mul(x = var_43923_cast_fp16, y = var_43924_to_fp16)[name = tensor("aw_chunk_4313_cast_fp16")]; + tensor var_43927_equation_0 = const()[name = tensor("op_43927_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43927_cast_fp16 = einsum(equation = var_43927_equation_0, values = (var_43537_cast_fp16, var_43442_cast_fp16))[name = tensor("op_43927_cast_fp16")]; + tensor var_43928_to_fp16 = const()[name = tensor("op_43928_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4315_cast_fp16 = mul(x = var_43927_cast_fp16, y = var_43928_to_fp16)[name = tensor("aw_chunk_4315_cast_fp16")]; + tensor var_43931_equation_0 = const()[name = tensor("op_43931_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43931_cast_fp16 = einsum(equation = var_43931_equation_0, values = (var_43537_cast_fp16, var_43449_cast_fp16))[name = tensor("op_43931_cast_fp16")]; + tensor var_43932_to_fp16 = const()[name = tensor("op_43932_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4317_cast_fp16 = mul(x = var_43931_cast_fp16, y = var_43932_to_fp16)[name = tensor("aw_chunk_4317_cast_fp16")]; + tensor var_43935_equation_0 = const()[name = tensor("op_43935_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_43935_cast_fp16 = einsum(equation = var_43935_equation_0, values = (var_43537_cast_fp16, var_43456_cast_fp16))[name = tensor("op_43935_cast_fp16")]; + tensor var_43936_to_fp16 = const()[name = tensor("op_43936_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4319_cast_fp16 = mul(x = var_43935_cast_fp16, y = var_43936_to_fp16)[name = tensor("aw_chunk_4319_cast_fp16")]; + tensor var_43938_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4161_cast_fp16)[name = tensor("op_43938_cast_fp16")]; + tensor var_43939_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4163_cast_fp16)[name = tensor("op_43939_cast_fp16")]; + tensor var_43940_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4165_cast_fp16)[name = tensor("op_43940_cast_fp16")]; + tensor var_43941_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4167_cast_fp16)[name = tensor("op_43941_cast_fp16")]; + tensor var_43942_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4169_cast_fp16)[name = tensor("op_43942_cast_fp16")]; + tensor var_43943_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4171_cast_fp16)[name = tensor("op_43943_cast_fp16")]; + tensor var_43944_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4173_cast_fp16)[name = tensor("op_43944_cast_fp16")]; + tensor var_43945_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4175_cast_fp16)[name = tensor("op_43945_cast_fp16")]; + tensor var_43946_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4177_cast_fp16)[name = tensor("op_43946_cast_fp16")]; + tensor var_43947_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4179_cast_fp16)[name = tensor("op_43947_cast_fp16")]; + tensor var_43948_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4181_cast_fp16)[name = tensor("op_43948_cast_fp16")]; + tensor var_43949_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4183_cast_fp16)[name = tensor("op_43949_cast_fp16")]; + tensor var_43950_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4185_cast_fp16)[name = tensor("op_43950_cast_fp16")]; + tensor var_43951_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4187_cast_fp16)[name = tensor("op_43951_cast_fp16")]; + tensor var_43952_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4189_cast_fp16)[name = tensor("op_43952_cast_fp16")]; + tensor var_43953_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4191_cast_fp16)[name = tensor("op_43953_cast_fp16")]; + tensor var_43954_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4193_cast_fp16)[name = tensor("op_43954_cast_fp16")]; + tensor var_43955_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4195_cast_fp16)[name = tensor("op_43955_cast_fp16")]; + tensor var_43956_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4197_cast_fp16)[name = tensor("op_43956_cast_fp16")]; + tensor var_43957_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4199_cast_fp16)[name = tensor("op_43957_cast_fp16")]; + tensor var_43958_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4201_cast_fp16)[name = tensor("op_43958_cast_fp16")]; + tensor var_43959_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4203_cast_fp16)[name = tensor("op_43959_cast_fp16")]; + tensor var_43960_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4205_cast_fp16)[name = tensor("op_43960_cast_fp16")]; + tensor var_43961_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4207_cast_fp16)[name = tensor("op_43961_cast_fp16")]; + tensor var_43962_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4209_cast_fp16)[name = tensor("op_43962_cast_fp16")]; + tensor var_43963_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4211_cast_fp16)[name = tensor("op_43963_cast_fp16")]; + tensor var_43964_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4213_cast_fp16)[name = tensor("op_43964_cast_fp16")]; + tensor var_43965_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4215_cast_fp16)[name = tensor("op_43965_cast_fp16")]; + tensor var_43966_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4217_cast_fp16)[name = tensor("op_43966_cast_fp16")]; + tensor var_43967_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4219_cast_fp16)[name = tensor("op_43967_cast_fp16")]; + tensor var_43968_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4221_cast_fp16)[name = tensor("op_43968_cast_fp16")]; + tensor var_43969_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4223_cast_fp16)[name = tensor("op_43969_cast_fp16")]; + tensor var_43970_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4225_cast_fp16)[name = tensor("op_43970_cast_fp16")]; + tensor var_43971_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4227_cast_fp16)[name = tensor("op_43971_cast_fp16")]; + tensor var_43972_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4229_cast_fp16)[name = tensor("op_43972_cast_fp16")]; + tensor var_43973_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4231_cast_fp16)[name = tensor("op_43973_cast_fp16")]; + tensor var_43974_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4233_cast_fp16)[name = tensor("op_43974_cast_fp16")]; + tensor var_43975_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4235_cast_fp16)[name = tensor("op_43975_cast_fp16")]; + tensor var_43976_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4237_cast_fp16)[name = tensor("op_43976_cast_fp16")]; + tensor var_43977_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4239_cast_fp16)[name = tensor("op_43977_cast_fp16")]; + tensor var_43978_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4241_cast_fp16)[name = tensor("op_43978_cast_fp16")]; + tensor var_43979_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4243_cast_fp16)[name = tensor("op_43979_cast_fp16")]; + tensor var_43980_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4245_cast_fp16)[name = tensor("op_43980_cast_fp16")]; + tensor var_43981_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4247_cast_fp16)[name = tensor("op_43981_cast_fp16")]; + tensor var_43982_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4249_cast_fp16)[name = tensor("op_43982_cast_fp16")]; + tensor var_43983_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4251_cast_fp16)[name = tensor("op_43983_cast_fp16")]; + tensor var_43984_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4253_cast_fp16)[name = tensor("op_43984_cast_fp16")]; + tensor var_43985_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4255_cast_fp16)[name = tensor("op_43985_cast_fp16")]; + tensor var_43986_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4257_cast_fp16)[name = tensor("op_43986_cast_fp16")]; + tensor var_43987_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4259_cast_fp16)[name = tensor("op_43987_cast_fp16")]; + tensor var_43988_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4261_cast_fp16)[name = tensor("op_43988_cast_fp16")]; + tensor var_43989_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4263_cast_fp16)[name = tensor("op_43989_cast_fp16")]; + tensor var_43990_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4265_cast_fp16)[name = tensor("op_43990_cast_fp16")]; + tensor var_43991_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4267_cast_fp16)[name = tensor("op_43991_cast_fp16")]; + tensor var_43992_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4269_cast_fp16)[name = tensor("op_43992_cast_fp16")]; + tensor var_43993_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4271_cast_fp16)[name = tensor("op_43993_cast_fp16")]; + tensor var_43994_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4273_cast_fp16)[name = tensor("op_43994_cast_fp16")]; + tensor var_43995_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4275_cast_fp16)[name = tensor("op_43995_cast_fp16")]; + tensor var_43996_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4277_cast_fp16)[name = tensor("op_43996_cast_fp16")]; + tensor var_43997_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4279_cast_fp16)[name = tensor("op_43997_cast_fp16")]; + tensor var_43998_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4281_cast_fp16)[name = tensor("op_43998_cast_fp16")]; + tensor var_43999_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4283_cast_fp16)[name = tensor("op_43999_cast_fp16")]; + tensor var_44000_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4285_cast_fp16)[name = tensor("op_44000_cast_fp16")]; + tensor var_44001_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4287_cast_fp16)[name = tensor("op_44001_cast_fp16")]; + tensor var_44002_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4289_cast_fp16)[name = tensor("op_44002_cast_fp16")]; + tensor var_44003_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4291_cast_fp16)[name = tensor("op_44003_cast_fp16")]; + tensor var_44004_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4293_cast_fp16)[name = tensor("op_44004_cast_fp16")]; + tensor var_44005_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4295_cast_fp16)[name = tensor("op_44005_cast_fp16")]; + tensor var_44006_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4297_cast_fp16)[name = tensor("op_44006_cast_fp16")]; + tensor var_44007_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4299_cast_fp16)[name = tensor("op_44007_cast_fp16")]; + tensor var_44008_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4301_cast_fp16)[name = tensor("op_44008_cast_fp16")]; + tensor var_44009_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4303_cast_fp16)[name = tensor("op_44009_cast_fp16")]; + tensor var_44010_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4305_cast_fp16)[name = tensor("op_44010_cast_fp16")]; + tensor var_44011_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4307_cast_fp16)[name = tensor("op_44011_cast_fp16")]; + tensor var_44012_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4309_cast_fp16)[name = tensor("op_44012_cast_fp16")]; + tensor var_44013_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4311_cast_fp16)[name = tensor("op_44013_cast_fp16")]; + tensor var_44014_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4313_cast_fp16)[name = tensor("op_44014_cast_fp16")]; + tensor var_44015_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4315_cast_fp16)[name = tensor("op_44015_cast_fp16")]; + tensor var_44016_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4317_cast_fp16)[name = tensor("op_44016_cast_fp16")]; + tensor var_44017_cast_fp16 = softmax(axis = var_42709, x = aw_chunk_4319_cast_fp16)[name = tensor("op_44017_cast_fp16")]; + tensor var_44019_equation_0 = const()[name = tensor("op_44019_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44019_cast_fp16 = einsum(equation = var_44019_equation_0, values = (var_43539_cast_fp16, var_43938_cast_fp16))[name = tensor("op_44019_cast_fp16")]; + tensor var_44021_equation_0 = const()[name = tensor("op_44021_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44021_cast_fp16 = einsum(equation = var_44021_equation_0, values = (var_43539_cast_fp16, var_43939_cast_fp16))[name = tensor("op_44021_cast_fp16")]; + tensor var_44023_equation_0 = const()[name = tensor("op_44023_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44023_cast_fp16 = einsum(equation = var_44023_equation_0, values = (var_43539_cast_fp16, var_43940_cast_fp16))[name = tensor("op_44023_cast_fp16")]; + tensor var_44025_equation_0 = const()[name = tensor("op_44025_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44025_cast_fp16 = einsum(equation = var_44025_equation_0, values = (var_43539_cast_fp16, var_43941_cast_fp16))[name = tensor("op_44025_cast_fp16")]; + tensor var_44027_equation_0 = const()[name = tensor("op_44027_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44027_cast_fp16 = einsum(equation = var_44027_equation_0, values = (var_43543_cast_fp16, var_43942_cast_fp16))[name = tensor("op_44027_cast_fp16")]; + tensor var_44029_equation_0 = const()[name = tensor("op_44029_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44029_cast_fp16 = einsum(equation = var_44029_equation_0, values = (var_43543_cast_fp16, var_43943_cast_fp16))[name = tensor("op_44029_cast_fp16")]; + tensor var_44031_equation_0 = const()[name = tensor("op_44031_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44031_cast_fp16 = einsum(equation = var_44031_equation_0, values = (var_43543_cast_fp16, var_43944_cast_fp16))[name = tensor("op_44031_cast_fp16")]; + tensor var_44033_equation_0 = const()[name = tensor("op_44033_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44033_cast_fp16 = einsum(equation = var_44033_equation_0, values = (var_43543_cast_fp16, var_43945_cast_fp16))[name = tensor("op_44033_cast_fp16")]; + tensor var_44035_equation_0 = const()[name = tensor("op_44035_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44035_cast_fp16 = einsum(equation = var_44035_equation_0, values = (var_43547_cast_fp16, var_43946_cast_fp16))[name = tensor("op_44035_cast_fp16")]; + tensor var_44037_equation_0 = const()[name = tensor("op_44037_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44037_cast_fp16 = einsum(equation = var_44037_equation_0, values = (var_43547_cast_fp16, var_43947_cast_fp16))[name = tensor("op_44037_cast_fp16")]; + tensor var_44039_equation_0 = const()[name = tensor("op_44039_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44039_cast_fp16 = einsum(equation = var_44039_equation_0, values = (var_43547_cast_fp16, var_43948_cast_fp16))[name = tensor("op_44039_cast_fp16")]; + tensor var_44041_equation_0 = const()[name = tensor("op_44041_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44041_cast_fp16 = einsum(equation = var_44041_equation_0, values = (var_43547_cast_fp16, var_43949_cast_fp16))[name = tensor("op_44041_cast_fp16")]; + tensor var_44043_equation_0 = const()[name = tensor("op_44043_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44043_cast_fp16 = einsum(equation = var_44043_equation_0, values = (var_43551_cast_fp16, var_43950_cast_fp16))[name = tensor("op_44043_cast_fp16")]; + tensor var_44045_equation_0 = const()[name = tensor("op_44045_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44045_cast_fp16 = einsum(equation = var_44045_equation_0, values = (var_43551_cast_fp16, var_43951_cast_fp16))[name = tensor("op_44045_cast_fp16")]; + tensor var_44047_equation_0 = const()[name = tensor("op_44047_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44047_cast_fp16 = einsum(equation = var_44047_equation_0, values = (var_43551_cast_fp16, var_43952_cast_fp16))[name = tensor("op_44047_cast_fp16")]; + tensor var_44049_equation_0 = const()[name = tensor("op_44049_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44049_cast_fp16 = einsum(equation = var_44049_equation_0, values = (var_43551_cast_fp16, var_43953_cast_fp16))[name = tensor("op_44049_cast_fp16")]; + tensor var_44051_equation_0 = const()[name = tensor("op_44051_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44051_cast_fp16 = einsum(equation = var_44051_equation_0, values = (var_43555_cast_fp16, var_43954_cast_fp16))[name = tensor("op_44051_cast_fp16")]; + tensor var_44053_equation_0 = const()[name = tensor("op_44053_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44053_cast_fp16 = einsum(equation = var_44053_equation_0, values = (var_43555_cast_fp16, var_43955_cast_fp16))[name = tensor("op_44053_cast_fp16")]; + tensor var_44055_equation_0 = const()[name = tensor("op_44055_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44055_cast_fp16 = einsum(equation = var_44055_equation_0, values = (var_43555_cast_fp16, var_43956_cast_fp16))[name = tensor("op_44055_cast_fp16")]; + tensor var_44057_equation_0 = const()[name = tensor("op_44057_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44057_cast_fp16 = einsum(equation = var_44057_equation_0, values = (var_43555_cast_fp16, var_43957_cast_fp16))[name = tensor("op_44057_cast_fp16")]; + tensor var_44059_equation_0 = const()[name = tensor("op_44059_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44059_cast_fp16 = einsum(equation = var_44059_equation_0, values = (var_43559_cast_fp16, var_43958_cast_fp16))[name = tensor("op_44059_cast_fp16")]; + tensor var_44061_equation_0 = const()[name = tensor("op_44061_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44061_cast_fp16 = einsum(equation = var_44061_equation_0, values = (var_43559_cast_fp16, var_43959_cast_fp16))[name = tensor("op_44061_cast_fp16")]; + tensor var_44063_equation_0 = const()[name = tensor("op_44063_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44063_cast_fp16 = einsum(equation = var_44063_equation_0, values = (var_43559_cast_fp16, var_43960_cast_fp16))[name = tensor("op_44063_cast_fp16")]; + tensor var_44065_equation_0 = const()[name = tensor("op_44065_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44065_cast_fp16 = einsum(equation = var_44065_equation_0, values = (var_43559_cast_fp16, var_43961_cast_fp16))[name = tensor("op_44065_cast_fp16")]; + tensor var_44067_equation_0 = const()[name = tensor("op_44067_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44067_cast_fp16 = einsum(equation = var_44067_equation_0, values = (var_43563_cast_fp16, var_43962_cast_fp16))[name = tensor("op_44067_cast_fp16")]; + tensor var_44069_equation_0 = const()[name = tensor("op_44069_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44069_cast_fp16 = einsum(equation = var_44069_equation_0, values = (var_43563_cast_fp16, var_43963_cast_fp16))[name = tensor("op_44069_cast_fp16")]; + tensor var_44071_equation_0 = const()[name = tensor("op_44071_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44071_cast_fp16 = einsum(equation = var_44071_equation_0, values = (var_43563_cast_fp16, var_43964_cast_fp16))[name = tensor("op_44071_cast_fp16")]; + tensor var_44073_equation_0 = const()[name = tensor("op_44073_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44073_cast_fp16 = einsum(equation = var_44073_equation_0, values = (var_43563_cast_fp16, var_43965_cast_fp16))[name = tensor("op_44073_cast_fp16")]; + tensor var_44075_equation_0 = const()[name = tensor("op_44075_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44075_cast_fp16 = einsum(equation = var_44075_equation_0, values = (var_43567_cast_fp16, var_43966_cast_fp16))[name = tensor("op_44075_cast_fp16")]; + tensor var_44077_equation_0 = const()[name = tensor("op_44077_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44077_cast_fp16 = einsum(equation = var_44077_equation_0, values = (var_43567_cast_fp16, var_43967_cast_fp16))[name = tensor("op_44077_cast_fp16")]; + tensor var_44079_equation_0 = const()[name = tensor("op_44079_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44079_cast_fp16 = einsum(equation = var_44079_equation_0, values = (var_43567_cast_fp16, var_43968_cast_fp16))[name = tensor("op_44079_cast_fp16")]; + tensor var_44081_equation_0 = const()[name = tensor("op_44081_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44081_cast_fp16 = einsum(equation = var_44081_equation_0, values = (var_43567_cast_fp16, var_43969_cast_fp16))[name = tensor("op_44081_cast_fp16")]; + tensor var_44083_equation_0 = const()[name = tensor("op_44083_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44083_cast_fp16 = einsum(equation = var_44083_equation_0, values = (var_43571_cast_fp16, var_43970_cast_fp16))[name = tensor("op_44083_cast_fp16")]; + tensor var_44085_equation_0 = const()[name = tensor("op_44085_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44085_cast_fp16 = einsum(equation = var_44085_equation_0, values = (var_43571_cast_fp16, var_43971_cast_fp16))[name = tensor("op_44085_cast_fp16")]; + tensor var_44087_equation_0 = const()[name = tensor("op_44087_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44087_cast_fp16 = einsum(equation = var_44087_equation_0, values = (var_43571_cast_fp16, var_43972_cast_fp16))[name = tensor("op_44087_cast_fp16")]; + tensor var_44089_equation_0 = const()[name = tensor("op_44089_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44089_cast_fp16 = einsum(equation = var_44089_equation_0, values = (var_43571_cast_fp16, var_43973_cast_fp16))[name = tensor("op_44089_cast_fp16")]; + tensor var_44091_equation_0 = const()[name = tensor("op_44091_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44091_cast_fp16 = einsum(equation = var_44091_equation_0, values = (var_43575_cast_fp16, var_43974_cast_fp16))[name = tensor("op_44091_cast_fp16")]; + tensor var_44093_equation_0 = const()[name = tensor("op_44093_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44093_cast_fp16 = einsum(equation = var_44093_equation_0, values = (var_43575_cast_fp16, var_43975_cast_fp16))[name = tensor("op_44093_cast_fp16")]; + tensor var_44095_equation_0 = const()[name = tensor("op_44095_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44095_cast_fp16 = einsum(equation = var_44095_equation_0, values = (var_43575_cast_fp16, var_43976_cast_fp16))[name = tensor("op_44095_cast_fp16")]; + tensor var_44097_equation_0 = const()[name = tensor("op_44097_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44097_cast_fp16 = einsum(equation = var_44097_equation_0, values = (var_43575_cast_fp16, var_43977_cast_fp16))[name = tensor("op_44097_cast_fp16")]; + tensor var_44099_equation_0 = const()[name = tensor("op_44099_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44099_cast_fp16 = einsum(equation = var_44099_equation_0, values = (var_43579_cast_fp16, var_43978_cast_fp16))[name = tensor("op_44099_cast_fp16")]; + tensor var_44101_equation_0 = const()[name = tensor("op_44101_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44101_cast_fp16 = einsum(equation = var_44101_equation_0, values = (var_43579_cast_fp16, var_43979_cast_fp16))[name = tensor("op_44101_cast_fp16")]; + tensor var_44103_equation_0 = const()[name = tensor("op_44103_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44103_cast_fp16 = einsum(equation = var_44103_equation_0, values = (var_43579_cast_fp16, var_43980_cast_fp16))[name = tensor("op_44103_cast_fp16")]; + tensor var_44105_equation_0 = const()[name = tensor("op_44105_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44105_cast_fp16 = einsum(equation = var_44105_equation_0, values = (var_43579_cast_fp16, var_43981_cast_fp16))[name = tensor("op_44105_cast_fp16")]; + tensor var_44107_equation_0 = const()[name = tensor("op_44107_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44107_cast_fp16 = einsum(equation = var_44107_equation_0, values = (var_43583_cast_fp16, var_43982_cast_fp16))[name = tensor("op_44107_cast_fp16")]; + tensor var_44109_equation_0 = const()[name = tensor("op_44109_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44109_cast_fp16 = einsum(equation = var_44109_equation_0, values = (var_43583_cast_fp16, var_43983_cast_fp16))[name = tensor("op_44109_cast_fp16")]; + tensor var_44111_equation_0 = const()[name = tensor("op_44111_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44111_cast_fp16 = einsum(equation = var_44111_equation_0, values = (var_43583_cast_fp16, var_43984_cast_fp16))[name = tensor("op_44111_cast_fp16")]; + tensor var_44113_equation_0 = const()[name = tensor("op_44113_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44113_cast_fp16 = einsum(equation = var_44113_equation_0, values = (var_43583_cast_fp16, var_43985_cast_fp16))[name = tensor("op_44113_cast_fp16")]; + tensor var_44115_equation_0 = const()[name = tensor("op_44115_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44115_cast_fp16 = einsum(equation = var_44115_equation_0, values = (var_43587_cast_fp16, var_43986_cast_fp16))[name = tensor("op_44115_cast_fp16")]; + tensor var_44117_equation_0 = const()[name = tensor("op_44117_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44117_cast_fp16 = einsum(equation = var_44117_equation_0, values = (var_43587_cast_fp16, var_43987_cast_fp16))[name = tensor("op_44117_cast_fp16")]; + tensor var_44119_equation_0 = const()[name = tensor("op_44119_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44119_cast_fp16 = einsum(equation = var_44119_equation_0, values = (var_43587_cast_fp16, var_43988_cast_fp16))[name = tensor("op_44119_cast_fp16")]; + tensor var_44121_equation_0 = const()[name = tensor("op_44121_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44121_cast_fp16 = einsum(equation = var_44121_equation_0, values = (var_43587_cast_fp16, var_43989_cast_fp16))[name = tensor("op_44121_cast_fp16")]; + tensor var_44123_equation_0 = const()[name = tensor("op_44123_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44123_cast_fp16 = einsum(equation = var_44123_equation_0, values = (var_43591_cast_fp16, var_43990_cast_fp16))[name = tensor("op_44123_cast_fp16")]; + tensor var_44125_equation_0 = const()[name = tensor("op_44125_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44125_cast_fp16 = einsum(equation = var_44125_equation_0, values = (var_43591_cast_fp16, var_43991_cast_fp16))[name = tensor("op_44125_cast_fp16")]; + tensor var_44127_equation_0 = const()[name = tensor("op_44127_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44127_cast_fp16 = einsum(equation = var_44127_equation_0, values = (var_43591_cast_fp16, var_43992_cast_fp16))[name = tensor("op_44127_cast_fp16")]; + tensor var_44129_equation_0 = const()[name = tensor("op_44129_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44129_cast_fp16 = einsum(equation = var_44129_equation_0, values = (var_43591_cast_fp16, var_43993_cast_fp16))[name = tensor("op_44129_cast_fp16")]; + tensor var_44131_equation_0 = const()[name = tensor("op_44131_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44131_cast_fp16 = einsum(equation = var_44131_equation_0, values = (var_43595_cast_fp16, var_43994_cast_fp16))[name = tensor("op_44131_cast_fp16")]; + tensor var_44133_equation_0 = const()[name = tensor("op_44133_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44133_cast_fp16 = einsum(equation = var_44133_equation_0, values = (var_43595_cast_fp16, var_43995_cast_fp16))[name = tensor("op_44133_cast_fp16")]; + tensor var_44135_equation_0 = const()[name = tensor("op_44135_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44135_cast_fp16 = einsum(equation = var_44135_equation_0, values = (var_43595_cast_fp16, var_43996_cast_fp16))[name = tensor("op_44135_cast_fp16")]; + tensor var_44137_equation_0 = const()[name = tensor("op_44137_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44137_cast_fp16 = einsum(equation = var_44137_equation_0, values = (var_43595_cast_fp16, var_43997_cast_fp16))[name = tensor("op_44137_cast_fp16")]; + tensor var_44139_equation_0 = const()[name = tensor("op_44139_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44139_cast_fp16 = einsum(equation = var_44139_equation_0, values = (var_43599_cast_fp16, var_43998_cast_fp16))[name = tensor("op_44139_cast_fp16")]; + tensor var_44141_equation_0 = const()[name = tensor("op_44141_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44141_cast_fp16 = einsum(equation = var_44141_equation_0, values = (var_43599_cast_fp16, var_43999_cast_fp16))[name = tensor("op_44141_cast_fp16")]; + tensor var_44143_equation_0 = const()[name = tensor("op_44143_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44143_cast_fp16 = einsum(equation = var_44143_equation_0, values = (var_43599_cast_fp16, var_44000_cast_fp16))[name = tensor("op_44143_cast_fp16")]; + tensor var_44145_equation_0 = const()[name = tensor("op_44145_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44145_cast_fp16 = einsum(equation = var_44145_equation_0, values = (var_43599_cast_fp16, var_44001_cast_fp16))[name = tensor("op_44145_cast_fp16")]; + tensor var_44147_equation_0 = const()[name = tensor("op_44147_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44147_cast_fp16 = einsum(equation = var_44147_equation_0, values = (var_43603_cast_fp16, var_44002_cast_fp16))[name = tensor("op_44147_cast_fp16")]; + tensor var_44149_equation_0 = const()[name = tensor("op_44149_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44149_cast_fp16 = einsum(equation = var_44149_equation_0, values = (var_43603_cast_fp16, var_44003_cast_fp16))[name = tensor("op_44149_cast_fp16")]; + tensor var_44151_equation_0 = const()[name = tensor("op_44151_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44151_cast_fp16 = einsum(equation = var_44151_equation_0, values = (var_43603_cast_fp16, var_44004_cast_fp16))[name = tensor("op_44151_cast_fp16")]; + tensor var_44153_equation_0 = const()[name = tensor("op_44153_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44153_cast_fp16 = einsum(equation = var_44153_equation_0, values = (var_43603_cast_fp16, var_44005_cast_fp16))[name = tensor("op_44153_cast_fp16")]; + tensor var_44155_equation_0 = const()[name = tensor("op_44155_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44155_cast_fp16 = einsum(equation = var_44155_equation_0, values = (var_43607_cast_fp16, var_44006_cast_fp16))[name = tensor("op_44155_cast_fp16")]; + tensor var_44157_equation_0 = const()[name = tensor("op_44157_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44157_cast_fp16 = einsum(equation = var_44157_equation_0, values = (var_43607_cast_fp16, var_44007_cast_fp16))[name = tensor("op_44157_cast_fp16")]; + tensor var_44159_equation_0 = const()[name = tensor("op_44159_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44159_cast_fp16 = einsum(equation = var_44159_equation_0, values = (var_43607_cast_fp16, var_44008_cast_fp16))[name = tensor("op_44159_cast_fp16")]; + tensor var_44161_equation_0 = const()[name = tensor("op_44161_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44161_cast_fp16 = einsum(equation = var_44161_equation_0, values = (var_43607_cast_fp16, var_44009_cast_fp16))[name = tensor("op_44161_cast_fp16")]; + tensor var_44163_equation_0 = const()[name = tensor("op_44163_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44163_cast_fp16 = einsum(equation = var_44163_equation_0, values = (var_43611_cast_fp16, var_44010_cast_fp16))[name = tensor("op_44163_cast_fp16")]; + tensor var_44165_equation_0 = const()[name = tensor("op_44165_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44165_cast_fp16 = einsum(equation = var_44165_equation_0, values = (var_43611_cast_fp16, var_44011_cast_fp16))[name = tensor("op_44165_cast_fp16")]; + tensor var_44167_equation_0 = const()[name = tensor("op_44167_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44167_cast_fp16 = einsum(equation = var_44167_equation_0, values = (var_43611_cast_fp16, var_44012_cast_fp16))[name = tensor("op_44167_cast_fp16")]; + tensor var_44169_equation_0 = const()[name = tensor("op_44169_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44169_cast_fp16 = einsum(equation = var_44169_equation_0, values = (var_43611_cast_fp16, var_44013_cast_fp16))[name = tensor("op_44169_cast_fp16")]; + tensor var_44171_equation_0 = const()[name = tensor("op_44171_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44171_cast_fp16 = einsum(equation = var_44171_equation_0, values = (var_43615_cast_fp16, var_44014_cast_fp16))[name = tensor("op_44171_cast_fp16")]; + tensor var_44173_equation_0 = const()[name = tensor("op_44173_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44173_cast_fp16 = einsum(equation = var_44173_equation_0, values = (var_43615_cast_fp16, var_44015_cast_fp16))[name = tensor("op_44173_cast_fp16")]; + tensor var_44175_equation_0 = const()[name = tensor("op_44175_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44175_cast_fp16 = einsum(equation = var_44175_equation_0, values = (var_43615_cast_fp16, var_44016_cast_fp16))[name = tensor("op_44175_cast_fp16")]; + tensor var_44177_equation_0 = const()[name = tensor("op_44177_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_44177_cast_fp16 = einsum(equation = var_44177_equation_0, values = (var_43615_cast_fp16, var_44017_cast_fp16))[name = tensor("op_44177_cast_fp16")]; + tensor var_44179_interleave_0 = const()[name = tensor("op_44179_interleave_0"), val = tensor(false)]; + tensor var_44179_cast_fp16 = concat(axis = var_42684, interleave = var_44179_interleave_0, values = (var_44019_cast_fp16, var_44021_cast_fp16, var_44023_cast_fp16, var_44025_cast_fp16))[name = tensor("op_44179_cast_fp16")]; + tensor var_44181_interleave_0 = const()[name = tensor("op_44181_interleave_0"), val = tensor(false)]; + tensor var_44181_cast_fp16 = concat(axis = var_42684, interleave = var_44181_interleave_0, values = (var_44027_cast_fp16, var_44029_cast_fp16, var_44031_cast_fp16, var_44033_cast_fp16))[name = tensor("op_44181_cast_fp16")]; + tensor var_44183_interleave_0 = const()[name = tensor("op_44183_interleave_0"), val = tensor(false)]; + tensor var_44183_cast_fp16 = concat(axis = var_42684, interleave = var_44183_interleave_0, values = (var_44035_cast_fp16, var_44037_cast_fp16, var_44039_cast_fp16, var_44041_cast_fp16))[name = tensor("op_44183_cast_fp16")]; + tensor var_44185_interleave_0 = const()[name = tensor("op_44185_interleave_0"), val = tensor(false)]; + tensor var_44185_cast_fp16 = concat(axis = var_42684, interleave = var_44185_interleave_0, values = (var_44043_cast_fp16, var_44045_cast_fp16, var_44047_cast_fp16, var_44049_cast_fp16))[name = tensor("op_44185_cast_fp16")]; + tensor var_44187_interleave_0 = const()[name = tensor("op_44187_interleave_0"), val = tensor(false)]; + tensor var_44187_cast_fp16 = concat(axis = var_42684, interleave = var_44187_interleave_0, values = (var_44051_cast_fp16, var_44053_cast_fp16, var_44055_cast_fp16, var_44057_cast_fp16))[name = tensor("op_44187_cast_fp16")]; + tensor var_44189_interleave_0 = const()[name = tensor("op_44189_interleave_0"), val = tensor(false)]; + tensor var_44189_cast_fp16 = concat(axis = var_42684, interleave = var_44189_interleave_0, values = (var_44059_cast_fp16, var_44061_cast_fp16, var_44063_cast_fp16, var_44065_cast_fp16))[name = tensor("op_44189_cast_fp16")]; + tensor var_44191_interleave_0 = const()[name = tensor("op_44191_interleave_0"), val = tensor(false)]; + tensor var_44191_cast_fp16 = concat(axis = var_42684, interleave = var_44191_interleave_0, values = (var_44067_cast_fp16, var_44069_cast_fp16, var_44071_cast_fp16, var_44073_cast_fp16))[name = tensor("op_44191_cast_fp16")]; + tensor var_44193_interleave_0 = const()[name = tensor("op_44193_interleave_0"), val = tensor(false)]; + tensor var_44193_cast_fp16 = concat(axis = var_42684, interleave = var_44193_interleave_0, values = (var_44075_cast_fp16, var_44077_cast_fp16, var_44079_cast_fp16, var_44081_cast_fp16))[name = tensor("op_44193_cast_fp16")]; + tensor var_44195_interleave_0 = const()[name = tensor("op_44195_interleave_0"), val = tensor(false)]; + tensor var_44195_cast_fp16 = concat(axis = var_42684, interleave = var_44195_interleave_0, values = (var_44083_cast_fp16, var_44085_cast_fp16, var_44087_cast_fp16, var_44089_cast_fp16))[name = tensor("op_44195_cast_fp16")]; + tensor var_44197_interleave_0 = const()[name = tensor("op_44197_interleave_0"), val = tensor(false)]; + tensor var_44197_cast_fp16 = concat(axis = var_42684, interleave = var_44197_interleave_0, values = (var_44091_cast_fp16, var_44093_cast_fp16, var_44095_cast_fp16, var_44097_cast_fp16))[name = tensor("op_44197_cast_fp16")]; + tensor var_44199_interleave_0 = const()[name = tensor("op_44199_interleave_0"), val = tensor(false)]; + tensor var_44199_cast_fp16 = concat(axis = var_42684, interleave = var_44199_interleave_0, values = (var_44099_cast_fp16, var_44101_cast_fp16, var_44103_cast_fp16, var_44105_cast_fp16))[name = tensor("op_44199_cast_fp16")]; + tensor var_44201_interleave_0 = const()[name = tensor("op_44201_interleave_0"), val = tensor(false)]; + tensor var_44201_cast_fp16 = concat(axis = var_42684, interleave = var_44201_interleave_0, values = (var_44107_cast_fp16, var_44109_cast_fp16, var_44111_cast_fp16, var_44113_cast_fp16))[name = tensor("op_44201_cast_fp16")]; + tensor var_44203_interleave_0 = const()[name = tensor("op_44203_interleave_0"), val = tensor(false)]; + tensor var_44203_cast_fp16 = concat(axis = var_42684, interleave = var_44203_interleave_0, values = (var_44115_cast_fp16, var_44117_cast_fp16, var_44119_cast_fp16, var_44121_cast_fp16))[name = tensor("op_44203_cast_fp16")]; + tensor var_44205_interleave_0 = const()[name = tensor("op_44205_interleave_0"), val = tensor(false)]; + tensor var_44205_cast_fp16 = concat(axis = var_42684, interleave = var_44205_interleave_0, values = (var_44123_cast_fp16, var_44125_cast_fp16, var_44127_cast_fp16, var_44129_cast_fp16))[name = tensor("op_44205_cast_fp16")]; + tensor var_44207_interleave_0 = const()[name = tensor("op_44207_interleave_0"), val = tensor(false)]; + tensor var_44207_cast_fp16 = concat(axis = var_42684, interleave = var_44207_interleave_0, values = (var_44131_cast_fp16, var_44133_cast_fp16, var_44135_cast_fp16, var_44137_cast_fp16))[name = tensor("op_44207_cast_fp16")]; + tensor var_44209_interleave_0 = const()[name = tensor("op_44209_interleave_0"), val = tensor(false)]; + tensor var_44209_cast_fp16 = concat(axis = var_42684, interleave = var_44209_interleave_0, values = (var_44139_cast_fp16, var_44141_cast_fp16, var_44143_cast_fp16, var_44145_cast_fp16))[name = tensor("op_44209_cast_fp16")]; + tensor var_44211_interleave_0 = const()[name = tensor("op_44211_interleave_0"), val = tensor(false)]; + tensor var_44211_cast_fp16 = concat(axis = var_42684, interleave = var_44211_interleave_0, values = (var_44147_cast_fp16, var_44149_cast_fp16, var_44151_cast_fp16, var_44153_cast_fp16))[name = tensor("op_44211_cast_fp16")]; + tensor var_44213_interleave_0 = const()[name = tensor("op_44213_interleave_0"), val = tensor(false)]; + tensor var_44213_cast_fp16 = concat(axis = var_42684, interleave = var_44213_interleave_0, values = (var_44155_cast_fp16, var_44157_cast_fp16, var_44159_cast_fp16, var_44161_cast_fp16))[name = tensor("op_44213_cast_fp16")]; + tensor var_44215_interleave_0 = const()[name = tensor("op_44215_interleave_0"), val = tensor(false)]; + tensor var_44215_cast_fp16 = concat(axis = var_42684, interleave = var_44215_interleave_0, values = (var_44163_cast_fp16, var_44165_cast_fp16, var_44167_cast_fp16, var_44169_cast_fp16))[name = tensor("op_44215_cast_fp16")]; + tensor var_44217_interleave_0 = const()[name = tensor("op_44217_interleave_0"), val = tensor(false)]; + tensor var_44217_cast_fp16 = concat(axis = var_42684, interleave = var_44217_interleave_0, values = (var_44171_cast_fp16, var_44173_cast_fp16, var_44175_cast_fp16, var_44177_cast_fp16))[name = tensor("op_44217_cast_fp16")]; + tensor input_527_interleave_0 = const()[name = tensor("input_527_interleave_0"), val = tensor(false)]; + tensor input_527_cast_fp16 = concat(axis = var_42709, interleave = input_527_interleave_0, values = (var_44179_cast_fp16, var_44181_cast_fp16, var_44183_cast_fp16, var_44185_cast_fp16, var_44187_cast_fp16, var_44189_cast_fp16, var_44191_cast_fp16, var_44193_cast_fp16, var_44195_cast_fp16, var_44197_cast_fp16, var_44199_cast_fp16, var_44201_cast_fp16, var_44203_cast_fp16, var_44205_cast_fp16, var_44207_cast_fp16, var_44209_cast_fp16, var_44211_cast_fp16, var_44213_cast_fp16, var_44215_cast_fp16, var_44217_cast_fp16))[name = tensor("input_527_cast_fp16")]; + tensor var_44225 = const()[name = tensor("op_44225"), val = tensor([1, 1])]; + tensor var_44227 = const()[name = tensor("op_44227"), val = tensor([1, 1])]; + tensor pretrained_out_319_pad_type_0 = const()[name = tensor("pretrained_out_319_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_319_pad_0 = const()[name = tensor("pretrained_out_319_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292643968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293463232))), name = tensor("layers_26_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_26_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_26_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293463360)))]; + tensor pretrained_out_319_cast_fp16 = conv(bias = layers_26_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_44227, groups = var_42709, pad = pretrained_out_319_pad_0, pad_type = pretrained_out_319_pad_type_0, strides = var_44225, weight = layers_26_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_527_cast_fp16)[name = tensor("pretrained_out_319_cast_fp16")]; + tensor var_44231 = const()[name = tensor("op_44231"), val = tensor([1, 1])]; + tensor var_44233 = const()[name = tensor("op_44233"), val = tensor([1, 1])]; + tensor input_529_pad_type_0 = const()[name = tensor("input_529_pad_type_0"), val = tensor("custom")]; + tensor input_529_pad_0 = const()[name = tensor("input_529_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_26_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293465984)))]; + tensor input_529_cast_fp16 = conv(dilations = var_44233, groups = var_42709, pad = input_529_pad_0, pad_type = input_529_pad_type_0, strides = var_44231, weight = layers_26_self_attn_o_proj_loraA_weight_to_fp16, x = input_527_cast_fp16)[name = tensor("input_529_cast_fp16")]; + tensor var_44237 = const()[name = tensor("op_44237"), val = tensor([1, 1])]; + tensor var_44239 = const()[name = tensor("op_44239"), val = tensor([1, 1])]; + tensor lora_out_637_pad_type_0 = const()[name = tensor("lora_out_637_pad_type_0"), val = tensor("custom")]; + tensor lora_out_637_pad_0 = const()[name = tensor("lora_out_637_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_639_weight_0_to_fp16 = const()[name = tensor("lora_out_639_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293507008)))]; + tensor lora_out_639_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_44239, groups = var_42709, pad = lora_out_637_pad_0, pad_type = lora_out_637_pad_type_0, strides = var_44237, weight = lora_out_639_weight_0_to_fp16, x = input_529_cast_fp16)[name = tensor("lora_out_639_cast_fp16")]; + tensor obj_107_cast_fp16 = add(x = pretrained_out_319_cast_fp16, y = lora_out_639_cast_fp16)[name = tensor("obj_107_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = obj_107_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; + tensor var_44248 = const()[name = tensor("op_44248"), val = tensor([1])]; + tensor channels_mean_107_cast_fp16 = reduce_mean(axes = var_44248, keep_dims = var_42710, x = inputs_107_cast_fp16)[name = tensor("channels_mean_107_cast_fp16")]; + tensor zero_mean_107_cast_fp16 = sub(x = inputs_107_cast_fp16, y = channels_mean_107_cast_fp16)[name = tensor("zero_mean_107_cast_fp16")]; + tensor zero_mean_sq_107_cast_fp16 = mul(x = zero_mean_107_cast_fp16, y = zero_mean_107_cast_fp16)[name = tensor("zero_mean_sq_107_cast_fp16")]; + tensor var_44252 = const()[name = tensor("op_44252"), val = tensor([1])]; + tensor var_44253_cast_fp16 = reduce_mean(axes = var_44252, keep_dims = var_42710, x = zero_mean_sq_107_cast_fp16)[name = tensor("op_44253_cast_fp16")]; + tensor var_44254_to_fp16 = const()[name = tensor("op_44254_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_44255_cast_fp16 = add(x = var_44253_cast_fp16, y = var_44254_to_fp16)[name = tensor("op_44255_cast_fp16")]; + tensor denom_107_epsilon_0 = const()[name = tensor("denom_107_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_107_cast_fp16 = rsqrt(epsilon = denom_107_epsilon_0, x = var_44255_cast_fp16)[name = tensor("denom_107_cast_fp16")]; + tensor out_107_cast_fp16 = mul(x = zero_mean_107_cast_fp16, y = denom_107_cast_fp16)[name = tensor("out_107_cast_fp16")]; + tensor input_531_gamma_0_to_fp16 = const()[name = tensor("input_531_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293548032)))]; + tensor input_531_beta_0_to_fp16 = const()[name = tensor("input_531_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293550656)))]; + tensor input_531_epsilon_0_to_fp16 = const()[name = tensor("input_531_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_531_cast_fp16 = batch_norm(beta = input_531_beta_0_to_fp16, epsilon = input_531_epsilon_0_to_fp16, gamma = input_531_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_107_cast_fp16)[name = tensor("input_531_cast_fp16")]; + tensor var_44269 = const()[name = tensor("op_44269"), val = tensor([1, 1])]; + tensor var_44271 = const()[name = tensor("op_44271"), val = tensor([1, 1])]; + tensor pretrained_out_321_pad_type_0 = const()[name = tensor("pretrained_out_321_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_321_pad_0 = const()[name = tensor("pretrained_out_321_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293553280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296830144))), name = tensor("layers_26_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_26_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_26_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296830272)))]; + tensor pretrained_out_321_cast_fp16 = conv(bias = layers_26_fc1_pretrained_bias_to_fp16, dilations = var_44271, groups = var_42709, pad = pretrained_out_321_pad_0, pad_type = pretrained_out_321_pad_type_0, strides = var_44269, weight = layers_26_fc1_pretrained_weight_to_fp16_palettized, x = input_531_cast_fp16)[name = tensor("pretrained_out_321_cast_fp16")]; + tensor var_44275 = const()[name = tensor("op_44275"), val = tensor([1, 1])]; + tensor var_44277 = const()[name = tensor("op_44277"), val = tensor([1, 1])]; + tensor input_533_pad_type_0 = const()[name = tensor("input_533_pad_type_0"), val = tensor("custom")]; + tensor input_533_pad_0 = const()[name = tensor("input_533_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_26_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296840576)))]; + tensor input_533_cast_fp16 = conv(dilations = var_44277, groups = var_42709, pad = input_533_pad_0, pad_type = input_533_pad_type_0, strides = var_44275, weight = layers_26_fc1_loraA_weight_to_fp16, x = input_531_cast_fp16)[name = tensor("input_533_cast_fp16")]; + tensor var_44281 = const()[name = tensor("op_44281"), val = tensor([1, 1])]; + tensor var_44283 = const()[name = tensor("op_44283"), val = tensor([1, 1])]; + tensor lora_out_641_pad_type_0 = const()[name = tensor("lora_out_641_pad_type_0"), val = tensor("custom")]; + tensor lora_out_641_pad_0 = const()[name = tensor("lora_out_641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_643_weight_0_to_fp16 = const()[name = tensor("lora_out_643_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296881600)))]; + tensor lora_out_643_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_44283, groups = var_42709, pad = lora_out_641_pad_0, pad_type = lora_out_641_pad_type_0, strides = var_44281, weight = lora_out_643_weight_0_to_fp16, x = input_533_cast_fp16)[name = tensor("lora_out_643_cast_fp16")]; + tensor input_535_cast_fp16 = add(x = pretrained_out_321_cast_fp16, y = lora_out_643_cast_fp16)[name = tensor("input_535_cast_fp16")]; + tensor input_537_mode_0 = const()[name = tensor("input_537_mode_0"), val = tensor("EXACT")]; + tensor input_537_cast_fp16 = gelu(mode = input_537_mode_0, x = input_535_cast_fp16)[name = tensor("input_537_cast_fp16")]; + tensor var_44295 = const()[name = tensor("op_44295"), val = tensor([1, 1])]; + tensor var_44297 = const()[name = tensor("op_44297"), val = tensor([1, 1])]; + tensor pretrained_out_323_pad_type_0 = const()[name = tensor("pretrained_out_323_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_323_pad_0 = const()[name = tensor("pretrained_out_323_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(297045504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300322368))), name = tensor("layers_26_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_26_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_26_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300322496)))]; + tensor pretrained_out_323_cast_fp16 = conv(bias = layers_26_fc2_pretrained_bias_to_fp16, dilations = var_44297, groups = var_42709, pad = pretrained_out_323_pad_0, pad_type = pretrained_out_323_pad_type_0, strides = var_44295, weight = layers_26_fc2_pretrained_weight_to_fp16_palettized, x = input_537_cast_fp16)[name = tensor("pretrained_out_323_cast_fp16")]; + tensor var_44301 = const()[name = tensor("op_44301"), val = tensor([1, 1])]; + tensor var_44303 = const()[name = tensor("op_44303"), val = tensor([1, 1])]; + tensor input_539_pad_type_0 = const()[name = tensor("input_539_pad_type_0"), val = tensor("custom")]; + tensor input_539_pad_0 = const()[name = tensor("input_539_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_26_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_26_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300325120)))]; + tensor input_539_cast_fp16 = conv(dilations = var_44303, groups = var_42709, pad = input_539_pad_0, pad_type = input_539_pad_type_0, strides = var_44301, weight = layers_26_fc2_loraA_weight_to_fp16, x = input_537_cast_fp16)[name = tensor("input_539_cast_fp16")]; + tensor var_44307 = const()[name = tensor("op_44307"), val = tensor([1, 1])]; + tensor var_44309 = const()[name = tensor("op_44309"), val = tensor([1, 1])]; + tensor lora_out_645_pad_type_0 = const()[name = tensor("lora_out_645_pad_type_0"), val = tensor("custom")]; + tensor lora_out_645_pad_0 = const()[name = tensor("lora_out_645_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_647_weight_0_to_fp16 = const()[name = tensor("lora_out_647_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300489024)))]; + tensor lora_out_647_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_44309, groups = var_42709, pad = lora_out_645_pad_0, pad_type = lora_out_645_pad_type_0, strides = var_44307, weight = lora_out_647_weight_0_to_fp16, x = input_539_cast_fp16)[name = tensor("lora_out_647_cast_fp16")]; + tensor hidden_states_57_cast_fp16 = add(x = pretrained_out_323_cast_fp16, y = lora_out_647_cast_fp16)[name = tensor("hidden_states_57_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = hidden_states_57_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; + tensor var_44319 = const()[name = tensor("op_44319"), val = tensor(3)]; + tensor var_44344 = const()[name = tensor("op_44344"), val = tensor(1)]; + tensor var_44345 = const()[name = tensor("op_44345"), val = tensor(true)]; + tensor var_44355 = const()[name = tensor("op_44355"), val = tensor([1])]; + tensor channels_mean_109_cast_fp16 = reduce_mean(axes = var_44355, keep_dims = var_44345, x = inputs_109_cast_fp16)[name = tensor("channels_mean_109_cast_fp16")]; + tensor zero_mean_109_cast_fp16 = sub(x = inputs_109_cast_fp16, y = channels_mean_109_cast_fp16)[name = tensor("zero_mean_109_cast_fp16")]; + tensor zero_mean_sq_109_cast_fp16 = mul(x = zero_mean_109_cast_fp16, y = zero_mean_109_cast_fp16)[name = tensor("zero_mean_sq_109_cast_fp16")]; + tensor var_44359 = const()[name = tensor("op_44359"), val = tensor([1])]; + tensor var_44360_cast_fp16 = reduce_mean(axes = var_44359, keep_dims = var_44345, x = zero_mean_sq_109_cast_fp16)[name = tensor("op_44360_cast_fp16")]; + tensor var_44361_to_fp16 = const()[name = tensor("op_44361_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_44362_cast_fp16 = add(x = var_44360_cast_fp16, y = var_44361_to_fp16)[name = tensor("op_44362_cast_fp16")]; + tensor denom_109_epsilon_0 = const()[name = tensor("denom_109_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_109_cast_fp16 = rsqrt(epsilon = denom_109_epsilon_0, x = var_44362_cast_fp16)[name = tensor("denom_109_cast_fp16")]; + tensor out_109_cast_fp16 = mul(x = zero_mean_109_cast_fp16, y = denom_109_cast_fp16)[name = tensor("out_109_cast_fp16")]; + tensor obj_109_gamma_0_to_fp16 = const()[name = tensor("obj_109_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300530048)))]; + tensor obj_109_beta_0_to_fp16 = const()[name = tensor("obj_109_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300532672)))]; + tensor obj_109_epsilon_0_to_fp16 = const()[name = tensor("obj_109_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_109_cast_fp16 = batch_norm(beta = obj_109_beta_0_to_fp16, epsilon = obj_109_epsilon_0_to_fp16, gamma = obj_109_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_109_cast_fp16)[name = tensor("obj_109_cast_fp16")]; + tensor var_44380 = const()[name = tensor("op_44380"), val = tensor([1, 1])]; + tensor var_44382 = const()[name = tensor("op_44382"), val = tensor([1, 1])]; + tensor pretrained_out_325_pad_type_0 = const()[name = tensor("pretrained_out_325_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_325_pad_0 = const()[name = tensor("pretrained_out_325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300535296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301354560))), name = tensor("layers_27_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_27_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301354688)))]; + tensor pretrained_out_325_cast_fp16 = conv(bias = layers_27_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_44382, groups = var_44344, pad = pretrained_out_325_pad_0, pad_type = pretrained_out_325_pad_type_0, strides = var_44380, weight = layers_27_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_109_cast_fp16)[name = tensor("pretrained_out_325_cast_fp16")]; + tensor var_44386 = const()[name = tensor("op_44386"), val = tensor([1, 1])]; + tensor var_44388 = const()[name = tensor("op_44388"), val = tensor([1, 1])]; + tensor input_541_pad_type_0 = const()[name = tensor("input_541_pad_type_0"), val = tensor("custom")]; + tensor input_541_pad_0 = const()[name = tensor("input_541_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301357312)))]; + tensor input_541_cast_fp16 = conv(dilations = var_44388, groups = var_44344, pad = input_541_pad_0, pad_type = input_541_pad_type_0, strides = var_44386, weight = layers_27_self_attn_q_proj_loraA_weight_to_fp16, x = obj_109_cast_fp16)[name = tensor("input_541_cast_fp16")]; + tensor var_44392 = const()[name = tensor("op_44392"), val = tensor([1, 1])]; + tensor var_44394 = const()[name = tensor("op_44394"), val = tensor([1, 1])]; + tensor lora_out_649_pad_type_0 = const()[name = tensor("lora_out_649_pad_type_0"), val = tensor("custom")]; + tensor lora_out_649_pad_0 = const()[name = tensor("lora_out_649_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_651_weight_0_to_fp16 = const()[name = tensor("lora_out_651_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301398336)))]; + tensor lora_out_651_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_44394, groups = var_44344, pad = lora_out_649_pad_0, pad_type = lora_out_649_pad_type_0, strides = var_44392, weight = lora_out_651_weight_0_to_fp16, x = input_541_cast_fp16)[name = tensor("lora_out_651_cast_fp16")]; + tensor query_55_cast_fp16 = add(x = pretrained_out_325_cast_fp16, y = lora_out_651_cast_fp16)[name = tensor("query_55_cast_fp16")]; + tensor var_44404 = const()[name = tensor("op_44404"), val = tensor([1, 1])]; + tensor var_44406 = const()[name = tensor("op_44406"), val = tensor([1, 1])]; + tensor pretrained_out_327_pad_type_0 = const()[name = tensor("pretrained_out_327_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_327_pad_0 = const()[name = tensor("pretrained_out_327_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(301439360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302258624))), name = tensor("layers_27_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_327_cast_fp16 = conv(dilations = var_44406, groups = var_44344, pad = pretrained_out_327_pad_0, pad_type = pretrained_out_327_pad_type_0, strides = var_44404, weight = layers_27_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_109_cast_fp16)[name = tensor("pretrained_out_327_cast_fp16")]; + tensor var_44410 = const()[name = tensor("op_44410"), val = tensor([1, 1])]; + tensor var_44412 = const()[name = tensor("op_44412"), val = tensor([1, 1])]; + tensor input_543_pad_type_0 = const()[name = tensor("input_543_pad_type_0"), val = tensor("custom")]; + tensor input_543_pad_0 = const()[name = tensor("input_543_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302258752)))]; + tensor input_543_cast_fp16 = conv(dilations = var_44412, groups = var_44344, pad = input_543_pad_0, pad_type = input_543_pad_type_0, strides = var_44410, weight = layers_27_self_attn_k_proj_loraA_weight_to_fp16, x = obj_109_cast_fp16)[name = tensor("input_543_cast_fp16")]; + tensor var_44416 = const()[name = tensor("op_44416"), val = tensor([1, 1])]; + tensor var_44418 = const()[name = tensor("op_44418"), val = tensor([1, 1])]; + tensor lora_out_653_pad_type_0 = const()[name = tensor("lora_out_653_pad_type_0"), val = tensor("custom")]; + tensor lora_out_653_pad_0 = const()[name = tensor("lora_out_653_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_655_weight_0_to_fp16 = const()[name = tensor("lora_out_655_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302299776)))]; + tensor lora_out_655_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_44418, groups = var_44344, pad = lora_out_653_pad_0, pad_type = lora_out_653_pad_type_0, strides = var_44416, weight = lora_out_655_weight_0_to_fp16, x = input_543_cast_fp16)[name = tensor("lora_out_655_cast_fp16")]; + tensor key_55_cast_fp16 = add(x = pretrained_out_327_cast_fp16, y = lora_out_655_cast_fp16)[name = tensor("key_55_cast_fp16")]; + tensor var_44429 = const()[name = tensor("op_44429"), val = tensor([1, 1])]; + tensor var_44431 = const()[name = tensor("op_44431"), val = tensor([1, 1])]; + tensor pretrained_out_329_pad_type_0 = const()[name = tensor("pretrained_out_329_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_329_pad_0 = const()[name = tensor("pretrained_out_329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302340800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303160064))), name = tensor("layers_27_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_27_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303160192)))]; + tensor pretrained_out_329_cast_fp16 = conv(bias = layers_27_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_44431, groups = var_44344, pad = pretrained_out_329_pad_0, pad_type = pretrained_out_329_pad_type_0, strides = var_44429, weight = layers_27_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_109_cast_fp16)[name = tensor("pretrained_out_329_cast_fp16")]; + tensor var_44435 = const()[name = tensor("op_44435"), val = tensor([1, 1])]; + tensor var_44437 = const()[name = tensor("op_44437"), val = tensor([1, 1])]; + tensor input_545_pad_type_0 = const()[name = tensor("input_545_pad_type_0"), val = tensor("custom")]; + tensor input_545_pad_0 = const()[name = tensor("input_545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303162816)))]; + tensor input_545_cast_fp16 = conv(dilations = var_44437, groups = var_44344, pad = input_545_pad_0, pad_type = input_545_pad_type_0, strides = var_44435, weight = layers_27_self_attn_v_proj_loraA_weight_to_fp16, x = obj_109_cast_fp16)[name = tensor("input_545_cast_fp16")]; + tensor var_44441 = const()[name = tensor("op_44441"), val = tensor([1, 1])]; + tensor var_44443 = const()[name = tensor("op_44443"), val = tensor([1, 1])]; + tensor lora_out_657_pad_type_0 = const()[name = tensor("lora_out_657_pad_type_0"), val = tensor("custom")]; + tensor lora_out_657_pad_0 = const()[name = tensor("lora_out_657_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_659_weight_0_to_fp16 = const()[name = tensor("lora_out_659_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303203840)))]; + tensor lora_out_659_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_44443, groups = var_44344, pad = lora_out_657_pad_0, pad_type = lora_out_657_pad_type_0, strides = var_44441, weight = lora_out_659_weight_0_to_fp16, x = input_545_cast_fp16)[name = tensor("lora_out_659_cast_fp16")]; + tensor value_55_cast_fp16 = add(x = pretrained_out_329_cast_fp16, y = lora_out_659_cast_fp16)[name = tensor("value_55_cast_fp16")]; + tensor var_44453_begin_0 = const()[name = tensor("op_44453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44453_end_0 = const()[name = tensor("op_44453_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44453_end_mask_0 = const()[name = tensor("op_44453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44453_cast_fp16 = slice_by_index(begin = var_44453_begin_0, end = var_44453_end_0, end_mask = var_44453_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44453_cast_fp16")]; + tensor var_44457_begin_0 = const()[name = tensor("op_44457_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_44457_end_0 = const()[name = tensor("op_44457_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_44457_end_mask_0 = const()[name = tensor("op_44457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44457_cast_fp16 = slice_by_index(begin = var_44457_begin_0, end = var_44457_end_0, end_mask = var_44457_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44457_cast_fp16")]; + tensor var_44461_begin_0 = const()[name = tensor("op_44461_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_44461_end_0 = const()[name = tensor("op_44461_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_44461_end_mask_0 = const()[name = tensor("op_44461_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44461_cast_fp16 = slice_by_index(begin = var_44461_begin_0, end = var_44461_end_0, end_mask = var_44461_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44461_cast_fp16")]; + tensor var_44465_begin_0 = const()[name = tensor("op_44465_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_44465_end_0 = const()[name = tensor("op_44465_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_44465_end_mask_0 = const()[name = tensor("op_44465_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44465_cast_fp16 = slice_by_index(begin = var_44465_begin_0, end = var_44465_end_0, end_mask = var_44465_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44465_cast_fp16")]; + tensor var_44469_begin_0 = const()[name = tensor("op_44469_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_44469_end_0 = const()[name = tensor("op_44469_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_44469_end_mask_0 = const()[name = tensor("op_44469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44469_cast_fp16 = slice_by_index(begin = var_44469_begin_0, end = var_44469_end_0, end_mask = var_44469_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44469_cast_fp16")]; + tensor var_44473_begin_0 = const()[name = tensor("op_44473_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_44473_end_0 = const()[name = tensor("op_44473_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_44473_end_mask_0 = const()[name = tensor("op_44473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44473_cast_fp16 = slice_by_index(begin = var_44473_begin_0, end = var_44473_end_0, end_mask = var_44473_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44473_cast_fp16")]; + tensor var_44477_begin_0 = const()[name = tensor("op_44477_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_44477_end_0 = const()[name = tensor("op_44477_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_44477_end_mask_0 = const()[name = tensor("op_44477_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44477_cast_fp16 = slice_by_index(begin = var_44477_begin_0, end = var_44477_end_0, end_mask = var_44477_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44477_cast_fp16")]; + tensor var_44481_begin_0 = const()[name = tensor("op_44481_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_44481_end_0 = const()[name = tensor("op_44481_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_44481_end_mask_0 = const()[name = tensor("op_44481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44481_cast_fp16 = slice_by_index(begin = var_44481_begin_0, end = var_44481_end_0, end_mask = var_44481_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44481_cast_fp16")]; + tensor var_44485_begin_0 = const()[name = tensor("op_44485_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_44485_end_0 = const()[name = tensor("op_44485_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_44485_end_mask_0 = const()[name = tensor("op_44485_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44485_cast_fp16 = slice_by_index(begin = var_44485_begin_0, end = var_44485_end_0, end_mask = var_44485_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44485_cast_fp16")]; + tensor var_44489_begin_0 = const()[name = tensor("op_44489_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_44489_end_0 = const()[name = tensor("op_44489_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_44489_end_mask_0 = const()[name = tensor("op_44489_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44489_cast_fp16 = slice_by_index(begin = var_44489_begin_0, end = var_44489_end_0, end_mask = var_44489_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44489_cast_fp16")]; + tensor var_44493_begin_0 = const()[name = tensor("op_44493_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_44493_end_0 = const()[name = tensor("op_44493_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_44493_end_mask_0 = const()[name = tensor("op_44493_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44493_cast_fp16 = slice_by_index(begin = var_44493_begin_0, end = var_44493_end_0, end_mask = var_44493_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44493_cast_fp16")]; + tensor var_44497_begin_0 = const()[name = tensor("op_44497_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_44497_end_0 = const()[name = tensor("op_44497_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_44497_end_mask_0 = const()[name = tensor("op_44497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44497_cast_fp16 = slice_by_index(begin = var_44497_begin_0, end = var_44497_end_0, end_mask = var_44497_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44497_cast_fp16")]; + tensor var_44501_begin_0 = const()[name = tensor("op_44501_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_44501_end_0 = const()[name = tensor("op_44501_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_44501_end_mask_0 = const()[name = tensor("op_44501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44501_cast_fp16 = slice_by_index(begin = var_44501_begin_0, end = var_44501_end_0, end_mask = var_44501_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44501_cast_fp16")]; + tensor var_44505_begin_0 = const()[name = tensor("op_44505_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_44505_end_0 = const()[name = tensor("op_44505_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_44505_end_mask_0 = const()[name = tensor("op_44505_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44505_cast_fp16 = slice_by_index(begin = var_44505_begin_0, end = var_44505_end_0, end_mask = var_44505_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44505_cast_fp16")]; + tensor var_44509_begin_0 = const()[name = tensor("op_44509_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_44509_end_0 = const()[name = tensor("op_44509_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_44509_end_mask_0 = const()[name = tensor("op_44509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44509_cast_fp16 = slice_by_index(begin = var_44509_begin_0, end = var_44509_end_0, end_mask = var_44509_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44509_cast_fp16")]; + tensor var_44513_begin_0 = const()[name = tensor("op_44513_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_44513_end_0 = const()[name = tensor("op_44513_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_44513_end_mask_0 = const()[name = tensor("op_44513_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44513_cast_fp16 = slice_by_index(begin = var_44513_begin_0, end = var_44513_end_0, end_mask = var_44513_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44513_cast_fp16")]; + tensor var_44517_begin_0 = const()[name = tensor("op_44517_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_44517_end_0 = const()[name = tensor("op_44517_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_44517_end_mask_0 = const()[name = tensor("op_44517_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44517_cast_fp16 = slice_by_index(begin = var_44517_begin_0, end = var_44517_end_0, end_mask = var_44517_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44517_cast_fp16")]; + tensor var_44521_begin_0 = const()[name = tensor("op_44521_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_44521_end_0 = const()[name = tensor("op_44521_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_44521_end_mask_0 = const()[name = tensor("op_44521_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44521_cast_fp16 = slice_by_index(begin = var_44521_begin_0, end = var_44521_end_0, end_mask = var_44521_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44521_cast_fp16")]; + tensor var_44525_begin_0 = const()[name = tensor("op_44525_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_44525_end_0 = const()[name = tensor("op_44525_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_44525_end_mask_0 = const()[name = tensor("op_44525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44525_cast_fp16 = slice_by_index(begin = var_44525_begin_0, end = var_44525_end_0, end_mask = var_44525_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44525_cast_fp16")]; + tensor var_44529_begin_0 = const()[name = tensor("op_44529_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_44529_end_0 = const()[name = tensor("op_44529_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_44529_end_mask_0 = const()[name = tensor("op_44529_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_44529_cast_fp16 = slice_by_index(begin = var_44529_begin_0, end = var_44529_end_0, end_mask = var_44529_end_mask_0, x = query_55_cast_fp16)[name = tensor("op_44529_cast_fp16")]; + tensor var_44538_begin_0 = const()[name = tensor("op_44538_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44538_end_0 = const()[name = tensor("op_44538_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44538_end_mask_0 = const()[name = tensor("op_44538_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44538_cast_fp16 = slice_by_index(begin = var_44538_begin_0, end = var_44538_end_0, end_mask = var_44538_end_mask_0, x = var_44453_cast_fp16)[name = tensor("op_44538_cast_fp16")]; + tensor var_44545_begin_0 = const()[name = tensor("op_44545_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44545_end_0 = const()[name = tensor("op_44545_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44545_end_mask_0 = const()[name = tensor("op_44545_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44545_cast_fp16 = slice_by_index(begin = var_44545_begin_0, end = var_44545_end_0, end_mask = var_44545_end_mask_0, x = var_44453_cast_fp16)[name = tensor("op_44545_cast_fp16")]; + tensor var_44552_begin_0 = const()[name = tensor("op_44552_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44552_end_0 = const()[name = tensor("op_44552_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44552_end_mask_0 = const()[name = tensor("op_44552_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44552_cast_fp16 = slice_by_index(begin = var_44552_begin_0, end = var_44552_end_0, end_mask = var_44552_end_mask_0, x = var_44453_cast_fp16)[name = tensor("op_44552_cast_fp16")]; + tensor var_44559_begin_0 = const()[name = tensor("op_44559_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44559_end_0 = const()[name = tensor("op_44559_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44559_end_mask_0 = const()[name = tensor("op_44559_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44559_cast_fp16 = slice_by_index(begin = var_44559_begin_0, end = var_44559_end_0, end_mask = var_44559_end_mask_0, x = var_44453_cast_fp16)[name = tensor("op_44559_cast_fp16")]; + tensor var_44566_begin_0 = const()[name = tensor("op_44566_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44566_end_0 = const()[name = tensor("op_44566_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44566_end_mask_0 = const()[name = tensor("op_44566_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44566_cast_fp16 = slice_by_index(begin = var_44566_begin_0, end = var_44566_end_0, end_mask = var_44566_end_mask_0, x = var_44457_cast_fp16)[name = tensor("op_44566_cast_fp16")]; + tensor var_44573_begin_0 = const()[name = tensor("op_44573_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44573_end_0 = const()[name = tensor("op_44573_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44573_end_mask_0 = const()[name = tensor("op_44573_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44573_cast_fp16 = slice_by_index(begin = var_44573_begin_0, end = var_44573_end_0, end_mask = var_44573_end_mask_0, x = var_44457_cast_fp16)[name = tensor("op_44573_cast_fp16")]; + tensor var_44580_begin_0 = const()[name = tensor("op_44580_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44580_end_0 = const()[name = tensor("op_44580_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44580_end_mask_0 = const()[name = tensor("op_44580_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44580_cast_fp16 = slice_by_index(begin = var_44580_begin_0, end = var_44580_end_0, end_mask = var_44580_end_mask_0, x = var_44457_cast_fp16)[name = tensor("op_44580_cast_fp16")]; + tensor var_44587_begin_0 = const()[name = tensor("op_44587_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44587_end_0 = const()[name = tensor("op_44587_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44587_end_mask_0 = const()[name = tensor("op_44587_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44587_cast_fp16 = slice_by_index(begin = var_44587_begin_0, end = var_44587_end_0, end_mask = var_44587_end_mask_0, x = var_44457_cast_fp16)[name = tensor("op_44587_cast_fp16")]; + tensor var_44594_begin_0 = const()[name = tensor("op_44594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44594_end_0 = const()[name = tensor("op_44594_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44594_end_mask_0 = const()[name = tensor("op_44594_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44594_cast_fp16 = slice_by_index(begin = var_44594_begin_0, end = var_44594_end_0, end_mask = var_44594_end_mask_0, x = var_44461_cast_fp16)[name = tensor("op_44594_cast_fp16")]; + tensor var_44601_begin_0 = const()[name = tensor("op_44601_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44601_end_0 = const()[name = tensor("op_44601_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44601_end_mask_0 = const()[name = tensor("op_44601_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44601_cast_fp16 = slice_by_index(begin = var_44601_begin_0, end = var_44601_end_0, end_mask = var_44601_end_mask_0, x = var_44461_cast_fp16)[name = tensor("op_44601_cast_fp16")]; + tensor var_44608_begin_0 = const()[name = tensor("op_44608_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44608_end_0 = const()[name = tensor("op_44608_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44608_end_mask_0 = const()[name = tensor("op_44608_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44608_cast_fp16 = slice_by_index(begin = var_44608_begin_0, end = var_44608_end_0, end_mask = var_44608_end_mask_0, x = var_44461_cast_fp16)[name = tensor("op_44608_cast_fp16")]; + tensor var_44615_begin_0 = const()[name = tensor("op_44615_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44615_end_0 = const()[name = tensor("op_44615_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44615_end_mask_0 = const()[name = tensor("op_44615_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44615_cast_fp16 = slice_by_index(begin = var_44615_begin_0, end = var_44615_end_0, end_mask = var_44615_end_mask_0, x = var_44461_cast_fp16)[name = tensor("op_44615_cast_fp16")]; + tensor var_44622_begin_0 = const()[name = tensor("op_44622_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44622_end_0 = const()[name = tensor("op_44622_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44622_end_mask_0 = const()[name = tensor("op_44622_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44622_cast_fp16 = slice_by_index(begin = var_44622_begin_0, end = var_44622_end_0, end_mask = var_44622_end_mask_0, x = var_44465_cast_fp16)[name = tensor("op_44622_cast_fp16")]; + tensor var_44629_begin_0 = const()[name = tensor("op_44629_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44629_end_0 = const()[name = tensor("op_44629_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44629_end_mask_0 = const()[name = tensor("op_44629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44629_cast_fp16 = slice_by_index(begin = var_44629_begin_0, end = var_44629_end_0, end_mask = var_44629_end_mask_0, x = var_44465_cast_fp16)[name = tensor("op_44629_cast_fp16")]; + tensor var_44636_begin_0 = const()[name = tensor("op_44636_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44636_end_0 = const()[name = tensor("op_44636_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44636_end_mask_0 = const()[name = tensor("op_44636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44636_cast_fp16 = slice_by_index(begin = var_44636_begin_0, end = var_44636_end_0, end_mask = var_44636_end_mask_0, x = var_44465_cast_fp16)[name = tensor("op_44636_cast_fp16")]; + tensor var_44643_begin_0 = const()[name = tensor("op_44643_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44643_end_0 = const()[name = tensor("op_44643_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44643_end_mask_0 = const()[name = tensor("op_44643_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44643_cast_fp16 = slice_by_index(begin = var_44643_begin_0, end = var_44643_end_0, end_mask = var_44643_end_mask_0, x = var_44465_cast_fp16)[name = tensor("op_44643_cast_fp16")]; + tensor var_44650_begin_0 = const()[name = tensor("op_44650_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44650_end_0 = const()[name = tensor("op_44650_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44650_end_mask_0 = const()[name = tensor("op_44650_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44650_cast_fp16 = slice_by_index(begin = var_44650_begin_0, end = var_44650_end_0, end_mask = var_44650_end_mask_0, x = var_44469_cast_fp16)[name = tensor("op_44650_cast_fp16")]; + tensor var_44657_begin_0 = const()[name = tensor("op_44657_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44657_end_0 = const()[name = tensor("op_44657_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44657_end_mask_0 = const()[name = tensor("op_44657_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44657_cast_fp16 = slice_by_index(begin = var_44657_begin_0, end = var_44657_end_0, end_mask = var_44657_end_mask_0, x = var_44469_cast_fp16)[name = tensor("op_44657_cast_fp16")]; + tensor var_44664_begin_0 = const()[name = tensor("op_44664_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44664_end_0 = const()[name = tensor("op_44664_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44664_end_mask_0 = const()[name = tensor("op_44664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44664_cast_fp16 = slice_by_index(begin = var_44664_begin_0, end = var_44664_end_0, end_mask = var_44664_end_mask_0, x = var_44469_cast_fp16)[name = tensor("op_44664_cast_fp16")]; + tensor var_44671_begin_0 = const()[name = tensor("op_44671_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44671_end_0 = const()[name = tensor("op_44671_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44671_end_mask_0 = const()[name = tensor("op_44671_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44671_cast_fp16 = slice_by_index(begin = var_44671_begin_0, end = var_44671_end_0, end_mask = var_44671_end_mask_0, x = var_44469_cast_fp16)[name = tensor("op_44671_cast_fp16")]; + tensor var_44678_begin_0 = const()[name = tensor("op_44678_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44678_end_0 = const()[name = tensor("op_44678_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44678_end_mask_0 = const()[name = tensor("op_44678_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44678_cast_fp16 = slice_by_index(begin = var_44678_begin_0, end = var_44678_end_0, end_mask = var_44678_end_mask_0, x = var_44473_cast_fp16)[name = tensor("op_44678_cast_fp16")]; + tensor var_44685_begin_0 = const()[name = tensor("op_44685_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44685_end_0 = const()[name = tensor("op_44685_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44685_end_mask_0 = const()[name = tensor("op_44685_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44685_cast_fp16 = slice_by_index(begin = var_44685_begin_0, end = var_44685_end_0, end_mask = var_44685_end_mask_0, x = var_44473_cast_fp16)[name = tensor("op_44685_cast_fp16")]; + tensor var_44692_begin_0 = const()[name = tensor("op_44692_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44692_end_0 = const()[name = tensor("op_44692_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44692_end_mask_0 = const()[name = tensor("op_44692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44692_cast_fp16 = slice_by_index(begin = var_44692_begin_0, end = var_44692_end_0, end_mask = var_44692_end_mask_0, x = var_44473_cast_fp16)[name = tensor("op_44692_cast_fp16")]; + tensor var_44699_begin_0 = const()[name = tensor("op_44699_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44699_end_0 = const()[name = tensor("op_44699_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44699_end_mask_0 = const()[name = tensor("op_44699_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44699_cast_fp16 = slice_by_index(begin = var_44699_begin_0, end = var_44699_end_0, end_mask = var_44699_end_mask_0, x = var_44473_cast_fp16)[name = tensor("op_44699_cast_fp16")]; + tensor var_44706_begin_0 = const()[name = tensor("op_44706_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44706_end_0 = const()[name = tensor("op_44706_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44706_end_mask_0 = const()[name = tensor("op_44706_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44706_cast_fp16 = slice_by_index(begin = var_44706_begin_0, end = var_44706_end_0, end_mask = var_44706_end_mask_0, x = var_44477_cast_fp16)[name = tensor("op_44706_cast_fp16")]; + tensor var_44713_begin_0 = const()[name = tensor("op_44713_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44713_end_0 = const()[name = tensor("op_44713_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44713_end_mask_0 = const()[name = tensor("op_44713_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44713_cast_fp16 = slice_by_index(begin = var_44713_begin_0, end = var_44713_end_0, end_mask = var_44713_end_mask_0, x = var_44477_cast_fp16)[name = tensor("op_44713_cast_fp16")]; + tensor var_44720_begin_0 = const()[name = tensor("op_44720_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44720_end_0 = const()[name = tensor("op_44720_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44720_end_mask_0 = const()[name = tensor("op_44720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44720_cast_fp16 = slice_by_index(begin = var_44720_begin_0, end = var_44720_end_0, end_mask = var_44720_end_mask_0, x = var_44477_cast_fp16)[name = tensor("op_44720_cast_fp16")]; + tensor var_44727_begin_0 = const()[name = tensor("op_44727_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44727_end_0 = const()[name = tensor("op_44727_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44727_end_mask_0 = const()[name = tensor("op_44727_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44727_cast_fp16 = slice_by_index(begin = var_44727_begin_0, end = var_44727_end_0, end_mask = var_44727_end_mask_0, x = var_44477_cast_fp16)[name = tensor("op_44727_cast_fp16")]; + tensor var_44734_begin_0 = const()[name = tensor("op_44734_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44734_end_0 = const()[name = tensor("op_44734_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44734_end_mask_0 = const()[name = tensor("op_44734_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44734_cast_fp16 = slice_by_index(begin = var_44734_begin_0, end = var_44734_end_0, end_mask = var_44734_end_mask_0, x = var_44481_cast_fp16)[name = tensor("op_44734_cast_fp16")]; + tensor var_44741_begin_0 = const()[name = tensor("op_44741_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44741_end_0 = const()[name = tensor("op_44741_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44741_end_mask_0 = const()[name = tensor("op_44741_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44741_cast_fp16 = slice_by_index(begin = var_44741_begin_0, end = var_44741_end_0, end_mask = var_44741_end_mask_0, x = var_44481_cast_fp16)[name = tensor("op_44741_cast_fp16")]; + tensor var_44748_begin_0 = const()[name = tensor("op_44748_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44748_end_0 = const()[name = tensor("op_44748_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44748_end_mask_0 = const()[name = tensor("op_44748_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44748_cast_fp16 = slice_by_index(begin = var_44748_begin_0, end = var_44748_end_0, end_mask = var_44748_end_mask_0, x = var_44481_cast_fp16)[name = tensor("op_44748_cast_fp16")]; + tensor var_44755_begin_0 = const()[name = tensor("op_44755_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44755_end_0 = const()[name = tensor("op_44755_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44755_end_mask_0 = const()[name = tensor("op_44755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44755_cast_fp16 = slice_by_index(begin = var_44755_begin_0, end = var_44755_end_0, end_mask = var_44755_end_mask_0, x = var_44481_cast_fp16)[name = tensor("op_44755_cast_fp16")]; + tensor var_44762_begin_0 = const()[name = tensor("op_44762_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44762_end_0 = const()[name = tensor("op_44762_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44762_end_mask_0 = const()[name = tensor("op_44762_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44762_cast_fp16 = slice_by_index(begin = var_44762_begin_0, end = var_44762_end_0, end_mask = var_44762_end_mask_0, x = var_44485_cast_fp16)[name = tensor("op_44762_cast_fp16")]; + tensor var_44769_begin_0 = const()[name = tensor("op_44769_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44769_end_0 = const()[name = tensor("op_44769_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44769_end_mask_0 = const()[name = tensor("op_44769_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44769_cast_fp16 = slice_by_index(begin = var_44769_begin_0, end = var_44769_end_0, end_mask = var_44769_end_mask_0, x = var_44485_cast_fp16)[name = tensor("op_44769_cast_fp16")]; + tensor var_44776_begin_0 = const()[name = tensor("op_44776_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44776_end_0 = const()[name = tensor("op_44776_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44776_end_mask_0 = const()[name = tensor("op_44776_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44776_cast_fp16 = slice_by_index(begin = var_44776_begin_0, end = var_44776_end_0, end_mask = var_44776_end_mask_0, x = var_44485_cast_fp16)[name = tensor("op_44776_cast_fp16")]; + tensor var_44783_begin_0 = const()[name = tensor("op_44783_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44783_end_0 = const()[name = tensor("op_44783_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44783_end_mask_0 = const()[name = tensor("op_44783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44783_cast_fp16 = slice_by_index(begin = var_44783_begin_0, end = var_44783_end_0, end_mask = var_44783_end_mask_0, x = var_44485_cast_fp16)[name = tensor("op_44783_cast_fp16")]; + tensor var_44790_begin_0 = const()[name = tensor("op_44790_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44790_end_0 = const()[name = tensor("op_44790_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44790_end_mask_0 = const()[name = tensor("op_44790_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44790_cast_fp16 = slice_by_index(begin = var_44790_begin_0, end = var_44790_end_0, end_mask = var_44790_end_mask_0, x = var_44489_cast_fp16)[name = tensor("op_44790_cast_fp16")]; + tensor var_44797_begin_0 = const()[name = tensor("op_44797_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44797_end_0 = const()[name = tensor("op_44797_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44797_end_mask_0 = const()[name = tensor("op_44797_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44797_cast_fp16 = slice_by_index(begin = var_44797_begin_0, end = var_44797_end_0, end_mask = var_44797_end_mask_0, x = var_44489_cast_fp16)[name = tensor("op_44797_cast_fp16")]; + tensor var_44804_begin_0 = const()[name = tensor("op_44804_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44804_end_0 = const()[name = tensor("op_44804_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44804_end_mask_0 = const()[name = tensor("op_44804_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44804_cast_fp16 = slice_by_index(begin = var_44804_begin_0, end = var_44804_end_0, end_mask = var_44804_end_mask_0, x = var_44489_cast_fp16)[name = tensor("op_44804_cast_fp16")]; + tensor var_44811_begin_0 = const()[name = tensor("op_44811_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44811_end_0 = const()[name = tensor("op_44811_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44811_end_mask_0 = const()[name = tensor("op_44811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44811_cast_fp16 = slice_by_index(begin = var_44811_begin_0, end = var_44811_end_0, end_mask = var_44811_end_mask_0, x = var_44489_cast_fp16)[name = tensor("op_44811_cast_fp16")]; + tensor var_44818_begin_0 = const()[name = tensor("op_44818_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44818_end_0 = const()[name = tensor("op_44818_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44818_end_mask_0 = const()[name = tensor("op_44818_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44818_cast_fp16 = slice_by_index(begin = var_44818_begin_0, end = var_44818_end_0, end_mask = var_44818_end_mask_0, x = var_44493_cast_fp16)[name = tensor("op_44818_cast_fp16")]; + tensor var_44825_begin_0 = const()[name = tensor("op_44825_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44825_end_0 = const()[name = tensor("op_44825_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44825_end_mask_0 = const()[name = tensor("op_44825_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44825_cast_fp16 = slice_by_index(begin = var_44825_begin_0, end = var_44825_end_0, end_mask = var_44825_end_mask_0, x = var_44493_cast_fp16)[name = tensor("op_44825_cast_fp16")]; + tensor var_44832_begin_0 = const()[name = tensor("op_44832_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44832_end_0 = const()[name = tensor("op_44832_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44832_end_mask_0 = const()[name = tensor("op_44832_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44832_cast_fp16 = slice_by_index(begin = var_44832_begin_0, end = var_44832_end_0, end_mask = var_44832_end_mask_0, x = var_44493_cast_fp16)[name = tensor("op_44832_cast_fp16")]; + tensor var_44839_begin_0 = const()[name = tensor("op_44839_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44839_end_0 = const()[name = tensor("op_44839_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44839_end_mask_0 = const()[name = tensor("op_44839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44839_cast_fp16 = slice_by_index(begin = var_44839_begin_0, end = var_44839_end_0, end_mask = var_44839_end_mask_0, x = var_44493_cast_fp16)[name = tensor("op_44839_cast_fp16")]; + tensor var_44846_begin_0 = const()[name = tensor("op_44846_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44846_end_0 = const()[name = tensor("op_44846_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44846_end_mask_0 = const()[name = tensor("op_44846_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44846_cast_fp16 = slice_by_index(begin = var_44846_begin_0, end = var_44846_end_0, end_mask = var_44846_end_mask_0, x = var_44497_cast_fp16)[name = tensor("op_44846_cast_fp16")]; + tensor var_44853_begin_0 = const()[name = tensor("op_44853_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44853_end_0 = const()[name = tensor("op_44853_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44853_end_mask_0 = const()[name = tensor("op_44853_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44853_cast_fp16 = slice_by_index(begin = var_44853_begin_0, end = var_44853_end_0, end_mask = var_44853_end_mask_0, x = var_44497_cast_fp16)[name = tensor("op_44853_cast_fp16")]; + tensor var_44860_begin_0 = const()[name = tensor("op_44860_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44860_end_0 = const()[name = tensor("op_44860_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44860_end_mask_0 = const()[name = tensor("op_44860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44860_cast_fp16 = slice_by_index(begin = var_44860_begin_0, end = var_44860_end_0, end_mask = var_44860_end_mask_0, x = var_44497_cast_fp16)[name = tensor("op_44860_cast_fp16")]; + tensor var_44867_begin_0 = const()[name = tensor("op_44867_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44867_end_0 = const()[name = tensor("op_44867_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44867_end_mask_0 = const()[name = tensor("op_44867_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44867_cast_fp16 = slice_by_index(begin = var_44867_begin_0, end = var_44867_end_0, end_mask = var_44867_end_mask_0, x = var_44497_cast_fp16)[name = tensor("op_44867_cast_fp16")]; + tensor var_44874_begin_0 = const()[name = tensor("op_44874_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44874_end_0 = const()[name = tensor("op_44874_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44874_end_mask_0 = const()[name = tensor("op_44874_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44874_cast_fp16 = slice_by_index(begin = var_44874_begin_0, end = var_44874_end_0, end_mask = var_44874_end_mask_0, x = var_44501_cast_fp16)[name = tensor("op_44874_cast_fp16")]; + tensor var_44881_begin_0 = const()[name = tensor("op_44881_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44881_end_0 = const()[name = tensor("op_44881_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44881_end_mask_0 = const()[name = tensor("op_44881_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44881_cast_fp16 = slice_by_index(begin = var_44881_begin_0, end = var_44881_end_0, end_mask = var_44881_end_mask_0, x = var_44501_cast_fp16)[name = tensor("op_44881_cast_fp16")]; + tensor var_44888_begin_0 = const()[name = tensor("op_44888_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44888_end_0 = const()[name = tensor("op_44888_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44888_end_mask_0 = const()[name = tensor("op_44888_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44888_cast_fp16 = slice_by_index(begin = var_44888_begin_0, end = var_44888_end_0, end_mask = var_44888_end_mask_0, x = var_44501_cast_fp16)[name = tensor("op_44888_cast_fp16")]; + tensor var_44895_begin_0 = const()[name = tensor("op_44895_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44895_end_0 = const()[name = tensor("op_44895_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44895_end_mask_0 = const()[name = tensor("op_44895_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44895_cast_fp16 = slice_by_index(begin = var_44895_begin_0, end = var_44895_end_0, end_mask = var_44895_end_mask_0, x = var_44501_cast_fp16)[name = tensor("op_44895_cast_fp16")]; + tensor var_44902_begin_0 = const()[name = tensor("op_44902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44902_end_0 = const()[name = tensor("op_44902_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44902_end_mask_0 = const()[name = tensor("op_44902_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44902_cast_fp16 = slice_by_index(begin = var_44902_begin_0, end = var_44902_end_0, end_mask = var_44902_end_mask_0, x = var_44505_cast_fp16)[name = tensor("op_44902_cast_fp16")]; + tensor var_44909_begin_0 = const()[name = tensor("op_44909_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44909_end_0 = const()[name = tensor("op_44909_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44909_end_mask_0 = const()[name = tensor("op_44909_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44909_cast_fp16 = slice_by_index(begin = var_44909_begin_0, end = var_44909_end_0, end_mask = var_44909_end_mask_0, x = var_44505_cast_fp16)[name = tensor("op_44909_cast_fp16")]; + tensor var_44916_begin_0 = const()[name = tensor("op_44916_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44916_end_0 = const()[name = tensor("op_44916_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44916_end_mask_0 = const()[name = tensor("op_44916_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44916_cast_fp16 = slice_by_index(begin = var_44916_begin_0, end = var_44916_end_0, end_mask = var_44916_end_mask_0, x = var_44505_cast_fp16)[name = tensor("op_44916_cast_fp16")]; + tensor var_44923_begin_0 = const()[name = tensor("op_44923_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44923_end_0 = const()[name = tensor("op_44923_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44923_end_mask_0 = const()[name = tensor("op_44923_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44923_cast_fp16 = slice_by_index(begin = var_44923_begin_0, end = var_44923_end_0, end_mask = var_44923_end_mask_0, x = var_44505_cast_fp16)[name = tensor("op_44923_cast_fp16")]; + tensor var_44930_begin_0 = const()[name = tensor("op_44930_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44930_end_0 = const()[name = tensor("op_44930_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44930_end_mask_0 = const()[name = tensor("op_44930_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44930_cast_fp16 = slice_by_index(begin = var_44930_begin_0, end = var_44930_end_0, end_mask = var_44930_end_mask_0, x = var_44509_cast_fp16)[name = tensor("op_44930_cast_fp16")]; + tensor var_44937_begin_0 = const()[name = tensor("op_44937_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44937_end_0 = const()[name = tensor("op_44937_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44937_end_mask_0 = const()[name = tensor("op_44937_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44937_cast_fp16 = slice_by_index(begin = var_44937_begin_0, end = var_44937_end_0, end_mask = var_44937_end_mask_0, x = var_44509_cast_fp16)[name = tensor("op_44937_cast_fp16")]; + tensor var_44944_begin_0 = const()[name = tensor("op_44944_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44944_end_0 = const()[name = tensor("op_44944_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44944_end_mask_0 = const()[name = tensor("op_44944_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44944_cast_fp16 = slice_by_index(begin = var_44944_begin_0, end = var_44944_end_0, end_mask = var_44944_end_mask_0, x = var_44509_cast_fp16)[name = tensor("op_44944_cast_fp16")]; + tensor var_44951_begin_0 = const()[name = tensor("op_44951_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44951_end_0 = const()[name = tensor("op_44951_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44951_end_mask_0 = const()[name = tensor("op_44951_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44951_cast_fp16 = slice_by_index(begin = var_44951_begin_0, end = var_44951_end_0, end_mask = var_44951_end_mask_0, x = var_44509_cast_fp16)[name = tensor("op_44951_cast_fp16")]; + tensor var_44958_begin_0 = const()[name = tensor("op_44958_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44958_end_0 = const()[name = tensor("op_44958_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44958_end_mask_0 = const()[name = tensor("op_44958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44958_cast_fp16 = slice_by_index(begin = var_44958_begin_0, end = var_44958_end_0, end_mask = var_44958_end_mask_0, x = var_44513_cast_fp16)[name = tensor("op_44958_cast_fp16")]; + tensor var_44965_begin_0 = const()[name = tensor("op_44965_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44965_end_0 = const()[name = tensor("op_44965_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44965_end_mask_0 = const()[name = tensor("op_44965_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44965_cast_fp16 = slice_by_index(begin = var_44965_begin_0, end = var_44965_end_0, end_mask = var_44965_end_mask_0, x = var_44513_cast_fp16)[name = tensor("op_44965_cast_fp16")]; + tensor var_44972_begin_0 = const()[name = tensor("op_44972_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_44972_end_0 = const()[name = tensor("op_44972_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_44972_end_mask_0 = const()[name = tensor("op_44972_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44972_cast_fp16 = slice_by_index(begin = var_44972_begin_0, end = var_44972_end_0, end_mask = var_44972_end_mask_0, x = var_44513_cast_fp16)[name = tensor("op_44972_cast_fp16")]; + tensor var_44979_begin_0 = const()[name = tensor("op_44979_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_44979_end_0 = const()[name = tensor("op_44979_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_44979_end_mask_0 = const()[name = tensor("op_44979_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44979_cast_fp16 = slice_by_index(begin = var_44979_begin_0, end = var_44979_end_0, end_mask = var_44979_end_mask_0, x = var_44513_cast_fp16)[name = tensor("op_44979_cast_fp16")]; + tensor var_44986_begin_0 = const()[name = tensor("op_44986_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_44986_end_0 = const()[name = tensor("op_44986_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_44986_end_mask_0 = const()[name = tensor("op_44986_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44986_cast_fp16 = slice_by_index(begin = var_44986_begin_0, end = var_44986_end_0, end_mask = var_44986_end_mask_0, x = var_44517_cast_fp16)[name = tensor("op_44986_cast_fp16")]; + tensor var_44993_begin_0 = const()[name = tensor("op_44993_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_44993_end_0 = const()[name = tensor("op_44993_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_44993_end_mask_0 = const()[name = tensor("op_44993_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_44993_cast_fp16 = slice_by_index(begin = var_44993_begin_0, end = var_44993_end_0, end_mask = var_44993_end_mask_0, x = var_44517_cast_fp16)[name = tensor("op_44993_cast_fp16")]; + tensor var_45000_begin_0 = const()[name = tensor("op_45000_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45000_end_0 = const()[name = tensor("op_45000_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45000_end_mask_0 = const()[name = tensor("op_45000_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45000_cast_fp16 = slice_by_index(begin = var_45000_begin_0, end = var_45000_end_0, end_mask = var_45000_end_mask_0, x = var_44517_cast_fp16)[name = tensor("op_45000_cast_fp16")]; + tensor var_45007_begin_0 = const()[name = tensor("op_45007_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45007_end_0 = const()[name = tensor("op_45007_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45007_end_mask_0 = const()[name = tensor("op_45007_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45007_cast_fp16 = slice_by_index(begin = var_45007_begin_0, end = var_45007_end_0, end_mask = var_45007_end_mask_0, x = var_44517_cast_fp16)[name = tensor("op_45007_cast_fp16")]; + tensor var_45014_begin_0 = const()[name = tensor("op_45014_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45014_end_0 = const()[name = tensor("op_45014_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45014_end_mask_0 = const()[name = tensor("op_45014_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45014_cast_fp16 = slice_by_index(begin = var_45014_begin_0, end = var_45014_end_0, end_mask = var_45014_end_mask_0, x = var_44521_cast_fp16)[name = tensor("op_45014_cast_fp16")]; + tensor var_45021_begin_0 = const()[name = tensor("op_45021_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45021_end_0 = const()[name = tensor("op_45021_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45021_end_mask_0 = const()[name = tensor("op_45021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45021_cast_fp16 = slice_by_index(begin = var_45021_begin_0, end = var_45021_end_0, end_mask = var_45021_end_mask_0, x = var_44521_cast_fp16)[name = tensor("op_45021_cast_fp16")]; + tensor var_45028_begin_0 = const()[name = tensor("op_45028_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45028_end_0 = const()[name = tensor("op_45028_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45028_end_mask_0 = const()[name = tensor("op_45028_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45028_cast_fp16 = slice_by_index(begin = var_45028_begin_0, end = var_45028_end_0, end_mask = var_45028_end_mask_0, x = var_44521_cast_fp16)[name = tensor("op_45028_cast_fp16")]; + tensor var_45035_begin_0 = const()[name = tensor("op_45035_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45035_end_0 = const()[name = tensor("op_45035_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45035_end_mask_0 = const()[name = tensor("op_45035_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45035_cast_fp16 = slice_by_index(begin = var_45035_begin_0, end = var_45035_end_0, end_mask = var_45035_end_mask_0, x = var_44521_cast_fp16)[name = tensor("op_45035_cast_fp16")]; + tensor var_45042_begin_0 = const()[name = tensor("op_45042_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45042_end_0 = const()[name = tensor("op_45042_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45042_end_mask_0 = const()[name = tensor("op_45042_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45042_cast_fp16 = slice_by_index(begin = var_45042_begin_0, end = var_45042_end_0, end_mask = var_45042_end_mask_0, x = var_44525_cast_fp16)[name = tensor("op_45042_cast_fp16")]; + tensor var_45049_begin_0 = const()[name = tensor("op_45049_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45049_end_0 = const()[name = tensor("op_45049_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45049_end_mask_0 = const()[name = tensor("op_45049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45049_cast_fp16 = slice_by_index(begin = var_45049_begin_0, end = var_45049_end_0, end_mask = var_45049_end_mask_0, x = var_44525_cast_fp16)[name = tensor("op_45049_cast_fp16")]; + tensor var_45056_begin_0 = const()[name = tensor("op_45056_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45056_end_0 = const()[name = tensor("op_45056_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45056_end_mask_0 = const()[name = tensor("op_45056_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45056_cast_fp16 = slice_by_index(begin = var_45056_begin_0, end = var_45056_end_0, end_mask = var_45056_end_mask_0, x = var_44525_cast_fp16)[name = tensor("op_45056_cast_fp16")]; + tensor var_45063_begin_0 = const()[name = tensor("op_45063_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45063_end_0 = const()[name = tensor("op_45063_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45063_end_mask_0 = const()[name = tensor("op_45063_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45063_cast_fp16 = slice_by_index(begin = var_45063_begin_0, end = var_45063_end_0, end_mask = var_45063_end_mask_0, x = var_44525_cast_fp16)[name = tensor("op_45063_cast_fp16")]; + tensor var_45070_begin_0 = const()[name = tensor("op_45070_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45070_end_0 = const()[name = tensor("op_45070_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_45070_end_mask_0 = const()[name = tensor("op_45070_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45070_cast_fp16 = slice_by_index(begin = var_45070_begin_0, end = var_45070_end_0, end_mask = var_45070_end_mask_0, x = var_44529_cast_fp16)[name = tensor("op_45070_cast_fp16")]; + tensor var_45077_begin_0 = const()[name = tensor("op_45077_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_45077_end_0 = const()[name = tensor("op_45077_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_45077_end_mask_0 = const()[name = tensor("op_45077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45077_cast_fp16 = slice_by_index(begin = var_45077_begin_0, end = var_45077_end_0, end_mask = var_45077_end_mask_0, x = var_44529_cast_fp16)[name = tensor("op_45077_cast_fp16")]; + tensor var_45084_begin_0 = const()[name = tensor("op_45084_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_45084_end_0 = const()[name = tensor("op_45084_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_45084_end_mask_0 = const()[name = tensor("op_45084_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45084_cast_fp16 = slice_by_index(begin = var_45084_begin_0, end = var_45084_end_0, end_mask = var_45084_end_mask_0, x = var_44529_cast_fp16)[name = tensor("op_45084_cast_fp16")]; + tensor var_45091_begin_0 = const()[name = tensor("op_45091_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_45091_end_0 = const()[name = tensor("op_45091_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45091_end_mask_0 = const()[name = tensor("op_45091_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45091_cast_fp16 = slice_by_index(begin = var_45091_begin_0, end = var_45091_end_0, end_mask = var_45091_end_mask_0, x = var_44529_cast_fp16)[name = tensor("op_45091_cast_fp16")]; + tensor k_55_perm_0 = const()[name = tensor("k_55_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_45096_begin_0 = const()[name = tensor("op_45096_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45096_end_0 = const()[name = tensor("op_45096_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_45096_end_mask_0 = const()[name = tensor("op_45096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_4 = transpose(perm = k_55_perm_0, x = key_55_cast_fp16)[name = tensor("transpose_4")]; + tensor var_45096_cast_fp16 = slice_by_index(begin = var_45096_begin_0, end = var_45096_end_0, end_mask = var_45096_end_mask_0, x = transpose_4)[name = tensor("op_45096_cast_fp16")]; + tensor var_45100_begin_0 = const()[name = tensor("op_45100_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_45100_end_0 = const()[name = tensor("op_45100_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_45100_end_mask_0 = const()[name = tensor("op_45100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45100_cast_fp16 = slice_by_index(begin = var_45100_begin_0, end = var_45100_end_0, end_mask = var_45100_end_mask_0, x = transpose_4)[name = tensor("op_45100_cast_fp16")]; + tensor var_45104_begin_0 = const()[name = tensor("op_45104_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_45104_end_0 = const()[name = tensor("op_45104_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_45104_end_mask_0 = const()[name = tensor("op_45104_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45104_cast_fp16 = slice_by_index(begin = var_45104_begin_0, end = var_45104_end_0, end_mask = var_45104_end_mask_0, x = transpose_4)[name = tensor("op_45104_cast_fp16")]; + tensor var_45108_begin_0 = const()[name = tensor("op_45108_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_45108_end_0 = const()[name = tensor("op_45108_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_45108_end_mask_0 = const()[name = tensor("op_45108_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45108_cast_fp16 = slice_by_index(begin = var_45108_begin_0, end = var_45108_end_0, end_mask = var_45108_end_mask_0, x = transpose_4)[name = tensor("op_45108_cast_fp16")]; + tensor var_45112_begin_0 = const()[name = tensor("op_45112_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_45112_end_0 = const()[name = tensor("op_45112_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_45112_end_mask_0 = const()[name = tensor("op_45112_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45112_cast_fp16 = slice_by_index(begin = var_45112_begin_0, end = var_45112_end_0, end_mask = var_45112_end_mask_0, x = transpose_4)[name = tensor("op_45112_cast_fp16")]; + tensor var_45116_begin_0 = const()[name = tensor("op_45116_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_45116_end_0 = const()[name = tensor("op_45116_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_45116_end_mask_0 = const()[name = tensor("op_45116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45116_cast_fp16 = slice_by_index(begin = var_45116_begin_0, end = var_45116_end_0, end_mask = var_45116_end_mask_0, x = transpose_4)[name = tensor("op_45116_cast_fp16")]; + tensor var_45120_begin_0 = const()[name = tensor("op_45120_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_45120_end_0 = const()[name = tensor("op_45120_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_45120_end_mask_0 = const()[name = tensor("op_45120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45120_cast_fp16 = slice_by_index(begin = var_45120_begin_0, end = var_45120_end_0, end_mask = var_45120_end_mask_0, x = transpose_4)[name = tensor("op_45120_cast_fp16")]; + tensor var_45124_begin_0 = const()[name = tensor("op_45124_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_45124_end_0 = const()[name = tensor("op_45124_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_45124_end_mask_0 = const()[name = tensor("op_45124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45124_cast_fp16 = slice_by_index(begin = var_45124_begin_0, end = var_45124_end_0, end_mask = var_45124_end_mask_0, x = transpose_4)[name = tensor("op_45124_cast_fp16")]; + tensor var_45128_begin_0 = const()[name = tensor("op_45128_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_45128_end_0 = const()[name = tensor("op_45128_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_45128_end_mask_0 = const()[name = tensor("op_45128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45128_cast_fp16 = slice_by_index(begin = var_45128_begin_0, end = var_45128_end_0, end_mask = var_45128_end_mask_0, x = transpose_4)[name = tensor("op_45128_cast_fp16")]; + tensor var_45132_begin_0 = const()[name = tensor("op_45132_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_45132_end_0 = const()[name = tensor("op_45132_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_45132_end_mask_0 = const()[name = tensor("op_45132_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45132_cast_fp16 = slice_by_index(begin = var_45132_begin_0, end = var_45132_end_0, end_mask = var_45132_end_mask_0, x = transpose_4)[name = tensor("op_45132_cast_fp16")]; + tensor var_45136_begin_0 = const()[name = tensor("op_45136_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_45136_end_0 = const()[name = tensor("op_45136_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_45136_end_mask_0 = const()[name = tensor("op_45136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45136_cast_fp16 = slice_by_index(begin = var_45136_begin_0, end = var_45136_end_0, end_mask = var_45136_end_mask_0, x = transpose_4)[name = tensor("op_45136_cast_fp16")]; + tensor var_45140_begin_0 = const()[name = tensor("op_45140_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_45140_end_0 = const()[name = tensor("op_45140_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_45140_end_mask_0 = const()[name = tensor("op_45140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45140_cast_fp16 = slice_by_index(begin = var_45140_begin_0, end = var_45140_end_0, end_mask = var_45140_end_mask_0, x = transpose_4)[name = tensor("op_45140_cast_fp16")]; + tensor var_45144_begin_0 = const()[name = tensor("op_45144_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_45144_end_0 = const()[name = tensor("op_45144_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_45144_end_mask_0 = const()[name = tensor("op_45144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45144_cast_fp16 = slice_by_index(begin = var_45144_begin_0, end = var_45144_end_0, end_mask = var_45144_end_mask_0, x = transpose_4)[name = tensor("op_45144_cast_fp16")]; + tensor var_45148_begin_0 = const()[name = tensor("op_45148_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_45148_end_0 = const()[name = tensor("op_45148_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_45148_end_mask_0 = const()[name = tensor("op_45148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45148_cast_fp16 = slice_by_index(begin = var_45148_begin_0, end = var_45148_end_0, end_mask = var_45148_end_mask_0, x = transpose_4)[name = tensor("op_45148_cast_fp16")]; + tensor var_45152_begin_0 = const()[name = tensor("op_45152_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_45152_end_0 = const()[name = tensor("op_45152_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_45152_end_mask_0 = const()[name = tensor("op_45152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45152_cast_fp16 = slice_by_index(begin = var_45152_begin_0, end = var_45152_end_0, end_mask = var_45152_end_mask_0, x = transpose_4)[name = tensor("op_45152_cast_fp16")]; + tensor var_45156_begin_0 = const()[name = tensor("op_45156_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_45156_end_0 = const()[name = tensor("op_45156_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_45156_end_mask_0 = const()[name = tensor("op_45156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45156_cast_fp16 = slice_by_index(begin = var_45156_begin_0, end = var_45156_end_0, end_mask = var_45156_end_mask_0, x = transpose_4)[name = tensor("op_45156_cast_fp16")]; + tensor var_45160_begin_0 = const()[name = tensor("op_45160_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_45160_end_0 = const()[name = tensor("op_45160_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_45160_end_mask_0 = const()[name = tensor("op_45160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45160_cast_fp16 = slice_by_index(begin = var_45160_begin_0, end = var_45160_end_0, end_mask = var_45160_end_mask_0, x = transpose_4)[name = tensor("op_45160_cast_fp16")]; + tensor var_45164_begin_0 = const()[name = tensor("op_45164_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_45164_end_0 = const()[name = tensor("op_45164_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_45164_end_mask_0 = const()[name = tensor("op_45164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45164_cast_fp16 = slice_by_index(begin = var_45164_begin_0, end = var_45164_end_0, end_mask = var_45164_end_mask_0, x = transpose_4)[name = tensor("op_45164_cast_fp16")]; + tensor var_45168_begin_0 = const()[name = tensor("op_45168_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_45168_end_0 = const()[name = tensor("op_45168_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_45168_end_mask_0 = const()[name = tensor("op_45168_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45168_cast_fp16 = slice_by_index(begin = var_45168_begin_0, end = var_45168_end_0, end_mask = var_45168_end_mask_0, x = transpose_4)[name = tensor("op_45168_cast_fp16")]; + tensor var_45172_begin_0 = const()[name = tensor("op_45172_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_45172_end_0 = const()[name = tensor("op_45172_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_45172_end_mask_0 = const()[name = tensor("op_45172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_45172_cast_fp16 = slice_by_index(begin = var_45172_begin_0, end = var_45172_end_0, end_mask = var_45172_end_mask_0, x = transpose_4)[name = tensor("op_45172_cast_fp16")]; + tensor var_45174_begin_0 = const()[name = tensor("op_45174_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_45174_end_0 = const()[name = tensor("op_45174_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_45174_end_mask_0 = const()[name = tensor("op_45174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45174_cast_fp16 = slice_by_index(begin = var_45174_begin_0, end = var_45174_end_0, end_mask = var_45174_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45174_cast_fp16")]; + tensor var_45178_begin_0 = const()[name = tensor("op_45178_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_45178_end_0 = const()[name = tensor("op_45178_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_45178_end_mask_0 = const()[name = tensor("op_45178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45178_cast_fp16 = slice_by_index(begin = var_45178_begin_0, end = var_45178_end_0, end_mask = var_45178_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45178_cast_fp16")]; + tensor var_45182_begin_0 = const()[name = tensor("op_45182_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_45182_end_0 = const()[name = tensor("op_45182_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_45182_end_mask_0 = const()[name = tensor("op_45182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45182_cast_fp16 = slice_by_index(begin = var_45182_begin_0, end = var_45182_end_0, end_mask = var_45182_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45182_cast_fp16")]; + tensor var_45186_begin_0 = const()[name = tensor("op_45186_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_45186_end_0 = const()[name = tensor("op_45186_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_45186_end_mask_0 = const()[name = tensor("op_45186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45186_cast_fp16 = slice_by_index(begin = var_45186_begin_0, end = var_45186_end_0, end_mask = var_45186_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45186_cast_fp16")]; + tensor var_45190_begin_0 = const()[name = tensor("op_45190_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_45190_end_0 = const()[name = tensor("op_45190_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_45190_end_mask_0 = const()[name = tensor("op_45190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45190_cast_fp16 = slice_by_index(begin = var_45190_begin_0, end = var_45190_end_0, end_mask = var_45190_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45190_cast_fp16")]; + tensor var_45194_begin_0 = const()[name = tensor("op_45194_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_45194_end_0 = const()[name = tensor("op_45194_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_45194_end_mask_0 = const()[name = tensor("op_45194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45194_cast_fp16 = slice_by_index(begin = var_45194_begin_0, end = var_45194_end_0, end_mask = var_45194_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45194_cast_fp16")]; + tensor var_45198_begin_0 = const()[name = tensor("op_45198_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_45198_end_0 = const()[name = tensor("op_45198_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_45198_end_mask_0 = const()[name = tensor("op_45198_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45198_cast_fp16 = slice_by_index(begin = var_45198_begin_0, end = var_45198_end_0, end_mask = var_45198_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45198_cast_fp16")]; + tensor var_45202_begin_0 = const()[name = tensor("op_45202_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_45202_end_0 = const()[name = tensor("op_45202_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_45202_end_mask_0 = const()[name = tensor("op_45202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45202_cast_fp16 = slice_by_index(begin = var_45202_begin_0, end = var_45202_end_0, end_mask = var_45202_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45202_cast_fp16")]; + tensor var_45206_begin_0 = const()[name = tensor("op_45206_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_45206_end_0 = const()[name = tensor("op_45206_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_45206_end_mask_0 = const()[name = tensor("op_45206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45206_cast_fp16 = slice_by_index(begin = var_45206_begin_0, end = var_45206_end_0, end_mask = var_45206_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45206_cast_fp16")]; + tensor var_45210_begin_0 = const()[name = tensor("op_45210_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_45210_end_0 = const()[name = tensor("op_45210_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_45210_end_mask_0 = const()[name = tensor("op_45210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45210_cast_fp16 = slice_by_index(begin = var_45210_begin_0, end = var_45210_end_0, end_mask = var_45210_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45210_cast_fp16")]; + tensor var_45214_begin_0 = const()[name = tensor("op_45214_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_45214_end_0 = const()[name = tensor("op_45214_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_45214_end_mask_0 = const()[name = tensor("op_45214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45214_cast_fp16 = slice_by_index(begin = var_45214_begin_0, end = var_45214_end_0, end_mask = var_45214_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45214_cast_fp16")]; + tensor var_45218_begin_0 = const()[name = tensor("op_45218_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_45218_end_0 = const()[name = tensor("op_45218_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_45218_end_mask_0 = const()[name = tensor("op_45218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45218_cast_fp16 = slice_by_index(begin = var_45218_begin_0, end = var_45218_end_0, end_mask = var_45218_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45218_cast_fp16")]; + tensor var_45222_begin_0 = const()[name = tensor("op_45222_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_45222_end_0 = const()[name = tensor("op_45222_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_45222_end_mask_0 = const()[name = tensor("op_45222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45222_cast_fp16 = slice_by_index(begin = var_45222_begin_0, end = var_45222_end_0, end_mask = var_45222_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45222_cast_fp16")]; + tensor var_45226_begin_0 = const()[name = tensor("op_45226_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_45226_end_0 = const()[name = tensor("op_45226_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_45226_end_mask_0 = const()[name = tensor("op_45226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45226_cast_fp16 = slice_by_index(begin = var_45226_begin_0, end = var_45226_end_0, end_mask = var_45226_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45226_cast_fp16")]; + tensor var_45230_begin_0 = const()[name = tensor("op_45230_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_45230_end_0 = const()[name = tensor("op_45230_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_45230_end_mask_0 = const()[name = tensor("op_45230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45230_cast_fp16 = slice_by_index(begin = var_45230_begin_0, end = var_45230_end_0, end_mask = var_45230_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45230_cast_fp16")]; + tensor var_45234_begin_0 = const()[name = tensor("op_45234_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_45234_end_0 = const()[name = tensor("op_45234_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_45234_end_mask_0 = const()[name = tensor("op_45234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45234_cast_fp16 = slice_by_index(begin = var_45234_begin_0, end = var_45234_end_0, end_mask = var_45234_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45234_cast_fp16")]; + tensor var_45238_begin_0 = const()[name = tensor("op_45238_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_45238_end_0 = const()[name = tensor("op_45238_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_45238_end_mask_0 = const()[name = tensor("op_45238_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45238_cast_fp16 = slice_by_index(begin = var_45238_begin_0, end = var_45238_end_0, end_mask = var_45238_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45238_cast_fp16")]; + tensor var_45242_begin_0 = const()[name = tensor("op_45242_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_45242_end_0 = const()[name = tensor("op_45242_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_45242_end_mask_0 = const()[name = tensor("op_45242_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45242_cast_fp16 = slice_by_index(begin = var_45242_begin_0, end = var_45242_end_0, end_mask = var_45242_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45242_cast_fp16")]; + tensor var_45246_begin_0 = const()[name = tensor("op_45246_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_45246_end_0 = const()[name = tensor("op_45246_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_45246_end_mask_0 = const()[name = tensor("op_45246_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45246_cast_fp16 = slice_by_index(begin = var_45246_begin_0, end = var_45246_end_0, end_mask = var_45246_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45246_cast_fp16")]; + tensor var_45250_begin_0 = const()[name = tensor("op_45250_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_45250_end_0 = const()[name = tensor("op_45250_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_45250_end_mask_0 = const()[name = tensor("op_45250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_45250_cast_fp16 = slice_by_index(begin = var_45250_begin_0, end = var_45250_end_0, end_mask = var_45250_end_mask_0, x = value_55_cast_fp16)[name = tensor("op_45250_cast_fp16")]; + tensor var_45254_equation_0 = const()[name = tensor("op_45254_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45254_cast_fp16 = einsum(equation = var_45254_equation_0, values = (var_45096_cast_fp16, var_44538_cast_fp16))[name = tensor("op_45254_cast_fp16")]; + tensor var_45255_to_fp16 = const()[name = tensor("op_45255_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4321_cast_fp16 = mul(x = var_45254_cast_fp16, y = var_45255_to_fp16)[name = tensor("aw_chunk_4321_cast_fp16")]; + tensor var_45258_equation_0 = const()[name = tensor("op_45258_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45258_cast_fp16 = einsum(equation = var_45258_equation_0, values = (var_45096_cast_fp16, var_44545_cast_fp16))[name = tensor("op_45258_cast_fp16")]; + tensor var_45259_to_fp16 = const()[name = tensor("op_45259_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4323_cast_fp16 = mul(x = var_45258_cast_fp16, y = var_45259_to_fp16)[name = tensor("aw_chunk_4323_cast_fp16")]; + tensor var_45262_equation_0 = const()[name = tensor("op_45262_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45262_cast_fp16 = einsum(equation = var_45262_equation_0, values = (var_45096_cast_fp16, var_44552_cast_fp16))[name = tensor("op_45262_cast_fp16")]; + tensor var_45263_to_fp16 = const()[name = tensor("op_45263_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4325_cast_fp16 = mul(x = var_45262_cast_fp16, y = var_45263_to_fp16)[name = tensor("aw_chunk_4325_cast_fp16")]; + tensor var_45266_equation_0 = const()[name = tensor("op_45266_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45266_cast_fp16 = einsum(equation = var_45266_equation_0, values = (var_45096_cast_fp16, var_44559_cast_fp16))[name = tensor("op_45266_cast_fp16")]; + tensor var_45267_to_fp16 = const()[name = tensor("op_45267_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4327_cast_fp16 = mul(x = var_45266_cast_fp16, y = var_45267_to_fp16)[name = tensor("aw_chunk_4327_cast_fp16")]; + tensor var_45270_equation_0 = const()[name = tensor("op_45270_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45270_cast_fp16 = einsum(equation = var_45270_equation_0, values = (var_45100_cast_fp16, var_44566_cast_fp16))[name = tensor("op_45270_cast_fp16")]; + tensor var_45271_to_fp16 = const()[name = tensor("op_45271_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4329_cast_fp16 = mul(x = var_45270_cast_fp16, y = var_45271_to_fp16)[name = tensor("aw_chunk_4329_cast_fp16")]; + tensor var_45274_equation_0 = const()[name = tensor("op_45274_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45274_cast_fp16 = einsum(equation = var_45274_equation_0, values = (var_45100_cast_fp16, var_44573_cast_fp16))[name = tensor("op_45274_cast_fp16")]; + tensor var_45275_to_fp16 = const()[name = tensor("op_45275_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4331_cast_fp16 = mul(x = var_45274_cast_fp16, y = var_45275_to_fp16)[name = tensor("aw_chunk_4331_cast_fp16")]; + tensor var_45278_equation_0 = const()[name = tensor("op_45278_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45278_cast_fp16 = einsum(equation = var_45278_equation_0, values = (var_45100_cast_fp16, var_44580_cast_fp16))[name = tensor("op_45278_cast_fp16")]; + tensor var_45279_to_fp16 = const()[name = tensor("op_45279_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4333_cast_fp16 = mul(x = var_45278_cast_fp16, y = var_45279_to_fp16)[name = tensor("aw_chunk_4333_cast_fp16")]; + tensor var_45282_equation_0 = const()[name = tensor("op_45282_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45282_cast_fp16 = einsum(equation = var_45282_equation_0, values = (var_45100_cast_fp16, var_44587_cast_fp16))[name = tensor("op_45282_cast_fp16")]; + tensor var_45283_to_fp16 = const()[name = tensor("op_45283_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4335_cast_fp16 = mul(x = var_45282_cast_fp16, y = var_45283_to_fp16)[name = tensor("aw_chunk_4335_cast_fp16")]; + tensor var_45286_equation_0 = const()[name = tensor("op_45286_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45286_cast_fp16 = einsum(equation = var_45286_equation_0, values = (var_45104_cast_fp16, var_44594_cast_fp16))[name = tensor("op_45286_cast_fp16")]; + tensor var_45287_to_fp16 = const()[name = tensor("op_45287_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4337_cast_fp16 = mul(x = var_45286_cast_fp16, y = var_45287_to_fp16)[name = tensor("aw_chunk_4337_cast_fp16")]; + tensor var_45290_equation_0 = const()[name = tensor("op_45290_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45290_cast_fp16 = einsum(equation = var_45290_equation_0, values = (var_45104_cast_fp16, var_44601_cast_fp16))[name = tensor("op_45290_cast_fp16")]; + tensor var_45291_to_fp16 = const()[name = tensor("op_45291_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4339_cast_fp16 = mul(x = var_45290_cast_fp16, y = var_45291_to_fp16)[name = tensor("aw_chunk_4339_cast_fp16")]; + tensor var_45294_equation_0 = const()[name = tensor("op_45294_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45294_cast_fp16 = einsum(equation = var_45294_equation_0, values = (var_45104_cast_fp16, var_44608_cast_fp16))[name = tensor("op_45294_cast_fp16")]; + tensor var_45295_to_fp16 = const()[name = tensor("op_45295_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4341_cast_fp16 = mul(x = var_45294_cast_fp16, y = var_45295_to_fp16)[name = tensor("aw_chunk_4341_cast_fp16")]; + tensor var_45298_equation_0 = const()[name = tensor("op_45298_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45298_cast_fp16 = einsum(equation = var_45298_equation_0, values = (var_45104_cast_fp16, var_44615_cast_fp16))[name = tensor("op_45298_cast_fp16")]; + tensor var_45299_to_fp16 = const()[name = tensor("op_45299_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4343_cast_fp16 = mul(x = var_45298_cast_fp16, y = var_45299_to_fp16)[name = tensor("aw_chunk_4343_cast_fp16")]; + tensor var_45302_equation_0 = const()[name = tensor("op_45302_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45302_cast_fp16 = einsum(equation = var_45302_equation_0, values = (var_45108_cast_fp16, var_44622_cast_fp16))[name = tensor("op_45302_cast_fp16")]; + tensor var_45303_to_fp16 = const()[name = tensor("op_45303_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4345_cast_fp16 = mul(x = var_45302_cast_fp16, y = var_45303_to_fp16)[name = tensor("aw_chunk_4345_cast_fp16")]; + tensor var_45306_equation_0 = const()[name = tensor("op_45306_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45306_cast_fp16 = einsum(equation = var_45306_equation_0, values = (var_45108_cast_fp16, var_44629_cast_fp16))[name = tensor("op_45306_cast_fp16")]; + tensor var_45307_to_fp16 = const()[name = tensor("op_45307_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4347_cast_fp16 = mul(x = var_45306_cast_fp16, y = var_45307_to_fp16)[name = tensor("aw_chunk_4347_cast_fp16")]; + tensor var_45310_equation_0 = const()[name = tensor("op_45310_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45310_cast_fp16 = einsum(equation = var_45310_equation_0, values = (var_45108_cast_fp16, var_44636_cast_fp16))[name = tensor("op_45310_cast_fp16")]; + tensor var_45311_to_fp16 = const()[name = tensor("op_45311_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4349_cast_fp16 = mul(x = var_45310_cast_fp16, y = var_45311_to_fp16)[name = tensor("aw_chunk_4349_cast_fp16")]; + tensor var_45314_equation_0 = const()[name = tensor("op_45314_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45314_cast_fp16 = einsum(equation = var_45314_equation_0, values = (var_45108_cast_fp16, var_44643_cast_fp16))[name = tensor("op_45314_cast_fp16")]; + tensor var_45315_to_fp16 = const()[name = tensor("op_45315_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4351_cast_fp16 = mul(x = var_45314_cast_fp16, y = var_45315_to_fp16)[name = tensor("aw_chunk_4351_cast_fp16")]; + tensor var_45318_equation_0 = const()[name = tensor("op_45318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45318_cast_fp16 = einsum(equation = var_45318_equation_0, values = (var_45112_cast_fp16, var_44650_cast_fp16))[name = tensor("op_45318_cast_fp16")]; + tensor var_45319_to_fp16 = const()[name = tensor("op_45319_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4353_cast_fp16 = mul(x = var_45318_cast_fp16, y = var_45319_to_fp16)[name = tensor("aw_chunk_4353_cast_fp16")]; + tensor var_45322_equation_0 = const()[name = tensor("op_45322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45322_cast_fp16 = einsum(equation = var_45322_equation_0, values = (var_45112_cast_fp16, var_44657_cast_fp16))[name = tensor("op_45322_cast_fp16")]; + tensor var_45323_to_fp16 = const()[name = tensor("op_45323_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4355_cast_fp16 = mul(x = var_45322_cast_fp16, y = var_45323_to_fp16)[name = tensor("aw_chunk_4355_cast_fp16")]; + tensor var_45326_equation_0 = const()[name = tensor("op_45326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45326_cast_fp16 = einsum(equation = var_45326_equation_0, values = (var_45112_cast_fp16, var_44664_cast_fp16))[name = tensor("op_45326_cast_fp16")]; + tensor var_45327_to_fp16 = const()[name = tensor("op_45327_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4357_cast_fp16 = mul(x = var_45326_cast_fp16, y = var_45327_to_fp16)[name = tensor("aw_chunk_4357_cast_fp16")]; + tensor var_45330_equation_0 = const()[name = tensor("op_45330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45330_cast_fp16 = einsum(equation = var_45330_equation_0, values = (var_45112_cast_fp16, var_44671_cast_fp16))[name = tensor("op_45330_cast_fp16")]; + tensor var_45331_to_fp16 = const()[name = tensor("op_45331_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4359_cast_fp16 = mul(x = var_45330_cast_fp16, y = var_45331_to_fp16)[name = tensor("aw_chunk_4359_cast_fp16")]; + tensor var_45334_equation_0 = const()[name = tensor("op_45334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45334_cast_fp16 = einsum(equation = var_45334_equation_0, values = (var_45116_cast_fp16, var_44678_cast_fp16))[name = tensor("op_45334_cast_fp16")]; + tensor var_45335_to_fp16 = const()[name = tensor("op_45335_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4361_cast_fp16 = mul(x = var_45334_cast_fp16, y = var_45335_to_fp16)[name = tensor("aw_chunk_4361_cast_fp16")]; + tensor var_45338_equation_0 = const()[name = tensor("op_45338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45338_cast_fp16 = einsum(equation = var_45338_equation_0, values = (var_45116_cast_fp16, var_44685_cast_fp16))[name = tensor("op_45338_cast_fp16")]; + tensor var_45339_to_fp16 = const()[name = tensor("op_45339_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4363_cast_fp16 = mul(x = var_45338_cast_fp16, y = var_45339_to_fp16)[name = tensor("aw_chunk_4363_cast_fp16")]; + tensor var_45342_equation_0 = const()[name = tensor("op_45342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45342_cast_fp16 = einsum(equation = var_45342_equation_0, values = (var_45116_cast_fp16, var_44692_cast_fp16))[name = tensor("op_45342_cast_fp16")]; + tensor var_45343_to_fp16 = const()[name = tensor("op_45343_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4365_cast_fp16 = mul(x = var_45342_cast_fp16, y = var_45343_to_fp16)[name = tensor("aw_chunk_4365_cast_fp16")]; + tensor var_45346_equation_0 = const()[name = tensor("op_45346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45346_cast_fp16 = einsum(equation = var_45346_equation_0, values = (var_45116_cast_fp16, var_44699_cast_fp16))[name = tensor("op_45346_cast_fp16")]; + tensor var_45347_to_fp16 = const()[name = tensor("op_45347_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4367_cast_fp16 = mul(x = var_45346_cast_fp16, y = var_45347_to_fp16)[name = tensor("aw_chunk_4367_cast_fp16")]; + tensor var_45350_equation_0 = const()[name = tensor("op_45350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45350_cast_fp16 = einsum(equation = var_45350_equation_0, values = (var_45120_cast_fp16, var_44706_cast_fp16))[name = tensor("op_45350_cast_fp16")]; + tensor var_45351_to_fp16 = const()[name = tensor("op_45351_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4369_cast_fp16 = mul(x = var_45350_cast_fp16, y = var_45351_to_fp16)[name = tensor("aw_chunk_4369_cast_fp16")]; + tensor var_45354_equation_0 = const()[name = tensor("op_45354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45354_cast_fp16 = einsum(equation = var_45354_equation_0, values = (var_45120_cast_fp16, var_44713_cast_fp16))[name = tensor("op_45354_cast_fp16")]; + tensor var_45355_to_fp16 = const()[name = tensor("op_45355_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4371_cast_fp16 = mul(x = var_45354_cast_fp16, y = var_45355_to_fp16)[name = tensor("aw_chunk_4371_cast_fp16")]; + tensor var_45358_equation_0 = const()[name = tensor("op_45358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45358_cast_fp16 = einsum(equation = var_45358_equation_0, values = (var_45120_cast_fp16, var_44720_cast_fp16))[name = tensor("op_45358_cast_fp16")]; + tensor var_45359_to_fp16 = const()[name = tensor("op_45359_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4373_cast_fp16 = mul(x = var_45358_cast_fp16, y = var_45359_to_fp16)[name = tensor("aw_chunk_4373_cast_fp16")]; + tensor var_45362_equation_0 = const()[name = tensor("op_45362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45362_cast_fp16 = einsum(equation = var_45362_equation_0, values = (var_45120_cast_fp16, var_44727_cast_fp16))[name = tensor("op_45362_cast_fp16")]; + tensor var_45363_to_fp16 = const()[name = tensor("op_45363_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4375_cast_fp16 = mul(x = var_45362_cast_fp16, y = var_45363_to_fp16)[name = tensor("aw_chunk_4375_cast_fp16")]; + tensor var_45366_equation_0 = const()[name = tensor("op_45366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45366_cast_fp16 = einsum(equation = var_45366_equation_0, values = (var_45124_cast_fp16, var_44734_cast_fp16))[name = tensor("op_45366_cast_fp16")]; + tensor var_45367_to_fp16 = const()[name = tensor("op_45367_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4377_cast_fp16 = mul(x = var_45366_cast_fp16, y = var_45367_to_fp16)[name = tensor("aw_chunk_4377_cast_fp16")]; + tensor var_45370_equation_0 = const()[name = tensor("op_45370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45370_cast_fp16 = einsum(equation = var_45370_equation_0, values = (var_45124_cast_fp16, var_44741_cast_fp16))[name = tensor("op_45370_cast_fp16")]; + tensor var_45371_to_fp16 = const()[name = tensor("op_45371_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4379_cast_fp16 = mul(x = var_45370_cast_fp16, y = var_45371_to_fp16)[name = tensor("aw_chunk_4379_cast_fp16")]; + tensor var_45374_equation_0 = const()[name = tensor("op_45374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45374_cast_fp16 = einsum(equation = var_45374_equation_0, values = (var_45124_cast_fp16, var_44748_cast_fp16))[name = tensor("op_45374_cast_fp16")]; + tensor var_45375_to_fp16 = const()[name = tensor("op_45375_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4381_cast_fp16 = mul(x = var_45374_cast_fp16, y = var_45375_to_fp16)[name = tensor("aw_chunk_4381_cast_fp16")]; + tensor var_45378_equation_0 = const()[name = tensor("op_45378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45378_cast_fp16 = einsum(equation = var_45378_equation_0, values = (var_45124_cast_fp16, var_44755_cast_fp16))[name = tensor("op_45378_cast_fp16")]; + tensor var_45379_to_fp16 = const()[name = tensor("op_45379_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4383_cast_fp16 = mul(x = var_45378_cast_fp16, y = var_45379_to_fp16)[name = tensor("aw_chunk_4383_cast_fp16")]; + tensor var_45382_equation_0 = const()[name = tensor("op_45382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45382_cast_fp16 = einsum(equation = var_45382_equation_0, values = (var_45128_cast_fp16, var_44762_cast_fp16))[name = tensor("op_45382_cast_fp16")]; + tensor var_45383_to_fp16 = const()[name = tensor("op_45383_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4385_cast_fp16 = mul(x = var_45382_cast_fp16, y = var_45383_to_fp16)[name = tensor("aw_chunk_4385_cast_fp16")]; + tensor var_45386_equation_0 = const()[name = tensor("op_45386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45386_cast_fp16 = einsum(equation = var_45386_equation_0, values = (var_45128_cast_fp16, var_44769_cast_fp16))[name = tensor("op_45386_cast_fp16")]; + tensor var_45387_to_fp16 = const()[name = tensor("op_45387_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4387_cast_fp16 = mul(x = var_45386_cast_fp16, y = var_45387_to_fp16)[name = tensor("aw_chunk_4387_cast_fp16")]; + tensor var_45390_equation_0 = const()[name = tensor("op_45390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45390_cast_fp16 = einsum(equation = var_45390_equation_0, values = (var_45128_cast_fp16, var_44776_cast_fp16))[name = tensor("op_45390_cast_fp16")]; + tensor var_45391_to_fp16 = const()[name = tensor("op_45391_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4389_cast_fp16 = mul(x = var_45390_cast_fp16, y = var_45391_to_fp16)[name = tensor("aw_chunk_4389_cast_fp16")]; + tensor var_45394_equation_0 = const()[name = tensor("op_45394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45394_cast_fp16 = einsum(equation = var_45394_equation_0, values = (var_45128_cast_fp16, var_44783_cast_fp16))[name = tensor("op_45394_cast_fp16")]; + tensor var_45395_to_fp16 = const()[name = tensor("op_45395_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4391_cast_fp16 = mul(x = var_45394_cast_fp16, y = var_45395_to_fp16)[name = tensor("aw_chunk_4391_cast_fp16")]; + tensor var_45398_equation_0 = const()[name = tensor("op_45398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45398_cast_fp16 = einsum(equation = var_45398_equation_0, values = (var_45132_cast_fp16, var_44790_cast_fp16))[name = tensor("op_45398_cast_fp16")]; + tensor var_45399_to_fp16 = const()[name = tensor("op_45399_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4393_cast_fp16 = mul(x = var_45398_cast_fp16, y = var_45399_to_fp16)[name = tensor("aw_chunk_4393_cast_fp16")]; + tensor var_45402_equation_0 = const()[name = tensor("op_45402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45402_cast_fp16 = einsum(equation = var_45402_equation_0, values = (var_45132_cast_fp16, var_44797_cast_fp16))[name = tensor("op_45402_cast_fp16")]; + tensor var_45403_to_fp16 = const()[name = tensor("op_45403_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4395_cast_fp16 = mul(x = var_45402_cast_fp16, y = var_45403_to_fp16)[name = tensor("aw_chunk_4395_cast_fp16")]; + tensor var_45406_equation_0 = const()[name = tensor("op_45406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45406_cast_fp16 = einsum(equation = var_45406_equation_0, values = (var_45132_cast_fp16, var_44804_cast_fp16))[name = tensor("op_45406_cast_fp16")]; + tensor var_45407_to_fp16 = const()[name = tensor("op_45407_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4397_cast_fp16 = mul(x = var_45406_cast_fp16, y = var_45407_to_fp16)[name = tensor("aw_chunk_4397_cast_fp16")]; + tensor var_45410_equation_0 = const()[name = tensor("op_45410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45410_cast_fp16 = einsum(equation = var_45410_equation_0, values = (var_45132_cast_fp16, var_44811_cast_fp16))[name = tensor("op_45410_cast_fp16")]; + tensor var_45411_to_fp16 = const()[name = tensor("op_45411_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4399_cast_fp16 = mul(x = var_45410_cast_fp16, y = var_45411_to_fp16)[name = tensor("aw_chunk_4399_cast_fp16")]; + tensor var_45414_equation_0 = const()[name = tensor("op_45414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45414_cast_fp16 = einsum(equation = var_45414_equation_0, values = (var_45136_cast_fp16, var_44818_cast_fp16))[name = tensor("op_45414_cast_fp16")]; + tensor var_45415_to_fp16 = const()[name = tensor("op_45415_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4401_cast_fp16 = mul(x = var_45414_cast_fp16, y = var_45415_to_fp16)[name = tensor("aw_chunk_4401_cast_fp16")]; + tensor var_45418_equation_0 = const()[name = tensor("op_45418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45418_cast_fp16 = einsum(equation = var_45418_equation_0, values = (var_45136_cast_fp16, var_44825_cast_fp16))[name = tensor("op_45418_cast_fp16")]; + tensor var_45419_to_fp16 = const()[name = tensor("op_45419_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4403_cast_fp16 = mul(x = var_45418_cast_fp16, y = var_45419_to_fp16)[name = tensor("aw_chunk_4403_cast_fp16")]; + tensor var_45422_equation_0 = const()[name = tensor("op_45422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45422_cast_fp16 = einsum(equation = var_45422_equation_0, values = (var_45136_cast_fp16, var_44832_cast_fp16))[name = tensor("op_45422_cast_fp16")]; + tensor var_45423_to_fp16 = const()[name = tensor("op_45423_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4405_cast_fp16 = mul(x = var_45422_cast_fp16, y = var_45423_to_fp16)[name = tensor("aw_chunk_4405_cast_fp16")]; + tensor var_45426_equation_0 = const()[name = tensor("op_45426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45426_cast_fp16 = einsum(equation = var_45426_equation_0, values = (var_45136_cast_fp16, var_44839_cast_fp16))[name = tensor("op_45426_cast_fp16")]; + tensor var_45427_to_fp16 = const()[name = tensor("op_45427_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4407_cast_fp16 = mul(x = var_45426_cast_fp16, y = var_45427_to_fp16)[name = tensor("aw_chunk_4407_cast_fp16")]; + tensor var_45430_equation_0 = const()[name = tensor("op_45430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45430_cast_fp16 = einsum(equation = var_45430_equation_0, values = (var_45140_cast_fp16, var_44846_cast_fp16))[name = tensor("op_45430_cast_fp16")]; + tensor var_45431_to_fp16 = const()[name = tensor("op_45431_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4409_cast_fp16 = mul(x = var_45430_cast_fp16, y = var_45431_to_fp16)[name = tensor("aw_chunk_4409_cast_fp16")]; + tensor var_45434_equation_0 = const()[name = tensor("op_45434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45434_cast_fp16 = einsum(equation = var_45434_equation_0, values = (var_45140_cast_fp16, var_44853_cast_fp16))[name = tensor("op_45434_cast_fp16")]; + tensor var_45435_to_fp16 = const()[name = tensor("op_45435_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4411_cast_fp16 = mul(x = var_45434_cast_fp16, y = var_45435_to_fp16)[name = tensor("aw_chunk_4411_cast_fp16")]; + tensor var_45438_equation_0 = const()[name = tensor("op_45438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45438_cast_fp16 = einsum(equation = var_45438_equation_0, values = (var_45140_cast_fp16, var_44860_cast_fp16))[name = tensor("op_45438_cast_fp16")]; + tensor var_45439_to_fp16 = const()[name = tensor("op_45439_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4413_cast_fp16 = mul(x = var_45438_cast_fp16, y = var_45439_to_fp16)[name = tensor("aw_chunk_4413_cast_fp16")]; + tensor var_45442_equation_0 = const()[name = tensor("op_45442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45442_cast_fp16 = einsum(equation = var_45442_equation_0, values = (var_45140_cast_fp16, var_44867_cast_fp16))[name = tensor("op_45442_cast_fp16")]; + tensor var_45443_to_fp16 = const()[name = tensor("op_45443_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4415_cast_fp16 = mul(x = var_45442_cast_fp16, y = var_45443_to_fp16)[name = tensor("aw_chunk_4415_cast_fp16")]; + tensor var_45446_equation_0 = const()[name = tensor("op_45446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45446_cast_fp16 = einsum(equation = var_45446_equation_0, values = (var_45144_cast_fp16, var_44874_cast_fp16))[name = tensor("op_45446_cast_fp16")]; + tensor var_45447_to_fp16 = const()[name = tensor("op_45447_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4417_cast_fp16 = mul(x = var_45446_cast_fp16, y = var_45447_to_fp16)[name = tensor("aw_chunk_4417_cast_fp16")]; + tensor var_45450_equation_0 = const()[name = tensor("op_45450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45450_cast_fp16 = einsum(equation = var_45450_equation_0, values = (var_45144_cast_fp16, var_44881_cast_fp16))[name = tensor("op_45450_cast_fp16")]; + tensor var_45451_to_fp16 = const()[name = tensor("op_45451_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4419_cast_fp16 = mul(x = var_45450_cast_fp16, y = var_45451_to_fp16)[name = tensor("aw_chunk_4419_cast_fp16")]; + tensor var_45454_equation_0 = const()[name = tensor("op_45454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45454_cast_fp16 = einsum(equation = var_45454_equation_0, values = (var_45144_cast_fp16, var_44888_cast_fp16))[name = tensor("op_45454_cast_fp16")]; + tensor var_45455_to_fp16 = const()[name = tensor("op_45455_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4421_cast_fp16 = mul(x = var_45454_cast_fp16, y = var_45455_to_fp16)[name = tensor("aw_chunk_4421_cast_fp16")]; + tensor var_45458_equation_0 = const()[name = tensor("op_45458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45458_cast_fp16 = einsum(equation = var_45458_equation_0, values = (var_45144_cast_fp16, var_44895_cast_fp16))[name = tensor("op_45458_cast_fp16")]; + tensor var_45459_to_fp16 = const()[name = tensor("op_45459_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4423_cast_fp16 = mul(x = var_45458_cast_fp16, y = var_45459_to_fp16)[name = tensor("aw_chunk_4423_cast_fp16")]; + tensor var_45462_equation_0 = const()[name = tensor("op_45462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45462_cast_fp16 = einsum(equation = var_45462_equation_0, values = (var_45148_cast_fp16, var_44902_cast_fp16))[name = tensor("op_45462_cast_fp16")]; + tensor var_45463_to_fp16 = const()[name = tensor("op_45463_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4425_cast_fp16 = mul(x = var_45462_cast_fp16, y = var_45463_to_fp16)[name = tensor("aw_chunk_4425_cast_fp16")]; + tensor var_45466_equation_0 = const()[name = tensor("op_45466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45466_cast_fp16 = einsum(equation = var_45466_equation_0, values = (var_45148_cast_fp16, var_44909_cast_fp16))[name = tensor("op_45466_cast_fp16")]; + tensor var_45467_to_fp16 = const()[name = tensor("op_45467_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4427_cast_fp16 = mul(x = var_45466_cast_fp16, y = var_45467_to_fp16)[name = tensor("aw_chunk_4427_cast_fp16")]; + tensor var_45470_equation_0 = const()[name = tensor("op_45470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45470_cast_fp16 = einsum(equation = var_45470_equation_0, values = (var_45148_cast_fp16, var_44916_cast_fp16))[name = tensor("op_45470_cast_fp16")]; + tensor var_45471_to_fp16 = const()[name = tensor("op_45471_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4429_cast_fp16 = mul(x = var_45470_cast_fp16, y = var_45471_to_fp16)[name = tensor("aw_chunk_4429_cast_fp16")]; + tensor var_45474_equation_0 = const()[name = tensor("op_45474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45474_cast_fp16 = einsum(equation = var_45474_equation_0, values = (var_45148_cast_fp16, var_44923_cast_fp16))[name = tensor("op_45474_cast_fp16")]; + tensor var_45475_to_fp16 = const()[name = tensor("op_45475_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4431_cast_fp16 = mul(x = var_45474_cast_fp16, y = var_45475_to_fp16)[name = tensor("aw_chunk_4431_cast_fp16")]; + tensor var_45478_equation_0 = const()[name = tensor("op_45478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45478_cast_fp16 = einsum(equation = var_45478_equation_0, values = (var_45152_cast_fp16, var_44930_cast_fp16))[name = tensor("op_45478_cast_fp16")]; + tensor var_45479_to_fp16 = const()[name = tensor("op_45479_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4433_cast_fp16 = mul(x = var_45478_cast_fp16, y = var_45479_to_fp16)[name = tensor("aw_chunk_4433_cast_fp16")]; + tensor var_45482_equation_0 = const()[name = tensor("op_45482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45482_cast_fp16 = einsum(equation = var_45482_equation_0, values = (var_45152_cast_fp16, var_44937_cast_fp16))[name = tensor("op_45482_cast_fp16")]; + tensor var_45483_to_fp16 = const()[name = tensor("op_45483_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4435_cast_fp16 = mul(x = var_45482_cast_fp16, y = var_45483_to_fp16)[name = tensor("aw_chunk_4435_cast_fp16")]; + tensor var_45486_equation_0 = const()[name = tensor("op_45486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45486_cast_fp16 = einsum(equation = var_45486_equation_0, values = (var_45152_cast_fp16, var_44944_cast_fp16))[name = tensor("op_45486_cast_fp16")]; + tensor var_45487_to_fp16 = const()[name = tensor("op_45487_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4437_cast_fp16 = mul(x = var_45486_cast_fp16, y = var_45487_to_fp16)[name = tensor("aw_chunk_4437_cast_fp16")]; + tensor var_45490_equation_0 = const()[name = tensor("op_45490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45490_cast_fp16 = einsum(equation = var_45490_equation_0, values = (var_45152_cast_fp16, var_44951_cast_fp16))[name = tensor("op_45490_cast_fp16")]; + tensor var_45491_to_fp16 = const()[name = tensor("op_45491_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4439_cast_fp16 = mul(x = var_45490_cast_fp16, y = var_45491_to_fp16)[name = tensor("aw_chunk_4439_cast_fp16")]; + tensor var_45494_equation_0 = const()[name = tensor("op_45494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45494_cast_fp16 = einsum(equation = var_45494_equation_0, values = (var_45156_cast_fp16, var_44958_cast_fp16))[name = tensor("op_45494_cast_fp16")]; + tensor var_45495_to_fp16 = const()[name = tensor("op_45495_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4441_cast_fp16 = mul(x = var_45494_cast_fp16, y = var_45495_to_fp16)[name = tensor("aw_chunk_4441_cast_fp16")]; + tensor var_45498_equation_0 = const()[name = tensor("op_45498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45498_cast_fp16 = einsum(equation = var_45498_equation_0, values = (var_45156_cast_fp16, var_44965_cast_fp16))[name = tensor("op_45498_cast_fp16")]; + tensor var_45499_to_fp16 = const()[name = tensor("op_45499_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4443_cast_fp16 = mul(x = var_45498_cast_fp16, y = var_45499_to_fp16)[name = tensor("aw_chunk_4443_cast_fp16")]; + tensor var_45502_equation_0 = const()[name = tensor("op_45502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45502_cast_fp16 = einsum(equation = var_45502_equation_0, values = (var_45156_cast_fp16, var_44972_cast_fp16))[name = tensor("op_45502_cast_fp16")]; + tensor var_45503_to_fp16 = const()[name = tensor("op_45503_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4445_cast_fp16 = mul(x = var_45502_cast_fp16, y = var_45503_to_fp16)[name = tensor("aw_chunk_4445_cast_fp16")]; + tensor var_45506_equation_0 = const()[name = tensor("op_45506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45506_cast_fp16 = einsum(equation = var_45506_equation_0, values = (var_45156_cast_fp16, var_44979_cast_fp16))[name = tensor("op_45506_cast_fp16")]; + tensor var_45507_to_fp16 = const()[name = tensor("op_45507_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4447_cast_fp16 = mul(x = var_45506_cast_fp16, y = var_45507_to_fp16)[name = tensor("aw_chunk_4447_cast_fp16")]; + tensor var_45510_equation_0 = const()[name = tensor("op_45510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45510_cast_fp16 = einsum(equation = var_45510_equation_0, values = (var_45160_cast_fp16, var_44986_cast_fp16))[name = tensor("op_45510_cast_fp16")]; + tensor var_45511_to_fp16 = const()[name = tensor("op_45511_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4449_cast_fp16 = mul(x = var_45510_cast_fp16, y = var_45511_to_fp16)[name = tensor("aw_chunk_4449_cast_fp16")]; + tensor var_45514_equation_0 = const()[name = tensor("op_45514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45514_cast_fp16 = einsum(equation = var_45514_equation_0, values = (var_45160_cast_fp16, var_44993_cast_fp16))[name = tensor("op_45514_cast_fp16")]; + tensor var_45515_to_fp16 = const()[name = tensor("op_45515_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4451_cast_fp16 = mul(x = var_45514_cast_fp16, y = var_45515_to_fp16)[name = tensor("aw_chunk_4451_cast_fp16")]; + tensor var_45518_equation_0 = const()[name = tensor("op_45518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45518_cast_fp16 = einsum(equation = var_45518_equation_0, values = (var_45160_cast_fp16, var_45000_cast_fp16))[name = tensor("op_45518_cast_fp16")]; + tensor var_45519_to_fp16 = const()[name = tensor("op_45519_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4453_cast_fp16 = mul(x = var_45518_cast_fp16, y = var_45519_to_fp16)[name = tensor("aw_chunk_4453_cast_fp16")]; + tensor var_45522_equation_0 = const()[name = tensor("op_45522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45522_cast_fp16 = einsum(equation = var_45522_equation_0, values = (var_45160_cast_fp16, var_45007_cast_fp16))[name = tensor("op_45522_cast_fp16")]; + tensor var_45523_to_fp16 = const()[name = tensor("op_45523_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4455_cast_fp16 = mul(x = var_45522_cast_fp16, y = var_45523_to_fp16)[name = tensor("aw_chunk_4455_cast_fp16")]; + tensor var_45526_equation_0 = const()[name = tensor("op_45526_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45526_cast_fp16 = einsum(equation = var_45526_equation_0, values = (var_45164_cast_fp16, var_45014_cast_fp16))[name = tensor("op_45526_cast_fp16")]; + tensor var_45527_to_fp16 = const()[name = tensor("op_45527_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4457_cast_fp16 = mul(x = var_45526_cast_fp16, y = var_45527_to_fp16)[name = tensor("aw_chunk_4457_cast_fp16")]; + tensor var_45530_equation_0 = const()[name = tensor("op_45530_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45530_cast_fp16 = einsum(equation = var_45530_equation_0, values = (var_45164_cast_fp16, var_45021_cast_fp16))[name = tensor("op_45530_cast_fp16")]; + tensor var_45531_to_fp16 = const()[name = tensor("op_45531_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4459_cast_fp16 = mul(x = var_45530_cast_fp16, y = var_45531_to_fp16)[name = tensor("aw_chunk_4459_cast_fp16")]; + tensor var_45534_equation_0 = const()[name = tensor("op_45534_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45534_cast_fp16 = einsum(equation = var_45534_equation_0, values = (var_45164_cast_fp16, var_45028_cast_fp16))[name = tensor("op_45534_cast_fp16")]; + tensor var_45535_to_fp16 = const()[name = tensor("op_45535_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4461_cast_fp16 = mul(x = var_45534_cast_fp16, y = var_45535_to_fp16)[name = tensor("aw_chunk_4461_cast_fp16")]; + tensor var_45538_equation_0 = const()[name = tensor("op_45538_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45538_cast_fp16 = einsum(equation = var_45538_equation_0, values = (var_45164_cast_fp16, var_45035_cast_fp16))[name = tensor("op_45538_cast_fp16")]; + tensor var_45539_to_fp16 = const()[name = tensor("op_45539_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4463_cast_fp16 = mul(x = var_45538_cast_fp16, y = var_45539_to_fp16)[name = tensor("aw_chunk_4463_cast_fp16")]; + tensor var_45542_equation_0 = const()[name = tensor("op_45542_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45542_cast_fp16 = einsum(equation = var_45542_equation_0, values = (var_45168_cast_fp16, var_45042_cast_fp16))[name = tensor("op_45542_cast_fp16")]; + tensor var_45543_to_fp16 = const()[name = tensor("op_45543_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4465_cast_fp16 = mul(x = var_45542_cast_fp16, y = var_45543_to_fp16)[name = tensor("aw_chunk_4465_cast_fp16")]; + tensor var_45546_equation_0 = const()[name = tensor("op_45546_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45546_cast_fp16 = einsum(equation = var_45546_equation_0, values = (var_45168_cast_fp16, var_45049_cast_fp16))[name = tensor("op_45546_cast_fp16")]; + tensor var_45547_to_fp16 = const()[name = tensor("op_45547_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4467_cast_fp16 = mul(x = var_45546_cast_fp16, y = var_45547_to_fp16)[name = tensor("aw_chunk_4467_cast_fp16")]; + tensor var_45550_equation_0 = const()[name = tensor("op_45550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45550_cast_fp16 = einsum(equation = var_45550_equation_0, values = (var_45168_cast_fp16, var_45056_cast_fp16))[name = tensor("op_45550_cast_fp16")]; + tensor var_45551_to_fp16 = const()[name = tensor("op_45551_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4469_cast_fp16 = mul(x = var_45550_cast_fp16, y = var_45551_to_fp16)[name = tensor("aw_chunk_4469_cast_fp16")]; + tensor var_45554_equation_0 = const()[name = tensor("op_45554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45554_cast_fp16 = einsum(equation = var_45554_equation_0, values = (var_45168_cast_fp16, var_45063_cast_fp16))[name = tensor("op_45554_cast_fp16")]; + tensor var_45555_to_fp16 = const()[name = tensor("op_45555_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4471_cast_fp16 = mul(x = var_45554_cast_fp16, y = var_45555_to_fp16)[name = tensor("aw_chunk_4471_cast_fp16")]; + tensor var_45558_equation_0 = const()[name = tensor("op_45558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45558_cast_fp16 = einsum(equation = var_45558_equation_0, values = (var_45172_cast_fp16, var_45070_cast_fp16))[name = tensor("op_45558_cast_fp16")]; + tensor var_45559_to_fp16 = const()[name = tensor("op_45559_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4473_cast_fp16 = mul(x = var_45558_cast_fp16, y = var_45559_to_fp16)[name = tensor("aw_chunk_4473_cast_fp16")]; + tensor var_45562_equation_0 = const()[name = tensor("op_45562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45562_cast_fp16 = einsum(equation = var_45562_equation_0, values = (var_45172_cast_fp16, var_45077_cast_fp16))[name = tensor("op_45562_cast_fp16")]; + tensor var_45563_to_fp16 = const()[name = tensor("op_45563_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4475_cast_fp16 = mul(x = var_45562_cast_fp16, y = var_45563_to_fp16)[name = tensor("aw_chunk_4475_cast_fp16")]; + tensor var_45566_equation_0 = const()[name = tensor("op_45566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45566_cast_fp16 = einsum(equation = var_45566_equation_0, values = (var_45172_cast_fp16, var_45084_cast_fp16))[name = tensor("op_45566_cast_fp16")]; + tensor var_45567_to_fp16 = const()[name = tensor("op_45567_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4477_cast_fp16 = mul(x = var_45566_cast_fp16, y = var_45567_to_fp16)[name = tensor("aw_chunk_4477_cast_fp16")]; + tensor var_45570_equation_0 = const()[name = tensor("op_45570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_45570_cast_fp16 = einsum(equation = var_45570_equation_0, values = (var_45172_cast_fp16, var_45091_cast_fp16))[name = tensor("op_45570_cast_fp16")]; + tensor var_45571_to_fp16 = const()[name = tensor("op_45571_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4479_cast_fp16 = mul(x = var_45570_cast_fp16, y = var_45571_to_fp16)[name = tensor("aw_chunk_4479_cast_fp16")]; + tensor var_45573_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4321_cast_fp16)[name = tensor("op_45573_cast_fp16")]; + tensor var_45574_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4323_cast_fp16)[name = tensor("op_45574_cast_fp16")]; + tensor var_45575_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4325_cast_fp16)[name = tensor("op_45575_cast_fp16")]; + tensor var_45576_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4327_cast_fp16)[name = tensor("op_45576_cast_fp16")]; + tensor var_45577_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4329_cast_fp16)[name = tensor("op_45577_cast_fp16")]; + tensor var_45578_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4331_cast_fp16)[name = tensor("op_45578_cast_fp16")]; + tensor var_45579_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4333_cast_fp16)[name = tensor("op_45579_cast_fp16")]; + tensor var_45580_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4335_cast_fp16)[name = tensor("op_45580_cast_fp16")]; + tensor var_45581_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4337_cast_fp16)[name = tensor("op_45581_cast_fp16")]; + tensor var_45582_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4339_cast_fp16)[name = tensor("op_45582_cast_fp16")]; + tensor var_45583_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4341_cast_fp16)[name = tensor("op_45583_cast_fp16")]; + tensor var_45584_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4343_cast_fp16)[name = tensor("op_45584_cast_fp16")]; + tensor var_45585_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4345_cast_fp16)[name = tensor("op_45585_cast_fp16")]; + tensor var_45586_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4347_cast_fp16)[name = tensor("op_45586_cast_fp16")]; + tensor var_45587_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4349_cast_fp16)[name = tensor("op_45587_cast_fp16")]; + tensor var_45588_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4351_cast_fp16)[name = tensor("op_45588_cast_fp16")]; + tensor var_45589_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4353_cast_fp16)[name = tensor("op_45589_cast_fp16")]; + tensor var_45590_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4355_cast_fp16)[name = tensor("op_45590_cast_fp16")]; + tensor var_45591_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4357_cast_fp16)[name = tensor("op_45591_cast_fp16")]; + tensor var_45592_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4359_cast_fp16)[name = tensor("op_45592_cast_fp16")]; + tensor var_45593_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4361_cast_fp16)[name = tensor("op_45593_cast_fp16")]; + tensor var_45594_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4363_cast_fp16)[name = tensor("op_45594_cast_fp16")]; + tensor var_45595_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4365_cast_fp16)[name = tensor("op_45595_cast_fp16")]; + tensor var_45596_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4367_cast_fp16)[name = tensor("op_45596_cast_fp16")]; + tensor var_45597_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4369_cast_fp16)[name = tensor("op_45597_cast_fp16")]; + tensor var_45598_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4371_cast_fp16)[name = tensor("op_45598_cast_fp16")]; + tensor var_45599_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4373_cast_fp16)[name = tensor("op_45599_cast_fp16")]; + tensor var_45600_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4375_cast_fp16)[name = tensor("op_45600_cast_fp16")]; + tensor var_45601_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4377_cast_fp16)[name = tensor("op_45601_cast_fp16")]; + tensor var_45602_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4379_cast_fp16)[name = tensor("op_45602_cast_fp16")]; + tensor var_45603_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4381_cast_fp16)[name = tensor("op_45603_cast_fp16")]; + tensor var_45604_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4383_cast_fp16)[name = tensor("op_45604_cast_fp16")]; + tensor var_45605_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4385_cast_fp16)[name = tensor("op_45605_cast_fp16")]; + tensor var_45606_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4387_cast_fp16)[name = tensor("op_45606_cast_fp16")]; + tensor var_45607_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4389_cast_fp16)[name = tensor("op_45607_cast_fp16")]; + tensor var_45608_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4391_cast_fp16)[name = tensor("op_45608_cast_fp16")]; + tensor var_45609_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4393_cast_fp16)[name = tensor("op_45609_cast_fp16")]; + tensor var_45610_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4395_cast_fp16)[name = tensor("op_45610_cast_fp16")]; + tensor var_45611_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4397_cast_fp16)[name = tensor("op_45611_cast_fp16")]; + tensor var_45612_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4399_cast_fp16)[name = tensor("op_45612_cast_fp16")]; + tensor var_45613_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4401_cast_fp16)[name = tensor("op_45613_cast_fp16")]; + tensor var_45614_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4403_cast_fp16)[name = tensor("op_45614_cast_fp16")]; + tensor var_45615_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4405_cast_fp16)[name = tensor("op_45615_cast_fp16")]; + tensor var_45616_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4407_cast_fp16)[name = tensor("op_45616_cast_fp16")]; + tensor var_45617_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4409_cast_fp16)[name = tensor("op_45617_cast_fp16")]; + tensor var_45618_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4411_cast_fp16)[name = tensor("op_45618_cast_fp16")]; + tensor var_45619_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4413_cast_fp16)[name = tensor("op_45619_cast_fp16")]; + tensor var_45620_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4415_cast_fp16)[name = tensor("op_45620_cast_fp16")]; + tensor var_45621_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4417_cast_fp16)[name = tensor("op_45621_cast_fp16")]; + tensor var_45622_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4419_cast_fp16)[name = tensor("op_45622_cast_fp16")]; + tensor var_45623_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4421_cast_fp16)[name = tensor("op_45623_cast_fp16")]; + tensor var_45624_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4423_cast_fp16)[name = tensor("op_45624_cast_fp16")]; + tensor var_45625_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4425_cast_fp16)[name = tensor("op_45625_cast_fp16")]; + tensor var_45626_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4427_cast_fp16)[name = tensor("op_45626_cast_fp16")]; + tensor var_45627_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4429_cast_fp16)[name = tensor("op_45627_cast_fp16")]; + tensor var_45628_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4431_cast_fp16)[name = tensor("op_45628_cast_fp16")]; + tensor var_45629_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4433_cast_fp16)[name = tensor("op_45629_cast_fp16")]; + tensor var_45630_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4435_cast_fp16)[name = tensor("op_45630_cast_fp16")]; + tensor var_45631_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4437_cast_fp16)[name = tensor("op_45631_cast_fp16")]; + tensor var_45632_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4439_cast_fp16)[name = tensor("op_45632_cast_fp16")]; + tensor var_45633_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4441_cast_fp16)[name = tensor("op_45633_cast_fp16")]; + tensor var_45634_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4443_cast_fp16)[name = tensor("op_45634_cast_fp16")]; + tensor var_45635_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4445_cast_fp16)[name = tensor("op_45635_cast_fp16")]; + tensor var_45636_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4447_cast_fp16)[name = tensor("op_45636_cast_fp16")]; + tensor var_45637_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4449_cast_fp16)[name = tensor("op_45637_cast_fp16")]; + tensor var_45638_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4451_cast_fp16)[name = tensor("op_45638_cast_fp16")]; + tensor var_45639_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4453_cast_fp16)[name = tensor("op_45639_cast_fp16")]; + tensor var_45640_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4455_cast_fp16)[name = tensor("op_45640_cast_fp16")]; + tensor var_45641_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4457_cast_fp16)[name = tensor("op_45641_cast_fp16")]; + tensor var_45642_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4459_cast_fp16)[name = tensor("op_45642_cast_fp16")]; + tensor var_45643_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4461_cast_fp16)[name = tensor("op_45643_cast_fp16")]; + tensor var_45644_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4463_cast_fp16)[name = tensor("op_45644_cast_fp16")]; + tensor var_45645_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4465_cast_fp16)[name = tensor("op_45645_cast_fp16")]; + tensor var_45646_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4467_cast_fp16)[name = tensor("op_45646_cast_fp16")]; + tensor var_45647_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4469_cast_fp16)[name = tensor("op_45647_cast_fp16")]; + tensor var_45648_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4471_cast_fp16)[name = tensor("op_45648_cast_fp16")]; + tensor var_45649_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4473_cast_fp16)[name = tensor("op_45649_cast_fp16")]; + tensor var_45650_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4475_cast_fp16)[name = tensor("op_45650_cast_fp16")]; + tensor var_45651_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4477_cast_fp16)[name = tensor("op_45651_cast_fp16")]; + tensor var_45652_cast_fp16 = softmax(axis = var_44344, x = aw_chunk_4479_cast_fp16)[name = tensor("op_45652_cast_fp16")]; + tensor var_45654_equation_0 = const()[name = tensor("op_45654_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45654_cast_fp16 = einsum(equation = var_45654_equation_0, values = (var_45174_cast_fp16, var_45573_cast_fp16))[name = tensor("op_45654_cast_fp16")]; + tensor var_45656_equation_0 = const()[name = tensor("op_45656_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45656_cast_fp16 = einsum(equation = var_45656_equation_0, values = (var_45174_cast_fp16, var_45574_cast_fp16))[name = tensor("op_45656_cast_fp16")]; + tensor var_45658_equation_0 = const()[name = tensor("op_45658_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45658_cast_fp16 = einsum(equation = var_45658_equation_0, values = (var_45174_cast_fp16, var_45575_cast_fp16))[name = tensor("op_45658_cast_fp16")]; + tensor var_45660_equation_0 = const()[name = tensor("op_45660_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45660_cast_fp16 = einsum(equation = var_45660_equation_0, values = (var_45174_cast_fp16, var_45576_cast_fp16))[name = tensor("op_45660_cast_fp16")]; + tensor var_45662_equation_0 = const()[name = tensor("op_45662_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45662_cast_fp16 = einsum(equation = var_45662_equation_0, values = (var_45178_cast_fp16, var_45577_cast_fp16))[name = tensor("op_45662_cast_fp16")]; + tensor var_45664_equation_0 = const()[name = tensor("op_45664_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45664_cast_fp16 = einsum(equation = var_45664_equation_0, values = (var_45178_cast_fp16, var_45578_cast_fp16))[name = tensor("op_45664_cast_fp16")]; + tensor var_45666_equation_0 = const()[name = tensor("op_45666_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45666_cast_fp16 = einsum(equation = var_45666_equation_0, values = (var_45178_cast_fp16, var_45579_cast_fp16))[name = tensor("op_45666_cast_fp16")]; + tensor var_45668_equation_0 = const()[name = tensor("op_45668_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45668_cast_fp16 = einsum(equation = var_45668_equation_0, values = (var_45178_cast_fp16, var_45580_cast_fp16))[name = tensor("op_45668_cast_fp16")]; + tensor var_45670_equation_0 = const()[name = tensor("op_45670_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45670_cast_fp16 = einsum(equation = var_45670_equation_0, values = (var_45182_cast_fp16, var_45581_cast_fp16))[name = tensor("op_45670_cast_fp16")]; + tensor var_45672_equation_0 = const()[name = tensor("op_45672_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45672_cast_fp16 = einsum(equation = var_45672_equation_0, values = (var_45182_cast_fp16, var_45582_cast_fp16))[name = tensor("op_45672_cast_fp16")]; + tensor var_45674_equation_0 = const()[name = tensor("op_45674_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45674_cast_fp16 = einsum(equation = var_45674_equation_0, values = (var_45182_cast_fp16, var_45583_cast_fp16))[name = tensor("op_45674_cast_fp16")]; + tensor var_45676_equation_0 = const()[name = tensor("op_45676_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45676_cast_fp16 = einsum(equation = var_45676_equation_0, values = (var_45182_cast_fp16, var_45584_cast_fp16))[name = tensor("op_45676_cast_fp16")]; + tensor var_45678_equation_0 = const()[name = tensor("op_45678_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45678_cast_fp16 = einsum(equation = var_45678_equation_0, values = (var_45186_cast_fp16, var_45585_cast_fp16))[name = tensor("op_45678_cast_fp16")]; + tensor var_45680_equation_0 = const()[name = tensor("op_45680_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45680_cast_fp16 = einsum(equation = var_45680_equation_0, values = (var_45186_cast_fp16, var_45586_cast_fp16))[name = tensor("op_45680_cast_fp16")]; + tensor var_45682_equation_0 = const()[name = tensor("op_45682_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45682_cast_fp16 = einsum(equation = var_45682_equation_0, values = (var_45186_cast_fp16, var_45587_cast_fp16))[name = tensor("op_45682_cast_fp16")]; + tensor var_45684_equation_0 = const()[name = tensor("op_45684_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45684_cast_fp16 = einsum(equation = var_45684_equation_0, values = (var_45186_cast_fp16, var_45588_cast_fp16))[name = tensor("op_45684_cast_fp16")]; + tensor var_45686_equation_0 = const()[name = tensor("op_45686_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45686_cast_fp16 = einsum(equation = var_45686_equation_0, values = (var_45190_cast_fp16, var_45589_cast_fp16))[name = tensor("op_45686_cast_fp16")]; + tensor var_45688_equation_0 = const()[name = tensor("op_45688_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45688_cast_fp16 = einsum(equation = var_45688_equation_0, values = (var_45190_cast_fp16, var_45590_cast_fp16))[name = tensor("op_45688_cast_fp16")]; + tensor var_45690_equation_0 = const()[name = tensor("op_45690_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45690_cast_fp16 = einsum(equation = var_45690_equation_0, values = (var_45190_cast_fp16, var_45591_cast_fp16))[name = tensor("op_45690_cast_fp16")]; + tensor var_45692_equation_0 = const()[name = tensor("op_45692_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45692_cast_fp16 = einsum(equation = var_45692_equation_0, values = (var_45190_cast_fp16, var_45592_cast_fp16))[name = tensor("op_45692_cast_fp16")]; + tensor var_45694_equation_0 = const()[name = tensor("op_45694_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45694_cast_fp16 = einsum(equation = var_45694_equation_0, values = (var_45194_cast_fp16, var_45593_cast_fp16))[name = tensor("op_45694_cast_fp16")]; + tensor var_45696_equation_0 = const()[name = tensor("op_45696_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45696_cast_fp16 = einsum(equation = var_45696_equation_0, values = (var_45194_cast_fp16, var_45594_cast_fp16))[name = tensor("op_45696_cast_fp16")]; + tensor var_45698_equation_0 = const()[name = tensor("op_45698_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45698_cast_fp16 = einsum(equation = var_45698_equation_0, values = (var_45194_cast_fp16, var_45595_cast_fp16))[name = tensor("op_45698_cast_fp16")]; + tensor var_45700_equation_0 = const()[name = tensor("op_45700_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45700_cast_fp16 = einsum(equation = var_45700_equation_0, values = (var_45194_cast_fp16, var_45596_cast_fp16))[name = tensor("op_45700_cast_fp16")]; + tensor var_45702_equation_0 = const()[name = tensor("op_45702_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45702_cast_fp16 = einsum(equation = var_45702_equation_0, values = (var_45198_cast_fp16, var_45597_cast_fp16))[name = tensor("op_45702_cast_fp16")]; + tensor var_45704_equation_0 = const()[name = tensor("op_45704_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45704_cast_fp16 = einsum(equation = var_45704_equation_0, values = (var_45198_cast_fp16, var_45598_cast_fp16))[name = tensor("op_45704_cast_fp16")]; + tensor var_45706_equation_0 = const()[name = tensor("op_45706_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45706_cast_fp16 = einsum(equation = var_45706_equation_0, values = (var_45198_cast_fp16, var_45599_cast_fp16))[name = tensor("op_45706_cast_fp16")]; + tensor var_45708_equation_0 = const()[name = tensor("op_45708_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45708_cast_fp16 = einsum(equation = var_45708_equation_0, values = (var_45198_cast_fp16, var_45600_cast_fp16))[name = tensor("op_45708_cast_fp16")]; + tensor var_45710_equation_0 = const()[name = tensor("op_45710_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45710_cast_fp16 = einsum(equation = var_45710_equation_0, values = (var_45202_cast_fp16, var_45601_cast_fp16))[name = tensor("op_45710_cast_fp16")]; + tensor var_45712_equation_0 = const()[name = tensor("op_45712_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45712_cast_fp16 = einsum(equation = var_45712_equation_0, values = (var_45202_cast_fp16, var_45602_cast_fp16))[name = tensor("op_45712_cast_fp16")]; + tensor var_45714_equation_0 = const()[name = tensor("op_45714_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45714_cast_fp16 = einsum(equation = var_45714_equation_0, values = (var_45202_cast_fp16, var_45603_cast_fp16))[name = tensor("op_45714_cast_fp16")]; + tensor var_45716_equation_0 = const()[name = tensor("op_45716_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45716_cast_fp16 = einsum(equation = var_45716_equation_0, values = (var_45202_cast_fp16, var_45604_cast_fp16))[name = tensor("op_45716_cast_fp16")]; + tensor var_45718_equation_0 = const()[name = tensor("op_45718_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45718_cast_fp16 = einsum(equation = var_45718_equation_0, values = (var_45206_cast_fp16, var_45605_cast_fp16))[name = tensor("op_45718_cast_fp16")]; + tensor var_45720_equation_0 = const()[name = tensor("op_45720_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45720_cast_fp16 = einsum(equation = var_45720_equation_0, values = (var_45206_cast_fp16, var_45606_cast_fp16))[name = tensor("op_45720_cast_fp16")]; + tensor var_45722_equation_0 = const()[name = tensor("op_45722_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45722_cast_fp16 = einsum(equation = var_45722_equation_0, values = (var_45206_cast_fp16, var_45607_cast_fp16))[name = tensor("op_45722_cast_fp16")]; + tensor var_45724_equation_0 = const()[name = tensor("op_45724_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45724_cast_fp16 = einsum(equation = var_45724_equation_0, values = (var_45206_cast_fp16, var_45608_cast_fp16))[name = tensor("op_45724_cast_fp16")]; + tensor var_45726_equation_0 = const()[name = tensor("op_45726_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45726_cast_fp16 = einsum(equation = var_45726_equation_0, values = (var_45210_cast_fp16, var_45609_cast_fp16))[name = tensor("op_45726_cast_fp16")]; + tensor var_45728_equation_0 = const()[name = tensor("op_45728_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45728_cast_fp16 = einsum(equation = var_45728_equation_0, values = (var_45210_cast_fp16, var_45610_cast_fp16))[name = tensor("op_45728_cast_fp16")]; + tensor var_45730_equation_0 = const()[name = tensor("op_45730_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45730_cast_fp16 = einsum(equation = var_45730_equation_0, values = (var_45210_cast_fp16, var_45611_cast_fp16))[name = tensor("op_45730_cast_fp16")]; + tensor var_45732_equation_0 = const()[name = tensor("op_45732_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45732_cast_fp16 = einsum(equation = var_45732_equation_0, values = (var_45210_cast_fp16, var_45612_cast_fp16))[name = tensor("op_45732_cast_fp16")]; + tensor var_45734_equation_0 = const()[name = tensor("op_45734_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45734_cast_fp16 = einsum(equation = var_45734_equation_0, values = (var_45214_cast_fp16, var_45613_cast_fp16))[name = tensor("op_45734_cast_fp16")]; + tensor var_45736_equation_0 = const()[name = tensor("op_45736_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45736_cast_fp16 = einsum(equation = var_45736_equation_0, values = (var_45214_cast_fp16, var_45614_cast_fp16))[name = tensor("op_45736_cast_fp16")]; + tensor var_45738_equation_0 = const()[name = tensor("op_45738_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45738_cast_fp16 = einsum(equation = var_45738_equation_0, values = (var_45214_cast_fp16, var_45615_cast_fp16))[name = tensor("op_45738_cast_fp16")]; + tensor var_45740_equation_0 = const()[name = tensor("op_45740_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45740_cast_fp16 = einsum(equation = var_45740_equation_0, values = (var_45214_cast_fp16, var_45616_cast_fp16))[name = tensor("op_45740_cast_fp16")]; + tensor var_45742_equation_0 = const()[name = tensor("op_45742_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45742_cast_fp16 = einsum(equation = var_45742_equation_0, values = (var_45218_cast_fp16, var_45617_cast_fp16))[name = tensor("op_45742_cast_fp16")]; + tensor var_45744_equation_0 = const()[name = tensor("op_45744_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45744_cast_fp16 = einsum(equation = var_45744_equation_0, values = (var_45218_cast_fp16, var_45618_cast_fp16))[name = tensor("op_45744_cast_fp16")]; + tensor var_45746_equation_0 = const()[name = tensor("op_45746_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45746_cast_fp16 = einsum(equation = var_45746_equation_0, values = (var_45218_cast_fp16, var_45619_cast_fp16))[name = tensor("op_45746_cast_fp16")]; + tensor var_45748_equation_0 = const()[name = tensor("op_45748_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45748_cast_fp16 = einsum(equation = var_45748_equation_0, values = (var_45218_cast_fp16, var_45620_cast_fp16))[name = tensor("op_45748_cast_fp16")]; + tensor var_45750_equation_0 = const()[name = tensor("op_45750_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45750_cast_fp16 = einsum(equation = var_45750_equation_0, values = (var_45222_cast_fp16, var_45621_cast_fp16))[name = tensor("op_45750_cast_fp16")]; + tensor var_45752_equation_0 = const()[name = tensor("op_45752_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45752_cast_fp16 = einsum(equation = var_45752_equation_0, values = (var_45222_cast_fp16, var_45622_cast_fp16))[name = tensor("op_45752_cast_fp16")]; + tensor var_45754_equation_0 = const()[name = tensor("op_45754_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45754_cast_fp16 = einsum(equation = var_45754_equation_0, values = (var_45222_cast_fp16, var_45623_cast_fp16))[name = tensor("op_45754_cast_fp16")]; + tensor var_45756_equation_0 = const()[name = tensor("op_45756_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45756_cast_fp16 = einsum(equation = var_45756_equation_0, values = (var_45222_cast_fp16, var_45624_cast_fp16))[name = tensor("op_45756_cast_fp16")]; + tensor var_45758_equation_0 = const()[name = tensor("op_45758_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45758_cast_fp16 = einsum(equation = var_45758_equation_0, values = (var_45226_cast_fp16, var_45625_cast_fp16))[name = tensor("op_45758_cast_fp16")]; + tensor var_45760_equation_0 = const()[name = tensor("op_45760_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45760_cast_fp16 = einsum(equation = var_45760_equation_0, values = (var_45226_cast_fp16, var_45626_cast_fp16))[name = tensor("op_45760_cast_fp16")]; + tensor var_45762_equation_0 = const()[name = tensor("op_45762_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45762_cast_fp16 = einsum(equation = var_45762_equation_0, values = (var_45226_cast_fp16, var_45627_cast_fp16))[name = tensor("op_45762_cast_fp16")]; + tensor var_45764_equation_0 = const()[name = tensor("op_45764_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45764_cast_fp16 = einsum(equation = var_45764_equation_0, values = (var_45226_cast_fp16, var_45628_cast_fp16))[name = tensor("op_45764_cast_fp16")]; + tensor var_45766_equation_0 = const()[name = tensor("op_45766_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45766_cast_fp16 = einsum(equation = var_45766_equation_0, values = (var_45230_cast_fp16, var_45629_cast_fp16))[name = tensor("op_45766_cast_fp16")]; + tensor var_45768_equation_0 = const()[name = tensor("op_45768_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45768_cast_fp16 = einsum(equation = var_45768_equation_0, values = (var_45230_cast_fp16, var_45630_cast_fp16))[name = tensor("op_45768_cast_fp16")]; + tensor var_45770_equation_0 = const()[name = tensor("op_45770_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45770_cast_fp16 = einsum(equation = var_45770_equation_0, values = (var_45230_cast_fp16, var_45631_cast_fp16))[name = tensor("op_45770_cast_fp16")]; + tensor var_45772_equation_0 = const()[name = tensor("op_45772_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45772_cast_fp16 = einsum(equation = var_45772_equation_0, values = (var_45230_cast_fp16, var_45632_cast_fp16))[name = tensor("op_45772_cast_fp16")]; + tensor var_45774_equation_0 = const()[name = tensor("op_45774_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45774_cast_fp16 = einsum(equation = var_45774_equation_0, values = (var_45234_cast_fp16, var_45633_cast_fp16))[name = tensor("op_45774_cast_fp16")]; + tensor var_45776_equation_0 = const()[name = tensor("op_45776_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45776_cast_fp16 = einsum(equation = var_45776_equation_0, values = (var_45234_cast_fp16, var_45634_cast_fp16))[name = tensor("op_45776_cast_fp16")]; + tensor var_45778_equation_0 = const()[name = tensor("op_45778_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45778_cast_fp16 = einsum(equation = var_45778_equation_0, values = (var_45234_cast_fp16, var_45635_cast_fp16))[name = tensor("op_45778_cast_fp16")]; + tensor var_45780_equation_0 = const()[name = tensor("op_45780_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45780_cast_fp16 = einsum(equation = var_45780_equation_0, values = (var_45234_cast_fp16, var_45636_cast_fp16))[name = tensor("op_45780_cast_fp16")]; + tensor var_45782_equation_0 = const()[name = tensor("op_45782_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45782_cast_fp16 = einsum(equation = var_45782_equation_0, values = (var_45238_cast_fp16, var_45637_cast_fp16))[name = tensor("op_45782_cast_fp16")]; + tensor var_45784_equation_0 = const()[name = tensor("op_45784_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45784_cast_fp16 = einsum(equation = var_45784_equation_0, values = (var_45238_cast_fp16, var_45638_cast_fp16))[name = tensor("op_45784_cast_fp16")]; + tensor var_45786_equation_0 = const()[name = tensor("op_45786_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45786_cast_fp16 = einsum(equation = var_45786_equation_0, values = (var_45238_cast_fp16, var_45639_cast_fp16))[name = tensor("op_45786_cast_fp16")]; + tensor var_45788_equation_0 = const()[name = tensor("op_45788_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45788_cast_fp16 = einsum(equation = var_45788_equation_0, values = (var_45238_cast_fp16, var_45640_cast_fp16))[name = tensor("op_45788_cast_fp16")]; + tensor var_45790_equation_0 = const()[name = tensor("op_45790_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45790_cast_fp16 = einsum(equation = var_45790_equation_0, values = (var_45242_cast_fp16, var_45641_cast_fp16))[name = tensor("op_45790_cast_fp16")]; + tensor var_45792_equation_0 = const()[name = tensor("op_45792_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45792_cast_fp16 = einsum(equation = var_45792_equation_0, values = (var_45242_cast_fp16, var_45642_cast_fp16))[name = tensor("op_45792_cast_fp16")]; + tensor var_45794_equation_0 = const()[name = tensor("op_45794_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45794_cast_fp16 = einsum(equation = var_45794_equation_0, values = (var_45242_cast_fp16, var_45643_cast_fp16))[name = tensor("op_45794_cast_fp16")]; + tensor var_45796_equation_0 = const()[name = tensor("op_45796_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45796_cast_fp16 = einsum(equation = var_45796_equation_0, values = (var_45242_cast_fp16, var_45644_cast_fp16))[name = tensor("op_45796_cast_fp16")]; + tensor var_45798_equation_0 = const()[name = tensor("op_45798_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45798_cast_fp16 = einsum(equation = var_45798_equation_0, values = (var_45246_cast_fp16, var_45645_cast_fp16))[name = tensor("op_45798_cast_fp16")]; + tensor var_45800_equation_0 = const()[name = tensor("op_45800_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45800_cast_fp16 = einsum(equation = var_45800_equation_0, values = (var_45246_cast_fp16, var_45646_cast_fp16))[name = tensor("op_45800_cast_fp16")]; + tensor var_45802_equation_0 = const()[name = tensor("op_45802_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45802_cast_fp16 = einsum(equation = var_45802_equation_0, values = (var_45246_cast_fp16, var_45647_cast_fp16))[name = tensor("op_45802_cast_fp16")]; + tensor var_45804_equation_0 = const()[name = tensor("op_45804_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45804_cast_fp16 = einsum(equation = var_45804_equation_0, values = (var_45246_cast_fp16, var_45648_cast_fp16))[name = tensor("op_45804_cast_fp16")]; + tensor var_45806_equation_0 = const()[name = tensor("op_45806_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45806_cast_fp16 = einsum(equation = var_45806_equation_0, values = (var_45250_cast_fp16, var_45649_cast_fp16))[name = tensor("op_45806_cast_fp16")]; + tensor var_45808_equation_0 = const()[name = tensor("op_45808_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45808_cast_fp16 = einsum(equation = var_45808_equation_0, values = (var_45250_cast_fp16, var_45650_cast_fp16))[name = tensor("op_45808_cast_fp16")]; + tensor var_45810_equation_0 = const()[name = tensor("op_45810_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45810_cast_fp16 = einsum(equation = var_45810_equation_0, values = (var_45250_cast_fp16, var_45651_cast_fp16))[name = tensor("op_45810_cast_fp16")]; + tensor var_45812_equation_0 = const()[name = tensor("op_45812_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_45812_cast_fp16 = einsum(equation = var_45812_equation_0, values = (var_45250_cast_fp16, var_45652_cast_fp16))[name = tensor("op_45812_cast_fp16")]; + tensor var_45814_interleave_0 = const()[name = tensor("op_45814_interleave_0"), val = tensor(false)]; + tensor var_45814_cast_fp16 = concat(axis = var_44319, interleave = var_45814_interleave_0, values = (var_45654_cast_fp16, var_45656_cast_fp16, var_45658_cast_fp16, var_45660_cast_fp16))[name = tensor("op_45814_cast_fp16")]; + tensor var_45816_interleave_0 = const()[name = tensor("op_45816_interleave_0"), val = tensor(false)]; + tensor var_45816_cast_fp16 = concat(axis = var_44319, interleave = var_45816_interleave_0, values = (var_45662_cast_fp16, var_45664_cast_fp16, var_45666_cast_fp16, var_45668_cast_fp16))[name = tensor("op_45816_cast_fp16")]; + tensor var_45818_interleave_0 = const()[name = tensor("op_45818_interleave_0"), val = tensor(false)]; + tensor var_45818_cast_fp16 = concat(axis = var_44319, interleave = var_45818_interleave_0, values = (var_45670_cast_fp16, var_45672_cast_fp16, var_45674_cast_fp16, var_45676_cast_fp16))[name = tensor("op_45818_cast_fp16")]; + tensor var_45820_interleave_0 = const()[name = tensor("op_45820_interleave_0"), val = tensor(false)]; + tensor var_45820_cast_fp16 = concat(axis = var_44319, interleave = var_45820_interleave_0, values = (var_45678_cast_fp16, var_45680_cast_fp16, var_45682_cast_fp16, var_45684_cast_fp16))[name = tensor("op_45820_cast_fp16")]; + tensor var_45822_interleave_0 = const()[name = tensor("op_45822_interleave_0"), val = tensor(false)]; + tensor var_45822_cast_fp16 = concat(axis = var_44319, interleave = var_45822_interleave_0, values = (var_45686_cast_fp16, var_45688_cast_fp16, var_45690_cast_fp16, var_45692_cast_fp16))[name = tensor("op_45822_cast_fp16")]; + tensor var_45824_interleave_0 = const()[name = tensor("op_45824_interleave_0"), val = tensor(false)]; + tensor var_45824_cast_fp16 = concat(axis = var_44319, interleave = var_45824_interleave_0, values = (var_45694_cast_fp16, var_45696_cast_fp16, var_45698_cast_fp16, var_45700_cast_fp16))[name = tensor("op_45824_cast_fp16")]; + tensor var_45826_interleave_0 = const()[name = tensor("op_45826_interleave_0"), val = tensor(false)]; + tensor var_45826_cast_fp16 = concat(axis = var_44319, interleave = var_45826_interleave_0, values = (var_45702_cast_fp16, var_45704_cast_fp16, var_45706_cast_fp16, var_45708_cast_fp16))[name = tensor("op_45826_cast_fp16")]; + tensor var_45828_interleave_0 = const()[name = tensor("op_45828_interleave_0"), val = tensor(false)]; + tensor var_45828_cast_fp16 = concat(axis = var_44319, interleave = var_45828_interleave_0, values = (var_45710_cast_fp16, var_45712_cast_fp16, var_45714_cast_fp16, var_45716_cast_fp16))[name = tensor("op_45828_cast_fp16")]; + tensor var_45830_interleave_0 = const()[name = tensor("op_45830_interleave_0"), val = tensor(false)]; + tensor var_45830_cast_fp16 = concat(axis = var_44319, interleave = var_45830_interleave_0, values = (var_45718_cast_fp16, var_45720_cast_fp16, var_45722_cast_fp16, var_45724_cast_fp16))[name = tensor("op_45830_cast_fp16")]; + tensor var_45832_interleave_0 = const()[name = tensor("op_45832_interleave_0"), val = tensor(false)]; + tensor var_45832_cast_fp16 = concat(axis = var_44319, interleave = var_45832_interleave_0, values = (var_45726_cast_fp16, var_45728_cast_fp16, var_45730_cast_fp16, var_45732_cast_fp16))[name = tensor("op_45832_cast_fp16")]; + tensor var_45834_interleave_0 = const()[name = tensor("op_45834_interleave_0"), val = tensor(false)]; + tensor var_45834_cast_fp16 = concat(axis = var_44319, interleave = var_45834_interleave_0, values = (var_45734_cast_fp16, var_45736_cast_fp16, var_45738_cast_fp16, var_45740_cast_fp16))[name = tensor("op_45834_cast_fp16")]; + tensor var_45836_interleave_0 = const()[name = tensor("op_45836_interleave_0"), val = tensor(false)]; + tensor var_45836_cast_fp16 = concat(axis = var_44319, interleave = var_45836_interleave_0, values = (var_45742_cast_fp16, var_45744_cast_fp16, var_45746_cast_fp16, var_45748_cast_fp16))[name = tensor("op_45836_cast_fp16")]; + tensor var_45838_interleave_0 = const()[name = tensor("op_45838_interleave_0"), val = tensor(false)]; + tensor var_45838_cast_fp16 = concat(axis = var_44319, interleave = var_45838_interleave_0, values = (var_45750_cast_fp16, var_45752_cast_fp16, var_45754_cast_fp16, var_45756_cast_fp16))[name = tensor("op_45838_cast_fp16")]; + tensor var_45840_interleave_0 = const()[name = tensor("op_45840_interleave_0"), val = tensor(false)]; + tensor var_45840_cast_fp16 = concat(axis = var_44319, interleave = var_45840_interleave_0, values = (var_45758_cast_fp16, var_45760_cast_fp16, var_45762_cast_fp16, var_45764_cast_fp16))[name = tensor("op_45840_cast_fp16")]; + tensor var_45842_interleave_0 = const()[name = tensor("op_45842_interleave_0"), val = tensor(false)]; + tensor var_45842_cast_fp16 = concat(axis = var_44319, interleave = var_45842_interleave_0, values = (var_45766_cast_fp16, var_45768_cast_fp16, var_45770_cast_fp16, var_45772_cast_fp16))[name = tensor("op_45842_cast_fp16")]; + tensor var_45844_interleave_0 = const()[name = tensor("op_45844_interleave_0"), val = tensor(false)]; + tensor var_45844_cast_fp16 = concat(axis = var_44319, interleave = var_45844_interleave_0, values = (var_45774_cast_fp16, var_45776_cast_fp16, var_45778_cast_fp16, var_45780_cast_fp16))[name = tensor("op_45844_cast_fp16")]; + tensor var_45846_interleave_0 = const()[name = tensor("op_45846_interleave_0"), val = tensor(false)]; + tensor var_45846_cast_fp16 = concat(axis = var_44319, interleave = var_45846_interleave_0, values = (var_45782_cast_fp16, var_45784_cast_fp16, var_45786_cast_fp16, var_45788_cast_fp16))[name = tensor("op_45846_cast_fp16")]; + tensor var_45848_interleave_0 = const()[name = tensor("op_45848_interleave_0"), val = tensor(false)]; + tensor var_45848_cast_fp16 = concat(axis = var_44319, interleave = var_45848_interleave_0, values = (var_45790_cast_fp16, var_45792_cast_fp16, var_45794_cast_fp16, var_45796_cast_fp16))[name = tensor("op_45848_cast_fp16")]; + tensor var_45850_interleave_0 = const()[name = tensor("op_45850_interleave_0"), val = tensor(false)]; + tensor var_45850_cast_fp16 = concat(axis = var_44319, interleave = var_45850_interleave_0, values = (var_45798_cast_fp16, var_45800_cast_fp16, var_45802_cast_fp16, var_45804_cast_fp16))[name = tensor("op_45850_cast_fp16")]; + tensor var_45852_interleave_0 = const()[name = tensor("op_45852_interleave_0"), val = tensor(false)]; + tensor var_45852_cast_fp16 = concat(axis = var_44319, interleave = var_45852_interleave_0, values = (var_45806_cast_fp16, var_45808_cast_fp16, var_45810_cast_fp16, var_45812_cast_fp16))[name = tensor("op_45852_cast_fp16")]; + tensor input_547_interleave_0 = const()[name = tensor("input_547_interleave_0"), val = tensor(false)]; + tensor input_547_cast_fp16 = concat(axis = var_44344, interleave = input_547_interleave_0, values = (var_45814_cast_fp16, var_45816_cast_fp16, var_45818_cast_fp16, var_45820_cast_fp16, var_45822_cast_fp16, var_45824_cast_fp16, var_45826_cast_fp16, var_45828_cast_fp16, var_45830_cast_fp16, var_45832_cast_fp16, var_45834_cast_fp16, var_45836_cast_fp16, var_45838_cast_fp16, var_45840_cast_fp16, var_45842_cast_fp16, var_45844_cast_fp16, var_45846_cast_fp16, var_45848_cast_fp16, var_45850_cast_fp16, var_45852_cast_fp16))[name = tensor("input_547_cast_fp16")]; + tensor var_45860 = const()[name = tensor("op_45860"), val = tensor([1, 1])]; + tensor var_45862 = const()[name = tensor("op_45862"), val = tensor([1, 1])]; + tensor pretrained_out_331_pad_type_0 = const()[name = tensor("pretrained_out_331_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_331_pad_0 = const()[name = tensor("pretrained_out_331_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303244864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304064128))), name = tensor("layers_27_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_27_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_27_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304064256)))]; + tensor pretrained_out_331_cast_fp16 = conv(bias = layers_27_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_45862, groups = var_44344, pad = pretrained_out_331_pad_0, pad_type = pretrained_out_331_pad_type_0, strides = var_45860, weight = layers_27_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_547_cast_fp16)[name = tensor("pretrained_out_331_cast_fp16")]; + tensor var_45866 = const()[name = tensor("op_45866"), val = tensor([1, 1])]; + tensor var_45868 = const()[name = tensor("op_45868"), val = tensor([1, 1])]; + tensor input_549_pad_type_0 = const()[name = tensor("input_549_pad_type_0"), val = tensor("custom")]; + tensor input_549_pad_0 = const()[name = tensor("input_549_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_27_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304066880)))]; + tensor input_549_cast_fp16 = conv(dilations = var_45868, groups = var_44344, pad = input_549_pad_0, pad_type = input_549_pad_type_0, strides = var_45866, weight = layers_27_self_attn_o_proj_loraA_weight_to_fp16, x = input_547_cast_fp16)[name = tensor("input_549_cast_fp16")]; + tensor var_45872 = const()[name = tensor("op_45872"), val = tensor([1, 1])]; + tensor var_45874 = const()[name = tensor("op_45874"), val = tensor([1, 1])]; + tensor lora_out_661_pad_type_0 = const()[name = tensor("lora_out_661_pad_type_0"), val = tensor("custom")]; + tensor lora_out_661_pad_0 = const()[name = tensor("lora_out_661_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_663_weight_0_to_fp16 = const()[name = tensor("lora_out_663_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304107904)))]; + tensor lora_out_663_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_45874, groups = var_44344, pad = lora_out_661_pad_0, pad_type = lora_out_661_pad_type_0, strides = var_45872, weight = lora_out_663_weight_0_to_fp16, x = input_549_cast_fp16)[name = tensor("lora_out_663_cast_fp16")]; + tensor obj_111_cast_fp16 = add(x = pretrained_out_331_cast_fp16, y = lora_out_663_cast_fp16)[name = tensor("obj_111_cast_fp16")]; + tensor inputs_111_cast_fp16 = add(x = inputs_109_cast_fp16, y = obj_111_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; + tensor var_45883 = const()[name = tensor("op_45883"), val = tensor([1])]; + tensor channels_mean_111_cast_fp16 = reduce_mean(axes = var_45883, keep_dims = var_44345, x = inputs_111_cast_fp16)[name = tensor("channels_mean_111_cast_fp16")]; + tensor zero_mean_111_cast_fp16 = sub(x = inputs_111_cast_fp16, y = channels_mean_111_cast_fp16)[name = tensor("zero_mean_111_cast_fp16")]; + tensor zero_mean_sq_111_cast_fp16 = mul(x = zero_mean_111_cast_fp16, y = zero_mean_111_cast_fp16)[name = tensor("zero_mean_sq_111_cast_fp16")]; + tensor var_45887 = const()[name = tensor("op_45887"), val = tensor([1])]; + tensor var_45888_cast_fp16 = reduce_mean(axes = var_45887, keep_dims = var_44345, x = zero_mean_sq_111_cast_fp16)[name = tensor("op_45888_cast_fp16")]; + tensor var_45889_to_fp16 = const()[name = tensor("op_45889_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_45890_cast_fp16 = add(x = var_45888_cast_fp16, y = var_45889_to_fp16)[name = tensor("op_45890_cast_fp16")]; + tensor denom_111_epsilon_0 = const()[name = tensor("denom_111_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_111_cast_fp16 = rsqrt(epsilon = denom_111_epsilon_0, x = var_45890_cast_fp16)[name = tensor("denom_111_cast_fp16")]; + tensor out_111_cast_fp16 = mul(x = zero_mean_111_cast_fp16, y = denom_111_cast_fp16)[name = tensor("out_111_cast_fp16")]; + tensor input_551_gamma_0_to_fp16 = const()[name = tensor("input_551_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304148928)))]; + tensor input_551_beta_0_to_fp16 = const()[name = tensor("input_551_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304151552)))]; + tensor input_551_epsilon_0_to_fp16 = const()[name = tensor("input_551_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_551_cast_fp16 = batch_norm(beta = input_551_beta_0_to_fp16, epsilon = input_551_epsilon_0_to_fp16, gamma = input_551_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_111_cast_fp16)[name = tensor("input_551_cast_fp16")]; + tensor var_45904 = const()[name = tensor("op_45904"), val = tensor([1, 1])]; + tensor var_45906 = const()[name = tensor("op_45906"), val = tensor([1, 1])]; + tensor pretrained_out_333_pad_type_0 = const()[name = tensor("pretrained_out_333_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_333_pad_0 = const()[name = tensor("pretrained_out_333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304154176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307431040))), name = tensor("layers_27_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_27_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_27_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307431168)))]; + tensor pretrained_out_333_cast_fp16 = conv(bias = layers_27_fc1_pretrained_bias_to_fp16, dilations = var_45906, groups = var_44344, pad = pretrained_out_333_pad_0, pad_type = pretrained_out_333_pad_type_0, strides = var_45904, weight = layers_27_fc1_pretrained_weight_to_fp16_palettized, x = input_551_cast_fp16)[name = tensor("pretrained_out_333_cast_fp16")]; + tensor var_45910 = const()[name = tensor("op_45910"), val = tensor([1, 1])]; + tensor var_45912 = const()[name = tensor("op_45912"), val = tensor([1, 1])]; + tensor input_553_pad_type_0 = const()[name = tensor("input_553_pad_type_0"), val = tensor("custom")]; + tensor input_553_pad_0 = const()[name = tensor("input_553_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_27_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307441472)))]; + tensor input_553_cast_fp16 = conv(dilations = var_45912, groups = var_44344, pad = input_553_pad_0, pad_type = input_553_pad_type_0, strides = var_45910, weight = layers_27_fc1_loraA_weight_to_fp16, x = input_551_cast_fp16)[name = tensor("input_553_cast_fp16")]; + tensor var_45916 = const()[name = tensor("op_45916"), val = tensor([1, 1])]; + tensor var_45918 = const()[name = tensor("op_45918"), val = tensor([1, 1])]; + tensor lora_out_665_pad_type_0 = const()[name = tensor("lora_out_665_pad_type_0"), val = tensor("custom")]; + tensor lora_out_665_pad_0 = const()[name = tensor("lora_out_665_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_667_weight_0_to_fp16 = const()[name = tensor("lora_out_667_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307482496)))]; + tensor lora_out_667_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_45918, groups = var_44344, pad = lora_out_665_pad_0, pad_type = lora_out_665_pad_type_0, strides = var_45916, weight = lora_out_667_weight_0_to_fp16, x = input_553_cast_fp16)[name = tensor("lora_out_667_cast_fp16")]; + tensor input_555_cast_fp16 = add(x = pretrained_out_333_cast_fp16, y = lora_out_667_cast_fp16)[name = tensor("input_555_cast_fp16")]; + tensor input_557_mode_0 = const()[name = tensor("input_557_mode_0"), val = tensor("EXACT")]; + tensor input_557_cast_fp16 = gelu(mode = input_557_mode_0, x = input_555_cast_fp16)[name = tensor("input_557_cast_fp16")]; + tensor var_45930 = const()[name = tensor("op_45930"), val = tensor([1, 1])]; + tensor var_45932 = const()[name = tensor("op_45932"), val = tensor([1, 1])]; + tensor pretrained_out_335_pad_type_0 = const()[name = tensor("pretrained_out_335_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_335_pad_0 = const()[name = tensor("pretrained_out_335_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307646400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310923264))), name = tensor("layers_27_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_27_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_27_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310923392)))]; + tensor pretrained_out_335_cast_fp16 = conv(bias = layers_27_fc2_pretrained_bias_to_fp16, dilations = var_45932, groups = var_44344, pad = pretrained_out_335_pad_0, pad_type = pretrained_out_335_pad_type_0, strides = var_45930, weight = layers_27_fc2_pretrained_weight_to_fp16_palettized, x = input_557_cast_fp16)[name = tensor("pretrained_out_335_cast_fp16")]; + tensor var_45936 = const()[name = tensor("op_45936"), val = tensor([1, 1])]; + tensor var_45938 = const()[name = tensor("op_45938"), val = tensor([1, 1])]; + tensor input_559_pad_type_0 = const()[name = tensor("input_559_pad_type_0"), val = tensor("custom")]; + tensor input_559_pad_0 = const()[name = tensor("input_559_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_27_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_27_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(310926016)))]; + tensor input_559_cast_fp16 = conv(dilations = var_45938, groups = var_44344, pad = input_559_pad_0, pad_type = input_559_pad_type_0, strides = var_45936, weight = layers_27_fc2_loraA_weight_to_fp16, x = input_557_cast_fp16)[name = tensor("input_559_cast_fp16")]; + tensor var_45942 = const()[name = tensor("op_45942"), val = tensor([1, 1])]; + tensor var_45944 = const()[name = tensor("op_45944"), val = tensor([1, 1])]; + tensor lora_out_669_pad_type_0 = const()[name = tensor("lora_out_669_pad_type_0"), val = tensor("custom")]; + tensor lora_out_669_pad_0 = const()[name = tensor("lora_out_669_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_671_weight_0_to_fp16 = const()[name = tensor("lora_out_671_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311089920)))]; + tensor lora_out_671_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_45944, groups = var_44344, pad = lora_out_669_pad_0, pad_type = lora_out_669_pad_type_0, strides = var_45942, weight = lora_out_671_weight_0_to_fp16, x = input_559_cast_fp16)[name = tensor("lora_out_671_cast_fp16")]; + tensor hidden_states_59_cast_fp16 = add(x = pretrained_out_335_cast_fp16, y = lora_out_671_cast_fp16)[name = tensor("hidden_states_59_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = hidden_states_59_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; + tensor var_45954 = const()[name = tensor("op_45954"), val = tensor(3)]; + tensor var_45979 = const()[name = tensor("op_45979"), val = tensor(1)]; + tensor var_45980 = const()[name = tensor("op_45980"), val = tensor(true)]; + tensor var_45990 = const()[name = tensor("op_45990"), val = tensor([1])]; + tensor channels_mean_113_cast_fp16 = reduce_mean(axes = var_45990, keep_dims = var_45980, x = inputs_113_cast_fp16)[name = tensor("channels_mean_113_cast_fp16")]; + tensor zero_mean_113_cast_fp16 = sub(x = inputs_113_cast_fp16, y = channels_mean_113_cast_fp16)[name = tensor("zero_mean_113_cast_fp16")]; + tensor zero_mean_sq_113_cast_fp16 = mul(x = zero_mean_113_cast_fp16, y = zero_mean_113_cast_fp16)[name = tensor("zero_mean_sq_113_cast_fp16")]; + tensor var_45994 = const()[name = tensor("op_45994"), val = tensor([1])]; + tensor var_45995_cast_fp16 = reduce_mean(axes = var_45994, keep_dims = var_45980, x = zero_mean_sq_113_cast_fp16)[name = tensor("op_45995_cast_fp16")]; + tensor var_45996_to_fp16 = const()[name = tensor("op_45996_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_45997_cast_fp16 = add(x = var_45995_cast_fp16, y = var_45996_to_fp16)[name = tensor("op_45997_cast_fp16")]; + tensor denom_113_epsilon_0 = const()[name = tensor("denom_113_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_113_cast_fp16 = rsqrt(epsilon = denom_113_epsilon_0, x = var_45997_cast_fp16)[name = tensor("denom_113_cast_fp16")]; + tensor out_113_cast_fp16 = mul(x = zero_mean_113_cast_fp16, y = denom_113_cast_fp16)[name = tensor("out_113_cast_fp16")]; + tensor obj_113_gamma_0_to_fp16 = const()[name = tensor("obj_113_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311130944)))]; + tensor obj_113_beta_0_to_fp16 = const()[name = tensor("obj_113_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311133568)))]; + tensor obj_113_epsilon_0_to_fp16 = const()[name = tensor("obj_113_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_113_cast_fp16 = batch_norm(beta = obj_113_beta_0_to_fp16, epsilon = obj_113_epsilon_0_to_fp16, gamma = obj_113_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_113_cast_fp16)[name = tensor("obj_113_cast_fp16")]; + tensor var_46015 = const()[name = tensor("op_46015"), val = tensor([1, 1])]; + tensor var_46017 = const()[name = tensor("op_46017"), val = tensor([1, 1])]; + tensor pretrained_out_337_pad_type_0 = const()[name = tensor("pretrained_out_337_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_337_pad_0 = const()[name = tensor("pretrained_out_337_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311136192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311955456))), name = tensor("layers_28_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_28_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311955584)))]; + tensor pretrained_out_337_cast_fp16 = conv(bias = layers_28_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_46017, groups = var_45979, pad = pretrained_out_337_pad_0, pad_type = pretrained_out_337_pad_type_0, strides = var_46015, weight = layers_28_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("pretrained_out_337_cast_fp16")]; + tensor var_46021 = const()[name = tensor("op_46021"), val = tensor([1, 1])]; + tensor var_46023 = const()[name = tensor("op_46023"), val = tensor([1, 1])]; + tensor input_561_pad_type_0 = const()[name = tensor("input_561_pad_type_0"), val = tensor("custom")]; + tensor input_561_pad_0 = const()[name = tensor("input_561_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311958208)))]; + tensor input_561_cast_fp16 = conv(dilations = var_46023, groups = var_45979, pad = input_561_pad_0, pad_type = input_561_pad_type_0, strides = var_46021, weight = layers_28_self_attn_q_proj_loraA_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("input_561_cast_fp16")]; + tensor var_46027 = const()[name = tensor("op_46027"), val = tensor([1, 1])]; + tensor var_46029 = const()[name = tensor("op_46029"), val = tensor([1, 1])]; + tensor lora_out_673_pad_type_0 = const()[name = tensor("lora_out_673_pad_type_0"), val = tensor("custom")]; + tensor lora_out_673_pad_0 = const()[name = tensor("lora_out_673_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_675_weight_0_to_fp16 = const()[name = tensor("lora_out_675_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311999232)))]; + tensor lora_out_675_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_46029, groups = var_45979, pad = lora_out_673_pad_0, pad_type = lora_out_673_pad_type_0, strides = var_46027, weight = lora_out_675_weight_0_to_fp16, x = input_561_cast_fp16)[name = tensor("lora_out_675_cast_fp16")]; + tensor query_57_cast_fp16 = add(x = pretrained_out_337_cast_fp16, y = lora_out_675_cast_fp16)[name = tensor("query_57_cast_fp16")]; + tensor var_46039 = const()[name = tensor("op_46039"), val = tensor([1, 1])]; + tensor var_46041 = const()[name = tensor("op_46041"), val = tensor([1, 1])]; + tensor pretrained_out_339_pad_type_0 = const()[name = tensor("pretrained_out_339_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_339_pad_0 = const()[name = tensor("pretrained_out_339_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312040256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312859520))), name = tensor("layers_28_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_339_cast_fp16 = conv(dilations = var_46041, groups = var_45979, pad = pretrained_out_339_pad_0, pad_type = pretrained_out_339_pad_type_0, strides = var_46039, weight = layers_28_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("pretrained_out_339_cast_fp16")]; + tensor var_46045 = const()[name = tensor("op_46045"), val = tensor([1, 1])]; + tensor var_46047 = const()[name = tensor("op_46047"), val = tensor([1, 1])]; + tensor input_563_pad_type_0 = const()[name = tensor("input_563_pad_type_0"), val = tensor("custom")]; + tensor input_563_pad_0 = const()[name = tensor("input_563_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312859648)))]; + tensor input_563_cast_fp16 = conv(dilations = var_46047, groups = var_45979, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = var_46045, weight = layers_28_self_attn_k_proj_loraA_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("input_563_cast_fp16")]; + tensor var_46051 = const()[name = tensor("op_46051"), val = tensor([1, 1])]; + tensor var_46053 = const()[name = tensor("op_46053"), val = tensor([1, 1])]; + tensor lora_out_677_pad_type_0 = const()[name = tensor("lora_out_677_pad_type_0"), val = tensor("custom")]; + tensor lora_out_677_pad_0 = const()[name = tensor("lora_out_677_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_679_weight_0_to_fp16 = const()[name = tensor("lora_out_679_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312900672)))]; + tensor lora_out_679_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_46053, groups = var_45979, pad = lora_out_677_pad_0, pad_type = lora_out_677_pad_type_0, strides = var_46051, weight = lora_out_679_weight_0_to_fp16, x = input_563_cast_fp16)[name = tensor("lora_out_679_cast_fp16")]; + tensor key_57_cast_fp16 = add(x = pretrained_out_339_cast_fp16, y = lora_out_679_cast_fp16)[name = tensor("key_57_cast_fp16")]; + tensor var_46064 = const()[name = tensor("op_46064"), val = tensor([1, 1])]; + tensor var_46066 = const()[name = tensor("op_46066"), val = tensor([1, 1])]; + tensor pretrained_out_341_pad_type_0 = const()[name = tensor("pretrained_out_341_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_341_pad_0 = const()[name = tensor("pretrained_out_341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312941696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313760960))), name = tensor("layers_28_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_28_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313761088)))]; + tensor pretrained_out_341_cast_fp16 = conv(bias = layers_28_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_46066, groups = var_45979, pad = pretrained_out_341_pad_0, pad_type = pretrained_out_341_pad_type_0, strides = var_46064, weight = layers_28_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_113_cast_fp16)[name = tensor("pretrained_out_341_cast_fp16")]; + tensor var_46070 = const()[name = tensor("op_46070"), val = tensor([1, 1])]; + tensor var_46072 = const()[name = tensor("op_46072"), val = tensor([1, 1])]; + tensor input_565_pad_type_0 = const()[name = tensor("input_565_pad_type_0"), val = tensor("custom")]; + tensor input_565_pad_0 = const()[name = tensor("input_565_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313763712)))]; + tensor input_565_cast_fp16 = conv(dilations = var_46072, groups = var_45979, pad = input_565_pad_0, pad_type = input_565_pad_type_0, strides = var_46070, weight = layers_28_self_attn_v_proj_loraA_weight_to_fp16, x = obj_113_cast_fp16)[name = tensor("input_565_cast_fp16")]; + tensor var_46076 = const()[name = tensor("op_46076"), val = tensor([1, 1])]; + tensor var_46078 = const()[name = tensor("op_46078"), val = tensor([1, 1])]; + tensor lora_out_681_pad_type_0 = const()[name = tensor("lora_out_681_pad_type_0"), val = tensor("custom")]; + tensor lora_out_681_pad_0 = const()[name = tensor("lora_out_681_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_683_weight_0_to_fp16 = const()[name = tensor("lora_out_683_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313804736)))]; + tensor lora_out_683_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_46078, groups = var_45979, pad = lora_out_681_pad_0, pad_type = lora_out_681_pad_type_0, strides = var_46076, weight = lora_out_683_weight_0_to_fp16, x = input_565_cast_fp16)[name = tensor("lora_out_683_cast_fp16")]; + tensor value_57_cast_fp16 = add(x = pretrained_out_341_cast_fp16, y = lora_out_683_cast_fp16)[name = tensor("value_57_cast_fp16")]; + tensor var_46088_begin_0 = const()[name = tensor("op_46088_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46088_end_0 = const()[name = tensor("op_46088_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46088_end_mask_0 = const()[name = tensor("op_46088_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46088_cast_fp16 = slice_by_index(begin = var_46088_begin_0, end = var_46088_end_0, end_mask = var_46088_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46088_cast_fp16")]; + tensor var_46092_begin_0 = const()[name = tensor("op_46092_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46092_end_0 = const()[name = tensor("op_46092_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46092_end_mask_0 = const()[name = tensor("op_46092_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46092_cast_fp16 = slice_by_index(begin = var_46092_begin_0, end = var_46092_end_0, end_mask = var_46092_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46092_cast_fp16")]; + tensor var_46096_begin_0 = const()[name = tensor("op_46096_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46096_end_0 = const()[name = tensor("op_46096_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46096_end_mask_0 = const()[name = tensor("op_46096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46096_cast_fp16 = slice_by_index(begin = var_46096_begin_0, end = var_46096_end_0, end_mask = var_46096_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46096_cast_fp16")]; + tensor var_46100_begin_0 = const()[name = tensor("op_46100_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46100_end_0 = const()[name = tensor("op_46100_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46100_end_mask_0 = const()[name = tensor("op_46100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46100_cast_fp16 = slice_by_index(begin = var_46100_begin_0, end = var_46100_end_0, end_mask = var_46100_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46100_cast_fp16")]; + tensor var_46104_begin_0 = const()[name = tensor("op_46104_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46104_end_0 = const()[name = tensor("op_46104_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46104_end_mask_0 = const()[name = tensor("op_46104_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46104_cast_fp16 = slice_by_index(begin = var_46104_begin_0, end = var_46104_end_0, end_mask = var_46104_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46104_cast_fp16")]; + tensor var_46108_begin_0 = const()[name = tensor("op_46108_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46108_end_0 = const()[name = tensor("op_46108_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46108_end_mask_0 = const()[name = tensor("op_46108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46108_cast_fp16 = slice_by_index(begin = var_46108_begin_0, end = var_46108_end_0, end_mask = var_46108_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46108_cast_fp16")]; + tensor var_46112_begin_0 = const()[name = tensor("op_46112_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46112_end_0 = const()[name = tensor("op_46112_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46112_end_mask_0 = const()[name = tensor("op_46112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46112_cast_fp16 = slice_by_index(begin = var_46112_begin_0, end = var_46112_end_0, end_mask = var_46112_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46112_cast_fp16")]; + tensor var_46116_begin_0 = const()[name = tensor("op_46116_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46116_end_0 = const()[name = tensor("op_46116_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46116_end_mask_0 = const()[name = tensor("op_46116_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46116_cast_fp16 = slice_by_index(begin = var_46116_begin_0, end = var_46116_end_0, end_mask = var_46116_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46116_cast_fp16")]; + tensor var_46120_begin_0 = const()[name = tensor("op_46120_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46120_end_0 = const()[name = tensor("op_46120_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46120_end_mask_0 = const()[name = tensor("op_46120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46120_cast_fp16 = slice_by_index(begin = var_46120_begin_0, end = var_46120_end_0, end_mask = var_46120_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46120_cast_fp16")]; + tensor var_46124_begin_0 = const()[name = tensor("op_46124_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46124_end_0 = const()[name = tensor("op_46124_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46124_end_mask_0 = const()[name = tensor("op_46124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46124_cast_fp16 = slice_by_index(begin = var_46124_begin_0, end = var_46124_end_0, end_mask = var_46124_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46124_cast_fp16")]; + tensor var_46128_begin_0 = const()[name = tensor("op_46128_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46128_end_0 = const()[name = tensor("op_46128_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46128_end_mask_0 = const()[name = tensor("op_46128_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46128_cast_fp16 = slice_by_index(begin = var_46128_begin_0, end = var_46128_end_0, end_mask = var_46128_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46128_cast_fp16")]; + tensor var_46132_begin_0 = const()[name = tensor("op_46132_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46132_end_0 = const()[name = tensor("op_46132_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46132_end_mask_0 = const()[name = tensor("op_46132_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46132_cast_fp16 = slice_by_index(begin = var_46132_begin_0, end = var_46132_end_0, end_mask = var_46132_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46132_cast_fp16")]; + tensor var_46136_begin_0 = const()[name = tensor("op_46136_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46136_end_0 = const()[name = tensor("op_46136_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46136_end_mask_0 = const()[name = tensor("op_46136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46136_cast_fp16 = slice_by_index(begin = var_46136_begin_0, end = var_46136_end_0, end_mask = var_46136_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46136_cast_fp16")]; + tensor var_46140_begin_0 = const()[name = tensor("op_46140_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46140_end_0 = const()[name = tensor("op_46140_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46140_end_mask_0 = const()[name = tensor("op_46140_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46140_cast_fp16 = slice_by_index(begin = var_46140_begin_0, end = var_46140_end_0, end_mask = var_46140_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46140_cast_fp16")]; + tensor var_46144_begin_0 = const()[name = tensor("op_46144_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46144_end_0 = const()[name = tensor("op_46144_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46144_end_mask_0 = const()[name = tensor("op_46144_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46144_cast_fp16 = slice_by_index(begin = var_46144_begin_0, end = var_46144_end_0, end_mask = var_46144_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46144_cast_fp16")]; + tensor var_46148_begin_0 = const()[name = tensor("op_46148_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46148_end_0 = const()[name = tensor("op_46148_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46148_end_mask_0 = const()[name = tensor("op_46148_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46148_cast_fp16 = slice_by_index(begin = var_46148_begin_0, end = var_46148_end_0, end_mask = var_46148_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46148_cast_fp16")]; + tensor var_46152_begin_0 = const()[name = tensor("op_46152_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46152_end_0 = const()[name = tensor("op_46152_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46152_end_mask_0 = const()[name = tensor("op_46152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46152_cast_fp16 = slice_by_index(begin = var_46152_begin_0, end = var_46152_end_0, end_mask = var_46152_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46152_cast_fp16")]; + tensor var_46156_begin_0 = const()[name = tensor("op_46156_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46156_end_0 = const()[name = tensor("op_46156_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46156_end_mask_0 = const()[name = tensor("op_46156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46156_cast_fp16 = slice_by_index(begin = var_46156_begin_0, end = var_46156_end_0, end_mask = var_46156_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46156_cast_fp16")]; + tensor var_46160_begin_0 = const()[name = tensor("op_46160_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46160_end_0 = const()[name = tensor("op_46160_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46160_end_mask_0 = const()[name = tensor("op_46160_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46160_cast_fp16 = slice_by_index(begin = var_46160_begin_0, end = var_46160_end_0, end_mask = var_46160_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46160_cast_fp16")]; + tensor var_46164_begin_0 = const()[name = tensor("op_46164_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46164_end_0 = const()[name = tensor("op_46164_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46164_end_mask_0 = const()[name = tensor("op_46164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46164_cast_fp16 = slice_by_index(begin = var_46164_begin_0, end = var_46164_end_0, end_mask = var_46164_end_mask_0, x = query_57_cast_fp16)[name = tensor("op_46164_cast_fp16")]; + tensor var_46173_begin_0 = const()[name = tensor("op_46173_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46173_end_0 = const()[name = tensor("op_46173_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46173_end_mask_0 = const()[name = tensor("op_46173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46173_cast_fp16 = slice_by_index(begin = var_46173_begin_0, end = var_46173_end_0, end_mask = var_46173_end_mask_0, x = var_46088_cast_fp16)[name = tensor("op_46173_cast_fp16")]; + tensor var_46180_begin_0 = const()[name = tensor("op_46180_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46180_end_0 = const()[name = tensor("op_46180_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46180_end_mask_0 = const()[name = tensor("op_46180_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46180_cast_fp16 = slice_by_index(begin = var_46180_begin_0, end = var_46180_end_0, end_mask = var_46180_end_mask_0, x = var_46088_cast_fp16)[name = tensor("op_46180_cast_fp16")]; + tensor var_46187_begin_0 = const()[name = tensor("op_46187_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46187_end_0 = const()[name = tensor("op_46187_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46187_end_mask_0 = const()[name = tensor("op_46187_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46187_cast_fp16 = slice_by_index(begin = var_46187_begin_0, end = var_46187_end_0, end_mask = var_46187_end_mask_0, x = var_46088_cast_fp16)[name = tensor("op_46187_cast_fp16")]; + tensor var_46194_begin_0 = const()[name = tensor("op_46194_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46194_end_0 = const()[name = tensor("op_46194_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46194_end_mask_0 = const()[name = tensor("op_46194_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46194_cast_fp16 = slice_by_index(begin = var_46194_begin_0, end = var_46194_end_0, end_mask = var_46194_end_mask_0, x = var_46088_cast_fp16)[name = tensor("op_46194_cast_fp16")]; + tensor var_46201_begin_0 = const()[name = tensor("op_46201_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46201_end_0 = const()[name = tensor("op_46201_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46201_end_mask_0 = const()[name = tensor("op_46201_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46201_cast_fp16 = slice_by_index(begin = var_46201_begin_0, end = var_46201_end_0, end_mask = var_46201_end_mask_0, x = var_46092_cast_fp16)[name = tensor("op_46201_cast_fp16")]; + tensor var_46208_begin_0 = const()[name = tensor("op_46208_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46208_end_0 = const()[name = tensor("op_46208_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46208_end_mask_0 = const()[name = tensor("op_46208_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46208_cast_fp16 = slice_by_index(begin = var_46208_begin_0, end = var_46208_end_0, end_mask = var_46208_end_mask_0, x = var_46092_cast_fp16)[name = tensor("op_46208_cast_fp16")]; + tensor var_46215_begin_0 = const()[name = tensor("op_46215_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46215_end_0 = const()[name = tensor("op_46215_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46215_end_mask_0 = const()[name = tensor("op_46215_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46215_cast_fp16 = slice_by_index(begin = var_46215_begin_0, end = var_46215_end_0, end_mask = var_46215_end_mask_0, x = var_46092_cast_fp16)[name = tensor("op_46215_cast_fp16")]; + tensor var_46222_begin_0 = const()[name = tensor("op_46222_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46222_end_0 = const()[name = tensor("op_46222_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46222_end_mask_0 = const()[name = tensor("op_46222_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46222_cast_fp16 = slice_by_index(begin = var_46222_begin_0, end = var_46222_end_0, end_mask = var_46222_end_mask_0, x = var_46092_cast_fp16)[name = tensor("op_46222_cast_fp16")]; + tensor var_46229_begin_0 = const()[name = tensor("op_46229_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46229_end_0 = const()[name = tensor("op_46229_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46229_end_mask_0 = const()[name = tensor("op_46229_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46229_cast_fp16 = slice_by_index(begin = var_46229_begin_0, end = var_46229_end_0, end_mask = var_46229_end_mask_0, x = var_46096_cast_fp16)[name = tensor("op_46229_cast_fp16")]; + tensor var_46236_begin_0 = const()[name = tensor("op_46236_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46236_end_0 = const()[name = tensor("op_46236_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46236_end_mask_0 = const()[name = tensor("op_46236_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46236_cast_fp16 = slice_by_index(begin = var_46236_begin_0, end = var_46236_end_0, end_mask = var_46236_end_mask_0, x = var_46096_cast_fp16)[name = tensor("op_46236_cast_fp16")]; + tensor var_46243_begin_0 = const()[name = tensor("op_46243_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46243_end_0 = const()[name = tensor("op_46243_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46243_end_mask_0 = const()[name = tensor("op_46243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46243_cast_fp16 = slice_by_index(begin = var_46243_begin_0, end = var_46243_end_0, end_mask = var_46243_end_mask_0, x = var_46096_cast_fp16)[name = tensor("op_46243_cast_fp16")]; + tensor var_46250_begin_0 = const()[name = tensor("op_46250_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46250_end_0 = const()[name = tensor("op_46250_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46250_end_mask_0 = const()[name = tensor("op_46250_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46250_cast_fp16 = slice_by_index(begin = var_46250_begin_0, end = var_46250_end_0, end_mask = var_46250_end_mask_0, x = var_46096_cast_fp16)[name = tensor("op_46250_cast_fp16")]; + tensor var_46257_begin_0 = const()[name = tensor("op_46257_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46257_end_0 = const()[name = tensor("op_46257_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46257_end_mask_0 = const()[name = tensor("op_46257_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46257_cast_fp16 = slice_by_index(begin = var_46257_begin_0, end = var_46257_end_0, end_mask = var_46257_end_mask_0, x = var_46100_cast_fp16)[name = tensor("op_46257_cast_fp16")]; + tensor var_46264_begin_0 = const()[name = tensor("op_46264_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46264_end_0 = const()[name = tensor("op_46264_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46264_end_mask_0 = const()[name = tensor("op_46264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46264_cast_fp16 = slice_by_index(begin = var_46264_begin_0, end = var_46264_end_0, end_mask = var_46264_end_mask_0, x = var_46100_cast_fp16)[name = tensor("op_46264_cast_fp16")]; + tensor var_46271_begin_0 = const()[name = tensor("op_46271_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46271_end_0 = const()[name = tensor("op_46271_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46271_end_mask_0 = const()[name = tensor("op_46271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46271_cast_fp16 = slice_by_index(begin = var_46271_begin_0, end = var_46271_end_0, end_mask = var_46271_end_mask_0, x = var_46100_cast_fp16)[name = tensor("op_46271_cast_fp16")]; + tensor var_46278_begin_0 = const()[name = tensor("op_46278_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46278_end_0 = const()[name = tensor("op_46278_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46278_end_mask_0 = const()[name = tensor("op_46278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46278_cast_fp16 = slice_by_index(begin = var_46278_begin_0, end = var_46278_end_0, end_mask = var_46278_end_mask_0, x = var_46100_cast_fp16)[name = tensor("op_46278_cast_fp16")]; + tensor var_46285_begin_0 = const()[name = tensor("op_46285_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46285_end_0 = const()[name = tensor("op_46285_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46285_end_mask_0 = const()[name = tensor("op_46285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46285_cast_fp16 = slice_by_index(begin = var_46285_begin_0, end = var_46285_end_0, end_mask = var_46285_end_mask_0, x = var_46104_cast_fp16)[name = tensor("op_46285_cast_fp16")]; + tensor var_46292_begin_0 = const()[name = tensor("op_46292_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46292_end_0 = const()[name = tensor("op_46292_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46292_end_mask_0 = const()[name = tensor("op_46292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46292_cast_fp16 = slice_by_index(begin = var_46292_begin_0, end = var_46292_end_0, end_mask = var_46292_end_mask_0, x = var_46104_cast_fp16)[name = tensor("op_46292_cast_fp16")]; + tensor var_46299_begin_0 = const()[name = tensor("op_46299_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46299_end_0 = const()[name = tensor("op_46299_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46299_end_mask_0 = const()[name = tensor("op_46299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46299_cast_fp16 = slice_by_index(begin = var_46299_begin_0, end = var_46299_end_0, end_mask = var_46299_end_mask_0, x = var_46104_cast_fp16)[name = tensor("op_46299_cast_fp16")]; + tensor var_46306_begin_0 = const()[name = tensor("op_46306_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46306_end_0 = const()[name = tensor("op_46306_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46306_end_mask_0 = const()[name = tensor("op_46306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46306_cast_fp16 = slice_by_index(begin = var_46306_begin_0, end = var_46306_end_0, end_mask = var_46306_end_mask_0, x = var_46104_cast_fp16)[name = tensor("op_46306_cast_fp16")]; + tensor var_46313_begin_0 = const()[name = tensor("op_46313_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46313_end_0 = const()[name = tensor("op_46313_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46313_end_mask_0 = const()[name = tensor("op_46313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46313_cast_fp16 = slice_by_index(begin = var_46313_begin_0, end = var_46313_end_0, end_mask = var_46313_end_mask_0, x = var_46108_cast_fp16)[name = tensor("op_46313_cast_fp16")]; + tensor var_46320_begin_0 = const()[name = tensor("op_46320_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46320_end_0 = const()[name = tensor("op_46320_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46320_end_mask_0 = const()[name = tensor("op_46320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46320_cast_fp16 = slice_by_index(begin = var_46320_begin_0, end = var_46320_end_0, end_mask = var_46320_end_mask_0, x = var_46108_cast_fp16)[name = tensor("op_46320_cast_fp16")]; + tensor var_46327_begin_0 = const()[name = tensor("op_46327_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46327_end_0 = const()[name = tensor("op_46327_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46327_end_mask_0 = const()[name = tensor("op_46327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46327_cast_fp16 = slice_by_index(begin = var_46327_begin_0, end = var_46327_end_0, end_mask = var_46327_end_mask_0, x = var_46108_cast_fp16)[name = tensor("op_46327_cast_fp16")]; + tensor var_46334_begin_0 = const()[name = tensor("op_46334_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46334_end_0 = const()[name = tensor("op_46334_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46334_end_mask_0 = const()[name = tensor("op_46334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46334_cast_fp16 = slice_by_index(begin = var_46334_begin_0, end = var_46334_end_0, end_mask = var_46334_end_mask_0, x = var_46108_cast_fp16)[name = tensor("op_46334_cast_fp16")]; + tensor var_46341_begin_0 = const()[name = tensor("op_46341_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46341_end_0 = const()[name = tensor("op_46341_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46341_end_mask_0 = const()[name = tensor("op_46341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46341_cast_fp16 = slice_by_index(begin = var_46341_begin_0, end = var_46341_end_0, end_mask = var_46341_end_mask_0, x = var_46112_cast_fp16)[name = tensor("op_46341_cast_fp16")]; + tensor var_46348_begin_0 = const()[name = tensor("op_46348_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46348_end_0 = const()[name = tensor("op_46348_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46348_end_mask_0 = const()[name = tensor("op_46348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46348_cast_fp16 = slice_by_index(begin = var_46348_begin_0, end = var_46348_end_0, end_mask = var_46348_end_mask_0, x = var_46112_cast_fp16)[name = tensor("op_46348_cast_fp16")]; + tensor var_46355_begin_0 = const()[name = tensor("op_46355_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46355_end_0 = const()[name = tensor("op_46355_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46355_end_mask_0 = const()[name = tensor("op_46355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46355_cast_fp16 = slice_by_index(begin = var_46355_begin_0, end = var_46355_end_0, end_mask = var_46355_end_mask_0, x = var_46112_cast_fp16)[name = tensor("op_46355_cast_fp16")]; + tensor var_46362_begin_0 = const()[name = tensor("op_46362_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46362_end_0 = const()[name = tensor("op_46362_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46362_end_mask_0 = const()[name = tensor("op_46362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46362_cast_fp16 = slice_by_index(begin = var_46362_begin_0, end = var_46362_end_0, end_mask = var_46362_end_mask_0, x = var_46112_cast_fp16)[name = tensor("op_46362_cast_fp16")]; + tensor var_46369_begin_0 = const()[name = tensor("op_46369_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46369_end_0 = const()[name = tensor("op_46369_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46369_end_mask_0 = const()[name = tensor("op_46369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46369_cast_fp16 = slice_by_index(begin = var_46369_begin_0, end = var_46369_end_0, end_mask = var_46369_end_mask_0, x = var_46116_cast_fp16)[name = tensor("op_46369_cast_fp16")]; + tensor var_46376_begin_0 = const()[name = tensor("op_46376_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46376_end_0 = const()[name = tensor("op_46376_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46376_end_mask_0 = const()[name = tensor("op_46376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46376_cast_fp16 = slice_by_index(begin = var_46376_begin_0, end = var_46376_end_0, end_mask = var_46376_end_mask_0, x = var_46116_cast_fp16)[name = tensor("op_46376_cast_fp16")]; + tensor var_46383_begin_0 = const()[name = tensor("op_46383_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46383_end_0 = const()[name = tensor("op_46383_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46383_end_mask_0 = const()[name = tensor("op_46383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46383_cast_fp16 = slice_by_index(begin = var_46383_begin_0, end = var_46383_end_0, end_mask = var_46383_end_mask_0, x = var_46116_cast_fp16)[name = tensor("op_46383_cast_fp16")]; + tensor var_46390_begin_0 = const()[name = tensor("op_46390_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46390_end_0 = const()[name = tensor("op_46390_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46390_end_mask_0 = const()[name = tensor("op_46390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46390_cast_fp16 = slice_by_index(begin = var_46390_begin_0, end = var_46390_end_0, end_mask = var_46390_end_mask_0, x = var_46116_cast_fp16)[name = tensor("op_46390_cast_fp16")]; + tensor var_46397_begin_0 = const()[name = tensor("op_46397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46397_end_0 = const()[name = tensor("op_46397_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46397_end_mask_0 = const()[name = tensor("op_46397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46397_cast_fp16 = slice_by_index(begin = var_46397_begin_0, end = var_46397_end_0, end_mask = var_46397_end_mask_0, x = var_46120_cast_fp16)[name = tensor("op_46397_cast_fp16")]; + tensor var_46404_begin_0 = const()[name = tensor("op_46404_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46404_end_0 = const()[name = tensor("op_46404_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46404_end_mask_0 = const()[name = tensor("op_46404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46404_cast_fp16 = slice_by_index(begin = var_46404_begin_0, end = var_46404_end_0, end_mask = var_46404_end_mask_0, x = var_46120_cast_fp16)[name = tensor("op_46404_cast_fp16")]; + tensor var_46411_begin_0 = const()[name = tensor("op_46411_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46411_end_0 = const()[name = tensor("op_46411_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46411_end_mask_0 = const()[name = tensor("op_46411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46411_cast_fp16 = slice_by_index(begin = var_46411_begin_0, end = var_46411_end_0, end_mask = var_46411_end_mask_0, x = var_46120_cast_fp16)[name = tensor("op_46411_cast_fp16")]; + tensor var_46418_begin_0 = const()[name = tensor("op_46418_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46418_end_0 = const()[name = tensor("op_46418_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46418_end_mask_0 = const()[name = tensor("op_46418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46418_cast_fp16 = slice_by_index(begin = var_46418_begin_0, end = var_46418_end_0, end_mask = var_46418_end_mask_0, x = var_46120_cast_fp16)[name = tensor("op_46418_cast_fp16")]; + tensor var_46425_begin_0 = const()[name = tensor("op_46425_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46425_end_0 = const()[name = tensor("op_46425_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46425_end_mask_0 = const()[name = tensor("op_46425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46425_cast_fp16 = slice_by_index(begin = var_46425_begin_0, end = var_46425_end_0, end_mask = var_46425_end_mask_0, x = var_46124_cast_fp16)[name = tensor("op_46425_cast_fp16")]; + tensor var_46432_begin_0 = const()[name = tensor("op_46432_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46432_end_0 = const()[name = tensor("op_46432_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46432_end_mask_0 = const()[name = tensor("op_46432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46432_cast_fp16 = slice_by_index(begin = var_46432_begin_0, end = var_46432_end_0, end_mask = var_46432_end_mask_0, x = var_46124_cast_fp16)[name = tensor("op_46432_cast_fp16")]; + tensor var_46439_begin_0 = const()[name = tensor("op_46439_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46439_end_0 = const()[name = tensor("op_46439_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46439_end_mask_0 = const()[name = tensor("op_46439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46439_cast_fp16 = slice_by_index(begin = var_46439_begin_0, end = var_46439_end_0, end_mask = var_46439_end_mask_0, x = var_46124_cast_fp16)[name = tensor("op_46439_cast_fp16")]; + tensor var_46446_begin_0 = const()[name = tensor("op_46446_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46446_end_0 = const()[name = tensor("op_46446_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46446_end_mask_0 = const()[name = tensor("op_46446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46446_cast_fp16 = slice_by_index(begin = var_46446_begin_0, end = var_46446_end_0, end_mask = var_46446_end_mask_0, x = var_46124_cast_fp16)[name = tensor("op_46446_cast_fp16")]; + tensor var_46453_begin_0 = const()[name = tensor("op_46453_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46453_end_0 = const()[name = tensor("op_46453_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46453_end_mask_0 = const()[name = tensor("op_46453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46453_cast_fp16 = slice_by_index(begin = var_46453_begin_0, end = var_46453_end_0, end_mask = var_46453_end_mask_0, x = var_46128_cast_fp16)[name = tensor("op_46453_cast_fp16")]; + tensor var_46460_begin_0 = const()[name = tensor("op_46460_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46460_end_0 = const()[name = tensor("op_46460_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46460_end_mask_0 = const()[name = tensor("op_46460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46460_cast_fp16 = slice_by_index(begin = var_46460_begin_0, end = var_46460_end_0, end_mask = var_46460_end_mask_0, x = var_46128_cast_fp16)[name = tensor("op_46460_cast_fp16")]; + tensor var_46467_begin_0 = const()[name = tensor("op_46467_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46467_end_0 = const()[name = tensor("op_46467_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46467_end_mask_0 = const()[name = tensor("op_46467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46467_cast_fp16 = slice_by_index(begin = var_46467_begin_0, end = var_46467_end_0, end_mask = var_46467_end_mask_0, x = var_46128_cast_fp16)[name = tensor("op_46467_cast_fp16")]; + tensor var_46474_begin_0 = const()[name = tensor("op_46474_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46474_end_0 = const()[name = tensor("op_46474_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46474_end_mask_0 = const()[name = tensor("op_46474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46474_cast_fp16 = slice_by_index(begin = var_46474_begin_0, end = var_46474_end_0, end_mask = var_46474_end_mask_0, x = var_46128_cast_fp16)[name = tensor("op_46474_cast_fp16")]; + tensor var_46481_begin_0 = const()[name = tensor("op_46481_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46481_end_0 = const()[name = tensor("op_46481_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46481_end_mask_0 = const()[name = tensor("op_46481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46481_cast_fp16 = slice_by_index(begin = var_46481_begin_0, end = var_46481_end_0, end_mask = var_46481_end_mask_0, x = var_46132_cast_fp16)[name = tensor("op_46481_cast_fp16")]; + tensor var_46488_begin_0 = const()[name = tensor("op_46488_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46488_end_0 = const()[name = tensor("op_46488_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46488_end_mask_0 = const()[name = tensor("op_46488_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46488_cast_fp16 = slice_by_index(begin = var_46488_begin_0, end = var_46488_end_0, end_mask = var_46488_end_mask_0, x = var_46132_cast_fp16)[name = tensor("op_46488_cast_fp16")]; + tensor var_46495_begin_0 = const()[name = tensor("op_46495_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46495_end_0 = const()[name = tensor("op_46495_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46495_end_mask_0 = const()[name = tensor("op_46495_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46495_cast_fp16 = slice_by_index(begin = var_46495_begin_0, end = var_46495_end_0, end_mask = var_46495_end_mask_0, x = var_46132_cast_fp16)[name = tensor("op_46495_cast_fp16")]; + tensor var_46502_begin_0 = const()[name = tensor("op_46502_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46502_end_0 = const()[name = tensor("op_46502_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46502_end_mask_0 = const()[name = tensor("op_46502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46502_cast_fp16 = slice_by_index(begin = var_46502_begin_0, end = var_46502_end_0, end_mask = var_46502_end_mask_0, x = var_46132_cast_fp16)[name = tensor("op_46502_cast_fp16")]; + tensor var_46509_begin_0 = const()[name = tensor("op_46509_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46509_end_0 = const()[name = tensor("op_46509_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46509_end_mask_0 = const()[name = tensor("op_46509_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46509_cast_fp16 = slice_by_index(begin = var_46509_begin_0, end = var_46509_end_0, end_mask = var_46509_end_mask_0, x = var_46136_cast_fp16)[name = tensor("op_46509_cast_fp16")]; + tensor var_46516_begin_0 = const()[name = tensor("op_46516_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46516_end_0 = const()[name = tensor("op_46516_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46516_end_mask_0 = const()[name = tensor("op_46516_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46516_cast_fp16 = slice_by_index(begin = var_46516_begin_0, end = var_46516_end_0, end_mask = var_46516_end_mask_0, x = var_46136_cast_fp16)[name = tensor("op_46516_cast_fp16")]; + tensor var_46523_begin_0 = const()[name = tensor("op_46523_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46523_end_0 = const()[name = tensor("op_46523_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46523_end_mask_0 = const()[name = tensor("op_46523_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46523_cast_fp16 = slice_by_index(begin = var_46523_begin_0, end = var_46523_end_0, end_mask = var_46523_end_mask_0, x = var_46136_cast_fp16)[name = tensor("op_46523_cast_fp16")]; + tensor var_46530_begin_0 = const()[name = tensor("op_46530_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46530_end_0 = const()[name = tensor("op_46530_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46530_end_mask_0 = const()[name = tensor("op_46530_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46530_cast_fp16 = slice_by_index(begin = var_46530_begin_0, end = var_46530_end_0, end_mask = var_46530_end_mask_0, x = var_46136_cast_fp16)[name = tensor("op_46530_cast_fp16")]; + tensor var_46537_begin_0 = const()[name = tensor("op_46537_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46537_end_0 = const()[name = tensor("op_46537_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46537_end_mask_0 = const()[name = tensor("op_46537_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46537_cast_fp16 = slice_by_index(begin = var_46537_begin_0, end = var_46537_end_0, end_mask = var_46537_end_mask_0, x = var_46140_cast_fp16)[name = tensor("op_46537_cast_fp16")]; + tensor var_46544_begin_0 = const()[name = tensor("op_46544_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46544_end_0 = const()[name = tensor("op_46544_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46544_end_mask_0 = const()[name = tensor("op_46544_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46544_cast_fp16 = slice_by_index(begin = var_46544_begin_0, end = var_46544_end_0, end_mask = var_46544_end_mask_0, x = var_46140_cast_fp16)[name = tensor("op_46544_cast_fp16")]; + tensor var_46551_begin_0 = const()[name = tensor("op_46551_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46551_end_0 = const()[name = tensor("op_46551_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46551_end_mask_0 = const()[name = tensor("op_46551_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46551_cast_fp16 = slice_by_index(begin = var_46551_begin_0, end = var_46551_end_0, end_mask = var_46551_end_mask_0, x = var_46140_cast_fp16)[name = tensor("op_46551_cast_fp16")]; + tensor var_46558_begin_0 = const()[name = tensor("op_46558_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46558_end_0 = const()[name = tensor("op_46558_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46558_end_mask_0 = const()[name = tensor("op_46558_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46558_cast_fp16 = slice_by_index(begin = var_46558_begin_0, end = var_46558_end_0, end_mask = var_46558_end_mask_0, x = var_46140_cast_fp16)[name = tensor("op_46558_cast_fp16")]; + tensor var_46565_begin_0 = const()[name = tensor("op_46565_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46565_end_0 = const()[name = tensor("op_46565_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46565_end_mask_0 = const()[name = tensor("op_46565_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46565_cast_fp16 = slice_by_index(begin = var_46565_begin_0, end = var_46565_end_0, end_mask = var_46565_end_mask_0, x = var_46144_cast_fp16)[name = tensor("op_46565_cast_fp16")]; + tensor var_46572_begin_0 = const()[name = tensor("op_46572_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46572_end_0 = const()[name = tensor("op_46572_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46572_end_mask_0 = const()[name = tensor("op_46572_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46572_cast_fp16 = slice_by_index(begin = var_46572_begin_0, end = var_46572_end_0, end_mask = var_46572_end_mask_0, x = var_46144_cast_fp16)[name = tensor("op_46572_cast_fp16")]; + tensor var_46579_begin_0 = const()[name = tensor("op_46579_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46579_end_0 = const()[name = tensor("op_46579_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46579_end_mask_0 = const()[name = tensor("op_46579_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46579_cast_fp16 = slice_by_index(begin = var_46579_begin_0, end = var_46579_end_0, end_mask = var_46579_end_mask_0, x = var_46144_cast_fp16)[name = tensor("op_46579_cast_fp16")]; + tensor var_46586_begin_0 = const()[name = tensor("op_46586_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46586_end_0 = const()[name = tensor("op_46586_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46586_end_mask_0 = const()[name = tensor("op_46586_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46586_cast_fp16 = slice_by_index(begin = var_46586_begin_0, end = var_46586_end_0, end_mask = var_46586_end_mask_0, x = var_46144_cast_fp16)[name = tensor("op_46586_cast_fp16")]; + tensor var_46593_begin_0 = const()[name = tensor("op_46593_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46593_end_0 = const()[name = tensor("op_46593_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46593_end_mask_0 = const()[name = tensor("op_46593_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46593_cast_fp16 = slice_by_index(begin = var_46593_begin_0, end = var_46593_end_0, end_mask = var_46593_end_mask_0, x = var_46148_cast_fp16)[name = tensor("op_46593_cast_fp16")]; + tensor var_46600_begin_0 = const()[name = tensor("op_46600_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46600_end_0 = const()[name = tensor("op_46600_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46600_end_mask_0 = const()[name = tensor("op_46600_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46600_cast_fp16 = slice_by_index(begin = var_46600_begin_0, end = var_46600_end_0, end_mask = var_46600_end_mask_0, x = var_46148_cast_fp16)[name = tensor("op_46600_cast_fp16")]; + tensor var_46607_begin_0 = const()[name = tensor("op_46607_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46607_end_0 = const()[name = tensor("op_46607_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46607_end_mask_0 = const()[name = tensor("op_46607_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46607_cast_fp16 = slice_by_index(begin = var_46607_begin_0, end = var_46607_end_0, end_mask = var_46607_end_mask_0, x = var_46148_cast_fp16)[name = tensor("op_46607_cast_fp16")]; + tensor var_46614_begin_0 = const()[name = tensor("op_46614_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46614_end_0 = const()[name = tensor("op_46614_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46614_end_mask_0 = const()[name = tensor("op_46614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46614_cast_fp16 = slice_by_index(begin = var_46614_begin_0, end = var_46614_end_0, end_mask = var_46614_end_mask_0, x = var_46148_cast_fp16)[name = tensor("op_46614_cast_fp16")]; + tensor var_46621_begin_0 = const()[name = tensor("op_46621_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46621_end_0 = const()[name = tensor("op_46621_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46621_end_mask_0 = const()[name = tensor("op_46621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46621_cast_fp16 = slice_by_index(begin = var_46621_begin_0, end = var_46621_end_0, end_mask = var_46621_end_mask_0, x = var_46152_cast_fp16)[name = tensor("op_46621_cast_fp16")]; + tensor var_46628_begin_0 = const()[name = tensor("op_46628_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46628_end_0 = const()[name = tensor("op_46628_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46628_end_mask_0 = const()[name = tensor("op_46628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46628_cast_fp16 = slice_by_index(begin = var_46628_begin_0, end = var_46628_end_0, end_mask = var_46628_end_mask_0, x = var_46152_cast_fp16)[name = tensor("op_46628_cast_fp16")]; + tensor var_46635_begin_0 = const()[name = tensor("op_46635_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46635_end_0 = const()[name = tensor("op_46635_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46635_end_mask_0 = const()[name = tensor("op_46635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46635_cast_fp16 = slice_by_index(begin = var_46635_begin_0, end = var_46635_end_0, end_mask = var_46635_end_mask_0, x = var_46152_cast_fp16)[name = tensor("op_46635_cast_fp16")]; + tensor var_46642_begin_0 = const()[name = tensor("op_46642_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46642_end_0 = const()[name = tensor("op_46642_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46642_end_mask_0 = const()[name = tensor("op_46642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46642_cast_fp16 = slice_by_index(begin = var_46642_begin_0, end = var_46642_end_0, end_mask = var_46642_end_mask_0, x = var_46152_cast_fp16)[name = tensor("op_46642_cast_fp16")]; + tensor var_46649_begin_0 = const()[name = tensor("op_46649_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46649_end_0 = const()[name = tensor("op_46649_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46649_end_mask_0 = const()[name = tensor("op_46649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46649_cast_fp16 = slice_by_index(begin = var_46649_begin_0, end = var_46649_end_0, end_mask = var_46649_end_mask_0, x = var_46156_cast_fp16)[name = tensor("op_46649_cast_fp16")]; + tensor var_46656_begin_0 = const()[name = tensor("op_46656_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46656_end_0 = const()[name = tensor("op_46656_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46656_end_mask_0 = const()[name = tensor("op_46656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46656_cast_fp16 = slice_by_index(begin = var_46656_begin_0, end = var_46656_end_0, end_mask = var_46656_end_mask_0, x = var_46156_cast_fp16)[name = tensor("op_46656_cast_fp16")]; + tensor var_46663_begin_0 = const()[name = tensor("op_46663_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46663_end_0 = const()[name = tensor("op_46663_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46663_end_mask_0 = const()[name = tensor("op_46663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46663_cast_fp16 = slice_by_index(begin = var_46663_begin_0, end = var_46663_end_0, end_mask = var_46663_end_mask_0, x = var_46156_cast_fp16)[name = tensor("op_46663_cast_fp16")]; + tensor var_46670_begin_0 = const()[name = tensor("op_46670_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46670_end_0 = const()[name = tensor("op_46670_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46670_end_mask_0 = const()[name = tensor("op_46670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46670_cast_fp16 = slice_by_index(begin = var_46670_begin_0, end = var_46670_end_0, end_mask = var_46670_end_mask_0, x = var_46156_cast_fp16)[name = tensor("op_46670_cast_fp16")]; + tensor var_46677_begin_0 = const()[name = tensor("op_46677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46677_end_0 = const()[name = tensor("op_46677_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46677_end_mask_0 = const()[name = tensor("op_46677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46677_cast_fp16 = slice_by_index(begin = var_46677_begin_0, end = var_46677_end_0, end_mask = var_46677_end_mask_0, x = var_46160_cast_fp16)[name = tensor("op_46677_cast_fp16")]; + tensor var_46684_begin_0 = const()[name = tensor("op_46684_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46684_end_0 = const()[name = tensor("op_46684_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46684_end_mask_0 = const()[name = tensor("op_46684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46684_cast_fp16 = slice_by_index(begin = var_46684_begin_0, end = var_46684_end_0, end_mask = var_46684_end_mask_0, x = var_46160_cast_fp16)[name = tensor("op_46684_cast_fp16")]; + tensor var_46691_begin_0 = const()[name = tensor("op_46691_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46691_end_0 = const()[name = tensor("op_46691_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46691_end_mask_0 = const()[name = tensor("op_46691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46691_cast_fp16 = slice_by_index(begin = var_46691_begin_0, end = var_46691_end_0, end_mask = var_46691_end_mask_0, x = var_46160_cast_fp16)[name = tensor("op_46691_cast_fp16")]; + tensor var_46698_begin_0 = const()[name = tensor("op_46698_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46698_end_0 = const()[name = tensor("op_46698_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46698_end_mask_0 = const()[name = tensor("op_46698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46698_cast_fp16 = slice_by_index(begin = var_46698_begin_0, end = var_46698_end_0, end_mask = var_46698_end_mask_0, x = var_46160_cast_fp16)[name = tensor("op_46698_cast_fp16")]; + tensor var_46705_begin_0 = const()[name = tensor("op_46705_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46705_end_0 = const()[name = tensor("op_46705_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_46705_end_mask_0 = const()[name = tensor("op_46705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46705_cast_fp16 = slice_by_index(begin = var_46705_begin_0, end = var_46705_end_0, end_mask = var_46705_end_mask_0, x = var_46164_cast_fp16)[name = tensor("op_46705_cast_fp16")]; + tensor var_46712_begin_0 = const()[name = tensor("op_46712_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_46712_end_0 = const()[name = tensor("op_46712_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_46712_end_mask_0 = const()[name = tensor("op_46712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46712_cast_fp16 = slice_by_index(begin = var_46712_begin_0, end = var_46712_end_0, end_mask = var_46712_end_mask_0, x = var_46164_cast_fp16)[name = tensor("op_46712_cast_fp16")]; + tensor var_46719_begin_0 = const()[name = tensor("op_46719_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_46719_end_0 = const()[name = tensor("op_46719_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_46719_end_mask_0 = const()[name = tensor("op_46719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46719_cast_fp16 = slice_by_index(begin = var_46719_begin_0, end = var_46719_end_0, end_mask = var_46719_end_mask_0, x = var_46164_cast_fp16)[name = tensor("op_46719_cast_fp16")]; + tensor var_46726_begin_0 = const()[name = tensor("op_46726_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_46726_end_0 = const()[name = tensor("op_46726_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46726_end_mask_0 = const()[name = tensor("op_46726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46726_cast_fp16 = slice_by_index(begin = var_46726_begin_0, end = var_46726_end_0, end_mask = var_46726_end_mask_0, x = var_46164_cast_fp16)[name = tensor("op_46726_cast_fp16")]; + tensor k_57_perm_0 = const()[name = tensor("k_57_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_46731_begin_0 = const()[name = tensor("op_46731_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46731_end_0 = const()[name = tensor("op_46731_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_46731_end_mask_0 = const()[name = tensor("op_46731_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_3 = transpose(perm = k_57_perm_0, x = key_57_cast_fp16)[name = tensor("transpose_3")]; + tensor var_46731_cast_fp16 = slice_by_index(begin = var_46731_begin_0, end = var_46731_end_0, end_mask = var_46731_end_mask_0, x = transpose_3)[name = tensor("op_46731_cast_fp16")]; + tensor var_46735_begin_0 = const()[name = tensor("op_46735_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_46735_end_0 = const()[name = tensor("op_46735_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_46735_end_mask_0 = const()[name = tensor("op_46735_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46735_cast_fp16 = slice_by_index(begin = var_46735_begin_0, end = var_46735_end_0, end_mask = var_46735_end_mask_0, x = transpose_3)[name = tensor("op_46735_cast_fp16")]; + tensor var_46739_begin_0 = const()[name = tensor("op_46739_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_46739_end_0 = const()[name = tensor("op_46739_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_46739_end_mask_0 = const()[name = tensor("op_46739_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46739_cast_fp16 = slice_by_index(begin = var_46739_begin_0, end = var_46739_end_0, end_mask = var_46739_end_mask_0, x = transpose_3)[name = tensor("op_46739_cast_fp16")]; + tensor var_46743_begin_0 = const()[name = tensor("op_46743_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_46743_end_0 = const()[name = tensor("op_46743_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_46743_end_mask_0 = const()[name = tensor("op_46743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46743_cast_fp16 = slice_by_index(begin = var_46743_begin_0, end = var_46743_end_0, end_mask = var_46743_end_mask_0, x = transpose_3)[name = tensor("op_46743_cast_fp16")]; + tensor var_46747_begin_0 = const()[name = tensor("op_46747_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_46747_end_0 = const()[name = tensor("op_46747_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_46747_end_mask_0 = const()[name = tensor("op_46747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46747_cast_fp16 = slice_by_index(begin = var_46747_begin_0, end = var_46747_end_0, end_mask = var_46747_end_mask_0, x = transpose_3)[name = tensor("op_46747_cast_fp16")]; + tensor var_46751_begin_0 = const()[name = tensor("op_46751_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_46751_end_0 = const()[name = tensor("op_46751_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_46751_end_mask_0 = const()[name = tensor("op_46751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46751_cast_fp16 = slice_by_index(begin = var_46751_begin_0, end = var_46751_end_0, end_mask = var_46751_end_mask_0, x = transpose_3)[name = tensor("op_46751_cast_fp16")]; + tensor var_46755_begin_0 = const()[name = tensor("op_46755_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_46755_end_0 = const()[name = tensor("op_46755_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_46755_end_mask_0 = const()[name = tensor("op_46755_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46755_cast_fp16 = slice_by_index(begin = var_46755_begin_0, end = var_46755_end_0, end_mask = var_46755_end_mask_0, x = transpose_3)[name = tensor("op_46755_cast_fp16")]; + tensor var_46759_begin_0 = const()[name = tensor("op_46759_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_46759_end_0 = const()[name = tensor("op_46759_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_46759_end_mask_0 = const()[name = tensor("op_46759_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46759_cast_fp16 = slice_by_index(begin = var_46759_begin_0, end = var_46759_end_0, end_mask = var_46759_end_mask_0, x = transpose_3)[name = tensor("op_46759_cast_fp16")]; + tensor var_46763_begin_0 = const()[name = tensor("op_46763_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_46763_end_0 = const()[name = tensor("op_46763_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_46763_end_mask_0 = const()[name = tensor("op_46763_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46763_cast_fp16 = slice_by_index(begin = var_46763_begin_0, end = var_46763_end_0, end_mask = var_46763_end_mask_0, x = transpose_3)[name = tensor("op_46763_cast_fp16")]; + tensor var_46767_begin_0 = const()[name = tensor("op_46767_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_46767_end_0 = const()[name = tensor("op_46767_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_46767_end_mask_0 = const()[name = tensor("op_46767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46767_cast_fp16 = slice_by_index(begin = var_46767_begin_0, end = var_46767_end_0, end_mask = var_46767_end_mask_0, x = transpose_3)[name = tensor("op_46767_cast_fp16")]; + tensor var_46771_begin_0 = const()[name = tensor("op_46771_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_46771_end_0 = const()[name = tensor("op_46771_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_46771_end_mask_0 = const()[name = tensor("op_46771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46771_cast_fp16 = slice_by_index(begin = var_46771_begin_0, end = var_46771_end_0, end_mask = var_46771_end_mask_0, x = transpose_3)[name = tensor("op_46771_cast_fp16")]; + tensor var_46775_begin_0 = const()[name = tensor("op_46775_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_46775_end_0 = const()[name = tensor("op_46775_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_46775_end_mask_0 = const()[name = tensor("op_46775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46775_cast_fp16 = slice_by_index(begin = var_46775_begin_0, end = var_46775_end_0, end_mask = var_46775_end_mask_0, x = transpose_3)[name = tensor("op_46775_cast_fp16")]; + tensor var_46779_begin_0 = const()[name = tensor("op_46779_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_46779_end_0 = const()[name = tensor("op_46779_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_46779_end_mask_0 = const()[name = tensor("op_46779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46779_cast_fp16 = slice_by_index(begin = var_46779_begin_0, end = var_46779_end_0, end_mask = var_46779_end_mask_0, x = transpose_3)[name = tensor("op_46779_cast_fp16")]; + tensor var_46783_begin_0 = const()[name = tensor("op_46783_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_46783_end_0 = const()[name = tensor("op_46783_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_46783_end_mask_0 = const()[name = tensor("op_46783_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46783_cast_fp16 = slice_by_index(begin = var_46783_begin_0, end = var_46783_end_0, end_mask = var_46783_end_mask_0, x = transpose_3)[name = tensor("op_46783_cast_fp16")]; + tensor var_46787_begin_0 = const()[name = tensor("op_46787_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_46787_end_0 = const()[name = tensor("op_46787_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_46787_end_mask_0 = const()[name = tensor("op_46787_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46787_cast_fp16 = slice_by_index(begin = var_46787_begin_0, end = var_46787_end_0, end_mask = var_46787_end_mask_0, x = transpose_3)[name = tensor("op_46787_cast_fp16")]; + tensor var_46791_begin_0 = const()[name = tensor("op_46791_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_46791_end_0 = const()[name = tensor("op_46791_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_46791_end_mask_0 = const()[name = tensor("op_46791_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46791_cast_fp16 = slice_by_index(begin = var_46791_begin_0, end = var_46791_end_0, end_mask = var_46791_end_mask_0, x = transpose_3)[name = tensor("op_46791_cast_fp16")]; + tensor var_46795_begin_0 = const()[name = tensor("op_46795_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_46795_end_0 = const()[name = tensor("op_46795_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_46795_end_mask_0 = const()[name = tensor("op_46795_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46795_cast_fp16 = slice_by_index(begin = var_46795_begin_0, end = var_46795_end_0, end_mask = var_46795_end_mask_0, x = transpose_3)[name = tensor("op_46795_cast_fp16")]; + tensor var_46799_begin_0 = const()[name = tensor("op_46799_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_46799_end_0 = const()[name = tensor("op_46799_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_46799_end_mask_0 = const()[name = tensor("op_46799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46799_cast_fp16 = slice_by_index(begin = var_46799_begin_0, end = var_46799_end_0, end_mask = var_46799_end_mask_0, x = transpose_3)[name = tensor("op_46799_cast_fp16")]; + tensor var_46803_begin_0 = const()[name = tensor("op_46803_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_46803_end_0 = const()[name = tensor("op_46803_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_46803_end_mask_0 = const()[name = tensor("op_46803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46803_cast_fp16 = slice_by_index(begin = var_46803_begin_0, end = var_46803_end_0, end_mask = var_46803_end_mask_0, x = transpose_3)[name = tensor("op_46803_cast_fp16")]; + tensor var_46807_begin_0 = const()[name = tensor("op_46807_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_46807_end_0 = const()[name = tensor("op_46807_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_46807_end_mask_0 = const()[name = tensor("op_46807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_46807_cast_fp16 = slice_by_index(begin = var_46807_begin_0, end = var_46807_end_0, end_mask = var_46807_end_mask_0, x = transpose_3)[name = tensor("op_46807_cast_fp16")]; + tensor var_46809_begin_0 = const()[name = tensor("op_46809_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_46809_end_0 = const()[name = tensor("op_46809_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_46809_end_mask_0 = const()[name = tensor("op_46809_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46809_cast_fp16 = slice_by_index(begin = var_46809_begin_0, end = var_46809_end_0, end_mask = var_46809_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46809_cast_fp16")]; + tensor var_46813_begin_0 = const()[name = tensor("op_46813_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_46813_end_0 = const()[name = tensor("op_46813_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_46813_end_mask_0 = const()[name = tensor("op_46813_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46813_cast_fp16 = slice_by_index(begin = var_46813_begin_0, end = var_46813_end_0, end_mask = var_46813_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46813_cast_fp16")]; + tensor var_46817_begin_0 = const()[name = tensor("op_46817_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_46817_end_0 = const()[name = tensor("op_46817_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_46817_end_mask_0 = const()[name = tensor("op_46817_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46817_cast_fp16 = slice_by_index(begin = var_46817_begin_0, end = var_46817_end_0, end_mask = var_46817_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46817_cast_fp16")]; + tensor var_46821_begin_0 = const()[name = tensor("op_46821_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_46821_end_0 = const()[name = tensor("op_46821_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_46821_end_mask_0 = const()[name = tensor("op_46821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46821_cast_fp16 = slice_by_index(begin = var_46821_begin_0, end = var_46821_end_0, end_mask = var_46821_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46821_cast_fp16")]; + tensor var_46825_begin_0 = const()[name = tensor("op_46825_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_46825_end_0 = const()[name = tensor("op_46825_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_46825_end_mask_0 = const()[name = tensor("op_46825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46825_cast_fp16 = slice_by_index(begin = var_46825_begin_0, end = var_46825_end_0, end_mask = var_46825_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46825_cast_fp16")]; + tensor var_46829_begin_0 = const()[name = tensor("op_46829_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_46829_end_0 = const()[name = tensor("op_46829_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_46829_end_mask_0 = const()[name = tensor("op_46829_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46829_cast_fp16 = slice_by_index(begin = var_46829_begin_0, end = var_46829_end_0, end_mask = var_46829_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46829_cast_fp16")]; + tensor var_46833_begin_0 = const()[name = tensor("op_46833_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_46833_end_0 = const()[name = tensor("op_46833_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_46833_end_mask_0 = const()[name = tensor("op_46833_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46833_cast_fp16 = slice_by_index(begin = var_46833_begin_0, end = var_46833_end_0, end_mask = var_46833_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46833_cast_fp16")]; + tensor var_46837_begin_0 = const()[name = tensor("op_46837_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_46837_end_0 = const()[name = tensor("op_46837_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_46837_end_mask_0 = const()[name = tensor("op_46837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46837_cast_fp16 = slice_by_index(begin = var_46837_begin_0, end = var_46837_end_0, end_mask = var_46837_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46837_cast_fp16")]; + tensor var_46841_begin_0 = const()[name = tensor("op_46841_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_46841_end_0 = const()[name = tensor("op_46841_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_46841_end_mask_0 = const()[name = tensor("op_46841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46841_cast_fp16 = slice_by_index(begin = var_46841_begin_0, end = var_46841_end_0, end_mask = var_46841_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46841_cast_fp16")]; + tensor var_46845_begin_0 = const()[name = tensor("op_46845_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_46845_end_0 = const()[name = tensor("op_46845_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_46845_end_mask_0 = const()[name = tensor("op_46845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46845_cast_fp16 = slice_by_index(begin = var_46845_begin_0, end = var_46845_end_0, end_mask = var_46845_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46845_cast_fp16")]; + tensor var_46849_begin_0 = const()[name = tensor("op_46849_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_46849_end_0 = const()[name = tensor("op_46849_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_46849_end_mask_0 = const()[name = tensor("op_46849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46849_cast_fp16 = slice_by_index(begin = var_46849_begin_0, end = var_46849_end_0, end_mask = var_46849_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46849_cast_fp16")]; + tensor var_46853_begin_0 = const()[name = tensor("op_46853_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_46853_end_0 = const()[name = tensor("op_46853_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_46853_end_mask_0 = const()[name = tensor("op_46853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46853_cast_fp16 = slice_by_index(begin = var_46853_begin_0, end = var_46853_end_0, end_mask = var_46853_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46853_cast_fp16")]; + tensor var_46857_begin_0 = const()[name = tensor("op_46857_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_46857_end_0 = const()[name = tensor("op_46857_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_46857_end_mask_0 = const()[name = tensor("op_46857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46857_cast_fp16 = slice_by_index(begin = var_46857_begin_0, end = var_46857_end_0, end_mask = var_46857_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46857_cast_fp16")]; + tensor var_46861_begin_0 = const()[name = tensor("op_46861_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_46861_end_0 = const()[name = tensor("op_46861_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_46861_end_mask_0 = const()[name = tensor("op_46861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46861_cast_fp16 = slice_by_index(begin = var_46861_begin_0, end = var_46861_end_0, end_mask = var_46861_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46861_cast_fp16")]; + tensor var_46865_begin_0 = const()[name = tensor("op_46865_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_46865_end_0 = const()[name = tensor("op_46865_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_46865_end_mask_0 = const()[name = tensor("op_46865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46865_cast_fp16 = slice_by_index(begin = var_46865_begin_0, end = var_46865_end_0, end_mask = var_46865_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46865_cast_fp16")]; + tensor var_46869_begin_0 = const()[name = tensor("op_46869_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_46869_end_0 = const()[name = tensor("op_46869_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_46869_end_mask_0 = const()[name = tensor("op_46869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46869_cast_fp16 = slice_by_index(begin = var_46869_begin_0, end = var_46869_end_0, end_mask = var_46869_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46869_cast_fp16")]; + tensor var_46873_begin_0 = const()[name = tensor("op_46873_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_46873_end_0 = const()[name = tensor("op_46873_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_46873_end_mask_0 = const()[name = tensor("op_46873_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46873_cast_fp16 = slice_by_index(begin = var_46873_begin_0, end = var_46873_end_0, end_mask = var_46873_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46873_cast_fp16")]; + tensor var_46877_begin_0 = const()[name = tensor("op_46877_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_46877_end_0 = const()[name = tensor("op_46877_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_46877_end_mask_0 = const()[name = tensor("op_46877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46877_cast_fp16 = slice_by_index(begin = var_46877_begin_0, end = var_46877_end_0, end_mask = var_46877_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46877_cast_fp16")]; + tensor var_46881_begin_0 = const()[name = tensor("op_46881_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_46881_end_0 = const()[name = tensor("op_46881_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_46881_end_mask_0 = const()[name = tensor("op_46881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46881_cast_fp16 = slice_by_index(begin = var_46881_begin_0, end = var_46881_end_0, end_mask = var_46881_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46881_cast_fp16")]; + tensor var_46885_begin_0 = const()[name = tensor("op_46885_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_46885_end_0 = const()[name = tensor("op_46885_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_46885_end_mask_0 = const()[name = tensor("op_46885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_46885_cast_fp16 = slice_by_index(begin = var_46885_begin_0, end = var_46885_end_0, end_mask = var_46885_end_mask_0, x = value_57_cast_fp16)[name = tensor("op_46885_cast_fp16")]; + tensor var_46889_equation_0 = const()[name = tensor("op_46889_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46889_cast_fp16 = einsum(equation = var_46889_equation_0, values = (var_46731_cast_fp16, var_46173_cast_fp16))[name = tensor("op_46889_cast_fp16")]; + tensor var_46890_to_fp16 = const()[name = tensor("op_46890_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4481_cast_fp16 = mul(x = var_46889_cast_fp16, y = var_46890_to_fp16)[name = tensor("aw_chunk_4481_cast_fp16")]; + tensor var_46893_equation_0 = const()[name = tensor("op_46893_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46893_cast_fp16 = einsum(equation = var_46893_equation_0, values = (var_46731_cast_fp16, var_46180_cast_fp16))[name = tensor("op_46893_cast_fp16")]; + tensor var_46894_to_fp16 = const()[name = tensor("op_46894_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4483_cast_fp16 = mul(x = var_46893_cast_fp16, y = var_46894_to_fp16)[name = tensor("aw_chunk_4483_cast_fp16")]; + tensor var_46897_equation_0 = const()[name = tensor("op_46897_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46897_cast_fp16 = einsum(equation = var_46897_equation_0, values = (var_46731_cast_fp16, var_46187_cast_fp16))[name = tensor("op_46897_cast_fp16")]; + tensor var_46898_to_fp16 = const()[name = tensor("op_46898_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4485_cast_fp16 = mul(x = var_46897_cast_fp16, y = var_46898_to_fp16)[name = tensor("aw_chunk_4485_cast_fp16")]; + tensor var_46901_equation_0 = const()[name = tensor("op_46901_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46901_cast_fp16 = einsum(equation = var_46901_equation_0, values = (var_46731_cast_fp16, var_46194_cast_fp16))[name = tensor("op_46901_cast_fp16")]; + tensor var_46902_to_fp16 = const()[name = tensor("op_46902_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4487_cast_fp16 = mul(x = var_46901_cast_fp16, y = var_46902_to_fp16)[name = tensor("aw_chunk_4487_cast_fp16")]; + tensor var_46905_equation_0 = const()[name = tensor("op_46905_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46905_cast_fp16 = einsum(equation = var_46905_equation_0, values = (var_46735_cast_fp16, var_46201_cast_fp16))[name = tensor("op_46905_cast_fp16")]; + tensor var_46906_to_fp16 = const()[name = tensor("op_46906_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4489_cast_fp16 = mul(x = var_46905_cast_fp16, y = var_46906_to_fp16)[name = tensor("aw_chunk_4489_cast_fp16")]; + tensor var_46909_equation_0 = const()[name = tensor("op_46909_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46909_cast_fp16 = einsum(equation = var_46909_equation_0, values = (var_46735_cast_fp16, var_46208_cast_fp16))[name = tensor("op_46909_cast_fp16")]; + tensor var_46910_to_fp16 = const()[name = tensor("op_46910_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4491_cast_fp16 = mul(x = var_46909_cast_fp16, y = var_46910_to_fp16)[name = tensor("aw_chunk_4491_cast_fp16")]; + tensor var_46913_equation_0 = const()[name = tensor("op_46913_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46913_cast_fp16 = einsum(equation = var_46913_equation_0, values = (var_46735_cast_fp16, var_46215_cast_fp16))[name = tensor("op_46913_cast_fp16")]; + tensor var_46914_to_fp16 = const()[name = tensor("op_46914_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4493_cast_fp16 = mul(x = var_46913_cast_fp16, y = var_46914_to_fp16)[name = tensor("aw_chunk_4493_cast_fp16")]; + tensor var_46917_equation_0 = const()[name = tensor("op_46917_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46917_cast_fp16 = einsum(equation = var_46917_equation_0, values = (var_46735_cast_fp16, var_46222_cast_fp16))[name = tensor("op_46917_cast_fp16")]; + tensor var_46918_to_fp16 = const()[name = tensor("op_46918_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4495_cast_fp16 = mul(x = var_46917_cast_fp16, y = var_46918_to_fp16)[name = tensor("aw_chunk_4495_cast_fp16")]; + tensor var_46921_equation_0 = const()[name = tensor("op_46921_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46921_cast_fp16 = einsum(equation = var_46921_equation_0, values = (var_46739_cast_fp16, var_46229_cast_fp16))[name = tensor("op_46921_cast_fp16")]; + tensor var_46922_to_fp16 = const()[name = tensor("op_46922_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4497_cast_fp16 = mul(x = var_46921_cast_fp16, y = var_46922_to_fp16)[name = tensor("aw_chunk_4497_cast_fp16")]; + tensor var_46925_equation_0 = const()[name = tensor("op_46925_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46925_cast_fp16 = einsum(equation = var_46925_equation_0, values = (var_46739_cast_fp16, var_46236_cast_fp16))[name = tensor("op_46925_cast_fp16")]; + tensor var_46926_to_fp16 = const()[name = tensor("op_46926_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4499_cast_fp16 = mul(x = var_46925_cast_fp16, y = var_46926_to_fp16)[name = tensor("aw_chunk_4499_cast_fp16")]; + tensor var_46929_equation_0 = const()[name = tensor("op_46929_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46929_cast_fp16 = einsum(equation = var_46929_equation_0, values = (var_46739_cast_fp16, var_46243_cast_fp16))[name = tensor("op_46929_cast_fp16")]; + tensor var_46930_to_fp16 = const()[name = tensor("op_46930_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4501_cast_fp16 = mul(x = var_46929_cast_fp16, y = var_46930_to_fp16)[name = tensor("aw_chunk_4501_cast_fp16")]; + tensor var_46933_equation_0 = const()[name = tensor("op_46933_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46933_cast_fp16 = einsum(equation = var_46933_equation_0, values = (var_46739_cast_fp16, var_46250_cast_fp16))[name = tensor("op_46933_cast_fp16")]; + tensor var_46934_to_fp16 = const()[name = tensor("op_46934_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4503_cast_fp16 = mul(x = var_46933_cast_fp16, y = var_46934_to_fp16)[name = tensor("aw_chunk_4503_cast_fp16")]; + tensor var_46937_equation_0 = const()[name = tensor("op_46937_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46937_cast_fp16 = einsum(equation = var_46937_equation_0, values = (var_46743_cast_fp16, var_46257_cast_fp16))[name = tensor("op_46937_cast_fp16")]; + tensor var_46938_to_fp16 = const()[name = tensor("op_46938_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4505_cast_fp16 = mul(x = var_46937_cast_fp16, y = var_46938_to_fp16)[name = tensor("aw_chunk_4505_cast_fp16")]; + tensor var_46941_equation_0 = const()[name = tensor("op_46941_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46941_cast_fp16 = einsum(equation = var_46941_equation_0, values = (var_46743_cast_fp16, var_46264_cast_fp16))[name = tensor("op_46941_cast_fp16")]; + tensor var_46942_to_fp16 = const()[name = tensor("op_46942_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4507_cast_fp16 = mul(x = var_46941_cast_fp16, y = var_46942_to_fp16)[name = tensor("aw_chunk_4507_cast_fp16")]; + tensor var_46945_equation_0 = const()[name = tensor("op_46945_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46945_cast_fp16 = einsum(equation = var_46945_equation_0, values = (var_46743_cast_fp16, var_46271_cast_fp16))[name = tensor("op_46945_cast_fp16")]; + tensor var_46946_to_fp16 = const()[name = tensor("op_46946_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4509_cast_fp16 = mul(x = var_46945_cast_fp16, y = var_46946_to_fp16)[name = tensor("aw_chunk_4509_cast_fp16")]; + tensor var_46949_equation_0 = const()[name = tensor("op_46949_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46949_cast_fp16 = einsum(equation = var_46949_equation_0, values = (var_46743_cast_fp16, var_46278_cast_fp16))[name = tensor("op_46949_cast_fp16")]; + tensor var_46950_to_fp16 = const()[name = tensor("op_46950_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4511_cast_fp16 = mul(x = var_46949_cast_fp16, y = var_46950_to_fp16)[name = tensor("aw_chunk_4511_cast_fp16")]; + tensor var_46953_equation_0 = const()[name = tensor("op_46953_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46953_cast_fp16 = einsum(equation = var_46953_equation_0, values = (var_46747_cast_fp16, var_46285_cast_fp16))[name = tensor("op_46953_cast_fp16")]; + tensor var_46954_to_fp16 = const()[name = tensor("op_46954_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4513_cast_fp16 = mul(x = var_46953_cast_fp16, y = var_46954_to_fp16)[name = tensor("aw_chunk_4513_cast_fp16")]; + tensor var_46957_equation_0 = const()[name = tensor("op_46957_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46957_cast_fp16 = einsum(equation = var_46957_equation_0, values = (var_46747_cast_fp16, var_46292_cast_fp16))[name = tensor("op_46957_cast_fp16")]; + tensor var_46958_to_fp16 = const()[name = tensor("op_46958_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4515_cast_fp16 = mul(x = var_46957_cast_fp16, y = var_46958_to_fp16)[name = tensor("aw_chunk_4515_cast_fp16")]; + tensor var_46961_equation_0 = const()[name = tensor("op_46961_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46961_cast_fp16 = einsum(equation = var_46961_equation_0, values = (var_46747_cast_fp16, var_46299_cast_fp16))[name = tensor("op_46961_cast_fp16")]; + tensor var_46962_to_fp16 = const()[name = tensor("op_46962_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4517_cast_fp16 = mul(x = var_46961_cast_fp16, y = var_46962_to_fp16)[name = tensor("aw_chunk_4517_cast_fp16")]; + tensor var_46965_equation_0 = const()[name = tensor("op_46965_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46965_cast_fp16 = einsum(equation = var_46965_equation_0, values = (var_46747_cast_fp16, var_46306_cast_fp16))[name = tensor("op_46965_cast_fp16")]; + tensor var_46966_to_fp16 = const()[name = tensor("op_46966_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4519_cast_fp16 = mul(x = var_46965_cast_fp16, y = var_46966_to_fp16)[name = tensor("aw_chunk_4519_cast_fp16")]; + tensor var_46969_equation_0 = const()[name = tensor("op_46969_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46969_cast_fp16 = einsum(equation = var_46969_equation_0, values = (var_46751_cast_fp16, var_46313_cast_fp16))[name = tensor("op_46969_cast_fp16")]; + tensor var_46970_to_fp16 = const()[name = tensor("op_46970_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4521_cast_fp16 = mul(x = var_46969_cast_fp16, y = var_46970_to_fp16)[name = tensor("aw_chunk_4521_cast_fp16")]; + tensor var_46973_equation_0 = const()[name = tensor("op_46973_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46973_cast_fp16 = einsum(equation = var_46973_equation_0, values = (var_46751_cast_fp16, var_46320_cast_fp16))[name = tensor("op_46973_cast_fp16")]; + tensor var_46974_to_fp16 = const()[name = tensor("op_46974_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4523_cast_fp16 = mul(x = var_46973_cast_fp16, y = var_46974_to_fp16)[name = tensor("aw_chunk_4523_cast_fp16")]; + tensor var_46977_equation_0 = const()[name = tensor("op_46977_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46977_cast_fp16 = einsum(equation = var_46977_equation_0, values = (var_46751_cast_fp16, var_46327_cast_fp16))[name = tensor("op_46977_cast_fp16")]; + tensor var_46978_to_fp16 = const()[name = tensor("op_46978_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4525_cast_fp16 = mul(x = var_46977_cast_fp16, y = var_46978_to_fp16)[name = tensor("aw_chunk_4525_cast_fp16")]; + tensor var_46981_equation_0 = const()[name = tensor("op_46981_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46981_cast_fp16 = einsum(equation = var_46981_equation_0, values = (var_46751_cast_fp16, var_46334_cast_fp16))[name = tensor("op_46981_cast_fp16")]; + tensor var_46982_to_fp16 = const()[name = tensor("op_46982_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4527_cast_fp16 = mul(x = var_46981_cast_fp16, y = var_46982_to_fp16)[name = tensor("aw_chunk_4527_cast_fp16")]; + tensor var_46985_equation_0 = const()[name = tensor("op_46985_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46985_cast_fp16 = einsum(equation = var_46985_equation_0, values = (var_46755_cast_fp16, var_46341_cast_fp16))[name = tensor("op_46985_cast_fp16")]; + tensor var_46986_to_fp16 = const()[name = tensor("op_46986_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4529_cast_fp16 = mul(x = var_46985_cast_fp16, y = var_46986_to_fp16)[name = tensor("aw_chunk_4529_cast_fp16")]; + tensor var_46989_equation_0 = const()[name = tensor("op_46989_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46989_cast_fp16 = einsum(equation = var_46989_equation_0, values = (var_46755_cast_fp16, var_46348_cast_fp16))[name = tensor("op_46989_cast_fp16")]; + tensor var_46990_to_fp16 = const()[name = tensor("op_46990_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4531_cast_fp16 = mul(x = var_46989_cast_fp16, y = var_46990_to_fp16)[name = tensor("aw_chunk_4531_cast_fp16")]; + tensor var_46993_equation_0 = const()[name = tensor("op_46993_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46993_cast_fp16 = einsum(equation = var_46993_equation_0, values = (var_46755_cast_fp16, var_46355_cast_fp16))[name = tensor("op_46993_cast_fp16")]; + tensor var_46994_to_fp16 = const()[name = tensor("op_46994_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4533_cast_fp16 = mul(x = var_46993_cast_fp16, y = var_46994_to_fp16)[name = tensor("aw_chunk_4533_cast_fp16")]; + tensor var_46997_equation_0 = const()[name = tensor("op_46997_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_46997_cast_fp16 = einsum(equation = var_46997_equation_0, values = (var_46755_cast_fp16, var_46362_cast_fp16))[name = tensor("op_46997_cast_fp16")]; + tensor var_46998_to_fp16 = const()[name = tensor("op_46998_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4535_cast_fp16 = mul(x = var_46997_cast_fp16, y = var_46998_to_fp16)[name = tensor("aw_chunk_4535_cast_fp16")]; + tensor var_47001_equation_0 = const()[name = tensor("op_47001_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47001_cast_fp16 = einsum(equation = var_47001_equation_0, values = (var_46759_cast_fp16, var_46369_cast_fp16))[name = tensor("op_47001_cast_fp16")]; + tensor var_47002_to_fp16 = const()[name = tensor("op_47002_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4537_cast_fp16 = mul(x = var_47001_cast_fp16, y = var_47002_to_fp16)[name = tensor("aw_chunk_4537_cast_fp16")]; + tensor var_47005_equation_0 = const()[name = tensor("op_47005_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47005_cast_fp16 = einsum(equation = var_47005_equation_0, values = (var_46759_cast_fp16, var_46376_cast_fp16))[name = tensor("op_47005_cast_fp16")]; + tensor var_47006_to_fp16 = const()[name = tensor("op_47006_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4539_cast_fp16 = mul(x = var_47005_cast_fp16, y = var_47006_to_fp16)[name = tensor("aw_chunk_4539_cast_fp16")]; + tensor var_47009_equation_0 = const()[name = tensor("op_47009_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47009_cast_fp16 = einsum(equation = var_47009_equation_0, values = (var_46759_cast_fp16, var_46383_cast_fp16))[name = tensor("op_47009_cast_fp16")]; + tensor var_47010_to_fp16 = const()[name = tensor("op_47010_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4541_cast_fp16 = mul(x = var_47009_cast_fp16, y = var_47010_to_fp16)[name = tensor("aw_chunk_4541_cast_fp16")]; + tensor var_47013_equation_0 = const()[name = tensor("op_47013_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47013_cast_fp16 = einsum(equation = var_47013_equation_0, values = (var_46759_cast_fp16, var_46390_cast_fp16))[name = tensor("op_47013_cast_fp16")]; + tensor var_47014_to_fp16 = const()[name = tensor("op_47014_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4543_cast_fp16 = mul(x = var_47013_cast_fp16, y = var_47014_to_fp16)[name = tensor("aw_chunk_4543_cast_fp16")]; + tensor var_47017_equation_0 = const()[name = tensor("op_47017_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47017_cast_fp16 = einsum(equation = var_47017_equation_0, values = (var_46763_cast_fp16, var_46397_cast_fp16))[name = tensor("op_47017_cast_fp16")]; + tensor var_47018_to_fp16 = const()[name = tensor("op_47018_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4545_cast_fp16 = mul(x = var_47017_cast_fp16, y = var_47018_to_fp16)[name = tensor("aw_chunk_4545_cast_fp16")]; + tensor var_47021_equation_0 = const()[name = tensor("op_47021_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47021_cast_fp16 = einsum(equation = var_47021_equation_0, values = (var_46763_cast_fp16, var_46404_cast_fp16))[name = tensor("op_47021_cast_fp16")]; + tensor var_47022_to_fp16 = const()[name = tensor("op_47022_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4547_cast_fp16 = mul(x = var_47021_cast_fp16, y = var_47022_to_fp16)[name = tensor("aw_chunk_4547_cast_fp16")]; + tensor var_47025_equation_0 = const()[name = tensor("op_47025_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47025_cast_fp16 = einsum(equation = var_47025_equation_0, values = (var_46763_cast_fp16, var_46411_cast_fp16))[name = tensor("op_47025_cast_fp16")]; + tensor var_47026_to_fp16 = const()[name = tensor("op_47026_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4549_cast_fp16 = mul(x = var_47025_cast_fp16, y = var_47026_to_fp16)[name = tensor("aw_chunk_4549_cast_fp16")]; + tensor var_47029_equation_0 = const()[name = tensor("op_47029_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47029_cast_fp16 = einsum(equation = var_47029_equation_0, values = (var_46763_cast_fp16, var_46418_cast_fp16))[name = tensor("op_47029_cast_fp16")]; + tensor var_47030_to_fp16 = const()[name = tensor("op_47030_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4551_cast_fp16 = mul(x = var_47029_cast_fp16, y = var_47030_to_fp16)[name = tensor("aw_chunk_4551_cast_fp16")]; + tensor var_47033_equation_0 = const()[name = tensor("op_47033_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47033_cast_fp16 = einsum(equation = var_47033_equation_0, values = (var_46767_cast_fp16, var_46425_cast_fp16))[name = tensor("op_47033_cast_fp16")]; + tensor var_47034_to_fp16 = const()[name = tensor("op_47034_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4553_cast_fp16 = mul(x = var_47033_cast_fp16, y = var_47034_to_fp16)[name = tensor("aw_chunk_4553_cast_fp16")]; + tensor var_47037_equation_0 = const()[name = tensor("op_47037_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47037_cast_fp16 = einsum(equation = var_47037_equation_0, values = (var_46767_cast_fp16, var_46432_cast_fp16))[name = tensor("op_47037_cast_fp16")]; + tensor var_47038_to_fp16 = const()[name = tensor("op_47038_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4555_cast_fp16 = mul(x = var_47037_cast_fp16, y = var_47038_to_fp16)[name = tensor("aw_chunk_4555_cast_fp16")]; + tensor var_47041_equation_0 = const()[name = tensor("op_47041_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47041_cast_fp16 = einsum(equation = var_47041_equation_0, values = (var_46767_cast_fp16, var_46439_cast_fp16))[name = tensor("op_47041_cast_fp16")]; + tensor var_47042_to_fp16 = const()[name = tensor("op_47042_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4557_cast_fp16 = mul(x = var_47041_cast_fp16, y = var_47042_to_fp16)[name = tensor("aw_chunk_4557_cast_fp16")]; + tensor var_47045_equation_0 = const()[name = tensor("op_47045_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47045_cast_fp16 = einsum(equation = var_47045_equation_0, values = (var_46767_cast_fp16, var_46446_cast_fp16))[name = tensor("op_47045_cast_fp16")]; + tensor var_47046_to_fp16 = const()[name = tensor("op_47046_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4559_cast_fp16 = mul(x = var_47045_cast_fp16, y = var_47046_to_fp16)[name = tensor("aw_chunk_4559_cast_fp16")]; + tensor var_47049_equation_0 = const()[name = tensor("op_47049_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47049_cast_fp16 = einsum(equation = var_47049_equation_0, values = (var_46771_cast_fp16, var_46453_cast_fp16))[name = tensor("op_47049_cast_fp16")]; + tensor var_47050_to_fp16 = const()[name = tensor("op_47050_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4561_cast_fp16 = mul(x = var_47049_cast_fp16, y = var_47050_to_fp16)[name = tensor("aw_chunk_4561_cast_fp16")]; + tensor var_47053_equation_0 = const()[name = tensor("op_47053_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47053_cast_fp16 = einsum(equation = var_47053_equation_0, values = (var_46771_cast_fp16, var_46460_cast_fp16))[name = tensor("op_47053_cast_fp16")]; + tensor var_47054_to_fp16 = const()[name = tensor("op_47054_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4563_cast_fp16 = mul(x = var_47053_cast_fp16, y = var_47054_to_fp16)[name = tensor("aw_chunk_4563_cast_fp16")]; + tensor var_47057_equation_0 = const()[name = tensor("op_47057_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47057_cast_fp16 = einsum(equation = var_47057_equation_0, values = (var_46771_cast_fp16, var_46467_cast_fp16))[name = tensor("op_47057_cast_fp16")]; + tensor var_47058_to_fp16 = const()[name = tensor("op_47058_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4565_cast_fp16 = mul(x = var_47057_cast_fp16, y = var_47058_to_fp16)[name = tensor("aw_chunk_4565_cast_fp16")]; + tensor var_47061_equation_0 = const()[name = tensor("op_47061_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47061_cast_fp16 = einsum(equation = var_47061_equation_0, values = (var_46771_cast_fp16, var_46474_cast_fp16))[name = tensor("op_47061_cast_fp16")]; + tensor var_47062_to_fp16 = const()[name = tensor("op_47062_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4567_cast_fp16 = mul(x = var_47061_cast_fp16, y = var_47062_to_fp16)[name = tensor("aw_chunk_4567_cast_fp16")]; + tensor var_47065_equation_0 = const()[name = tensor("op_47065_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47065_cast_fp16 = einsum(equation = var_47065_equation_0, values = (var_46775_cast_fp16, var_46481_cast_fp16))[name = tensor("op_47065_cast_fp16")]; + tensor var_47066_to_fp16 = const()[name = tensor("op_47066_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4569_cast_fp16 = mul(x = var_47065_cast_fp16, y = var_47066_to_fp16)[name = tensor("aw_chunk_4569_cast_fp16")]; + tensor var_47069_equation_0 = const()[name = tensor("op_47069_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47069_cast_fp16 = einsum(equation = var_47069_equation_0, values = (var_46775_cast_fp16, var_46488_cast_fp16))[name = tensor("op_47069_cast_fp16")]; + tensor var_47070_to_fp16 = const()[name = tensor("op_47070_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4571_cast_fp16 = mul(x = var_47069_cast_fp16, y = var_47070_to_fp16)[name = tensor("aw_chunk_4571_cast_fp16")]; + tensor var_47073_equation_0 = const()[name = tensor("op_47073_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47073_cast_fp16 = einsum(equation = var_47073_equation_0, values = (var_46775_cast_fp16, var_46495_cast_fp16))[name = tensor("op_47073_cast_fp16")]; + tensor var_47074_to_fp16 = const()[name = tensor("op_47074_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4573_cast_fp16 = mul(x = var_47073_cast_fp16, y = var_47074_to_fp16)[name = tensor("aw_chunk_4573_cast_fp16")]; + tensor var_47077_equation_0 = const()[name = tensor("op_47077_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47077_cast_fp16 = einsum(equation = var_47077_equation_0, values = (var_46775_cast_fp16, var_46502_cast_fp16))[name = tensor("op_47077_cast_fp16")]; + tensor var_47078_to_fp16 = const()[name = tensor("op_47078_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4575_cast_fp16 = mul(x = var_47077_cast_fp16, y = var_47078_to_fp16)[name = tensor("aw_chunk_4575_cast_fp16")]; + tensor var_47081_equation_0 = const()[name = tensor("op_47081_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47081_cast_fp16 = einsum(equation = var_47081_equation_0, values = (var_46779_cast_fp16, var_46509_cast_fp16))[name = tensor("op_47081_cast_fp16")]; + tensor var_47082_to_fp16 = const()[name = tensor("op_47082_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4577_cast_fp16 = mul(x = var_47081_cast_fp16, y = var_47082_to_fp16)[name = tensor("aw_chunk_4577_cast_fp16")]; + tensor var_47085_equation_0 = const()[name = tensor("op_47085_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47085_cast_fp16 = einsum(equation = var_47085_equation_0, values = (var_46779_cast_fp16, var_46516_cast_fp16))[name = tensor("op_47085_cast_fp16")]; + tensor var_47086_to_fp16 = const()[name = tensor("op_47086_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4579_cast_fp16 = mul(x = var_47085_cast_fp16, y = var_47086_to_fp16)[name = tensor("aw_chunk_4579_cast_fp16")]; + tensor var_47089_equation_0 = const()[name = tensor("op_47089_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47089_cast_fp16 = einsum(equation = var_47089_equation_0, values = (var_46779_cast_fp16, var_46523_cast_fp16))[name = tensor("op_47089_cast_fp16")]; + tensor var_47090_to_fp16 = const()[name = tensor("op_47090_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4581_cast_fp16 = mul(x = var_47089_cast_fp16, y = var_47090_to_fp16)[name = tensor("aw_chunk_4581_cast_fp16")]; + tensor var_47093_equation_0 = const()[name = tensor("op_47093_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47093_cast_fp16 = einsum(equation = var_47093_equation_0, values = (var_46779_cast_fp16, var_46530_cast_fp16))[name = tensor("op_47093_cast_fp16")]; + tensor var_47094_to_fp16 = const()[name = tensor("op_47094_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4583_cast_fp16 = mul(x = var_47093_cast_fp16, y = var_47094_to_fp16)[name = tensor("aw_chunk_4583_cast_fp16")]; + tensor var_47097_equation_0 = const()[name = tensor("op_47097_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47097_cast_fp16 = einsum(equation = var_47097_equation_0, values = (var_46783_cast_fp16, var_46537_cast_fp16))[name = tensor("op_47097_cast_fp16")]; + tensor var_47098_to_fp16 = const()[name = tensor("op_47098_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4585_cast_fp16 = mul(x = var_47097_cast_fp16, y = var_47098_to_fp16)[name = tensor("aw_chunk_4585_cast_fp16")]; + tensor var_47101_equation_0 = const()[name = tensor("op_47101_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47101_cast_fp16 = einsum(equation = var_47101_equation_0, values = (var_46783_cast_fp16, var_46544_cast_fp16))[name = tensor("op_47101_cast_fp16")]; + tensor var_47102_to_fp16 = const()[name = tensor("op_47102_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4587_cast_fp16 = mul(x = var_47101_cast_fp16, y = var_47102_to_fp16)[name = tensor("aw_chunk_4587_cast_fp16")]; + tensor var_47105_equation_0 = const()[name = tensor("op_47105_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47105_cast_fp16 = einsum(equation = var_47105_equation_0, values = (var_46783_cast_fp16, var_46551_cast_fp16))[name = tensor("op_47105_cast_fp16")]; + tensor var_47106_to_fp16 = const()[name = tensor("op_47106_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4589_cast_fp16 = mul(x = var_47105_cast_fp16, y = var_47106_to_fp16)[name = tensor("aw_chunk_4589_cast_fp16")]; + tensor var_47109_equation_0 = const()[name = tensor("op_47109_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47109_cast_fp16 = einsum(equation = var_47109_equation_0, values = (var_46783_cast_fp16, var_46558_cast_fp16))[name = tensor("op_47109_cast_fp16")]; + tensor var_47110_to_fp16 = const()[name = tensor("op_47110_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4591_cast_fp16 = mul(x = var_47109_cast_fp16, y = var_47110_to_fp16)[name = tensor("aw_chunk_4591_cast_fp16")]; + tensor var_47113_equation_0 = const()[name = tensor("op_47113_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47113_cast_fp16 = einsum(equation = var_47113_equation_0, values = (var_46787_cast_fp16, var_46565_cast_fp16))[name = tensor("op_47113_cast_fp16")]; + tensor var_47114_to_fp16 = const()[name = tensor("op_47114_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4593_cast_fp16 = mul(x = var_47113_cast_fp16, y = var_47114_to_fp16)[name = tensor("aw_chunk_4593_cast_fp16")]; + tensor var_47117_equation_0 = const()[name = tensor("op_47117_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47117_cast_fp16 = einsum(equation = var_47117_equation_0, values = (var_46787_cast_fp16, var_46572_cast_fp16))[name = tensor("op_47117_cast_fp16")]; + tensor var_47118_to_fp16 = const()[name = tensor("op_47118_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4595_cast_fp16 = mul(x = var_47117_cast_fp16, y = var_47118_to_fp16)[name = tensor("aw_chunk_4595_cast_fp16")]; + tensor var_47121_equation_0 = const()[name = tensor("op_47121_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47121_cast_fp16 = einsum(equation = var_47121_equation_0, values = (var_46787_cast_fp16, var_46579_cast_fp16))[name = tensor("op_47121_cast_fp16")]; + tensor var_47122_to_fp16 = const()[name = tensor("op_47122_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4597_cast_fp16 = mul(x = var_47121_cast_fp16, y = var_47122_to_fp16)[name = tensor("aw_chunk_4597_cast_fp16")]; + tensor var_47125_equation_0 = const()[name = tensor("op_47125_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47125_cast_fp16 = einsum(equation = var_47125_equation_0, values = (var_46787_cast_fp16, var_46586_cast_fp16))[name = tensor("op_47125_cast_fp16")]; + tensor var_47126_to_fp16 = const()[name = tensor("op_47126_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4599_cast_fp16 = mul(x = var_47125_cast_fp16, y = var_47126_to_fp16)[name = tensor("aw_chunk_4599_cast_fp16")]; + tensor var_47129_equation_0 = const()[name = tensor("op_47129_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47129_cast_fp16 = einsum(equation = var_47129_equation_0, values = (var_46791_cast_fp16, var_46593_cast_fp16))[name = tensor("op_47129_cast_fp16")]; + tensor var_47130_to_fp16 = const()[name = tensor("op_47130_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4601_cast_fp16 = mul(x = var_47129_cast_fp16, y = var_47130_to_fp16)[name = tensor("aw_chunk_4601_cast_fp16")]; + tensor var_47133_equation_0 = const()[name = tensor("op_47133_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47133_cast_fp16 = einsum(equation = var_47133_equation_0, values = (var_46791_cast_fp16, var_46600_cast_fp16))[name = tensor("op_47133_cast_fp16")]; + tensor var_47134_to_fp16 = const()[name = tensor("op_47134_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4603_cast_fp16 = mul(x = var_47133_cast_fp16, y = var_47134_to_fp16)[name = tensor("aw_chunk_4603_cast_fp16")]; + tensor var_47137_equation_0 = const()[name = tensor("op_47137_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47137_cast_fp16 = einsum(equation = var_47137_equation_0, values = (var_46791_cast_fp16, var_46607_cast_fp16))[name = tensor("op_47137_cast_fp16")]; + tensor var_47138_to_fp16 = const()[name = tensor("op_47138_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4605_cast_fp16 = mul(x = var_47137_cast_fp16, y = var_47138_to_fp16)[name = tensor("aw_chunk_4605_cast_fp16")]; + tensor var_47141_equation_0 = const()[name = tensor("op_47141_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47141_cast_fp16 = einsum(equation = var_47141_equation_0, values = (var_46791_cast_fp16, var_46614_cast_fp16))[name = tensor("op_47141_cast_fp16")]; + tensor var_47142_to_fp16 = const()[name = tensor("op_47142_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4607_cast_fp16 = mul(x = var_47141_cast_fp16, y = var_47142_to_fp16)[name = tensor("aw_chunk_4607_cast_fp16")]; + tensor var_47145_equation_0 = const()[name = tensor("op_47145_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47145_cast_fp16 = einsum(equation = var_47145_equation_0, values = (var_46795_cast_fp16, var_46621_cast_fp16))[name = tensor("op_47145_cast_fp16")]; + tensor var_47146_to_fp16 = const()[name = tensor("op_47146_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4609_cast_fp16 = mul(x = var_47145_cast_fp16, y = var_47146_to_fp16)[name = tensor("aw_chunk_4609_cast_fp16")]; + tensor var_47149_equation_0 = const()[name = tensor("op_47149_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47149_cast_fp16 = einsum(equation = var_47149_equation_0, values = (var_46795_cast_fp16, var_46628_cast_fp16))[name = tensor("op_47149_cast_fp16")]; + tensor var_47150_to_fp16 = const()[name = tensor("op_47150_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4611_cast_fp16 = mul(x = var_47149_cast_fp16, y = var_47150_to_fp16)[name = tensor("aw_chunk_4611_cast_fp16")]; + tensor var_47153_equation_0 = const()[name = tensor("op_47153_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47153_cast_fp16 = einsum(equation = var_47153_equation_0, values = (var_46795_cast_fp16, var_46635_cast_fp16))[name = tensor("op_47153_cast_fp16")]; + tensor var_47154_to_fp16 = const()[name = tensor("op_47154_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4613_cast_fp16 = mul(x = var_47153_cast_fp16, y = var_47154_to_fp16)[name = tensor("aw_chunk_4613_cast_fp16")]; + tensor var_47157_equation_0 = const()[name = tensor("op_47157_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47157_cast_fp16 = einsum(equation = var_47157_equation_0, values = (var_46795_cast_fp16, var_46642_cast_fp16))[name = tensor("op_47157_cast_fp16")]; + tensor var_47158_to_fp16 = const()[name = tensor("op_47158_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4615_cast_fp16 = mul(x = var_47157_cast_fp16, y = var_47158_to_fp16)[name = tensor("aw_chunk_4615_cast_fp16")]; + tensor var_47161_equation_0 = const()[name = tensor("op_47161_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47161_cast_fp16 = einsum(equation = var_47161_equation_0, values = (var_46799_cast_fp16, var_46649_cast_fp16))[name = tensor("op_47161_cast_fp16")]; + tensor var_47162_to_fp16 = const()[name = tensor("op_47162_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4617_cast_fp16 = mul(x = var_47161_cast_fp16, y = var_47162_to_fp16)[name = tensor("aw_chunk_4617_cast_fp16")]; + tensor var_47165_equation_0 = const()[name = tensor("op_47165_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47165_cast_fp16 = einsum(equation = var_47165_equation_0, values = (var_46799_cast_fp16, var_46656_cast_fp16))[name = tensor("op_47165_cast_fp16")]; + tensor var_47166_to_fp16 = const()[name = tensor("op_47166_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4619_cast_fp16 = mul(x = var_47165_cast_fp16, y = var_47166_to_fp16)[name = tensor("aw_chunk_4619_cast_fp16")]; + tensor var_47169_equation_0 = const()[name = tensor("op_47169_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47169_cast_fp16 = einsum(equation = var_47169_equation_0, values = (var_46799_cast_fp16, var_46663_cast_fp16))[name = tensor("op_47169_cast_fp16")]; + tensor var_47170_to_fp16 = const()[name = tensor("op_47170_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4621_cast_fp16 = mul(x = var_47169_cast_fp16, y = var_47170_to_fp16)[name = tensor("aw_chunk_4621_cast_fp16")]; + tensor var_47173_equation_0 = const()[name = tensor("op_47173_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47173_cast_fp16 = einsum(equation = var_47173_equation_0, values = (var_46799_cast_fp16, var_46670_cast_fp16))[name = tensor("op_47173_cast_fp16")]; + tensor var_47174_to_fp16 = const()[name = tensor("op_47174_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4623_cast_fp16 = mul(x = var_47173_cast_fp16, y = var_47174_to_fp16)[name = tensor("aw_chunk_4623_cast_fp16")]; + tensor var_47177_equation_0 = const()[name = tensor("op_47177_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47177_cast_fp16 = einsum(equation = var_47177_equation_0, values = (var_46803_cast_fp16, var_46677_cast_fp16))[name = tensor("op_47177_cast_fp16")]; + tensor var_47178_to_fp16 = const()[name = tensor("op_47178_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4625_cast_fp16 = mul(x = var_47177_cast_fp16, y = var_47178_to_fp16)[name = tensor("aw_chunk_4625_cast_fp16")]; + tensor var_47181_equation_0 = const()[name = tensor("op_47181_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47181_cast_fp16 = einsum(equation = var_47181_equation_0, values = (var_46803_cast_fp16, var_46684_cast_fp16))[name = tensor("op_47181_cast_fp16")]; + tensor var_47182_to_fp16 = const()[name = tensor("op_47182_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4627_cast_fp16 = mul(x = var_47181_cast_fp16, y = var_47182_to_fp16)[name = tensor("aw_chunk_4627_cast_fp16")]; + tensor var_47185_equation_0 = const()[name = tensor("op_47185_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47185_cast_fp16 = einsum(equation = var_47185_equation_0, values = (var_46803_cast_fp16, var_46691_cast_fp16))[name = tensor("op_47185_cast_fp16")]; + tensor var_47186_to_fp16 = const()[name = tensor("op_47186_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4629_cast_fp16 = mul(x = var_47185_cast_fp16, y = var_47186_to_fp16)[name = tensor("aw_chunk_4629_cast_fp16")]; + tensor var_47189_equation_0 = const()[name = tensor("op_47189_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47189_cast_fp16 = einsum(equation = var_47189_equation_0, values = (var_46803_cast_fp16, var_46698_cast_fp16))[name = tensor("op_47189_cast_fp16")]; + tensor var_47190_to_fp16 = const()[name = tensor("op_47190_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4631_cast_fp16 = mul(x = var_47189_cast_fp16, y = var_47190_to_fp16)[name = tensor("aw_chunk_4631_cast_fp16")]; + tensor var_47193_equation_0 = const()[name = tensor("op_47193_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47193_cast_fp16 = einsum(equation = var_47193_equation_0, values = (var_46807_cast_fp16, var_46705_cast_fp16))[name = tensor("op_47193_cast_fp16")]; + tensor var_47194_to_fp16 = const()[name = tensor("op_47194_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4633_cast_fp16 = mul(x = var_47193_cast_fp16, y = var_47194_to_fp16)[name = tensor("aw_chunk_4633_cast_fp16")]; + tensor var_47197_equation_0 = const()[name = tensor("op_47197_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47197_cast_fp16 = einsum(equation = var_47197_equation_0, values = (var_46807_cast_fp16, var_46712_cast_fp16))[name = tensor("op_47197_cast_fp16")]; + tensor var_47198_to_fp16 = const()[name = tensor("op_47198_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4635_cast_fp16 = mul(x = var_47197_cast_fp16, y = var_47198_to_fp16)[name = tensor("aw_chunk_4635_cast_fp16")]; + tensor var_47201_equation_0 = const()[name = tensor("op_47201_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47201_cast_fp16 = einsum(equation = var_47201_equation_0, values = (var_46807_cast_fp16, var_46719_cast_fp16))[name = tensor("op_47201_cast_fp16")]; + tensor var_47202_to_fp16 = const()[name = tensor("op_47202_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4637_cast_fp16 = mul(x = var_47201_cast_fp16, y = var_47202_to_fp16)[name = tensor("aw_chunk_4637_cast_fp16")]; + tensor var_47205_equation_0 = const()[name = tensor("op_47205_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_47205_cast_fp16 = einsum(equation = var_47205_equation_0, values = (var_46807_cast_fp16, var_46726_cast_fp16))[name = tensor("op_47205_cast_fp16")]; + tensor var_47206_to_fp16 = const()[name = tensor("op_47206_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4639_cast_fp16 = mul(x = var_47205_cast_fp16, y = var_47206_to_fp16)[name = tensor("aw_chunk_4639_cast_fp16")]; + tensor var_47208_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4481_cast_fp16)[name = tensor("op_47208_cast_fp16")]; + tensor var_47209_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4483_cast_fp16)[name = tensor("op_47209_cast_fp16")]; + tensor var_47210_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4485_cast_fp16)[name = tensor("op_47210_cast_fp16")]; + tensor var_47211_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4487_cast_fp16)[name = tensor("op_47211_cast_fp16")]; + tensor var_47212_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4489_cast_fp16)[name = tensor("op_47212_cast_fp16")]; + tensor var_47213_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4491_cast_fp16)[name = tensor("op_47213_cast_fp16")]; + tensor var_47214_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4493_cast_fp16)[name = tensor("op_47214_cast_fp16")]; + tensor var_47215_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4495_cast_fp16)[name = tensor("op_47215_cast_fp16")]; + tensor var_47216_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4497_cast_fp16)[name = tensor("op_47216_cast_fp16")]; + tensor var_47217_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4499_cast_fp16)[name = tensor("op_47217_cast_fp16")]; + tensor var_47218_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4501_cast_fp16)[name = tensor("op_47218_cast_fp16")]; + tensor var_47219_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4503_cast_fp16)[name = tensor("op_47219_cast_fp16")]; + tensor var_47220_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4505_cast_fp16)[name = tensor("op_47220_cast_fp16")]; + tensor var_47221_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4507_cast_fp16)[name = tensor("op_47221_cast_fp16")]; + tensor var_47222_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4509_cast_fp16)[name = tensor("op_47222_cast_fp16")]; + tensor var_47223_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4511_cast_fp16)[name = tensor("op_47223_cast_fp16")]; + tensor var_47224_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4513_cast_fp16)[name = tensor("op_47224_cast_fp16")]; + tensor var_47225_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4515_cast_fp16)[name = tensor("op_47225_cast_fp16")]; + tensor var_47226_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4517_cast_fp16)[name = tensor("op_47226_cast_fp16")]; + tensor var_47227_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4519_cast_fp16)[name = tensor("op_47227_cast_fp16")]; + tensor var_47228_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4521_cast_fp16)[name = tensor("op_47228_cast_fp16")]; + tensor var_47229_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4523_cast_fp16)[name = tensor("op_47229_cast_fp16")]; + tensor var_47230_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4525_cast_fp16)[name = tensor("op_47230_cast_fp16")]; + tensor var_47231_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4527_cast_fp16)[name = tensor("op_47231_cast_fp16")]; + tensor var_47232_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4529_cast_fp16)[name = tensor("op_47232_cast_fp16")]; + tensor var_47233_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4531_cast_fp16)[name = tensor("op_47233_cast_fp16")]; + tensor var_47234_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4533_cast_fp16)[name = tensor("op_47234_cast_fp16")]; + tensor var_47235_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4535_cast_fp16)[name = tensor("op_47235_cast_fp16")]; + tensor var_47236_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4537_cast_fp16)[name = tensor("op_47236_cast_fp16")]; + tensor var_47237_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4539_cast_fp16)[name = tensor("op_47237_cast_fp16")]; + tensor var_47238_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4541_cast_fp16)[name = tensor("op_47238_cast_fp16")]; + tensor var_47239_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4543_cast_fp16)[name = tensor("op_47239_cast_fp16")]; + tensor var_47240_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4545_cast_fp16)[name = tensor("op_47240_cast_fp16")]; + tensor var_47241_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4547_cast_fp16)[name = tensor("op_47241_cast_fp16")]; + tensor var_47242_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4549_cast_fp16)[name = tensor("op_47242_cast_fp16")]; + tensor var_47243_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4551_cast_fp16)[name = tensor("op_47243_cast_fp16")]; + tensor var_47244_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4553_cast_fp16)[name = tensor("op_47244_cast_fp16")]; + tensor var_47245_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4555_cast_fp16)[name = tensor("op_47245_cast_fp16")]; + tensor var_47246_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4557_cast_fp16)[name = tensor("op_47246_cast_fp16")]; + tensor var_47247_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4559_cast_fp16)[name = tensor("op_47247_cast_fp16")]; + tensor var_47248_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4561_cast_fp16)[name = tensor("op_47248_cast_fp16")]; + tensor var_47249_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4563_cast_fp16)[name = tensor("op_47249_cast_fp16")]; + tensor var_47250_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4565_cast_fp16)[name = tensor("op_47250_cast_fp16")]; + tensor var_47251_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4567_cast_fp16)[name = tensor("op_47251_cast_fp16")]; + tensor var_47252_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4569_cast_fp16)[name = tensor("op_47252_cast_fp16")]; + tensor var_47253_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4571_cast_fp16)[name = tensor("op_47253_cast_fp16")]; + tensor var_47254_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4573_cast_fp16)[name = tensor("op_47254_cast_fp16")]; + tensor var_47255_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4575_cast_fp16)[name = tensor("op_47255_cast_fp16")]; + tensor var_47256_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4577_cast_fp16)[name = tensor("op_47256_cast_fp16")]; + tensor var_47257_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4579_cast_fp16)[name = tensor("op_47257_cast_fp16")]; + tensor var_47258_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4581_cast_fp16)[name = tensor("op_47258_cast_fp16")]; + tensor var_47259_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4583_cast_fp16)[name = tensor("op_47259_cast_fp16")]; + tensor var_47260_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4585_cast_fp16)[name = tensor("op_47260_cast_fp16")]; + tensor var_47261_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4587_cast_fp16)[name = tensor("op_47261_cast_fp16")]; + tensor var_47262_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4589_cast_fp16)[name = tensor("op_47262_cast_fp16")]; + tensor var_47263_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4591_cast_fp16)[name = tensor("op_47263_cast_fp16")]; + tensor var_47264_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4593_cast_fp16)[name = tensor("op_47264_cast_fp16")]; + tensor var_47265_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4595_cast_fp16)[name = tensor("op_47265_cast_fp16")]; + tensor var_47266_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4597_cast_fp16)[name = tensor("op_47266_cast_fp16")]; + tensor var_47267_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4599_cast_fp16)[name = tensor("op_47267_cast_fp16")]; + tensor var_47268_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4601_cast_fp16)[name = tensor("op_47268_cast_fp16")]; + tensor var_47269_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4603_cast_fp16)[name = tensor("op_47269_cast_fp16")]; + tensor var_47270_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4605_cast_fp16)[name = tensor("op_47270_cast_fp16")]; + tensor var_47271_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4607_cast_fp16)[name = tensor("op_47271_cast_fp16")]; + tensor var_47272_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4609_cast_fp16)[name = tensor("op_47272_cast_fp16")]; + tensor var_47273_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4611_cast_fp16)[name = tensor("op_47273_cast_fp16")]; + tensor var_47274_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4613_cast_fp16)[name = tensor("op_47274_cast_fp16")]; + tensor var_47275_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4615_cast_fp16)[name = tensor("op_47275_cast_fp16")]; + tensor var_47276_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4617_cast_fp16)[name = tensor("op_47276_cast_fp16")]; + tensor var_47277_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4619_cast_fp16)[name = tensor("op_47277_cast_fp16")]; + tensor var_47278_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4621_cast_fp16)[name = tensor("op_47278_cast_fp16")]; + tensor var_47279_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4623_cast_fp16)[name = tensor("op_47279_cast_fp16")]; + tensor var_47280_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4625_cast_fp16)[name = tensor("op_47280_cast_fp16")]; + tensor var_47281_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4627_cast_fp16)[name = tensor("op_47281_cast_fp16")]; + tensor var_47282_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4629_cast_fp16)[name = tensor("op_47282_cast_fp16")]; + tensor var_47283_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4631_cast_fp16)[name = tensor("op_47283_cast_fp16")]; + tensor var_47284_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4633_cast_fp16)[name = tensor("op_47284_cast_fp16")]; + tensor var_47285_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4635_cast_fp16)[name = tensor("op_47285_cast_fp16")]; + tensor var_47286_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4637_cast_fp16)[name = tensor("op_47286_cast_fp16")]; + tensor var_47287_cast_fp16 = softmax(axis = var_45979, x = aw_chunk_4639_cast_fp16)[name = tensor("op_47287_cast_fp16")]; + tensor var_47289_equation_0 = const()[name = tensor("op_47289_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47289_cast_fp16 = einsum(equation = var_47289_equation_0, values = (var_46809_cast_fp16, var_47208_cast_fp16))[name = tensor("op_47289_cast_fp16")]; + tensor var_47291_equation_0 = const()[name = tensor("op_47291_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47291_cast_fp16 = einsum(equation = var_47291_equation_0, values = (var_46809_cast_fp16, var_47209_cast_fp16))[name = tensor("op_47291_cast_fp16")]; + tensor var_47293_equation_0 = const()[name = tensor("op_47293_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47293_cast_fp16 = einsum(equation = var_47293_equation_0, values = (var_46809_cast_fp16, var_47210_cast_fp16))[name = tensor("op_47293_cast_fp16")]; + tensor var_47295_equation_0 = const()[name = tensor("op_47295_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47295_cast_fp16 = einsum(equation = var_47295_equation_0, values = (var_46809_cast_fp16, var_47211_cast_fp16))[name = tensor("op_47295_cast_fp16")]; + tensor var_47297_equation_0 = const()[name = tensor("op_47297_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47297_cast_fp16 = einsum(equation = var_47297_equation_0, values = (var_46813_cast_fp16, var_47212_cast_fp16))[name = tensor("op_47297_cast_fp16")]; + tensor var_47299_equation_0 = const()[name = tensor("op_47299_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47299_cast_fp16 = einsum(equation = var_47299_equation_0, values = (var_46813_cast_fp16, var_47213_cast_fp16))[name = tensor("op_47299_cast_fp16")]; + tensor var_47301_equation_0 = const()[name = tensor("op_47301_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47301_cast_fp16 = einsum(equation = var_47301_equation_0, values = (var_46813_cast_fp16, var_47214_cast_fp16))[name = tensor("op_47301_cast_fp16")]; + tensor var_47303_equation_0 = const()[name = tensor("op_47303_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47303_cast_fp16 = einsum(equation = var_47303_equation_0, values = (var_46813_cast_fp16, var_47215_cast_fp16))[name = tensor("op_47303_cast_fp16")]; + tensor var_47305_equation_0 = const()[name = tensor("op_47305_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47305_cast_fp16 = einsum(equation = var_47305_equation_0, values = (var_46817_cast_fp16, var_47216_cast_fp16))[name = tensor("op_47305_cast_fp16")]; + tensor var_47307_equation_0 = const()[name = tensor("op_47307_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47307_cast_fp16 = einsum(equation = var_47307_equation_0, values = (var_46817_cast_fp16, var_47217_cast_fp16))[name = tensor("op_47307_cast_fp16")]; + tensor var_47309_equation_0 = const()[name = tensor("op_47309_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47309_cast_fp16 = einsum(equation = var_47309_equation_0, values = (var_46817_cast_fp16, var_47218_cast_fp16))[name = tensor("op_47309_cast_fp16")]; + tensor var_47311_equation_0 = const()[name = tensor("op_47311_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47311_cast_fp16 = einsum(equation = var_47311_equation_0, values = (var_46817_cast_fp16, var_47219_cast_fp16))[name = tensor("op_47311_cast_fp16")]; + tensor var_47313_equation_0 = const()[name = tensor("op_47313_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47313_cast_fp16 = einsum(equation = var_47313_equation_0, values = (var_46821_cast_fp16, var_47220_cast_fp16))[name = tensor("op_47313_cast_fp16")]; + tensor var_47315_equation_0 = const()[name = tensor("op_47315_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47315_cast_fp16 = einsum(equation = var_47315_equation_0, values = (var_46821_cast_fp16, var_47221_cast_fp16))[name = tensor("op_47315_cast_fp16")]; + tensor var_47317_equation_0 = const()[name = tensor("op_47317_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47317_cast_fp16 = einsum(equation = var_47317_equation_0, values = (var_46821_cast_fp16, var_47222_cast_fp16))[name = tensor("op_47317_cast_fp16")]; + tensor var_47319_equation_0 = const()[name = tensor("op_47319_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47319_cast_fp16 = einsum(equation = var_47319_equation_0, values = (var_46821_cast_fp16, var_47223_cast_fp16))[name = tensor("op_47319_cast_fp16")]; + tensor var_47321_equation_0 = const()[name = tensor("op_47321_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47321_cast_fp16 = einsum(equation = var_47321_equation_0, values = (var_46825_cast_fp16, var_47224_cast_fp16))[name = tensor("op_47321_cast_fp16")]; + tensor var_47323_equation_0 = const()[name = tensor("op_47323_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47323_cast_fp16 = einsum(equation = var_47323_equation_0, values = (var_46825_cast_fp16, var_47225_cast_fp16))[name = tensor("op_47323_cast_fp16")]; + tensor var_47325_equation_0 = const()[name = tensor("op_47325_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47325_cast_fp16 = einsum(equation = var_47325_equation_0, values = (var_46825_cast_fp16, var_47226_cast_fp16))[name = tensor("op_47325_cast_fp16")]; + tensor var_47327_equation_0 = const()[name = tensor("op_47327_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47327_cast_fp16 = einsum(equation = var_47327_equation_0, values = (var_46825_cast_fp16, var_47227_cast_fp16))[name = tensor("op_47327_cast_fp16")]; + tensor var_47329_equation_0 = const()[name = tensor("op_47329_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47329_cast_fp16 = einsum(equation = var_47329_equation_0, values = (var_46829_cast_fp16, var_47228_cast_fp16))[name = tensor("op_47329_cast_fp16")]; + tensor var_47331_equation_0 = const()[name = tensor("op_47331_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47331_cast_fp16 = einsum(equation = var_47331_equation_0, values = (var_46829_cast_fp16, var_47229_cast_fp16))[name = tensor("op_47331_cast_fp16")]; + tensor var_47333_equation_0 = const()[name = tensor("op_47333_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47333_cast_fp16 = einsum(equation = var_47333_equation_0, values = (var_46829_cast_fp16, var_47230_cast_fp16))[name = tensor("op_47333_cast_fp16")]; + tensor var_47335_equation_0 = const()[name = tensor("op_47335_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47335_cast_fp16 = einsum(equation = var_47335_equation_0, values = (var_46829_cast_fp16, var_47231_cast_fp16))[name = tensor("op_47335_cast_fp16")]; + tensor var_47337_equation_0 = const()[name = tensor("op_47337_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47337_cast_fp16 = einsum(equation = var_47337_equation_0, values = (var_46833_cast_fp16, var_47232_cast_fp16))[name = tensor("op_47337_cast_fp16")]; + tensor var_47339_equation_0 = const()[name = tensor("op_47339_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47339_cast_fp16 = einsum(equation = var_47339_equation_0, values = (var_46833_cast_fp16, var_47233_cast_fp16))[name = tensor("op_47339_cast_fp16")]; + tensor var_47341_equation_0 = const()[name = tensor("op_47341_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47341_cast_fp16 = einsum(equation = var_47341_equation_0, values = (var_46833_cast_fp16, var_47234_cast_fp16))[name = tensor("op_47341_cast_fp16")]; + tensor var_47343_equation_0 = const()[name = tensor("op_47343_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47343_cast_fp16 = einsum(equation = var_47343_equation_0, values = (var_46833_cast_fp16, var_47235_cast_fp16))[name = tensor("op_47343_cast_fp16")]; + tensor var_47345_equation_0 = const()[name = tensor("op_47345_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47345_cast_fp16 = einsum(equation = var_47345_equation_0, values = (var_46837_cast_fp16, var_47236_cast_fp16))[name = tensor("op_47345_cast_fp16")]; + tensor var_47347_equation_0 = const()[name = tensor("op_47347_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47347_cast_fp16 = einsum(equation = var_47347_equation_0, values = (var_46837_cast_fp16, var_47237_cast_fp16))[name = tensor("op_47347_cast_fp16")]; + tensor var_47349_equation_0 = const()[name = tensor("op_47349_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47349_cast_fp16 = einsum(equation = var_47349_equation_0, values = (var_46837_cast_fp16, var_47238_cast_fp16))[name = tensor("op_47349_cast_fp16")]; + tensor var_47351_equation_0 = const()[name = tensor("op_47351_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47351_cast_fp16 = einsum(equation = var_47351_equation_0, values = (var_46837_cast_fp16, var_47239_cast_fp16))[name = tensor("op_47351_cast_fp16")]; + tensor var_47353_equation_0 = const()[name = tensor("op_47353_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47353_cast_fp16 = einsum(equation = var_47353_equation_0, values = (var_46841_cast_fp16, var_47240_cast_fp16))[name = tensor("op_47353_cast_fp16")]; + tensor var_47355_equation_0 = const()[name = tensor("op_47355_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47355_cast_fp16 = einsum(equation = var_47355_equation_0, values = (var_46841_cast_fp16, var_47241_cast_fp16))[name = tensor("op_47355_cast_fp16")]; + tensor var_47357_equation_0 = const()[name = tensor("op_47357_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47357_cast_fp16 = einsum(equation = var_47357_equation_0, values = (var_46841_cast_fp16, var_47242_cast_fp16))[name = tensor("op_47357_cast_fp16")]; + tensor var_47359_equation_0 = const()[name = tensor("op_47359_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47359_cast_fp16 = einsum(equation = var_47359_equation_0, values = (var_46841_cast_fp16, var_47243_cast_fp16))[name = tensor("op_47359_cast_fp16")]; + tensor var_47361_equation_0 = const()[name = tensor("op_47361_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47361_cast_fp16 = einsum(equation = var_47361_equation_0, values = (var_46845_cast_fp16, var_47244_cast_fp16))[name = tensor("op_47361_cast_fp16")]; + tensor var_47363_equation_0 = const()[name = tensor("op_47363_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47363_cast_fp16 = einsum(equation = var_47363_equation_0, values = (var_46845_cast_fp16, var_47245_cast_fp16))[name = tensor("op_47363_cast_fp16")]; + tensor var_47365_equation_0 = const()[name = tensor("op_47365_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47365_cast_fp16 = einsum(equation = var_47365_equation_0, values = (var_46845_cast_fp16, var_47246_cast_fp16))[name = tensor("op_47365_cast_fp16")]; + tensor var_47367_equation_0 = const()[name = tensor("op_47367_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47367_cast_fp16 = einsum(equation = var_47367_equation_0, values = (var_46845_cast_fp16, var_47247_cast_fp16))[name = tensor("op_47367_cast_fp16")]; + tensor var_47369_equation_0 = const()[name = tensor("op_47369_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47369_cast_fp16 = einsum(equation = var_47369_equation_0, values = (var_46849_cast_fp16, var_47248_cast_fp16))[name = tensor("op_47369_cast_fp16")]; + tensor var_47371_equation_0 = const()[name = tensor("op_47371_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47371_cast_fp16 = einsum(equation = var_47371_equation_0, values = (var_46849_cast_fp16, var_47249_cast_fp16))[name = tensor("op_47371_cast_fp16")]; + tensor var_47373_equation_0 = const()[name = tensor("op_47373_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47373_cast_fp16 = einsum(equation = var_47373_equation_0, values = (var_46849_cast_fp16, var_47250_cast_fp16))[name = tensor("op_47373_cast_fp16")]; + tensor var_47375_equation_0 = const()[name = tensor("op_47375_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47375_cast_fp16 = einsum(equation = var_47375_equation_0, values = (var_46849_cast_fp16, var_47251_cast_fp16))[name = tensor("op_47375_cast_fp16")]; + tensor var_47377_equation_0 = const()[name = tensor("op_47377_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47377_cast_fp16 = einsum(equation = var_47377_equation_0, values = (var_46853_cast_fp16, var_47252_cast_fp16))[name = tensor("op_47377_cast_fp16")]; + tensor var_47379_equation_0 = const()[name = tensor("op_47379_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47379_cast_fp16 = einsum(equation = var_47379_equation_0, values = (var_46853_cast_fp16, var_47253_cast_fp16))[name = tensor("op_47379_cast_fp16")]; + tensor var_47381_equation_0 = const()[name = tensor("op_47381_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47381_cast_fp16 = einsum(equation = var_47381_equation_0, values = (var_46853_cast_fp16, var_47254_cast_fp16))[name = tensor("op_47381_cast_fp16")]; + tensor var_47383_equation_0 = const()[name = tensor("op_47383_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47383_cast_fp16 = einsum(equation = var_47383_equation_0, values = (var_46853_cast_fp16, var_47255_cast_fp16))[name = tensor("op_47383_cast_fp16")]; + tensor var_47385_equation_0 = const()[name = tensor("op_47385_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47385_cast_fp16 = einsum(equation = var_47385_equation_0, values = (var_46857_cast_fp16, var_47256_cast_fp16))[name = tensor("op_47385_cast_fp16")]; + tensor var_47387_equation_0 = const()[name = tensor("op_47387_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47387_cast_fp16 = einsum(equation = var_47387_equation_0, values = (var_46857_cast_fp16, var_47257_cast_fp16))[name = tensor("op_47387_cast_fp16")]; + tensor var_47389_equation_0 = const()[name = tensor("op_47389_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47389_cast_fp16 = einsum(equation = var_47389_equation_0, values = (var_46857_cast_fp16, var_47258_cast_fp16))[name = tensor("op_47389_cast_fp16")]; + tensor var_47391_equation_0 = const()[name = tensor("op_47391_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47391_cast_fp16 = einsum(equation = var_47391_equation_0, values = (var_46857_cast_fp16, var_47259_cast_fp16))[name = tensor("op_47391_cast_fp16")]; + tensor var_47393_equation_0 = const()[name = tensor("op_47393_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47393_cast_fp16 = einsum(equation = var_47393_equation_0, values = (var_46861_cast_fp16, var_47260_cast_fp16))[name = tensor("op_47393_cast_fp16")]; + tensor var_47395_equation_0 = const()[name = tensor("op_47395_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47395_cast_fp16 = einsum(equation = var_47395_equation_0, values = (var_46861_cast_fp16, var_47261_cast_fp16))[name = tensor("op_47395_cast_fp16")]; + tensor var_47397_equation_0 = const()[name = tensor("op_47397_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47397_cast_fp16 = einsum(equation = var_47397_equation_0, values = (var_46861_cast_fp16, var_47262_cast_fp16))[name = tensor("op_47397_cast_fp16")]; + tensor var_47399_equation_0 = const()[name = tensor("op_47399_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47399_cast_fp16 = einsum(equation = var_47399_equation_0, values = (var_46861_cast_fp16, var_47263_cast_fp16))[name = tensor("op_47399_cast_fp16")]; + tensor var_47401_equation_0 = const()[name = tensor("op_47401_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47401_cast_fp16 = einsum(equation = var_47401_equation_0, values = (var_46865_cast_fp16, var_47264_cast_fp16))[name = tensor("op_47401_cast_fp16")]; + tensor var_47403_equation_0 = const()[name = tensor("op_47403_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47403_cast_fp16 = einsum(equation = var_47403_equation_0, values = (var_46865_cast_fp16, var_47265_cast_fp16))[name = tensor("op_47403_cast_fp16")]; + tensor var_47405_equation_0 = const()[name = tensor("op_47405_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47405_cast_fp16 = einsum(equation = var_47405_equation_0, values = (var_46865_cast_fp16, var_47266_cast_fp16))[name = tensor("op_47405_cast_fp16")]; + tensor var_47407_equation_0 = const()[name = tensor("op_47407_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47407_cast_fp16 = einsum(equation = var_47407_equation_0, values = (var_46865_cast_fp16, var_47267_cast_fp16))[name = tensor("op_47407_cast_fp16")]; + tensor var_47409_equation_0 = const()[name = tensor("op_47409_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47409_cast_fp16 = einsum(equation = var_47409_equation_0, values = (var_46869_cast_fp16, var_47268_cast_fp16))[name = tensor("op_47409_cast_fp16")]; + tensor var_47411_equation_0 = const()[name = tensor("op_47411_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47411_cast_fp16 = einsum(equation = var_47411_equation_0, values = (var_46869_cast_fp16, var_47269_cast_fp16))[name = tensor("op_47411_cast_fp16")]; + tensor var_47413_equation_0 = const()[name = tensor("op_47413_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47413_cast_fp16 = einsum(equation = var_47413_equation_0, values = (var_46869_cast_fp16, var_47270_cast_fp16))[name = tensor("op_47413_cast_fp16")]; + tensor var_47415_equation_0 = const()[name = tensor("op_47415_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47415_cast_fp16 = einsum(equation = var_47415_equation_0, values = (var_46869_cast_fp16, var_47271_cast_fp16))[name = tensor("op_47415_cast_fp16")]; + tensor var_47417_equation_0 = const()[name = tensor("op_47417_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47417_cast_fp16 = einsum(equation = var_47417_equation_0, values = (var_46873_cast_fp16, var_47272_cast_fp16))[name = tensor("op_47417_cast_fp16")]; + tensor var_47419_equation_0 = const()[name = tensor("op_47419_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47419_cast_fp16 = einsum(equation = var_47419_equation_0, values = (var_46873_cast_fp16, var_47273_cast_fp16))[name = tensor("op_47419_cast_fp16")]; + tensor var_47421_equation_0 = const()[name = tensor("op_47421_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47421_cast_fp16 = einsum(equation = var_47421_equation_0, values = (var_46873_cast_fp16, var_47274_cast_fp16))[name = tensor("op_47421_cast_fp16")]; + tensor var_47423_equation_0 = const()[name = tensor("op_47423_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47423_cast_fp16 = einsum(equation = var_47423_equation_0, values = (var_46873_cast_fp16, var_47275_cast_fp16))[name = tensor("op_47423_cast_fp16")]; + tensor var_47425_equation_0 = const()[name = tensor("op_47425_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47425_cast_fp16 = einsum(equation = var_47425_equation_0, values = (var_46877_cast_fp16, var_47276_cast_fp16))[name = tensor("op_47425_cast_fp16")]; + tensor var_47427_equation_0 = const()[name = tensor("op_47427_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47427_cast_fp16 = einsum(equation = var_47427_equation_0, values = (var_46877_cast_fp16, var_47277_cast_fp16))[name = tensor("op_47427_cast_fp16")]; + tensor var_47429_equation_0 = const()[name = tensor("op_47429_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47429_cast_fp16 = einsum(equation = var_47429_equation_0, values = (var_46877_cast_fp16, var_47278_cast_fp16))[name = tensor("op_47429_cast_fp16")]; + tensor var_47431_equation_0 = const()[name = tensor("op_47431_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47431_cast_fp16 = einsum(equation = var_47431_equation_0, values = (var_46877_cast_fp16, var_47279_cast_fp16))[name = tensor("op_47431_cast_fp16")]; + tensor var_47433_equation_0 = const()[name = tensor("op_47433_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47433_cast_fp16 = einsum(equation = var_47433_equation_0, values = (var_46881_cast_fp16, var_47280_cast_fp16))[name = tensor("op_47433_cast_fp16")]; + tensor var_47435_equation_0 = const()[name = tensor("op_47435_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47435_cast_fp16 = einsum(equation = var_47435_equation_0, values = (var_46881_cast_fp16, var_47281_cast_fp16))[name = tensor("op_47435_cast_fp16")]; + tensor var_47437_equation_0 = const()[name = tensor("op_47437_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47437_cast_fp16 = einsum(equation = var_47437_equation_0, values = (var_46881_cast_fp16, var_47282_cast_fp16))[name = tensor("op_47437_cast_fp16")]; + tensor var_47439_equation_0 = const()[name = tensor("op_47439_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47439_cast_fp16 = einsum(equation = var_47439_equation_0, values = (var_46881_cast_fp16, var_47283_cast_fp16))[name = tensor("op_47439_cast_fp16")]; + tensor var_47441_equation_0 = const()[name = tensor("op_47441_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47441_cast_fp16 = einsum(equation = var_47441_equation_0, values = (var_46885_cast_fp16, var_47284_cast_fp16))[name = tensor("op_47441_cast_fp16")]; + tensor var_47443_equation_0 = const()[name = tensor("op_47443_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47443_cast_fp16 = einsum(equation = var_47443_equation_0, values = (var_46885_cast_fp16, var_47285_cast_fp16))[name = tensor("op_47443_cast_fp16")]; + tensor var_47445_equation_0 = const()[name = tensor("op_47445_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47445_cast_fp16 = einsum(equation = var_47445_equation_0, values = (var_46885_cast_fp16, var_47286_cast_fp16))[name = tensor("op_47445_cast_fp16")]; + tensor var_47447_equation_0 = const()[name = tensor("op_47447_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_47447_cast_fp16 = einsum(equation = var_47447_equation_0, values = (var_46885_cast_fp16, var_47287_cast_fp16))[name = tensor("op_47447_cast_fp16")]; + tensor var_47449_interleave_0 = const()[name = tensor("op_47449_interleave_0"), val = tensor(false)]; + tensor var_47449_cast_fp16 = concat(axis = var_45954, interleave = var_47449_interleave_0, values = (var_47289_cast_fp16, var_47291_cast_fp16, var_47293_cast_fp16, var_47295_cast_fp16))[name = tensor("op_47449_cast_fp16")]; + tensor var_47451_interleave_0 = const()[name = tensor("op_47451_interleave_0"), val = tensor(false)]; + tensor var_47451_cast_fp16 = concat(axis = var_45954, interleave = var_47451_interleave_0, values = (var_47297_cast_fp16, var_47299_cast_fp16, var_47301_cast_fp16, var_47303_cast_fp16))[name = tensor("op_47451_cast_fp16")]; + tensor var_47453_interleave_0 = const()[name = tensor("op_47453_interleave_0"), val = tensor(false)]; + tensor var_47453_cast_fp16 = concat(axis = var_45954, interleave = var_47453_interleave_0, values = (var_47305_cast_fp16, var_47307_cast_fp16, var_47309_cast_fp16, var_47311_cast_fp16))[name = tensor("op_47453_cast_fp16")]; + tensor var_47455_interleave_0 = const()[name = tensor("op_47455_interleave_0"), val = tensor(false)]; + tensor var_47455_cast_fp16 = concat(axis = var_45954, interleave = var_47455_interleave_0, values = (var_47313_cast_fp16, var_47315_cast_fp16, var_47317_cast_fp16, var_47319_cast_fp16))[name = tensor("op_47455_cast_fp16")]; + tensor var_47457_interleave_0 = const()[name = tensor("op_47457_interleave_0"), val = tensor(false)]; + tensor var_47457_cast_fp16 = concat(axis = var_45954, interleave = var_47457_interleave_0, values = (var_47321_cast_fp16, var_47323_cast_fp16, var_47325_cast_fp16, var_47327_cast_fp16))[name = tensor("op_47457_cast_fp16")]; + tensor var_47459_interleave_0 = const()[name = tensor("op_47459_interleave_0"), val = tensor(false)]; + tensor var_47459_cast_fp16 = concat(axis = var_45954, interleave = var_47459_interleave_0, values = (var_47329_cast_fp16, var_47331_cast_fp16, var_47333_cast_fp16, var_47335_cast_fp16))[name = tensor("op_47459_cast_fp16")]; + tensor var_47461_interleave_0 = const()[name = tensor("op_47461_interleave_0"), val = tensor(false)]; + tensor var_47461_cast_fp16 = concat(axis = var_45954, interleave = var_47461_interleave_0, values = (var_47337_cast_fp16, var_47339_cast_fp16, var_47341_cast_fp16, var_47343_cast_fp16))[name = tensor("op_47461_cast_fp16")]; + tensor var_47463_interleave_0 = const()[name = tensor("op_47463_interleave_0"), val = tensor(false)]; + tensor var_47463_cast_fp16 = concat(axis = var_45954, interleave = var_47463_interleave_0, values = (var_47345_cast_fp16, var_47347_cast_fp16, var_47349_cast_fp16, var_47351_cast_fp16))[name = tensor("op_47463_cast_fp16")]; + tensor var_47465_interleave_0 = const()[name = tensor("op_47465_interleave_0"), val = tensor(false)]; + tensor var_47465_cast_fp16 = concat(axis = var_45954, interleave = var_47465_interleave_0, values = (var_47353_cast_fp16, var_47355_cast_fp16, var_47357_cast_fp16, var_47359_cast_fp16))[name = tensor("op_47465_cast_fp16")]; + tensor var_47467_interleave_0 = const()[name = tensor("op_47467_interleave_0"), val = tensor(false)]; + tensor var_47467_cast_fp16 = concat(axis = var_45954, interleave = var_47467_interleave_0, values = (var_47361_cast_fp16, var_47363_cast_fp16, var_47365_cast_fp16, var_47367_cast_fp16))[name = tensor("op_47467_cast_fp16")]; + tensor var_47469_interleave_0 = const()[name = tensor("op_47469_interleave_0"), val = tensor(false)]; + tensor var_47469_cast_fp16 = concat(axis = var_45954, interleave = var_47469_interleave_0, values = (var_47369_cast_fp16, var_47371_cast_fp16, var_47373_cast_fp16, var_47375_cast_fp16))[name = tensor("op_47469_cast_fp16")]; + tensor var_47471_interleave_0 = const()[name = tensor("op_47471_interleave_0"), val = tensor(false)]; + tensor var_47471_cast_fp16 = concat(axis = var_45954, interleave = var_47471_interleave_0, values = (var_47377_cast_fp16, var_47379_cast_fp16, var_47381_cast_fp16, var_47383_cast_fp16))[name = tensor("op_47471_cast_fp16")]; + tensor var_47473_interleave_0 = const()[name = tensor("op_47473_interleave_0"), val = tensor(false)]; + tensor var_47473_cast_fp16 = concat(axis = var_45954, interleave = var_47473_interleave_0, values = (var_47385_cast_fp16, var_47387_cast_fp16, var_47389_cast_fp16, var_47391_cast_fp16))[name = tensor("op_47473_cast_fp16")]; + tensor var_47475_interleave_0 = const()[name = tensor("op_47475_interleave_0"), val = tensor(false)]; + tensor var_47475_cast_fp16 = concat(axis = var_45954, interleave = var_47475_interleave_0, values = (var_47393_cast_fp16, var_47395_cast_fp16, var_47397_cast_fp16, var_47399_cast_fp16))[name = tensor("op_47475_cast_fp16")]; + tensor var_47477_interleave_0 = const()[name = tensor("op_47477_interleave_0"), val = tensor(false)]; + tensor var_47477_cast_fp16 = concat(axis = var_45954, interleave = var_47477_interleave_0, values = (var_47401_cast_fp16, var_47403_cast_fp16, var_47405_cast_fp16, var_47407_cast_fp16))[name = tensor("op_47477_cast_fp16")]; + tensor var_47479_interleave_0 = const()[name = tensor("op_47479_interleave_0"), val = tensor(false)]; + tensor var_47479_cast_fp16 = concat(axis = var_45954, interleave = var_47479_interleave_0, values = (var_47409_cast_fp16, var_47411_cast_fp16, var_47413_cast_fp16, var_47415_cast_fp16))[name = tensor("op_47479_cast_fp16")]; + tensor var_47481_interleave_0 = const()[name = tensor("op_47481_interleave_0"), val = tensor(false)]; + tensor var_47481_cast_fp16 = concat(axis = var_45954, interleave = var_47481_interleave_0, values = (var_47417_cast_fp16, var_47419_cast_fp16, var_47421_cast_fp16, var_47423_cast_fp16))[name = tensor("op_47481_cast_fp16")]; + tensor var_47483_interleave_0 = const()[name = tensor("op_47483_interleave_0"), val = tensor(false)]; + tensor var_47483_cast_fp16 = concat(axis = var_45954, interleave = var_47483_interleave_0, values = (var_47425_cast_fp16, var_47427_cast_fp16, var_47429_cast_fp16, var_47431_cast_fp16))[name = tensor("op_47483_cast_fp16")]; + tensor var_47485_interleave_0 = const()[name = tensor("op_47485_interleave_0"), val = tensor(false)]; + tensor var_47485_cast_fp16 = concat(axis = var_45954, interleave = var_47485_interleave_0, values = (var_47433_cast_fp16, var_47435_cast_fp16, var_47437_cast_fp16, var_47439_cast_fp16))[name = tensor("op_47485_cast_fp16")]; + tensor var_47487_interleave_0 = const()[name = tensor("op_47487_interleave_0"), val = tensor(false)]; + tensor var_47487_cast_fp16 = concat(axis = var_45954, interleave = var_47487_interleave_0, values = (var_47441_cast_fp16, var_47443_cast_fp16, var_47445_cast_fp16, var_47447_cast_fp16))[name = tensor("op_47487_cast_fp16")]; + tensor input_567_interleave_0 = const()[name = tensor("input_567_interleave_0"), val = tensor(false)]; + tensor input_567_cast_fp16 = concat(axis = var_45979, interleave = input_567_interleave_0, values = (var_47449_cast_fp16, var_47451_cast_fp16, var_47453_cast_fp16, var_47455_cast_fp16, var_47457_cast_fp16, var_47459_cast_fp16, var_47461_cast_fp16, var_47463_cast_fp16, var_47465_cast_fp16, var_47467_cast_fp16, var_47469_cast_fp16, var_47471_cast_fp16, var_47473_cast_fp16, var_47475_cast_fp16, var_47477_cast_fp16, var_47479_cast_fp16, var_47481_cast_fp16, var_47483_cast_fp16, var_47485_cast_fp16, var_47487_cast_fp16))[name = tensor("input_567_cast_fp16")]; + tensor var_47495 = const()[name = tensor("op_47495"), val = tensor([1, 1])]; + tensor var_47497 = const()[name = tensor("op_47497"), val = tensor([1, 1])]; + tensor pretrained_out_343_pad_type_0 = const()[name = tensor("pretrained_out_343_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_343_pad_0 = const()[name = tensor("pretrained_out_343_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313845760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314665024))), name = tensor("layers_28_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_28_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_28_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314665152)))]; + tensor pretrained_out_343_cast_fp16 = conv(bias = layers_28_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_47497, groups = var_45979, pad = pretrained_out_343_pad_0, pad_type = pretrained_out_343_pad_type_0, strides = var_47495, weight = layers_28_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_567_cast_fp16)[name = tensor("pretrained_out_343_cast_fp16")]; + tensor var_47501 = const()[name = tensor("op_47501"), val = tensor([1, 1])]; + tensor var_47503 = const()[name = tensor("op_47503"), val = tensor([1, 1])]; + tensor input_569_pad_type_0 = const()[name = tensor("input_569_pad_type_0"), val = tensor("custom")]; + tensor input_569_pad_0 = const()[name = tensor("input_569_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_28_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314667776)))]; + tensor input_569_cast_fp16 = conv(dilations = var_47503, groups = var_45979, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = var_47501, weight = layers_28_self_attn_o_proj_loraA_weight_to_fp16, x = input_567_cast_fp16)[name = tensor("input_569_cast_fp16")]; + tensor var_47507 = const()[name = tensor("op_47507"), val = tensor([1, 1])]; + tensor var_47509 = const()[name = tensor("op_47509"), val = tensor([1, 1])]; + tensor lora_out_685_pad_type_0 = const()[name = tensor("lora_out_685_pad_type_0"), val = tensor("custom")]; + tensor lora_out_685_pad_0 = const()[name = tensor("lora_out_685_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_687_weight_0_to_fp16 = const()[name = tensor("lora_out_687_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314708800)))]; + tensor lora_out_687_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_47509, groups = var_45979, pad = lora_out_685_pad_0, pad_type = lora_out_685_pad_type_0, strides = var_47507, weight = lora_out_687_weight_0_to_fp16, x = input_569_cast_fp16)[name = tensor("lora_out_687_cast_fp16")]; + tensor obj_115_cast_fp16 = add(x = pretrained_out_343_cast_fp16, y = lora_out_687_cast_fp16)[name = tensor("obj_115_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_115_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; + tensor var_47518 = const()[name = tensor("op_47518"), val = tensor([1])]; + tensor channels_mean_115_cast_fp16 = reduce_mean(axes = var_47518, keep_dims = var_45980, x = inputs_115_cast_fp16)[name = tensor("channels_mean_115_cast_fp16")]; + tensor zero_mean_115_cast_fp16 = sub(x = inputs_115_cast_fp16, y = channels_mean_115_cast_fp16)[name = tensor("zero_mean_115_cast_fp16")]; + tensor zero_mean_sq_115_cast_fp16 = mul(x = zero_mean_115_cast_fp16, y = zero_mean_115_cast_fp16)[name = tensor("zero_mean_sq_115_cast_fp16")]; + tensor var_47522 = const()[name = tensor("op_47522"), val = tensor([1])]; + tensor var_47523_cast_fp16 = reduce_mean(axes = var_47522, keep_dims = var_45980, x = zero_mean_sq_115_cast_fp16)[name = tensor("op_47523_cast_fp16")]; + tensor var_47524_to_fp16 = const()[name = tensor("op_47524_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_47525_cast_fp16 = add(x = var_47523_cast_fp16, y = var_47524_to_fp16)[name = tensor("op_47525_cast_fp16")]; + tensor denom_115_epsilon_0 = const()[name = tensor("denom_115_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_115_cast_fp16 = rsqrt(epsilon = denom_115_epsilon_0, x = var_47525_cast_fp16)[name = tensor("denom_115_cast_fp16")]; + tensor out_115_cast_fp16 = mul(x = zero_mean_115_cast_fp16, y = denom_115_cast_fp16)[name = tensor("out_115_cast_fp16")]; + tensor input_571_gamma_0_to_fp16 = const()[name = tensor("input_571_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314749824)))]; + tensor input_571_beta_0_to_fp16 = const()[name = tensor("input_571_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314752448)))]; + tensor input_571_epsilon_0_to_fp16 = const()[name = tensor("input_571_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_571_cast_fp16 = batch_norm(beta = input_571_beta_0_to_fp16, epsilon = input_571_epsilon_0_to_fp16, gamma = input_571_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_115_cast_fp16)[name = tensor("input_571_cast_fp16")]; + tensor var_47539 = const()[name = tensor("op_47539"), val = tensor([1, 1])]; + tensor var_47541 = const()[name = tensor("op_47541"), val = tensor([1, 1])]; + tensor pretrained_out_345_pad_type_0 = const()[name = tensor("pretrained_out_345_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_345_pad_0 = const()[name = tensor("pretrained_out_345_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314755072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318031936))), name = tensor("layers_28_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_28_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_28_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318032064)))]; + tensor pretrained_out_345_cast_fp16 = conv(bias = layers_28_fc1_pretrained_bias_to_fp16, dilations = var_47541, groups = var_45979, pad = pretrained_out_345_pad_0, pad_type = pretrained_out_345_pad_type_0, strides = var_47539, weight = layers_28_fc1_pretrained_weight_to_fp16_palettized, x = input_571_cast_fp16)[name = tensor("pretrained_out_345_cast_fp16")]; + tensor var_47545 = const()[name = tensor("op_47545"), val = tensor([1, 1])]; + tensor var_47547 = const()[name = tensor("op_47547"), val = tensor([1, 1])]; + tensor input_573_pad_type_0 = const()[name = tensor("input_573_pad_type_0"), val = tensor("custom")]; + tensor input_573_pad_0 = const()[name = tensor("input_573_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_28_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318042368)))]; + tensor input_573_cast_fp16 = conv(dilations = var_47547, groups = var_45979, pad = input_573_pad_0, pad_type = input_573_pad_type_0, strides = var_47545, weight = layers_28_fc1_loraA_weight_to_fp16, x = input_571_cast_fp16)[name = tensor("input_573_cast_fp16")]; + tensor var_47551 = const()[name = tensor("op_47551"), val = tensor([1, 1])]; + tensor var_47553 = const()[name = tensor("op_47553"), val = tensor([1, 1])]; + tensor lora_out_689_pad_type_0 = const()[name = tensor("lora_out_689_pad_type_0"), val = tensor("custom")]; + tensor lora_out_689_pad_0 = const()[name = tensor("lora_out_689_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_691_weight_0_to_fp16 = const()[name = tensor("lora_out_691_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318083392)))]; + tensor lora_out_691_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_47553, groups = var_45979, pad = lora_out_689_pad_0, pad_type = lora_out_689_pad_type_0, strides = var_47551, weight = lora_out_691_weight_0_to_fp16, x = input_573_cast_fp16)[name = tensor("lora_out_691_cast_fp16")]; + tensor input_575_cast_fp16 = add(x = pretrained_out_345_cast_fp16, y = lora_out_691_cast_fp16)[name = tensor("input_575_cast_fp16")]; + tensor input_577_mode_0 = const()[name = tensor("input_577_mode_0"), val = tensor("EXACT")]; + tensor input_577_cast_fp16 = gelu(mode = input_577_mode_0, x = input_575_cast_fp16)[name = tensor("input_577_cast_fp16")]; + tensor var_47565 = const()[name = tensor("op_47565"), val = tensor([1, 1])]; + tensor var_47567 = const()[name = tensor("op_47567"), val = tensor([1, 1])]; + tensor pretrained_out_347_pad_type_0 = const()[name = tensor("pretrained_out_347_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_347_pad_0 = const()[name = tensor("pretrained_out_347_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318247296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321524160))), name = tensor("layers_28_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_28_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_28_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321524288)))]; + tensor pretrained_out_347_cast_fp16 = conv(bias = layers_28_fc2_pretrained_bias_to_fp16, dilations = var_47567, groups = var_45979, pad = pretrained_out_347_pad_0, pad_type = pretrained_out_347_pad_type_0, strides = var_47565, weight = layers_28_fc2_pretrained_weight_to_fp16_palettized, x = input_577_cast_fp16)[name = tensor("pretrained_out_347_cast_fp16")]; + tensor var_47571 = const()[name = tensor("op_47571"), val = tensor([1, 1])]; + tensor var_47573 = const()[name = tensor("op_47573"), val = tensor([1, 1])]; + tensor input_579_pad_type_0 = const()[name = tensor("input_579_pad_type_0"), val = tensor("custom")]; + tensor input_579_pad_0 = const()[name = tensor("input_579_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_28_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_28_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321526912)))]; + tensor input_579_cast_fp16 = conv(dilations = var_47573, groups = var_45979, pad = input_579_pad_0, pad_type = input_579_pad_type_0, strides = var_47571, weight = layers_28_fc2_loraA_weight_to_fp16, x = input_577_cast_fp16)[name = tensor("input_579_cast_fp16")]; + tensor var_47577 = const()[name = tensor("op_47577"), val = tensor([1, 1])]; + tensor var_47579 = const()[name = tensor("op_47579"), val = tensor([1, 1])]; + tensor lora_out_693_pad_type_0 = const()[name = tensor("lora_out_693_pad_type_0"), val = tensor("custom")]; + tensor lora_out_693_pad_0 = const()[name = tensor("lora_out_693_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_695_weight_0_to_fp16 = const()[name = tensor("lora_out_695_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321690816)))]; + tensor lora_out_695_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_47579, groups = var_45979, pad = lora_out_693_pad_0, pad_type = lora_out_693_pad_type_0, strides = var_47577, weight = lora_out_695_weight_0_to_fp16, x = input_579_cast_fp16)[name = tensor("lora_out_695_cast_fp16")]; + tensor hidden_states_61_cast_fp16 = add(x = pretrained_out_347_cast_fp16, y = lora_out_695_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = hidden_states_61_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; + tensor var_47589 = const()[name = tensor("op_47589"), val = tensor(3)]; + tensor var_47614 = const()[name = tensor("op_47614"), val = tensor(1)]; + tensor var_47615 = const()[name = tensor("op_47615"), val = tensor(true)]; + tensor var_47625 = const()[name = tensor("op_47625"), val = tensor([1])]; + tensor channels_mean_117_cast_fp16 = reduce_mean(axes = var_47625, keep_dims = var_47615, x = inputs_117_cast_fp16)[name = tensor("channels_mean_117_cast_fp16")]; + tensor zero_mean_117_cast_fp16 = sub(x = inputs_117_cast_fp16, y = channels_mean_117_cast_fp16)[name = tensor("zero_mean_117_cast_fp16")]; + tensor zero_mean_sq_117_cast_fp16 = mul(x = zero_mean_117_cast_fp16, y = zero_mean_117_cast_fp16)[name = tensor("zero_mean_sq_117_cast_fp16")]; + tensor var_47629 = const()[name = tensor("op_47629"), val = tensor([1])]; + tensor var_47630_cast_fp16 = reduce_mean(axes = var_47629, keep_dims = var_47615, x = zero_mean_sq_117_cast_fp16)[name = tensor("op_47630_cast_fp16")]; + tensor var_47631_to_fp16 = const()[name = tensor("op_47631_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_47632_cast_fp16 = add(x = var_47630_cast_fp16, y = var_47631_to_fp16)[name = tensor("op_47632_cast_fp16")]; + tensor denom_117_epsilon_0 = const()[name = tensor("denom_117_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_117_cast_fp16 = rsqrt(epsilon = denom_117_epsilon_0, x = var_47632_cast_fp16)[name = tensor("denom_117_cast_fp16")]; + tensor out_117_cast_fp16 = mul(x = zero_mean_117_cast_fp16, y = denom_117_cast_fp16)[name = tensor("out_117_cast_fp16")]; + tensor obj_117_gamma_0_to_fp16 = const()[name = tensor("obj_117_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321731840)))]; + tensor obj_117_beta_0_to_fp16 = const()[name = tensor("obj_117_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321734464)))]; + tensor obj_117_epsilon_0_to_fp16 = const()[name = tensor("obj_117_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_117_cast_fp16 = batch_norm(beta = obj_117_beta_0_to_fp16, epsilon = obj_117_epsilon_0_to_fp16, gamma = obj_117_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_117_cast_fp16)[name = tensor("obj_117_cast_fp16")]; + tensor var_47650 = const()[name = tensor("op_47650"), val = tensor([1, 1])]; + tensor var_47652 = const()[name = tensor("op_47652"), val = tensor([1, 1])]; + tensor pretrained_out_349_pad_type_0 = const()[name = tensor("pretrained_out_349_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_349_pad_0 = const()[name = tensor("pretrained_out_349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(321737088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322556352))), name = tensor("layers_29_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_29_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322556480)))]; + tensor pretrained_out_349_cast_fp16 = conv(bias = layers_29_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_47652, groups = var_47614, pad = pretrained_out_349_pad_0, pad_type = pretrained_out_349_pad_type_0, strides = var_47650, weight = layers_29_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_117_cast_fp16)[name = tensor("pretrained_out_349_cast_fp16")]; + tensor var_47656 = const()[name = tensor("op_47656"), val = tensor([1, 1])]; + tensor var_47658 = const()[name = tensor("op_47658"), val = tensor([1, 1])]; + tensor input_581_pad_type_0 = const()[name = tensor("input_581_pad_type_0"), val = tensor("custom")]; + tensor input_581_pad_0 = const()[name = tensor("input_581_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322559104)))]; + tensor input_581_cast_fp16 = conv(dilations = var_47658, groups = var_47614, pad = input_581_pad_0, pad_type = input_581_pad_type_0, strides = var_47656, weight = layers_29_self_attn_q_proj_loraA_weight_to_fp16, x = obj_117_cast_fp16)[name = tensor("input_581_cast_fp16")]; + tensor var_47662 = const()[name = tensor("op_47662"), val = tensor([1, 1])]; + tensor var_47664 = const()[name = tensor("op_47664"), val = tensor([1, 1])]; + tensor lora_out_697_pad_type_0 = const()[name = tensor("lora_out_697_pad_type_0"), val = tensor("custom")]; + tensor lora_out_697_pad_0 = const()[name = tensor("lora_out_697_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_699_weight_0_to_fp16 = const()[name = tensor("lora_out_699_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322600128)))]; + tensor lora_out_699_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_47664, groups = var_47614, pad = lora_out_697_pad_0, pad_type = lora_out_697_pad_type_0, strides = var_47662, weight = lora_out_699_weight_0_to_fp16, x = input_581_cast_fp16)[name = tensor("lora_out_699_cast_fp16")]; + tensor query_59_cast_fp16 = add(x = pretrained_out_349_cast_fp16, y = lora_out_699_cast_fp16)[name = tensor("query_59_cast_fp16")]; + tensor var_47674 = const()[name = tensor("op_47674"), val = tensor([1, 1])]; + tensor var_47676 = const()[name = tensor("op_47676"), val = tensor([1, 1])]; + tensor pretrained_out_351_pad_type_0 = const()[name = tensor("pretrained_out_351_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_351_pad_0 = const()[name = tensor("pretrained_out_351_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322641152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323460416))), name = tensor("layers_29_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_351_cast_fp16 = conv(dilations = var_47676, groups = var_47614, pad = pretrained_out_351_pad_0, pad_type = pretrained_out_351_pad_type_0, strides = var_47674, weight = layers_29_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_117_cast_fp16)[name = tensor("pretrained_out_351_cast_fp16")]; + tensor var_47680 = const()[name = tensor("op_47680"), val = tensor([1, 1])]; + tensor var_47682 = const()[name = tensor("op_47682"), val = tensor([1, 1])]; + tensor input_583_pad_type_0 = const()[name = tensor("input_583_pad_type_0"), val = tensor("custom")]; + tensor input_583_pad_0 = const()[name = tensor("input_583_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323460544)))]; + tensor input_583_cast_fp16 = conv(dilations = var_47682, groups = var_47614, pad = input_583_pad_0, pad_type = input_583_pad_type_0, strides = var_47680, weight = layers_29_self_attn_k_proj_loraA_weight_to_fp16, x = obj_117_cast_fp16)[name = tensor("input_583_cast_fp16")]; + tensor var_47686 = const()[name = tensor("op_47686"), val = tensor([1, 1])]; + tensor var_47688 = const()[name = tensor("op_47688"), val = tensor([1, 1])]; + tensor lora_out_701_pad_type_0 = const()[name = tensor("lora_out_701_pad_type_0"), val = tensor("custom")]; + tensor lora_out_701_pad_0 = const()[name = tensor("lora_out_701_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_703_weight_0_to_fp16 = const()[name = tensor("lora_out_703_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323501568)))]; + tensor lora_out_703_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_47688, groups = var_47614, pad = lora_out_701_pad_0, pad_type = lora_out_701_pad_type_0, strides = var_47686, weight = lora_out_703_weight_0_to_fp16, x = input_583_cast_fp16)[name = tensor("lora_out_703_cast_fp16")]; + tensor key_59_cast_fp16 = add(x = pretrained_out_351_cast_fp16, y = lora_out_703_cast_fp16)[name = tensor("key_59_cast_fp16")]; + tensor var_47699 = const()[name = tensor("op_47699"), val = tensor([1, 1])]; + tensor var_47701 = const()[name = tensor("op_47701"), val = tensor([1, 1])]; + tensor pretrained_out_353_pad_type_0 = const()[name = tensor("pretrained_out_353_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_353_pad_0 = const()[name = tensor("pretrained_out_353_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323542592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324361856))), name = tensor("layers_29_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_29_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324361984)))]; + tensor pretrained_out_353_cast_fp16 = conv(bias = layers_29_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_47701, groups = var_47614, pad = pretrained_out_353_pad_0, pad_type = pretrained_out_353_pad_type_0, strides = var_47699, weight = layers_29_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_117_cast_fp16)[name = tensor("pretrained_out_353_cast_fp16")]; + tensor var_47705 = const()[name = tensor("op_47705"), val = tensor([1, 1])]; + tensor var_47707 = const()[name = tensor("op_47707"), val = tensor([1, 1])]; + tensor input_585_pad_type_0 = const()[name = tensor("input_585_pad_type_0"), val = tensor("custom")]; + tensor input_585_pad_0 = const()[name = tensor("input_585_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324364608)))]; + tensor input_585_cast_fp16 = conv(dilations = var_47707, groups = var_47614, pad = input_585_pad_0, pad_type = input_585_pad_type_0, strides = var_47705, weight = layers_29_self_attn_v_proj_loraA_weight_to_fp16, x = obj_117_cast_fp16)[name = tensor("input_585_cast_fp16")]; + tensor var_47711 = const()[name = tensor("op_47711"), val = tensor([1, 1])]; + tensor var_47713 = const()[name = tensor("op_47713"), val = tensor([1, 1])]; + tensor lora_out_705_pad_type_0 = const()[name = tensor("lora_out_705_pad_type_0"), val = tensor("custom")]; + tensor lora_out_705_pad_0 = const()[name = tensor("lora_out_705_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_707_weight_0_to_fp16 = const()[name = tensor("lora_out_707_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324405632)))]; + tensor lora_out_707_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_47713, groups = var_47614, pad = lora_out_705_pad_0, pad_type = lora_out_705_pad_type_0, strides = var_47711, weight = lora_out_707_weight_0_to_fp16, x = input_585_cast_fp16)[name = tensor("lora_out_707_cast_fp16")]; + tensor value_59_cast_fp16 = add(x = pretrained_out_353_cast_fp16, y = lora_out_707_cast_fp16)[name = tensor("value_59_cast_fp16")]; + tensor var_47723_begin_0 = const()[name = tensor("op_47723_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47723_end_0 = const()[name = tensor("op_47723_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47723_end_mask_0 = const()[name = tensor("op_47723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47723_cast_fp16 = slice_by_index(begin = var_47723_begin_0, end = var_47723_end_0, end_mask = var_47723_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47723_cast_fp16")]; + tensor var_47727_begin_0 = const()[name = tensor("op_47727_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_47727_end_0 = const()[name = tensor("op_47727_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_47727_end_mask_0 = const()[name = tensor("op_47727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47727_cast_fp16 = slice_by_index(begin = var_47727_begin_0, end = var_47727_end_0, end_mask = var_47727_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47727_cast_fp16")]; + tensor var_47731_begin_0 = const()[name = tensor("op_47731_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_47731_end_0 = const()[name = tensor("op_47731_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_47731_end_mask_0 = const()[name = tensor("op_47731_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47731_cast_fp16 = slice_by_index(begin = var_47731_begin_0, end = var_47731_end_0, end_mask = var_47731_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47731_cast_fp16")]; + tensor var_47735_begin_0 = const()[name = tensor("op_47735_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_47735_end_0 = const()[name = tensor("op_47735_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_47735_end_mask_0 = const()[name = tensor("op_47735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47735_cast_fp16 = slice_by_index(begin = var_47735_begin_0, end = var_47735_end_0, end_mask = var_47735_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47735_cast_fp16")]; + tensor var_47739_begin_0 = const()[name = tensor("op_47739_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_47739_end_0 = const()[name = tensor("op_47739_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_47739_end_mask_0 = const()[name = tensor("op_47739_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47739_cast_fp16 = slice_by_index(begin = var_47739_begin_0, end = var_47739_end_0, end_mask = var_47739_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47739_cast_fp16")]; + tensor var_47743_begin_0 = const()[name = tensor("op_47743_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_47743_end_0 = const()[name = tensor("op_47743_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_47743_end_mask_0 = const()[name = tensor("op_47743_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47743_cast_fp16 = slice_by_index(begin = var_47743_begin_0, end = var_47743_end_0, end_mask = var_47743_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47743_cast_fp16")]; + tensor var_47747_begin_0 = const()[name = tensor("op_47747_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_47747_end_0 = const()[name = tensor("op_47747_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_47747_end_mask_0 = const()[name = tensor("op_47747_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47747_cast_fp16 = slice_by_index(begin = var_47747_begin_0, end = var_47747_end_0, end_mask = var_47747_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47747_cast_fp16")]; + tensor var_47751_begin_0 = const()[name = tensor("op_47751_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_47751_end_0 = const()[name = tensor("op_47751_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_47751_end_mask_0 = const()[name = tensor("op_47751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47751_cast_fp16 = slice_by_index(begin = var_47751_begin_0, end = var_47751_end_0, end_mask = var_47751_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47751_cast_fp16")]; + tensor var_47755_begin_0 = const()[name = tensor("op_47755_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_47755_end_0 = const()[name = tensor("op_47755_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_47755_end_mask_0 = const()[name = tensor("op_47755_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47755_cast_fp16 = slice_by_index(begin = var_47755_begin_0, end = var_47755_end_0, end_mask = var_47755_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47755_cast_fp16")]; + tensor var_47759_begin_0 = const()[name = tensor("op_47759_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_47759_end_0 = const()[name = tensor("op_47759_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_47759_end_mask_0 = const()[name = tensor("op_47759_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47759_cast_fp16 = slice_by_index(begin = var_47759_begin_0, end = var_47759_end_0, end_mask = var_47759_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47759_cast_fp16")]; + tensor var_47763_begin_0 = const()[name = tensor("op_47763_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_47763_end_0 = const()[name = tensor("op_47763_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_47763_end_mask_0 = const()[name = tensor("op_47763_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47763_cast_fp16 = slice_by_index(begin = var_47763_begin_0, end = var_47763_end_0, end_mask = var_47763_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47763_cast_fp16")]; + tensor var_47767_begin_0 = const()[name = tensor("op_47767_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_47767_end_0 = const()[name = tensor("op_47767_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_47767_end_mask_0 = const()[name = tensor("op_47767_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47767_cast_fp16 = slice_by_index(begin = var_47767_begin_0, end = var_47767_end_0, end_mask = var_47767_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47767_cast_fp16")]; + tensor var_47771_begin_0 = const()[name = tensor("op_47771_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_47771_end_0 = const()[name = tensor("op_47771_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_47771_end_mask_0 = const()[name = tensor("op_47771_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47771_cast_fp16 = slice_by_index(begin = var_47771_begin_0, end = var_47771_end_0, end_mask = var_47771_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47771_cast_fp16")]; + tensor var_47775_begin_0 = const()[name = tensor("op_47775_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_47775_end_0 = const()[name = tensor("op_47775_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_47775_end_mask_0 = const()[name = tensor("op_47775_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47775_cast_fp16 = slice_by_index(begin = var_47775_begin_0, end = var_47775_end_0, end_mask = var_47775_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47775_cast_fp16")]; + tensor var_47779_begin_0 = const()[name = tensor("op_47779_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_47779_end_0 = const()[name = tensor("op_47779_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_47779_end_mask_0 = const()[name = tensor("op_47779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47779_cast_fp16 = slice_by_index(begin = var_47779_begin_0, end = var_47779_end_0, end_mask = var_47779_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47779_cast_fp16")]; + tensor var_47783_begin_0 = const()[name = tensor("op_47783_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_47783_end_0 = const()[name = tensor("op_47783_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_47783_end_mask_0 = const()[name = tensor("op_47783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47783_cast_fp16 = slice_by_index(begin = var_47783_begin_0, end = var_47783_end_0, end_mask = var_47783_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47783_cast_fp16")]; + tensor var_47787_begin_0 = const()[name = tensor("op_47787_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_47787_end_0 = const()[name = tensor("op_47787_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_47787_end_mask_0 = const()[name = tensor("op_47787_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47787_cast_fp16 = slice_by_index(begin = var_47787_begin_0, end = var_47787_end_0, end_mask = var_47787_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47787_cast_fp16")]; + tensor var_47791_begin_0 = const()[name = tensor("op_47791_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_47791_end_0 = const()[name = tensor("op_47791_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_47791_end_mask_0 = const()[name = tensor("op_47791_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47791_cast_fp16 = slice_by_index(begin = var_47791_begin_0, end = var_47791_end_0, end_mask = var_47791_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47791_cast_fp16")]; + tensor var_47795_begin_0 = const()[name = tensor("op_47795_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_47795_end_0 = const()[name = tensor("op_47795_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_47795_end_mask_0 = const()[name = tensor("op_47795_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47795_cast_fp16 = slice_by_index(begin = var_47795_begin_0, end = var_47795_end_0, end_mask = var_47795_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47795_cast_fp16")]; + tensor var_47799_begin_0 = const()[name = tensor("op_47799_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_47799_end_0 = const()[name = tensor("op_47799_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_47799_end_mask_0 = const()[name = tensor("op_47799_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_47799_cast_fp16 = slice_by_index(begin = var_47799_begin_0, end = var_47799_end_0, end_mask = var_47799_end_mask_0, x = query_59_cast_fp16)[name = tensor("op_47799_cast_fp16")]; + tensor var_47808_begin_0 = const()[name = tensor("op_47808_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47808_end_0 = const()[name = tensor("op_47808_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47808_end_mask_0 = const()[name = tensor("op_47808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47808_cast_fp16 = slice_by_index(begin = var_47808_begin_0, end = var_47808_end_0, end_mask = var_47808_end_mask_0, x = var_47723_cast_fp16)[name = tensor("op_47808_cast_fp16")]; + tensor var_47815_begin_0 = const()[name = tensor("op_47815_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47815_end_0 = const()[name = tensor("op_47815_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47815_end_mask_0 = const()[name = tensor("op_47815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47815_cast_fp16 = slice_by_index(begin = var_47815_begin_0, end = var_47815_end_0, end_mask = var_47815_end_mask_0, x = var_47723_cast_fp16)[name = tensor("op_47815_cast_fp16")]; + tensor var_47822_begin_0 = const()[name = tensor("op_47822_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47822_end_0 = const()[name = tensor("op_47822_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47822_end_mask_0 = const()[name = tensor("op_47822_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47822_cast_fp16 = slice_by_index(begin = var_47822_begin_0, end = var_47822_end_0, end_mask = var_47822_end_mask_0, x = var_47723_cast_fp16)[name = tensor("op_47822_cast_fp16")]; + tensor var_47829_begin_0 = const()[name = tensor("op_47829_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47829_end_0 = const()[name = tensor("op_47829_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47829_end_mask_0 = const()[name = tensor("op_47829_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47829_cast_fp16 = slice_by_index(begin = var_47829_begin_0, end = var_47829_end_0, end_mask = var_47829_end_mask_0, x = var_47723_cast_fp16)[name = tensor("op_47829_cast_fp16")]; + tensor var_47836_begin_0 = const()[name = tensor("op_47836_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47836_end_0 = const()[name = tensor("op_47836_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47836_end_mask_0 = const()[name = tensor("op_47836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47836_cast_fp16 = slice_by_index(begin = var_47836_begin_0, end = var_47836_end_0, end_mask = var_47836_end_mask_0, x = var_47727_cast_fp16)[name = tensor("op_47836_cast_fp16")]; + tensor var_47843_begin_0 = const()[name = tensor("op_47843_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47843_end_0 = const()[name = tensor("op_47843_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47843_end_mask_0 = const()[name = tensor("op_47843_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47843_cast_fp16 = slice_by_index(begin = var_47843_begin_0, end = var_47843_end_0, end_mask = var_47843_end_mask_0, x = var_47727_cast_fp16)[name = tensor("op_47843_cast_fp16")]; + tensor var_47850_begin_0 = const()[name = tensor("op_47850_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47850_end_0 = const()[name = tensor("op_47850_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47850_end_mask_0 = const()[name = tensor("op_47850_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47850_cast_fp16 = slice_by_index(begin = var_47850_begin_0, end = var_47850_end_0, end_mask = var_47850_end_mask_0, x = var_47727_cast_fp16)[name = tensor("op_47850_cast_fp16")]; + tensor var_47857_begin_0 = const()[name = tensor("op_47857_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47857_end_0 = const()[name = tensor("op_47857_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47857_end_mask_0 = const()[name = tensor("op_47857_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47857_cast_fp16 = slice_by_index(begin = var_47857_begin_0, end = var_47857_end_0, end_mask = var_47857_end_mask_0, x = var_47727_cast_fp16)[name = tensor("op_47857_cast_fp16")]; + tensor var_47864_begin_0 = const()[name = tensor("op_47864_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47864_end_0 = const()[name = tensor("op_47864_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47864_end_mask_0 = const()[name = tensor("op_47864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47864_cast_fp16 = slice_by_index(begin = var_47864_begin_0, end = var_47864_end_0, end_mask = var_47864_end_mask_0, x = var_47731_cast_fp16)[name = tensor("op_47864_cast_fp16")]; + tensor var_47871_begin_0 = const()[name = tensor("op_47871_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47871_end_0 = const()[name = tensor("op_47871_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47871_end_mask_0 = const()[name = tensor("op_47871_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47871_cast_fp16 = slice_by_index(begin = var_47871_begin_0, end = var_47871_end_0, end_mask = var_47871_end_mask_0, x = var_47731_cast_fp16)[name = tensor("op_47871_cast_fp16")]; + tensor var_47878_begin_0 = const()[name = tensor("op_47878_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47878_end_0 = const()[name = tensor("op_47878_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47878_end_mask_0 = const()[name = tensor("op_47878_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47878_cast_fp16 = slice_by_index(begin = var_47878_begin_0, end = var_47878_end_0, end_mask = var_47878_end_mask_0, x = var_47731_cast_fp16)[name = tensor("op_47878_cast_fp16")]; + tensor var_47885_begin_0 = const()[name = tensor("op_47885_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47885_end_0 = const()[name = tensor("op_47885_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47885_end_mask_0 = const()[name = tensor("op_47885_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47885_cast_fp16 = slice_by_index(begin = var_47885_begin_0, end = var_47885_end_0, end_mask = var_47885_end_mask_0, x = var_47731_cast_fp16)[name = tensor("op_47885_cast_fp16")]; + tensor var_47892_begin_0 = const()[name = tensor("op_47892_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47892_end_0 = const()[name = tensor("op_47892_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47892_end_mask_0 = const()[name = tensor("op_47892_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47892_cast_fp16 = slice_by_index(begin = var_47892_begin_0, end = var_47892_end_0, end_mask = var_47892_end_mask_0, x = var_47735_cast_fp16)[name = tensor("op_47892_cast_fp16")]; + tensor var_47899_begin_0 = const()[name = tensor("op_47899_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47899_end_0 = const()[name = tensor("op_47899_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47899_end_mask_0 = const()[name = tensor("op_47899_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47899_cast_fp16 = slice_by_index(begin = var_47899_begin_0, end = var_47899_end_0, end_mask = var_47899_end_mask_0, x = var_47735_cast_fp16)[name = tensor("op_47899_cast_fp16")]; + tensor var_47906_begin_0 = const()[name = tensor("op_47906_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47906_end_0 = const()[name = tensor("op_47906_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47906_end_mask_0 = const()[name = tensor("op_47906_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47906_cast_fp16 = slice_by_index(begin = var_47906_begin_0, end = var_47906_end_0, end_mask = var_47906_end_mask_0, x = var_47735_cast_fp16)[name = tensor("op_47906_cast_fp16")]; + tensor var_47913_begin_0 = const()[name = tensor("op_47913_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47913_end_0 = const()[name = tensor("op_47913_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47913_end_mask_0 = const()[name = tensor("op_47913_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47913_cast_fp16 = slice_by_index(begin = var_47913_begin_0, end = var_47913_end_0, end_mask = var_47913_end_mask_0, x = var_47735_cast_fp16)[name = tensor("op_47913_cast_fp16")]; + tensor var_47920_begin_0 = const()[name = tensor("op_47920_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47920_end_0 = const()[name = tensor("op_47920_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47920_end_mask_0 = const()[name = tensor("op_47920_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47920_cast_fp16 = slice_by_index(begin = var_47920_begin_0, end = var_47920_end_0, end_mask = var_47920_end_mask_0, x = var_47739_cast_fp16)[name = tensor("op_47920_cast_fp16")]; + tensor var_47927_begin_0 = const()[name = tensor("op_47927_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47927_end_0 = const()[name = tensor("op_47927_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47927_end_mask_0 = const()[name = tensor("op_47927_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47927_cast_fp16 = slice_by_index(begin = var_47927_begin_0, end = var_47927_end_0, end_mask = var_47927_end_mask_0, x = var_47739_cast_fp16)[name = tensor("op_47927_cast_fp16")]; + tensor var_47934_begin_0 = const()[name = tensor("op_47934_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47934_end_0 = const()[name = tensor("op_47934_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47934_end_mask_0 = const()[name = tensor("op_47934_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47934_cast_fp16 = slice_by_index(begin = var_47934_begin_0, end = var_47934_end_0, end_mask = var_47934_end_mask_0, x = var_47739_cast_fp16)[name = tensor("op_47934_cast_fp16")]; + tensor var_47941_begin_0 = const()[name = tensor("op_47941_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47941_end_0 = const()[name = tensor("op_47941_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47941_end_mask_0 = const()[name = tensor("op_47941_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47941_cast_fp16 = slice_by_index(begin = var_47941_begin_0, end = var_47941_end_0, end_mask = var_47941_end_mask_0, x = var_47739_cast_fp16)[name = tensor("op_47941_cast_fp16")]; + tensor var_47948_begin_0 = const()[name = tensor("op_47948_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47948_end_0 = const()[name = tensor("op_47948_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47948_end_mask_0 = const()[name = tensor("op_47948_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47948_cast_fp16 = slice_by_index(begin = var_47948_begin_0, end = var_47948_end_0, end_mask = var_47948_end_mask_0, x = var_47743_cast_fp16)[name = tensor("op_47948_cast_fp16")]; + tensor var_47955_begin_0 = const()[name = tensor("op_47955_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47955_end_0 = const()[name = tensor("op_47955_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47955_end_mask_0 = const()[name = tensor("op_47955_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47955_cast_fp16 = slice_by_index(begin = var_47955_begin_0, end = var_47955_end_0, end_mask = var_47955_end_mask_0, x = var_47743_cast_fp16)[name = tensor("op_47955_cast_fp16")]; + tensor var_47962_begin_0 = const()[name = tensor("op_47962_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47962_end_0 = const()[name = tensor("op_47962_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47962_end_mask_0 = const()[name = tensor("op_47962_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47962_cast_fp16 = slice_by_index(begin = var_47962_begin_0, end = var_47962_end_0, end_mask = var_47962_end_mask_0, x = var_47743_cast_fp16)[name = tensor("op_47962_cast_fp16")]; + tensor var_47969_begin_0 = const()[name = tensor("op_47969_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47969_end_0 = const()[name = tensor("op_47969_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47969_end_mask_0 = const()[name = tensor("op_47969_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47969_cast_fp16 = slice_by_index(begin = var_47969_begin_0, end = var_47969_end_0, end_mask = var_47969_end_mask_0, x = var_47743_cast_fp16)[name = tensor("op_47969_cast_fp16")]; + tensor var_47976_begin_0 = const()[name = tensor("op_47976_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_47976_end_0 = const()[name = tensor("op_47976_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_47976_end_mask_0 = const()[name = tensor("op_47976_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47976_cast_fp16 = slice_by_index(begin = var_47976_begin_0, end = var_47976_end_0, end_mask = var_47976_end_mask_0, x = var_47747_cast_fp16)[name = tensor("op_47976_cast_fp16")]; + tensor var_47983_begin_0 = const()[name = tensor("op_47983_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_47983_end_0 = const()[name = tensor("op_47983_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_47983_end_mask_0 = const()[name = tensor("op_47983_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47983_cast_fp16 = slice_by_index(begin = var_47983_begin_0, end = var_47983_end_0, end_mask = var_47983_end_mask_0, x = var_47747_cast_fp16)[name = tensor("op_47983_cast_fp16")]; + tensor var_47990_begin_0 = const()[name = tensor("op_47990_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_47990_end_0 = const()[name = tensor("op_47990_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_47990_end_mask_0 = const()[name = tensor("op_47990_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47990_cast_fp16 = slice_by_index(begin = var_47990_begin_0, end = var_47990_end_0, end_mask = var_47990_end_mask_0, x = var_47747_cast_fp16)[name = tensor("op_47990_cast_fp16")]; + tensor var_47997_begin_0 = const()[name = tensor("op_47997_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_47997_end_0 = const()[name = tensor("op_47997_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_47997_end_mask_0 = const()[name = tensor("op_47997_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_47997_cast_fp16 = slice_by_index(begin = var_47997_begin_0, end = var_47997_end_0, end_mask = var_47997_end_mask_0, x = var_47747_cast_fp16)[name = tensor("op_47997_cast_fp16")]; + tensor var_48004_begin_0 = const()[name = tensor("op_48004_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48004_end_0 = const()[name = tensor("op_48004_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48004_end_mask_0 = const()[name = tensor("op_48004_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48004_cast_fp16 = slice_by_index(begin = var_48004_begin_0, end = var_48004_end_0, end_mask = var_48004_end_mask_0, x = var_47751_cast_fp16)[name = tensor("op_48004_cast_fp16")]; + tensor var_48011_begin_0 = const()[name = tensor("op_48011_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48011_end_0 = const()[name = tensor("op_48011_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48011_end_mask_0 = const()[name = tensor("op_48011_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48011_cast_fp16 = slice_by_index(begin = var_48011_begin_0, end = var_48011_end_0, end_mask = var_48011_end_mask_0, x = var_47751_cast_fp16)[name = tensor("op_48011_cast_fp16")]; + tensor var_48018_begin_0 = const()[name = tensor("op_48018_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48018_end_0 = const()[name = tensor("op_48018_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48018_end_mask_0 = const()[name = tensor("op_48018_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48018_cast_fp16 = slice_by_index(begin = var_48018_begin_0, end = var_48018_end_0, end_mask = var_48018_end_mask_0, x = var_47751_cast_fp16)[name = tensor("op_48018_cast_fp16")]; + tensor var_48025_begin_0 = const()[name = tensor("op_48025_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48025_end_0 = const()[name = tensor("op_48025_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48025_end_mask_0 = const()[name = tensor("op_48025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48025_cast_fp16 = slice_by_index(begin = var_48025_begin_0, end = var_48025_end_0, end_mask = var_48025_end_mask_0, x = var_47751_cast_fp16)[name = tensor("op_48025_cast_fp16")]; + tensor var_48032_begin_0 = const()[name = tensor("op_48032_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48032_end_0 = const()[name = tensor("op_48032_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48032_end_mask_0 = const()[name = tensor("op_48032_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48032_cast_fp16 = slice_by_index(begin = var_48032_begin_0, end = var_48032_end_0, end_mask = var_48032_end_mask_0, x = var_47755_cast_fp16)[name = tensor("op_48032_cast_fp16")]; + tensor var_48039_begin_0 = const()[name = tensor("op_48039_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48039_end_0 = const()[name = tensor("op_48039_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48039_end_mask_0 = const()[name = tensor("op_48039_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48039_cast_fp16 = slice_by_index(begin = var_48039_begin_0, end = var_48039_end_0, end_mask = var_48039_end_mask_0, x = var_47755_cast_fp16)[name = tensor("op_48039_cast_fp16")]; + tensor var_48046_begin_0 = const()[name = tensor("op_48046_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48046_end_0 = const()[name = tensor("op_48046_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48046_end_mask_0 = const()[name = tensor("op_48046_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48046_cast_fp16 = slice_by_index(begin = var_48046_begin_0, end = var_48046_end_0, end_mask = var_48046_end_mask_0, x = var_47755_cast_fp16)[name = tensor("op_48046_cast_fp16")]; + tensor var_48053_begin_0 = const()[name = tensor("op_48053_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48053_end_0 = const()[name = tensor("op_48053_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48053_end_mask_0 = const()[name = tensor("op_48053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48053_cast_fp16 = slice_by_index(begin = var_48053_begin_0, end = var_48053_end_0, end_mask = var_48053_end_mask_0, x = var_47755_cast_fp16)[name = tensor("op_48053_cast_fp16")]; + tensor var_48060_begin_0 = const()[name = tensor("op_48060_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48060_end_0 = const()[name = tensor("op_48060_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48060_end_mask_0 = const()[name = tensor("op_48060_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48060_cast_fp16 = slice_by_index(begin = var_48060_begin_0, end = var_48060_end_0, end_mask = var_48060_end_mask_0, x = var_47759_cast_fp16)[name = tensor("op_48060_cast_fp16")]; + tensor var_48067_begin_0 = const()[name = tensor("op_48067_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48067_end_0 = const()[name = tensor("op_48067_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48067_end_mask_0 = const()[name = tensor("op_48067_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48067_cast_fp16 = slice_by_index(begin = var_48067_begin_0, end = var_48067_end_0, end_mask = var_48067_end_mask_0, x = var_47759_cast_fp16)[name = tensor("op_48067_cast_fp16")]; + tensor var_48074_begin_0 = const()[name = tensor("op_48074_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48074_end_0 = const()[name = tensor("op_48074_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48074_end_mask_0 = const()[name = tensor("op_48074_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48074_cast_fp16 = slice_by_index(begin = var_48074_begin_0, end = var_48074_end_0, end_mask = var_48074_end_mask_0, x = var_47759_cast_fp16)[name = tensor("op_48074_cast_fp16")]; + tensor var_48081_begin_0 = const()[name = tensor("op_48081_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48081_end_0 = const()[name = tensor("op_48081_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48081_end_mask_0 = const()[name = tensor("op_48081_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48081_cast_fp16 = slice_by_index(begin = var_48081_begin_0, end = var_48081_end_0, end_mask = var_48081_end_mask_0, x = var_47759_cast_fp16)[name = tensor("op_48081_cast_fp16")]; + tensor var_48088_begin_0 = const()[name = tensor("op_48088_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48088_end_0 = const()[name = tensor("op_48088_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48088_end_mask_0 = const()[name = tensor("op_48088_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48088_cast_fp16 = slice_by_index(begin = var_48088_begin_0, end = var_48088_end_0, end_mask = var_48088_end_mask_0, x = var_47763_cast_fp16)[name = tensor("op_48088_cast_fp16")]; + tensor var_48095_begin_0 = const()[name = tensor("op_48095_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48095_end_0 = const()[name = tensor("op_48095_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48095_end_mask_0 = const()[name = tensor("op_48095_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48095_cast_fp16 = slice_by_index(begin = var_48095_begin_0, end = var_48095_end_0, end_mask = var_48095_end_mask_0, x = var_47763_cast_fp16)[name = tensor("op_48095_cast_fp16")]; + tensor var_48102_begin_0 = const()[name = tensor("op_48102_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48102_end_0 = const()[name = tensor("op_48102_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48102_end_mask_0 = const()[name = tensor("op_48102_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48102_cast_fp16 = slice_by_index(begin = var_48102_begin_0, end = var_48102_end_0, end_mask = var_48102_end_mask_0, x = var_47763_cast_fp16)[name = tensor("op_48102_cast_fp16")]; + tensor var_48109_begin_0 = const()[name = tensor("op_48109_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48109_end_0 = const()[name = tensor("op_48109_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48109_end_mask_0 = const()[name = tensor("op_48109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48109_cast_fp16 = slice_by_index(begin = var_48109_begin_0, end = var_48109_end_0, end_mask = var_48109_end_mask_0, x = var_47763_cast_fp16)[name = tensor("op_48109_cast_fp16")]; + tensor var_48116_begin_0 = const()[name = tensor("op_48116_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48116_end_0 = const()[name = tensor("op_48116_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48116_end_mask_0 = const()[name = tensor("op_48116_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48116_cast_fp16 = slice_by_index(begin = var_48116_begin_0, end = var_48116_end_0, end_mask = var_48116_end_mask_0, x = var_47767_cast_fp16)[name = tensor("op_48116_cast_fp16")]; + tensor var_48123_begin_0 = const()[name = tensor("op_48123_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48123_end_0 = const()[name = tensor("op_48123_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48123_end_mask_0 = const()[name = tensor("op_48123_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48123_cast_fp16 = slice_by_index(begin = var_48123_begin_0, end = var_48123_end_0, end_mask = var_48123_end_mask_0, x = var_47767_cast_fp16)[name = tensor("op_48123_cast_fp16")]; + tensor var_48130_begin_0 = const()[name = tensor("op_48130_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48130_end_0 = const()[name = tensor("op_48130_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48130_end_mask_0 = const()[name = tensor("op_48130_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48130_cast_fp16 = slice_by_index(begin = var_48130_begin_0, end = var_48130_end_0, end_mask = var_48130_end_mask_0, x = var_47767_cast_fp16)[name = tensor("op_48130_cast_fp16")]; + tensor var_48137_begin_0 = const()[name = tensor("op_48137_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48137_end_0 = const()[name = tensor("op_48137_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48137_end_mask_0 = const()[name = tensor("op_48137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48137_cast_fp16 = slice_by_index(begin = var_48137_begin_0, end = var_48137_end_0, end_mask = var_48137_end_mask_0, x = var_47767_cast_fp16)[name = tensor("op_48137_cast_fp16")]; + tensor var_48144_begin_0 = const()[name = tensor("op_48144_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48144_end_0 = const()[name = tensor("op_48144_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48144_end_mask_0 = const()[name = tensor("op_48144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48144_cast_fp16 = slice_by_index(begin = var_48144_begin_0, end = var_48144_end_0, end_mask = var_48144_end_mask_0, x = var_47771_cast_fp16)[name = tensor("op_48144_cast_fp16")]; + tensor var_48151_begin_0 = const()[name = tensor("op_48151_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48151_end_0 = const()[name = tensor("op_48151_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48151_end_mask_0 = const()[name = tensor("op_48151_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48151_cast_fp16 = slice_by_index(begin = var_48151_begin_0, end = var_48151_end_0, end_mask = var_48151_end_mask_0, x = var_47771_cast_fp16)[name = tensor("op_48151_cast_fp16")]; + tensor var_48158_begin_0 = const()[name = tensor("op_48158_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48158_end_0 = const()[name = tensor("op_48158_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48158_end_mask_0 = const()[name = tensor("op_48158_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48158_cast_fp16 = slice_by_index(begin = var_48158_begin_0, end = var_48158_end_0, end_mask = var_48158_end_mask_0, x = var_47771_cast_fp16)[name = tensor("op_48158_cast_fp16")]; + tensor var_48165_begin_0 = const()[name = tensor("op_48165_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48165_end_0 = const()[name = tensor("op_48165_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48165_end_mask_0 = const()[name = tensor("op_48165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48165_cast_fp16 = slice_by_index(begin = var_48165_begin_0, end = var_48165_end_0, end_mask = var_48165_end_mask_0, x = var_47771_cast_fp16)[name = tensor("op_48165_cast_fp16")]; + tensor var_48172_begin_0 = const()[name = tensor("op_48172_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48172_end_0 = const()[name = tensor("op_48172_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48172_end_mask_0 = const()[name = tensor("op_48172_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48172_cast_fp16 = slice_by_index(begin = var_48172_begin_0, end = var_48172_end_0, end_mask = var_48172_end_mask_0, x = var_47775_cast_fp16)[name = tensor("op_48172_cast_fp16")]; + tensor var_48179_begin_0 = const()[name = tensor("op_48179_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48179_end_0 = const()[name = tensor("op_48179_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48179_end_mask_0 = const()[name = tensor("op_48179_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48179_cast_fp16 = slice_by_index(begin = var_48179_begin_0, end = var_48179_end_0, end_mask = var_48179_end_mask_0, x = var_47775_cast_fp16)[name = tensor("op_48179_cast_fp16")]; + tensor var_48186_begin_0 = const()[name = tensor("op_48186_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48186_end_0 = const()[name = tensor("op_48186_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48186_end_mask_0 = const()[name = tensor("op_48186_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48186_cast_fp16 = slice_by_index(begin = var_48186_begin_0, end = var_48186_end_0, end_mask = var_48186_end_mask_0, x = var_47775_cast_fp16)[name = tensor("op_48186_cast_fp16")]; + tensor var_48193_begin_0 = const()[name = tensor("op_48193_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48193_end_0 = const()[name = tensor("op_48193_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48193_end_mask_0 = const()[name = tensor("op_48193_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48193_cast_fp16 = slice_by_index(begin = var_48193_begin_0, end = var_48193_end_0, end_mask = var_48193_end_mask_0, x = var_47775_cast_fp16)[name = tensor("op_48193_cast_fp16")]; + tensor var_48200_begin_0 = const()[name = tensor("op_48200_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48200_end_0 = const()[name = tensor("op_48200_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48200_end_mask_0 = const()[name = tensor("op_48200_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48200_cast_fp16 = slice_by_index(begin = var_48200_begin_0, end = var_48200_end_0, end_mask = var_48200_end_mask_0, x = var_47779_cast_fp16)[name = tensor("op_48200_cast_fp16")]; + tensor var_48207_begin_0 = const()[name = tensor("op_48207_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48207_end_0 = const()[name = tensor("op_48207_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48207_end_mask_0 = const()[name = tensor("op_48207_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48207_cast_fp16 = slice_by_index(begin = var_48207_begin_0, end = var_48207_end_0, end_mask = var_48207_end_mask_0, x = var_47779_cast_fp16)[name = tensor("op_48207_cast_fp16")]; + tensor var_48214_begin_0 = const()[name = tensor("op_48214_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48214_end_0 = const()[name = tensor("op_48214_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48214_end_mask_0 = const()[name = tensor("op_48214_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48214_cast_fp16 = slice_by_index(begin = var_48214_begin_0, end = var_48214_end_0, end_mask = var_48214_end_mask_0, x = var_47779_cast_fp16)[name = tensor("op_48214_cast_fp16")]; + tensor var_48221_begin_0 = const()[name = tensor("op_48221_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48221_end_0 = const()[name = tensor("op_48221_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48221_end_mask_0 = const()[name = tensor("op_48221_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48221_cast_fp16 = slice_by_index(begin = var_48221_begin_0, end = var_48221_end_0, end_mask = var_48221_end_mask_0, x = var_47779_cast_fp16)[name = tensor("op_48221_cast_fp16")]; + tensor var_48228_begin_0 = const()[name = tensor("op_48228_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48228_end_0 = const()[name = tensor("op_48228_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48228_end_mask_0 = const()[name = tensor("op_48228_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48228_cast_fp16 = slice_by_index(begin = var_48228_begin_0, end = var_48228_end_0, end_mask = var_48228_end_mask_0, x = var_47783_cast_fp16)[name = tensor("op_48228_cast_fp16")]; + tensor var_48235_begin_0 = const()[name = tensor("op_48235_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48235_end_0 = const()[name = tensor("op_48235_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48235_end_mask_0 = const()[name = tensor("op_48235_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48235_cast_fp16 = slice_by_index(begin = var_48235_begin_0, end = var_48235_end_0, end_mask = var_48235_end_mask_0, x = var_47783_cast_fp16)[name = tensor("op_48235_cast_fp16")]; + tensor var_48242_begin_0 = const()[name = tensor("op_48242_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48242_end_0 = const()[name = tensor("op_48242_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48242_end_mask_0 = const()[name = tensor("op_48242_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48242_cast_fp16 = slice_by_index(begin = var_48242_begin_0, end = var_48242_end_0, end_mask = var_48242_end_mask_0, x = var_47783_cast_fp16)[name = tensor("op_48242_cast_fp16")]; + tensor var_48249_begin_0 = const()[name = tensor("op_48249_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48249_end_0 = const()[name = tensor("op_48249_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48249_end_mask_0 = const()[name = tensor("op_48249_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48249_cast_fp16 = slice_by_index(begin = var_48249_begin_0, end = var_48249_end_0, end_mask = var_48249_end_mask_0, x = var_47783_cast_fp16)[name = tensor("op_48249_cast_fp16")]; + tensor var_48256_begin_0 = const()[name = tensor("op_48256_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48256_end_0 = const()[name = tensor("op_48256_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48256_end_mask_0 = const()[name = tensor("op_48256_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48256_cast_fp16 = slice_by_index(begin = var_48256_begin_0, end = var_48256_end_0, end_mask = var_48256_end_mask_0, x = var_47787_cast_fp16)[name = tensor("op_48256_cast_fp16")]; + tensor var_48263_begin_0 = const()[name = tensor("op_48263_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48263_end_0 = const()[name = tensor("op_48263_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48263_end_mask_0 = const()[name = tensor("op_48263_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48263_cast_fp16 = slice_by_index(begin = var_48263_begin_0, end = var_48263_end_0, end_mask = var_48263_end_mask_0, x = var_47787_cast_fp16)[name = tensor("op_48263_cast_fp16")]; + tensor var_48270_begin_0 = const()[name = tensor("op_48270_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48270_end_0 = const()[name = tensor("op_48270_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48270_end_mask_0 = const()[name = tensor("op_48270_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48270_cast_fp16 = slice_by_index(begin = var_48270_begin_0, end = var_48270_end_0, end_mask = var_48270_end_mask_0, x = var_47787_cast_fp16)[name = tensor("op_48270_cast_fp16")]; + tensor var_48277_begin_0 = const()[name = tensor("op_48277_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48277_end_0 = const()[name = tensor("op_48277_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48277_end_mask_0 = const()[name = tensor("op_48277_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48277_cast_fp16 = slice_by_index(begin = var_48277_begin_0, end = var_48277_end_0, end_mask = var_48277_end_mask_0, x = var_47787_cast_fp16)[name = tensor("op_48277_cast_fp16")]; + tensor var_48284_begin_0 = const()[name = tensor("op_48284_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48284_end_0 = const()[name = tensor("op_48284_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48284_end_mask_0 = const()[name = tensor("op_48284_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48284_cast_fp16 = slice_by_index(begin = var_48284_begin_0, end = var_48284_end_0, end_mask = var_48284_end_mask_0, x = var_47791_cast_fp16)[name = tensor("op_48284_cast_fp16")]; + tensor var_48291_begin_0 = const()[name = tensor("op_48291_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48291_end_0 = const()[name = tensor("op_48291_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48291_end_mask_0 = const()[name = tensor("op_48291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48291_cast_fp16 = slice_by_index(begin = var_48291_begin_0, end = var_48291_end_0, end_mask = var_48291_end_mask_0, x = var_47791_cast_fp16)[name = tensor("op_48291_cast_fp16")]; + tensor var_48298_begin_0 = const()[name = tensor("op_48298_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48298_end_0 = const()[name = tensor("op_48298_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48298_end_mask_0 = const()[name = tensor("op_48298_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48298_cast_fp16 = slice_by_index(begin = var_48298_begin_0, end = var_48298_end_0, end_mask = var_48298_end_mask_0, x = var_47791_cast_fp16)[name = tensor("op_48298_cast_fp16")]; + tensor var_48305_begin_0 = const()[name = tensor("op_48305_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48305_end_0 = const()[name = tensor("op_48305_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48305_end_mask_0 = const()[name = tensor("op_48305_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48305_cast_fp16 = slice_by_index(begin = var_48305_begin_0, end = var_48305_end_0, end_mask = var_48305_end_mask_0, x = var_47791_cast_fp16)[name = tensor("op_48305_cast_fp16")]; + tensor var_48312_begin_0 = const()[name = tensor("op_48312_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48312_end_0 = const()[name = tensor("op_48312_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48312_end_mask_0 = const()[name = tensor("op_48312_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48312_cast_fp16 = slice_by_index(begin = var_48312_begin_0, end = var_48312_end_0, end_mask = var_48312_end_mask_0, x = var_47795_cast_fp16)[name = tensor("op_48312_cast_fp16")]; + tensor var_48319_begin_0 = const()[name = tensor("op_48319_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48319_end_0 = const()[name = tensor("op_48319_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48319_end_mask_0 = const()[name = tensor("op_48319_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48319_cast_fp16 = slice_by_index(begin = var_48319_begin_0, end = var_48319_end_0, end_mask = var_48319_end_mask_0, x = var_47795_cast_fp16)[name = tensor("op_48319_cast_fp16")]; + tensor var_48326_begin_0 = const()[name = tensor("op_48326_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48326_end_0 = const()[name = tensor("op_48326_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48326_end_mask_0 = const()[name = tensor("op_48326_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48326_cast_fp16 = slice_by_index(begin = var_48326_begin_0, end = var_48326_end_0, end_mask = var_48326_end_mask_0, x = var_47795_cast_fp16)[name = tensor("op_48326_cast_fp16")]; + tensor var_48333_begin_0 = const()[name = tensor("op_48333_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48333_end_0 = const()[name = tensor("op_48333_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48333_end_mask_0 = const()[name = tensor("op_48333_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48333_cast_fp16 = slice_by_index(begin = var_48333_begin_0, end = var_48333_end_0, end_mask = var_48333_end_mask_0, x = var_47795_cast_fp16)[name = tensor("op_48333_cast_fp16")]; + tensor var_48340_begin_0 = const()[name = tensor("op_48340_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48340_end_0 = const()[name = tensor("op_48340_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_48340_end_mask_0 = const()[name = tensor("op_48340_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48340_cast_fp16 = slice_by_index(begin = var_48340_begin_0, end = var_48340_end_0, end_mask = var_48340_end_mask_0, x = var_47799_cast_fp16)[name = tensor("op_48340_cast_fp16")]; + tensor var_48347_begin_0 = const()[name = tensor("op_48347_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_48347_end_0 = const()[name = tensor("op_48347_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_48347_end_mask_0 = const()[name = tensor("op_48347_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48347_cast_fp16 = slice_by_index(begin = var_48347_begin_0, end = var_48347_end_0, end_mask = var_48347_end_mask_0, x = var_47799_cast_fp16)[name = tensor("op_48347_cast_fp16")]; + tensor var_48354_begin_0 = const()[name = tensor("op_48354_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_48354_end_0 = const()[name = tensor("op_48354_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_48354_end_mask_0 = const()[name = tensor("op_48354_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48354_cast_fp16 = slice_by_index(begin = var_48354_begin_0, end = var_48354_end_0, end_mask = var_48354_end_mask_0, x = var_47799_cast_fp16)[name = tensor("op_48354_cast_fp16")]; + tensor var_48361_begin_0 = const()[name = tensor("op_48361_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_48361_end_0 = const()[name = tensor("op_48361_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48361_end_mask_0 = const()[name = tensor("op_48361_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48361_cast_fp16 = slice_by_index(begin = var_48361_begin_0, end = var_48361_end_0, end_mask = var_48361_end_mask_0, x = var_47799_cast_fp16)[name = tensor("op_48361_cast_fp16")]; + tensor k_59_perm_0 = const()[name = tensor("k_59_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_48366_begin_0 = const()[name = tensor("op_48366_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48366_end_0 = const()[name = tensor("op_48366_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_48366_end_mask_0 = const()[name = tensor("op_48366_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_2 = transpose(perm = k_59_perm_0, x = key_59_cast_fp16)[name = tensor("transpose_2")]; + tensor var_48366_cast_fp16 = slice_by_index(begin = var_48366_begin_0, end = var_48366_end_0, end_mask = var_48366_end_mask_0, x = transpose_2)[name = tensor("op_48366_cast_fp16")]; + tensor var_48370_begin_0 = const()[name = tensor("op_48370_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_48370_end_0 = const()[name = tensor("op_48370_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_48370_end_mask_0 = const()[name = tensor("op_48370_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48370_cast_fp16 = slice_by_index(begin = var_48370_begin_0, end = var_48370_end_0, end_mask = var_48370_end_mask_0, x = transpose_2)[name = tensor("op_48370_cast_fp16")]; + tensor var_48374_begin_0 = const()[name = tensor("op_48374_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_48374_end_0 = const()[name = tensor("op_48374_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_48374_end_mask_0 = const()[name = tensor("op_48374_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48374_cast_fp16 = slice_by_index(begin = var_48374_begin_0, end = var_48374_end_0, end_mask = var_48374_end_mask_0, x = transpose_2)[name = tensor("op_48374_cast_fp16")]; + tensor var_48378_begin_0 = const()[name = tensor("op_48378_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_48378_end_0 = const()[name = tensor("op_48378_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_48378_end_mask_0 = const()[name = tensor("op_48378_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48378_cast_fp16 = slice_by_index(begin = var_48378_begin_0, end = var_48378_end_0, end_mask = var_48378_end_mask_0, x = transpose_2)[name = tensor("op_48378_cast_fp16")]; + tensor var_48382_begin_0 = const()[name = tensor("op_48382_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_48382_end_0 = const()[name = tensor("op_48382_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_48382_end_mask_0 = const()[name = tensor("op_48382_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48382_cast_fp16 = slice_by_index(begin = var_48382_begin_0, end = var_48382_end_0, end_mask = var_48382_end_mask_0, x = transpose_2)[name = tensor("op_48382_cast_fp16")]; + tensor var_48386_begin_0 = const()[name = tensor("op_48386_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_48386_end_0 = const()[name = tensor("op_48386_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_48386_end_mask_0 = const()[name = tensor("op_48386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48386_cast_fp16 = slice_by_index(begin = var_48386_begin_0, end = var_48386_end_0, end_mask = var_48386_end_mask_0, x = transpose_2)[name = tensor("op_48386_cast_fp16")]; + tensor var_48390_begin_0 = const()[name = tensor("op_48390_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_48390_end_0 = const()[name = tensor("op_48390_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_48390_end_mask_0 = const()[name = tensor("op_48390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48390_cast_fp16 = slice_by_index(begin = var_48390_begin_0, end = var_48390_end_0, end_mask = var_48390_end_mask_0, x = transpose_2)[name = tensor("op_48390_cast_fp16")]; + tensor var_48394_begin_0 = const()[name = tensor("op_48394_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_48394_end_0 = const()[name = tensor("op_48394_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_48394_end_mask_0 = const()[name = tensor("op_48394_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48394_cast_fp16 = slice_by_index(begin = var_48394_begin_0, end = var_48394_end_0, end_mask = var_48394_end_mask_0, x = transpose_2)[name = tensor("op_48394_cast_fp16")]; + tensor var_48398_begin_0 = const()[name = tensor("op_48398_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_48398_end_0 = const()[name = tensor("op_48398_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_48398_end_mask_0 = const()[name = tensor("op_48398_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48398_cast_fp16 = slice_by_index(begin = var_48398_begin_0, end = var_48398_end_0, end_mask = var_48398_end_mask_0, x = transpose_2)[name = tensor("op_48398_cast_fp16")]; + tensor var_48402_begin_0 = const()[name = tensor("op_48402_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_48402_end_0 = const()[name = tensor("op_48402_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_48402_end_mask_0 = const()[name = tensor("op_48402_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48402_cast_fp16 = slice_by_index(begin = var_48402_begin_0, end = var_48402_end_0, end_mask = var_48402_end_mask_0, x = transpose_2)[name = tensor("op_48402_cast_fp16")]; + tensor var_48406_begin_0 = const()[name = tensor("op_48406_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_48406_end_0 = const()[name = tensor("op_48406_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_48406_end_mask_0 = const()[name = tensor("op_48406_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48406_cast_fp16 = slice_by_index(begin = var_48406_begin_0, end = var_48406_end_0, end_mask = var_48406_end_mask_0, x = transpose_2)[name = tensor("op_48406_cast_fp16")]; + tensor var_48410_begin_0 = const()[name = tensor("op_48410_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_48410_end_0 = const()[name = tensor("op_48410_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_48410_end_mask_0 = const()[name = tensor("op_48410_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48410_cast_fp16 = slice_by_index(begin = var_48410_begin_0, end = var_48410_end_0, end_mask = var_48410_end_mask_0, x = transpose_2)[name = tensor("op_48410_cast_fp16")]; + tensor var_48414_begin_0 = const()[name = tensor("op_48414_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_48414_end_0 = const()[name = tensor("op_48414_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_48414_end_mask_0 = const()[name = tensor("op_48414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48414_cast_fp16 = slice_by_index(begin = var_48414_begin_0, end = var_48414_end_0, end_mask = var_48414_end_mask_0, x = transpose_2)[name = tensor("op_48414_cast_fp16")]; + tensor var_48418_begin_0 = const()[name = tensor("op_48418_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_48418_end_0 = const()[name = tensor("op_48418_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_48418_end_mask_0 = const()[name = tensor("op_48418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48418_cast_fp16 = slice_by_index(begin = var_48418_begin_0, end = var_48418_end_0, end_mask = var_48418_end_mask_0, x = transpose_2)[name = tensor("op_48418_cast_fp16")]; + tensor var_48422_begin_0 = const()[name = tensor("op_48422_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_48422_end_0 = const()[name = tensor("op_48422_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_48422_end_mask_0 = const()[name = tensor("op_48422_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48422_cast_fp16 = slice_by_index(begin = var_48422_begin_0, end = var_48422_end_0, end_mask = var_48422_end_mask_0, x = transpose_2)[name = tensor("op_48422_cast_fp16")]; + tensor var_48426_begin_0 = const()[name = tensor("op_48426_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_48426_end_0 = const()[name = tensor("op_48426_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_48426_end_mask_0 = const()[name = tensor("op_48426_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48426_cast_fp16 = slice_by_index(begin = var_48426_begin_0, end = var_48426_end_0, end_mask = var_48426_end_mask_0, x = transpose_2)[name = tensor("op_48426_cast_fp16")]; + tensor var_48430_begin_0 = const()[name = tensor("op_48430_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_48430_end_0 = const()[name = tensor("op_48430_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_48430_end_mask_0 = const()[name = tensor("op_48430_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48430_cast_fp16 = slice_by_index(begin = var_48430_begin_0, end = var_48430_end_0, end_mask = var_48430_end_mask_0, x = transpose_2)[name = tensor("op_48430_cast_fp16")]; + tensor var_48434_begin_0 = const()[name = tensor("op_48434_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_48434_end_0 = const()[name = tensor("op_48434_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_48434_end_mask_0 = const()[name = tensor("op_48434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48434_cast_fp16 = slice_by_index(begin = var_48434_begin_0, end = var_48434_end_0, end_mask = var_48434_end_mask_0, x = transpose_2)[name = tensor("op_48434_cast_fp16")]; + tensor var_48438_begin_0 = const()[name = tensor("op_48438_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_48438_end_0 = const()[name = tensor("op_48438_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_48438_end_mask_0 = const()[name = tensor("op_48438_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48438_cast_fp16 = slice_by_index(begin = var_48438_begin_0, end = var_48438_end_0, end_mask = var_48438_end_mask_0, x = transpose_2)[name = tensor("op_48438_cast_fp16")]; + tensor var_48442_begin_0 = const()[name = tensor("op_48442_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_48442_end_0 = const()[name = tensor("op_48442_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_48442_end_mask_0 = const()[name = tensor("op_48442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_48442_cast_fp16 = slice_by_index(begin = var_48442_begin_0, end = var_48442_end_0, end_mask = var_48442_end_mask_0, x = transpose_2)[name = tensor("op_48442_cast_fp16")]; + tensor var_48444_begin_0 = const()[name = tensor("op_48444_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_48444_end_0 = const()[name = tensor("op_48444_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_48444_end_mask_0 = const()[name = tensor("op_48444_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48444_cast_fp16 = slice_by_index(begin = var_48444_begin_0, end = var_48444_end_0, end_mask = var_48444_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48444_cast_fp16")]; + tensor var_48448_begin_0 = const()[name = tensor("op_48448_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_48448_end_0 = const()[name = tensor("op_48448_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_48448_end_mask_0 = const()[name = tensor("op_48448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48448_cast_fp16 = slice_by_index(begin = var_48448_begin_0, end = var_48448_end_0, end_mask = var_48448_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48448_cast_fp16")]; + tensor var_48452_begin_0 = const()[name = tensor("op_48452_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_48452_end_0 = const()[name = tensor("op_48452_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_48452_end_mask_0 = const()[name = tensor("op_48452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48452_cast_fp16 = slice_by_index(begin = var_48452_begin_0, end = var_48452_end_0, end_mask = var_48452_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48452_cast_fp16")]; + tensor var_48456_begin_0 = const()[name = tensor("op_48456_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_48456_end_0 = const()[name = tensor("op_48456_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_48456_end_mask_0 = const()[name = tensor("op_48456_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48456_cast_fp16 = slice_by_index(begin = var_48456_begin_0, end = var_48456_end_0, end_mask = var_48456_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48456_cast_fp16")]; + tensor var_48460_begin_0 = const()[name = tensor("op_48460_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_48460_end_0 = const()[name = tensor("op_48460_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_48460_end_mask_0 = const()[name = tensor("op_48460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48460_cast_fp16 = slice_by_index(begin = var_48460_begin_0, end = var_48460_end_0, end_mask = var_48460_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48460_cast_fp16")]; + tensor var_48464_begin_0 = const()[name = tensor("op_48464_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_48464_end_0 = const()[name = tensor("op_48464_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_48464_end_mask_0 = const()[name = tensor("op_48464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48464_cast_fp16 = slice_by_index(begin = var_48464_begin_0, end = var_48464_end_0, end_mask = var_48464_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48464_cast_fp16")]; + tensor var_48468_begin_0 = const()[name = tensor("op_48468_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_48468_end_0 = const()[name = tensor("op_48468_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_48468_end_mask_0 = const()[name = tensor("op_48468_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48468_cast_fp16 = slice_by_index(begin = var_48468_begin_0, end = var_48468_end_0, end_mask = var_48468_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48468_cast_fp16")]; + tensor var_48472_begin_0 = const()[name = tensor("op_48472_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_48472_end_0 = const()[name = tensor("op_48472_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_48472_end_mask_0 = const()[name = tensor("op_48472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48472_cast_fp16 = slice_by_index(begin = var_48472_begin_0, end = var_48472_end_0, end_mask = var_48472_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48472_cast_fp16")]; + tensor var_48476_begin_0 = const()[name = tensor("op_48476_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_48476_end_0 = const()[name = tensor("op_48476_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_48476_end_mask_0 = const()[name = tensor("op_48476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48476_cast_fp16 = slice_by_index(begin = var_48476_begin_0, end = var_48476_end_0, end_mask = var_48476_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48476_cast_fp16")]; + tensor var_48480_begin_0 = const()[name = tensor("op_48480_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_48480_end_0 = const()[name = tensor("op_48480_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_48480_end_mask_0 = const()[name = tensor("op_48480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48480_cast_fp16 = slice_by_index(begin = var_48480_begin_0, end = var_48480_end_0, end_mask = var_48480_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48480_cast_fp16")]; + tensor var_48484_begin_0 = const()[name = tensor("op_48484_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_48484_end_0 = const()[name = tensor("op_48484_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_48484_end_mask_0 = const()[name = tensor("op_48484_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48484_cast_fp16 = slice_by_index(begin = var_48484_begin_0, end = var_48484_end_0, end_mask = var_48484_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48484_cast_fp16")]; + tensor var_48488_begin_0 = const()[name = tensor("op_48488_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_48488_end_0 = const()[name = tensor("op_48488_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_48488_end_mask_0 = const()[name = tensor("op_48488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48488_cast_fp16 = slice_by_index(begin = var_48488_begin_0, end = var_48488_end_0, end_mask = var_48488_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48488_cast_fp16")]; + tensor var_48492_begin_0 = const()[name = tensor("op_48492_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_48492_end_0 = const()[name = tensor("op_48492_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_48492_end_mask_0 = const()[name = tensor("op_48492_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48492_cast_fp16 = slice_by_index(begin = var_48492_begin_0, end = var_48492_end_0, end_mask = var_48492_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48492_cast_fp16")]; + tensor var_48496_begin_0 = const()[name = tensor("op_48496_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_48496_end_0 = const()[name = tensor("op_48496_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_48496_end_mask_0 = const()[name = tensor("op_48496_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48496_cast_fp16 = slice_by_index(begin = var_48496_begin_0, end = var_48496_end_0, end_mask = var_48496_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48496_cast_fp16")]; + tensor var_48500_begin_0 = const()[name = tensor("op_48500_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_48500_end_0 = const()[name = tensor("op_48500_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_48500_end_mask_0 = const()[name = tensor("op_48500_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48500_cast_fp16 = slice_by_index(begin = var_48500_begin_0, end = var_48500_end_0, end_mask = var_48500_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48500_cast_fp16")]; + tensor var_48504_begin_0 = const()[name = tensor("op_48504_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_48504_end_0 = const()[name = tensor("op_48504_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_48504_end_mask_0 = const()[name = tensor("op_48504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48504_cast_fp16 = slice_by_index(begin = var_48504_begin_0, end = var_48504_end_0, end_mask = var_48504_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48504_cast_fp16")]; + tensor var_48508_begin_0 = const()[name = tensor("op_48508_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_48508_end_0 = const()[name = tensor("op_48508_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_48508_end_mask_0 = const()[name = tensor("op_48508_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48508_cast_fp16 = slice_by_index(begin = var_48508_begin_0, end = var_48508_end_0, end_mask = var_48508_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48508_cast_fp16")]; + tensor var_48512_begin_0 = const()[name = tensor("op_48512_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_48512_end_0 = const()[name = tensor("op_48512_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_48512_end_mask_0 = const()[name = tensor("op_48512_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48512_cast_fp16 = slice_by_index(begin = var_48512_begin_0, end = var_48512_end_0, end_mask = var_48512_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48512_cast_fp16")]; + tensor var_48516_begin_0 = const()[name = tensor("op_48516_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_48516_end_0 = const()[name = tensor("op_48516_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_48516_end_mask_0 = const()[name = tensor("op_48516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48516_cast_fp16 = slice_by_index(begin = var_48516_begin_0, end = var_48516_end_0, end_mask = var_48516_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48516_cast_fp16")]; + tensor var_48520_begin_0 = const()[name = tensor("op_48520_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_48520_end_0 = const()[name = tensor("op_48520_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_48520_end_mask_0 = const()[name = tensor("op_48520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_48520_cast_fp16 = slice_by_index(begin = var_48520_begin_0, end = var_48520_end_0, end_mask = var_48520_end_mask_0, x = value_59_cast_fp16)[name = tensor("op_48520_cast_fp16")]; + tensor var_48524_equation_0 = const()[name = tensor("op_48524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48524_cast_fp16 = einsum(equation = var_48524_equation_0, values = (var_48366_cast_fp16, var_47808_cast_fp16))[name = tensor("op_48524_cast_fp16")]; + tensor var_48525_to_fp16 = const()[name = tensor("op_48525_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4641_cast_fp16 = mul(x = var_48524_cast_fp16, y = var_48525_to_fp16)[name = tensor("aw_chunk_4641_cast_fp16")]; + tensor var_48528_equation_0 = const()[name = tensor("op_48528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48528_cast_fp16 = einsum(equation = var_48528_equation_0, values = (var_48366_cast_fp16, var_47815_cast_fp16))[name = tensor("op_48528_cast_fp16")]; + tensor var_48529_to_fp16 = const()[name = tensor("op_48529_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4643_cast_fp16 = mul(x = var_48528_cast_fp16, y = var_48529_to_fp16)[name = tensor("aw_chunk_4643_cast_fp16")]; + tensor var_48532_equation_0 = const()[name = tensor("op_48532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48532_cast_fp16 = einsum(equation = var_48532_equation_0, values = (var_48366_cast_fp16, var_47822_cast_fp16))[name = tensor("op_48532_cast_fp16")]; + tensor var_48533_to_fp16 = const()[name = tensor("op_48533_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4645_cast_fp16 = mul(x = var_48532_cast_fp16, y = var_48533_to_fp16)[name = tensor("aw_chunk_4645_cast_fp16")]; + tensor var_48536_equation_0 = const()[name = tensor("op_48536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48536_cast_fp16 = einsum(equation = var_48536_equation_0, values = (var_48366_cast_fp16, var_47829_cast_fp16))[name = tensor("op_48536_cast_fp16")]; + tensor var_48537_to_fp16 = const()[name = tensor("op_48537_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4647_cast_fp16 = mul(x = var_48536_cast_fp16, y = var_48537_to_fp16)[name = tensor("aw_chunk_4647_cast_fp16")]; + tensor var_48540_equation_0 = const()[name = tensor("op_48540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48540_cast_fp16 = einsum(equation = var_48540_equation_0, values = (var_48370_cast_fp16, var_47836_cast_fp16))[name = tensor("op_48540_cast_fp16")]; + tensor var_48541_to_fp16 = const()[name = tensor("op_48541_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4649_cast_fp16 = mul(x = var_48540_cast_fp16, y = var_48541_to_fp16)[name = tensor("aw_chunk_4649_cast_fp16")]; + tensor var_48544_equation_0 = const()[name = tensor("op_48544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48544_cast_fp16 = einsum(equation = var_48544_equation_0, values = (var_48370_cast_fp16, var_47843_cast_fp16))[name = tensor("op_48544_cast_fp16")]; + tensor var_48545_to_fp16 = const()[name = tensor("op_48545_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4651_cast_fp16 = mul(x = var_48544_cast_fp16, y = var_48545_to_fp16)[name = tensor("aw_chunk_4651_cast_fp16")]; + tensor var_48548_equation_0 = const()[name = tensor("op_48548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48548_cast_fp16 = einsum(equation = var_48548_equation_0, values = (var_48370_cast_fp16, var_47850_cast_fp16))[name = tensor("op_48548_cast_fp16")]; + tensor var_48549_to_fp16 = const()[name = tensor("op_48549_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4653_cast_fp16 = mul(x = var_48548_cast_fp16, y = var_48549_to_fp16)[name = tensor("aw_chunk_4653_cast_fp16")]; + tensor var_48552_equation_0 = const()[name = tensor("op_48552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48552_cast_fp16 = einsum(equation = var_48552_equation_0, values = (var_48370_cast_fp16, var_47857_cast_fp16))[name = tensor("op_48552_cast_fp16")]; + tensor var_48553_to_fp16 = const()[name = tensor("op_48553_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4655_cast_fp16 = mul(x = var_48552_cast_fp16, y = var_48553_to_fp16)[name = tensor("aw_chunk_4655_cast_fp16")]; + tensor var_48556_equation_0 = const()[name = tensor("op_48556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48556_cast_fp16 = einsum(equation = var_48556_equation_0, values = (var_48374_cast_fp16, var_47864_cast_fp16))[name = tensor("op_48556_cast_fp16")]; + tensor var_48557_to_fp16 = const()[name = tensor("op_48557_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4657_cast_fp16 = mul(x = var_48556_cast_fp16, y = var_48557_to_fp16)[name = tensor("aw_chunk_4657_cast_fp16")]; + tensor var_48560_equation_0 = const()[name = tensor("op_48560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48560_cast_fp16 = einsum(equation = var_48560_equation_0, values = (var_48374_cast_fp16, var_47871_cast_fp16))[name = tensor("op_48560_cast_fp16")]; + tensor var_48561_to_fp16 = const()[name = tensor("op_48561_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4659_cast_fp16 = mul(x = var_48560_cast_fp16, y = var_48561_to_fp16)[name = tensor("aw_chunk_4659_cast_fp16")]; + tensor var_48564_equation_0 = const()[name = tensor("op_48564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48564_cast_fp16 = einsum(equation = var_48564_equation_0, values = (var_48374_cast_fp16, var_47878_cast_fp16))[name = tensor("op_48564_cast_fp16")]; + tensor var_48565_to_fp16 = const()[name = tensor("op_48565_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4661_cast_fp16 = mul(x = var_48564_cast_fp16, y = var_48565_to_fp16)[name = tensor("aw_chunk_4661_cast_fp16")]; + tensor var_48568_equation_0 = const()[name = tensor("op_48568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48568_cast_fp16 = einsum(equation = var_48568_equation_0, values = (var_48374_cast_fp16, var_47885_cast_fp16))[name = tensor("op_48568_cast_fp16")]; + tensor var_48569_to_fp16 = const()[name = tensor("op_48569_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4663_cast_fp16 = mul(x = var_48568_cast_fp16, y = var_48569_to_fp16)[name = tensor("aw_chunk_4663_cast_fp16")]; + tensor var_48572_equation_0 = const()[name = tensor("op_48572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48572_cast_fp16 = einsum(equation = var_48572_equation_0, values = (var_48378_cast_fp16, var_47892_cast_fp16))[name = tensor("op_48572_cast_fp16")]; + tensor var_48573_to_fp16 = const()[name = tensor("op_48573_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4665_cast_fp16 = mul(x = var_48572_cast_fp16, y = var_48573_to_fp16)[name = tensor("aw_chunk_4665_cast_fp16")]; + tensor var_48576_equation_0 = const()[name = tensor("op_48576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48576_cast_fp16 = einsum(equation = var_48576_equation_0, values = (var_48378_cast_fp16, var_47899_cast_fp16))[name = tensor("op_48576_cast_fp16")]; + tensor var_48577_to_fp16 = const()[name = tensor("op_48577_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4667_cast_fp16 = mul(x = var_48576_cast_fp16, y = var_48577_to_fp16)[name = tensor("aw_chunk_4667_cast_fp16")]; + tensor var_48580_equation_0 = const()[name = tensor("op_48580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48580_cast_fp16 = einsum(equation = var_48580_equation_0, values = (var_48378_cast_fp16, var_47906_cast_fp16))[name = tensor("op_48580_cast_fp16")]; + tensor var_48581_to_fp16 = const()[name = tensor("op_48581_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4669_cast_fp16 = mul(x = var_48580_cast_fp16, y = var_48581_to_fp16)[name = tensor("aw_chunk_4669_cast_fp16")]; + tensor var_48584_equation_0 = const()[name = tensor("op_48584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48584_cast_fp16 = einsum(equation = var_48584_equation_0, values = (var_48378_cast_fp16, var_47913_cast_fp16))[name = tensor("op_48584_cast_fp16")]; + tensor var_48585_to_fp16 = const()[name = tensor("op_48585_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4671_cast_fp16 = mul(x = var_48584_cast_fp16, y = var_48585_to_fp16)[name = tensor("aw_chunk_4671_cast_fp16")]; + tensor var_48588_equation_0 = const()[name = tensor("op_48588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48588_cast_fp16 = einsum(equation = var_48588_equation_0, values = (var_48382_cast_fp16, var_47920_cast_fp16))[name = tensor("op_48588_cast_fp16")]; + tensor var_48589_to_fp16 = const()[name = tensor("op_48589_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4673_cast_fp16 = mul(x = var_48588_cast_fp16, y = var_48589_to_fp16)[name = tensor("aw_chunk_4673_cast_fp16")]; + tensor var_48592_equation_0 = const()[name = tensor("op_48592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48592_cast_fp16 = einsum(equation = var_48592_equation_0, values = (var_48382_cast_fp16, var_47927_cast_fp16))[name = tensor("op_48592_cast_fp16")]; + tensor var_48593_to_fp16 = const()[name = tensor("op_48593_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4675_cast_fp16 = mul(x = var_48592_cast_fp16, y = var_48593_to_fp16)[name = tensor("aw_chunk_4675_cast_fp16")]; + tensor var_48596_equation_0 = const()[name = tensor("op_48596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48596_cast_fp16 = einsum(equation = var_48596_equation_0, values = (var_48382_cast_fp16, var_47934_cast_fp16))[name = tensor("op_48596_cast_fp16")]; + tensor var_48597_to_fp16 = const()[name = tensor("op_48597_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4677_cast_fp16 = mul(x = var_48596_cast_fp16, y = var_48597_to_fp16)[name = tensor("aw_chunk_4677_cast_fp16")]; + tensor var_48600_equation_0 = const()[name = tensor("op_48600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48600_cast_fp16 = einsum(equation = var_48600_equation_0, values = (var_48382_cast_fp16, var_47941_cast_fp16))[name = tensor("op_48600_cast_fp16")]; + tensor var_48601_to_fp16 = const()[name = tensor("op_48601_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4679_cast_fp16 = mul(x = var_48600_cast_fp16, y = var_48601_to_fp16)[name = tensor("aw_chunk_4679_cast_fp16")]; + tensor var_48604_equation_0 = const()[name = tensor("op_48604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48604_cast_fp16 = einsum(equation = var_48604_equation_0, values = (var_48386_cast_fp16, var_47948_cast_fp16))[name = tensor("op_48604_cast_fp16")]; + tensor var_48605_to_fp16 = const()[name = tensor("op_48605_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4681_cast_fp16 = mul(x = var_48604_cast_fp16, y = var_48605_to_fp16)[name = tensor("aw_chunk_4681_cast_fp16")]; + tensor var_48608_equation_0 = const()[name = tensor("op_48608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48608_cast_fp16 = einsum(equation = var_48608_equation_0, values = (var_48386_cast_fp16, var_47955_cast_fp16))[name = tensor("op_48608_cast_fp16")]; + tensor var_48609_to_fp16 = const()[name = tensor("op_48609_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4683_cast_fp16 = mul(x = var_48608_cast_fp16, y = var_48609_to_fp16)[name = tensor("aw_chunk_4683_cast_fp16")]; + tensor var_48612_equation_0 = const()[name = tensor("op_48612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48612_cast_fp16 = einsum(equation = var_48612_equation_0, values = (var_48386_cast_fp16, var_47962_cast_fp16))[name = tensor("op_48612_cast_fp16")]; + tensor var_48613_to_fp16 = const()[name = tensor("op_48613_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4685_cast_fp16 = mul(x = var_48612_cast_fp16, y = var_48613_to_fp16)[name = tensor("aw_chunk_4685_cast_fp16")]; + tensor var_48616_equation_0 = const()[name = tensor("op_48616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48616_cast_fp16 = einsum(equation = var_48616_equation_0, values = (var_48386_cast_fp16, var_47969_cast_fp16))[name = tensor("op_48616_cast_fp16")]; + tensor var_48617_to_fp16 = const()[name = tensor("op_48617_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4687_cast_fp16 = mul(x = var_48616_cast_fp16, y = var_48617_to_fp16)[name = tensor("aw_chunk_4687_cast_fp16")]; + tensor var_48620_equation_0 = const()[name = tensor("op_48620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48620_cast_fp16 = einsum(equation = var_48620_equation_0, values = (var_48390_cast_fp16, var_47976_cast_fp16))[name = tensor("op_48620_cast_fp16")]; + tensor var_48621_to_fp16 = const()[name = tensor("op_48621_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4689_cast_fp16 = mul(x = var_48620_cast_fp16, y = var_48621_to_fp16)[name = tensor("aw_chunk_4689_cast_fp16")]; + tensor var_48624_equation_0 = const()[name = tensor("op_48624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48624_cast_fp16 = einsum(equation = var_48624_equation_0, values = (var_48390_cast_fp16, var_47983_cast_fp16))[name = tensor("op_48624_cast_fp16")]; + tensor var_48625_to_fp16 = const()[name = tensor("op_48625_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4691_cast_fp16 = mul(x = var_48624_cast_fp16, y = var_48625_to_fp16)[name = tensor("aw_chunk_4691_cast_fp16")]; + tensor var_48628_equation_0 = const()[name = tensor("op_48628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48628_cast_fp16 = einsum(equation = var_48628_equation_0, values = (var_48390_cast_fp16, var_47990_cast_fp16))[name = tensor("op_48628_cast_fp16")]; + tensor var_48629_to_fp16 = const()[name = tensor("op_48629_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4693_cast_fp16 = mul(x = var_48628_cast_fp16, y = var_48629_to_fp16)[name = tensor("aw_chunk_4693_cast_fp16")]; + tensor var_48632_equation_0 = const()[name = tensor("op_48632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48632_cast_fp16 = einsum(equation = var_48632_equation_0, values = (var_48390_cast_fp16, var_47997_cast_fp16))[name = tensor("op_48632_cast_fp16")]; + tensor var_48633_to_fp16 = const()[name = tensor("op_48633_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4695_cast_fp16 = mul(x = var_48632_cast_fp16, y = var_48633_to_fp16)[name = tensor("aw_chunk_4695_cast_fp16")]; + tensor var_48636_equation_0 = const()[name = tensor("op_48636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48636_cast_fp16 = einsum(equation = var_48636_equation_0, values = (var_48394_cast_fp16, var_48004_cast_fp16))[name = tensor("op_48636_cast_fp16")]; + tensor var_48637_to_fp16 = const()[name = tensor("op_48637_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4697_cast_fp16 = mul(x = var_48636_cast_fp16, y = var_48637_to_fp16)[name = tensor("aw_chunk_4697_cast_fp16")]; + tensor var_48640_equation_0 = const()[name = tensor("op_48640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48640_cast_fp16 = einsum(equation = var_48640_equation_0, values = (var_48394_cast_fp16, var_48011_cast_fp16))[name = tensor("op_48640_cast_fp16")]; + tensor var_48641_to_fp16 = const()[name = tensor("op_48641_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4699_cast_fp16 = mul(x = var_48640_cast_fp16, y = var_48641_to_fp16)[name = tensor("aw_chunk_4699_cast_fp16")]; + tensor var_48644_equation_0 = const()[name = tensor("op_48644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48644_cast_fp16 = einsum(equation = var_48644_equation_0, values = (var_48394_cast_fp16, var_48018_cast_fp16))[name = tensor("op_48644_cast_fp16")]; + tensor var_48645_to_fp16 = const()[name = tensor("op_48645_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4701_cast_fp16 = mul(x = var_48644_cast_fp16, y = var_48645_to_fp16)[name = tensor("aw_chunk_4701_cast_fp16")]; + tensor var_48648_equation_0 = const()[name = tensor("op_48648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48648_cast_fp16 = einsum(equation = var_48648_equation_0, values = (var_48394_cast_fp16, var_48025_cast_fp16))[name = tensor("op_48648_cast_fp16")]; + tensor var_48649_to_fp16 = const()[name = tensor("op_48649_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4703_cast_fp16 = mul(x = var_48648_cast_fp16, y = var_48649_to_fp16)[name = tensor("aw_chunk_4703_cast_fp16")]; + tensor var_48652_equation_0 = const()[name = tensor("op_48652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48652_cast_fp16 = einsum(equation = var_48652_equation_0, values = (var_48398_cast_fp16, var_48032_cast_fp16))[name = tensor("op_48652_cast_fp16")]; + tensor var_48653_to_fp16 = const()[name = tensor("op_48653_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4705_cast_fp16 = mul(x = var_48652_cast_fp16, y = var_48653_to_fp16)[name = tensor("aw_chunk_4705_cast_fp16")]; + tensor var_48656_equation_0 = const()[name = tensor("op_48656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48656_cast_fp16 = einsum(equation = var_48656_equation_0, values = (var_48398_cast_fp16, var_48039_cast_fp16))[name = tensor("op_48656_cast_fp16")]; + tensor var_48657_to_fp16 = const()[name = tensor("op_48657_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4707_cast_fp16 = mul(x = var_48656_cast_fp16, y = var_48657_to_fp16)[name = tensor("aw_chunk_4707_cast_fp16")]; + tensor var_48660_equation_0 = const()[name = tensor("op_48660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48660_cast_fp16 = einsum(equation = var_48660_equation_0, values = (var_48398_cast_fp16, var_48046_cast_fp16))[name = tensor("op_48660_cast_fp16")]; + tensor var_48661_to_fp16 = const()[name = tensor("op_48661_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4709_cast_fp16 = mul(x = var_48660_cast_fp16, y = var_48661_to_fp16)[name = tensor("aw_chunk_4709_cast_fp16")]; + tensor var_48664_equation_0 = const()[name = tensor("op_48664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48664_cast_fp16 = einsum(equation = var_48664_equation_0, values = (var_48398_cast_fp16, var_48053_cast_fp16))[name = tensor("op_48664_cast_fp16")]; + tensor var_48665_to_fp16 = const()[name = tensor("op_48665_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4711_cast_fp16 = mul(x = var_48664_cast_fp16, y = var_48665_to_fp16)[name = tensor("aw_chunk_4711_cast_fp16")]; + tensor var_48668_equation_0 = const()[name = tensor("op_48668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48668_cast_fp16 = einsum(equation = var_48668_equation_0, values = (var_48402_cast_fp16, var_48060_cast_fp16))[name = tensor("op_48668_cast_fp16")]; + tensor var_48669_to_fp16 = const()[name = tensor("op_48669_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4713_cast_fp16 = mul(x = var_48668_cast_fp16, y = var_48669_to_fp16)[name = tensor("aw_chunk_4713_cast_fp16")]; + tensor var_48672_equation_0 = const()[name = tensor("op_48672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48672_cast_fp16 = einsum(equation = var_48672_equation_0, values = (var_48402_cast_fp16, var_48067_cast_fp16))[name = tensor("op_48672_cast_fp16")]; + tensor var_48673_to_fp16 = const()[name = tensor("op_48673_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4715_cast_fp16 = mul(x = var_48672_cast_fp16, y = var_48673_to_fp16)[name = tensor("aw_chunk_4715_cast_fp16")]; + tensor var_48676_equation_0 = const()[name = tensor("op_48676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48676_cast_fp16 = einsum(equation = var_48676_equation_0, values = (var_48402_cast_fp16, var_48074_cast_fp16))[name = tensor("op_48676_cast_fp16")]; + tensor var_48677_to_fp16 = const()[name = tensor("op_48677_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4717_cast_fp16 = mul(x = var_48676_cast_fp16, y = var_48677_to_fp16)[name = tensor("aw_chunk_4717_cast_fp16")]; + tensor var_48680_equation_0 = const()[name = tensor("op_48680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48680_cast_fp16 = einsum(equation = var_48680_equation_0, values = (var_48402_cast_fp16, var_48081_cast_fp16))[name = tensor("op_48680_cast_fp16")]; + tensor var_48681_to_fp16 = const()[name = tensor("op_48681_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4719_cast_fp16 = mul(x = var_48680_cast_fp16, y = var_48681_to_fp16)[name = tensor("aw_chunk_4719_cast_fp16")]; + tensor var_48684_equation_0 = const()[name = tensor("op_48684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48684_cast_fp16 = einsum(equation = var_48684_equation_0, values = (var_48406_cast_fp16, var_48088_cast_fp16))[name = tensor("op_48684_cast_fp16")]; + tensor var_48685_to_fp16 = const()[name = tensor("op_48685_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4721_cast_fp16 = mul(x = var_48684_cast_fp16, y = var_48685_to_fp16)[name = tensor("aw_chunk_4721_cast_fp16")]; + tensor var_48688_equation_0 = const()[name = tensor("op_48688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48688_cast_fp16 = einsum(equation = var_48688_equation_0, values = (var_48406_cast_fp16, var_48095_cast_fp16))[name = tensor("op_48688_cast_fp16")]; + tensor var_48689_to_fp16 = const()[name = tensor("op_48689_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4723_cast_fp16 = mul(x = var_48688_cast_fp16, y = var_48689_to_fp16)[name = tensor("aw_chunk_4723_cast_fp16")]; + tensor var_48692_equation_0 = const()[name = tensor("op_48692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48692_cast_fp16 = einsum(equation = var_48692_equation_0, values = (var_48406_cast_fp16, var_48102_cast_fp16))[name = tensor("op_48692_cast_fp16")]; + tensor var_48693_to_fp16 = const()[name = tensor("op_48693_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4725_cast_fp16 = mul(x = var_48692_cast_fp16, y = var_48693_to_fp16)[name = tensor("aw_chunk_4725_cast_fp16")]; + tensor var_48696_equation_0 = const()[name = tensor("op_48696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48696_cast_fp16 = einsum(equation = var_48696_equation_0, values = (var_48406_cast_fp16, var_48109_cast_fp16))[name = tensor("op_48696_cast_fp16")]; + tensor var_48697_to_fp16 = const()[name = tensor("op_48697_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4727_cast_fp16 = mul(x = var_48696_cast_fp16, y = var_48697_to_fp16)[name = tensor("aw_chunk_4727_cast_fp16")]; + tensor var_48700_equation_0 = const()[name = tensor("op_48700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48700_cast_fp16 = einsum(equation = var_48700_equation_0, values = (var_48410_cast_fp16, var_48116_cast_fp16))[name = tensor("op_48700_cast_fp16")]; + tensor var_48701_to_fp16 = const()[name = tensor("op_48701_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4729_cast_fp16 = mul(x = var_48700_cast_fp16, y = var_48701_to_fp16)[name = tensor("aw_chunk_4729_cast_fp16")]; + tensor var_48704_equation_0 = const()[name = tensor("op_48704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48704_cast_fp16 = einsum(equation = var_48704_equation_0, values = (var_48410_cast_fp16, var_48123_cast_fp16))[name = tensor("op_48704_cast_fp16")]; + tensor var_48705_to_fp16 = const()[name = tensor("op_48705_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4731_cast_fp16 = mul(x = var_48704_cast_fp16, y = var_48705_to_fp16)[name = tensor("aw_chunk_4731_cast_fp16")]; + tensor var_48708_equation_0 = const()[name = tensor("op_48708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48708_cast_fp16 = einsum(equation = var_48708_equation_0, values = (var_48410_cast_fp16, var_48130_cast_fp16))[name = tensor("op_48708_cast_fp16")]; + tensor var_48709_to_fp16 = const()[name = tensor("op_48709_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4733_cast_fp16 = mul(x = var_48708_cast_fp16, y = var_48709_to_fp16)[name = tensor("aw_chunk_4733_cast_fp16")]; + tensor var_48712_equation_0 = const()[name = tensor("op_48712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48712_cast_fp16 = einsum(equation = var_48712_equation_0, values = (var_48410_cast_fp16, var_48137_cast_fp16))[name = tensor("op_48712_cast_fp16")]; + tensor var_48713_to_fp16 = const()[name = tensor("op_48713_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4735_cast_fp16 = mul(x = var_48712_cast_fp16, y = var_48713_to_fp16)[name = tensor("aw_chunk_4735_cast_fp16")]; + tensor var_48716_equation_0 = const()[name = tensor("op_48716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48716_cast_fp16 = einsum(equation = var_48716_equation_0, values = (var_48414_cast_fp16, var_48144_cast_fp16))[name = tensor("op_48716_cast_fp16")]; + tensor var_48717_to_fp16 = const()[name = tensor("op_48717_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4737_cast_fp16 = mul(x = var_48716_cast_fp16, y = var_48717_to_fp16)[name = tensor("aw_chunk_4737_cast_fp16")]; + tensor var_48720_equation_0 = const()[name = tensor("op_48720_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48720_cast_fp16 = einsum(equation = var_48720_equation_0, values = (var_48414_cast_fp16, var_48151_cast_fp16))[name = tensor("op_48720_cast_fp16")]; + tensor var_48721_to_fp16 = const()[name = tensor("op_48721_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4739_cast_fp16 = mul(x = var_48720_cast_fp16, y = var_48721_to_fp16)[name = tensor("aw_chunk_4739_cast_fp16")]; + tensor var_48724_equation_0 = const()[name = tensor("op_48724_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48724_cast_fp16 = einsum(equation = var_48724_equation_0, values = (var_48414_cast_fp16, var_48158_cast_fp16))[name = tensor("op_48724_cast_fp16")]; + tensor var_48725_to_fp16 = const()[name = tensor("op_48725_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4741_cast_fp16 = mul(x = var_48724_cast_fp16, y = var_48725_to_fp16)[name = tensor("aw_chunk_4741_cast_fp16")]; + tensor var_48728_equation_0 = const()[name = tensor("op_48728_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48728_cast_fp16 = einsum(equation = var_48728_equation_0, values = (var_48414_cast_fp16, var_48165_cast_fp16))[name = tensor("op_48728_cast_fp16")]; + tensor var_48729_to_fp16 = const()[name = tensor("op_48729_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4743_cast_fp16 = mul(x = var_48728_cast_fp16, y = var_48729_to_fp16)[name = tensor("aw_chunk_4743_cast_fp16")]; + tensor var_48732_equation_0 = const()[name = tensor("op_48732_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48732_cast_fp16 = einsum(equation = var_48732_equation_0, values = (var_48418_cast_fp16, var_48172_cast_fp16))[name = tensor("op_48732_cast_fp16")]; + tensor var_48733_to_fp16 = const()[name = tensor("op_48733_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4745_cast_fp16 = mul(x = var_48732_cast_fp16, y = var_48733_to_fp16)[name = tensor("aw_chunk_4745_cast_fp16")]; + tensor var_48736_equation_0 = const()[name = tensor("op_48736_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48736_cast_fp16 = einsum(equation = var_48736_equation_0, values = (var_48418_cast_fp16, var_48179_cast_fp16))[name = tensor("op_48736_cast_fp16")]; + tensor var_48737_to_fp16 = const()[name = tensor("op_48737_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4747_cast_fp16 = mul(x = var_48736_cast_fp16, y = var_48737_to_fp16)[name = tensor("aw_chunk_4747_cast_fp16")]; + tensor var_48740_equation_0 = const()[name = tensor("op_48740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48740_cast_fp16 = einsum(equation = var_48740_equation_0, values = (var_48418_cast_fp16, var_48186_cast_fp16))[name = tensor("op_48740_cast_fp16")]; + tensor var_48741_to_fp16 = const()[name = tensor("op_48741_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4749_cast_fp16 = mul(x = var_48740_cast_fp16, y = var_48741_to_fp16)[name = tensor("aw_chunk_4749_cast_fp16")]; + tensor var_48744_equation_0 = const()[name = tensor("op_48744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48744_cast_fp16 = einsum(equation = var_48744_equation_0, values = (var_48418_cast_fp16, var_48193_cast_fp16))[name = tensor("op_48744_cast_fp16")]; + tensor var_48745_to_fp16 = const()[name = tensor("op_48745_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4751_cast_fp16 = mul(x = var_48744_cast_fp16, y = var_48745_to_fp16)[name = tensor("aw_chunk_4751_cast_fp16")]; + tensor var_48748_equation_0 = const()[name = tensor("op_48748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48748_cast_fp16 = einsum(equation = var_48748_equation_0, values = (var_48422_cast_fp16, var_48200_cast_fp16))[name = tensor("op_48748_cast_fp16")]; + tensor var_48749_to_fp16 = const()[name = tensor("op_48749_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4753_cast_fp16 = mul(x = var_48748_cast_fp16, y = var_48749_to_fp16)[name = tensor("aw_chunk_4753_cast_fp16")]; + tensor var_48752_equation_0 = const()[name = tensor("op_48752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48752_cast_fp16 = einsum(equation = var_48752_equation_0, values = (var_48422_cast_fp16, var_48207_cast_fp16))[name = tensor("op_48752_cast_fp16")]; + tensor var_48753_to_fp16 = const()[name = tensor("op_48753_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4755_cast_fp16 = mul(x = var_48752_cast_fp16, y = var_48753_to_fp16)[name = tensor("aw_chunk_4755_cast_fp16")]; + tensor var_48756_equation_0 = const()[name = tensor("op_48756_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48756_cast_fp16 = einsum(equation = var_48756_equation_0, values = (var_48422_cast_fp16, var_48214_cast_fp16))[name = tensor("op_48756_cast_fp16")]; + tensor var_48757_to_fp16 = const()[name = tensor("op_48757_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4757_cast_fp16 = mul(x = var_48756_cast_fp16, y = var_48757_to_fp16)[name = tensor("aw_chunk_4757_cast_fp16")]; + tensor var_48760_equation_0 = const()[name = tensor("op_48760_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48760_cast_fp16 = einsum(equation = var_48760_equation_0, values = (var_48422_cast_fp16, var_48221_cast_fp16))[name = tensor("op_48760_cast_fp16")]; + tensor var_48761_to_fp16 = const()[name = tensor("op_48761_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4759_cast_fp16 = mul(x = var_48760_cast_fp16, y = var_48761_to_fp16)[name = tensor("aw_chunk_4759_cast_fp16")]; + tensor var_48764_equation_0 = const()[name = tensor("op_48764_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48764_cast_fp16 = einsum(equation = var_48764_equation_0, values = (var_48426_cast_fp16, var_48228_cast_fp16))[name = tensor("op_48764_cast_fp16")]; + tensor var_48765_to_fp16 = const()[name = tensor("op_48765_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4761_cast_fp16 = mul(x = var_48764_cast_fp16, y = var_48765_to_fp16)[name = tensor("aw_chunk_4761_cast_fp16")]; + tensor var_48768_equation_0 = const()[name = tensor("op_48768_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48768_cast_fp16 = einsum(equation = var_48768_equation_0, values = (var_48426_cast_fp16, var_48235_cast_fp16))[name = tensor("op_48768_cast_fp16")]; + tensor var_48769_to_fp16 = const()[name = tensor("op_48769_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4763_cast_fp16 = mul(x = var_48768_cast_fp16, y = var_48769_to_fp16)[name = tensor("aw_chunk_4763_cast_fp16")]; + tensor var_48772_equation_0 = const()[name = tensor("op_48772_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48772_cast_fp16 = einsum(equation = var_48772_equation_0, values = (var_48426_cast_fp16, var_48242_cast_fp16))[name = tensor("op_48772_cast_fp16")]; + tensor var_48773_to_fp16 = const()[name = tensor("op_48773_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4765_cast_fp16 = mul(x = var_48772_cast_fp16, y = var_48773_to_fp16)[name = tensor("aw_chunk_4765_cast_fp16")]; + tensor var_48776_equation_0 = const()[name = tensor("op_48776_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48776_cast_fp16 = einsum(equation = var_48776_equation_0, values = (var_48426_cast_fp16, var_48249_cast_fp16))[name = tensor("op_48776_cast_fp16")]; + tensor var_48777_to_fp16 = const()[name = tensor("op_48777_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4767_cast_fp16 = mul(x = var_48776_cast_fp16, y = var_48777_to_fp16)[name = tensor("aw_chunk_4767_cast_fp16")]; + tensor var_48780_equation_0 = const()[name = tensor("op_48780_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48780_cast_fp16 = einsum(equation = var_48780_equation_0, values = (var_48430_cast_fp16, var_48256_cast_fp16))[name = tensor("op_48780_cast_fp16")]; + tensor var_48781_to_fp16 = const()[name = tensor("op_48781_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4769_cast_fp16 = mul(x = var_48780_cast_fp16, y = var_48781_to_fp16)[name = tensor("aw_chunk_4769_cast_fp16")]; + tensor var_48784_equation_0 = const()[name = tensor("op_48784_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48784_cast_fp16 = einsum(equation = var_48784_equation_0, values = (var_48430_cast_fp16, var_48263_cast_fp16))[name = tensor("op_48784_cast_fp16")]; + tensor var_48785_to_fp16 = const()[name = tensor("op_48785_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4771_cast_fp16 = mul(x = var_48784_cast_fp16, y = var_48785_to_fp16)[name = tensor("aw_chunk_4771_cast_fp16")]; + tensor var_48788_equation_0 = const()[name = tensor("op_48788_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48788_cast_fp16 = einsum(equation = var_48788_equation_0, values = (var_48430_cast_fp16, var_48270_cast_fp16))[name = tensor("op_48788_cast_fp16")]; + tensor var_48789_to_fp16 = const()[name = tensor("op_48789_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4773_cast_fp16 = mul(x = var_48788_cast_fp16, y = var_48789_to_fp16)[name = tensor("aw_chunk_4773_cast_fp16")]; + tensor var_48792_equation_0 = const()[name = tensor("op_48792_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48792_cast_fp16 = einsum(equation = var_48792_equation_0, values = (var_48430_cast_fp16, var_48277_cast_fp16))[name = tensor("op_48792_cast_fp16")]; + tensor var_48793_to_fp16 = const()[name = tensor("op_48793_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4775_cast_fp16 = mul(x = var_48792_cast_fp16, y = var_48793_to_fp16)[name = tensor("aw_chunk_4775_cast_fp16")]; + tensor var_48796_equation_0 = const()[name = tensor("op_48796_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48796_cast_fp16 = einsum(equation = var_48796_equation_0, values = (var_48434_cast_fp16, var_48284_cast_fp16))[name = tensor("op_48796_cast_fp16")]; + tensor var_48797_to_fp16 = const()[name = tensor("op_48797_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4777_cast_fp16 = mul(x = var_48796_cast_fp16, y = var_48797_to_fp16)[name = tensor("aw_chunk_4777_cast_fp16")]; + tensor var_48800_equation_0 = const()[name = tensor("op_48800_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48800_cast_fp16 = einsum(equation = var_48800_equation_0, values = (var_48434_cast_fp16, var_48291_cast_fp16))[name = tensor("op_48800_cast_fp16")]; + tensor var_48801_to_fp16 = const()[name = tensor("op_48801_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4779_cast_fp16 = mul(x = var_48800_cast_fp16, y = var_48801_to_fp16)[name = tensor("aw_chunk_4779_cast_fp16")]; + tensor var_48804_equation_0 = const()[name = tensor("op_48804_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48804_cast_fp16 = einsum(equation = var_48804_equation_0, values = (var_48434_cast_fp16, var_48298_cast_fp16))[name = tensor("op_48804_cast_fp16")]; + tensor var_48805_to_fp16 = const()[name = tensor("op_48805_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4781_cast_fp16 = mul(x = var_48804_cast_fp16, y = var_48805_to_fp16)[name = tensor("aw_chunk_4781_cast_fp16")]; + tensor var_48808_equation_0 = const()[name = tensor("op_48808_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48808_cast_fp16 = einsum(equation = var_48808_equation_0, values = (var_48434_cast_fp16, var_48305_cast_fp16))[name = tensor("op_48808_cast_fp16")]; + tensor var_48809_to_fp16 = const()[name = tensor("op_48809_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4783_cast_fp16 = mul(x = var_48808_cast_fp16, y = var_48809_to_fp16)[name = tensor("aw_chunk_4783_cast_fp16")]; + tensor var_48812_equation_0 = const()[name = tensor("op_48812_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48812_cast_fp16 = einsum(equation = var_48812_equation_0, values = (var_48438_cast_fp16, var_48312_cast_fp16))[name = tensor("op_48812_cast_fp16")]; + tensor var_48813_to_fp16 = const()[name = tensor("op_48813_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4785_cast_fp16 = mul(x = var_48812_cast_fp16, y = var_48813_to_fp16)[name = tensor("aw_chunk_4785_cast_fp16")]; + tensor var_48816_equation_0 = const()[name = tensor("op_48816_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48816_cast_fp16 = einsum(equation = var_48816_equation_0, values = (var_48438_cast_fp16, var_48319_cast_fp16))[name = tensor("op_48816_cast_fp16")]; + tensor var_48817_to_fp16 = const()[name = tensor("op_48817_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4787_cast_fp16 = mul(x = var_48816_cast_fp16, y = var_48817_to_fp16)[name = tensor("aw_chunk_4787_cast_fp16")]; + tensor var_48820_equation_0 = const()[name = tensor("op_48820_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48820_cast_fp16 = einsum(equation = var_48820_equation_0, values = (var_48438_cast_fp16, var_48326_cast_fp16))[name = tensor("op_48820_cast_fp16")]; + tensor var_48821_to_fp16 = const()[name = tensor("op_48821_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4789_cast_fp16 = mul(x = var_48820_cast_fp16, y = var_48821_to_fp16)[name = tensor("aw_chunk_4789_cast_fp16")]; + tensor var_48824_equation_0 = const()[name = tensor("op_48824_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48824_cast_fp16 = einsum(equation = var_48824_equation_0, values = (var_48438_cast_fp16, var_48333_cast_fp16))[name = tensor("op_48824_cast_fp16")]; + tensor var_48825_to_fp16 = const()[name = tensor("op_48825_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4791_cast_fp16 = mul(x = var_48824_cast_fp16, y = var_48825_to_fp16)[name = tensor("aw_chunk_4791_cast_fp16")]; + tensor var_48828_equation_0 = const()[name = tensor("op_48828_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48828_cast_fp16 = einsum(equation = var_48828_equation_0, values = (var_48442_cast_fp16, var_48340_cast_fp16))[name = tensor("op_48828_cast_fp16")]; + tensor var_48829_to_fp16 = const()[name = tensor("op_48829_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4793_cast_fp16 = mul(x = var_48828_cast_fp16, y = var_48829_to_fp16)[name = tensor("aw_chunk_4793_cast_fp16")]; + tensor var_48832_equation_0 = const()[name = tensor("op_48832_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48832_cast_fp16 = einsum(equation = var_48832_equation_0, values = (var_48442_cast_fp16, var_48347_cast_fp16))[name = tensor("op_48832_cast_fp16")]; + tensor var_48833_to_fp16 = const()[name = tensor("op_48833_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4795_cast_fp16 = mul(x = var_48832_cast_fp16, y = var_48833_to_fp16)[name = tensor("aw_chunk_4795_cast_fp16")]; + tensor var_48836_equation_0 = const()[name = tensor("op_48836_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48836_cast_fp16 = einsum(equation = var_48836_equation_0, values = (var_48442_cast_fp16, var_48354_cast_fp16))[name = tensor("op_48836_cast_fp16")]; + tensor var_48837_to_fp16 = const()[name = tensor("op_48837_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4797_cast_fp16 = mul(x = var_48836_cast_fp16, y = var_48837_to_fp16)[name = tensor("aw_chunk_4797_cast_fp16")]; + tensor var_48840_equation_0 = const()[name = tensor("op_48840_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_48840_cast_fp16 = einsum(equation = var_48840_equation_0, values = (var_48442_cast_fp16, var_48361_cast_fp16))[name = tensor("op_48840_cast_fp16")]; + tensor var_48841_to_fp16 = const()[name = tensor("op_48841_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4799_cast_fp16 = mul(x = var_48840_cast_fp16, y = var_48841_to_fp16)[name = tensor("aw_chunk_4799_cast_fp16")]; + tensor var_48843_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4641_cast_fp16)[name = tensor("op_48843_cast_fp16")]; + tensor var_48844_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4643_cast_fp16)[name = tensor("op_48844_cast_fp16")]; + tensor var_48845_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4645_cast_fp16)[name = tensor("op_48845_cast_fp16")]; + tensor var_48846_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4647_cast_fp16)[name = tensor("op_48846_cast_fp16")]; + tensor var_48847_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4649_cast_fp16)[name = tensor("op_48847_cast_fp16")]; + tensor var_48848_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4651_cast_fp16)[name = tensor("op_48848_cast_fp16")]; + tensor var_48849_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4653_cast_fp16)[name = tensor("op_48849_cast_fp16")]; + tensor var_48850_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4655_cast_fp16)[name = tensor("op_48850_cast_fp16")]; + tensor var_48851_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4657_cast_fp16)[name = tensor("op_48851_cast_fp16")]; + tensor var_48852_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4659_cast_fp16)[name = tensor("op_48852_cast_fp16")]; + tensor var_48853_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4661_cast_fp16)[name = tensor("op_48853_cast_fp16")]; + tensor var_48854_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4663_cast_fp16)[name = tensor("op_48854_cast_fp16")]; + tensor var_48855_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4665_cast_fp16)[name = tensor("op_48855_cast_fp16")]; + tensor var_48856_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4667_cast_fp16)[name = tensor("op_48856_cast_fp16")]; + tensor var_48857_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4669_cast_fp16)[name = tensor("op_48857_cast_fp16")]; + tensor var_48858_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4671_cast_fp16)[name = tensor("op_48858_cast_fp16")]; + tensor var_48859_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4673_cast_fp16)[name = tensor("op_48859_cast_fp16")]; + tensor var_48860_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4675_cast_fp16)[name = tensor("op_48860_cast_fp16")]; + tensor var_48861_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4677_cast_fp16)[name = tensor("op_48861_cast_fp16")]; + tensor var_48862_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4679_cast_fp16)[name = tensor("op_48862_cast_fp16")]; + tensor var_48863_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4681_cast_fp16)[name = tensor("op_48863_cast_fp16")]; + tensor var_48864_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4683_cast_fp16)[name = tensor("op_48864_cast_fp16")]; + tensor var_48865_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4685_cast_fp16)[name = tensor("op_48865_cast_fp16")]; + tensor var_48866_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4687_cast_fp16)[name = tensor("op_48866_cast_fp16")]; + tensor var_48867_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4689_cast_fp16)[name = tensor("op_48867_cast_fp16")]; + tensor var_48868_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4691_cast_fp16)[name = tensor("op_48868_cast_fp16")]; + tensor var_48869_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4693_cast_fp16)[name = tensor("op_48869_cast_fp16")]; + tensor var_48870_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4695_cast_fp16)[name = tensor("op_48870_cast_fp16")]; + tensor var_48871_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4697_cast_fp16)[name = tensor("op_48871_cast_fp16")]; + tensor var_48872_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4699_cast_fp16)[name = tensor("op_48872_cast_fp16")]; + tensor var_48873_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4701_cast_fp16)[name = tensor("op_48873_cast_fp16")]; + tensor var_48874_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4703_cast_fp16)[name = tensor("op_48874_cast_fp16")]; + tensor var_48875_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4705_cast_fp16)[name = tensor("op_48875_cast_fp16")]; + tensor var_48876_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4707_cast_fp16)[name = tensor("op_48876_cast_fp16")]; + tensor var_48877_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4709_cast_fp16)[name = tensor("op_48877_cast_fp16")]; + tensor var_48878_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4711_cast_fp16)[name = tensor("op_48878_cast_fp16")]; + tensor var_48879_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4713_cast_fp16)[name = tensor("op_48879_cast_fp16")]; + tensor var_48880_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4715_cast_fp16)[name = tensor("op_48880_cast_fp16")]; + tensor var_48881_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4717_cast_fp16)[name = tensor("op_48881_cast_fp16")]; + tensor var_48882_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4719_cast_fp16)[name = tensor("op_48882_cast_fp16")]; + tensor var_48883_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4721_cast_fp16)[name = tensor("op_48883_cast_fp16")]; + tensor var_48884_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4723_cast_fp16)[name = tensor("op_48884_cast_fp16")]; + tensor var_48885_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4725_cast_fp16)[name = tensor("op_48885_cast_fp16")]; + tensor var_48886_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4727_cast_fp16)[name = tensor("op_48886_cast_fp16")]; + tensor var_48887_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4729_cast_fp16)[name = tensor("op_48887_cast_fp16")]; + tensor var_48888_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4731_cast_fp16)[name = tensor("op_48888_cast_fp16")]; + tensor var_48889_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4733_cast_fp16)[name = tensor("op_48889_cast_fp16")]; + tensor var_48890_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4735_cast_fp16)[name = tensor("op_48890_cast_fp16")]; + tensor var_48891_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4737_cast_fp16)[name = tensor("op_48891_cast_fp16")]; + tensor var_48892_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4739_cast_fp16)[name = tensor("op_48892_cast_fp16")]; + tensor var_48893_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4741_cast_fp16)[name = tensor("op_48893_cast_fp16")]; + tensor var_48894_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4743_cast_fp16)[name = tensor("op_48894_cast_fp16")]; + tensor var_48895_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4745_cast_fp16)[name = tensor("op_48895_cast_fp16")]; + tensor var_48896_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4747_cast_fp16)[name = tensor("op_48896_cast_fp16")]; + tensor var_48897_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4749_cast_fp16)[name = tensor("op_48897_cast_fp16")]; + tensor var_48898_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4751_cast_fp16)[name = tensor("op_48898_cast_fp16")]; + tensor var_48899_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4753_cast_fp16)[name = tensor("op_48899_cast_fp16")]; + tensor var_48900_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4755_cast_fp16)[name = tensor("op_48900_cast_fp16")]; + tensor var_48901_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4757_cast_fp16)[name = tensor("op_48901_cast_fp16")]; + tensor var_48902_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4759_cast_fp16)[name = tensor("op_48902_cast_fp16")]; + tensor var_48903_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4761_cast_fp16)[name = tensor("op_48903_cast_fp16")]; + tensor var_48904_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4763_cast_fp16)[name = tensor("op_48904_cast_fp16")]; + tensor var_48905_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4765_cast_fp16)[name = tensor("op_48905_cast_fp16")]; + tensor var_48906_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4767_cast_fp16)[name = tensor("op_48906_cast_fp16")]; + tensor var_48907_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4769_cast_fp16)[name = tensor("op_48907_cast_fp16")]; + tensor var_48908_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4771_cast_fp16)[name = tensor("op_48908_cast_fp16")]; + tensor var_48909_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4773_cast_fp16)[name = tensor("op_48909_cast_fp16")]; + tensor var_48910_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4775_cast_fp16)[name = tensor("op_48910_cast_fp16")]; + tensor var_48911_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4777_cast_fp16)[name = tensor("op_48911_cast_fp16")]; + tensor var_48912_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4779_cast_fp16)[name = tensor("op_48912_cast_fp16")]; + tensor var_48913_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4781_cast_fp16)[name = tensor("op_48913_cast_fp16")]; + tensor var_48914_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4783_cast_fp16)[name = tensor("op_48914_cast_fp16")]; + tensor var_48915_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4785_cast_fp16)[name = tensor("op_48915_cast_fp16")]; + tensor var_48916_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4787_cast_fp16)[name = tensor("op_48916_cast_fp16")]; + tensor var_48917_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4789_cast_fp16)[name = tensor("op_48917_cast_fp16")]; + tensor var_48918_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4791_cast_fp16)[name = tensor("op_48918_cast_fp16")]; + tensor var_48919_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4793_cast_fp16)[name = tensor("op_48919_cast_fp16")]; + tensor var_48920_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4795_cast_fp16)[name = tensor("op_48920_cast_fp16")]; + tensor var_48921_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4797_cast_fp16)[name = tensor("op_48921_cast_fp16")]; + tensor var_48922_cast_fp16 = softmax(axis = var_47614, x = aw_chunk_4799_cast_fp16)[name = tensor("op_48922_cast_fp16")]; + tensor var_48924_equation_0 = const()[name = tensor("op_48924_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48924_cast_fp16 = einsum(equation = var_48924_equation_0, values = (var_48444_cast_fp16, var_48843_cast_fp16))[name = tensor("op_48924_cast_fp16")]; + tensor var_48926_equation_0 = const()[name = tensor("op_48926_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48926_cast_fp16 = einsum(equation = var_48926_equation_0, values = (var_48444_cast_fp16, var_48844_cast_fp16))[name = tensor("op_48926_cast_fp16")]; + tensor var_48928_equation_0 = const()[name = tensor("op_48928_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48928_cast_fp16 = einsum(equation = var_48928_equation_0, values = (var_48444_cast_fp16, var_48845_cast_fp16))[name = tensor("op_48928_cast_fp16")]; + tensor var_48930_equation_0 = const()[name = tensor("op_48930_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48930_cast_fp16 = einsum(equation = var_48930_equation_0, values = (var_48444_cast_fp16, var_48846_cast_fp16))[name = tensor("op_48930_cast_fp16")]; + tensor var_48932_equation_0 = const()[name = tensor("op_48932_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48932_cast_fp16 = einsum(equation = var_48932_equation_0, values = (var_48448_cast_fp16, var_48847_cast_fp16))[name = tensor("op_48932_cast_fp16")]; + tensor var_48934_equation_0 = const()[name = tensor("op_48934_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48934_cast_fp16 = einsum(equation = var_48934_equation_0, values = (var_48448_cast_fp16, var_48848_cast_fp16))[name = tensor("op_48934_cast_fp16")]; + tensor var_48936_equation_0 = const()[name = tensor("op_48936_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48936_cast_fp16 = einsum(equation = var_48936_equation_0, values = (var_48448_cast_fp16, var_48849_cast_fp16))[name = tensor("op_48936_cast_fp16")]; + tensor var_48938_equation_0 = const()[name = tensor("op_48938_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48938_cast_fp16 = einsum(equation = var_48938_equation_0, values = (var_48448_cast_fp16, var_48850_cast_fp16))[name = tensor("op_48938_cast_fp16")]; + tensor var_48940_equation_0 = const()[name = tensor("op_48940_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48940_cast_fp16 = einsum(equation = var_48940_equation_0, values = (var_48452_cast_fp16, var_48851_cast_fp16))[name = tensor("op_48940_cast_fp16")]; + tensor var_48942_equation_0 = const()[name = tensor("op_48942_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48942_cast_fp16 = einsum(equation = var_48942_equation_0, values = (var_48452_cast_fp16, var_48852_cast_fp16))[name = tensor("op_48942_cast_fp16")]; + tensor var_48944_equation_0 = const()[name = tensor("op_48944_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48944_cast_fp16 = einsum(equation = var_48944_equation_0, values = (var_48452_cast_fp16, var_48853_cast_fp16))[name = tensor("op_48944_cast_fp16")]; + tensor var_48946_equation_0 = const()[name = tensor("op_48946_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48946_cast_fp16 = einsum(equation = var_48946_equation_0, values = (var_48452_cast_fp16, var_48854_cast_fp16))[name = tensor("op_48946_cast_fp16")]; + tensor var_48948_equation_0 = const()[name = tensor("op_48948_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48948_cast_fp16 = einsum(equation = var_48948_equation_0, values = (var_48456_cast_fp16, var_48855_cast_fp16))[name = tensor("op_48948_cast_fp16")]; + tensor var_48950_equation_0 = const()[name = tensor("op_48950_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48950_cast_fp16 = einsum(equation = var_48950_equation_0, values = (var_48456_cast_fp16, var_48856_cast_fp16))[name = tensor("op_48950_cast_fp16")]; + tensor var_48952_equation_0 = const()[name = tensor("op_48952_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48952_cast_fp16 = einsum(equation = var_48952_equation_0, values = (var_48456_cast_fp16, var_48857_cast_fp16))[name = tensor("op_48952_cast_fp16")]; + tensor var_48954_equation_0 = const()[name = tensor("op_48954_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48954_cast_fp16 = einsum(equation = var_48954_equation_0, values = (var_48456_cast_fp16, var_48858_cast_fp16))[name = tensor("op_48954_cast_fp16")]; + tensor var_48956_equation_0 = const()[name = tensor("op_48956_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48956_cast_fp16 = einsum(equation = var_48956_equation_0, values = (var_48460_cast_fp16, var_48859_cast_fp16))[name = tensor("op_48956_cast_fp16")]; + tensor var_48958_equation_0 = const()[name = tensor("op_48958_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48958_cast_fp16 = einsum(equation = var_48958_equation_0, values = (var_48460_cast_fp16, var_48860_cast_fp16))[name = tensor("op_48958_cast_fp16")]; + tensor var_48960_equation_0 = const()[name = tensor("op_48960_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48960_cast_fp16 = einsum(equation = var_48960_equation_0, values = (var_48460_cast_fp16, var_48861_cast_fp16))[name = tensor("op_48960_cast_fp16")]; + tensor var_48962_equation_0 = const()[name = tensor("op_48962_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48962_cast_fp16 = einsum(equation = var_48962_equation_0, values = (var_48460_cast_fp16, var_48862_cast_fp16))[name = tensor("op_48962_cast_fp16")]; + tensor var_48964_equation_0 = const()[name = tensor("op_48964_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48964_cast_fp16 = einsum(equation = var_48964_equation_0, values = (var_48464_cast_fp16, var_48863_cast_fp16))[name = tensor("op_48964_cast_fp16")]; + tensor var_48966_equation_0 = const()[name = tensor("op_48966_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48966_cast_fp16 = einsum(equation = var_48966_equation_0, values = (var_48464_cast_fp16, var_48864_cast_fp16))[name = tensor("op_48966_cast_fp16")]; + tensor var_48968_equation_0 = const()[name = tensor("op_48968_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48968_cast_fp16 = einsum(equation = var_48968_equation_0, values = (var_48464_cast_fp16, var_48865_cast_fp16))[name = tensor("op_48968_cast_fp16")]; + tensor var_48970_equation_0 = const()[name = tensor("op_48970_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48970_cast_fp16 = einsum(equation = var_48970_equation_0, values = (var_48464_cast_fp16, var_48866_cast_fp16))[name = tensor("op_48970_cast_fp16")]; + tensor var_48972_equation_0 = const()[name = tensor("op_48972_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48972_cast_fp16 = einsum(equation = var_48972_equation_0, values = (var_48468_cast_fp16, var_48867_cast_fp16))[name = tensor("op_48972_cast_fp16")]; + tensor var_48974_equation_0 = const()[name = tensor("op_48974_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48974_cast_fp16 = einsum(equation = var_48974_equation_0, values = (var_48468_cast_fp16, var_48868_cast_fp16))[name = tensor("op_48974_cast_fp16")]; + tensor var_48976_equation_0 = const()[name = tensor("op_48976_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48976_cast_fp16 = einsum(equation = var_48976_equation_0, values = (var_48468_cast_fp16, var_48869_cast_fp16))[name = tensor("op_48976_cast_fp16")]; + tensor var_48978_equation_0 = const()[name = tensor("op_48978_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48978_cast_fp16 = einsum(equation = var_48978_equation_0, values = (var_48468_cast_fp16, var_48870_cast_fp16))[name = tensor("op_48978_cast_fp16")]; + tensor var_48980_equation_0 = const()[name = tensor("op_48980_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48980_cast_fp16 = einsum(equation = var_48980_equation_0, values = (var_48472_cast_fp16, var_48871_cast_fp16))[name = tensor("op_48980_cast_fp16")]; + tensor var_48982_equation_0 = const()[name = tensor("op_48982_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48982_cast_fp16 = einsum(equation = var_48982_equation_0, values = (var_48472_cast_fp16, var_48872_cast_fp16))[name = tensor("op_48982_cast_fp16")]; + tensor var_48984_equation_0 = const()[name = tensor("op_48984_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48984_cast_fp16 = einsum(equation = var_48984_equation_0, values = (var_48472_cast_fp16, var_48873_cast_fp16))[name = tensor("op_48984_cast_fp16")]; + tensor var_48986_equation_0 = const()[name = tensor("op_48986_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48986_cast_fp16 = einsum(equation = var_48986_equation_0, values = (var_48472_cast_fp16, var_48874_cast_fp16))[name = tensor("op_48986_cast_fp16")]; + tensor var_48988_equation_0 = const()[name = tensor("op_48988_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48988_cast_fp16 = einsum(equation = var_48988_equation_0, values = (var_48476_cast_fp16, var_48875_cast_fp16))[name = tensor("op_48988_cast_fp16")]; + tensor var_48990_equation_0 = const()[name = tensor("op_48990_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48990_cast_fp16 = einsum(equation = var_48990_equation_0, values = (var_48476_cast_fp16, var_48876_cast_fp16))[name = tensor("op_48990_cast_fp16")]; + tensor var_48992_equation_0 = const()[name = tensor("op_48992_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48992_cast_fp16 = einsum(equation = var_48992_equation_0, values = (var_48476_cast_fp16, var_48877_cast_fp16))[name = tensor("op_48992_cast_fp16")]; + tensor var_48994_equation_0 = const()[name = tensor("op_48994_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48994_cast_fp16 = einsum(equation = var_48994_equation_0, values = (var_48476_cast_fp16, var_48878_cast_fp16))[name = tensor("op_48994_cast_fp16")]; + tensor var_48996_equation_0 = const()[name = tensor("op_48996_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48996_cast_fp16 = einsum(equation = var_48996_equation_0, values = (var_48480_cast_fp16, var_48879_cast_fp16))[name = tensor("op_48996_cast_fp16")]; + tensor var_48998_equation_0 = const()[name = tensor("op_48998_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_48998_cast_fp16 = einsum(equation = var_48998_equation_0, values = (var_48480_cast_fp16, var_48880_cast_fp16))[name = tensor("op_48998_cast_fp16")]; + tensor var_49000_equation_0 = const()[name = tensor("op_49000_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49000_cast_fp16 = einsum(equation = var_49000_equation_0, values = (var_48480_cast_fp16, var_48881_cast_fp16))[name = tensor("op_49000_cast_fp16")]; + tensor var_49002_equation_0 = const()[name = tensor("op_49002_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49002_cast_fp16 = einsum(equation = var_49002_equation_0, values = (var_48480_cast_fp16, var_48882_cast_fp16))[name = tensor("op_49002_cast_fp16")]; + tensor var_49004_equation_0 = const()[name = tensor("op_49004_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49004_cast_fp16 = einsum(equation = var_49004_equation_0, values = (var_48484_cast_fp16, var_48883_cast_fp16))[name = tensor("op_49004_cast_fp16")]; + tensor var_49006_equation_0 = const()[name = tensor("op_49006_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49006_cast_fp16 = einsum(equation = var_49006_equation_0, values = (var_48484_cast_fp16, var_48884_cast_fp16))[name = tensor("op_49006_cast_fp16")]; + tensor var_49008_equation_0 = const()[name = tensor("op_49008_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49008_cast_fp16 = einsum(equation = var_49008_equation_0, values = (var_48484_cast_fp16, var_48885_cast_fp16))[name = tensor("op_49008_cast_fp16")]; + tensor var_49010_equation_0 = const()[name = tensor("op_49010_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49010_cast_fp16 = einsum(equation = var_49010_equation_0, values = (var_48484_cast_fp16, var_48886_cast_fp16))[name = tensor("op_49010_cast_fp16")]; + tensor var_49012_equation_0 = const()[name = tensor("op_49012_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49012_cast_fp16 = einsum(equation = var_49012_equation_0, values = (var_48488_cast_fp16, var_48887_cast_fp16))[name = tensor("op_49012_cast_fp16")]; + tensor var_49014_equation_0 = const()[name = tensor("op_49014_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49014_cast_fp16 = einsum(equation = var_49014_equation_0, values = (var_48488_cast_fp16, var_48888_cast_fp16))[name = tensor("op_49014_cast_fp16")]; + tensor var_49016_equation_0 = const()[name = tensor("op_49016_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49016_cast_fp16 = einsum(equation = var_49016_equation_0, values = (var_48488_cast_fp16, var_48889_cast_fp16))[name = tensor("op_49016_cast_fp16")]; + tensor var_49018_equation_0 = const()[name = tensor("op_49018_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49018_cast_fp16 = einsum(equation = var_49018_equation_0, values = (var_48488_cast_fp16, var_48890_cast_fp16))[name = tensor("op_49018_cast_fp16")]; + tensor var_49020_equation_0 = const()[name = tensor("op_49020_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49020_cast_fp16 = einsum(equation = var_49020_equation_0, values = (var_48492_cast_fp16, var_48891_cast_fp16))[name = tensor("op_49020_cast_fp16")]; + tensor var_49022_equation_0 = const()[name = tensor("op_49022_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49022_cast_fp16 = einsum(equation = var_49022_equation_0, values = (var_48492_cast_fp16, var_48892_cast_fp16))[name = tensor("op_49022_cast_fp16")]; + tensor var_49024_equation_0 = const()[name = tensor("op_49024_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49024_cast_fp16 = einsum(equation = var_49024_equation_0, values = (var_48492_cast_fp16, var_48893_cast_fp16))[name = tensor("op_49024_cast_fp16")]; + tensor var_49026_equation_0 = const()[name = tensor("op_49026_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49026_cast_fp16 = einsum(equation = var_49026_equation_0, values = (var_48492_cast_fp16, var_48894_cast_fp16))[name = tensor("op_49026_cast_fp16")]; + tensor var_49028_equation_0 = const()[name = tensor("op_49028_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49028_cast_fp16 = einsum(equation = var_49028_equation_0, values = (var_48496_cast_fp16, var_48895_cast_fp16))[name = tensor("op_49028_cast_fp16")]; + tensor var_49030_equation_0 = const()[name = tensor("op_49030_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49030_cast_fp16 = einsum(equation = var_49030_equation_0, values = (var_48496_cast_fp16, var_48896_cast_fp16))[name = tensor("op_49030_cast_fp16")]; + tensor var_49032_equation_0 = const()[name = tensor("op_49032_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49032_cast_fp16 = einsum(equation = var_49032_equation_0, values = (var_48496_cast_fp16, var_48897_cast_fp16))[name = tensor("op_49032_cast_fp16")]; + tensor var_49034_equation_0 = const()[name = tensor("op_49034_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49034_cast_fp16 = einsum(equation = var_49034_equation_0, values = (var_48496_cast_fp16, var_48898_cast_fp16))[name = tensor("op_49034_cast_fp16")]; + tensor var_49036_equation_0 = const()[name = tensor("op_49036_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49036_cast_fp16 = einsum(equation = var_49036_equation_0, values = (var_48500_cast_fp16, var_48899_cast_fp16))[name = tensor("op_49036_cast_fp16")]; + tensor var_49038_equation_0 = const()[name = tensor("op_49038_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49038_cast_fp16 = einsum(equation = var_49038_equation_0, values = (var_48500_cast_fp16, var_48900_cast_fp16))[name = tensor("op_49038_cast_fp16")]; + tensor var_49040_equation_0 = const()[name = tensor("op_49040_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49040_cast_fp16 = einsum(equation = var_49040_equation_0, values = (var_48500_cast_fp16, var_48901_cast_fp16))[name = tensor("op_49040_cast_fp16")]; + tensor var_49042_equation_0 = const()[name = tensor("op_49042_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49042_cast_fp16 = einsum(equation = var_49042_equation_0, values = (var_48500_cast_fp16, var_48902_cast_fp16))[name = tensor("op_49042_cast_fp16")]; + tensor var_49044_equation_0 = const()[name = tensor("op_49044_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49044_cast_fp16 = einsum(equation = var_49044_equation_0, values = (var_48504_cast_fp16, var_48903_cast_fp16))[name = tensor("op_49044_cast_fp16")]; + tensor var_49046_equation_0 = const()[name = tensor("op_49046_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49046_cast_fp16 = einsum(equation = var_49046_equation_0, values = (var_48504_cast_fp16, var_48904_cast_fp16))[name = tensor("op_49046_cast_fp16")]; + tensor var_49048_equation_0 = const()[name = tensor("op_49048_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49048_cast_fp16 = einsum(equation = var_49048_equation_0, values = (var_48504_cast_fp16, var_48905_cast_fp16))[name = tensor("op_49048_cast_fp16")]; + tensor var_49050_equation_0 = const()[name = tensor("op_49050_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49050_cast_fp16 = einsum(equation = var_49050_equation_0, values = (var_48504_cast_fp16, var_48906_cast_fp16))[name = tensor("op_49050_cast_fp16")]; + tensor var_49052_equation_0 = const()[name = tensor("op_49052_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49052_cast_fp16 = einsum(equation = var_49052_equation_0, values = (var_48508_cast_fp16, var_48907_cast_fp16))[name = tensor("op_49052_cast_fp16")]; + tensor var_49054_equation_0 = const()[name = tensor("op_49054_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49054_cast_fp16 = einsum(equation = var_49054_equation_0, values = (var_48508_cast_fp16, var_48908_cast_fp16))[name = tensor("op_49054_cast_fp16")]; + tensor var_49056_equation_0 = const()[name = tensor("op_49056_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49056_cast_fp16 = einsum(equation = var_49056_equation_0, values = (var_48508_cast_fp16, var_48909_cast_fp16))[name = tensor("op_49056_cast_fp16")]; + tensor var_49058_equation_0 = const()[name = tensor("op_49058_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49058_cast_fp16 = einsum(equation = var_49058_equation_0, values = (var_48508_cast_fp16, var_48910_cast_fp16))[name = tensor("op_49058_cast_fp16")]; + tensor var_49060_equation_0 = const()[name = tensor("op_49060_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49060_cast_fp16 = einsum(equation = var_49060_equation_0, values = (var_48512_cast_fp16, var_48911_cast_fp16))[name = tensor("op_49060_cast_fp16")]; + tensor var_49062_equation_0 = const()[name = tensor("op_49062_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49062_cast_fp16 = einsum(equation = var_49062_equation_0, values = (var_48512_cast_fp16, var_48912_cast_fp16))[name = tensor("op_49062_cast_fp16")]; + tensor var_49064_equation_0 = const()[name = tensor("op_49064_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49064_cast_fp16 = einsum(equation = var_49064_equation_0, values = (var_48512_cast_fp16, var_48913_cast_fp16))[name = tensor("op_49064_cast_fp16")]; + tensor var_49066_equation_0 = const()[name = tensor("op_49066_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49066_cast_fp16 = einsum(equation = var_49066_equation_0, values = (var_48512_cast_fp16, var_48914_cast_fp16))[name = tensor("op_49066_cast_fp16")]; + tensor var_49068_equation_0 = const()[name = tensor("op_49068_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49068_cast_fp16 = einsum(equation = var_49068_equation_0, values = (var_48516_cast_fp16, var_48915_cast_fp16))[name = tensor("op_49068_cast_fp16")]; + tensor var_49070_equation_0 = const()[name = tensor("op_49070_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49070_cast_fp16 = einsum(equation = var_49070_equation_0, values = (var_48516_cast_fp16, var_48916_cast_fp16))[name = tensor("op_49070_cast_fp16")]; + tensor var_49072_equation_0 = const()[name = tensor("op_49072_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49072_cast_fp16 = einsum(equation = var_49072_equation_0, values = (var_48516_cast_fp16, var_48917_cast_fp16))[name = tensor("op_49072_cast_fp16")]; + tensor var_49074_equation_0 = const()[name = tensor("op_49074_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49074_cast_fp16 = einsum(equation = var_49074_equation_0, values = (var_48516_cast_fp16, var_48918_cast_fp16))[name = tensor("op_49074_cast_fp16")]; + tensor var_49076_equation_0 = const()[name = tensor("op_49076_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49076_cast_fp16 = einsum(equation = var_49076_equation_0, values = (var_48520_cast_fp16, var_48919_cast_fp16))[name = tensor("op_49076_cast_fp16")]; + tensor var_49078_equation_0 = const()[name = tensor("op_49078_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49078_cast_fp16 = einsum(equation = var_49078_equation_0, values = (var_48520_cast_fp16, var_48920_cast_fp16))[name = tensor("op_49078_cast_fp16")]; + tensor var_49080_equation_0 = const()[name = tensor("op_49080_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49080_cast_fp16 = einsum(equation = var_49080_equation_0, values = (var_48520_cast_fp16, var_48921_cast_fp16))[name = tensor("op_49080_cast_fp16")]; + tensor var_49082_equation_0 = const()[name = tensor("op_49082_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_49082_cast_fp16 = einsum(equation = var_49082_equation_0, values = (var_48520_cast_fp16, var_48922_cast_fp16))[name = tensor("op_49082_cast_fp16")]; + tensor var_49084_interleave_0 = const()[name = tensor("op_49084_interleave_0"), val = tensor(false)]; + tensor var_49084_cast_fp16 = concat(axis = var_47589, interleave = var_49084_interleave_0, values = (var_48924_cast_fp16, var_48926_cast_fp16, var_48928_cast_fp16, var_48930_cast_fp16))[name = tensor("op_49084_cast_fp16")]; + tensor var_49086_interleave_0 = const()[name = tensor("op_49086_interleave_0"), val = tensor(false)]; + tensor var_49086_cast_fp16 = concat(axis = var_47589, interleave = var_49086_interleave_0, values = (var_48932_cast_fp16, var_48934_cast_fp16, var_48936_cast_fp16, var_48938_cast_fp16))[name = tensor("op_49086_cast_fp16")]; + tensor var_49088_interleave_0 = const()[name = tensor("op_49088_interleave_0"), val = tensor(false)]; + tensor var_49088_cast_fp16 = concat(axis = var_47589, interleave = var_49088_interleave_0, values = (var_48940_cast_fp16, var_48942_cast_fp16, var_48944_cast_fp16, var_48946_cast_fp16))[name = tensor("op_49088_cast_fp16")]; + tensor var_49090_interleave_0 = const()[name = tensor("op_49090_interleave_0"), val = tensor(false)]; + tensor var_49090_cast_fp16 = concat(axis = var_47589, interleave = var_49090_interleave_0, values = (var_48948_cast_fp16, var_48950_cast_fp16, var_48952_cast_fp16, var_48954_cast_fp16))[name = tensor("op_49090_cast_fp16")]; + tensor var_49092_interleave_0 = const()[name = tensor("op_49092_interleave_0"), val = tensor(false)]; + tensor var_49092_cast_fp16 = concat(axis = var_47589, interleave = var_49092_interleave_0, values = (var_48956_cast_fp16, var_48958_cast_fp16, var_48960_cast_fp16, var_48962_cast_fp16))[name = tensor("op_49092_cast_fp16")]; + tensor var_49094_interleave_0 = const()[name = tensor("op_49094_interleave_0"), val = tensor(false)]; + tensor var_49094_cast_fp16 = concat(axis = var_47589, interleave = var_49094_interleave_0, values = (var_48964_cast_fp16, var_48966_cast_fp16, var_48968_cast_fp16, var_48970_cast_fp16))[name = tensor("op_49094_cast_fp16")]; + tensor var_49096_interleave_0 = const()[name = tensor("op_49096_interleave_0"), val = tensor(false)]; + tensor var_49096_cast_fp16 = concat(axis = var_47589, interleave = var_49096_interleave_0, values = (var_48972_cast_fp16, var_48974_cast_fp16, var_48976_cast_fp16, var_48978_cast_fp16))[name = tensor("op_49096_cast_fp16")]; + tensor var_49098_interleave_0 = const()[name = tensor("op_49098_interleave_0"), val = tensor(false)]; + tensor var_49098_cast_fp16 = concat(axis = var_47589, interleave = var_49098_interleave_0, values = (var_48980_cast_fp16, var_48982_cast_fp16, var_48984_cast_fp16, var_48986_cast_fp16))[name = tensor("op_49098_cast_fp16")]; + tensor var_49100_interleave_0 = const()[name = tensor("op_49100_interleave_0"), val = tensor(false)]; + tensor var_49100_cast_fp16 = concat(axis = var_47589, interleave = var_49100_interleave_0, values = (var_48988_cast_fp16, var_48990_cast_fp16, var_48992_cast_fp16, var_48994_cast_fp16))[name = tensor("op_49100_cast_fp16")]; + tensor var_49102_interleave_0 = const()[name = tensor("op_49102_interleave_0"), val = tensor(false)]; + tensor var_49102_cast_fp16 = concat(axis = var_47589, interleave = var_49102_interleave_0, values = (var_48996_cast_fp16, var_48998_cast_fp16, var_49000_cast_fp16, var_49002_cast_fp16))[name = tensor("op_49102_cast_fp16")]; + tensor var_49104_interleave_0 = const()[name = tensor("op_49104_interleave_0"), val = tensor(false)]; + tensor var_49104_cast_fp16 = concat(axis = var_47589, interleave = var_49104_interleave_0, values = (var_49004_cast_fp16, var_49006_cast_fp16, var_49008_cast_fp16, var_49010_cast_fp16))[name = tensor("op_49104_cast_fp16")]; + tensor var_49106_interleave_0 = const()[name = tensor("op_49106_interleave_0"), val = tensor(false)]; + tensor var_49106_cast_fp16 = concat(axis = var_47589, interleave = var_49106_interleave_0, values = (var_49012_cast_fp16, var_49014_cast_fp16, var_49016_cast_fp16, var_49018_cast_fp16))[name = tensor("op_49106_cast_fp16")]; + tensor var_49108_interleave_0 = const()[name = tensor("op_49108_interleave_0"), val = tensor(false)]; + tensor var_49108_cast_fp16 = concat(axis = var_47589, interleave = var_49108_interleave_0, values = (var_49020_cast_fp16, var_49022_cast_fp16, var_49024_cast_fp16, var_49026_cast_fp16))[name = tensor("op_49108_cast_fp16")]; + tensor var_49110_interleave_0 = const()[name = tensor("op_49110_interleave_0"), val = tensor(false)]; + tensor var_49110_cast_fp16 = concat(axis = var_47589, interleave = var_49110_interleave_0, values = (var_49028_cast_fp16, var_49030_cast_fp16, var_49032_cast_fp16, var_49034_cast_fp16))[name = tensor("op_49110_cast_fp16")]; + tensor var_49112_interleave_0 = const()[name = tensor("op_49112_interleave_0"), val = tensor(false)]; + tensor var_49112_cast_fp16 = concat(axis = var_47589, interleave = var_49112_interleave_0, values = (var_49036_cast_fp16, var_49038_cast_fp16, var_49040_cast_fp16, var_49042_cast_fp16))[name = tensor("op_49112_cast_fp16")]; + tensor var_49114_interleave_0 = const()[name = tensor("op_49114_interleave_0"), val = tensor(false)]; + tensor var_49114_cast_fp16 = concat(axis = var_47589, interleave = var_49114_interleave_0, values = (var_49044_cast_fp16, var_49046_cast_fp16, var_49048_cast_fp16, var_49050_cast_fp16))[name = tensor("op_49114_cast_fp16")]; + tensor var_49116_interleave_0 = const()[name = tensor("op_49116_interleave_0"), val = tensor(false)]; + tensor var_49116_cast_fp16 = concat(axis = var_47589, interleave = var_49116_interleave_0, values = (var_49052_cast_fp16, var_49054_cast_fp16, var_49056_cast_fp16, var_49058_cast_fp16))[name = tensor("op_49116_cast_fp16")]; + tensor var_49118_interleave_0 = const()[name = tensor("op_49118_interleave_0"), val = tensor(false)]; + tensor var_49118_cast_fp16 = concat(axis = var_47589, interleave = var_49118_interleave_0, values = (var_49060_cast_fp16, var_49062_cast_fp16, var_49064_cast_fp16, var_49066_cast_fp16))[name = tensor("op_49118_cast_fp16")]; + tensor var_49120_interleave_0 = const()[name = tensor("op_49120_interleave_0"), val = tensor(false)]; + tensor var_49120_cast_fp16 = concat(axis = var_47589, interleave = var_49120_interleave_0, values = (var_49068_cast_fp16, var_49070_cast_fp16, var_49072_cast_fp16, var_49074_cast_fp16))[name = tensor("op_49120_cast_fp16")]; + tensor var_49122_interleave_0 = const()[name = tensor("op_49122_interleave_0"), val = tensor(false)]; + tensor var_49122_cast_fp16 = concat(axis = var_47589, interleave = var_49122_interleave_0, values = (var_49076_cast_fp16, var_49078_cast_fp16, var_49080_cast_fp16, var_49082_cast_fp16))[name = tensor("op_49122_cast_fp16")]; + tensor input_587_interleave_0 = const()[name = tensor("input_587_interleave_0"), val = tensor(false)]; + tensor input_587_cast_fp16 = concat(axis = var_47614, interleave = input_587_interleave_0, values = (var_49084_cast_fp16, var_49086_cast_fp16, var_49088_cast_fp16, var_49090_cast_fp16, var_49092_cast_fp16, var_49094_cast_fp16, var_49096_cast_fp16, var_49098_cast_fp16, var_49100_cast_fp16, var_49102_cast_fp16, var_49104_cast_fp16, var_49106_cast_fp16, var_49108_cast_fp16, var_49110_cast_fp16, var_49112_cast_fp16, var_49114_cast_fp16, var_49116_cast_fp16, var_49118_cast_fp16, var_49120_cast_fp16, var_49122_cast_fp16))[name = tensor("input_587_cast_fp16")]; + tensor var_49130 = const()[name = tensor("op_49130"), val = tensor([1, 1])]; + tensor var_49132 = const()[name = tensor("op_49132"), val = tensor([1, 1])]; + tensor pretrained_out_355_pad_type_0 = const()[name = tensor("pretrained_out_355_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_355_pad_0 = const()[name = tensor("pretrained_out_355_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(324446656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325265920))), name = tensor("layers_29_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_29_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_29_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325266048)))]; + tensor pretrained_out_355_cast_fp16 = conv(bias = layers_29_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_49132, groups = var_47614, pad = pretrained_out_355_pad_0, pad_type = pretrained_out_355_pad_type_0, strides = var_49130, weight = layers_29_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_587_cast_fp16)[name = tensor("pretrained_out_355_cast_fp16")]; + tensor var_49136 = const()[name = tensor("op_49136"), val = tensor([1, 1])]; + tensor var_49138 = const()[name = tensor("op_49138"), val = tensor([1, 1])]; + tensor input_589_pad_type_0 = const()[name = tensor("input_589_pad_type_0"), val = tensor("custom")]; + tensor input_589_pad_0 = const()[name = tensor("input_589_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_29_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325268672)))]; + tensor input_589_cast_fp16 = conv(dilations = var_49138, groups = var_47614, pad = input_589_pad_0, pad_type = input_589_pad_type_0, strides = var_49136, weight = layers_29_self_attn_o_proj_loraA_weight_to_fp16, x = input_587_cast_fp16)[name = tensor("input_589_cast_fp16")]; + tensor var_49142 = const()[name = tensor("op_49142"), val = tensor([1, 1])]; + tensor var_49144 = const()[name = tensor("op_49144"), val = tensor([1, 1])]; + tensor lora_out_709_pad_type_0 = const()[name = tensor("lora_out_709_pad_type_0"), val = tensor("custom")]; + tensor lora_out_709_pad_0 = const()[name = tensor("lora_out_709_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_711_weight_0_to_fp16 = const()[name = tensor("lora_out_711_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325309696)))]; + tensor lora_out_711_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_49144, groups = var_47614, pad = lora_out_709_pad_0, pad_type = lora_out_709_pad_type_0, strides = var_49142, weight = lora_out_711_weight_0_to_fp16, x = input_589_cast_fp16)[name = tensor("lora_out_711_cast_fp16")]; + tensor obj_119_cast_fp16 = add(x = pretrained_out_355_cast_fp16, y = lora_out_711_cast_fp16)[name = tensor("obj_119_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = obj_119_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; + tensor var_49153 = const()[name = tensor("op_49153"), val = tensor([1])]; + tensor channels_mean_119_cast_fp16 = reduce_mean(axes = var_49153, keep_dims = var_47615, x = inputs_119_cast_fp16)[name = tensor("channels_mean_119_cast_fp16")]; + tensor zero_mean_119_cast_fp16 = sub(x = inputs_119_cast_fp16, y = channels_mean_119_cast_fp16)[name = tensor("zero_mean_119_cast_fp16")]; + tensor zero_mean_sq_119_cast_fp16 = mul(x = zero_mean_119_cast_fp16, y = zero_mean_119_cast_fp16)[name = tensor("zero_mean_sq_119_cast_fp16")]; + tensor var_49157 = const()[name = tensor("op_49157"), val = tensor([1])]; + tensor var_49158_cast_fp16 = reduce_mean(axes = var_49157, keep_dims = var_47615, x = zero_mean_sq_119_cast_fp16)[name = tensor("op_49158_cast_fp16")]; + tensor var_49159_to_fp16 = const()[name = tensor("op_49159_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_49160_cast_fp16 = add(x = var_49158_cast_fp16, y = var_49159_to_fp16)[name = tensor("op_49160_cast_fp16")]; + tensor denom_119_epsilon_0 = const()[name = tensor("denom_119_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_119_cast_fp16 = rsqrt(epsilon = denom_119_epsilon_0, x = var_49160_cast_fp16)[name = tensor("denom_119_cast_fp16")]; + tensor out_119_cast_fp16 = mul(x = zero_mean_119_cast_fp16, y = denom_119_cast_fp16)[name = tensor("out_119_cast_fp16")]; + tensor input_591_gamma_0_to_fp16 = const()[name = tensor("input_591_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325350720)))]; + tensor input_591_beta_0_to_fp16 = const()[name = tensor("input_591_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325353344)))]; + tensor input_591_epsilon_0_to_fp16 = const()[name = tensor("input_591_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_591_cast_fp16 = batch_norm(beta = input_591_beta_0_to_fp16, epsilon = input_591_epsilon_0_to_fp16, gamma = input_591_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_119_cast_fp16)[name = tensor("input_591_cast_fp16")]; + tensor var_49174 = const()[name = tensor("op_49174"), val = tensor([1, 1])]; + tensor var_49176 = const()[name = tensor("op_49176"), val = tensor([1, 1])]; + tensor pretrained_out_357_pad_type_0 = const()[name = tensor("pretrained_out_357_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_357_pad_0 = const()[name = tensor("pretrained_out_357_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(325355968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328632832))), name = tensor("layers_29_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_29_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_29_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328632960)))]; + tensor pretrained_out_357_cast_fp16 = conv(bias = layers_29_fc1_pretrained_bias_to_fp16, dilations = var_49176, groups = var_47614, pad = pretrained_out_357_pad_0, pad_type = pretrained_out_357_pad_type_0, strides = var_49174, weight = layers_29_fc1_pretrained_weight_to_fp16_palettized, x = input_591_cast_fp16)[name = tensor("pretrained_out_357_cast_fp16")]; + tensor var_49180 = const()[name = tensor("op_49180"), val = tensor([1, 1])]; + tensor var_49182 = const()[name = tensor("op_49182"), val = tensor([1, 1])]; + tensor input_593_pad_type_0 = const()[name = tensor("input_593_pad_type_0"), val = tensor("custom")]; + tensor input_593_pad_0 = const()[name = tensor("input_593_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_29_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328643264)))]; + tensor input_593_cast_fp16 = conv(dilations = var_49182, groups = var_47614, pad = input_593_pad_0, pad_type = input_593_pad_type_0, strides = var_49180, weight = layers_29_fc1_loraA_weight_to_fp16, x = input_591_cast_fp16)[name = tensor("input_593_cast_fp16")]; + tensor var_49186 = const()[name = tensor("op_49186"), val = tensor([1, 1])]; + tensor var_49188 = const()[name = tensor("op_49188"), val = tensor([1, 1])]; + tensor lora_out_713_pad_type_0 = const()[name = tensor("lora_out_713_pad_type_0"), val = tensor("custom")]; + tensor lora_out_713_pad_0 = const()[name = tensor("lora_out_713_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_715_weight_0_to_fp16 = const()[name = tensor("lora_out_715_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328684288)))]; + tensor lora_out_715_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_49188, groups = var_47614, pad = lora_out_713_pad_0, pad_type = lora_out_713_pad_type_0, strides = var_49186, weight = lora_out_715_weight_0_to_fp16, x = input_593_cast_fp16)[name = tensor("lora_out_715_cast_fp16")]; + tensor input_595_cast_fp16 = add(x = pretrained_out_357_cast_fp16, y = lora_out_715_cast_fp16)[name = tensor("input_595_cast_fp16")]; + tensor input_597_mode_0 = const()[name = tensor("input_597_mode_0"), val = tensor("EXACT")]; + tensor input_597_cast_fp16 = gelu(mode = input_597_mode_0, x = input_595_cast_fp16)[name = tensor("input_597_cast_fp16")]; + tensor var_49200 = const()[name = tensor("op_49200"), val = tensor([1, 1])]; + tensor var_49202 = const()[name = tensor("op_49202"), val = tensor([1, 1])]; + tensor pretrained_out_359_pad_type_0 = const()[name = tensor("pretrained_out_359_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_359_pad_0 = const()[name = tensor("pretrained_out_359_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328848192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332125056))), name = tensor("layers_29_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_29_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_29_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332125184)))]; + tensor pretrained_out_359_cast_fp16 = conv(bias = layers_29_fc2_pretrained_bias_to_fp16, dilations = var_49202, groups = var_47614, pad = pretrained_out_359_pad_0, pad_type = pretrained_out_359_pad_type_0, strides = var_49200, weight = layers_29_fc2_pretrained_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = tensor("pretrained_out_359_cast_fp16")]; + tensor var_49206 = const()[name = tensor("op_49206"), val = tensor([1, 1])]; + tensor var_49208 = const()[name = tensor("op_49208"), val = tensor([1, 1])]; + tensor input_599_pad_type_0 = const()[name = tensor("input_599_pad_type_0"), val = tensor("custom")]; + tensor input_599_pad_0 = const()[name = tensor("input_599_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_29_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_29_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332127808)))]; + tensor input_599_cast_fp16 = conv(dilations = var_49208, groups = var_47614, pad = input_599_pad_0, pad_type = input_599_pad_type_0, strides = var_49206, weight = layers_29_fc2_loraA_weight_to_fp16, x = input_597_cast_fp16)[name = tensor("input_599_cast_fp16")]; + tensor var_49212 = const()[name = tensor("op_49212"), val = tensor([1, 1])]; + tensor var_49214 = const()[name = tensor("op_49214"), val = tensor([1, 1])]; + tensor lora_out_717_pad_type_0 = const()[name = tensor("lora_out_717_pad_type_0"), val = tensor("custom")]; + tensor lora_out_717_pad_0 = const()[name = tensor("lora_out_717_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_719_weight_0_to_fp16 = const()[name = tensor("lora_out_719_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332291712)))]; + tensor lora_out_719_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_49214, groups = var_47614, pad = lora_out_717_pad_0, pad_type = lora_out_717_pad_type_0, strides = var_49212, weight = lora_out_719_weight_0_to_fp16, x = input_599_cast_fp16)[name = tensor("lora_out_719_cast_fp16")]; + tensor hidden_states_63_cast_fp16 = add(x = pretrained_out_359_cast_fp16, y = lora_out_719_cast_fp16)[name = tensor("hidden_states_63_cast_fp16")]; + tensor inputs_121_cast_fp16 = add(x = inputs_119_cast_fp16, y = hidden_states_63_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; + tensor var_49224 = const()[name = tensor("op_49224"), val = tensor(3)]; + tensor var_49249 = const()[name = tensor("op_49249"), val = tensor(1)]; + tensor var_49250 = const()[name = tensor("op_49250"), val = tensor(true)]; + tensor var_49260 = const()[name = tensor("op_49260"), val = tensor([1])]; + tensor channels_mean_121_cast_fp16 = reduce_mean(axes = var_49260, keep_dims = var_49250, x = inputs_121_cast_fp16)[name = tensor("channels_mean_121_cast_fp16")]; + tensor zero_mean_121_cast_fp16 = sub(x = inputs_121_cast_fp16, y = channels_mean_121_cast_fp16)[name = tensor("zero_mean_121_cast_fp16")]; + tensor zero_mean_sq_121_cast_fp16 = mul(x = zero_mean_121_cast_fp16, y = zero_mean_121_cast_fp16)[name = tensor("zero_mean_sq_121_cast_fp16")]; + tensor var_49264 = const()[name = tensor("op_49264"), val = tensor([1])]; + tensor var_49265_cast_fp16 = reduce_mean(axes = var_49264, keep_dims = var_49250, x = zero_mean_sq_121_cast_fp16)[name = tensor("op_49265_cast_fp16")]; + tensor var_49266_to_fp16 = const()[name = tensor("op_49266_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_49267_cast_fp16 = add(x = var_49265_cast_fp16, y = var_49266_to_fp16)[name = tensor("op_49267_cast_fp16")]; + tensor denom_121_epsilon_0 = const()[name = tensor("denom_121_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_121_cast_fp16 = rsqrt(epsilon = denom_121_epsilon_0, x = var_49267_cast_fp16)[name = tensor("denom_121_cast_fp16")]; + tensor out_121_cast_fp16 = mul(x = zero_mean_121_cast_fp16, y = denom_121_cast_fp16)[name = tensor("out_121_cast_fp16")]; + tensor obj_121_gamma_0_to_fp16 = const()[name = tensor("obj_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332332736)))]; + tensor obj_121_beta_0_to_fp16 = const()[name = tensor("obj_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332335360)))]; + tensor obj_121_epsilon_0_to_fp16 = const()[name = tensor("obj_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_121_cast_fp16 = batch_norm(beta = obj_121_beta_0_to_fp16, epsilon = obj_121_epsilon_0_to_fp16, gamma = obj_121_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_121_cast_fp16)[name = tensor("obj_121_cast_fp16")]; + tensor var_49285 = const()[name = tensor("op_49285"), val = tensor([1, 1])]; + tensor var_49287 = const()[name = tensor("op_49287"), val = tensor([1, 1])]; + tensor pretrained_out_361_pad_type_0 = const()[name = tensor("pretrained_out_361_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_361_pad_0 = const()[name = tensor("pretrained_out_361_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332337984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(333157248))), name = tensor("layers_30_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_30_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(333157376)))]; + tensor pretrained_out_361_cast_fp16 = conv(bias = layers_30_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_49287, groups = var_49249, pad = pretrained_out_361_pad_0, pad_type = pretrained_out_361_pad_type_0, strides = var_49285, weight = layers_30_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_121_cast_fp16)[name = tensor("pretrained_out_361_cast_fp16")]; + tensor var_49291 = const()[name = tensor("op_49291"), val = tensor([1, 1])]; + tensor var_49293 = const()[name = tensor("op_49293"), val = tensor([1, 1])]; + tensor input_601_pad_type_0 = const()[name = tensor("input_601_pad_type_0"), val = tensor("custom")]; + tensor input_601_pad_0 = const()[name = tensor("input_601_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(333160000)))]; + tensor input_601_cast_fp16 = conv(dilations = var_49293, groups = var_49249, pad = input_601_pad_0, pad_type = input_601_pad_type_0, strides = var_49291, weight = layers_30_self_attn_q_proj_loraA_weight_to_fp16, x = obj_121_cast_fp16)[name = tensor("input_601_cast_fp16")]; + tensor var_49297 = const()[name = tensor("op_49297"), val = tensor([1, 1])]; + tensor var_49299 = const()[name = tensor("op_49299"), val = tensor([1, 1])]; + tensor lora_out_721_pad_type_0 = const()[name = tensor("lora_out_721_pad_type_0"), val = tensor("custom")]; + tensor lora_out_721_pad_0 = const()[name = tensor("lora_out_721_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_723_weight_0_to_fp16 = const()[name = tensor("lora_out_723_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(333201024)))]; + tensor lora_out_723_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_49299, groups = var_49249, pad = lora_out_721_pad_0, pad_type = lora_out_721_pad_type_0, strides = var_49297, weight = lora_out_723_weight_0_to_fp16, x = input_601_cast_fp16)[name = tensor("lora_out_723_cast_fp16")]; + tensor query_61_cast_fp16 = add(x = pretrained_out_361_cast_fp16, y = lora_out_723_cast_fp16)[name = tensor("query_61_cast_fp16")]; + tensor var_49309 = const()[name = tensor("op_49309"), val = tensor([1, 1])]; + tensor var_49311 = const()[name = tensor("op_49311"), val = tensor([1, 1])]; + tensor pretrained_out_363_pad_type_0 = const()[name = tensor("pretrained_out_363_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_363_pad_0 = const()[name = tensor("pretrained_out_363_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(333242048))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334061312))), name = tensor("layers_30_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_363_cast_fp16 = conv(dilations = var_49311, groups = var_49249, pad = pretrained_out_363_pad_0, pad_type = pretrained_out_363_pad_type_0, strides = var_49309, weight = layers_30_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_121_cast_fp16)[name = tensor("pretrained_out_363_cast_fp16")]; + tensor var_49315 = const()[name = tensor("op_49315"), val = tensor([1, 1])]; + tensor var_49317 = const()[name = tensor("op_49317"), val = tensor([1, 1])]; + tensor input_603_pad_type_0 = const()[name = tensor("input_603_pad_type_0"), val = tensor("custom")]; + tensor input_603_pad_0 = const()[name = tensor("input_603_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334061440)))]; + tensor input_603_cast_fp16 = conv(dilations = var_49317, groups = var_49249, pad = input_603_pad_0, pad_type = input_603_pad_type_0, strides = var_49315, weight = layers_30_self_attn_k_proj_loraA_weight_to_fp16, x = obj_121_cast_fp16)[name = tensor("input_603_cast_fp16")]; + tensor var_49321 = const()[name = tensor("op_49321"), val = tensor([1, 1])]; + tensor var_49323 = const()[name = tensor("op_49323"), val = tensor([1, 1])]; + tensor lora_out_725_pad_type_0 = const()[name = tensor("lora_out_725_pad_type_0"), val = tensor("custom")]; + tensor lora_out_725_pad_0 = const()[name = tensor("lora_out_725_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_727_weight_0_to_fp16 = const()[name = tensor("lora_out_727_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334102464)))]; + tensor lora_out_727_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_49323, groups = var_49249, pad = lora_out_725_pad_0, pad_type = lora_out_725_pad_type_0, strides = var_49321, weight = lora_out_727_weight_0_to_fp16, x = input_603_cast_fp16)[name = tensor("lora_out_727_cast_fp16")]; + tensor key_61_cast_fp16 = add(x = pretrained_out_363_cast_fp16, y = lora_out_727_cast_fp16)[name = tensor("key_61_cast_fp16")]; + tensor var_49334 = const()[name = tensor("op_49334"), val = tensor([1, 1])]; + tensor var_49336 = const()[name = tensor("op_49336"), val = tensor([1, 1])]; + tensor pretrained_out_365_pad_type_0 = const()[name = tensor("pretrained_out_365_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_365_pad_0 = const()[name = tensor("pretrained_out_365_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334143488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334962752))), name = tensor("layers_30_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_30_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334962880)))]; + tensor pretrained_out_365_cast_fp16 = conv(bias = layers_30_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_49336, groups = var_49249, pad = pretrained_out_365_pad_0, pad_type = pretrained_out_365_pad_type_0, strides = var_49334, weight = layers_30_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_121_cast_fp16)[name = tensor("pretrained_out_365_cast_fp16")]; + tensor var_49340 = const()[name = tensor("op_49340"), val = tensor([1, 1])]; + tensor var_49342 = const()[name = tensor("op_49342"), val = tensor([1, 1])]; + tensor input_605_pad_type_0 = const()[name = tensor("input_605_pad_type_0"), val = tensor("custom")]; + tensor input_605_pad_0 = const()[name = tensor("input_605_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334965504)))]; + tensor input_605_cast_fp16 = conv(dilations = var_49342, groups = var_49249, pad = input_605_pad_0, pad_type = input_605_pad_type_0, strides = var_49340, weight = layers_30_self_attn_v_proj_loraA_weight_to_fp16, x = obj_121_cast_fp16)[name = tensor("input_605_cast_fp16")]; + tensor var_49346 = const()[name = tensor("op_49346"), val = tensor([1, 1])]; + tensor var_49348 = const()[name = tensor("op_49348"), val = tensor([1, 1])]; + tensor lora_out_729_pad_type_0 = const()[name = tensor("lora_out_729_pad_type_0"), val = tensor("custom")]; + tensor lora_out_729_pad_0 = const()[name = tensor("lora_out_729_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_731_weight_0_to_fp16 = const()[name = tensor("lora_out_731_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335006528)))]; + tensor lora_out_731_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_49348, groups = var_49249, pad = lora_out_729_pad_0, pad_type = lora_out_729_pad_type_0, strides = var_49346, weight = lora_out_731_weight_0_to_fp16, x = input_605_cast_fp16)[name = tensor("lora_out_731_cast_fp16")]; + tensor value_61_cast_fp16 = add(x = pretrained_out_365_cast_fp16, y = lora_out_731_cast_fp16)[name = tensor("value_61_cast_fp16")]; + tensor var_49358_begin_0 = const()[name = tensor("op_49358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49358_end_0 = const()[name = tensor("op_49358_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49358_end_mask_0 = const()[name = tensor("op_49358_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49358_cast_fp16 = slice_by_index(begin = var_49358_begin_0, end = var_49358_end_0, end_mask = var_49358_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49358_cast_fp16")]; + tensor var_49362_begin_0 = const()[name = tensor("op_49362_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_49362_end_0 = const()[name = tensor("op_49362_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_49362_end_mask_0 = const()[name = tensor("op_49362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49362_cast_fp16 = slice_by_index(begin = var_49362_begin_0, end = var_49362_end_0, end_mask = var_49362_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49362_cast_fp16")]; + tensor var_49366_begin_0 = const()[name = tensor("op_49366_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_49366_end_0 = const()[name = tensor("op_49366_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_49366_end_mask_0 = const()[name = tensor("op_49366_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49366_cast_fp16 = slice_by_index(begin = var_49366_begin_0, end = var_49366_end_0, end_mask = var_49366_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49366_cast_fp16")]; + tensor var_49370_begin_0 = const()[name = tensor("op_49370_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_49370_end_0 = const()[name = tensor("op_49370_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_49370_end_mask_0 = const()[name = tensor("op_49370_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49370_cast_fp16 = slice_by_index(begin = var_49370_begin_0, end = var_49370_end_0, end_mask = var_49370_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49370_cast_fp16")]; + tensor var_49374_begin_0 = const()[name = tensor("op_49374_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_49374_end_0 = const()[name = tensor("op_49374_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_49374_end_mask_0 = const()[name = tensor("op_49374_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49374_cast_fp16 = slice_by_index(begin = var_49374_begin_0, end = var_49374_end_0, end_mask = var_49374_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49374_cast_fp16")]; + tensor var_49378_begin_0 = const()[name = tensor("op_49378_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_49378_end_0 = const()[name = tensor("op_49378_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_49378_end_mask_0 = const()[name = tensor("op_49378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49378_cast_fp16 = slice_by_index(begin = var_49378_begin_0, end = var_49378_end_0, end_mask = var_49378_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49378_cast_fp16")]; + tensor var_49382_begin_0 = const()[name = tensor("op_49382_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_49382_end_0 = const()[name = tensor("op_49382_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_49382_end_mask_0 = const()[name = tensor("op_49382_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49382_cast_fp16 = slice_by_index(begin = var_49382_begin_0, end = var_49382_end_0, end_mask = var_49382_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49382_cast_fp16")]; + tensor var_49386_begin_0 = const()[name = tensor("op_49386_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_49386_end_0 = const()[name = tensor("op_49386_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_49386_end_mask_0 = const()[name = tensor("op_49386_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49386_cast_fp16 = slice_by_index(begin = var_49386_begin_0, end = var_49386_end_0, end_mask = var_49386_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49386_cast_fp16")]; + tensor var_49390_begin_0 = const()[name = tensor("op_49390_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_49390_end_0 = const()[name = tensor("op_49390_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_49390_end_mask_0 = const()[name = tensor("op_49390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49390_cast_fp16 = slice_by_index(begin = var_49390_begin_0, end = var_49390_end_0, end_mask = var_49390_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49390_cast_fp16")]; + tensor var_49394_begin_0 = const()[name = tensor("op_49394_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_49394_end_0 = const()[name = tensor("op_49394_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_49394_end_mask_0 = const()[name = tensor("op_49394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49394_cast_fp16 = slice_by_index(begin = var_49394_begin_0, end = var_49394_end_0, end_mask = var_49394_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49394_cast_fp16")]; + tensor var_49398_begin_0 = const()[name = tensor("op_49398_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_49398_end_0 = const()[name = tensor("op_49398_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_49398_end_mask_0 = const()[name = tensor("op_49398_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49398_cast_fp16 = slice_by_index(begin = var_49398_begin_0, end = var_49398_end_0, end_mask = var_49398_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49398_cast_fp16")]; + tensor var_49402_begin_0 = const()[name = tensor("op_49402_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_49402_end_0 = const()[name = tensor("op_49402_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_49402_end_mask_0 = const()[name = tensor("op_49402_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49402_cast_fp16 = slice_by_index(begin = var_49402_begin_0, end = var_49402_end_0, end_mask = var_49402_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49402_cast_fp16")]; + tensor var_49406_begin_0 = const()[name = tensor("op_49406_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_49406_end_0 = const()[name = tensor("op_49406_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_49406_end_mask_0 = const()[name = tensor("op_49406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49406_cast_fp16 = slice_by_index(begin = var_49406_begin_0, end = var_49406_end_0, end_mask = var_49406_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49406_cast_fp16")]; + tensor var_49410_begin_0 = const()[name = tensor("op_49410_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_49410_end_0 = const()[name = tensor("op_49410_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_49410_end_mask_0 = const()[name = tensor("op_49410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49410_cast_fp16 = slice_by_index(begin = var_49410_begin_0, end = var_49410_end_0, end_mask = var_49410_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49410_cast_fp16")]; + tensor var_49414_begin_0 = const()[name = tensor("op_49414_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_49414_end_0 = const()[name = tensor("op_49414_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_49414_end_mask_0 = const()[name = tensor("op_49414_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49414_cast_fp16 = slice_by_index(begin = var_49414_begin_0, end = var_49414_end_0, end_mask = var_49414_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49414_cast_fp16")]; + tensor var_49418_begin_0 = const()[name = tensor("op_49418_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_49418_end_0 = const()[name = tensor("op_49418_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_49418_end_mask_0 = const()[name = tensor("op_49418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49418_cast_fp16 = slice_by_index(begin = var_49418_begin_0, end = var_49418_end_0, end_mask = var_49418_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49418_cast_fp16")]; + tensor var_49422_begin_0 = const()[name = tensor("op_49422_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_49422_end_0 = const()[name = tensor("op_49422_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_49422_end_mask_0 = const()[name = tensor("op_49422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49422_cast_fp16 = slice_by_index(begin = var_49422_begin_0, end = var_49422_end_0, end_mask = var_49422_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49422_cast_fp16")]; + tensor var_49426_begin_0 = const()[name = tensor("op_49426_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_49426_end_0 = const()[name = tensor("op_49426_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_49426_end_mask_0 = const()[name = tensor("op_49426_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49426_cast_fp16 = slice_by_index(begin = var_49426_begin_0, end = var_49426_end_0, end_mask = var_49426_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49426_cast_fp16")]; + tensor var_49430_begin_0 = const()[name = tensor("op_49430_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_49430_end_0 = const()[name = tensor("op_49430_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_49430_end_mask_0 = const()[name = tensor("op_49430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49430_cast_fp16 = slice_by_index(begin = var_49430_begin_0, end = var_49430_end_0, end_mask = var_49430_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49430_cast_fp16")]; + tensor var_49434_begin_0 = const()[name = tensor("op_49434_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_49434_end_0 = const()[name = tensor("op_49434_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_49434_end_mask_0 = const()[name = tensor("op_49434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_49434_cast_fp16 = slice_by_index(begin = var_49434_begin_0, end = var_49434_end_0, end_mask = var_49434_end_mask_0, x = query_61_cast_fp16)[name = tensor("op_49434_cast_fp16")]; + tensor var_49443_begin_0 = const()[name = tensor("op_49443_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49443_end_0 = const()[name = tensor("op_49443_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49443_end_mask_0 = const()[name = tensor("op_49443_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49443_cast_fp16 = slice_by_index(begin = var_49443_begin_0, end = var_49443_end_0, end_mask = var_49443_end_mask_0, x = var_49358_cast_fp16)[name = tensor("op_49443_cast_fp16")]; + tensor var_49450_begin_0 = const()[name = tensor("op_49450_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49450_end_0 = const()[name = tensor("op_49450_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49450_end_mask_0 = const()[name = tensor("op_49450_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49450_cast_fp16 = slice_by_index(begin = var_49450_begin_0, end = var_49450_end_0, end_mask = var_49450_end_mask_0, x = var_49358_cast_fp16)[name = tensor("op_49450_cast_fp16")]; + tensor var_49457_begin_0 = const()[name = tensor("op_49457_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49457_end_0 = const()[name = tensor("op_49457_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49457_end_mask_0 = const()[name = tensor("op_49457_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49457_cast_fp16 = slice_by_index(begin = var_49457_begin_0, end = var_49457_end_0, end_mask = var_49457_end_mask_0, x = var_49358_cast_fp16)[name = tensor("op_49457_cast_fp16")]; + tensor var_49464_begin_0 = const()[name = tensor("op_49464_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49464_end_0 = const()[name = tensor("op_49464_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49464_end_mask_0 = const()[name = tensor("op_49464_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49464_cast_fp16 = slice_by_index(begin = var_49464_begin_0, end = var_49464_end_0, end_mask = var_49464_end_mask_0, x = var_49358_cast_fp16)[name = tensor("op_49464_cast_fp16")]; + tensor var_49471_begin_0 = const()[name = tensor("op_49471_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49471_end_0 = const()[name = tensor("op_49471_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49471_end_mask_0 = const()[name = tensor("op_49471_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49471_cast_fp16 = slice_by_index(begin = var_49471_begin_0, end = var_49471_end_0, end_mask = var_49471_end_mask_0, x = var_49362_cast_fp16)[name = tensor("op_49471_cast_fp16")]; + tensor var_49478_begin_0 = const()[name = tensor("op_49478_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49478_end_0 = const()[name = tensor("op_49478_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49478_end_mask_0 = const()[name = tensor("op_49478_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49478_cast_fp16 = slice_by_index(begin = var_49478_begin_0, end = var_49478_end_0, end_mask = var_49478_end_mask_0, x = var_49362_cast_fp16)[name = tensor("op_49478_cast_fp16")]; + tensor var_49485_begin_0 = const()[name = tensor("op_49485_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49485_end_0 = const()[name = tensor("op_49485_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49485_end_mask_0 = const()[name = tensor("op_49485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49485_cast_fp16 = slice_by_index(begin = var_49485_begin_0, end = var_49485_end_0, end_mask = var_49485_end_mask_0, x = var_49362_cast_fp16)[name = tensor("op_49485_cast_fp16")]; + tensor var_49492_begin_0 = const()[name = tensor("op_49492_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49492_end_0 = const()[name = tensor("op_49492_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49492_end_mask_0 = const()[name = tensor("op_49492_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49492_cast_fp16 = slice_by_index(begin = var_49492_begin_0, end = var_49492_end_0, end_mask = var_49492_end_mask_0, x = var_49362_cast_fp16)[name = tensor("op_49492_cast_fp16")]; + tensor var_49499_begin_0 = const()[name = tensor("op_49499_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49499_end_0 = const()[name = tensor("op_49499_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49499_end_mask_0 = const()[name = tensor("op_49499_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49499_cast_fp16 = slice_by_index(begin = var_49499_begin_0, end = var_49499_end_0, end_mask = var_49499_end_mask_0, x = var_49366_cast_fp16)[name = tensor("op_49499_cast_fp16")]; + tensor var_49506_begin_0 = const()[name = tensor("op_49506_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49506_end_0 = const()[name = tensor("op_49506_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49506_end_mask_0 = const()[name = tensor("op_49506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49506_cast_fp16 = slice_by_index(begin = var_49506_begin_0, end = var_49506_end_0, end_mask = var_49506_end_mask_0, x = var_49366_cast_fp16)[name = tensor("op_49506_cast_fp16")]; + tensor var_49513_begin_0 = const()[name = tensor("op_49513_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49513_end_0 = const()[name = tensor("op_49513_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49513_end_mask_0 = const()[name = tensor("op_49513_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49513_cast_fp16 = slice_by_index(begin = var_49513_begin_0, end = var_49513_end_0, end_mask = var_49513_end_mask_0, x = var_49366_cast_fp16)[name = tensor("op_49513_cast_fp16")]; + tensor var_49520_begin_0 = const()[name = tensor("op_49520_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49520_end_0 = const()[name = tensor("op_49520_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49520_end_mask_0 = const()[name = tensor("op_49520_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49520_cast_fp16 = slice_by_index(begin = var_49520_begin_0, end = var_49520_end_0, end_mask = var_49520_end_mask_0, x = var_49366_cast_fp16)[name = tensor("op_49520_cast_fp16")]; + tensor var_49527_begin_0 = const()[name = tensor("op_49527_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49527_end_0 = const()[name = tensor("op_49527_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49527_end_mask_0 = const()[name = tensor("op_49527_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49527_cast_fp16 = slice_by_index(begin = var_49527_begin_0, end = var_49527_end_0, end_mask = var_49527_end_mask_0, x = var_49370_cast_fp16)[name = tensor("op_49527_cast_fp16")]; + tensor var_49534_begin_0 = const()[name = tensor("op_49534_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49534_end_0 = const()[name = tensor("op_49534_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49534_end_mask_0 = const()[name = tensor("op_49534_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49534_cast_fp16 = slice_by_index(begin = var_49534_begin_0, end = var_49534_end_0, end_mask = var_49534_end_mask_0, x = var_49370_cast_fp16)[name = tensor("op_49534_cast_fp16")]; + tensor var_49541_begin_0 = const()[name = tensor("op_49541_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49541_end_0 = const()[name = tensor("op_49541_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49541_end_mask_0 = const()[name = tensor("op_49541_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49541_cast_fp16 = slice_by_index(begin = var_49541_begin_0, end = var_49541_end_0, end_mask = var_49541_end_mask_0, x = var_49370_cast_fp16)[name = tensor("op_49541_cast_fp16")]; + tensor var_49548_begin_0 = const()[name = tensor("op_49548_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49548_end_0 = const()[name = tensor("op_49548_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49548_end_mask_0 = const()[name = tensor("op_49548_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49548_cast_fp16 = slice_by_index(begin = var_49548_begin_0, end = var_49548_end_0, end_mask = var_49548_end_mask_0, x = var_49370_cast_fp16)[name = tensor("op_49548_cast_fp16")]; + tensor var_49555_begin_0 = const()[name = tensor("op_49555_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49555_end_0 = const()[name = tensor("op_49555_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49555_end_mask_0 = const()[name = tensor("op_49555_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49555_cast_fp16 = slice_by_index(begin = var_49555_begin_0, end = var_49555_end_0, end_mask = var_49555_end_mask_0, x = var_49374_cast_fp16)[name = tensor("op_49555_cast_fp16")]; + tensor var_49562_begin_0 = const()[name = tensor("op_49562_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49562_end_0 = const()[name = tensor("op_49562_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49562_end_mask_0 = const()[name = tensor("op_49562_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49562_cast_fp16 = slice_by_index(begin = var_49562_begin_0, end = var_49562_end_0, end_mask = var_49562_end_mask_0, x = var_49374_cast_fp16)[name = tensor("op_49562_cast_fp16")]; + tensor var_49569_begin_0 = const()[name = tensor("op_49569_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49569_end_0 = const()[name = tensor("op_49569_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49569_end_mask_0 = const()[name = tensor("op_49569_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49569_cast_fp16 = slice_by_index(begin = var_49569_begin_0, end = var_49569_end_0, end_mask = var_49569_end_mask_0, x = var_49374_cast_fp16)[name = tensor("op_49569_cast_fp16")]; + tensor var_49576_begin_0 = const()[name = tensor("op_49576_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49576_end_0 = const()[name = tensor("op_49576_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49576_end_mask_0 = const()[name = tensor("op_49576_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49576_cast_fp16 = slice_by_index(begin = var_49576_begin_0, end = var_49576_end_0, end_mask = var_49576_end_mask_0, x = var_49374_cast_fp16)[name = tensor("op_49576_cast_fp16")]; + tensor var_49583_begin_0 = const()[name = tensor("op_49583_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49583_end_0 = const()[name = tensor("op_49583_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49583_end_mask_0 = const()[name = tensor("op_49583_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49583_cast_fp16 = slice_by_index(begin = var_49583_begin_0, end = var_49583_end_0, end_mask = var_49583_end_mask_0, x = var_49378_cast_fp16)[name = tensor("op_49583_cast_fp16")]; + tensor var_49590_begin_0 = const()[name = tensor("op_49590_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49590_end_0 = const()[name = tensor("op_49590_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49590_end_mask_0 = const()[name = tensor("op_49590_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49590_cast_fp16 = slice_by_index(begin = var_49590_begin_0, end = var_49590_end_0, end_mask = var_49590_end_mask_0, x = var_49378_cast_fp16)[name = tensor("op_49590_cast_fp16")]; + tensor var_49597_begin_0 = const()[name = tensor("op_49597_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49597_end_0 = const()[name = tensor("op_49597_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49597_end_mask_0 = const()[name = tensor("op_49597_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49597_cast_fp16 = slice_by_index(begin = var_49597_begin_0, end = var_49597_end_0, end_mask = var_49597_end_mask_0, x = var_49378_cast_fp16)[name = tensor("op_49597_cast_fp16")]; + tensor var_49604_begin_0 = const()[name = tensor("op_49604_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49604_end_0 = const()[name = tensor("op_49604_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49604_end_mask_0 = const()[name = tensor("op_49604_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49604_cast_fp16 = slice_by_index(begin = var_49604_begin_0, end = var_49604_end_0, end_mask = var_49604_end_mask_0, x = var_49378_cast_fp16)[name = tensor("op_49604_cast_fp16")]; + tensor var_49611_begin_0 = const()[name = tensor("op_49611_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49611_end_0 = const()[name = tensor("op_49611_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49611_end_mask_0 = const()[name = tensor("op_49611_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49611_cast_fp16 = slice_by_index(begin = var_49611_begin_0, end = var_49611_end_0, end_mask = var_49611_end_mask_0, x = var_49382_cast_fp16)[name = tensor("op_49611_cast_fp16")]; + tensor var_49618_begin_0 = const()[name = tensor("op_49618_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49618_end_0 = const()[name = tensor("op_49618_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49618_end_mask_0 = const()[name = tensor("op_49618_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49618_cast_fp16 = slice_by_index(begin = var_49618_begin_0, end = var_49618_end_0, end_mask = var_49618_end_mask_0, x = var_49382_cast_fp16)[name = tensor("op_49618_cast_fp16")]; + tensor var_49625_begin_0 = const()[name = tensor("op_49625_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49625_end_0 = const()[name = tensor("op_49625_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49625_end_mask_0 = const()[name = tensor("op_49625_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49625_cast_fp16 = slice_by_index(begin = var_49625_begin_0, end = var_49625_end_0, end_mask = var_49625_end_mask_0, x = var_49382_cast_fp16)[name = tensor("op_49625_cast_fp16")]; + tensor var_49632_begin_0 = const()[name = tensor("op_49632_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49632_end_0 = const()[name = tensor("op_49632_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49632_end_mask_0 = const()[name = tensor("op_49632_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49632_cast_fp16 = slice_by_index(begin = var_49632_begin_0, end = var_49632_end_0, end_mask = var_49632_end_mask_0, x = var_49382_cast_fp16)[name = tensor("op_49632_cast_fp16")]; + tensor var_49639_begin_0 = const()[name = tensor("op_49639_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49639_end_0 = const()[name = tensor("op_49639_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49639_end_mask_0 = const()[name = tensor("op_49639_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49639_cast_fp16 = slice_by_index(begin = var_49639_begin_0, end = var_49639_end_0, end_mask = var_49639_end_mask_0, x = var_49386_cast_fp16)[name = tensor("op_49639_cast_fp16")]; + tensor var_49646_begin_0 = const()[name = tensor("op_49646_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49646_end_0 = const()[name = tensor("op_49646_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49646_end_mask_0 = const()[name = tensor("op_49646_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49646_cast_fp16 = slice_by_index(begin = var_49646_begin_0, end = var_49646_end_0, end_mask = var_49646_end_mask_0, x = var_49386_cast_fp16)[name = tensor("op_49646_cast_fp16")]; + tensor var_49653_begin_0 = const()[name = tensor("op_49653_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49653_end_0 = const()[name = tensor("op_49653_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49653_end_mask_0 = const()[name = tensor("op_49653_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49653_cast_fp16 = slice_by_index(begin = var_49653_begin_0, end = var_49653_end_0, end_mask = var_49653_end_mask_0, x = var_49386_cast_fp16)[name = tensor("op_49653_cast_fp16")]; + tensor var_49660_begin_0 = const()[name = tensor("op_49660_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49660_end_0 = const()[name = tensor("op_49660_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49660_end_mask_0 = const()[name = tensor("op_49660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49660_cast_fp16 = slice_by_index(begin = var_49660_begin_0, end = var_49660_end_0, end_mask = var_49660_end_mask_0, x = var_49386_cast_fp16)[name = tensor("op_49660_cast_fp16")]; + tensor var_49667_begin_0 = const()[name = tensor("op_49667_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49667_end_0 = const()[name = tensor("op_49667_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49667_end_mask_0 = const()[name = tensor("op_49667_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49667_cast_fp16 = slice_by_index(begin = var_49667_begin_0, end = var_49667_end_0, end_mask = var_49667_end_mask_0, x = var_49390_cast_fp16)[name = tensor("op_49667_cast_fp16")]; + tensor var_49674_begin_0 = const()[name = tensor("op_49674_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49674_end_0 = const()[name = tensor("op_49674_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49674_end_mask_0 = const()[name = tensor("op_49674_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49674_cast_fp16 = slice_by_index(begin = var_49674_begin_0, end = var_49674_end_0, end_mask = var_49674_end_mask_0, x = var_49390_cast_fp16)[name = tensor("op_49674_cast_fp16")]; + tensor var_49681_begin_0 = const()[name = tensor("op_49681_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49681_end_0 = const()[name = tensor("op_49681_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49681_end_mask_0 = const()[name = tensor("op_49681_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49681_cast_fp16 = slice_by_index(begin = var_49681_begin_0, end = var_49681_end_0, end_mask = var_49681_end_mask_0, x = var_49390_cast_fp16)[name = tensor("op_49681_cast_fp16")]; + tensor var_49688_begin_0 = const()[name = tensor("op_49688_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49688_end_0 = const()[name = tensor("op_49688_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49688_end_mask_0 = const()[name = tensor("op_49688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49688_cast_fp16 = slice_by_index(begin = var_49688_begin_0, end = var_49688_end_0, end_mask = var_49688_end_mask_0, x = var_49390_cast_fp16)[name = tensor("op_49688_cast_fp16")]; + tensor var_49695_begin_0 = const()[name = tensor("op_49695_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49695_end_0 = const()[name = tensor("op_49695_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49695_end_mask_0 = const()[name = tensor("op_49695_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49695_cast_fp16 = slice_by_index(begin = var_49695_begin_0, end = var_49695_end_0, end_mask = var_49695_end_mask_0, x = var_49394_cast_fp16)[name = tensor("op_49695_cast_fp16")]; + tensor var_49702_begin_0 = const()[name = tensor("op_49702_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49702_end_0 = const()[name = tensor("op_49702_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49702_end_mask_0 = const()[name = tensor("op_49702_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49702_cast_fp16 = slice_by_index(begin = var_49702_begin_0, end = var_49702_end_0, end_mask = var_49702_end_mask_0, x = var_49394_cast_fp16)[name = tensor("op_49702_cast_fp16")]; + tensor var_49709_begin_0 = const()[name = tensor("op_49709_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49709_end_0 = const()[name = tensor("op_49709_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49709_end_mask_0 = const()[name = tensor("op_49709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49709_cast_fp16 = slice_by_index(begin = var_49709_begin_0, end = var_49709_end_0, end_mask = var_49709_end_mask_0, x = var_49394_cast_fp16)[name = tensor("op_49709_cast_fp16")]; + tensor var_49716_begin_0 = const()[name = tensor("op_49716_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49716_end_0 = const()[name = tensor("op_49716_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49716_end_mask_0 = const()[name = tensor("op_49716_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49716_cast_fp16 = slice_by_index(begin = var_49716_begin_0, end = var_49716_end_0, end_mask = var_49716_end_mask_0, x = var_49394_cast_fp16)[name = tensor("op_49716_cast_fp16")]; + tensor var_49723_begin_0 = const()[name = tensor("op_49723_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49723_end_0 = const()[name = tensor("op_49723_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49723_end_mask_0 = const()[name = tensor("op_49723_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49723_cast_fp16 = slice_by_index(begin = var_49723_begin_0, end = var_49723_end_0, end_mask = var_49723_end_mask_0, x = var_49398_cast_fp16)[name = tensor("op_49723_cast_fp16")]; + tensor var_49730_begin_0 = const()[name = tensor("op_49730_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49730_end_0 = const()[name = tensor("op_49730_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49730_end_mask_0 = const()[name = tensor("op_49730_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49730_cast_fp16 = slice_by_index(begin = var_49730_begin_0, end = var_49730_end_0, end_mask = var_49730_end_mask_0, x = var_49398_cast_fp16)[name = tensor("op_49730_cast_fp16")]; + tensor var_49737_begin_0 = const()[name = tensor("op_49737_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49737_end_0 = const()[name = tensor("op_49737_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49737_end_mask_0 = const()[name = tensor("op_49737_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49737_cast_fp16 = slice_by_index(begin = var_49737_begin_0, end = var_49737_end_0, end_mask = var_49737_end_mask_0, x = var_49398_cast_fp16)[name = tensor("op_49737_cast_fp16")]; + tensor var_49744_begin_0 = const()[name = tensor("op_49744_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49744_end_0 = const()[name = tensor("op_49744_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49744_end_mask_0 = const()[name = tensor("op_49744_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49744_cast_fp16 = slice_by_index(begin = var_49744_begin_0, end = var_49744_end_0, end_mask = var_49744_end_mask_0, x = var_49398_cast_fp16)[name = tensor("op_49744_cast_fp16")]; + tensor var_49751_begin_0 = const()[name = tensor("op_49751_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49751_end_0 = const()[name = tensor("op_49751_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49751_end_mask_0 = const()[name = tensor("op_49751_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49751_cast_fp16 = slice_by_index(begin = var_49751_begin_0, end = var_49751_end_0, end_mask = var_49751_end_mask_0, x = var_49402_cast_fp16)[name = tensor("op_49751_cast_fp16")]; + tensor var_49758_begin_0 = const()[name = tensor("op_49758_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49758_end_0 = const()[name = tensor("op_49758_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49758_end_mask_0 = const()[name = tensor("op_49758_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49758_cast_fp16 = slice_by_index(begin = var_49758_begin_0, end = var_49758_end_0, end_mask = var_49758_end_mask_0, x = var_49402_cast_fp16)[name = tensor("op_49758_cast_fp16")]; + tensor var_49765_begin_0 = const()[name = tensor("op_49765_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49765_end_0 = const()[name = tensor("op_49765_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49765_end_mask_0 = const()[name = tensor("op_49765_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49765_cast_fp16 = slice_by_index(begin = var_49765_begin_0, end = var_49765_end_0, end_mask = var_49765_end_mask_0, x = var_49402_cast_fp16)[name = tensor("op_49765_cast_fp16")]; + tensor var_49772_begin_0 = const()[name = tensor("op_49772_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49772_end_0 = const()[name = tensor("op_49772_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49772_end_mask_0 = const()[name = tensor("op_49772_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49772_cast_fp16 = slice_by_index(begin = var_49772_begin_0, end = var_49772_end_0, end_mask = var_49772_end_mask_0, x = var_49402_cast_fp16)[name = tensor("op_49772_cast_fp16")]; + tensor var_49779_begin_0 = const()[name = tensor("op_49779_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49779_end_0 = const()[name = tensor("op_49779_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49779_end_mask_0 = const()[name = tensor("op_49779_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49779_cast_fp16 = slice_by_index(begin = var_49779_begin_0, end = var_49779_end_0, end_mask = var_49779_end_mask_0, x = var_49406_cast_fp16)[name = tensor("op_49779_cast_fp16")]; + tensor var_49786_begin_0 = const()[name = tensor("op_49786_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49786_end_0 = const()[name = tensor("op_49786_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49786_end_mask_0 = const()[name = tensor("op_49786_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49786_cast_fp16 = slice_by_index(begin = var_49786_begin_0, end = var_49786_end_0, end_mask = var_49786_end_mask_0, x = var_49406_cast_fp16)[name = tensor("op_49786_cast_fp16")]; + tensor var_49793_begin_0 = const()[name = tensor("op_49793_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49793_end_0 = const()[name = tensor("op_49793_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49793_end_mask_0 = const()[name = tensor("op_49793_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49793_cast_fp16 = slice_by_index(begin = var_49793_begin_0, end = var_49793_end_0, end_mask = var_49793_end_mask_0, x = var_49406_cast_fp16)[name = tensor("op_49793_cast_fp16")]; + tensor var_49800_begin_0 = const()[name = tensor("op_49800_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49800_end_0 = const()[name = tensor("op_49800_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49800_end_mask_0 = const()[name = tensor("op_49800_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49800_cast_fp16 = slice_by_index(begin = var_49800_begin_0, end = var_49800_end_0, end_mask = var_49800_end_mask_0, x = var_49406_cast_fp16)[name = tensor("op_49800_cast_fp16")]; + tensor var_49807_begin_0 = const()[name = tensor("op_49807_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49807_end_0 = const()[name = tensor("op_49807_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49807_end_mask_0 = const()[name = tensor("op_49807_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49807_cast_fp16 = slice_by_index(begin = var_49807_begin_0, end = var_49807_end_0, end_mask = var_49807_end_mask_0, x = var_49410_cast_fp16)[name = tensor("op_49807_cast_fp16")]; + tensor var_49814_begin_0 = const()[name = tensor("op_49814_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49814_end_0 = const()[name = tensor("op_49814_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49814_end_mask_0 = const()[name = tensor("op_49814_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49814_cast_fp16 = slice_by_index(begin = var_49814_begin_0, end = var_49814_end_0, end_mask = var_49814_end_mask_0, x = var_49410_cast_fp16)[name = tensor("op_49814_cast_fp16")]; + tensor var_49821_begin_0 = const()[name = tensor("op_49821_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49821_end_0 = const()[name = tensor("op_49821_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49821_end_mask_0 = const()[name = tensor("op_49821_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49821_cast_fp16 = slice_by_index(begin = var_49821_begin_0, end = var_49821_end_0, end_mask = var_49821_end_mask_0, x = var_49410_cast_fp16)[name = tensor("op_49821_cast_fp16")]; + tensor var_49828_begin_0 = const()[name = tensor("op_49828_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49828_end_0 = const()[name = tensor("op_49828_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49828_end_mask_0 = const()[name = tensor("op_49828_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49828_cast_fp16 = slice_by_index(begin = var_49828_begin_0, end = var_49828_end_0, end_mask = var_49828_end_mask_0, x = var_49410_cast_fp16)[name = tensor("op_49828_cast_fp16")]; + tensor var_49835_begin_0 = const()[name = tensor("op_49835_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49835_end_0 = const()[name = tensor("op_49835_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49835_end_mask_0 = const()[name = tensor("op_49835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49835_cast_fp16 = slice_by_index(begin = var_49835_begin_0, end = var_49835_end_0, end_mask = var_49835_end_mask_0, x = var_49414_cast_fp16)[name = tensor("op_49835_cast_fp16")]; + tensor var_49842_begin_0 = const()[name = tensor("op_49842_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49842_end_0 = const()[name = tensor("op_49842_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49842_end_mask_0 = const()[name = tensor("op_49842_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49842_cast_fp16 = slice_by_index(begin = var_49842_begin_0, end = var_49842_end_0, end_mask = var_49842_end_mask_0, x = var_49414_cast_fp16)[name = tensor("op_49842_cast_fp16")]; + tensor var_49849_begin_0 = const()[name = tensor("op_49849_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49849_end_0 = const()[name = tensor("op_49849_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49849_end_mask_0 = const()[name = tensor("op_49849_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49849_cast_fp16 = slice_by_index(begin = var_49849_begin_0, end = var_49849_end_0, end_mask = var_49849_end_mask_0, x = var_49414_cast_fp16)[name = tensor("op_49849_cast_fp16")]; + tensor var_49856_begin_0 = const()[name = tensor("op_49856_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49856_end_0 = const()[name = tensor("op_49856_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49856_end_mask_0 = const()[name = tensor("op_49856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49856_cast_fp16 = slice_by_index(begin = var_49856_begin_0, end = var_49856_end_0, end_mask = var_49856_end_mask_0, x = var_49414_cast_fp16)[name = tensor("op_49856_cast_fp16")]; + tensor var_49863_begin_0 = const()[name = tensor("op_49863_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49863_end_0 = const()[name = tensor("op_49863_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49863_end_mask_0 = const()[name = tensor("op_49863_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49863_cast_fp16 = slice_by_index(begin = var_49863_begin_0, end = var_49863_end_0, end_mask = var_49863_end_mask_0, x = var_49418_cast_fp16)[name = tensor("op_49863_cast_fp16")]; + tensor var_49870_begin_0 = const()[name = tensor("op_49870_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49870_end_0 = const()[name = tensor("op_49870_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49870_end_mask_0 = const()[name = tensor("op_49870_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49870_cast_fp16 = slice_by_index(begin = var_49870_begin_0, end = var_49870_end_0, end_mask = var_49870_end_mask_0, x = var_49418_cast_fp16)[name = tensor("op_49870_cast_fp16")]; + tensor var_49877_begin_0 = const()[name = tensor("op_49877_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49877_end_0 = const()[name = tensor("op_49877_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49877_end_mask_0 = const()[name = tensor("op_49877_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49877_cast_fp16 = slice_by_index(begin = var_49877_begin_0, end = var_49877_end_0, end_mask = var_49877_end_mask_0, x = var_49418_cast_fp16)[name = tensor("op_49877_cast_fp16")]; + tensor var_49884_begin_0 = const()[name = tensor("op_49884_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49884_end_0 = const()[name = tensor("op_49884_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49884_end_mask_0 = const()[name = tensor("op_49884_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49884_cast_fp16 = slice_by_index(begin = var_49884_begin_0, end = var_49884_end_0, end_mask = var_49884_end_mask_0, x = var_49418_cast_fp16)[name = tensor("op_49884_cast_fp16")]; + tensor var_49891_begin_0 = const()[name = tensor("op_49891_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49891_end_0 = const()[name = tensor("op_49891_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49891_end_mask_0 = const()[name = tensor("op_49891_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49891_cast_fp16 = slice_by_index(begin = var_49891_begin_0, end = var_49891_end_0, end_mask = var_49891_end_mask_0, x = var_49422_cast_fp16)[name = tensor("op_49891_cast_fp16")]; + tensor var_49898_begin_0 = const()[name = tensor("op_49898_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49898_end_0 = const()[name = tensor("op_49898_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49898_end_mask_0 = const()[name = tensor("op_49898_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49898_cast_fp16 = slice_by_index(begin = var_49898_begin_0, end = var_49898_end_0, end_mask = var_49898_end_mask_0, x = var_49422_cast_fp16)[name = tensor("op_49898_cast_fp16")]; + tensor var_49905_begin_0 = const()[name = tensor("op_49905_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49905_end_0 = const()[name = tensor("op_49905_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49905_end_mask_0 = const()[name = tensor("op_49905_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49905_cast_fp16 = slice_by_index(begin = var_49905_begin_0, end = var_49905_end_0, end_mask = var_49905_end_mask_0, x = var_49422_cast_fp16)[name = tensor("op_49905_cast_fp16")]; + tensor var_49912_begin_0 = const()[name = tensor("op_49912_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49912_end_0 = const()[name = tensor("op_49912_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49912_end_mask_0 = const()[name = tensor("op_49912_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49912_cast_fp16 = slice_by_index(begin = var_49912_begin_0, end = var_49912_end_0, end_mask = var_49912_end_mask_0, x = var_49422_cast_fp16)[name = tensor("op_49912_cast_fp16")]; + tensor var_49919_begin_0 = const()[name = tensor("op_49919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49919_end_0 = const()[name = tensor("op_49919_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49919_end_mask_0 = const()[name = tensor("op_49919_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49919_cast_fp16 = slice_by_index(begin = var_49919_begin_0, end = var_49919_end_0, end_mask = var_49919_end_mask_0, x = var_49426_cast_fp16)[name = tensor("op_49919_cast_fp16")]; + tensor var_49926_begin_0 = const()[name = tensor("op_49926_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49926_end_0 = const()[name = tensor("op_49926_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49926_end_mask_0 = const()[name = tensor("op_49926_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49926_cast_fp16 = slice_by_index(begin = var_49926_begin_0, end = var_49926_end_0, end_mask = var_49926_end_mask_0, x = var_49426_cast_fp16)[name = tensor("op_49926_cast_fp16")]; + tensor var_49933_begin_0 = const()[name = tensor("op_49933_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49933_end_0 = const()[name = tensor("op_49933_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49933_end_mask_0 = const()[name = tensor("op_49933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49933_cast_fp16 = slice_by_index(begin = var_49933_begin_0, end = var_49933_end_0, end_mask = var_49933_end_mask_0, x = var_49426_cast_fp16)[name = tensor("op_49933_cast_fp16")]; + tensor var_49940_begin_0 = const()[name = tensor("op_49940_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49940_end_0 = const()[name = tensor("op_49940_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49940_end_mask_0 = const()[name = tensor("op_49940_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49940_cast_fp16 = slice_by_index(begin = var_49940_begin_0, end = var_49940_end_0, end_mask = var_49940_end_mask_0, x = var_49426_cast_fp16)[name = tensor("op_49940_cast_fp16")]; + tensor var_49947_begin_0 = const()[name = tensor("op_49947_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49947_end_0 = const()[name = tensor("op_49947_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49947_end_mask_0 = const()[name = tensor("op_49947_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49947_cast_fp16 = slice_by_index(begin = var_49947_begin_0, end = var_49947_end_0, end_mask = var_49947_end_mask_0, x = var_49430_cast_fp16)[name = tensor("op_49947_cast_fp16")]; + tensor var_49954_begin_0 = const()[name = tensor("op_49954_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49954_end_0 = const()[name = tensor("op_49954_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49954_end_mask_0 = const()[name = tensor("op_49954_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49954_cast_fp16 = slice_by_index(begin = var_49954_begin_0, end = var_49954_end_0, end_mask = var_49954_end_mask_0, x = var_49430_cast_fp16)[name = tensor("op_49954_cast_fp16")]; + tensor var_49961_begin_0 = const()[name = tensor("op_49961_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49961_end_0 = const()[name = tensor("op_49961_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49961_end_mask_0 = const()[name = tensor("op_49961_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49961_cast_fp16 = slice_by_index(begin = var_49961_begin_0, end = var_49961_end_0, end_mask = var_49961_end_mask_0, x = var_49430_cast_fp16)[name = tensor("op_49961_cast_fp16")]; + tensor var_49968_begin_0 = const()[name = tensor("op_49968_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49968_end_0 = const()[name = tensor("op_49968_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49968_end_mask_0 = const()[name = tensor("op_49968_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49968_cast_fp16 = slice_by_index(begin = var_49968_begin_0, end = var_49968_end_0, end_mask = var_49968_end_mask_0, x = var_49430_cast_fp16)[name = tensor("op_49968_cast_fp16")]; + tensor var_49975_begin_0 = const()[name = tensor("op_49975_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_49975_end_0 = const()[name = tensor("op_49975_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_49975_end_mask_0 = const()[name = tensor("op_49975_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49975_cast_fp16 = slice_by_index(begin = var_49975_begin_0, end = var_49975_end_0, end_mask = var_49975_end_mask_0, x = var_49434_cast_fp16)[name = tensor("op_49975_cast_fp16")]; + tensor var_49982_begin_0 = const()[name = tensor("op_49982_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_49982_end_0 = const()[name = tensor("op_49982_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_49982_end_mask_0 = const()[name = tensor("op_49982_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49982_cast_fp16 = slice_by_index(begin = var_49982_begin_0, end = var_49982_end_0, end_mask = var_49982_end_mask_0, x = var_49434_cast_fp16)[name = tensor("op_49982_cast_fp16")]; + tensor var_49989_begin_0 = const()[name = tensor("op_49989_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_49989_end_0 = const()[name = tensor("op_49989_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_49989_end_mask_0 = const()[name = tensor("op_49989_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49989_cast_fp16 = slice_by_index(begin = var_49989_begin_0, end = var_49989_end_0, end_mask = var_49989_end_mask_0, x = var_49434_cast_fp16)[name = tensor("op_49989_cast_fp16")]; + tensor var_49996_begin_0 = const()[name = tensor("op_49996_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_49996_end_0 = const()[name = tensor("op_49996_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_49996_end_mask_0 = const()[name = tensor("op_49996_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_49996_cast_fp16 = slice_by_index(begin = var_49996_begin_0, end = var_49996_end_0, end_mask = var_49996_end_mask_0, x = var_49434_cast_fp16)[name = tensor("op_49996_cast_fp16")]; + tensor k_61_perm_0 = const()[name = tensor("k_61_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_50001_begin_0 = const()[name = tensor("op_50001_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50001_end_0 = const()[name = tensor("op_50001_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_50001_end_mask_0 = const()[name = tensor("op_50001_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_1 = transpose(perm = k_61_perm_0, x = key_61_cast_fp16)[name = tensor("transpose_1")]; + tensor var_50001_cast_fp16 = slice_by_index(begin = var_50001_begin_0, end = var_50001_end_0, end_mask = var_50001_end_mask_0, x = transpose_1)[name = tensor("op_50001_cast_fp16")]; + tensor var_50005_begin_0 = const()[name = tensor("op_50005_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_50005_end_0 = const()[name = tensor("op_50005_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_50005_end_mask_0 = const()[name = tensor("op_50005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50005_cast_fp16 = slice_by_index(begin = var_50005_begin_0, end = var_50005_end_0, end_mask = var_50005_end_mask_0, x = transpose_1)[name = tensor("op_50005_cast_fp16")]; + tensor var_50009_begin_0 = const()[name = tensor("op_50009_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_50009_end_0 = const()[name = tensor("op_50009_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_50009_end_mask_0 = const()[name = tensor("op_50009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50009_cast_fp16 = slice_by_index(begin = var_50009_begin_0, end = var_50009_end_0, end_mask = var_50009_end_mask_0, x = transpose_1)[name = tensor("op_50009_cast_fp16")]; + tensor var_50013_begin_0 = const()[name = tensor("op_50013_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_50013_end_0 = const()[name = tensor("op_50013_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_50013_end_mask_0 = const()[name = tensor("op_50013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50013_cast_fp16 = slice_by_index(begin = var_50013_begin_0, end = var_50013_end_0, end_mask = var_50013_end_mask_0, x = transpose_1)[name = tensor("op_50013_cast_fp16")]; + tensor var_50017_begin_0 = const()[name = tensor("op_50017_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_50017_end_0 = const()[name = tensor("op_50017_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_50017_end_mask_0 = const()[name = tensor("op_50017_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50017_cast_fp16 = slice_by_index(begin = var_50017_begin_0, end = var_50017_end_0, end_mask = var_50017_end_mask_0, x = transpose_1)[name = tensor("op_50017_cast_fp16")]; + tensor var_50021_begin_0 = const()[name = tensor("op_50021_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_50021_end_0 = const()[name = tensor("op_50021_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_50021_end_mask_0 = const()[name = tensor("op_50021_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50021_cast_fp16 = slice_by_index(begin = var_50021_begin_0, end = var_50021_end_0, end_mask = var_50021_end_mask_0, x = transpose_1)[name = tensor("op_50021_cast_fp16")]; + tensor var_50025_begin_0 = const()[name = tensor("op_50025_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_50025_end_0 = const()[name = tensor("op_50025_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_50025_end_mask_0 = const()[name = tensor("op_50025_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50025_cast_fp16 = slice_by_index(begin = var_50025_begin_0, end = var_50025_end_0, end_mask = var_50025_end_mask_0, x = transpose_1)[name = tensor("op_50025_cast_fp16")]; + tensor var_50029_begin_0 = const()[name = tensor("op_50029_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_50029_end_0 = const()[name = tensor("op_50029_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_50029_end_mask_0 = const()[name = tensor("op_50029_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50029_cast_fp16 = slice_by_index(begin = var_50029_begin_0, end = var_50029_end_0, end_mask = var_50029_end_mask_0, x = transpose_1)[name = tensor("op_50029_cast_fp16")]; + tensor var_50033_begin_0 = const()[name = tensor("op_50033_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_50033_end_0 = const()[name = tensor("op_50033_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_50033_end_mask_0 = const()[name = tensor("op_50033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50033_cast_fp16 = slice_by_index(begin = var_50033_begin_0, end = var_50033_end_0, end_mask = var_50033_end_mask_0, x = transpose_1)[name = tensor("op_50033_cast_fp16")]; + tensor var_50037_begin_0 = const()[name = tensor("op_50037_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_50037_end_0 = const()[name = tensor("op_50037_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_50037_end_mask_0 = const()[name = tensor("op_50037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50037_cast_fp16 = slice_by_index(begin = var_50037_begin_0, end = var_50037_end_0, end_mask = var_50037_end_mask_0, x = transpose_1)[name = tensor("op_50037_cast_fp16")]; + tensor var_50041_begin_0 = const()[name = tensor("op_50041_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_50041_end_0 = const()[name = tensor("op_50041_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_50041_end_mask_0 = const()[name = tensor("op_50041_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50041_cast_fp16 = slice_by_index(begin = var_50041_begin_0, end = var_50041_end_0, end_mask = var_50041_end_mask_0, x = transpose_1)[name = tensor("op_50041_cast_fp16")]; + tensor var_50045_begin_0 = const()[name = tensor("op_50045_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_50045_end_0 = const()[name = tensor("op_50045_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_50045_end_mask_0 = const()[name = tensor("op_50045_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50045_cast_fp16 = slice_by_index(begin = var_50045_begin_0, end = var_50045_end_0, end_mask = var_50045_end_mask_0, x = transpose_1)[name = tensor("op_50045_cast_fp16")]; + tensor var_50049_begin_0 = const()[name = tensor("op_50049_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_50049_end_0 = const()[name = tensor("op_50049_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_50049_end_mask_0 = const()[name = tensor("op_50049_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50049_cast_fp16 = slice_by_index(begin = var_50049_begin_0, end = var_50049_end_0, end_mask = var_50049_end_mask_0, x = transpose_1)[name = tensor("op_50049_cast_fp16")]; + tensor var_50053_begin_0 = const()[name = tensor("op_50053_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_50053_end_0 = const()[name = tensor("op_50053_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_50053_end_mask_0 = const()[name = tensor("op_50053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50053_cast_fp16 = slice_by_index(begin = var_50053_begin_0, end = var_50053_end_0, end_mask = var_50053_end_mask_0, x = transpose_1)[name = tensor("op_50053_cast_fp16")]; + tensor var_50057_begin_0 = const()[name = tensor("op_50057_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_50057_end_0 = const()[name = tensor("op_50057_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_50057_end_mask_0 = const()[name = tensor("op_50057_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50057_cast_fp16 = slice_by_index(begin = var_50057_begin_0, end = var_50057_end_0, end_mask = var_50057_end_mask_0, x = transpose_1)[name = tensor("op_50057_cast_fp16")]; + tensor var_50061_begin_0 = const()[name = tensor("op_50061_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_50061_end_0 = const()[name = tensor("op_50061_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_50061_end_mask_0 = const()[name = tensor("op_50061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50061_cast_fp16 = slice_by_index(begin = var_50061_begin_0, end = var_50061_end_0, end_mask = var_50061_end_mask_0, x = transpose_1)[name = tensor("op_50061_cast_fp16")]; + tensor var_50065_begin_0 = const()[name = tensor("op_50065_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_50065_end_0 = const()[name = tensor("op_50065_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_50065_end_mask_0 = const()[name = tensor("op_50065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50065_cast_fp16 = slice_by_index(begin = var_50065_begin_0, end = var_50065_end_0, end_mask = var_50065_end_mask_0, x = transpose_1)[name = tensor("op_50065_cast_fp16")]; + tensor var_50069_begin_0 = const()[name = tensor("op_50069_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_50069_end_0 = const()[name = tensor("op_50069_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_50069_end_mask_0 = const()[name = tensor("op_50069_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50069_cast_fp16 = slice_by_index(begin = var_50069_begin_0, end = var_50069_end_0, end_mask = var_50069_end_mask_0, x = transpose_1)[name = tensor("op_50069_cast_fp16")]; + tensor var_50073_begin_0 = const()[name = tensor("op_50073_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_50073_end_0 = const()[name = tensor("op_50073_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_50073_end_mask_0 = const()[name = tensor("op_50073_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50073_cast_fp16 = slice_by_index(begin = var_50073_begin_0, end = var_50073_end_0, end_mask = var_50073_end_mask_0, x = transpose_1)[name = tensor("op_50073_cast_fp16")]; + tensor var_50077_begin_0 = const()[name = tensor("op_50077_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_50077_end_0 = const()[name = tensor("op_50077_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_50077_end_mask_0 = const()[name = tensor("op_50077_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_50077_cast_fp16 = slice_by_index(begin = var_50077_begin_0, end = var_50077_end_0, end_mask = var_50077_end_mask_0, x = transpose_1)[name = tensor("op_50077_cast_fp16")]; + tensor var_50079_begin_0 = const()[name = tensor("op_50079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50079_end_0 = const()[name = tensor("op_50079_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_50079_end_mask_0 = const()[name = tensor("op_50079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50079_cast_fp16 = slice_by_index(begin = var_50079_begin_0, end = var_50079_end_0, end_mask = var_50079_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50079_cast_fp16")]; + tensor var_50083_begin_0 = const()[name = tensor("op_50083_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50083_end_0 = const()[name = tensor("op_50083_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_50083_end_mask_0 = const()[name = tensor("op_50083_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50083_cast_fp16 = slice_by_index(begin = var_50083_begin_0, end = var_50083_end_0, end_mask = var_50083_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50083_cast_fp16")]; + tensor var_50087_begin_0 = const()[name = tensor("op_50087_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_50087_end_0 = const()[name = tensor("op_50087_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_50087_end_mask_0 = const()[name = tensor("op_50087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50087_cast_fp16 = slice_by_index(begin = var_50087_begin_0, end = var_50087_end_0, end_mask = var_50087_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50087_cast_fp16")]; + tensor var_50091_begin_0 = const()[name = tensor("op_50091_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_50091_end_0 = const()[name = tensor("op_50091_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_50091_end_mask_0 = const()[name = tensor("op_50091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50091_cast_fp16 = slice_by_index(begin = var_50091_begin_0, end = var_50091_end_0, end_mask = var_50091_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50091_cast_fp16")]; + tensor var_50095_begin_0 = const()[name = tensor("op_50095_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_50095_end_0 = const()[name = tensor("op_50095_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_50095_end_mask_0 = const()[name = tensor("op_50095_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50095_cast_fp16 = slice_by_index(begin = var_50095_begin_0, end = var_50095_end_0, end_mask = var_50095_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50095_cast_fp16")]; + tensor var_50099_begin_0 = const()[name = tensor("op_50099_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_50099_end_0 = const()[name = tensor("op_50099_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_50099_end_mask_0 = const()[name = tensor("op_50099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50099_cast_fp16 = slice_by_index(begin = var_50099_begin_0, end = var_50099_end_0, end_mask = var_50099_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50099_cast_fp16")]; + tensor var_50103_begin_0 = const()[name = tensor("op_50103_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_50103_end_0 = const()[name = tensor("op_50103_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_50103_end_mask_0 = const()[name = tensor("op_50103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50103_cast_fp16 = slice_by_index(begin = var_50103_begin_0, end = var_50103_end_0, end_mask = var_50103_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50103_cast_fp16")]; + tensor var_50107_begin_0 = const()[name = tensor("op_50107_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_50107_end_0 = const()[name = tensor("op_50107_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_50107_end_mask_0 = const()[name = tensor("op_50107_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50107_cast_fp16 = slice_by_index(begin = var_50107_begin_0, end = var_50107_end_0, end_mask = var_50107_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50107_cast_fp16")]; + tensor var_50111_begin_0 = const()[name = tensor("op_50111_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_50111_end_0 = const()[name = tensor("op_50111_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_50111_end_mask_0 = const()[name = tensor("op_50111_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50111_cast_fp16 = slice_by_index(begin = var_50111_begin_0, end = var_50111_end_0, end_mask = var_50111_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50111_cast_fp16")]; + tensor var_50115_begin_0 = const()[name = tensor("op_50115_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_50115_end_0 = const()[name = tensor("op_50115_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_50115_end_mask_0 = const()[name = tensor("op_50115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50115_cast_fp16 = slice_by_index(begin = var_50115_begin_0, end = var_50115_end_0, end_mask = var_50115_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50115_cast_fp16")]; + tensor var_50119_begin_0 = const()[name = tensor("op_50119_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_50119_end_0 = const()[name = tensor("op_50119_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_50119_end_mask_0 = const()[name = tensor("op_50119_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50119_cast_fp16 = slice_by_index(begin = var_50119_begin_0, end = var_50119_end_0, end_mask = var_50119_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50119_cast_fp16")]; + tensor var_50123_begin_0 = const()[name = tensor("op_50123_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_50123_end_0 = const()[name = tensor("op_50123_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_50123_end_mask_0 = const()[name = tensor("op_50123_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50123_cast_fp16 = slice_by_index(begin = var_50123_begin_0, end = var_50123_end_0, end_mask = var_50123_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50123_cast_fp16")]; + tensor var_50127_begin_0 = const()[name = tensor("op_50127_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_50127_end_0 = const()[name = tensor("op_50127_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_50127_end_mask_0 = const()[name = tensor("op_50127_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50127_cast_fp16 = slice_by_index(begin = var_50127_begin_0, end = var_50127_end_0, end_mask = var_50127_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50127_cast_fp16")]; + tensor var_50131_begin_0 = const()[name = tensor("op_50131_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_50131_end_0 = const()[name = tensor("op_50131_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_50131_end_mask_0 = const()[name = tensor("op_50131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50131_cast_fp16 = slice_by_index(begin = var_50131_begin_0, end = var_50131_end_0, end_mask = var_50131_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50131_cast_fp16")]; + tensor var_50135_begin_0 = const()[name = tensor("op_50135_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_50135_end_0 = const()[name = tensor("op_50135_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_50135_end_mask_0 = const()[name = tensor("op_50135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50135_cast_fp16 = slice_by_index(begin = var_50135_begin_0, end = var_50135_end_0, end_mask = var_50135_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50135_cast_fp16")]; + tensor var_50139_begin_0 = const()[name = tensor("op_50139_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_50139_end_0 = const()[name = tensor("op_50139_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_50139_end_mask_0 = const()[name = tensor("op_50139_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50139_cast_fp16 = slice_by_index(begin = var_50139_begin_0, end = var_50139_end_0, end_mask = var_50139_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50139_cast_fp16")]; + tensor var_50143_begin_0 = const()[name = tensor("op_50143_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_50143_end_0 = const()[name = tensor("op_50143_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_50143_end_mask_0 = const()[name = tensor("op_50143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50143_cast_fp16 = slice_by_index(begin = var_50143_begin_0, end = var_50143_end_0, end_mask = var_50143_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50143_cast_fp16")]; + tensor var_50147_begin_0 = const()[name = tensor("op_50147_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_50147_end_0 = const()[name = tensor("op_50147_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_50147_end_mask_0 = const()[name = tensor("op_50147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50147_cast_fp16 = slice_by_index(begin = var_50147_begin_0, end = var_50147_end_0, end_mask = var_50147_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50147_cast_fp16")]; + tensor var_50151_begin_0 = const()[name = tensor("op_50151_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_50151_end_0 = const()[name = tensor("op_50151_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_50151_end_mask_0 = const()[name = tensor("op_50151_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50151_cast_fp16 = slice_by_index(begin = var_50151_begin_0, end = var_50151_end_0, end_mask = var_50151_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50151_cast_fp16")]; + tensor var_50155_begin_0 = const()[name = tensor("op_50155_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_50155_end_0 = const()[name = tensor("op_50155_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_50155_end_mask_0 = const()[name = tensor("op_50155_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50155_cast_fp16 = slice_by_index(begin = var_50155_begin_0, end = var_50155_end_0, end_mask = var_50155_end_mask_0, x = value_61_cast_fp16)[name = tensor("op_50155_cast_fp16")]; + tensor var_50159_equation_0 = const()[name = tensor("op_50159_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50159_cast_fp16 = einsum(equation = var_50159_equation_0, values = (var_50001_cast_fp16, var_49443_cast_fp16))[name = tensor("op_50159_cast_fp16")]; + tensor var_50160_to_fp16 = const()[name = tensor("op_50160_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4801_cast_fp16 = mul(x = var_50159_cast_fp16, y = var_50160_to_fp16)[name = tensor("aw_chunk_4801_cast_fp16")]; + tensor var_50163_equation_0 = const()[name = tensor("op_50163_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50163_cast_fp16 = einsum(equation = var_50163_equation_0, values = (var_50001_cast_fp16, var_49450_cast_fp16))[name = tensor("op_50163_cast_fp16")]; + tensor var_50164_to_fp16 = const()[name = tensor("op_50164_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4803_cast_fp16 = mul(x = var_50163_cast_fp16, y = var_50164_to_fp16)[name = tensor("aw_chunk_4803_cast_fp16")]; + tensor var_50167_equation_0 = const()[name = tensor("op_50167_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50167_cast_fp16 = einsum(equation = var_50167_equation_0, values = (var_50001_cast_fp16, var_49457_cast_fp16))[name = tensor("op_50167_cast_fp16")]; + tensor var_50168_to_fp16 = const()[name = tensor("op_50168_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4805_cast_fp16 = mul(x = var_50167_cast_fp16, y = var_50168_to_fp16)[name = tensor("aw_chunk_4805_cast_fp16")]; + tensor var_50171_equation_0 = const()[name = tensor("op_50171_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50171_cast_fp16 = einsum(equation = var_50171_equation_0, values = (var_50001_cast_fp16, var_49464_cast_fp16))[name = tensor("op_50171_cast_fp16")]; + tensor var_50172_to_fp16 = const()[name = tensor("op_50172_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4807_cast_fp16 = mul(x = var_50171_cast_fp16, y = var_50172_to_fp16)[name = tensor("aw_chunk_4807_cast_fp16")]; + tensor var_50175_equation_0 = const()[name = tensor("op_50175_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50175_cast_fp16 = einsum(equation = var_50175_equation_0, values = (var_50005_cast_fp16, var_49471_cast_fp16))[name = tensor("op_50175_cast_fp16")]; + tensor var_50176_to_fp16 = const()[name = tensor("op_50176_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4809_cast_fp16 = mul(x = var_50175_cast_fp16, y = var_50176_to_fp16)[name = tensor("aw_chunk_4809_cast_fp16")]; + tensor var_50179_equation_0 = const()[name = tensor("op_50179_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50179_cast_fp16 = einsum(equation = var_50179_equation_0, values = (var_50005_cast_fp16, var_49478_cast_fp16))[name = tensor("op_50179_cast_fp16")]; + tensor var_50180_to_fp16 = const()[name = tensor("op_50180_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4811_cast_fp16 = mul(x = var_50179_cast_fp16, y = var_50180_to_fp16)[name = tensor("aw_chunk_4811_cast_fp16")]; + tensor var_50183_equation_0 = const()[name = tensor("op_50183_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50183_cast_fp16 = einsum(equation = var_50183_equation_0, values = (var_50005_cast_fp16, var_49485_cast_fp16))[name = tensor("op_50183_cast_fp16")]; + tensor var_50184_to_fp16 = const()[name = tensor("op_50184_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4813_cast_fp16 = mul(x = var_50183_cast_fp16, y = var_50184_to_fp16)[name = tensor("aw_chunk_4813_cast_fp16")]; + tensor var_50187_equation_0 = const()[name = tensor("op_50187_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50187_cast_fp16 = einsum(equation = var_50187_equation_0, values = (var_50005_cast_fp16, var_49492_cast_fp16))[name = tensor("op_50187_cast_fp16")]; + tensor var_50188_to_fp16 = const()[name = tensor("op_50188_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4815_cast_fp16 = mul(x = var_50187_cast_fp16, y = var_50188_to_fp16)[name = tensor("aw_chunk_4815_cast_fp16")]; + tensor var_50191_equation_0 = const()[name = tensor("op_50191_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50191_cast_fp16 = einsum(equation = var_50191_equation_0, values = (var_50009_cast_fp16, var_49499_cast_fp16))[name = tensor("op_50191_cast_fp16")]; + tensor var_50192_to_fp16 = const()[name = tensor("op_50192_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4817_cast_fp16 = mul(x = var_50191_cast_fp16, y = var_50192_to_fp16)[name = tensor("aw_chunk_4817_cast_fp16")]; + tensor var_50195_equation_0 = const()[name = tensor("op_50195_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50195_cast_fp16 = einsum(equation = var_50195_equation_0, values = (var_50009_cast_fp16, var_49506_cast_fp16))[name = tensor("op_50195_cast_fp16")]; + tensor var_50196_to_fp16 = const()[name = tensor("op_50196_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4819_cast_fp16 = mul(x = var_50195_cast_fp16, y = var_50196_to_fp16)[name = tensor("aw_chunk_4819_cast_fp16")]; + tensor var_50199_equation_0 = const()[name = tensor("op_50199_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50199_cast_fp16 = einsum(equation = var_50199_equation_0, values = (var_50009_cast_fp16, var_49513_cast_fp16))[name = tensor("op_50199_cast_fp16")]; + tensor var_50200_to_fp16 = const()[name = tensor("op_50200_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4821_cast_fp16 = mul(x = var_50199_cast_fp16, y = var_50200_to_fp16)[name = tensor("aw_chunk_4821_cast_fp16")]; + tensor var_50203_equation_0 = const()[name = tensor("op_50203_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50203_cast_fp16 = einsum(equation = var_50203_equation_0, values = (var_50009_cast_fp16, var_49520_cast_fp16))[name = tensor("op_50203_cast_fp16")]; + tensor var_50204_to_fp16 = const()[name = tensor("op_50204_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4823_cast_fp16 = mul(x = var_50203_cast_fp16, y = var_50204_to_fp16)[name = tensor("aw_chunk_4823_cast_fp16")]; + tensor var_50207_equation_0 = const()[name = tensor("op_50207_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50207_cast_fp16 = einsum(equation = var_50207_equation_0, values = (var_50013_cast_fp16, var_49527_cast_fp16))[name = tensor("op_50207_cast_fp16")]; + tensor var_50208_to_fp16 = const()[name = tensor("op_50208_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4825_cast_fp16 = mul(x = var_50207_cast_fp16, y = var_50208_to_fp16)[name = tensor("aw_chunk_4825_cast_fp16")]; + tensor var_50211_equation_0 = const()[name = tensor("op_50211_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50211_cast_fp16 = einsum(equation = var_50211_equation_0, values = (var_50013_cast_fp16, var_49534_cast_fp16))[name = tensor("op_50211_cast_fp16")]; + tensor var_50212_to_fp16 = const()[name = tensor("op_50212_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4827_cast_fp16 = mul(x = var_50211_cast_fp16, y = var_50212_to_fp16)[name = tensor("aw_chunk_4827_cast_fp16")]; + tensor var_50215_equation_0 = const()[name = tensor("op_50215_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50215_cast_fp16 = einsum(equation = var_50215_equation_0, values = (var_50013_cast_fp16, var_49541_cast_fp16))[name = tensor("op_50215_cast_fp16")]; + tensor var_50216_to_fp16 = const()[name = tensor("op_50216_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4829_cast_fp16 = mul(x = var_50215_cast_fp16, y = var_50216_to_fp16)[name = tensor("aw_chunk_4829_cast_fp16")]; + tensor var_50219_equation_0 = const()[name = tensor("op_50219_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50219_cast_fp16 = einsum(equation = var_50219_equation_0, values = (var_50013_cast_fp16, var_49548_cast_fp16))[name = tensor("op_50219_cast_fp16")]; + tensor var_50220_to_fp16 = const()[name = tensor("op_50220_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4831_cast_fp16 = mul(x = var_50219_cast_fp16, y = var_50220_to_fp16)[name = tensor("aw_chunk_4831_cast_fp16")]; + tensor var_50223_equation_0 = const()[name = tensor("op_50223_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50223_cast_fp16 = einsum(equation = var_50223_equation_0, values = (var_50017_cast_fp16, var_49555_cast_fp16))[name = tensor("op_50223_cast_fp16")]; + tensor var_50224_to_fp16 = const()[name = tensor("op_50224_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4833_cast_fp16 = mul(x = var_50223_cast_fp16, y = var_50224_to_fp16)[name = tensor("aw_chunk_4833_cast_fp16")]; + tensor var_50227_equation_0 = const()[name = tensor("op_50227_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50227_cast_fp16 = einsum(equation = var_50227_equation_0, values = (var_50017_cast_fp16, var_49562_cast_fp16))[name = tensor("op_50227_cast_fp16")]; + tensor var_50228_to_fp16 = const()[name = tensor("op_50228_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4835_cast_fp16 = mul(x = var_50227_cast_fp16, y = var_50228_to_fp16)[name = tensor("aw_chunk_4835_cast_fp16")]; + tensor var_50231_equation_0 = const()[name = tensor("op_50231_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50231_cast_fp16 = einsum(equation = var_50231_equation_0, values = (var_50017_cast_fp16, var_49569_cast_fp16))[name = tensor("op_50231_cast_fp16")]; + tensor var_50232_to_fp16 = const()[name = tensor("op_50232_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4837_cast_fp16 = mul(x = var_50231_cast_fp16, y = var_50232_to_fp16)[name = tensor("aw_chunk_4837_cast_fp16")]; + tensor var_50235_equation_0 = const()[name = tensor("op_50235_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50235_cast_fp16 = einsum(equation = var_50235_equation_0, values = (var_50017_cast_fp16, var_49576_cast_fp16))[name = tensor("op_50235_cast_fp16")]; + tensor var_50236_to_fp16 = const()[name = tensor("op_50236_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4839_cast_fp16 = mul(x = var_50235_cast_fp16, y = var_50236_to_fp16)[name = tensor("aw_chunk_4839_cast_fp16")]; + tensor var_50239_equation_0 = const()[name = tensor("op_50239_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50239_cast_fp16 = einsum(equation = var_50239_equation_0, values = (var_50021_cast_fp16, var_49583_cast_fp16))[name = tensor("op_50239_cast_fp16")]; + tensor var_50240_to_fp16 = const()[name = tensor("op_50240_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4841_cast_fp16 = mul(x = var_50239_cast_fp16, y = var_50240_to_fp16)[name = tensor("aw_chunk_4841_cast_fp16")]; + tensor var_50243_equation_0 = const()[name = tensor("op_50243_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50243_cast_fp16 = einsum(equation = var_50243_equation_0, values = (var_50021_cast_fp16, var_49590_cast_fp16))[name = tensor("op_50243_cast_fp16")]; + tensor var_50244_to_fp16 = const()[name = tensor("op_50244_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4843_cast_fp16 = mul(x = var_50243_cast_fp16, y = var_50244_to_fp16)[name = tensor("aw_chunk_4843_cast_fp16")]; + tensor var_50247_equation_0 = const()[name = tensor("op_50247_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50247_cast_fp16 = einsum(equation = var_50247_equation_0, values = (var_50021_cast_fp16, var_49597_cast_fp16))[name = tensor("op_50247_cast_fp16")]; + tensor var_50248_to_fp16 = const()[name = tensor("op_50248_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4845_cast_fp16 = mul(x = var_50247_cast_fp16, y = var_50248_to_fp16)[name = tensor("aw_chunk_4845_cast_fp16")]; + tensor var_50251_equation_0 = const()[name = tensor("op_50251_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50251_cast_fp16 = einsum(equation = var_50251_equation_0, values = (var_50021_cast_fp16, var_49604_cast_fp16))[name = tensor("op_50251_cast_fp16")]; + tensor var_50252_to_fp16 = const()[name = tensor("op_50252_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4847_cast_fp16 = mul(x = var_50251_cast_fp16, y = var_50252_to_fp16)[name = tensor("aw_chunk_4847_cast_fp16")]; + tensor var_50255_equation_0 = const()[name = tensor("op_50255_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50255_cast_fp16 = einsum(equation = var_50255_equation_0, values = (var_50025_cast_fp16, var_49611_cast_fp16))[name = tensor("op_50255_cast_fp16")]; + tensor var_50256_to_fp16 = const()[name = tensor("op_50256_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4849_cast_fp16 = mul(x = var_50255_cast_fp16, y = var_50256_to_fp16)[name = tensor("aw_chunk_4849_cast_fp16")]; + tensor var_50259_equation_0 = const()[name = tensor("op_50259_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50259_cast_fp16 = einsum(equation = var_50259_equation_0, values = (var_50025_cast_fp16, var_49618_cast_fp16))[name = tensor("op_50259_cast_fp16")]; + tensor var_50260_to_fp16 = const()[name = tensor("op_50260_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4851_cast_fp16 = mul(x = var_50259_cast_fp16, y = var_50260_to_fp16)[name = tensor("aw_chunk_4851_cast_fp16")]; + tensor var_50263_equation_0 = const()[name = tensor("op_50263_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50263_cast_fp16 = einsum(equation = var_50263_equation_0, values = (var_50025_cast_fp16, var_49625_cast_fp16))[name = tensor("op_50263_cast_fp16")]; + tensor var_50264_to_fp16 = const()[name = tensor("op_50264_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4853_cast_fp16 = mul(x = var_50263_cast_fp16, y = var_50264_to_fp16)[name = tensor("aw_chunk_4853_cast_fp16")]; + tensor var_50267_equation_0 = const()[name = tensor("op_50267_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50267_cast_fp16 = einsum(equation = var_50267_equation_0, values = (var_50025_cast_fp16, var_49632_cast_fp16))[name = tensor("op_50267_cast_fp16")]; + tensor var_50268_to_fp16 = const()[name = tensor("op_50268_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4855_cast_fp16 = mul(x = var_50267_cast_fp16, y = var_50268_to_fp16)[name = tensor("aw_chunk_4855_cast_fp16")]; + tensor var_50271_equation_0 = const()[name = tensor("op_50271_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50271_cast_fp16 = einsum(equation = var_50271_equation_0, values = (var_50029_cast_fp16, var_49639_cast_fp16))[name = tensor("op_50271_cast_fp16")]; + tensor var_50272_to_fp16 = const()[name = tensor("op_50272_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4857_cast_fp16 = mul(x = var_50271_cast_fp16, y = var_50272_to_fp16)[name = tensor("aw_chunk_4857_cast_fp16")]; + tensor var_50275_equation_0 = const()[name = tensor("op_50275_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50275_cast_fp16 = einsum(equation = var_50275_equation_0, values = (var_50029_cast_fp16, var_49646_cast_fp16))[name = tensor("op_50275_cast_fp16")]; + tensor var_50276_to_fp16 = const()[name = tensor("op_50276_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4859_cast_fp16 = mul(x = var_50275_cast_fp16, y = var_50276_to_fp16)[name = tensor("aw_chunk_4859_cast_fp16")]; + tensor var_50279_equation_0 = const()[name = tensor("op_50279_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50279_cast_fp16 = einsum(equation = var_50279_equation_0, values = (var_50029_cast_fp16, var_49653_cast_fp16))[name = tensor("op_50279_cast_fp16")]; + tensor var_50280_to_fp16 = const()[name = tensor("op_50280_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4861_cast_fp16 = mul(x = var_50279_cast_fp16, y = var_50280_to_fp16)[name = tensor("aw_chunk_4861_cast_fp16")]; + tensor var_50283_equation_0 = const()[name = tensor("op_50283_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50283_cast_fp16 = einsum(equation = var_50283_equation_0, values = (var_50029_cast_fp16, var_49660_cast_fp16))[name = tensor("op_50283_cast_fp16")]; + tensor var_50284_to_fp16 = const()[name = tensor("op_50284_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4863_cast_fp16 = mul(x = var_50283_cast_fp16, y = var_50284_to_fp16)[name = tensor("aw_chunk_4863_cast_fp16")]; + tensor var_50287_equation_0 = const()[name = tensor("op_50287_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50287_cast_fp16 = einsum(equation = var_50287_equation_0, values = (var_50033_cast_fp16, var_49667_cast_fp16))[name = tensor("op_50287_cast_fp16")]; + tensor var_50288_to_fp16 = const()[name = tensor("op_50288_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4865_cast_fp16 = mul(x = var_50287_cast_fp16, y = var_50288_to_fp16)[name = tensor("aw_chunk_4865_cast_fp16")]; + tensor var_50291_equation_0 = const()[name = tensor("op_50291_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50291_cast_fp16 = einsum(equation = var_50291_equation_0, values = (var_50033_cast_fp16, var_49674_cast_fp16))[name = tensor("op_50291_cast_fp16")]; + tensor var_50292_to_fp16 = const()[name = tensor("op_50292_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4867_cast_fp16 = mul(x = var_50291_cast_fp16, y = var_50292_to_fp16)[name = tensor("aw_chunk_4867_cast_fp16")]; + tensor var_50295_equation_0 = const()[name = tensor("op_50295_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50295_cast_fp16 = einsum(equation = var_50295_equation_0, values = (var_50033_cast_fp16, var_49681_cast_fp16))[name = tensor("op_50295_cast_fp16")]; + tensor var_50296_to_fp16 = const()[name = tensor("op_50296_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4869_cast_fp16 = mul(x = var_50295_cast_fp16, y = var_50296_to_fp16)[name = tensor("aw_chunk_4869_cast_fp16")]; + tensor var_50299_equation_0 = const()[name = tensor("op_50299_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50299_cast_fp16 = einsum(equation = var_50299_equation_0, values = (var_50033_cast_fp16, var_49688_cast_fp16))[name = tensor("op_50299_cast_fp16")]; + tensor var_50300_to_fp16 = const()[name = tensor("op_50300_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4871_cast_fp16 = mul(x = var_50299_cast_fp16, y = var_50300_to_fp16)[name = tensor("aw_chunk_4871_cast_fp16")]; + tensor var_50303_equation_0 = const()[name = tensor("op_50303_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50303_cast_fp16 = einsum(equation = var_50303_equation_0, values = (var_50037_cast_fp16, var_49695_cast_fp16))[name = tensor("op_50303_cast_fp16")]; + tensor var_50304_to_fp16 = const()[name = tensor("op_50304_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4873_cast_fp16 = mul(x = var_50303_cast_fp16, y = var_50304_to_fp16)[name = tensor("aw_chunk_4873_cast_fp16")]; + tensor var_50307_equation_0 = const()[name = tensor("op_50307_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50307_cast_fp16 = einsum(equation = var_50307_equation_0, values = (var_50037_cast_fp16, var_49702_cast_fp16))[name = tensor("op_50307_cast_fp16")]; + tensor var_50308_to_fp16 = const()[name = tensor("op_50308_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4875_cast_fp16 = mul(x = var_50307_cast_fp16, y = var_50308_to_fp16)[name = tensor("aw_chunk_4875_cast_fp16")]; + tensor var_50311_equation_0 = const()[name = tensor("op_50311_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50311_cast_fp16 = einsum(equation = var_50311_equation_0, values = (var_50037_cast_fp16, var_49709_cast_fp16))[name = tensor("op_50311_cast_fp16")]; + tensor var_50312_to_fp16 = const()[name = tensor("op_50312_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4877_cast_fp16 = mul(x = var_50311_cast_fp16, y = var_50312_to_fp16)[name = tensor("aw_chunk_4877_cast_fp16")]; + tensor var_50315_equation_0 = const()[name = tensor("op_50315_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50315_cast_fp16 = einsum(equation = var_50315_equation_0, values = (var_50037_cast_fp16, var_49716_cast_fp16))[name = tensor("op_50315_cast_fp16")]; + tensor var_50316_to_fp16 = const()[name = tensor("op_50316_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4879_cast_fp16 = mul(x = var_50315_cast_fp16, y = var_50316_to_fp16)[name = tensor("aw_chunk_4879_cast_fp16")]; + tensor var_50319_equation_0 = const()[name = tensor("op_50319_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50319_cast_fp16 = einsum(equation = var_50319_equation_0, values = (var_50041_cast_fp16, var_49723_cast_fp16))[name = tensor("op_50319_cast_fp16")]; + tensor var_50320_to_fp16 = const()[name = tensor("op_50320_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4881_cast_fp16 = mul(x = var_50319_cast_fp16, y = var_50320_to_fp16)[name = tensor("aw_chunk_4881_cast_fp16")]; + tensor var_50323_equation_0 = const()[name = tensor("op_50323_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50323_cast_fp16 = einsum(equation = var_50323_equation_0, values = (var_50041_cast_fp16, var_49730_cast_fp16))[name = tensor("op_50323_cast_fp16")]; + tensor var_50324_to_fp16 = const()[name = tensor("op_50324_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4883_cast_fp16 = mul(x = var_50323_cast_fp16, y = var_50324_to_fp16)[name = tensor("aw_chunk_4883_cast_fp16")]; + tensor var_50327_equation_0 = const()[name = tensor("op_50327_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50327_cast_fp16 = einsum(equation = var_50327_equation_0, values = (var_50041_cast_fp16, var_49737_cast_fp16))[name = tensor("op_50327_cast_fp16")]; + tensor var_50328_to_fp16 = const()[name = tensor("op_50328_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4885_cast_fp16 = mul(x = var_50327_cast_fp16, y = var_50328_to_fp16)[name = tensor("aw_chunk_4885_cast_fp16")]; + tensor var_50331_equation_0 = const()[name = tensor("op_50331_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50331_cast_fp16 = einsum(equation = var_50331_equation_0, values = (var_50041_cast_fp16, var_49744_cast_fp16))[name = tensor("op_50331_cast_fp16")]; + tensor var_50332_to_fp16 = const()[name = tensor("op_50332_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4887_cast_fp16 = mul(x = var_50331_cast_fp16, y = var_50332_to_fp16)[name = tensor("aw_chunk_4887_cast_fp16")]; + tensor var_50335_equation_0 = const()[name = tensor("op_50335_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50335_cast_fp16 = einsum(equation = var_50335_equation_0, values = (var_50045_cast_fp16, var_49751_cast_fp16))[name = tensor("op_50335_cast_fp16")]; + tensor var_50336_to_fp16 = const()[name = tensor("op_50336_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4889_cast_fp16 = mul(x = var_50335_cast_fp16, y = var_50336_to_fp16)[name = tensor("aw_chunk_4889_cast_fp16")]; + tensor var_50339_equation_0 = const()[name = tensor("op_50339_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50339_cast_fp16 = einsum(equation = var_50339_equation_0, values = (var_50045_cast_fp16, var_49758_cast_fp16))[name = tensor("op_50339_cast_fp16")]; + tensor var_50340_to_fp16 = const()[name = tensor("op_50340_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4891_cast_fp16 = mul(x = var_50339_cast_fp16, y = var_50340_to_fp16)[name = tensor("aw_chunk_4891_cast_fp16")]; + tensor var_50343_equation_0 = const()[name = tensor("op_50343_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50343_cast_fp16 = einsum(equation = var_50343_equation_0, values = (var_50045_cast_fp16, var_49765_cast_fp16))[name = tensor("op_50343_cast_fp16")]; + tensor var_50344_to_fp16 = const()[name = tensor("op_50344_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4893_cast_fp16 = mul(x = var_50343_cast_fp16, y = var_50344_to_fp16)[name = tensor("aw_chunk_4893_cast_fp16")]; + tensor var_50347_equation_0 = const()[name = tensor("op_50347_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50347_cast_fp16 = einsum(equation = var_50347_equation_0, values = (var_50045_cast_fp16, var_49772_cast_fp16))[name = tensor("op_50347_cast_fp16")]; + tensor var_50348_to_fp16 = const()[name = tensor("op_50348_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4895_cast_fp16 = mul(x = var_50347_cast_fp16, y = var_50348_to_fp16)[name = tensor("aw_chunk_4895_cast_fp16")]; + tensor var_50351_equation_0 = const()[name = tensor("op_50351_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50351_cast_fp16 = einsum(equation = var_50351_equation_0, values = (var_50049_cast_fp16, var_49779_cast_fp16))[name = tensor("op_50351_cast_fp16")]; + tensor var_50352_to_fp16 = const()[name = tensor("op_50352_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4897_cast_fp16 = mul(x = var_50351_cast_fp16, y = var_50352_to_fp16)[name = tensor("aw_chunk_4897_cast_fp16")]; + tensor var_50355_equation_0 = const()[name = tensor("op_50355_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50355_cast_fp16 = einsum(equation = var_50355_equation_0, values = (var_50049_cast_fp16, var_49786_cast_fp16))[name = tensor("op_50355_cast_fp16")]; + tensor var_50356_to_fp16 = const()[name = tensor("op_50356_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4899_cast_fp16 = mul(x = var_50355_cast_fp16, y = var_50356_to_fp16)[name = tensor("aw_chunk_4899_cast_fp16")]; + tensor var_50359_equation_0 = const()[name = tensor("op_50359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50359_cast_fp16 = einsum(equation = var_50359_equation_0, values = (var_50049_cast_fp16, var_49793_cast_fp16))[name = tensor("op_50359_cast_fp16")]; + tensor var_50360_to_fp16 = const()[name = tensor("op_50360_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4901_cast_fp16 = mul(x = var_50359_cast_fp16, y = var_50360_to_fp16)[name = tensor("aw_chunk_4901_cast_fp16")]; + tensor var_50363_equation_0 = const()[name = tensor("op_50363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50363_cast_fp16 = einsum(equation = var_50363_equation_0, values = (var_50049_cast_fp16, var_49800_cast_fp16))[name = tensor("op_50363_cast_fp16")]; + tensor var_50364_to_fp16 = const()[name = tensor("op_50364_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4903_cast_fp16 = mul(x = var_50363_cast_fp16, y = var_50364_to_fp16)[name = tensor("aw_chunk_4903_cast_fp16")]; + tensor var_50367_equation_0 = const()[name = tensor("op_50367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50367_cast_fp16 = einsum(equation = var_50367_equation_0, values = (var_50053_cast_fp16, var_49807_cast_fp16))[name = tensor("op_50367_cast_fp16")]; + tensor var_50368_to_fp16 = const()[name = tensor("op_50368_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4905_cast_fp16 = mul(x = var_50367_cast_fp16, y = var_50368_to_fp16)[name = tensor("aw_chunk_4905_cast_fp16")]; + tensor var_50371_equation_0 = const()[name = tensor("op_50371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50371_cast_fp16 = einsum(equation = var_50371_equation_0, values = (var_50053_cast_fp16, var_49814_cast_fp16))[name = tensor("op_50371_cast_fp16")]; + tensor var_50372_to_fp16 = const()[name = tensor("op_50372_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4907_cast_fp16 = mul(x = var_50371_cast_fp16, y = var_50372_to_fp16)[name = tensor("aw_chunk_4907_cast_fp16")]; + tensor var_50375_equation_0 = const()[name = tensor("op_50375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50375_cast_fp16 = einsum(equation = var_50375_equation_0, values = (var_50053_cast_fp16, var_49821_cast_fp16))[name = tensor("op_50375_cast_fp16")]; + tensor var_50376_to_fp16 = const()[name = tensor("op_50376_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4909_cast_fp16 = mul(x = var_50375_cast_fp16, y = var_50376_to_fp16)[name = tensor("aw_chunk_4909_cast_fp16")]; + tensor var_50379_equation_0 = const()[name = tensor("op_50379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50379_cast_fp16 = einsum(equation = var_50379_equation_0, values = (var_50053_cast_fp16, var_49828_cast_fp16))[name = tensor("op_50379_cast_fp16")]; + tensor var_50380_to_fp16 = const()[name = tensor("op_50380_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4911_cast_fp16 = mul(x = var_50379_cast_fp16, y = var_50380_to_fp16)[name = tensor("aw_chunk_4911_cast_fp16")]; + tensor var_50383_equation_0 = const()[name = tensor("op_50383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50383_cast_fp16 = einsum(equation = var_50383_equation_0, values = (var_50057_cast_fp16, var_49835_cast_fp16))[name = tensor("op_50383_cast_fp16")]; + tensor var_50384_to_fp16 = const()[name = tensor("op_50384_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4913_cast_fp16 = mul(x = var_50383_cast_fp16, y = var_50384_to_fp16)[name = tensor("aw_chunk_4913_cast_fp16")]; + tensor var_50387_equation_0 = const()[name = tensor("op_50387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50387_cast_fp16 = einsum(equation = var_50387_equation_0, values = (var_50057_cast_fp16, var_49842_cast_fp16))[name = tensor("op_50387_cast_fp16")]; + tensor var_50388_to_fp16 = const()[name = tensor("op_50388_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4915_cast_fp16 = mul(x = var_50387_cast_fp16, y = var_50388_to_fp16)[name = tensor("aw_chunk_4915_cast_fp16")]; + tensor var_50391_equation_0 = const()[name = tensor("op_50391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50391_cast_fp16 = einsum(equation = var_50391_equation_0, values = (var_50057_cast_fp16, var_49849_cast_fp16))[name = tensor("op_50391_cast_fp16")]; + tensor var_50392_to_fp16 = const()[name = tensor("op_50392_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4917_cast_fp16 = mul(x = var_50391_cast_fp16, y = var_50392_to_fp16)[name = tensor("aw_chunk_4917_cast_fp16")]; + tensor var_50395_equation_0 = const()[name = tensor("op_50395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50395_cast_fp16 = einsum(equation = var_50395_equation_0, values = (var_50057_cast_fp16, var_49856_cast_fp16))[name = tensor("op_50395_cast_fp16")]; + tensor var_50396_to_fp16 = const()[name = tensor("op_50396_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4919_cast_fp16 = mul(x = var_50395_cast_fp16, y = var_50396_to_fp16)[name = tensor("aw_chunk_4919_cast_fp16")]; + tensor var_50399_equation_0 = const()[name = tensor("op_50399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50399_cast_fp16 = einsum(equation = var_50399_equation_0, values = (var_50061_cast_fp16, var_49863_cast_fp16))[name = tensor("op_50399_cast_fp16")]; + tensor var_50400_to_fp16 = const()[name = tensor("op_50400_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4921_cast_fp16 = mul(x = var_50399_cast_fp16, y = var_50400_to_fp16)[name = tensor("aw_chunk_4921_cast_fp16")]; + tensor var_50403_equation_0 = const()[name = tensor("op_50403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50403_cast_fp16 = einsum(equation = var_50403_equation_0, values = (var_50061_cast_fp16, var_49870_cast_fp16))[name = tensor("op_50403_cast_fp16")]; + tensor var_50404_to_fp16 = const()[name = tensor("op_50404_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4923_cast_fp16 = mul(x = var_50403_cast_fp16, y = var_50404_to_fp16)[name = tensor("aw_chunk_4923_cast_fp16")]; + tensor var_50407_equation_0 = const()[name = tensor("op_50407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50407_cast_fp16 = einsum(equation = var_50407_equation_0, values = (var_50061_cast_fp16, var_49877_cast_fp16))[name = tensor("op_50407_cast_fp16")]; + tensor var_50408_to_fp16 = const()[name = tensor("op_50408_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4925_cast_fp16 = mul(x = var_50407_cast_fp16, y = var_50408_to_fp16)[name = tensor("aw_chunk_4925_cast_fp16")]; + tensor var_50411_equation_0 = const()[name = tensor("op_50411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50411_cast_fp16 = einsum(equation = var_50411_equation_0, values = (var_50061_cast_fp16, var_49884_cast_fp16))[name = tensor("op_50411_cast_fp16")]; + tensor var_50412_to_fp16 = const()[name = tensor("op_50412_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4927_cast_fp16 = mul(x = var_50411_cast_fp16, y = var_50412_to_fp16)[name = tensor("aw_chunk_4927_cast_fp16")]; + tensor var_50415_equation_0 = const()[name = tensor("op_50415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50415_cast_fp16 = einsum(equation = var_50415_equation_0, values = (var_50065_cast_fp16, var_49891_cast_fp16))[name = tensor("op_50415_cast_fp16")]; + tensor var_50416_to_fp16 = const()[name = tensor("op_50416_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4929_cast_fp16 = mul(x = var_50415_cast_fp16, y = var_50416_to_fp16)[name = tensor("aw_chunk_4929_cast_fp16")]; + tensor var_50419_equation_0 = const()[name = tensor("op_50419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50419_cast_fp16 = einsum(equation = var_50419_equation_0, values = (var_50065_cast_fp16, var_49898_cast_fp16))[name = tensor("op_50419_cast_fp16")]; + tensor var_50420_to_fp16 = const()[name = tensor("op_50420_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4931_cast_fp16 = mul(x = var_50419_cast_fp16, y = var_50420_to_fp16)[name = tensor("aw_chunk_4931_cast_fp16")]; + tensor var_50423_equation_0 = const()[name = tensor("op_50423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50423_cast_fp16 = einsum(equation = var_50423_equation_0, values = (var_50065_cast_fp16, var_49905_cast_fp16))[name = tensor("op_50423_cast_fp16")]; + tensor var_50424_to_fp16 = const()[name = tensor("op_50424_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4933_cast_fp16 = mul(x = var_50423_cast_fp16, y = var_50424_to_fp16)[name = tensor("aw_chunk_4933_cast_fp16")]; + tensor var_50427_equation_0 = const()[name = tensor("op_50427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50427_cast_fp16 = einsum(equation = var_50427_equation_0, values = (var_50065_cast_fp16, var_49912_cast_fp16))[name = tensor("op_50427_cast_fp16")]; + tensor var_50428_to_fp16 = const()[name = tensor("op_50428_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4935_cast_fp16 = mul(x = var_50427_cast_fp16, y = var_50428_to_fp16)[name = tensor("aw_chunk_4935_cast_fp16")]; + tensor var_50431_equation_0 = const()[name = tensor("op_50431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50431_cast_fp16 = einsum(equation = var_50431_equation_0, values = (var_50069_cast_fp16, var_49919_cast_fp16))[name = tensor("op_50431_cast_fp16")]; + tensor var_50432_to_fp16 = const()[name = tensor("op_50432_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4937_cast_fp16 = mul(x = var_50431_cast_fp16, y = var_50432_to_fp16)[name = tensor("aw_chunk_4937_cast_fp16")]; + tensor var_50435_equation_0 = const()[name = tensor("op_50435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50435_cast_fp16 = einsum(equation = var_50435_equation_0, values = (var_50069_cast_fp16, var_49926_cast_fp16))[name = tensor("op_50435_cast_fp16")]; + tensor var_50436_to_fp16 = const()[name = tensor("op_50436_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4939_cast_fp16 = mul(x = var_50435_cast_fp16, y = var_50436_to_fp16)[name = tensor("aw_chunk_4939_cast_fp16")]; + tensor var_50439_equation_0 = const()[name = tensor("op_50439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50439_cast_fp16 = einsum(equation = var_50439_equation_0, values = (var_50069_cast_fp16, var_49933_cast_fp16))[name = tensor("op_50439_cast_fp16")]; + tensor var_50440_to_fp16 = const()[name = tensor("op_50440_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4941_cast_fp16 = mul(x = var_50439_cast_fp16, y = var_50440_to_fp16)[name = tensor("aw_chunk_4941_cast_fp16")]; + tensor var_50443_equation_0 = const()[name = tensor("op_50443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50443_cast_fp16 = einsum(equation = var_50443_equation_0, values = (var_50069_cast_fp16, var_49940_cast_fp16))[name = tensor("op_50443_cast_fp16")]; + tensor var_50444_to_fp16 = const()[name = tensor("op_50444_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4943_cast_fp16 = mul(x = var_50443_cast_fp16, y = var_50444_to_fp16)[name = tensor("aw_chunk_4943_cast_fp16")]; + tensor var_50447_equation_0 = const()[name = tensor("op_50447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50447_cast_fp16 = einsum(equation = var_50447_equation_0, values = (var_50073_cast_fp16, var_49947_cast_fp16))[name = tensor("op_50447_cast_fp16")]; + tensor var_50448_to_fp16 = const()[name = tensor("op_50448_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4945_cast_fp16 = mul(x = var_50447_cast_fp16, y = var_50448_to_fp16)[name = tensor("aw_chunk_4945_cast_fp16")]; + tensor var_50451_equation_0 = const()[name = tensor("op_50451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50451_cast_fp16 = einsum(equation = var_50451_equation_0, values = (var_50073_cast_fp16, var_49954_cast_fp16))[name = tensor("op_50451_cast_fp16")]; + tensor var_50452_to_fp16 = const()[name = tensor("op_50452_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4947_cast_fp16 = mul(x = var_50451_cast_fp16, y = var_50452_to_fp16)[name = tensor("aw_chunk_4947_cast_fp16")]; + tensor var_50455_equation_0 = const()[name = tensor("op_50455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50455_cast_fp16 = einsum(equation = var_50455_equation_0, values = (var_50073_cast_fp16, var_49961_cast_fp16))[name = tensor("op_50455_cast_fp16")]; + tensor var_50456_to_fp16 = const()[name = tensor("op_50456_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4949_cast_fp16 = mul(x = var_50455_cast_fp16, y = var_50456_to_fp16)[name = tensor("aw_chunk_4949_cast_fp16")]; + tensor var_50459_equation_0 = const()[name = tensor("op_50459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50459_cast_fp16 = einsum(equation = var_50459_equation_0, values = (var_50073_cast_fp16, var_49968_cast_fp16))[name = tensor("op_50459_cast_fp16")]; + tensor var_50460_to_fp16 = const()[name = tensor("op_50460_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4951_cast_fp16 = mul(x = var_50459_cast_fp16, y = var_50460_to_fp16)[name = tensor("aw_chunk_4951_cast_fp16")]; + tensor var_50463_equation_0 = const()[name = tensor("op_50463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50463_cast_fp16 = einsum(equation = var_50463_equation_0, values = (var_50077_cast_fp16, var_49975_cast_fp16))[name = tensor("op_50463_cast_fp16")]; + tensor var_50464_to_fp16 = const()[name = tensor("op_50464_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4953_cast_fp16 = mul(x = var_50463_cast_fp16, y = var_50464_to_fp16)[name = tensor("aw_chunk_4953_cast_fp16")]; + tensor var_50467_equation_0 = const()[name = tensor("op_50467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50467_cast_fp16 = einsum(equation = var_50467_equation_0, values = (var_50077_cast_fp16, var_49982_cast_fp16))[name = tensor("op_50467_cast_fp16")]; + tensor var_50468_to_fp16 = const()[name = tensor("op_50468_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4955_cast_fp16 = mul(x = var_50467_cast_fp16, y = var_50468_to_fp16)[name = tensor("aw_chunk_4955_cast_fp16")]; + tensor var_50471_equation_0 = const()[name = tensor("op_50471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50471_cast_fp16 = einsum(equation = var_50471_equation_0, values = (var_50077_cast_fp16, var_49989_cast_fp16))[name = tensor("op_50471_cast_fp16")]; + tensor var_50472_to_fp16 = const()[name = tensor("op_50472_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4957_cast_fp16 = mul(x = var_50471_cast_fp16, y = var_50472_to_fp16)[name = tensor("aw_chunk_4957_cast_fp16")]; + tensor var_50475_equation_0 = const()[name = tensor("op_50475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_50475_cast_fp16 = einsum(equation = var_50475_equation_0, values = (var_50077_cast_fp16, var_49996_cast_fp16))[name = tensor("op_50475_cast_fp16")]; + tensor var_50476_to_fp16 = const()[name = tensor("op_50476_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4959_cast_fp16 = mul(x = var_50475_cast_fp16, y = var_50476_to_fp16)[name = tensor("aw_chunk_4959_cast_fp16")]; + tensor var_50478_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4801_cast_fp16)[name = tensor("op_50478_cast_fp16")]; + tensor var_50479_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4803_cast_fp16)[name = tensor("op_50479_cast_fp16")]; + tensor var_50480_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4805_cast_fp16)[name = tensor("op_50480_cast_fp16")]; + tensor var_50481_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4807_cast_fp16)[name = tensor("op_50481_cast_fp16")]; + tensor var_50482_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4809_cast_fp16)[name = tensor("op_50482_cast_fp16")]; + tensor var_50483_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4811_cast_fp16)[name = tensor("op_50483_cast_fp16")]; + tensor var_50484_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4813_cast_fp16)[name = tensor("op_50484_cast_fp16")]; + tensor var_50485_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4815_cast_fp16)[name = tensor("op_50485_cast_fp16")]; + tensor var_50486_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4817_cast_fp16)[name = tensor("op_50486_cast_fp16")]; + tensor var_50487_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4819_cast_fp16)[name = tensor("op_50487_cast_fp16")]; + tensor var_50488_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4821_cast_fp16)[name = tensor("op_50488_cast_fp16")]; + tensor var_50489_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4823_cast_fp16)[name = tensor("op_50489_cast_fp16")]; + tensor var_50490_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4825_cast_fp16)[name = tensor("op_50490_cast_fp16")]; + tensor var_50491_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4827_cast_fp16)[name = tensor("op_50491_cast_fp16")]; + tensor var_50492_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4829_cast_fp16)[name = tensor("op_50492_cast_fp16")]; + tensor var_50493_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4831_cast_fp16)[name = tensor("op_50493_cast_fp16")]; + tensor var_50494_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4833_cast_fp16)[name = tensor("op_50494_cast_fp16")]; + tensor var_50495_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4835_cast_fp16)[name = tensor("op_50495_cast_fp16")]; + tensor var_50496_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4837_cast_fp16)[name = tensor("op_50496_cast_fp16")]; + tensor var_50497_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4839_cast_fp16)[name = tensor("op_50497_cast_fp16")]; + tensor var_50498_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4841_cast_fp16)[name = tensor("op_50498_cast_fp16")]; + tensor var_50499_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4843_cast_fp16)[name = tensor("op_50499_cast_fp16")]; + tensor var_50500_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4845_cast_fp16)[name = tensor("op_50500_cast_fp16")]; + tensor var_50501_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4847_cast_fp16)[name = tensor("op_50501_cast_fp16")]; + tensor var_50502_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4849_cast_fp16)[name = tensor("op_50502_cast_fp16")]; + tensor var_50503_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4851_cast_fp16)[name = tensor("op_50503_cast_fp16")]; + tensor var_50504_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4853_cast_fp16)[name = tensor("op_50504_cast_fp16")]; + tensor var_50505_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4855_cast_fp16)[name = tensor("op_50505_cast_fp16")]; + tensor var_50506_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4857_cast_fp16)[name = tensor("op_50506_cast_fp16")]; + tensor var_50507_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4859_cast_fp16)[name = tensor("op_50507_cast_fp16")]; + tensor var_50508_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4861_cast_fp16)[name = tensor("op_50508_cast_fp16")]; + tensor var_50509_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4863_cast_fp16)[name = tensor("op_50509_cast_fp16")]; + tensor var_50510_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4865_cast_fp16)[name = tensor("op_50510_cast_fp16")]; + tensor var_50511_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4867_cast_fp16)[name = tensor("op_50511_cast_fp16")]; + tensor var_50512_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4869_cast_fp16)[name = tensor("op_50512_cast_fp16")]; + tensor var_50513_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4871_cast_fp16)[name = tensor("op_50513_cast_fp16")]; + tensor var_50514_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4873_cast_fp16)[name = tensor("op_50514_cast_fp16")]; + tensor var_50515_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4875_cast_fp16)[name = tensor("op_50515_cast_fp16")]; + tensor var_50516_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4877_cast_fp16)[name = tensor("op_50516_cast_fp16")]; + tensor var_50517_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4879_cast_fp16)[name = tensor("op_50517_cast_fp16")]; + tensor var_50518_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4881_cast_fp16)[name = tensor("op_50518_cast_fp16")]; + tensor var_50519_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4883_cast_fp16)[name = tensor("op_50519_cast_fp16")]; + tensor var_50520_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4885_cast_fp16)[name = tensor("op_50520_cast_fp16")]; + tensor var_50521_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4887_cast_fp16)[name = tensor("op_50521_cast_fp16")]; + tensor var_50522_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4889_cast_fp16)[name = tensor("op_50522_cast_fp16")]; + tensor var_50523_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4891_cast_fp16)[name = tensor("op_50523_cast_fp16")]; + tensor var_50524_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4893_cast_fp16)[name = tensor("op_50524_cast_fp16")]; + tensor var_50525_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4895_cast_fp16)[name = tensor("op_50525_cast_fp16")]; + tensor var_50526_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4897_cast_fp16)[name = tensor("op_50526_cast_fp16")]; + tensor var_50527_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4899_cast_fp16)[name = tensor("op_50527_cast_fp16")]; + tensor var_50528_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4901_cast_fp16)[name = tensor("op_50528_cast_fp16")]; + tensor var_50529_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4903_cast_fp16)[name = tensor("op_50529_cast_fp16")]; + tensor var_50530_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4905_cast_fp16)[name = tensor("op_50530_cast_fp16")]; + tensor var_50531_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4907_cast_fp16)[name = tensor("op_50531_cast_fp16")]; + tensor var_50532_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4909_cast_fp16)[name = tensor("op_50532_cast_fp16")]; + tensor var_50533_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4911_cast_fp16)[name = tensor("op_50533_cast_fp16")]; + tensor var_50534_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4913_cast_fp16)[name = tensor("op_50534_cast_fp16")]; + tensor var_50535_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4915_cast_fp16)[name = tensor("op_50535_cast_fp16")]; + tensor var_50536_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4917_cast_fp16)[name = tensor("op_50536_cast_fp16")]; + tensor var_50537_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4919_cast_fp16)[name = tensor("op_50537_cast_fp16")]; + tensor var_50538_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4921_cast_fp16)[name = tensor("op_50538_cast_fp16")]; + tensor var_50539_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4923_cast_fp16)[name = tensor("op_50539_cast_fp16")]; + tensor var_50540_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4925_cast_fp16)[name = tensor("op_50540_cast_fp16")]; + tensor var_50541_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4927_cast_fp16)[name = tensor("op_50541_cast_fp16")]; + tensor var_50542_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4929_cast_fp16)[name = tensor("op_50542_cast_fp16")]; + tensor var_50543_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4931_cast_fp16)[name = tensor("op_50543_cast_fp16")]; + tensor var_50544_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4933_cast_fp16)[name = tensor("op_50544_cast_fp16")]; + tensor var_50545_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4935_cast_fp16)[name = tensor("op_50545_cast_fp16")]; + tensor var_50546_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4937_cast_fp16)[name = tensor("op_50546_cast_fp16")]; + tensor var_50547_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4939_cast_fp16)[name = tensor("op_50547_cast_fp16")]; + tensor var_50548_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4941_cast_fp16)[name = tensor("op_50548_cast_fp16")]; + tensor var_50549_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4943_cast_fp16)[name = tensor("op_50549_cast_fp16")]; + tensor var_50550_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4945_cast_fp16)[name = tensor("op_50550_cast_fp16")]; + tensor var_50551_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4947_cast_fp16)[name = tensor("op_50551_cast_fp16")]; + tensor var_50552_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4949_cast_fp16)[name = tensor("op_50552_cast_fp16")]; + tensor var_50553_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4951_cast_fp16)[name = tensor("op_50553_cast_fp16")]; + tensor var_50554_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4953_cast_fp16)[name = tensor("op_50554_cast_fp16")]; + tensor var_50555_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4955_cast_fp16)[name = tensor("op_50555_cast_fp16")]; + tensor var_50556_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4957_cast_fp16)[name = tensor("op_50556_cast_fp16")]; + tensor var_50557_cast_fp16 = softmax(axis = var_49249, x = aw_chunk_4959_cast_fp16)[name = tensor("op_50557_cast_fp16")]; + tensor var_50559_equation_0 = const()[name = tensor("op_50559_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50559_cast_fp16 = einsum(equation = var_50559_equation_0, values = (var_50079_cast_fp16, var_50478_cast_fp16))[name = tensor("op_50559_cast_fp16")]; + tensor var_50561_equation_0 = const()[name = tensor("op_50561_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50561_cast_fp16 = einsum(equation = var_50561_equation_0, values = (var_50079_cast_fp16, var_50479_cast_fp16))[name = tensor("op_50561_cast_fp16")]; + tensor var_50563_equation_0 = const()[name = tensor("op_50563_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50563_cast_fp16 = einsum(equation = var_50563_equation_0, values = (var_50079_cast_fp16, var_50480_cast_fp16))[name = tensor("op_50563_cast_fp16")]; + tensor var_50565_equation_0 = const()[name = tensor("op_50565_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50565_cast_fp16 = einsum(equation = var_50565_equation_0, values = (var_50079_cast_fp16, var_50481_cast_fp16))[name = tensor("op_50565_cast_fp16")]; + tensor var_50567_equation_0 = const()[name = tensor("op_50567_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50567_cast_fp16 = einsum(equation = var_50567_equation_0, values = (var_50083_cast_fp16, var_50482_cast_fp16))[name = tensor("op_50567_cast_fp16")]; + tensor var_50569_equation_0 = const()[name = tensor("op_50569_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50569_cast_fp16 = einsum(equation = var_50569_equation_0, values = (var_50083_cast_fp16, var_50483_cast_fp16))[name = tensor("op_50569_cast_fp16")]; + tensor var_50571_equation_0 = const()[name = tensor("op_50571_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50571_cast_fp16 = einsum(equation = var_50571_equation_0, values = (var_50083_cast_fp16, var_50484_cast_fp16))[name = tensor("op_50571_cast_fp16")]; + tensor var_50573_equation_0 = const()[name = tensor("op_50573_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50573_cast_fp16 = einsum(equation = var_50573_equation_0, values = (var_50083_cast_fp16, var_50485_cast_fp16))[name = tensor("op_50573_cast_fp16")]; + tensor var_50575_equation_0 = const()[name = tensor("op_50575_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50575_cast_fp16 = einsum(equation = var_50575_equation_0, values = (var_50087_cast_fp16, var_50486_cast_fp16))[name = tensor("op_50575_cast_fp16")]; + tensor var_50577_equation_0 = const()[name = tensor("op_50577_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50577_cast_fp16 = einsum(equation = var_50577_equation_0, values = (var_50087_cast_fp16, var_50487_cast_fp16))[name = tensor("op_50577_cast_fp16")]; + tensor var_50579_equation_0 = const()[name = tensor("op_50579_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50579_cast_fp16 = einsum(equation = var_50579_equation_0, values = (var_50087_cast_fp16, var_50488_cast_fp16))[name = tensor("op_50579_cast_fp16")]; + tensor var_50581_equation_0 = const()[name = tensor("op_50581_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50581_cast_fp16 = einsum(equation = var_50581_equation_0, values = (var_50087_cast_fp16, var_50489_cast_fp16))[name = tensor("op_50581_cast_fp16")]; + tensor var_50583_equation_0 = const()[name = tensor("op_50583_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50583_cast_fp16 = einsum(equation = var_50583_equation_0, values = (var_50091_cast_fp16, var_50490_cast_fp16))[name = tensor("op_50583_cast_fp16")]; + tensor var_50585_equation_0 = const()[name = tensor("op_50585_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50585_cast_fp16 = einsum(equation = var_50585_equation_0, values = (var_50091_cast_fp16, var_50491_cast_fp16))[name = tensor("op_50585_cast_fp16")]; + tensor var_50587_equation_0 = const()[name = tensor("op_50587_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50587_cast_fp16 = einsum(equation = var_50587_equation_0, values = (var_50091_cast_fp16, var_50492_cast_fp16))[name = tensor("op_50587_cast_fp16")]; + tensor var_50589_equation_0 = const()[name = tensor("op_50589_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50589_cast_fp16 = einsum(equation = var_50589_equation_0, values = (var_50091_cast_fp16, var_50493_cast_fp16))[name = tensor("op_50589_cast_fp16")]; + tensor var_50591_equation_0 = const()[name = tensor("op_50591_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50591_cast_fp16 = einsum(equation = var_50591_equation_0, values = (var_50095_cast_fp16, var_50494_cast_fp16))[name = tensor("op_50591_cast_fp16")]; + tensor var_50593_equation_0 = const()[name = tensor("op_50593_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50593_cast_fp16 = einsum(equation = var_50593_equation_0, values = (var_50095_cast_fp16, var_50495_cast_fp16))[name = tensor("op_50593_cast_fp16")]; + tensor var_50595_equation_0 = const()[name = tensor("op_50595_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50595_cast_fp16 = einsum(equation = var_50595_equation_0, values = (var_50095_cast_fp16, var_50496_cast_fp16))[name = tensor("op_50595_cast_fp16")]; + tensor var_50597_equation_0 = const()[name = tensor("op_50597_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50597_cast_fp16 = einsum(equation = var_50597_equation_0, values = (var_50095_cast_fp16, var_50497_cast_fp16))[name = tensor("op_50597_cast_fp16")]; + tensor var_50599_equation_0 = const()[name = tensor("op_50599_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50599_cast_fp16 = einsum(equation = var_50599_equation_0, values = (var_50099_cast_fp16, var_50498_cast_fp16))[name = tensor("op_50599_cast_fp16")]; + tensor var_50601_equation_0 = const()[name = tensor("op_50601_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50601_cast_fp16 = einsum(equation = var_50601_equation_0, values = (var_50099_cast_fp16, var_50499_cast_fp16))[name = tensor("op_50601_cast_fp16")]; + tensor var_50603_equation_0 = const()[name = tensor("op_50603_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50603_cast_fp16 = einsum(equation = var_50603_equation_0, values = (var_50099_cast_fp16, var_50500_cast_fp16))[name = tensor("op_50603_cast_fp16")]; + tensor var_50605_equation_0 = const()[name = tensor("op_50605_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50605_cast_fp16 = einsum(equation = var_50605_equation_0, values = (var_50099_cast_fp16, var_50501_cast_fp16))[name = tensor("op_50605_cast_fp16")]; + tensor var_50607_equation_0 = const()[name = tensor("op_50607_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50607_cast_fp16 = einsum(equation = var_50607_equation_0, values = (var_50103_cast_fp16, var_50502_cast_fp16))[name = tensor("op_50607_cast_fp16")]; + tensor var_50609_equation_0 = const()[name = tensor("op_50609_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50609_cast_fp16 = einsum(equation = var_50609_equation_0, values = (var_50103_cast_fp16, var_50503_cast_fp16))[name = tensor("op_50609_cast_fp16")]; + tensor var_50611_equation_0 = const()[name = tensor("op_50611_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50611_cast_fp16 = einsum(equation = var_50611_equation_0, values = (var_50103_cast_fp16, var_50504_cast_fp16))[name = tensor("op_50611_cast_fp16")]; + tensor var_50613_equation_0 = const()[name = tensor("op_50613_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50613_cast_fp16 = einsum(equation = var_50613_equation_0, values = (var_50103_cast_fp16, var_50505_cast_fp16))[name = tensor("op_50613_cast_fp16")]; + tensor var_50615_equation_0 = const()[name = tensor("op_50615_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50615_cast_fp16 = einsum(equation = var_50615_equation_0, values = (var_50107_cast_fp16, var_50506_cast_fp16))[name = tensor("op_50615_cast_fp16")]; + tensor var_50617_equation_0 = const()[name = tensor("op_50617_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50617_cast_fp16 = einsum(equation = var_50617_equation_0, values = (var_50107_cast_fp16, var_50507_cast_fp16))[name = tensor("op_50617_cast_fp16")]; + tensor var_50619_equation_0 = const()[name = tensor("op_50619_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50619_cast_fp16 = einsum(equation = var_50619_equation_0, values = (var_50107_cast_fp16, var_50508_cast_fp16))[name = tensor("op_50619_cast_fp16")]; + tensor var_50621_equation_0 = const()[name = tensor("op_50621_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50621_cast_fp16 = einsum(equation = var_50621_equation_0, values = (var_50107_cast_fp16, var_50509_cast_fp16))[name = tensor("op_50621_cast_fp16")]; + tensor var_50623_equation_0 = const()[name = tensor("op_50623_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50623_cast_fp16 = einsum(equation = var_50623_equation_0, values = (var_50111_cast_fp16, var_50510_cast_fp16))[name = tensor("op_50623_cast_fp16")]; + tensor var_50625_equation_0 = const()[name = tensor("op_50625_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50625_cast_fp16 = einsum(equation = var_50625_equation_0, values = (var_50111_cast_fp16, var_50511_cast_fp16))[name = tensor("op_50625_cast_fp16")]; + tensor var_50627_equation_0 = const()[name = tensor("op_50627_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50627_cast_fp16 = einsum(equation = var_50627_equation_0, values = (var_50111_cast_fp16, var_50512_cast_fp16))[name = tensor("op_50627_cast_fp16")]; + tensor var_50629_equation_0 = const()[name = tensor("op_50629_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50629_cast_fp16 = einsum(equation = var_50629_equation_0, values = (var_50111_cast_fp16, var_50513_cast_fp16))[name = tensor("op_50629_cast_fp16")]; + tensor var_50631_equation_0 = const()[name = tensor("op_50631_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50631_cast_fp16 = einsum(equation = var_50631_equation_0, values = (var_50115_cast_fp16, var_50514_cast_fp16))[name = tensor("op_50631_cast_fp16")]; + tensor var_50633_equation_0 = const()[name = tensor("op_50633_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50633_cast_fp16 = einsum(equation = var_50633_equation_0, values = (var_50115_cast_fp16, var_50515_cast_fp16))[name = tensor("op_50633_cast_fp16")]; + tensor var_50635_equation_0 = const()[name = tensor("op_50635_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50635_cast_fp16 = einsum(equation = var_50635_equation_0, values = (var_50115_cast_fp16, var_50516_cast_fp16))[name = tensor("op_50635_cast_fp16")]; + tensor var_50637_equation_0 = const()[name = tensor("op_50637_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50637_cast_fp16 = einsum(equation = var_50637_equation_0, values = (var_50115_cast_fp16, var_50517_cast_fp16))[name = tensor("op_50637_cast_fp16")]; + tensor var_50639_equation_0 = const()[name = tensor("op_50639_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50639_cast_fp16 = einsum(equation = var_50639_equation_0, values = (var_50119_cast_fp16, var_50518_cast_fp16))[name = tensor("op_50639_cast_fp16")]; + tensor var_50641_equation_0 = const()[name = tensor("op_50641_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50641_cast_fp16 = einsum(equation = var_50641_equation_0, values = (var_50119_cast_fp16, var_50519_cast_fp16))[name = tensor("op_50641_cast_fp16")]; + tensor var_50643_equation_0 = const()[name = tensor("op_50643_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50643_cast_fp16 = einsum(equation = var_50643_equation_0, values = (var_50119_cast_fp16, var_50520_cast_fp16))[name = tensor("op_50643_cast_fp16")]; + tensor var_50645_equation_0 = const()[name = tensor("op_50645_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50645_cast_fp16 = einsum(equation = var_50645_equation_0, values = (var_50119_cast_fp16, var_50521_cast_fp16))[name = tensor("op_50645_cast_fp16")]; + tensor var_50647_equation_0 = const()[name = tensor("op_50647_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50647_cast_fp16 = einsum(equation = var_50647_equation_0, values = (var_50123_cast_fp16, var_50522_cast_fp16))[name = tensor("op_50647_cast_fp16")]; + tensor var_50649_equation_0 = const()[name = tensor("op_50649_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50649_cast_fp16 = einsum(equation = var_50649_equation_0, values = (var_50123_cast_fp16, var_50523_cast_fp16))[name = tensor("op_50649_cast_fp16")]; + tensor var_50651_equation_0 = const()[name = tensor("op_50651_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50651_cast_fp16 = einsum(equation = var_50651_equation_0, values = (var_50123_cast_fp16, var_50524_cast_fp16))[name = tensor("op_50651_cast_fp16")]; + tensor var_50653_equation_0 = const()[name = tensor("op_50653_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50653_cast_fp16 = einsum(equation = var_50653_equation_0, values = (var_50123_cast_fp16, var_50525_cast_fp16))[name = tensor("op_50653_cast_fp16")]; + tensor var_50655_equation_0 = const()[name = tensor("op_50655_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50655_cast_fp16 = einsum(equation = var_50655_equation_0, values = (var_50127_cast_fp16, var_50526_cast_fp16))[name = tensor("op_50655_cast_fp16")]; + tensor var_50657_equation_0 = const()[name = tensor("op_50657_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50657_cast_fp16 = einsum(equation = var_50657_equation_0, values = (var_50127_cast_fp16, var_50527_cast_fp16))[name = tensor("op_50657_cast_fp16")]; + tensor var_50659_equation_0 = const()[name = tensor("op_50659_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50659_cast_fp16 = einsum(equation = var_50659_equation_0, values = (var_50127_cast_fp16, var_50528_cast_fp16))[name = tensor("op_50659_cast_fp16")]; + tensor var_50661_equation_0 = const()[name = tensor("op_50661_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50661_cast_fp16 = einsum(equation = var_50661_equation_0, values = (var_50127_cast_fp16, var_50529_cast_fp16))[name = tensor("op_50661_cast_fp16")]; + tensor var_50663_equation_0 = const()[name = tensor("op_50663_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50663_cast_fp16 = einsum(equation = var_50663_equation_0, values = (var_50131_cast_fp16, var_50530_cast_fp16))[name = tensor("op_50663_cast_fp16")]; + tensor var_50665_equation_0 = const()[name = tensor("op_50665_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50665_cast_fp16 = einsum(equation = var_50665_equation_0, values = (var_50131_cast_fp16, var_50531_cast_fp16))[name = tensor("op_50665_cast_fp16")]; + tensor var_50667_equation_0 = const()[name = tensor("op_50667_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50667_cast_fp16 = einsum(equation = var_50667_equation_0, values = (var_50131_cast_fp16, var_50532_cast_fp16))[name = tensor("op_50667_cast_fp16")]; + tensor var_50669_equation_0 = const()[name = tensor("op_50669_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50669_cast_fp16 = einsum(equation = var_50669_equation_0, values = (var_50131_cast_fp16, var_50533_cast_fp16))[name = tensor("op_50669_cast_fp16")]; + tensor var_50671_equation_0 = const()[name = tensor("op_50671_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50671_cast_fp16 = einsum(equation = var_50671_equation_0, values = (var_50135_cast_fp16, var_50534_cast_fp16))[name = tensor("op_50671_cast_fp16")]; + tensor var_50673_equation_0 = const()[name = tensor("op_50673_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50673_cast_fp16 = einsum(equation = var_50673_equation_0, values = (var_50135_cast_fp16, var_50535_cast_fp16))[name = tensor("op_50673_cast_fp16")]; + tensor var_50675_equation_0 = const()[name = tensor("op_50675_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50675_cast_fp16 = einsum(equation = var_50675_equation_0, values = (var_50135_cast_fp16, var_50536_cast_fp16))[name = tensor("op_50675_cast_fp16")]; + tensor var_50677_equation_0 = const()[name = tensor("op_50677_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50677_cast_fp16 = einsum(equation = var_50677_equation_0, values = (var_50135_cast_fp16, var_50537_cast_fp16))[name = tensor("op_50677_cast_fp16")]; + tensor var_50679_equation_0 = const()[name = tensor("op_50679_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50679_cast_fp16 = einsum(equation = var_50679_equation_0, values = (var_50139_cast_fp16, var_50538_cast_fp16))[name = tensor("op_50679_cast_fp16")]; + tensor var_50681_equation_0 = const()[name = tensor("op_50681_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50681_cast_fp16 = einsum(equation = var_50681_equation_0, values = (var_50139_cast_fp16, var_50539_cast_fp16))[name = tensor("op_50681_cast_fp16")]; + tensor var_50683_equation_0 = const()[name = tensor("op_50683_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50683_cast_fp16 = einsum(equation = var_50683_equation_0, values = (var_50139_cast_fp16, var_50540_cast_fp16))[name = tensor("op_50683_cast_fp16")]; + tensor var_50685_equation_0 = const()[name = tensor("op_50685_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50685_cast_fp16 = einsum(equation = var_50685_equation_0, values = (var_50139_cast_fp16, var_50541_cast_fp16))[name = tensor("op_50685_cast_fp16")]; + tensor var_50687_equation_0 = const()[name = tensor("op_50687_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50687_cast_fp16 = einsum(equation = var_50687_equation_0, values = (var_50143_cast_fp16, var_50542_cast_fp16))[name = tensor("op_50687_cast_fp16")]; + tensor var_50689_equation_0 = const()[name = tensor("op_50689_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50689_cast_fp16 = einsum(equation = var_50689_equation_0, values = (var_50143_cast_fp16, var_50543_cast_fp16))[name = tensor("op_50689_cast_fp16")]; + tensor var_50691_equation_0 = const()[name = tensor("op_50691_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50691_cast_fp16 = einsum(equation = var_50691_equation_0, values = (var_50143_cast_fp16, var_50544_cast_fp16))[name = tensor("op_50691_cast_fp16")]; + tensor var_50693_equation_0 = const()[name = tensor("op_50693_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50693_cast_fp16 = einsum(equation = var_50693_equation_0, values = (var_50143_cast_fp16, var_50545_cast_fp16))[name = tensor("op_50693_cast_fp16")]; + tensor var_50695_equation_0 = const()[name = tensor("op_50695_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50695_cast_fp16 = einsum(equation = var_50695_equation_0, values = (var_50147_cast_fp16, var_50546_cast_fp16))[name = tensor("op_50695_cast_fp16")]; + tensor var_50697_equation_0 = const()[name = tensor("op_50697_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50697_cast_fp16 = einsum(equation = var_50697_equation_0, values = (var_50147_cast_fp16, var_50547_cast_fp16))[name = tensor("op_50697_cast_fp16")]; + tensor var_50699_equation_0 = const()[name = tensor("op_50699_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50699_cast_fp16 = einsum(equation = var_50699_equation_0, values = (var_50147_cast_fp16, var_50548_cast_fp16))[name = tensor("op_50699_cast_fp16")]; + tensor var_50701_equation_0 = const()[name = tensor("op_50701_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50701_cast_fp16 = einsum(equation = var_50701_equation_0, values = (var_50147_cast_fp16, var_50549_cast_fp16))[name = tensor("op_50701_cast_fp16")]; + tensor var_50703_equation_0 = const()[name = tensor("op_50703_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50703_cast_fp16 = einsum(equation = var_50703_equation_0, values = (var_50151_cast_fp16, var_50550_cast_fp16))[name = tensor("op_50703_cast_fp16")]; + tensor var_50705_equation_0 = const()[name = tensor("op_50705_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50705_cast_fp16 = einsum(equation = var_50705_equation_0, values = (var_50151_cast_fp16, var_50551_cast_fp16))[name = tensor("op_50705_cast_fp16")]; + tensor var_50707_equation_0 = const()[name = tensor("op_50707_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50707_cast_fp16 = einsum(equation = var_50707_equation_0, values = (var_50151_cast_fp16, var_50552_cast_fp16))[name = tensor("op_50707_cast_fp16")]; + tensor var_50709_equation_0 = const()[name = tensor("op_50709_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50709_cast_fp16 = einsum(equation = var_50709_equation_0, values = (var_50151_cast_fp16, var_50553_cast_fp16))[name = tensor("op_50709_cast_fp16")]; + tensor var_50711_equation_0 = const()[name = tensor("op_50711_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50711_cast_fp16 = einsum(equation = var_50711_equation_0, values = (var_50155_cast_fp16, var_50554_cast_fp16))[name = tensor("op_50711_cast_fp16")]; + tensor var_50713_equation_0 = const()[name = tensor("op_50713_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50713_cast_fp16 = einsum(equation = var_50713_equation_0, values = (var_50155_cast_fp16, var_50555_cast_fp16))[name = tensor("op_50713_cast_fp16")]; + tensor var_50715_equation_0 = const()[name = tensor("op_50715_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50715_cast_fp16 = einsum(equation = var_50715_equation_0, values = (var_50155_cast_fp16, var_50556_cast_fp16))[name = tensor("op_50715_cast_fp16")]; + tensor var_50717_equation_0 = const()[name = tensor("op_50717_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_50717_cast_fp16 = einsum(equation = var_50717_equation_0, values = (var_50155_cast_fp16, var_50557_cast_fp16))[name = tensor("op_50717_cast_fp16")]; + tensor var_50719_interleave_0 = const()[name = tensor("op_50719_interleave_0"), val = tensor(false)]; + tensor var_50719_cast_fp16 = concat(axis = var_49224, interleave = var_50719_interleave_0, values = (var_50559_cast_fp16, var_50561_cast_fp16, var_50563_cast_fp16, var_50565_cast_fp16))[name = tensor("op_50719_cast_fp16")]; + tensor var_50721_interleave_0 = const()[name = tensor("op_50721_interleave_0"), val = tensor(false)]; + tensor var_50721_cast_fp16 = concat(axis = var_49224, interleave = var_50721_interleave_0, values = (var_50567_cast_fp16, var_50569_cast_fp16, var_50571_cast_fp16, var_50573_cast_fp16))[name = tensor("op_50721_cast_fp16")]; + tensor var_50723_interleave_0 = const()[name = tensor("op_50723_interleave_0"), val = tensor(false)]; + tensor var_50723_cast_fp16 = concat(axis = var_49224, interleave = var_50723_interleave_0, values = (var_50575_cast_fp16, var_50577_cast_fp16, var_50579_cast_fp16, var_50581_cast_fp16))[name = tensor("op_50723_cast_fp16")]; + tensor var_50725_interleave_0 = const()[name = tensor("op_50725_interleave_0"), val = tensor(false)]; + tensor var_50725_cast_fp16 = concat(axis = var_49224, interleave = var_50725_interleave_0, values = (var_50583_cast_fp16, var_50585_cast_fp16, var_50587_cast_fp16, var_50589_cast_fp16))[name = tensor("op_50725_cast_fp16")]; + tensor var_50727_interleave_0 = const()[name = tensor("op_50727_interleave_0"), val = tensor(false)]; + tensor var_50727_cast_fp16 = concat(axis = var_49224, interleave = var_50727_interleave_0, values = (var_50591_cast_fp16, var_50593_cast_fp16, var_50595_cast_fp16, var_50597_cast_fp16))[name = tensor("op_50727_cast_fp16")]; + tensor var_50729_interleave_0 = const()[name = tensor("op_50729_interleave_0"), val = tensor(false)]; + tensor var_50729_cast_fp16 = concat(axis = var_49224, interleave = var_50729_interleave_0, values = (var_50599_cast_fp16, var_50601_cast_fp16, var_50603_cast_fp16, var_50605_cast_fp16))[name = tensor("op_50729_cast_fp16")]; + tensor var_50731_interleave_0 = const()[name = tensor("op_50731_interleave_0"), val = tensor(false)]; + tensor var_50731_cast_fp16 = concat(axis = var_49224, interleave = var_50731_interleave_0, values = (var_50607_cast_fp16, var_50609_cast_fp16, var_50611_cast_fp16, var_50613_cast_fp16))[name = tensor("op_50731_cast_fp16")]; + tensor var_50733_interleave_0 = const()[name = tensor("op_50733_interleave_0"), val = tensor(false)]; + tensor var_50733_cast_fp16 = concat(axis = var_49224, interleave = var_50733_interleave_0, values = (var_50615_cast_fp16, var_50617_cast_fp16, var_50619_cast_fp16, var_50621_cast_fp16))[name = tensor("op_50733_cast_fp16")]; + tensor var_50735_interleave_0 = const()[name = tensor("op_50735_interleave_0"), val = tensor(false)]; + tensor var_50735_cast_fp16 = concat(axis = var_49224, interleave = var_50735_interleave_0, values = (var_50623_cast_fp16, var_50625_cast_fp16, var_50627_cast_fp16, var_50629_cast_fp16))[name = tensor("op_50735_cast_fp16")]; + tensor var_50737_interleave_0 = const()[name = tensor("op_50737_interleave_0"), val = tensor(false)]; + tensor var_50737_cast_fp16 = concat(axis = var_49224, interleave = var_50737_interleave_0, values = (var_50631_cast_fp16, var_50633_cast_fp16, var_50635_cast_fp16, var_50637_cast_fp16))[name = tensor("op_50737_cast_fp16")]; + tensor var_50739_interleave_0 = const()[name = tensor("op_50739_interleave_0"), val = tensor(false)]; + tensor var_50739_cast_fp16 = concat(axis = var_49224, interleave = var_50739_interleave_0, values = (var_50639_cast_fp16, var_50641_cast_fp16, var_50643_cast_fp16, var_50645_cast_fp16))[name = tensor("op_50739_cast_fp16")]; + tensor var_50741_interleave_0 = const()[name = tensor("op_50741_interleave_0"), val = tensor(false)]; + tensor var_50741_cast_fp16 = concat(axis = var_49224, interleave = var_50741_interleave_0, values = (var_50647_cast_fp16, var_50649_cast_fp16, var_50651_cast_fp16, var_50653_cast_fp16))[name = tensor("op_50741_cast_fp16")]; + tensor var_50743_interleave_0 = const()[name = tensor("op_50743_interleave_0"), val = tensor(false)]; + tensor var_50743_cast_fp16 = concat(axis = var_49224, interleave = var_50743_interleave_0, values = (var_50655_cast_fp16, var_50657_cast_fp16, var_50659_cast_fp16, var_50661_cast_fp16))[name = tensor("op_50743_cast_fp16")]; + tensor var_50745_interleave_0 = const()[name = tensor("op_50745_interleave_0"), val = tensor(false)]; + tensor var_50745_cast_fp16 = concat(axis = var_49224, interleave = var_50745_interleave_0, values = (var_50663_cast_fp16, var_50665_cast_fp16, var_50667_cast_fp16, var_50669_cast_fp16))[name = tensor("op_50745_cast_fp16")]; + tensor var_50747_interleave_0 = const()[name = tensor("op_50747_interleave_0"), val = tensor(false)]; + tensor var_50747_cast_fp16 = concat(axis = var_49224, interleave = var_50747_interleave_0, values = (var_50671_cast_fp16, var_50673_cast_fp16, var_50675_cast_fp16, var_50677_cast_fp16))[name = tensor("op_50747_cast_fp16")]; + tensor var_50749_interleave_0 = const()[name = tensor("op_50749_interleave_0"), val = tensor(false)]; + tensor var_50749_cast_fp16 = concat(axis = var_49224, interleave = var_50749_interleave_0, values = (var_50679_cast_fp16, var_50681_cast_fp16, var_50683_cast_fp16, var_50685_cast_fp16))[name = tensor("op_50749_cast_fp16")]; + tensor var_50751_interleave_0 = const()[name = tensor("op_50751_interleave_0"), val = tensor(false)]; + tensor var_50751_cast_fp16 = concat(axis = var_49224, interleave = var_50751_interleave_0, values = (var_50687_cast_fp16, var_50689_cast_fp16, var_50691_cast_fp16, var_50693_cast_fp16))[name = tensor("op_50751_cast_fp16")]; + tensor var_50753_interleave_0 = const()[name = tensor("op_50753_interleave_0"), val = tensor(false)]; + tensor var_50753_cast_fp16 = concat(axis = var_49224, interleave = var_50753_interleave_0, values = (var_50695_cast_fp16, var_50697_cast_fp16, var_50699_cast_fp16, var_50701_cast_fp16))[name = tensor("op_50753_cast_fp16")]; + tensor var_50755_interleave_0 = const()[name = tensor("op_50755_interleave_0"), val = tensor(false)]; + tensor var_50755_cast_fp16 = concat(axis = var_49224, interleave = var_50755_interleave_0, values = (var_50703_cast_fp16, var_50705_cast_fp16, var_50707_cast_fp16, var_50709_cast_fp16))[name = tensor("op_50755_cast_fp16")]; + tensor var_50757_interleave_0 = const()[name = tensor("op_50757_interleave_0"), val = tensor(false)]; + tensor var_50757_cast_fp16 = concat(axis = var_49224, interleave = var_50757_interleave_0, values = (var_50711_cast_fp16, var_50713_cast_fp16, var_50715_cast_fp16, var_50717_cast_fp16))[name = tensor("op_50757_cast_fp16")]; + tensor input_607_interleave_0 = const()[name = tensor("input_607_interleave_0"), val = tensor(false)]; + tensor input_607_cast_fp16 = concat(axis = var_49249, interleave = input_607_interleave_0, values = (var_50719_cast_fp16, var_50721_cast_fp16, var_50723_cast_fp16, var_50725_cast_fp16, var_50727_cast_fp16, var_50729_cast_fp16, var_50731_cast_fp16, var_50733_cast_fp16, var_50735_cast_fp16, var_50737_cast_fp16, var_50739_cast_fp16, var_50741_cast_fp16, var_50743_cast_fp16, var_50745_cast_fp16, var_50747_cast_fp16, var_50749_cast_fp16, var_50751_cast_fp16, var_50753_cast_fp16, var_50755_cast_fp16, var_50757_cast_fp16))[name = tensor("input_607_cast_fp16")]; + tensor var_50765 = const()[name = tensor("op_50765"), val = tensor([1, 1])]; + tensor var_50767 = const()[name = tensor("op_50767"), val = tensor([1, 1])]; + tensor pretrained_out_367_pad_type_0 = const()[name = tensor("pretrained_out_367_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_367_pad_0 = const()[name = tensor("pretrained_out_367_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335047552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335866816))), name = tensor("layers_30_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_30_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_30_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335866944)))]; + tensor pretrained_out_367_cast_fp16 = conv(bias = layers_30_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_50767, groups = var_49249, pad = pretrained_out_367_pad_0, pad_type = pretrained_out_367_pad_type_0, strides = var_50765, weight = layers_30_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_607_cast_fp16)[name = tensor("pretrained_out_367_cast_fp16")]; + tensor var_50771 = const()[name = tensor("op_50771"), val = tensor([1, 1])]; + tensor var_50773 = const()[name = tensor("op_50773"), val = tensor([1, 1])]; + tensor input_609_pad_type_0 = const()[name = tensor("input_609_pad_type_0"), val = tensor("custom")]; + tensor input_609_pad_0 = const()[name = tensor("input_609_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_30_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335869568)))]; + tensor input_609_cast_fp16 = conv(dilations = var_50773, groups = var_49249, pad = input_609_pad_0, pad_type = input_609_pad_type_0, strides = var_50771, weight = layers_30_self_attn_o_proj_loraA_weight_to_fp16, x = input_607_cast_fp16)[name = tensor("input_609_cast_fp16")]; + tensor var_50777 = const()[name = tensor("op_50777"), val = tensor([1, 1])]; + tensor var_50779 = const()[name = tensor("op_50779"), val = tensor([1, 1])]; + tensor lora_out_733_pad_type_0 = const()[name = tensor("lora_out_733_pad_type_0"), val = tensor("custom")]; + tensor lora_out_733_pad_0 = const()[name = tensor("lora_out_733_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_735_weight_0_to_fp16 = const()[name = tensor("lora_out_735_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335910592)))]; + tensor lora_out_735_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_50779, groups = var_49249, pad = lora_out_733_pad_0, pad_type = lora_out_733_pad_type_0, strides = var_50777, weight = lora_out_735_weight_0_to_fp16, x = input_609_cast_fp16)[name = tensor("lora_out_735_cast_fp16")]; + tensor obj_123_cast_fp16 = add(x = pretrained_out_367_cast_fp16, y = lora_out_735_cast_fp16)[name = tensor("obj_123_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = obj_123_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; + tensor var_50788 = const()[name = tensor("op_50788"), val = tensor([1])]; + tensor channels_mean_123_cast_fp16 = reduce_mean(axes = var_50788, keep_dims = var_49250, x = inputs_123_cast_fp16)[name = tensor("channels_mean_123_cast_fp16")]; + tensor zero_mean_123_cast_fp16 = sub(x = inputs_123_cast_fp16, y = channels_mean_123_cast_fp16)[name = tensor("zero_mean_123_cast_fp16")]; + tensor zero_mean_sq_123_cast_fp16 = mul(x = zero_mean_123_cast_fp16, y = zero_mean_123_cast_fp16)[name = tensor("zero_mean_sq_123_cast_fp16")]; + tensor var_50792 = const()[name = tensor("op_50792"), val = tensor([1])]; + tensor var_50793_cast_fp16 = reduce_mean(axes = var_50792, keep_dims = var_49250, x = zero_mean_sq_123_cast_fp16)[name = tensor("op_50793_cast_fp16")]; + tensor var_50794_to_fp16 = const()[name = tensor("op_50794_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_50795_cast_fp16 = add(x = var_50793_cast_fp16, y = var_50794_to_fp16)[name = tensor("op_50795_cast_fp16")]; + tensor denom_123_epsilon_0 = const()[name = tensor("denom_123_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_123_cast_fp16 = rsqrt(epsilon = denom_123_epsilon_0, x = var_50795_cast_fp16)[name = tensor("denom_123_cast_fp16")]; + tensor out_123_cast_fp16 = mul(x = zero_mean_123_cast_fp16, y = denom_123_cast_fp16)[name = tensor("out_123_cast_fp16")]; + tensor input_611_gamma_0_to_fp16 = const()[name = tensor("input_611_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335951616)))]; + tensor input_611_beta_0_to_fp16 = const()[name = tensor("input_611_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335954240)))]; + tensor input_611_epsilon_0_to_fp16 = const()[name = tensor("input_611_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_611_cast_fp16 = batch_norm(beta = input_611_beta_0_to_fp16, epsilon = input_611_epsilon_0_to_fp16, gamma = input_611_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_123_cast_fp16)[name = tensor("input_611_cast_fp16")]; + tensor var_50809 = const()[name = tensor("op_50809"), val = tensor([1, 1])]; + tensor var_50811 = const()[name = tensor("op_50811"), val = tensor([1, 1])]; + tensor pretrained_out_369_pad_type_0 = const()[name = tensor("pretrained_out_369_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_369_pad_0 = const()[name = tensor("pretrained_out_369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335956864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339233728))), name = tensor("layers_30_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_30_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_30_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339233856)))]; + tensor pretrained_out_369_cast_fp16 = conv(bias = layers_30_fc1_pretrained_bias_to_fp16, dilations = var_50811, groups = var_49249, pad = pretrained_out_369_pad_0, pad_type = pretrained_out_369_pad_type_0, strides = var_50809, weight = layers_30_fc1_pretrained_weight_to_fp16_palettized, x = input_611_cast_fp16)[name = tensor("pretrained_out_369_cast_fp16")]; + tensor var_50815 = const()[name = tensor("op_50815"), val = tensor([1, 1])]; + tensor var_50817 = const()[name = tensor("op_50817"), val = tensor([1, 1])]; + tensor input_613_pad_type_0 = const()[name = tensor("input_613_pad_type_0"), val = tensor("custom")]; + tensor input_613_pad_0 = const()[name = tensor("input_613_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_30_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339244160)))]; + tensor input_613_cast_fp16 = conv(dilations = var_50817, groups = var_49249, pad = input_613_pad_0, pad_type = input_613_pad_type_0, strides = var_50815, weight = layers_30_fc1_loraA_weight_to_fp16, x = input_611_cast_fp16)[name = tensor("input_613_cast_fp16")]; + tensor var_50821 = const()[name = tensor("op_50821"), val = tensor([1, 1])]; + tensor var_50823 = const()[name = tensor("op_50823"), val = tensor([1, 1])]; + tensor lora_out_737_pad_type_0 = const()[name = tensor("lora_out_737_pad_type_0"), val = tensor("custom")]; + tensor lora_out_737_pad_0 = const()[name = tensor("lora_out_737_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_739_weight_0_to_fp16 = const()[name = tensor("lora_out_739_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339285184)))]; + tensor lora_out_739_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_50823, groups = var_49249, pad = lora_out_737_pad_0, pad_type = lora_out_737_pad_type_0, strides = var_50821, weight = lora_out_739_weight_0_to_fp16, x = input_613_cast_fp16)[name = tensor("lora_out_739_cast_fp16")]; + tensor input_615_cast_fp16 = add(x = pretrained_out_369_cast_fp16, y = lora_out_739_cast_fp16)[name = tensor("input_615_cast_fp16")]; + tensor input_617_mode_0 = const()[name = tensor("input_617_mode_0"), val = tensor("EXACT")]; + tensor input_617_cast_fp16 = gelu(mode = input_617_mode_0, x = input_615_cast_fp16)[name = tensor("input_617_cast_fp16")]; + tensor var_50835 = const()[name = tensor("op_50835"), val = tensor([1, 1])]; + tensor var_50837 = const()[name = tensor("op_50837"), val = tensor([1, 1])]; + tensor pretrained_out_371_pad_type_0 = const()[name = tensor("pretrained_out_371_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_371_pad_0 = const()[name = tensor("pretrained_out_371_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339449088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342725952))), name = tensor("layers_30_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_30_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_30_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342726080)))]; + tensor pretrained_out_371_cast_fp16 = conv(bias = layers_30_fc2_pretrained_bias_to_fp16, dilations = var_50837, groups = var_49249, pad = pretrained_out_371_pad_0, pad_type = pretrained_out_371_pad_type_0, strides = var_50835, weight = layers_30_fc2_pretrained_weight_to_fp16_palettized, x = input_617_cast_fp16)[name = tensor("pretrained_out_371_cast_fp16")]; + tensor var_50841 = const()[name = tensor("op_50841"), val = tensor([1, 1])]; + tensor var_50843 = const()[name = tensor("op_50843"), val = tensor([1, 1])]; + tensor input_619_pad_type_0 = const()[name = tensor("input_619_pad_type_0"), val = tensor("custom")]; + tensor input_619_pad_0 = const()[name = tensor("input_619_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_30_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_30_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342728704)))]; + tensor input_619_cast_fp16 = conv(dilations = var_50843, groups = var_49249, pad = input_619_pad_0, pad_type = input_619_pad_type_0, strides = var_50841, weight = layers_30_fc2_loraA_weight_to_fp16, x = input_617_cast_fp16)[name = tensor("input_619_cast_fp16")]; + tensor var_50847 = const()[name = tensor("op_50847"), val = tensor([1, 1])]; + tensor var_50849 = const()[name = tensor("op_50849"), val = tensor([1, 1])]; + tensor lora_out_741_pad_type_0 = const()[name = tensor("lora_out_741_pad_type_0"), val = tensor("custom")]; + tensor lora_out_741_pad_0 = const()[name = tensor("lora_out_741_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_743_weight_0_to_fp16 = const()[name = tensor("lora_out_743_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342892608)))]; + tensor lora_out_743_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_50849, groups = var_49249, pad = lora_out_741_pad_0, pad_type = lora_out_741_pad_type_0, strides = var_50847, weight = lora_out_743_weight_0_to_fp16, x = input_619_cast_fp16)[name = tensor("lora_out_743_cast_fp16")]; + tensor hidden_states_65_cast_fp16 = add(x = pretrained_out_371_cast_fp16, y = lora_out_743_cast_fp16)[name = tensor("hidden_states_65_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = hidden_states_65_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; + tensor var_50859 = const()[name = tensor("op_50859"), val = tensor(3)]; + tensor var_50884 = const()[name = tensor("op_50884"), val = tensor(1)]; + tensor var_50885 = const()[name = tensor("op_50885"), val = tensor(true)]; + tensor var_50895 = const()[name = tensor("op_50895"), val = tensor([1])]; + tensor channels_mean_125_cast_fp16 = reduce_mean(axes = var_50895, keep_dims = var_50885, x = inputs_125_cast_fp16)[name = tensor("channels_mean_125_cast_fp16")]; + tensor zero_mean_125_cast_fp16 = sub(x = inputs_125_cast_fp16, y = channels_mean_125_cast_fp16)[name = tensor("zero_mean_125_cast_fp16")]; + tensor zero_mean_sq_125_cast_fp16 = mul(x = zero_mean_125_cast_fp16, y = zero_mean_125_cast_fp16)[name = tensor("zero_mean_sq_125_cast_fp16")]; + tensor var_50899 = const()[name = tensor("op_50899"), val = tensor([1])]; + tensor var_50900_cast_fp16 = reduce_mean(axes = var_50899, keep_dims = var_50885, x = zero_mean_sq_125_cast_fp16)[name = tensor("op_50900_cast_fp16")]; + tensor var_50901_to_fp16 = const()[name = tensor("op_50901_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_50902_cast_fp16 = add(x = var_50900_cast_fp16, y = var_50901_to_fp16)[name = tensor("op_50902_cast_fp16")]; + tensor denom_125_epsilon_0 = const()[name = tensor("denom_125_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_125_cast_fp16 = rsqrt(epsilon = denom_125_epsilon_0, x = var_50902_cast_fp16)[name = tensor("denom_125_cast_fp16")]; + tensor out_125_cast_fp16 = mul(x = zero_mean_125_cast_fp16, y = denom_125_cast_fp16)[name = tensor("out_125_cast_fp16")]; + tensor obj_125_gamma_0_to_fp16 = const()[name = tensor("obj_125_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342933632)))]; + tensor obj_125_beta_0_to_fp16 = const()[name = tensor("obj_125_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342936256)))]; + tensor obj_125_epsilon_0_to_fp16 = const()[name = tensor("obj_125_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_125_cast_fp16 = batch_norm(beta = obj_125_beta_0_to_fp16, epsilon = obj_125_epsilon_0_to_fp16, gamma = obj_125_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_125_cast_fp16)[name = tensor("obj_125_cast_fp16")]; + tensor var_50920 = const()[name = tensor("op_50920"), val = tensor([1, 1])]; + tensor var_50922 = const()[name = tensor("op_50922"), val = tensor([1, 1])]; + tensor pretrained_out_373_pad_type_0 = const()[name = tensor("pretrained_out_373_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_373_pad_0 = const()[name = tensor("pretrained_out_373_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_q_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342938880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343758144))), name = tensor("layers_31_self_attn_q_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_31_self_attn_q_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_q_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343758272)))]; + tensor pretrained_out_373_cast_fp16 = conv(bias = layers_31_self_attn_q_proj_pretrained_bias_to_fp16, dilations = var_50922, groups = var_50884, pad = pretrained_out_373_pad_0, pad_type = pretrained_out_373_pad_type_0, strides = var_50920, weight = layers_31_self_attn_q_proj_pretrained_weight_to_fp16_palettized, x = obj_125_cast_fp16)[name = tensor("pretrained_out_373_cast_fp16")]; + tensor var_50926 = const()[name = tensor("op_50926"), val = tensor([1, 1])]; + tensor var_50928 = const()[name = tensor("op_50928"), val = tensor([1, 1])]; + tensor input_621_pad_type_0 = const()[name = tensor("input_621_pad_type_0"), val = tensor("custom")]; + tensor input_621_pad_0 = const()[name = tensor("input_621_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_q_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_q_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343760896)))]; + tensor input_621_cast_fp16 = conv(dilations = var_50928, groups = var_50884, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = var_50926, weight = layers_31_self_attn_q_proj_loraA_weight_to_fp16, x = obj_125_cast_fp16)[name = tensor("input_621_cast_fp16")]; + tensor var_50932 = const()[name = tensor("op_50932"), val = tensor([1, 1])]; + tensor var_50934 = const()[name = tensor("op_50934"), val = tensor([1, 1])]; + tensor lora_out_745_pad_type_0 = const()[name = tensor("lora_out_745_pad_type_0"), val = tensor("custom")]; + tensor lora_out_745_pad_0 = const()[name = tensor("lora_out_745_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_747_weight_0_to_fp16 = const()[name = tensor("lora_out_747_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343801920)))]; + tensor lora_out_747_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_50934, groups = var_50884, pad = lora_out_745_pad_0, pad_type = lora_out_745_pad_type_0, strides = var_50932, weight = lora_out_747_weight_0_to_fp16, x = input_621_cast_fp16)[name = tensor("lora_out_747_cast_fp16")]; + tensor query_cast_fp16 = add(x = pretrained_out_373_cast_fp16, y = lora_out_747_cast_fp16)[name = tensor("query_cast_fp16")]; + tensor var_50944 = const()[name = tensor("op_50944"), val = tensor([1, 1])]; + tensor var_50946 = const()[name = tensor("op_50946"), val = tensor([1, 1])]; + tensor pretrained_out_375_pad_type_0 = const()[name = tensor("pretrained_out_375_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_375_pad_0 = const()[name = tensor("pretrained_out_375_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_k_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343842944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(344662208))), name = tensor("layers_31_self_attn_k_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor pretrained_out_375_cast_fp16 = conv(dilations = var_50946, groups = var_50884, pad = pretrained_out_375_pad_0, pad_type = pretrained_out_375_pad_type_0, strides = var_50944, weight = layers_31_self_attn_k_proj_pretrained_weight_to_fp16_palettized, x = obj_125_cast_fp16)[name = tensor("pretrained_out_375_cast_fp16")]; + tensor var_50950 = const()[name = tensor("op_50950"), val = tensor([1, 1])]; + tensor var_50952 = const()[name = tensor("op_50952"), val = tensor([1, 1])]; + tensor input_623_pad_type_0 = const()[name = tensor("input_623_pad_type_0"), val = tensor("custom")]; + tensor input_623_pad_0 = const()[name = tensor("input_623_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_k_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_k_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(344662336)))]; + tensor input_623_cast_fp16 = conv(dilations = var_50952, groups = var_50884, pad = input_623_pad_0, pad_type = input_623_pad_type_0, strides = var_50950, weight = layers_31_self_attn_k_proj_loraA_weight_to_fp16, x = obj_125_cast_fp16)[name = tensor("input_623_cast_fp16")]; + tensor var_50956 = const()[name = tensor("op_50956"), val = tensor([1, 1])]; + tensor var_50958 = const()[name = tensor("op_50958"), val = tensor([1, 1])]; + tensor lora_out_749_pad_type_0 = const()[name = tensor("lora_out_749_pad_type_0"), val = tensor("custom")]; + tensor lora_out_749_pad_0 = const()[name = tensor("lora_out_749_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_751_weight_0_to_fp16 = const()[name = tensor("lora_out_751_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(344703360)))]; + tensor lora_out_751_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_50958, groups = var_50884, pad = lora_out_749_pad_0, pad_type = lora_out_749_pad_type_0, strides = var_50956, weight = lora_out_751_weight_0_to_fp16, x = input_623_cast_fp16)[name = tensor("lora_out_751_cast_fp16")]; + tensor key_cast_fp16 = add(x = pretrained_out_375_cast_fp16, y = lora_out_751_cast_fp16)[name = tensor("key_cast_fp16")]; + tensor var_50969 = const()[name = tensor("op_50969"), val = tensor([1, 1])]; + tensor var_50971 = const()[name = tensor("op_50971"), val = tensor([1, 1])]; + tensor pretrained_out_377_pad_type_0 = const()[name = tensor("pretrained_out_377_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_377_pad_0 = const()[name = tensor("pretrained_out_377_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_v_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(344744384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345563648))), name = tensor("layers_31_self_attn_v_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_31_self_attn_v_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_v_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345563776)))]; + tensor pretrained_out_377_cast_fp16 = conv(bias = layers_31_self_attn_v_proj_pretrained_bias_to_fp16, dilations = var_50971, groups = var_50884, pad = pretrained_out_377_pad_0, pad_type = pretrained_out_377_pad_type_0, strides = var_50969, weight = layers_31_self_attn_v_proj_pretrained_weight_to_fp16_palettized, x = obj_125_cast_fp16)[name = tensor("pretrained_out_377_cast_fp16")]; + tensor var_50975 = const()[name = tensor("op_50975"), val = tensor([1, 1])]; + tensor var_50977 = const()[name = tensor("op_50977"), val = tensor([1, 1])]; + tensor input_625_pad_type_0 = const()[name = tensor("input_625_pad_type_0"), val = tensor("custom")]; + tensor input_625_pad_0 = const()[name = tensor("input_625_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_v_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_v_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345566400)))]; + tensor input_625_cast_fp16 = conv(dilations = var_50977, groups = var_50884, pad = input_625_pad_0, pad_type = input_625_pad_type_0, strides = var_50975, weight = layers_31_self_attn_v_proj_loraA_weight_to_fp16, x = obj_125_cast_fp16)[name = tensor("input_625_cast_fp16")]; + tensor var_50981 = const()[name = tensor("op_50981"), val = tensor([1, 1])]; + tensor var_50983 = const()[name = tensor("op_50983"), val = tensor([1, 1])]; + tensor lora_out_753_pad_type_0 = const()[name = tensor("lora_out_753_pad_type_0"), val = tensor("custom")]; + tensor lora_out_753_pad_0 = const()[name = tensor("lora_out_753_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_755_weight_0_to_fp16 = const()[name = tensor("lora_out_755_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345607424)))]; + tensor lora_out_755_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_50983, groups = var_50884, pad = lora_out_753_pad_0, pad_type = lora_out_753_pad_type_0, strides = var_50981, weight = lora_out_755_weight_0_to_fp16, x = input_625_cast_fp16)[name = tensor("lora_out_755_cast_fp16")]; + tensor value_cast_fp16 = add(x = pretrained_out_377_cast_fp16, y = lora_out_755_cast_fp16)[name = tensor("value_cast_fp16")]; + tensor var_50993_begin_0 = const()[name = tensor("op_50993_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_50993_end_0 = const()[name = tensor("op_50993_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_50993_end_mask_0 = const()[name = tensor("op_50993_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50993_cast_fp16 = slice_by_index(begin = var_50993_begin_0, end = var_50993_end_0, end_mask = var_50993_end_mask_0, x = query_cast_fp16)[name = tensor("op_50993_cast_fp16")]; + tensor var_50997_begin_0 = const()[name = tensor("op_50997_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_50997_end_0 = const()[name = tensor("op_50997_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_50997_end_mask_0 = const()[name = tensor("op_50997_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_50997_cast_fp16 = slice_by_index(begin = var_50997_begin_0, end = var_50997_end_0, end_mask = var_50997_end_mask_0, x = query_cast_fp16)[name = tensor("op_50997_cast_fp16")]; + tensor var_51001_begin_0 = const()[name = tensor("op_51001_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51001_end_0 = const()[name = tensor("op_51001_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_51001_end_mask_0 = const()[name = tensor("op_51001_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51001_cast_fp16 = slice_by_index(begin = var_51001_begin_0, end = var_51001_end_0, end_mask = var_51001_end_mask_0, x = query_cast_fp16)[name = tensor("op_51001_cast_fp16")]; + tensor var_51005_begin_0 = const()[name = tensor("op_51005_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51005_end_0 = const()[name = tensor("op_51005_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_51005_end_mask_0 = const()[name = tensor("op_51005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51005_cast_fp16 = slice_by_index(begin = var_51005_begin_0, end = var_51005_end_0, end_mask = var_51005_end_mask_0, x = query_cast_fp16)[name = tensor("op_51005_cast_fp16")]; + tensor var_51009_begin_0 = const()[name = tensor("op_51009_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51009_end_0 = const()[name = tensor("op_51009_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_51009_end_mask_0 = const()[name = tensor("op_51009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51009_cast_fp16 = slice_by_index(begin = var_51009_begin_0, end = var_51009_end_0, end_mask = var_51009_end_mask_0, x = query_cast_fp16)[name = tensor("op_51009_cast_fp16")]; + tensor var_51013_begin_0 = const()[name = tensor("op_51013_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51013_end_0 = const()[name = tensor("op_51013_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_51013_end_mask_0 = const()[name = tensor("op_51013_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51013_cast_fp16 = slice_by_index(begin = var_51013_begin_0, end = var_51013_end_0, end_mask = var_51013_end_mask_0, x = query_cast_fp16)[name = tensor("op_51013_cast_fp16")]; + tensor var_51017_begin_0 = const()[name = tensor("op_51017_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51017_end_0 = const()[name = tensor("op_51017_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_51017_end_mask_0 = const()[name = tensor("op_51017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51017_cast_fp16 = slice_by_index(begin = var_51017_begin_0, end = var_51017_end_0, end_mask = var_51017_end_mask_0, x = query_cast_fp16)[name = tensor("op_51017_cast_fp16")]; + tensor var_51021_begin_0 = const()[name = tensor("op_51021_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51021_end_0 = const()[name = tensor("op_51021_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_51021_end_mask_0 = const()[name = tensor("op_51021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51021_cast_fp16 = slice_by_index(begin = var_51021_begin_0, end = var_51021_end_0, end_mask = var_51021_end_mask_0, x = query_cast_fp16)[name = tensor("op_51021_cast_fp16")]; + tensor var_51025_begin_0 = const()[name = tensor("op_51025_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51025_end_0 = const()[name = tensor("op_51025_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_51025_end_mask_0 = const()[name = tensor("op_51025_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51025_cast_fp16 = slice_by_index(begin = var_51025_begin_0, end = var_51025_end_0, end_mask = var_51025_end_mask_0, x = query_cast_fp16)[name = tensor("op_51025_cast_fp16")]; + tensor var_51029_begin_0 = const()[name = tensor("op_51029_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51029_end_0 = const()[name = tensor("op_51029_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_51029_end_mask_0 = const()[name = tensor("op_51029_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51029_cast_fp16 = slice_by_index(begin = var_51029_begin_0, end = var_51029_end_0, end_mask = var_51029_end_mask_0, x = query_cast_fp16)[name = tensor("op_51029_cast_fp16")]; + tensor var_51033_begin_0 = const()[name = tensor("op_51033_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51033_end_0 = const()[name = tensor("op_51033_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_51033_end_mask_0 = const()[name = tensor("op_51033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51033_cast_fp16 = slice_by_index(begin = var_51033_begin_0, end = var_51033_end_0, end_mask = var_51033_end_mask_0, x = query_cast_fp16)[name = tensor("op_51033_cast_fp16")]; + tensor var_51037_begin_0 = const()[name = tensor("op_51037_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51037_end_0 = const()[name = tensor("op_51037_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_51037_end_mask_0 = const()[name = tensor("op_51037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51037_cast_fp16 = slice_by_index(begin = var_51037_begin_0, end = var_51037_end_0, end_mask = var_51037_end_mask_0, x = query_cast_fp16)[name = tensor("op_51037_cast_fp16")]; + tensor var_51041_begin_0 = const()[name = tensor("op_51041_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51041_end_0 = const()[name = tensor("op_51041_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_51041_end_mask_0 = const()[name = tensor("op_51041_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51041_cast_fp16 = slice_by_index(begin = var_51041_begin_0, end = var_51041_end_0, end_mask = var_51041_end_mask_0, x = query_cast_fp16)[name = tensor("op_51041_cast_fp16")]; + tensor var_51045_begin_0 = const()[name = tensor("op_51045_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51045_end_0 = const()[name = tensor("op_51045_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_51045_end_mask_0 = const()[name = tensor("op_51045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51045_cast_fp16 = slice_by_index(begin = var_51045_begin_0, end = var_51045_end_0, end_mask = var_51045_end_mask_0, x = query_cast_fp16)[name = tensor("op_51045_cast_fp16")]; + tensor var_51049_begin_0 = const()[name = tensor("op_51049_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51049_end_0 = const()[name = tensor("op_51049_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_51049_end_mask_0 = const()[name = tensor("op_51049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51049_cast_fp16 = slice_by_index(begin = var_51049_begin_0, end = var_51049_end_0, end_mask = var_51049_end_mask_0, x = query_cast_fp16)[name = tensor("op_51049_cast_fp16")]; + tensor var_51053_begin_0 = const()[name = tensor("op_51053_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51053_end_0 = const()[name = tensor("op_51053_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_51053_end_mask_0 = const()[name = tensor("op_51053_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51053_cast_fp16 = slice_by_index(begin = var_51053_begin_0, end = var_51053_end_0, end_mask = var_51053_end_mask_0, x = query_cast_fp16)[name = tensor("op_51053_cast_fp16")]; + tensor var_51057_begin_0 = const()[name = tensor("op_51057_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51057_end_0 = const()[name = tensor("op_51057_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_51057_end_mask_0 = const()[name = tensor("op_51057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51057_cast_fp16 = slice_by_index(begin = var_51057_begin_0, end = var_51057_end_0, end_mask = var_51057_end_mask_0, x = query_cast_fp16)[name = tensor("op_51057_cast_fp16")]; + tensor var_51061_begin_0 = const()[name = tensor("op_51061_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51061_end_0 = const()[name = tensor("op_51061_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_51061_end_mask_0 = const()[name = tensor("op_51061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51061_cast_fp16 = slice_by_index(begin = var_51061_begin_0, end = var_51061_end_0, end_mask = var_51061_end_mask_0, x = query_cast_fp16)[name = tensor("op_51061_cast_fp16")]; + tensor var_51065_begin_0 = const()[name = tensor("op_51065_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51065_end_0 = const()[name = tensor("op_51065_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_51065_end_mask_0 = const()[name = tensor("op_51065_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51065_cast_fp16 = slice_by_index(begin = var_51065_begin_0, end = var_51065_end_0, end_mask = var_51065_end_mask_0, x = query_cast_fp16)[name = tensor("op_51065_cast_fp16")]; + tensor var_51069_begin_0 = const()[name = tensor("op_51069_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51069_end_0 = const()[name = tensor("op_51069_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_51069_end_mask_0 = const()[name = tensor("op_51069_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51069_cast_fp16 = slice_by_index(begin = var_51069_begin_0, end = var_51069_end_0, end_mask = var_51069_end_mask_0, x = query_cast_fp16)[name = tensor("op_51069_cast_fp16")]; + tensor var_51078_begin_0 = const()[name = tensor("op_51078_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51078_end_0 = const()[name = tensor("op_51078_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51078_end_mask_0 = const()[name = tensor("op_51078_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51078_cast_fp16 = slice_by_index(begin = var_51078_begin_0, end = var_51078_end_0, end_mask = var_51078_end_mask_0, x = var_50993_cast_fp16)[name = tensor("op_51078_cast_fp16")]; + tensor var_51085_begin_0 = const()[name = tensor("op_51085_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51085_end_0 = const()[name = tensor("op_51085_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51085_end_mask_0 = const()[name = tensor("op_51085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51085_cast_fp16 = slice_by_index(begin = var_51085_begin_0, end = var_51085_end_0, end_mask = var_51085_end_mask_0, x = var_50993_cast_fp16)[name = tensor("op_51085_cast_fp16")]; + tensor var_51092_begin_0 = const()[name = tensor("op_51092_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51092_end_0 = const()[name = tensor("op_51092_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51092_end_mask_0 = const()[name = tensor("op_51092_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51092_cast_fp16 = slice_by_index(begin = var_51092_begin_0, end = var_51092_end_0, end_mask = var_51092_end_mask_0, x = var_50993_cast_fp16)[name = tensor("op_51092_cast_fp16")]; + tensor var_51099_begin_0 = const()[name = tensor("op_51099_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51099_end_0 = const()[name = tensor("op_51099_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51099_end_mask_0 = const()[name = tensor("op_51099_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51099_cast_fp16 = slice_by_index(begin = var_51099_begin_0, end = var_51099_end_0, end_mask = var_51099_end_mask_0, x = var_50993_cast_fp16)[name = tensor("op_51099_cast_fp16")]; + tensor var_51106_begin_0 = const()[name = tensor("op_51106_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51106_end_0 = const()[name = tensor("op_51106_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51106_end_mask_0 = const()[name = tensor("op_51106_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51106_cast_fp16 = slice_by_index(begin = var_51106_begin_0, end = var_51106_end_0, end_mask = var_51106_end_mask_0, x = var_50997_cast_fp16)[name = tensor("op_51106_cast_fp16")]; + tensor var_51113_begin_0 = const()[name = tensor("op_51113_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51113_end_0 = const()[name = tensor("op_51113_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51113_end_mask_0 = const()[name = tensor("op_51113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51113_cast_fp16 = slice_by_index(begin = var_51113_begin_0, end = var_51113_end_0, end_mask = var_51113_end_mask_0, x = var_50997_cast_fp16)[name = tensor("op_51113_cast_fp16")]; + tensor var_51120_begin_0 = const()[name = tensor("op_51120_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51120_end_0 = const()[name = tensor("op_51120_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51120_end_mask_0 = const()[name = tensor("op_51120_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51120_cast_fp16 = slice_by_index(begin = var_51120_begin_0, end = var_51120_end_0, end_mask = var_51120_end_mask_0, x = var_50997_cast_fp16)[name = tensor("op_51120_cast_fp16")]; + tensor var_51127_begin_0 = const()[name = tensor("op_51127_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51127_end_0 = const()[name = tensor("op_51127_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51127_end_mask_0 = const()[name = tensor("op_51127_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51127_cast_fp16 = slice_by_index(begin = var_51127_begin_0, end = var_51127_end_0, end_mask = var_51127_end_mask_0, x = var_50997_cast_fp16)[name = tensor("op_51127_cast_fp16")]; + tensor var_51134_begin_0 = const()[name = tensor("op_51134_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51134_end_0 = const()[name = tensor("op_51134_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51134_end_mask_0 = const()[name = tensor("op_51134_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51134_cast_fp16 = slice_by_index(begin = var_51134_begin_0, end = var_51134_end_0, end_mask = var_51134_end_mask_0, x = var_51001_cast_fp16)[name = tensor("op_51134_cast_fp16")]; + tensor var_51141_begin_0 = const()[name = tensor("op_51141_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51141_end_0 = const()[name = tensor("op_51141_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51141_end_mask_0 = const()[name = tensor("op_51141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51141_cast_fp16 = slice_by_index(begin = var_51141_begin_0, end = var_51141_end_0, end_mask = var_51141_end_mask_0, x = var_51001_cast_fp16)[name = tensor("op_51141_cast_fp16")]; + tensor var_51148_begin_0 = const()[name = tensor("op_51148_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51148_end_0 = const()[name = tensor("op_51148_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51148_end_mask_0 = const()[name = tensor("op_51148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51148_cast_fp16 = slice_by_index(begin = var_51148_begin_0, end = var_51148_end_0, end_mask = var_51148_end_mask_0, x = var_51001_cast_fp16)[name = tensor("op_51148_cast_fp16")]; + tensor var_51155_begin_0 = const()[name = tensor("op_51155_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51155_end_0 = const()[name = tensor("op_51155_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51155_end_mask_0 = const()[name = tensor("op_51155_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51155_cast_fp16 = slice_by_index(begin = var_51155_begin_0, end = var_51155_end_0, end_mask = var_51155_end_mask_0, x = var_51001_cast_fp16)[name = tensor("op_51155_cast_fp16")]; + tensor var_51162_begin_0 = const()[name = tensor("op_51162_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51162_end_0 = const()[name = tensor("op_51162_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51162_end_mask_0 = const()[name = tensor("op_51162_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51162_cast_fp16 = slice_by_index(begin = var_51162_begin_0, end = var_51162_end_0, end_mask = var_51162_end_mask_0, x = var_51005_cast_fp16)[name = tensor("op_51162_cast_fp16")]; + tensor var_51169_begin_0 = const()[name = tensor("op_51169_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51169_end_0 = const()[name = tensor("op_51169_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51169_end_mask_0 = const()[name = tensor("op_51169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51169_cast_fp16 = slice_by_index(begin = var_51169_begin_0, end = var_51169_end_0, end_mask = var_51169_end_mask_0, x = var_51005_cast_fp16)[name = tensor("op_51169_cast_fp16")]; + tensor var_51176_begin_0 = const()[name = tensor("op_51176_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51176_end_0 = const()[name = tensor("op_51176_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51176_end_mask_0 = const()[name = tensor("op_51176_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51176_cast_fp16 = slice_by_index(begin = var_51176_begin_0, end = var_51176_end_0, end_mask = var_51176_end_mask_0, x = var_51005_cast_fp16)[name = tensor("op_51176_cast_fp16")]; + tensor var_51183_begin_0 = const()[name = tensor("op_51183_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51183_end_0 = const()[name = tensor("op_51183_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51183_end_mask_0 = const()[name = tensor("op_51183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51183_cast_fp16 = slice_by_index(begin = var_51183_begin_0, end = var_51183_end_0, end_mask = var_51183_end_mask_0, x = var_51005_cast_fp16)[name = tensor("op_51183_cast_fp16")]; + tensor var_51190_begin_0 = const()[name = tensor("op_51190_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51190_end_0 = const()[name = tensor("op_51190_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51190_end_mask_0 = const()[name = tensor("op_51190_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51190_cast_fp16 = slice_by_index(begin = var_51190_begin_0, end = var_51190_end_0, end_mask = var_51190_end_mask_0, x = var_51009_cast_fp16)[name = tensor("op_51190_cast_fp16")]; + tensor var_51197_begin_0 = const()[name = tensor("op_51197_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51197_end_0 = const()[name = tensor("op_51197_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51197_end_mask_0 = const()[name = tensor("op_51197_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51197_cast_fp16 = slice_by_index(begin = var_51197_begin_0, end = var_51197_end_0, end_mask = var_51197_end_mask_0, x = var_51009_cast_fp16)[name = tensor("op_51197_cast_fp16")]; + tensor var_51204_begin_0 = const()[name = tensor("op_51204_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51204_end_0 = const()[name = tensor("op_51204_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51204_end_mask_0 = const()[name = tensor("op_51204_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51204_cast_fp16 = slice_by_index(begin = var_51204_begin_0, end = var_51204_end_0, end_mask = var_51204_end_mask_0, x = var_51009_cast_fp16)[name = tensor("op_51204_cast_fp16")]; + tensor var_51211_begin_0 = const()[name = tensor("op_51211_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51211_end_0 = const()[name = tensor("op_51211_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51211_end_mask_0 = const()[name = tensor("op_51211_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51211_cast_fp16 = slice_by_index(begin = var_51211_begin_0, end = var_51211_end_0, end_mask = var_51211_end_mask_0, x = var_51009_cast_fp16)[name = tensor("op_51211_cast_fp16")]; + tensor var_51218_begin_0 = const()[name = tensor("op_51218_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51218_end_0 = const()[name = tensor("op_51218_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51218_end_mask_0 = const()[name = tensor("op_51218_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51218_cast_fp16 = slice_by_index(begin = var_51218_begin_0, end = var_51218_end_0, end_mask = var_51218_end_mask_0, x = var_51013_cast_fp16)[name = tensor("op_51218_cast_fp16")]; + tensor var_51225_begin_0 = const()[name = tensor("op_51225_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51225_end_0 = const()[name = tensor("op_51225_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51225_end_mask_0 = const()[name = tensor("op_51225_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51225_cast_fp16 = slice_by_index(begin = var_51225_begin_0, end = var_51225_end_0, end_mask = var_51225_end_mask_0, x = var_51013_cast_fp16)[name = tensor("op_51225_cast_fp16")]; + tensor var_51232_begin_0 = const()[name = tensor("op_51232_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51232_end_0 = const()[name = tensor("op_51232_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51232_end_mask_0 = const()[name = tensor("op_51232_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51232_cast_fp16 = slice_by_index(begin = var_51232_begin_0, end = var_51232_end_0, end_mask = var_51232_end_mask_0, x = var_51013_cast_fp16)[name = tensor("op_51232_cast_fp16")]; + tensor var_51239_begin_0 = const()[name = tensor("op_51239_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51239_end_0 = const()[name = tensor("op_51239_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51239_end_mask_0 = const()[name = tensor("op_51239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51239_cast_fp16 = slice_by_index(begin = var_51239_begin_0, end = var_51239_end_0, end_mask = var_51239_end_mask_0, x = var_51013_cast_fp16)[name = tensor("op_51239_cast_fp16")]; + tensor var_51246_begin_0 = const()[name = tensor("op_51246_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51246_end_0 = const()[name = tensor("op_51246_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51246_end_mask_0 = const()[name = tensor("op_51246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51246_cast_fp16 = slice_by_index(begin = var_51246_begin_0, end = var_51246_end_0, end_mask = var_51246_end_mask_0, x = var_51017_cast_fp16)[name = tensor("op_51246_cast_fp16")]; + tensor var_51253_begin_0 = const()[name = tensor("op_51253_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51253_end_0 = const()[name = tensor("op_51253_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51253_end_mask_0 = const()[name = tensor("op_51253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51253_cast_fp16 = slice_by_index(begin = var_51253_begin_0, end = var_51253_end_0, end_mask = var_51253_end_mask_0, x = var_51017_cast_fp16)[name = tensor("op_51253_cast_fp16")]; + tensor var_51260_begin_0 = const()[name = tensor("op_51260_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51260_end_0 = const()[name = tensor("op_51260_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51260_end_mask_0 = const()[name = tensor("op_51260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51260_cast_fp16 = slice_by_index(begin = var_51260_begin_0, end = var_51260_end_0, end_mask = var_51260_end_mask_0, x = var_51017_cast_fp16)[name = tensor("op_51260_cast_fp16")]; + tensor var_51267_begin_0 = const()[name = tensor("op_51267_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51267_end_0 = const()[name = tensor("op_51267_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51267_end_mask_0 = const()[name = tensor("op_51267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51267_cast_fp16 = slice_by_index(begin = var_51267_begin_0, end = var_51267_end_0, end_mask = var_51267_end_mask_0, x = var_51017_cast_fp16)[name = tensor("op_51267_cast_fp16")]; + tensor var_51274_begin_0 = const()[name = tensor("op_51274_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51274_end_0 = const()[name = tensor("op_51274_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51274_end_mask_0 = const()[name = tensor("op_51274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51274_cast_fp16 = slice_by_index(begin = var_51274_begin_0, end = var_51274_end_0, end_mask = var_51274_end_mask_0, x = var_51021_cast_fp16)[name = tensor("op_51274_cast_fp16")]; + tensor var_51281_begin_0 = const()[name = tensor("op_51281_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51281_end_0 = const()[name = tensor("op_51281_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51281_end_mask_0 = const()[name = tensor("op_51281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51281_cast_fp16 = slice_by_index(begin = var_51281_begin_0, end = var_51281_end_0, end_mask = var_51281_end_mask_0, x = var_51021_cast_fp16)[name = tensor("op_51281_cast_fp16")]; + tensor var_51288_begin_0 = const()[name = tensor("op_51288_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51288_end_0 = const()[name = tensor("op_51288_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51288_end_mask_0 = const()[name = tensor("op_51288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51288_cast_fp16 = slice_by_index(begin = var_51288_begin_0, end = var_51288_end_0, end_mask = var_51288_end_mask_0, x = var_51021_cast_fp16)[name = tensor("op_51288_cast_fp16")]; + tensor var_51295_begin_0 = const()[name = tensor("op_51295_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51295_end_0 = const()[name = tensor("op_51295_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51295_end_mask_0 = const()[name = tensor("op_51295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51295_cast_fp16 = slice_by_index(begin = var_51295_begin_0, end = var_51295_end_0, end_mask = var_51295_end_mask_0, x = var_51021_cast_fp16)[name = tensor("op_51295_cast_fp16")]; + tensor var_51302_begin_0 = const()[name = tensor("op_51302_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51302_end_0 = const()[name = tensor("op_51302_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51302_end_mask_0 = const()[name = tensor("op_51302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51302_cast_fp16 = slice_by_index(begin = var_51302_begin_0, end = var_51302_end_0, end_mask = var_51302_end_mask_0, x = var_51025_cast_fp16)[name = tensor("op_51302_cast_fp16")]; + tensor var_51309_begin_0 = const()[name = tensor("op_51309_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51309_end_0 = const()[name = tensor("op_51309_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51309_end_mask_0 = const()[name = tensor("op_51309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51309_cast_fp16 = slice_by_index(begin = var_51309_begin_0, end = var_51309_end_0, end_mask = var_51309_end_mask_0, x = var_51025_cast_fp16)[name = tensor("op_51309_cast_fp16")]; + tensor var_51316_begin_0 = const()[name = tensor("op_51316_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51316_end_0 = const()[name = tensor("op_51316_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51316_end_mask_0 = const()[name = tensor("op_51316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51316_cast_fp16 = slice_by_index(begin = var_51316_begin_0, end = var_51316_end_0, end_mask = var_51316_end_mask_0, x = var_51025_cast_fp16)[name = tensor("op_51316_cast_fp16")]; + tensor var_51323_begin_0 = const()[name = tensor("op_51323_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51323_end_0 = const()[name = tensor("op_51323_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51323_end_mask_0 = const()[name = tensor("op_51323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51323_cast_fp16 = slice_by_index(begin = var_51323_begin_0, end = var_51323_end_0, end_mask = var_51323_end_mask_0, x = var_51025_cast_fp16)[name = tensor("op_51323_cast_fp16")]; + tensor var_51330_begin_0 = const()[name = tensor("op_51330_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51330_end_0 = const()[name = tensor("op_51330_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51330_end_mask_0 = const()[name = tensor("op_51330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51330_cast_fp16 = slice_by_index(begin = var_51330_begin_0, end = var_51330_end_0, end_mask = var_51330_end_mask_0, x = var_51029_cast_fp16)[name = tensor("op_51330_cast_fp16")]; + tensor var_51337_begin_0 = const()[name = tensor("op_51337_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51337_end_0 = const()[name = tensor("op_51337_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51337_end_mask_0 = const()[name = tensor("op_51337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51337_cast_fp16 = slice_by_index(begin = var_51337_begin_0, end = var_51337_end_0, end_mask = var_51337_end_mask_0, x = var_51029_cast_fp16)[name = tensor("op_51337_cast_fp16")]; + tensor var_51344_begin_0 = const()[name = tensor("op_51344_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51344_end_0 = const()[name = tensor("op_51344_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51344_end_mask_0 = const()[name = tensor("op_51344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51344_cast_fp16 = slice_by_index(begin = var_51344_begin_0, end = var_51344_end_0, end_mask = var_51344_end_mask_0, x = var_51029_cast_fp16)[name = tensor("op_51344_cast_fp16")]; + tensor var_51351_begin_0 = const()[name = tensor("op_51351_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51351_end_0 = const()[name = tensor("op_51351_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51351_end_mask_0 = const()[name = tensor("op_51351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51351_cast_fp16 = slice_by_index(begin = var_51351_begin_0, end = var_51351_end_0, end_mask = var_51351_end_mask_0, x = var_51029_cast_fp16)[name = tensor("op_51351_cast_fp16")]; + tensor var_51358_begin_0 = const()[name = tensor("op_51358_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51358_end_0 = const()[name = tensor("op_51358_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51358_end_mask_0 = const()[name = tensor("op_51358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51358_cast_fp16 = slice_by_index(begin = var_51358_begin_0, end = var_51358_end_0, end_mask = var_51358_end_mask_0, x = var_51033_cast_fp16)[name = tensor("op_51358_cast_fp16")]; + tensor var_51365_begin_0 = const()[name = tensor("op_51365_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51365_end_0 = const()[name = tensor("op_51365_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51365_end_mask_0 = const()[name = tensor("op_51365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51365_cast_fp16 = slice_by_index(begin = var_51365_begin_0, end = var_51365_end_0, end_mask = var_51365_end_mask_0, x = var_51033_cast_fp16)[name = tensor("op_51365_cast_fp16")]; + tensor var_51372_begin_0 = const()[name = tensor("op_51372_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51372_end_0 = const()[name = tensor("op_51372_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51372_end_mask_0 = const()[name = tensor("op_51372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51372_cast_fp16 = slice_by_index(begin = var_51372_begin_0, end = var_51372_end_0, end_mask = var_51372_end_mask_0, x = var_51033_cast_fp16)[name = tensor("op_51372_cast_fp16")]; + tensor var_51379_begin_0 = const()[name = tensor("op_51379_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51379_end_0 = const()[name = tensor("op_51379_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51379_end_mask_0 = const()[name = tensor("op_51379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51379_cast_fp16 = slice_by_index(begin = var_51379_begin_0, end = var_51379_end_0, end_mask = var_51379_end_mask_0, x = var_51033_cast_fp16)[name = tensor("op_51379_cast_fp16")]; + tensor var_51386_begin_0 = const()[name = tensor("op_51386_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51386_end_0 = const()[name = tensor("op_51386_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51386_end_mask_0 = const()[name = tensor("op_51386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51386_cast_fp16 = slice_by_index(begin = var_51386_begin_0, end = var_51386_end_0, end_mask = var_51386_end_mask_0, x = var_51037_cast_fp16)[name = tensor("op_51386_cast_fp16")]; + tensor var_51393_begin_0 = const()[name = tensor("op_51393_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51393_end_0 = const()[name = tensor("op_51393_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51393_end_mask_0 = const()[name = tensor("op_51393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51393_cast_fp16 = slice_by_index(begin = var_51393_begin_0, end = var_51393_end_0, end_mask = var_51393_end_mask_0, x = var_51037_cast_fp16)[name = tensor("op_51393_cast_fp16")]; + tensor var_51400_begin_0 = const()[name = tensor("op_51400_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51400_end_0 = const()[name = tensor("op_51400_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51400_end_mask_0 = const()[name = tensor("op_51400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51400_cast_fp16 = slice_by_index(begin = var_51400_begin_0, end = var_51400_end_0, end_mask = var_51400_end_mask_0, x = var_51037_cast_fp16)[name = tensor("op_51400_cast_fp16")]; + tensor var_51407_begin_0 = const()[name = tensor("op_51407_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51407_end_0 = const()[name = tensor("op_51407_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51407_end_mask_0 = const()[name = tensor("op_51407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51407_cast_fp16 = slice_by_index(begin = var_51407_begin_0, end = var_51407_end_0, end_mask = var_51407_end_mask_0, x = var_51037_cast_fp16)[name = tensor("op_51407_cast_fp16")]; + tensor var_51414_begin_0 = const()[name = tensor("op_51414_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51414_end_0 = const()[name = tensor("op_51414_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51414_end_mask_0 = const()[name = tensor("op_51414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51414_cast_fp16 = slice_by_index(begin = var_51414_begin_0, end = var_51414_end_0, end_mask = var_51414_end_mask_0, x = var_51041_cast_fp16)[name = tensor("op_51414_cast_fp16")]; + tensor var_51421_begin_0 = const()[name = tensor("op_51421_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51421_end_0 = const()[name = tensor("op_51421_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51421_end_mask_0 = const()[name = tensor("op_51421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51421_cast_fp16 = slice_by_index(begin = var_51421_begin_0, end = var_51421_end_0, end_mask = var_51421_end_mask_0, x = var_51041_cast_fp16)[name = tensor("op_51421_cast_fp16")]; + tensor var_51428_begin_0 = const()[name = tensor("op_51428_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51428_end_0 = const()[name = tensor("op_51428_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51428_end_mask_0 = const()[name = tensor("op_51428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51428_cast_fp16 = slice_by_index(begin = var_51428_begin_0, end = var_51428_end_0, end_mask = var_51428_end_mask_0, x = var_51041_cast_fp16)[name = tensor("op_51428_cast_fp16")]; + tensor var_51435_begin_0 = const()[name = tensor("op_51435_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51435_end_0 = const()[name = tensor("op_51435_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51435_end_mask_0 = const()[name = tensor("op_51435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51435_cast_fp16 = slice_by_index(begin = var_51435_begin_0, end = var_51435_end_0, end_mask = var_51435_end_mask_0, x = var_51041_cast_fp16)[name = tensor("op_51435_cast_fp16")]; + tensor var_51442_begin_0 = const()[name = tensor("op_51442_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51442_end_0 = const()[name = tensor("op_51442_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51442_end_mask_0 = const()[name = tensor("op_51442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51442_cast_fp16 = slice_by_index(begin = var_51442_begin_0, end = var_51442_end_0, end_mask = var_51442_end_mask_0, x = var_51045_cast_fp16)[name = tensor("op_51442_cast_fp16")]; + tensor var_51449_begin_0 = const()[name = tensor("op_51449_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51449_end_0 = const()[name = tensor("op_51449_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51449_end_mask_0 = const()[name = tensor("op_51449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51449_cast_fp16 = slice_by_index(begin = var_51449_begin_0, end = var_51449_end_0, end_mask = var_51449_end_mask_0, x = var_51045_cast_fp16)[name = tensor("op_51449_cast_fp16")]; + tensor var_51456_begin_0 = const()[name = tensor("op_51456_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51456_end_0 = const()[name = tensor("op_51456_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51456_end_mask_0 = const()[name = tensor("op_51456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51456_cast_fp16 = slice_by_index(begin = var_51456_begin_0, end = var_51456_end_0, end_mask = var_51456_end_mask_0, x = var_51045_cast_fp16)[name = tensor("op_51456_cast_fp16")]; + tensor var_51463_begin_0 = const()[name = tensor("op_51463_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51463_end_0 = const()[name = tensor("op_51463_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51463_end_mask_0 = const()[name = tensor("op_51463_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51463_cast_fp16 = slice_by_index(begin = var_51463_begin_0, end = var_51463_end_0, end_mask = var_51463_end_mask_0, x = var_51045_cast_fp16)[name = tensor("op_51463_cast_fp16")]; + tensor var_51470_begin_0 = const()[name = tensor("op_51470_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51470_end_0 = const()[name = tensor("op_51470_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51470_end_mask_0 = const()[name = tensor("op_51470_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51470_cast_fp16 = slice_by_index(begin = var_51470_begin_0, end = var_51470_end_0, end_mask = var_51470_end_mask_0, x = var_51049_cast_fp16)[name = tensor("op_51470_cast_fp16")]; + tensor var_51477_begin_0 = const()[name = tensor("op_51477_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51477_end_0 = const()[name = tensor("op_51477_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51477_end_mask_0 = const()[name = tensor("op_51477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51477_cast_fp16 = slice_by_index(begin = var_51477_begin_0, end = var_51477_end_0, end_mask = var_51477_end_mask_0, x = var_51049_cast_fp16)[name = tensor("op_51477_cast_fp16")]; + tensor var_51484_begin_0 = const()[name = tensor("op_51484_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51484_end_0 = const()[name = tensor("op_51484_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51484_end_mask_0 = const()[name = tensor("op_51484_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51484_cast_fp16 = slice_by_index(begin = var_51484_begin_0, end = var_51484_end_0, end_mask = var_51484_end_mask_0, x = var_51049_cast_fp16)[name = tensor("op_51484_cast_fp16")]; + tensor var_51491_begin_0 = const()[name = tensor("op_51491_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51491_end_0 = const()[name = tensor("op_51491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51491_end_mask_0 = const()[name = tensor("op_51491_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51491_cast_fp16 = slice_by_index(begin = var_51491_begin_0, end = var_51491_end_0, end_mask = var_51491_end_mask_0, x = var_51049_cast_fp16)[name = tensor("op_51491_cast_fp16")]; + tensor var_51498_begin_0 = const()[name = tensor("op_51498_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51498_end_0 = const()[name = tensor("op_51498_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51498_end_mask_0 = const()[name = tensor("op_51498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51498_cast_fp16 = slice_by_index(begin = var_51498_begin_0, end = var_51498_end_0, end_mask = var_51498_end_mask_0, x = var_51053_cast_fp16)[name = tensor("op_51498_cast_fp16")]; + tensor var_51505_begin_0 = const()[name = tensor("op_51505_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51505_end_0 = const()[name = tensor("op_51505_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51505_end_mask_0 = const()[name = tensor("op_51505_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51505_cast_fp16 = slice_by_index(begin = var_51505_begin_0, end = var_51505_end_0, end_mask = var_51505_end_mask_0, x = var_51053_cast_fp16)[name = tensor("op_51505_cast_fp16")]; + tensor var_51512_begin_0 = const()[name = tensor("op_51512_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51512_end_0 = const()[name = tensor("op_51512_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51512_end_mask_0 = const()[name = tensor("op_51512_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51512_cast_fp16 = slice_by_index(begin = var_51512_begin_0, end = var_51512_end_0, end_mask = var_51512_end_mask_0, x = var_51053_cast_fp16)[name = tensor("op_51512_cast_fp16")]; + tensor var_51519_begin_0 = const()[name = tensor("op_51519_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51519_end_0 = const()[name = tensor("op_51519_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51519_end_mask_0 = const()[name = tensor("op_51519_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51519_cast_fp16 = slice_by_index(begin = var_51519_begin_0, end = var_51519_end_0, end_mask = var_51519_end_mask_0, x = var_51053_cast_fp16)[name = tensor("op_51519_cast_fp16")]; + tensor var_51526_begin_0 = const()[name = tensor("op_51526_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51526_end_0 = const()[name = tensor("op_51526_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51526_end_mask_0 = const()[name = tensor("op_51526_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51526_cast_fp16 = slice_by_index(begin = var_51526_begin_0, end = var_51526_end_0, end_mask = var_51526_end_mask_0, x = var_51057_cast_fp16)[name = tensor("op_51526_cast_fp16")]; + tensor var_51533_begin_0 = const()[name = tensor("op_51533_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51533_end_0 = const()[name = tensor("op_51533_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51533_end_mask_0 = const()[name = tensor("op_51533_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51533_cast_fp16 = slice_by_index(begin = var_51533_begin_0, end = var_51533_end_0, end_mask = var_51533_end_mask_0, x = var_51057_cast_fp16)[name = tensor("op_51533_cast_fp16")]; + tensor var_51540_begin_0 = const()[name = tensor("op_51540_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51540_end_0 = const()[name = tensor("op_51540_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51540_end_mask_0 = const()[name = tensor("op_51540_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51540_cast_fp16 = slice_by_index(begin = var_51540_begin_0, end = var_51540_end_0, end_mask = var_51540_end_mask_0, x = var_51057_cast_fp16)[name = tensor("op_51540_cast_fp16")]; + tensor var_51547_begin_0 = const()[name = tensor("op_51547_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51547_end_0 = const()[name = tensor("op_51547_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51547_end_mask_0 = const()[name = tensor("op_51547_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51547_cast_fp16 = slice_by_index(begin = var_51547_begin_0, end = var_51547_end_0, end_mask = var_51547_end_mask_0, x = var_51057_cast_fp16)[name = tensor("op_51547_cast_fp16")]; + tensor var_51554_begin_0 = const()[name = tensor("op_51554_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51554_end_0 = const()[name = tensor("op_51554_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51554_end_mask_0 = const()[name = tensor("op_51554_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51554_cast_fp16 = slice_by_index(begin = var_51554_begin_0, end = var_51554_end_0, end_mask = var_51554_end_mask_0, x = var_51061_cast_fp16)[name = tensor("op_51554_cast_fp16")]; + tensor var_51561_begin_0 = const()[name = tensor("op_51561_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51561_end_0 = const()[name = tensor("op_51561_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51561_end_mask_0 = const()[name = tensor("op_51561_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51561_cast_fp16 = slice_by_index(begin = var_51561_begin_0, end = var_51561_end_0, end_mask = var_51561_end_mask_0, x = var_51061_cast_fp16)[name = tensor("op_51561_cast_fp16")]; + tensor var_51568_begin_0 = const()[name = tensor("op_51568_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51568_end_0 = const()[name = tensor("op_51568_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51568_end_mask_0 = const()[name = tensor("op_51568_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51568_cast_fp16 = slice_by_index(begin = var_51568_begin_0, end = var_51568_end_0, end_mask = var_51568_end_mask_0, x = var_51061_cast_fp16)[name = tensor("op_51568_cast_fp16")]; + tensor var_51575_begin_0 = const()[name = tensor("op_51575_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51575_end_0 = const()[name = tensor("op_51575_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51575_end_mask_0 = const()[name = tensor("op_51575_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51575_cast_fp16 = slice_by_index(begin = var_51575_begin_0, end = var_51575_end_0, end_mask = var_51575_end_mask_0, x = var_51061_cast_fp16)[name = tensor("op_51575_cast_fp16")]; + tensor var_51582_begin_0 = const()[name = tensor("op_51582_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51582_end_0 = const()[name = tensor("op_51582_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51582_end_mask_0 = const()[name = tensor("op_51582_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51582_cast_fp16 = slice_by_index(begin = var_51582_begin_0, end = var_51582_end_0, end_mask = var_51582_end_mask_0, x = var_51065_cast_fp16)[name = tensor("op_51582_cast_fp16")]; + tensor var_51589_begin_0 = const()[name = tensor("op_51589_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51589_end_0 = const()[name = tensor("op_51589_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51589_end_mask_0 = const()[name = tensor("op_51589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51589_cast_fp16 = slice_by_index(begin = var_51589_begin_0, end = var_51589_end_0, end_mask = var_51589_end_mask_0, x = var_51065_cast_fp16)[name = tensor("op_51589_cast_fp16")]; + tensor var_51596_begin_0 = const()[name = tensor("op_51596_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51596_end_0 = const()[name = tensor("op_51596_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51596_end_mask_0 = const()[name = tensor("op_51596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51596_cast_fp16 = slice_by_index(begin = var_51596_begin_0, end = var_51596_end_0, end_mask = var_51596_end_mask_0, x = var_51065_cast_fp16)[name = tensor("op_51596_cast_fp16")]; + tensor var_51603_begin_0 = const()[name = tensor("op_51603_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51603_end_0 = const()[name = tensor("op_51603_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51603_end_mask_0 = const()[name = tensor("op_51603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51603_cast_fp16 = slice_by_index(begin = var_51603_begin_0, end = var_51603_end_0, end_mask = var_51603_end_mask_0, x = var_51065_cast_fp16)[name = tensor("op_51603_cast_fp16")]; + tensor var_51610_begin_0 = const()[name = tensor("op_51610_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51610_end_0 = const()[name = tensor("op_51610_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_51610_end_mask_0 = const()[name = tensor("op_51610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51610_cast_fp16 = slice_by_index(begin = var_51610_begin_0, end = var_51610_end_0, end_mask = var_51610_end_mask_0, x = var_51069_cast_fp16)[name = tensor("op_51610_cast_fp16")]; + tensor var_51617_begin_0 = const()[name = tensor("op_51617_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_51617_end_0 = const()[name = tensor("op_51617_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_51617_end_mask_0 = const()[name = tensor("op_51617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51617_cast_fp16 = slice_by_index(begin = var_51617_begin_0, end = var_51617_end_0, end_mask = var_51617_end_mask_0, x = var_51069_cast_fp16)[name = tensor("op_51617_cast_fp16")]; + tensor var_51624_begin_0 = const()[name = tensor("op_51624_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_51624_end_0 = const()[name = tensor("op_51624_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_51624_end_mask_0 = const()[name = tensor("op_51624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51624_cast_fp16 = slice_by_index(begin = var_51624_begin_0, end = var_51624_end_0, end_mask = var_51624_end_mask_0, x = var_51069_cast_fp16)[name = tensor("op_51624_cast_fp16")]; + tensor var_51631_begin_0 = const()[name = tensor("op_51631_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_51631_end_0 = const()[name = tensor("op_51631_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51631_end_mask_0 = const()[name = tensor("op_51631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51631_cast_fp16 = slice_by_index(begin = var_51631_begin_0, end = var_51631_end_0, end_mask = var_51631_end_mask_0, x = var_51069_cast_fp16)[name = tensor("op_51631_cast_fp16")]; + tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_51636_begin_0 = const()[name = tensor("op_51636_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51636_end_0 = const()[name = tensor("op_51636_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_51636_end_mask_0 = const()[name = tensor("op_51636_end_mask_0"), val = tensor([true, true, true, false])]; + tensor transpose_0 = transpose(perm = k_perm_0, x = key_cast_fp16)[name = tensor("transpose_0")]; + tensor var_51636_cast_fp16 = slice_by_index(begin = var_51636_begin_0, end = var_51636_end_0, end_mask = var_51636_end_mask_0, x = transpose_0)[name = tensor("op_51636_cast_fp16")]; + tensor var_51640_begin_0 = const()[name = tensor("op_51640_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_51640_end_0 = const()[name = tensor("op_51640_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_51640_end_mask_0 = const()[name = tensor("op_51640_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51640_cast_fp16 = slice_by_index(begin = var_51640_begin_0, end = var_51640_end_0, end_mask = var_51640_end_mask_0, x = transpose_0)[name = tensor("op_51640_cast_fp16")]; + tensor var_51644_begin_0 = const()[name = tensor("op_51644_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_51644_end_0 = const()[name = tensor("op_51644_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_51644_end_mask_0 = const()[name = tensor("op_51644_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51644_cast_fp16 = slice_by_index(begin = var_51644_begin_0, end = var_51644_end_0, end_mask = var_51644_end_mask_0, x = transpose_0)[name = tensor("op_51644_cast_fp16")]; + tensor var_51648_begin_0 = const()[name = tensor("op_51648_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_51648_end_0 = const()[name = tensor("op_51648_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_51648_end_mask_0 = const()[name = tensor("op_51648_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51648_cast_fp16 = slice_by_index(begin = var_51648_begin_0, end = var_51648_end_0, end_mask = var_51648_end_mask_0, x = transpose_0)[name = tensor("op_51648_cast_fp16")]; + tensor var_51652_begin_0 = const()[name = tensor("op_51652_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_51652_end_0 = const()[name = tensor("op_51652_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_51652_end_mask_0 = const()[name = tensor("op_51652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51652_cast_fp16 = slice_by_index(begin = var_51652_begin_0, end = var_51652_end_0, end_mask = var_51652_end_mask_0, x = transpose_0)[name = tensor("op_51652_cast_fp16")]; + tensor var_51656_begin_0 = const()[name = tensor("op_51656_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_51656_end_0 = const()[name = tensor("op_51656_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_51656_end_mask_0 = const()[name = tensor("op_51656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51656_cast_fp16 = slice_by_index(begin = var_51656_begin_0, end = var_51656_end_0, end_mask = var_51656_end_mask_0, x = transpose_0)[name = tensor("op_51656_cast_fp16")]; + tensor var_51660_begin_0 = const()[name = tensor("op_51660_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_51660_end_0 = const()[name = tensor("op_51660_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_51660_end_mask_0 = const()[name = tensor("op_51660_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51660_cast_fp16 = slice_by_index(begin = var_51660_begin_0, end = var_51660_end_0, end_mask = var_51660_end_mask_0, x = transpose_0)[name = tensor("op_51660_cast_fp16")]; + tensor var_51664_begin_0 = const()[name = tensor("op_51664_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_51664_end_0 = const()[name = tensor("op_51664_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_51664_end_mask_0 = const()[name = tensor("op_51664_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51664_cast_fp16 = slice_by_index(begin = var_51664_begin_0, end = var_51664_end_0, end_mask = var_51664_end_mask_0, x = transpose_0)[name = tensor("op_51664_cast_fp16")]; + tensor var_51668_begin_0 = const()[name = tensor("op_51668_begin_0"), val = tensor([0, 0, 0, 512])]; + tensor var_51668_end_0 = const()[name = tensor("op_51668_end_0"), val = tensor([1, 1500, 1, 576])]; + tensor var_51668_end_mask_0 = const()[name = tensor("op_51668_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51668_cast_fp16 = slice_by_index(begin = var_51668_begin_0, end = var_51668_end_0, end_mask = var_51668_end_mask_0, x = transpose_0)[name = tensor("op_51668_cast_fp16")]; + tensor var_51672_begin_0 = const()[name = tensor("op_51672_begin_0"), val = tensor([0, 0, 0, 576])]; + tensor var_51672_end_0 = const()[name = tensor("op_51672_end_0"), val = tensor([1, 1500, 1, 640])]; + tensor var_51672_end_mask_0 = const()[name = tensor("op_51672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51672_cast_fp16 = slice_by_index(begin = var_51672_begin_0, end = var_51672_end_0, end_mask = var_51672_end_mask_0, x = transpose_0)[name = tensor("op_51672_cast_fp16")]; + tensor var_51676_begin_0 = const()[name = tensor("op_51676_begin_0"), val = tensor([0, 0, 0, 640])]; + tensor var_51676_end_0 = const()[name = tensor("op_51676_end_0"), val = tensor([1, 1500, 1, 704])]; + tensor var_51676_end_mask_0 = const()[name = tensor("op_51676_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51676_cast_fp16 = slice_by_index(begin = var_51676_begin_0, end = var_51676_end_0, end_mask = var_51676_end_mask_0, x = transpose_0)[name = tensor("op_51676_cast_fp16")]; + tensor var_51680_begin_0 = const()[name = tensor("op_51680_begin_0"), val = tensor([0, 0, 0, 704])]; + tensor var_51680_end_0 = const()[name = tensor("op_51680_end_0"), val = tensor([1, 1500, 1, 768])]; + tensor var_51680_end_mask_0 = const()[name = tensor("op_51680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51680_cast_fp16 = slice_by_index(begin = var_51680_begin_0, end = var_51680_end_0, end_mask = var_51680_end_mask_0, x = transpose_0)[name = tensor("op_51680_cast_fp16")]; + tensor var_51684_begin_0 = const()[name = tensor("op_51684_begin_0"), val = tensor([0, 0, 0, 768])]; + tensor var_51684_end_0 = const()[name = tensor("op_51684_end_0"), val = tensor([1, 1500, 1, 832])]; + tensor var_51684_end_mask_0 = const()[name = tensor("op_51684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51684_cast_fp16 = slice_by_index(begin = var_51684_begin_0, end = var_51684_end_0, end_mask = var_51684_end_mask_0, x = transpose_0)[name = tensor("op_51684_cast_fp16")]; + tensor var_51688_begin_0 = const()[name = tensor("op_51688_begin_0"), val = tensor([0, 0, 0, 832])]; + tensor var_51688_end_0 = const()[name = tensor("op_51688_end_0"), val = tensor([1, 1500, 1, 896])]; + tensor var_51688_end_mask_0 = const()[name = tensor("op_51688_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51688_cast_fp16 = slice_by_index(begin = var_51688_begin_0, end = var_51688_end_0, end_mask = var_51688_end_mask_0, x = transpose_0)[name = tensor("op_51688_cast_fp16")]; + tensor var_51692_begin_0 = const()[name = tensor("op_51692_begin_0"), val = tensor([0, 0, 0, 896])]; + tensor var_51692_end_0 = const()[name = tensor("op_51692_end_0"), val = tensor([1, 1500, 1, 960])]; + tensor var_51692_end_mask_0 = const()[name = tensor("op_51692_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51692_cast_fp16 = slice_by_index(begin = var_51692_begin_0, end = var_51692_end_0, end_mask = var_51692_end_mask_0, x = transpose_0)[name = tensor("op_51692_cast_fp16")]; + tensor var_51696_begin_0 = const()[name = tensor("op_51696_begin_0"), val = tensor([0, 0, 0, 960])]; + tensor var_51696_end_0 = const()[name = tensor("op_51696_end_0"), val = tensor([1, 1500, 1, 1024])]; + tensor var_51696_end_mask_0 = const()[name = tensor("op_51696_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51696_cast_fp16 = slice_by_index(begin = var_51696_begin_0, end = var_51696_end_0, end_mask = var_51696_end_mask_0, x = transpose_0)[name = tensor("op_51696_cast_fp16")]; + tensor var_51700_begin_0 = const()[name = tensor("op_51700_begin_0"), val = tensor([0, 0, 0, 1024])]; + tensor var_51700_end_0 = const()[name = tensor("op_51700_end_0"), val = tensor([1, 1500, 1, 1088])]; + tensor var_51700_end_mask_0 = const()[name = tensor("op_51700_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51700_cast_fp16 = slice_by_index(begin = var_51700_begin_0, end = var_51700_end_0, end_mask = var_51700_end_mask_0, x = transpose_0)[name = tensor("op_51700_cast_fp16")]; + tensor var_51704_begin_0 = const()[name = tensor("op_51704_begin_0"), val = tensor([0, 0, 0, 1088])]; + tensor var_51704_end_0 = const()[name = tensor("op_51704_end_0"), val = tensor([1, 1500, 1, 1152])]; + tensor var_51704_end_mask_0 = const()[name = tensor("op_51704_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51704_cast_fp16 = slice_by_index(begin = var_51704_begin_0, end = var_51704_end_0, end_mask = var_51704_end_mask_0, x = transpose_0)[name = tensor("op_51704_cast_fp16")]; + tensor var_51708_begin_0 = const()[name = tensor("op_51708_begin_0"), val = tensor([0, 0, 0, 1152])]; + tensor var_51708_end_0 = const()[name = tensor("op_51708_end_0"), val = tensor([1, 1500, 1, 1216])]; + tensor var_51708_end_mask_0 = const()[name = tensor("op_51708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51708_cast_fp16 = slice_by_index(begin = var_51708_begin_0, end = var_51708_end_0, end_mask = var_51708_end_mask_0, x = transpose_0)[name = tensor("op_51708_cast_fp16")]; + tensor var_51712_begin_0 = const()[name = tensor("op_51712_begin_0"), val = tensor([0, 0, 0, 1216])]; + tensor var_51712_end_0 = const()[name = tensor("op_51712_end_0"), val = tensor([1, 1500, 1, 1280])]; + tensor var_51712_end_mask_0 = const()[name = tensor("op_51712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_51712_cast_fp16 = slice_by_index(begin = var_51712_begin_0, end = var_51712_end_0, end_mask = var_51712_end_mask_0, x = transpose_0)[name = tensor("op_51712_cast_fp16")]; + tensor var_51714_begin_0 = const()[name = tensor("op_51714_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_51714_end_0 = const()[name = tensor("op_51714_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_51714_end_mask_0 = const()[name = tensor("op_51714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51714_cast_fp16 = slice_by_index(begin = var_51714_begin_0, end = var_51714_end_0, end_mask = var_51714_end_mask_0, x = value_cast_fp16)[name = tensor("op_51714_cast_fp16")]; + tensor var_51718_begin_0 = const()[name = tensor("op_51718_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_51718_end_0 = const()[name = tensor("op_51718_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_51718_end_mask_0 = const()[name = tensor("op_51718_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51718_cast_fp16 = slice_by_index(begin = var_51718_begin_0, end = var_51718_end_0, end_mask = var_51718_end_mask_0, x = value_cast_fp16)[name = tensor("op_51718_cast_fp16")]; + tensor var_51722_begin_0 = const()[name = tensor("op_51722_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_51722_end_0 = const()[name = tensor("op_51722_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_51722_end_mask_0 = const()[name = tensor("op_51722_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51722_cast_fp16 = slice_by_index(begin = var_51722_begin_0, end = var_51722_end_0, end_mask = var_51722_end_mask_0, x = value_cast_fp16)[name = tensor("op_51722_cast_fp16")]; + tensor var_51726_begin_0 = const()[name = tensor("op_51726_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_51726_end_0 = const()[name = tensor("op_51726_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_51726_end_mask_0 = const()[name = tensor("op_51726_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51726_cast_fp16 = slice_by_index(begin = var_51726_begin_0, end = var_51726_end_0, end_mask = var_51726_end_mask_0, x = value_cast_fp16)[name = tensor("op_51726_cast_fp16")]; + tensor var_51730_begin_0 = const()[name = tensor("op_51730_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_51730_end_0 = const()[name = tensor("op_51730_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_51730_end_mask_0 = const()[name = tensor("op_51730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51730_cast_fp16 = slice_by_index(begin = var_51730_begin_0, end = var_51730_end_0, end_mask = var_51730_end_mask_0, x = value_cast_fp16)[name = tensor("op_51730_cast_fp16")]; + tensor var_51734_begin_0 = const()[name = tensor("op_51734_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_51734_end_0 = const()[name = tensor("op_51734_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_51734_end_mask_0 = const()[name = tensor("op_51734_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51734_cast_fp16 = slice_by_index(begin = var_51734_begin_0, end = var_51734_end_0, end_mask = var_51734_end_mask_0, x = value_cast_fp16)[name = tensor("op_51734_cast_fp16")]; + tensor var_51738_begin_0 = const()[name = tensor("op_51738_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_51738_end_0 = const()[name = tensor("op_51738_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_51738_end_mask_0 = const()[name = tensor("op_51738_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51738_cast_fp16 = slice_by_index(begin = var_51738_begin_0, end = var_51738_end_0, end_mask = var_51738_end_mask_0, x = value_cast_fp16)[name = tensor("op_51738_cast_fp16")]; + tensor var_51742_begin_0 = const()[name = tensor("op_51742_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_51742_end_0 = const()[name = tensor("op_51742_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_51742_end_mask_0 = const()[name = tensor("op_51742_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51742_cast_fp16 = slice_by_index(begin = var_51742_begin_0, end = var_51742_end_0, end_mask = var_51742_end_mask_0, x = value_cast_fp16)[name = tensor("op_51742_cast_fp16")]; + tensor var_51746_begin_0 = const()[name = tensor("op_51746_begin_0"), val = tensor([0, 512, 0, 0])]; + tensor var_51746_end_0 = const()[name = tensor("op_51746_end_0"), val = tensor([1, 576, 1, 1500])]; + tensor var_51746_end_mask_0 = const()[name = tensor("op_51746_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51746_cast_fp16 = slice_by_index(begin = var_51746_begin_0, end = var_51746_end_0, end_mask = var_51746_end_mask_0, x = value_cast_fp16)[name = tensor("op_51746_cast_fp16")]; + tensor var_51750_begin_0 = const()[name = tensor("op_51750_begin_0"), val = tensor([0, 576, 0, 0])]; + tensor var_51750_end_0 = const()[name = tensor("op_51750_end_0"), val = tensor([1, 640, 1, 1500])]; + tensor var_51750_end_mask_0 = const()[name = tensor("op_51750_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51750_cast_fp16 = slice_by_index(begin = var_51750_begin_0, end = var_51750_end_0, end_mask = var_51750_end_mask_0, x = value_cast_fp16)[name = tensor("op_51750_cast_fp16")]; + tensor var_51754_begin_0 = const()[name = tensor("op_51754_begin_0"), val = tensor([0, 640, 0, 0])]; + tensor var_51754_end_0 = const()[name = tensor("op_51754_end_0"), val = tensor([1, 704, 1, 1500])]; + tensor var_51754_end_mask_0 = const()[name = tensor("op_51754_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51754_cast_fp16 = slice_by_index(begin = var_51754_begin_0, end = var_51754_end_0, end_mask = var_51754_end_mask_0, x = value_cast_fp16)[name = tensor("op_51754_cast_fp16")]; + tensor var_51758_begin_0 = const()[name = tensor("op_51758_begin_0"), val = tensor([0, 704, 0, 0])]; + tensor var_51758_end_0 = const()[name = tensor("op_51758_end_0"), val = tensor([1, 768, 1, 1500])]; + tensor var_51758_end_mask_0 = const()[name = tensor("op_51758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51758_cast_fp16 = slice_by_index(begin = var_51758_begin_0, end = var_51758_end_0, end_mask = var_51758_end_mask_0, x = value_cast_fp16)[name = tensor("op_51758_cast_fp16")]; + tensor var_51762_begin_0 = const()[name = tensor("op_51762_begin_0"), val = tensor([0, 768, 0, 0])]; + tensor var_51762_end_0 = const()[name = tensor("op_51762_end_0"), val = tensor([1, 832, 1, 1500])]; + tensor var_51762_end_mask_0 = const()[name = tensor("op_51762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51762_cast_fp16 = slice_by_index(begin = var_51762_begin_0, end = var_51762_end_0, end_mask = var_51762_end_mask_0, x = value_cast_fp16)[name = tensor("op_51762_cast_fp16")]; + tensor var_51766_begin_0 = const()[name = tensor("op_51766_begin_0"), val = tensor([0, 832, 0, 0])]; + tensor var_51766_end_0 = const()[name = tensor("op_51766_end_0"), val = tensor([1, 896, 1, 1500])]; + tensor var_51766_end_mask_0 = const()[name = tensor("op_51766_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51766_cast_fp16 = slice_by_index(begin = var_51766_begin_0, end = var_51766_end_0, end_mask = var_51766_end_mask_0, x = value_cast_fp16)[name = tensor("op_51766_cast_fp16")]; + tensor var_51770_begin_0 = const()[name = tensor("op_51770_begin_0"), val = tensor([0, 896, 0, 0])]; + tensor var_51770_end_0 = const()[name = tensor("op_51770_end_0"), val = tensor([1, 960, 1, 1500])]; + tensor var_51770_end_mask_0 = const()[name = tensor("op_51770_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51770_cast_fp16 = slice_by_index(begin = var_51770_begin_0, end = var_51770_end_0, end_mask = var_51770_end_mask_0, x = value_cast_fp16)[name = tensor("op_51770_cast_fp16")]; + tensor var_51774_begin_0 = const()[name = tensor("op_51774_begin_0"), val = tensor([0, 960, 0, 0])]; + tensor var_51774_end_0 = const()[name = tensor("op_51774_end_0"), val = tensor([1, 1024, 1, 1500])]; + tensor var_51774_end_mask_0 = const()[name = tensor("op_51774_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51774_cast_fp16 = slice_by_index(begin = var_51774_begin_0, end = var_51774_end_0, end_mask = var_51774_end_mask_0, x = value_cast_fp16)[name = tensor("op_51774_cast_fp16")]; + tensor var_51778_begin_0 = const()[name = tensor("op_51778_begin_0"), val = tensor([0, 1024, 0, 0])]; + tensor var_51778_end_0 = const()[name = tensor("op_51778_end_0"), val = tensor([1, 1088, 1, 1500])]; + tensor var_51778_end_mask_0 = const()[name = tensor("op_51778_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51778_cast_fp16 = slice_by_index(begin = var_51778_begin_0, end = var_51778_end_0, end_mask = var_51778_end_mask_0, x = value_cast_fp16)[name = tensor("op_51778_cast_fp16")]; + tensor var_51782_begin_0 = const()[name = tensor("op_51782_begin_0"), val = tensor([0, 1088, 0, 0])]; + tensor var_51782_end_0 = const()[name = tensor("op_51782_end_0"), val = tensor([1, 1152, 1, 1500])]; + tensor var_51782_end_mask_0 = const()[name = tensor("op_51782_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51782_cast_fp16 = slice_by_index(begin = var_51782_begin_0, end = var_51782_end_0, end_mask = var_51782_end_mask_0, x = value_cast_fp16)[name = tensor("op_51782_cast_fp16")]; + tensor var_51786_begin_0 = const()[name = tensor("op_51786_begin_0"), val = tensor([0, 1152, 0, 0])]; + tensor var_51786_end_0 = const()[name = tensor("op_51786_end_0"), val = tensor([1, 1216, 1, 1500])]; + tensor var_51786_end_mask_0 = const()[name = tensor("op_51786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51786_cast_fp16 = slice_by_index(begin = var_51786_begin_0, end = var_51786_end_0, end_mask = var_51786_end_mask_0, x = value_cast_fp16)[name = tensor("op_51786_cast_fp16")]; + tensor var_51790_begin_0 = const()[name = tensor("op_51790_begin_0"), val = tensor([0, 1216, 0, 0])]; + tensor var_51790_end_0 = const()[name = tensor("op_51790_end_0"), val = tensor([1, 1280, 1, 1500])]; + tensor var_51790_end_mask_0 = const()[name = tensor("op_51790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_51790_cast_fp16 = slice_by_index(begin = var_51790_begin_0, end = var_51790_end_0, end_mask = var_51790_end_mask_0, x = value_cast_fp16)[name = tensor("op_51790_cast_fp16")]; + tensor var_51794_equation_0 = const()[name = tensor("op_51794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51794_cast_fp16 = einsum(equation = var_51794_equation_0, values = (var_51636_cast_fp16, var_51078_cast_fp16))[name = tensor("op_51794_cast_fp16")]; + tensor var_51795_to_fp16 = const()[name = tensor("op_51795_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4961_cast_fp16 = mul(x = var_51794_cast_fp16, y = var_51795_to_fp16)[name = tensor("aw_chunk_4961_cast_fp16")]; + tensor var_51798_equation_0 = const()[name = tensor("op_51798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51798_cast_fp16 = einsum(equation = var_51798_equation_0, values = (var_51636_cast_fp16, var_51085_cast_fp16))[name = tensor("op_51798_cast_fp16")]; + tensor var_51799_to_fp16 = const()[name = tensor("op_51799_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4963_cast_fp16 = mul(x = var_51798_cast_fp16, y = var_51799_to_fp16)[name = tensor("aw_chunk_4963_cast_fp16")]; + tensor var_51802_equation_0 = const()[name = tensor("op_51802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51802_cast_fp16 = einsum(equation = var_51802_equation_0, values = (var_51636_cast_fp16, var_51092_cast_fp16))[name = tensor("op_51802_cast_fp16")]; + tensor var_51803_to_fp16 = const()[name = tensor("op_51803_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4965_cast_fp16 = mul(x = var_51802_cast_fp16, y = var_51803_to_fp16)[name = tensor("aw_chunk_4965_cast_fp16")]; + tensor var_51806_equation_0 = const()[name = tensor("op_51806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51806_cast_fp16 = einsum(equation = var_51806_equation_0, values = (var_51636_cast_fp16, var_51099_cast_fp16))[name = tensor("op_51806_cast_fp16")]; + tensor var_51807_to_fp16 = const()[name = tensor("op_51807_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4967_cast_fp16 = mul(x = var_51806_cast_fp16, y = var_51807_to_fp16)[name = tensor("aw_chunk_4967_cast_fp16")]; + tensor var_51810_equation_0 = const()[name = tensor("op_51810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51810_cast_fp16 = einsum(equation = var_51810_equation_0, values = (var_51640_cast_fp16, var_51106_cast_fp16))[name = tensor("op_51810_cast_fp16")]; + tensor var_51811_to_fp16 = const()[name = tensor("op_51811_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4969_cast_fp16 = mul(x = var_51810_cast_fp16, y = var_51811_to_fp16)[name = tensor("aw_chunk_4969_cast_fp16")]; + tensor var_51814_equation_0 = const()[name = tensor("op_51814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51814_cast_fp16 = einsum(equation = var_51814_equation_0, values = (var_51640_cast_fp16, var_51113_cast_fp16))[name = tensor("op_51814_cast_fp16")]; + tensor var_51815_to_fp16 = const()[name = tensor("op_51815_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4971_cast_fp16 = mul(x = var_51814_cast_fp16, y = var_51815_to_fp16)[name = tensor("aw_chunk_4971_cast_fp16")]; + tensor var_51818_equation_0 = const()[name = tensor("op_51818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51818_cast_fp16 = einsum(equation = var_51818_equation_0, values = (var_51640_cast_fp16, var_51120_cast_fp16))[name = tensor("op_51818_cast_fp16")]; + tensor var_51819_to_fp16 = const()[name = tensor("op_51819_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4973_cast_fp16 = mul(x = var_51818_cast_fp16, y = var_51819_to_fp16)[name = tensor("aw_chunk_4973_cast_fp16")]; + tensor var_51822_equation_0 = const()[name = tensor("op_51822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51822_cast_fp16 = einsum(equation = var_51822_equation_0, values = (var_51640_cast_fp16, var_51127_cast_fp16))[name = tensor("op_51822_cast_fp16")]; + tensor var_51823_to_fp16 = const()[name = tensor("op_51823_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4975_cast_fp16 = mul(x = var_51822_cast_fp16, y = var_51823_to_fp16)[name = tensor("aw_chunk_4975_cast_fp16")]; + tensor var_51826_equation_0 = const()[name = tensor("op_51826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51826_cast_fp16 = einsum(equation = var_51826_equation_0, values = (var_51644_cast_fp16, var_51134_cast_fp16))[name = tensor("op_51826_cast_fp16")]; + tensor var_51827_to_fp16 = const()[name = tensor("op_51827_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4977_cast_fp16 = mul(x = var_51826_cast_fp16, y = var_51827_to_fp16)[name = tensor("aw_chunk_4977_cast_fp16")]; + tensor var_51830_equation_0 = const()[name = tensor("op_51830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51830_cast_fp16 = einsum(equation = var_51830_equation_0, values = (var_51644_cast_fp16, var_51141_cast_fp16))[name = tensor("op_51830_cast_fp16")]; + tensor var_51831_to_fp16 = const()[name = tensor("op_51831_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4979_cast_fp16 = mul(x = var_51830_cast_fp16, y = var_51831_to_fp16)[name = tensor("aw_chunk_4979_cast_fp16")]; + tensor var_51834_equation_0 = const()[name = tensor("op_51834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51834_cast_fp16 = einsum(equation = var_51834_equation_0, values = (var_51644_cast_fp16, var_51148_cast_fp16))[name = tensor("op_51834_cast_fp16")]; + tensor var_51835_to_fp16 = const()[name = tensor("op_51835_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4981_cast_fp16 = mul(x = var_51834_cast_fp16, y = var_51835_to_fp16)[name = tensor("aw_chunk_4981_cast_fp16")]; + tensor var_51838_equation_0 = const()[name = tensor("op_51838_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51838_cast_fp16 = einsum(equation = var_51838_equation_0, values = (var_51644_cast_fp16, var_51155_cast_fp16))[name = tensor("op_51838_cast_fp16")]; + tensor var_51839_to_fp16 = const()[name = tensor("op_51839_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4983_cast_fp16 = mul(x = var_51838_cast_fp16, y = var_51839_to_fp16)[name = tensor("aw_chunk_4983_cast_fp16")]; + tensor var_51842_equation_0 = const()[name = tensor("op_51842_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51842_cast_fp16 = einsum(equation = var_51842_equation_0, values = (var_51648_cast_fp16, var_51162_cast_fp16))[name = tensor("op_51842_cast_fp16")]; + tensor var_51843_to_fp16 = const()[name = tensor("op_51843_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4985_cast_fp16 = mul(x = var_51842_cast_fp16, y = var_51843_to_fp16)[name = tensor("aw_chunk_4985_cast_fp16")]; + tensor var_51846_equation_0 = const()[name = tensor("op_51846_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51846_cast_fp16 = einsum(equation = var_51846_equation_0, values = (var_51648_cast_fp16, var_51169_cast_fp16))[name = tensor("op_51846_cast_fp16")]; + tensor var_51847_to_fp16 = const()[name = tensor("op_51847_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4987_cast_fp16 = mul(x = var_51846_cast_fp16, y = var_51847_to_fp16)[name = tensor("aw_chunk_4987_cast_fp16")]; + tensor var_51850_equation_0 = const()[name = tensor("op_51850_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51850_cast_fp16 = einsum(equation = var_51850_equation_0, values = (var_51648_cast_fp16, var_51176_cast_fp16))[name = tensor("op_51850_cast_fp16")]; + tensor var_51851_to_fp16 = const()[name = tensor("op_51851_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4989_cast_fp16 = mul(x = var_51850_cast_fp16, y = var_51851_to_fp16)[name = tensor("aw_chunk_4989_cast_fp16")]; + tensor var_51854_equation_0 = const()[name = tensor("op_51854_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51854_cast_fp16 = einsum(equation = var_51854_equation_0, values = (var_51648_cast_fp16, var_51183_cast_fp16))[name = tensor("op_51854_cast_fp16")]; + tensor var_51855_to_fp16 = const()[name = tensor("op_51855_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4991_cast_fp16 = mul(x = var_51854_cast_fp16, y = var_51855_to_fp16)[name = tensor("aw_chunk_4991_cast_fp16")]; + tensor var_51858_equation_0 = const()[name = tensor("op_51858_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51858_cast_fp16 = einsum(equation = var_51858_equation_0, values = (var_51652_cast_fp16, var_51190_cast_fp16))[name = tensor("op_51858_cast_fp16")]; + tensor var_51859_to_fp16 = const()[name = tensor("op_51859_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4993_cast_fp16 = mul(x = var_51858_cast_fp16, y = var_51859_to_fp16)[name = tensor("aw_chunk_4993_cast_fp16")]; + tensor var_51862_equation_0 = const()[name = tensor("op_51862_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51862_cast_fp16 = einsum(equation = var_51862_equation_0, values = (var_51652_cast_fp16, var_51197_cast_fp16))[name = tensor("op_51862_cast_fp16")]; + tensor var_51863_to_fp16 = const()[name = tensor("op_51863_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4995_cast_fp16 = mul(x = var_51862_cast_fp16, y = var_51863_to_fp16)[name = tensor("aw_chunk_4995_cast_fp16")]; + tensor var_51866_equation_0 = const()[name = tensor("op_51866_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51866_cast_fp16 = einsum(equation = var_51866_equation_0, values = (var_51652_cast_fp16, var_51204_cast_fp16))[name = tensor("op_51866_cast_fp16")]; + tensor var_51867_to_fp16 = const()[name = tensor("op_51867_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4997_cast_fp16 = mul(x = var_51866_cast_fp16, y = var_51867_to_fp16)[name = tensor("aw_chunk_4997_cast_fp16")]; + tensor var_51870_equation_0 = const()[name = tensor("op_51870_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51870_cast_fp16 = einsum(equation = var_51870_equation_0, values = (var_51652_cast_fp16, var_51211_cast_fp16))[name = tensor("op_51870_cast_fp16")]; + tensor var_51871_to_fp16 = const()[name = tensor("op_51871_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_4999_cast_fp16 = mul(x = var_51870_cast_fp16, y = var_51871_to_fp16)[name = tensor("aw_chunk_4999_cast_fp16")]; + tensor var_51874_equation_0 = const()[name = tensor("op_51874_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51874_cast_fp16 = einsum(equation = var_51874_equation_0, values = (var_51656_cast_fp16, var_51218_cast_fp16))[name = tensor("op_51874_cast_fp16")]; + tensor var_51875_to_fp16 = const()[name = tensor("op_51875_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5001_cast_fp16 = mul(x = var_51874_cast_fp16, y = var_51875_to_fp16)[name = tensor("aw_chunk_5001_cast_fp16")]; + tensor var_51878_equation_0 = const()[name = tensor("op_51878_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51878_cast_fp16 = einsum(equation = var_51878_equation_0, values = (var_51656_cast_fp16, var_51225_cast_fp16))[name = tensor("op_51878_cast_fp16")]; + tensor var_51879_to_fp16 = const()[name = tensor("op_51879_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5003_cast_fp16 = mul(x = var_51878_cast_fp16, y = var_51879_to_fp16)[name = tensor("aw_chunk_5003_cast_fp16")]; + tensor var_51882_equation_0 = const()[name = tensor("op_51882_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51882_cast_fp16 = einsum(equation = var_51882_equation_0, values = (var_51656_cast_fp16, var_51232_cast_fp16))[name = tensor("op_51882_cast_fp16")]; + tensor var_51883_to_fp16 = const()[name = tensor("op_51883_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5005_cast_fp16 = mul(x = var_51882_cast_fp16, y = var_51883_to_fp16)[name = tensor("aw_chunk_5005_cast_fp16")]; + tensor var_51886_equation_0 = const()[name = tensor("op_51886_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51886_cast_fp16 = einsum(equation = var_51886_equation_0, values = (var_51656_cast_fp16, var_51239_cast_fp16))[name = tensor("op_51886_cast_fp16")]; + tensor var_51887_to_fp16 = const()[name = tensor("op_51887_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5007_cast_fp16 = mul(x = var_51886_cast_fp16, y = var_51887_to_fp16)[name = tensor("aw_chunk_5007_cast_fp16")]; + tensor var_51890_equation_0 = const()[name = tensor("op_51890_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51890_cast_fp16 = einsum(equation = var_51890_equation_0, values = (var_51660_cast_fp16, var_51246_cast_fp16))[name = tensor("op_51890_cast_fp16")]; + tensor var_51891_to_fp16 = const()[name = tensor("op_51891_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5009_cast_fp16 = mul(x = var_51890_cast_fp16, y = var_51891_to_fp16)[name = tensor("aw_chunk_5009_cast_fp16")]; + tensor var_51894_equation_0 = const()[name = tensor("op_51894_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51894_cast_fp16 = einsum(equation = var_51894_equation_0, values = (var_51660_cast_fp16, var_51253_cast_fp16))[name = tensor("op_51894_cast_fp16")]; + tensor var_51895_to_fp16 = const()[name = tensor("op_51895_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5011_cast_fp16 = mul(x = var_51894_cast_fp16, y = var_51895_to_fp16)[name = tensor("aw_chunk_5011_cast_fp16")]; + tensor var_51898_equation_0 = const()[name = tensor("op_51898_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51898_cast_fp16 = einsum(equation = var_51898_equation_0, values = (var_51660_cast_fp16, var_51260_cast_fp16))[name = tensor("op_51898_cast_fp16")]; + tensor var_51899_to_fp16 = const()[name = tensor("op_51899_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5013_cast_fp16 = mul(x = var_51898_cast_fp16, y = var_51899_to_fp16)[name = tensor("aw_chunk_5013_cast_fp16")]; + tensor var_51902_equation_0 = const()[name = tensor("op_51902_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51902_cast_fp16 = einsum(equation = var_51902_equation_0, values = (var_51660_cast_fp16, var_51267_cast_fp16))[name = tensor("op_51902_cast_fp16")]; + tensor var_51903_to_fp16 = const()[name = tensor("op_51903_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5015_cast_fp16 = mul(x = var_51902_cast_fp16, y = var_51903_to_fp16)[name = tensor("aw_chunk_5015_cast_fp16")]; + tensor var_51906_equation_0 = const()[name = tensor("op_51906_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51906_cast_fp16 = einsum(equation = var_51906_equation_0, values = (var_51664_cast_fp16, var_51274_cast_fp16))[name = tensor("op_51906_cast_fp16")]; + tensor var_51907_to_fp16 = const()[name = tensor("op_51907_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5017_cast_fp16 = mul(x = var_51906_cast_fp16, y = var_51907_to_fp16)[name = tensor("aw_chunk_5017_cast_fp16")]; + tensor var_51910_equation_0 = const()[name = tensor("op_51910_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51910_cast_fp16 = einsum(equation = var_51910_equation_0, values = (var_51664_cast_fp16, var_51281_cast_fp16))[name = tensor("op_51910_cast_fp16")]; + tensor var_51911_to_fp16 = const()[name = tensor("op_51911_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5019_cast_fp16 = mul(x = var_51910_cast_fp16, y = var_51911_to_fp16)[name = tensor("aw_chunk_5019_cast_fp16")]; + tensor var_51914_equation_0 = const()[name = tensor("op_51914_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51914_cast_fp16 = einsum(equation = var_51914_equation_0, values = (var_51664_cast_fp16, var_51288_cast_fp16))[name = tensor("op_51914_cast_fp16")]; + tensor var_51915_to_fp16 = const()[name = tensor("op_51915_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5021_cast_fp16 = mul(x = var_51914_cast_fp16, y = var_51915_to_fp16)[name = tensor("aw_chunk_5021_cast_fp16")]; + tensor var_51918_equation_0 = const()[name = tensor("op_51918_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51918_cast_fp16 = einsum(equation = var_51918_equation_0, values = (var_51664_cast_fp16, var_51295_cast_fp16))[name = tensor("op_51918_cast_fp16")]; + tensor var_51919_to_fp16 = const()[name = tensor("op_51919_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5023_cast_fp16 = mul(x = var_51918_cast_fp16, y = var_51919_to_fp16)[name = tensor("aw_chunk_5023_cast_fp16")]; + tensor var_51922_equation_0 = const()[name = tensor("op_51922_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51922_cast_fp16 = einsum(equation = var_51922_equation_0, values = (var_51668_cast_fp16, var_51302_cast_fp16))[name = tensor("op_51922_cast_fp16")]; + tensor var_51923_to_fp16 = const()[name = tensor("op_51923_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5025_cast_fp16 = mul(x = var_51922_cast_fp16, y = var_51923_to_fp16)[name = tensor("aw_chunk_5025_cast_fp16")]; + tensor var_51926_equation_0 = const()[name = tensor("op_51926_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51926_cast_fp16 = einsum(equation = var_51926_equation_0, values = (var_51668_cast_fp16, var_51309_cast_fp16))[name = tensor("op_51926_cast_fp16")]; + tensor var_51927_to_fp16 = const()[name = tensor("op_51927_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5027_cast_fp16 = mul(x = var_51926_cast_fp16, y = var_51927_to_fp16)[name = tensor("aw_chunk_5027_cast_fp16")]; + tensor var_51930_equation_0 = const()[name = tensor("op_51930_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51930_cast_fp16 = einsum(equation = var_51930_equation_0, values = (var_51668_cast_fp16, var_51316_cast_fp16))[name = tensor("op_51930_cast_fp16")]; + tensor var_51931_to_fp16 = const()[name = tensor("op_51931_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5029_cast_fp16 = mul(x = var_51930_cast_fp16, y = var_51931_to_fp16)[name = tensor("aw_chunk_5029_cast_fp16")]; + tensor var_51934_equation_0 = const()[name = tensor("op_51934_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51934_cast_fp16 = einsum(equation = var_51934_equation_0, values = (var_51668_cast_fp16, var_51323_cast_fp16))[name = tensor("op_51934_cast_fp16")]; + tensor var_51935_to_fp16 = const()[name = tensor("op_51935_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5031_cast_fp16 = mul(x = var_51934_cast_fp16, y = var_51935_to_fp16)[name = tensor("aw_chunk_5031_cast_fp16")]; + tensor var_51938_equation_0 = const()[name = tensor("op_51938_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51938_cast_fp16 = einsum(equation = var_51938_equation_0, values = (var_51672_cast_fp16, var_51330_cast_fp16))[name = tensor("op_51938_cast_fp16")]; + tensor var_51939_to_fp16 = const()[name = tensor("op_51939_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5033_cast_fp16 = mul(x = var_51938_cast_fp16, y = var_51939_to_fp16)[name = tensor("aw_chunk_5033_cast_fp16")]; + tensor var_51942_equation_0 = const()[name = tensor("op_51942_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51942_cast_fp16 = einsum(equation = var_51942_equation_0, values = (var_51672_cast_fp16, var_51337_cast_fp16))[name = tensor("op_51942_cast_fp16")]; + tensor var_51943_to_fp16 = const()[name = tensor("op_51943_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5035_cast_fp16 = mul(x = var_51942_cast_fp16, y = var_51943_to_fp16)[name = tensor("aw_chunk_5035_cast_fp16")]; + tensor var_51946_equation_0 = const()[name = tensor("op_51946_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51946_cast_fp16 = einsum(equation = var_51946_equation_0, values = (var_51672_cast_fp16, var_51344_cast_fp16))[name = tensor("op_51946_cast_fp16")]; + tensor var_51947_to_fp16 = const()[name = tensor("op_51947_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5037_cast_fp16 = mul(x = var_51946_cast_fp16, y = var_51947_to_fp16)[name = tensor("aw_chunk_5037_cast_fp16")]; + tensor var_51950_equation_0 = const()[name = tensor("op_51950_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51950_cast_fp16 = einsum(equation = var_51950_equation_0, values = (var_51672_cast_fp16, var_51351_cast_fp16))[name = tensor("op_51950_cast_fp16")]; + tensor var_51951_to_fp16 = const()[name = tensor("op_51951_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5039_cast_fp16 = mul(x = var_51950_cast_fp16, y = var_51951_to_fp16)[name = tensor("aw_chunk_5039_cast_fp16")]; + tensor var_51954_equation_0 = const()[name = tensor("op_51954_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51954_cast_fp16 = einsum(equation = var_51954_equation_0, values = (var_51676_cast_fp16, var_51358_cast_fp16))[name = tensor("op_51954_cast_fp16")]; + tensor var_51955_to_fp16 = const()[name = tensor("op_51955_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5041_cast_fp16 = mul(x = var_51954_cast_fp16, y = var_51955_to_fp16)[name = tensor("aw_chunk_5041_cast_fp16")]; + tensor var_51958_equation_0 = const()[name = tensor("op_51958_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51958_cast_fp16 = einsum(equation = var_51958_equation_0, values = (var_51676_cast_fp16, var_51365_cast_fp16))[name = tensor("op_51958_cast_fp16")]; + tensor var_51959_to_fp16 = const()[name = tensor("op_51959_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5043_cast_fp16 = mul(x = var_51958_cast_fp16, y = var_51959_to_fp16)[name = tensor("aw_chunk_5043_cast_fp16")]; + tensor var_51962_equation_0 = const()[name = tensor("op_51962_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51962_cast_fp16 = einsum(equation = var_51962_equation_0, values = (var_51676_cast_fp16, var_51372_cast_fp16))[name = tensor("op_51962_cast_fp16")]; + tensor var_51963_to_fp16 = const()[name = tensor("op_51963_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5045_cast_fp16 = mul(x = var_51962_cast_fp16, y = var_51963_to_fp16)[name = tensor("aw_chunk_5045_cast_fp16")]; + tensor var_51966_equation_0 = const()[name = tensor("op_51966_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51966_cast_fp16 = einsum(equation = var_51966_equation_0, values = (var_51676_cast_fp16, var_51379_cast_fp16))[name = tensor("op_51966_cast_fp16")]; + tensor var_51967_to_fp16 = const()[name = tensor("op_51967_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5047_cast_fp16 = mul(x = var_51966_cast_fp16, y = var_51967_to_fp16)[name = tensor("aw_chunk_5047_cast_fp16")]; + tensor var_51970_equation_0 = const()[name = tensor("op_51970_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51970_cast_fp16 = einsum(equation = var_51970_equation_0, values = (var_51680_cast_fp16, var_51386_cast_fp16))[name = tensor("op_51970_cast_fp16")]; + tensor var_51971_to_fp16 = const()[name = tensor("op_51971_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5049_cast_fp16 = mul(x = var_51970_cast_fp16, y = var_51971_to_fp16)[name = tensor("aw_chunk_5049_cast_fp16")]; + tensor var_51974_equation_0 = const()[name = tensor("op_51974_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51974_cast_fp16 = einsum(equation = var_51974_equation_0, values = (var_51680_cast_fp16, var_51393_cast_fp16))[name = tensor("op_51974_cast_fp16")]; + tensor var_51975_to_fp16 = const()[name = tensor("op_51975_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5051_cast_fp16 = mul(x = var_51974_cast_fp16, y = var_51975_to_fp16)[name = tensor("aw_chunk_5051_cast_fp16")]; + tensor var_51978_equation_0 = const()[name = tensor("op_51978_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51978_cast_fp16 = einsum(equation = var_51978_equation_0, values = (var_51680_cast_fp16, var_51400_cast_fp16))[name = tensor("op_51978_cast_fp16")]; + tensor var_51979_to_fp16 = const()[name = tensor("op_51979_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5053_cast_fp16 = mul(x = var_51978_cast_fp16, y = var_51979_to_fp16)[name = tensor("aw_chunk_5053_cast_fp16")]; + tensor var_51982_equation_0 = const()[name = tensor("op_51982_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51982_cast_fp16 = einsum(equation = var_51982_equation_0, values = (var_51680_cast_fp16, var_51407_cast_fp16))[name = tensor("op_51982_cast_fp16")]; + tensor var_51983_to_fp16 = const()[name = tensor("op_51983_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5055_cast_fp16 = mul(x = var_51982_cast_fp16, y = var_51983_to_fp16)[name = tensor("aw_chunk_5055_cast_fp16")]; + tensor var_51986_equation_0 = const()[name = tensor("op_51986_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51986_cast_fp16 = einsum(equation = var_51986_equation_0, values = (var_51684_cast_fp16, var_51414_cast_fp16))[name = tensor("op_51986_cast_fp16")]; + tensor var_51987_to_fp16 = const()[name = tensor("op_51987_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5057_cast_fp16 = mul(x = var_51986_cast_fp16, y = var_51987_to_fp16)[name = tensor("aw_chunk_5057_cast_fp16")]; + tensor var_51990_equation_0 = const()[name = tensor("op_51990_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51990_cast_fp16 = einsum(equation = var_51990_equation_0, values = (var_51684_cast_fp16, var_51421_cast_fp16))[name = tensor("op_51990_cast_fp16")]; + tensor var_51991_to_fp16 = const()[name = tensor("op_51991_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5059_cast_fp16 = mul(x = var_51990_cast_fp16, y = var_51991_to_fp16)[name = tensor("aw_chunk_5059_cast_fp16")]; + tensor var_51994_equation_0 = const()[name = tensor("op_51994_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51994_cast_fp16 = einsum(equation = var_51994_equation_0, values = (var_51684_cast_fp16, var_51428_cast_fp16))[name = tensor("op_51994_cast_fp16")]; + tensor var_51995_to_fp16 = const()[name = tensor("op_51995_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5061_cast_fp16 = mul(x = var_51994_cast_fp16, y = var_51995_to_fp16)[name = tensor("aw_chunk_5061_cast_fp16")]; + tensor var_51998_equation_0 = const()[name = tensor("op_51998_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_51998_cast_fp16 = einsum(equation = var_51998_equation_0, values = (var_51684_cast_fp16, var_51435_cast_fp16))[name = tensor("op_51998_cast_fp16")]; + tensor var_51999_to_fp16 = const()[name = tensor("op_51999_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5063_cast_fp16 = mul(x = var_51998_cast_fp16, y = var_51999_to_fp16)[name = tensor("aw_chunk_5063_cast_fp16")]; + tensor var_52002_equation_0 = const()[name = tensor("op_52002_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52002_cast_fp16 = einsum(equation = var_52002_equation_0, values = (var_51688_cast_fp16, var_51442_cast_fp16))[name = tensor("op_52002_cast_fp16")]; + tensor var_52003_to_fp16 = const()[name = tensor("op_52003_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5065_cast_fp16 = mul(x = var_52002_cast_fp16, y = var_52003_to_fp16)[name = tensor("aw_chunk_5065_cast_fp16")]; + tensor var_52006_equation_0 = const()[name = tensor("op_52006_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52006_cast_fp16 = einsum(equation = var_52006_equation_0, values = (var_51688_cast_fp16, var_51449_cast_fp16))[name = tensor("op_52006_cast_fp16")]; + tensor var_52007_to_fp16 = const()[name = tensor("op_52007_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5067_cast_fp16 = mul(x = var_52006_cast_fp16, y = var_52007_to_fp16)[name = tensor("aw_chunk_5067_cast_fp16")]; + tensor var_52010_equation_0 = const()[name = tensor("op_52010_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52010_cast_fp16 = einsum(equation = var_52010_equation_0, values = (var_51688_cast_fp16, var_51456_cast_fp16))[name = tensor("op_52010_cast_fp16")]; + tensor var_52011_to_fp16 = const()[name = tensor("op_52011_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5069_cast_fp16 = mul(x = var_52010_cast_fp16, y = var_52011_to_fp16)[name = tensor("aw_chunk_5069_cast_fp16")]; + tensor var_52014_equation_0 = const()[name = tensor("op_52014_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52014_cast_fp16 = einsum(equation = var_52014_equation_0, values = (var_51688_cast_fp16, var_51463_cast_fp16))[name = tensor("op_52014_cast_fp16")]; + tensor var_52015_to_fp16 = const()[name = tensor("op_52015_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5071_cast_fp16 = mul(x = var_52014_cast_fp16, y = var_52015_to_fp16)[name = tensor("aw_chunk_5071_cast_fp16")]; + tensor var_52018_equation_0 = const()[name = tensor("op_52018_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52018_cast_fp16 = einsum(equation = var_52018_equation_0, values = (var_51692_cast_fp16, var_51470_cast_fp16))[name = tensor("op_52018_cast_fp16")]; + tensor var_52019_to_fp16 = const()[name = tensor("op_52019_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5073_cast_fp16 = mul(x = var_52018_cast_fp16, y = var_52019_to_fp16)[name = tensor("aw_chunk_5073_cast_fp16")]; + tensor var_52022_equation_0 = const()[name = tensor("op_52022_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52022_cast_fp16 = einsum(equation = var_52022_equation_0, values = (var_51692_cast_fp16, var_51477_cast_fp16))[name = tensor("op_52022_cast_fp16")]; + tensor var_52023_to_fp16 = const()[name = tensor("op_52023_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5075_cast_fp16 = mul(x = var_52022_cast_fp16, y = var_52023_to_fp16)[name = tensor("aw_chunk_5075_cast_fp16")]; + tensor var_52026_equation_0 = const()[name = tensor("op_52026_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52026_cast_fp16 = einsum(equation = var_52026_equation_0, values = (var_51692_cast_fp16, var_51484_cast_fp16))[name = tensor("op_52026_cast_fp16")]; + tensor var_52027_to_fp16 = const()[name = tensor("op_52027_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5077_cast_fp16 = mul(x = var_52026_cast_fp16, y = var_52027_to_fp16)[name = tensor("aw_chunk_5077_cast_fp16")]; + tensor var_52030_equation_0 = const()[name = tensor("op_52030_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52030_cast_fp16 = einsum(equation = var_52030_equation_0, values = (var_51692_cast_fp16, var_51491_cast_fp16))[name = tensor("op_52030_cast_fp16")]; + tensor var_52031_to_fp16 = const()[name = tensor("op_52031_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5079_cast_fp16 = mul(x = var_52030_cast_fp16, y = var_52031_to_fp16)[name = tensor("aw_chunk_5079_cast_fp16")]; + tensor var_52034_equation_0 = const()[name = tensor("op_52034_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52034_cast_fp16 = einsum(equation = var_52034_equation_0, values = (var_51696_cast_fp16, var_51498_cast_fp16))[name = tensor("op_52034_cast_fp16")]; + tensor var_52035_to_fp16 = const()[name = tensor("op_52035_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5081_cast_fp16 = mul(x = var_52034_cast_fp16, y = var_52035_to_fp16)[name = tensor("aw_chunk_5081_cast_fp16")]; + tensor var_52038_equation_0 = const()[name = tensor("op_52038_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52038_cast_fp16 = einsum(equation = var_52038_equation_0, values = (var_51696_cast_fp16, var_51505_cast_fp16))[name = tensor("op_52038_cast_fp16")]; + tensor var_52039_to_fp16 = const()[name = tensor("op_52039_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5083_cast_fp16 = mul(x = var_52038_cast_fp16, y = var_52039_to_fp16)[name = tensor("aw_chunk_5083_cast_fp16")]; + tensor var_52042_equation_0 = const()[name = tensor("op_52042_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52042_cast_fp16 = einsum(equation = var_52042_equation_0, values = (var_51696_cast_fp16, var_51512_cast_fp16))[name = tensor("op_52042_cast_fp16")]; + tensor var_52043_to_fp16 = const()[name = tensor("op_52043_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5085_cast_fp16 = mul(x = var_52042_cast_fp16, y = var_52043_to_fp16)[name = tensor("aw_chunk_5085_cast_fp16")]; + tensor var_52046_equation_0 = const()[name = tensor("op_52046_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52046_cast_fp16 = einsum(equation = var_52046_equation_0, values = (var_51696_cast_fp16, var_51519_cast_fp16))[name = tensor("op_52046_cast_fp16")]; + tensor var_52047_to_fp16 = const()[name = tensor("op_52047_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5087_cast_fp16 = mul(x = var_52046_cast_fp16, y = var_52047_to_fp16)[name = tensor("aw_chunk_5087_cast_fp16")]; + tensor var_52050_equation_0 = const()[name = tensor("op_52050_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52050_cast_fp16 = einsum(equation = var_52050_equation_0, values = (var_51700_cast_fp16, var_51526_cast_fp16))[name = tensor("op_52050_cast_fp16")]; + tensor var_52051_to_fp16 = const()[name = tensor("op_52051_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5089_cast_fp16 = mul(x = var_52050_cast_fp16, y = var_52051_to_fp16)[name = tensor("aw_chunk_5089_cast_fp16")]; + tensor var_52054_equation_0 = const()[name = tensor("op_52054_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52054_cast_fp16 = einsum(equation = var_52054_equation_0, values = (var_51700_cast_fp16, var_51533_cast_fp16))[name = tensor("op_52054_cast_fp16")]; + tensor var_52055_to_fp16 = const()[name = tensor("op_52055_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5091_cast_fp16 = mul(x = var_52054_cast_fp16, y = var_52055_to_fp16)[name = tensor("aw_chunk_5091_cast_fp16")]; + tensor var_52058_equation_0 = const()[name = tensor("op_52058_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52058_cast_fp16 = einsum(equation = var_52058_equation_0, values = (var_51700_cast_fp16, var_51540_cast_fp16))[name = tensor("op_52058_cast_fp16")]; + tensor var_52059_to_fp16 = const()[name = tensor("op_52059_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5093_cast_fp16 = mul(x = var_52058_cast_fp16, y = var_52059_to_fp16)[name = tensor("aw_chunk_5093_cast_fp16")]; + tensor var_52062_equation_0 = const()[name = tensor("op_52062_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52062_cast_fp16 = einsum(equation = var_52062_equation_0, values = (var_51700_cast_fp16, var_51547_cast_fp16))[name = tensor("op_52062_cast_fp16")]; + tensor var_52063_to_fp16 = const()[name = tensor("op_52063_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5095_cast_fp16 = mul(x = var_52062_cast_fp16, y = var_52063_to_fp16)[name = tensor("aw_chunk_5095_cast_fp16")]; + tensor var_52066_equation_0 = const()[name = tensor("op_52066_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52066_cast_fp16 = einsum(equation = var_52066_equation_0, values = (var_51704_cast_fp16, var_51554_cast_fp16))[name = tensor("op_52066_cast_fp16")]; + tensor var_52067_to_fp16 = const()[name = tensor("op_52067_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5097_cast_fp16 = mul(x = var_52066_cast_fp16, y = var_52067_to_fp16)[name = tensor("aw_chunk_5097_cast_fp16")]; + tensor var_52070_equation_0 = const()[name = tensor("op_52070_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52070_cast_fp16 = einsum(equation = var_52070_equation_0, values = (var_51704_cast_fp16, var_51561_cast_fp16))[name = tensor("op_52070_cast_fp16")]; + tensor var_52071_to_fp16 = const()[name = tensor("op_52071_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5099_cast_fp16 = mul(x = var_52070_cast_fp16, y = var_52071_to_fp16)[name = tensor("aw_chunk_5099_cast_fp16")]; + tensor var_52074_equation_0 = const()[name = tensor("op_52074_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52074_cast_fp16 = einsum(equation = var_52074_equation_0, values = (var_51704_cast_fp16, var_51568_cast_fp16))[name = tensor("op_52074_cast_fp16")]; + tensor var_52075_to_fp16 = const()[name = tensor("op_52075_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5101_cast_fp16 = mul(x = var_52074_cast_fp16, y = var_52075_to_fp16)[name = tensor("aw_chunk_5101_cast_fp16")]; + tensor var_52078_equation_0 = const()[name = tensor("op_52078_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52078_cast_fp16 = einsum(equation = var_52078_equation_0, values = (var_51704_cast_fp16, var_51575_cast_fp16))[name = tensor("op_52078_cast_fp16")]; + tensor var_52079_to_fp16 = const()[name = tensor("op_52079_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5103_cast_fp16 = mul(x = var_52078_cast_fp16, y = var_52079_to_fp16)[name = tensor("aw_chunk_5103_cast_fp16")]; + tensor var_52082_equation_0 = const()[name = tensor("op_52082_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52082_cast_fp16 = einsum(equation = var_52082_equation_0, values = (var_51708_cast_fp16, var_51582_cast_fp16))[name = tensor("op_52082_cast_fp16")]; + tensor var_52083_to_fp16 = const()[name = tensor("op_52083_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5105_cast_fp16 = mul(x = var_52082_cast_fp16, y = var_52083_to_fp16)[name = tensor("aw_chunk_5105_cast_fp16")]; + tensor var_52086_equation_0 = const()[name = tensor("op_52086_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52086_cast_fp16 = einsum(equation = var_52086_equation_0, values = (var_51708_cast_fp16, var_51589_cast_fp16))[name = tensor("op_52086_cast_fp16")]; + tensor var_52087_to_fp16 = const()[name = tensor("op_52087_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5107_cast_fp16 = mul(x = var_52086_cast_fp16, y = var_52087_to_fp16)[name = tensor("aw_chunk_5107_cast_fp16")]; + tensor var_52090_equation_0 = const()[name = tensor("op_52090_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52090_cast_fp16 = einsum(equation = var_52090_equation_0, values = (var_51708_cast_fp16, var_51596_cast_fp16))[name = tensor("op_52090_cast_fp16")]; + tensor var_52091_to_fp16 = const()[name = tensor("op_52091_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5109_cast_fp16 = mul(x = var_52090_cast_fp16, y = var_52091_to_fp16)[name = tensor("aw_chunk_5109_cast_fp16")]; + tensor var_52094_equation_0 = const()[name = tensor("op_52094_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52094_cast_fp16 = einsum(equation = var_52094_equation_0, values = (var_51708_cast_fp16, var_51603_cast_fp16))[name = tensor("op_52094_cast_fp16")]; + tensor var_52095_to_fp16 = const()[name = tensor("op_52095_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5111_cast_fp16 = mul(x = var_52094_cast_fp16, y = var_52095_to_fp16)[name = tensor("aw_chunk_5111_cast_fp16")]; + tensor var_52098_equation_0 = const()[name = tensor("op_52098_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52098_cast_fp16 = einsum(equation = var_52098_equation_0, values = (var_51712_cast_fp16, var_51610_cast_fp16))[name = tensor("op_52098_cast_fp16")]; + tensor var_52099_to_fp16 = const()[name = tensor("op_52099_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5113_cast_fp16 = mul(x = var_52098_cast_fp16, y = var_52099_to_fp16)[name = tensor("aw_chunk_5113_cast_fp16")]; + tensor var_52102_equation_0 = const()[name = tensor("op_52102_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52102_cast_fp16 = einsum(equation = var_52102_equation_0, values = (var_51712_cast_fp16, var_51617_cast_fp16))[name = tensor("op_52102_cast_fp16")]; + tensor var_52103_to_fp16 = const()[name = tensor("op_52103_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5115_cast_fp16 = mul(x = var_52102_cast_fp16, y = var_52103_to_fp16)[name = tensor("aw_chunk_5115_cast_fp16")]; + tensor var_52106_equation_0 = const()[name = tensor("op_52106_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52106_cast_fp16 = einsum(equation = var_52106_equation_0, values = (var_51712_cast_fp16, var_51624_cast_fp16))[name = tensor("op_52106_cast_fp16")]; + tensor var_52107_to_fp16 = const()[name = tensor("op_52107_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_5117_cast_fp16 = mul(x = var_52106_cast_fp16, y = var_52107_to_fp16)[name = tensor("aw_chunk_5117_cast_fp16")]; + tensor var_52110_equation_0 = const()[name = tensor("op_52110_equation_0"), val = tensor("bkhc,bchq->bkhq")]; + tensor var_52110_cast_fp16 = einsum(equation = var_52110_equation_0, values = (var_51712_cast_fp16, var_51631_cast_fp16))[name = tensor("op_52110_cast_fp16")]; + tensor var_52111_to_fp16 = const()[name = tensor("op_52111_to_fp16"), val = tensor(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = var_52110_cast_fp16, y = var_52111_to_fp16)[name = tensor("aw_chunk_cast_fp16")]; + tensor var_52113_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4961_cast_fp16)[name = tensor("op_52113_cast_fp16")]; + tensor var_52114_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4963_cast_fp16)[name = tensor("op_52114_cast_fp16")]; + tensor var_52115_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4965_cast_fp16)[name = tensor("op_52115_cast_fp16")]; + tensor var_52116_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4967_cast_fp16)[name = tensor("op_52116_cast_fp16")]; + tensor var_52117_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4969_cast_fp16)[name = tensor("op_52117_cast_fp16")]; + tensor var_52118_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4971_cast_fp16)[name = tensor("op_52118_cast_fp16")]; + tensor var_52119_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4973_cast_fp16)[name = tensor("op_52119_cast_fp16")]; + tensor var_52120_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4975_cast_fp16)[name = tensor("op_52120_cast_fp16")]; + tensor var_52121_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4977_cast_fp16)[name = tensor("op_52121_cast_fp16")]; + tensor var_52122_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4979_cast_fp16)[name = tensor("op_52122_cast_fp16")]; + tensor var_52123_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4981_cast_fp16)[name = tensor("op_52123_cast_fp16")]; + tensor var_52124_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4983_cast_fp16)[name = tensor("op_52124_cast_fp16")]; + tensor var_52125_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4985_cast_fp16)[name = tensor("op_52125_cast_fp16")]; + tensor var_52126_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4987_cast_fp16)[name = tensor("op_52126_cast_fp16")]; + tensor var_52127_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4989_cast_fp16)[name = tensor("op_52127_cast_fp16")]; + tensor var_52128_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4991_cast_fp16)[name = tensor("op_52128_cast_fp16")]; + tensor var_52129_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4993_cast_fp16)[name = tensor("op_52129_cast_fp16")]; + tensor var_52130_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4995_cast_fp16)[name = tensor("op_52130_cast_fp16")]; + tensor var_52131_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4997_cast_fp16)[name = tensor("op_52131_cast_fp16")]; + tensor var_52132_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_4999_cast_fp16)[name = tensor("op_52132_cast_fp16")]; + tensor var_52133_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5001_cast_fp16)[name = tensor("op_52133_cast_fp16")]; + tensor var_52134_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5003_cast_fp16)[name = tensor("op_52134_cast_fp16")]; + tensor var_52135_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5005_cast_fp16)[name = tensor("op_52135_cast_fp16")]; + tensor var_52136_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5007_cast_fp16)[name = tensor("op_52136_cast_fp16")]; + tensor var_52137_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5009_cast_fp16)[name = tensor("op_52137_cast_fp16")]; + tensor var_52138_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5011_cast_fp16)[name = tensor("op_52138_cast_fp16")]; + tensor var_52139_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5013_cast_fp16)[name = tensor("op_52139_cast_fp16")]; + tensor var_52140_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5015_cast_fp16)[name = tensor("op_52140_cast_fp16")]; + tensor var_52141_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5017_cast_fp16)[name = tensor("op_52141_cast_fp16")]; + tensor var_52142_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5019_cast_fp16)[name = tensor("op_52142_cast_fp16")]; + tensor var_52143_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5021_cast_fp16)[name = tensor("op_52143_cast_fp16")]; + tensor var_52144_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5023_cast_fp16)[name = tensor("op_52144_cast_fp16")]; + tensor var_52145_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5025_cast_fp16)[name = tensor("op_52145_cast_fp16")]; + tensor var_52146_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5027_cast_fp16)[name = tensor("op_52146_cast_fp16")]; + tensor var_52147_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5029_cast_fp16)[name = tensor("op_52147_cast_fp16")]; + tensor var_52148_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5031_cast_fp16)[name = tensor("op_52148_cast_fp16")]; + tensor var_52149_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5033_cast_fp16)[name = tensor("op_52149_cast_fp16")]; + tensor var_52150_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5035_cast_fp16)[name = tensor("op_52150_cast_fp16")]; + tensor var_52151_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5037_cast_fp16)[name = tensor("op_52151_cast_fp16")]; + tensor var_52152_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5039_cast_fp16)[name = tensor("op_52152_cast_fp16")]; + tensor var_52153_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5041_cast_fp16)[name = tensor("op_52153_cast_fp16")]; + tensor var_52154_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5043_cast_fp16)[name = tensor("op_52154_cast_fp16")]; + tensor var_52155_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5045_cast_fp16)[name = tensor("op_52155_cast_fp16")]; + tensor var_52156_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5047_cast_fp16)[name = tensor("op_52156_cast_fp16")]; + tensor var_52157_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5049_cast_fp16)[name = tensor("op_52157_cast_fp16")]; + tensor var_52158_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5051_cast_fp16)[name = tensor("op_52158_cast_fp16")]; + tensor var_52159_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5053_cast_fp16)[name = tensor("op_52159_cast_fp16")]; + tensor var_52160_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5055_cast_fp16)[name = tensor("op_52160_cast_fp16")]; + tensor var_52161_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5057_cast_fp16)[name = tensor("op_52161_cast_fp16")]; + tensor var_52162_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5059_cast_fp16)[name = tensor("op_52162_cast_fp16")]; + tensor var_52163_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5061_cast_fp16)[name = tensor("op_52163_cast_fp16")]; + tensor var_52164_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5063_cast_fp16)[name = tensor("op_52164_cast_fp16")]; + tensor var_52165_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5065_cast_fp16)[name = tensor("op_52165_cast_fp16")]; + tensor var_52166_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5067_cast_fp16)[name = tensor("op_52166_cast_fp16")]; + tensor var_52167_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5069_cast_fp16)[name = tensor("op_52167_cast_fp16")]; + tensor var_52168_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5071_cast_fp16)[name = tensor("op_52168_cast_fp16")]; + tensor var_52169_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5073_cast_fp16)[name = tensor("op_52169_cast_fp16")]; + tensor var_52170_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5075_cast_fp16)[name = tensor("op_52170_cast_fp16")]; + tensor var_52171_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5077_cast_fp16)[name = tensor("op_52171_cast_fp16")]; + tensor var_52172_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5079_cast_fp16)[name = tensor("op_52172_cast_fp16")]; + tensor var_52173_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5081_cast_fp16)[name = tensor("op_52173_cast_fp16")]; + tensor var_52174_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5083_cast_fp16)[name = tensor("op_52174_cast_fp16")]; + tensor var_52175_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5085_cast_fp16)[name = tensor("op_52175_cast_fp16")]; + tensor var_52176_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5087_cast_fp16)[name = tensor("op_52176_cast_fp16")]; + tensor var_52177_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5089_cast_fp16)[name = tensor("op_52177_cast_fp16")]; + tensor var_52178_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5091_cast_fp16)[name = tensor("op_52178_cast_fp16")]; + tensor var_52179_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5093_cast_fp16)[name = tensor("op_52179_cast_fp16")]; + tensor var_52180_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5095_cast_fp16)[name = tensor("op_52180_cast_fp16")]; + tensor var_52181_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5097_cast_fp16)[name = tensor("op_52181_cast_fp16")]; + tensor var_52182_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5099_cast_fp16)[name = tensor("op_52182_cast_fp16")]; + tensor var_52183_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5101_cast_fp16)[name = tensor("op_52183_cast_fp16")]; + tensor var_52184_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5103_cast_fp16)[name = tensor("op_52184_cast_fp16")]; + tensor var_52185_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5105_cast_fp16)[name = tensor("op_52185_cast_fp16")]; + tensor var_52186_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5107_cast_fp16)[name = tensor("op_52186_cast_fp16")]; + tensor var_52187_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5109_cast_fp16)[name = tensor("op_52187_cast_fp16")]; + tensor var_52188_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5111_cast_fp16)[name = tensor("op_52188_cast_fp16")]; + tensor var_52189_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5113_cast_fp16)[name = tensor("op_52189_cast_fp16")]; + tensor var_52190_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5115_cast_fp16)[name = tensor("op_52190_cast_fp16")]; + tensor var_52191_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_5117_cast_fp16)[name = tensor("op_52191_cast_fp16")]; + tensor var_52192_cast_fp16 = softmax(axis = var_50884, x = aw_chunk_cast_fp16)[name = tensor("op_52192_cast_fp16")]; + tensor var_52194_equation_0 = const()[name = tensor("op_52194_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52194_cast_fp16 = einsum(equation = var_52194_equation_0, values = (var_51714_cast_fp16, var_52113_cast_fp16))[name = tensor("op_52194_cast_fp16")]; + tensor var_52196_equation_0 = const()[name = tensor("op_52196_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52196_cast_fp16 = einsum(equation = var_52196_equation_0, values = (var_51714_cast_fp16, var_52114_cast_fp16))[name = tensor("op_52196_cast_fp16")]; + tensor var_52198_equation_0 = const()[name = tensor("op_52198_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52198_cast_fp16 = einsum(equation = var_52198_equation_0, values = (var_51714_cast_fp16, var_52115_cast_fp16))[name = tensor("op_52198_cast_fp16")]; + tensor var_52200_equation_0 = const()[name = tensor("op_52200_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52200_cast_fp16 = einsum(equation = var_52200_equation_0, values = (var_51714_cast_fp16, var_52116_cast_fp16))[name = tensor("op_52200_cast_fp16")]; + tensor var_52202_equation_0 = const()[name = tensor("op_52202_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52202_cast_fp16 = einsum(equation = var_52202_equation_0, values = (var_51718_cast_fp16, var_52117_cast_fp16))[name = tensor("op_52202_cast_fp16")]; + tensor var_52204_equation_0 = const()[name = tensor("op_52204_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52204_cast_fp16 = einsum(equation = var_52204_equation_0, values = (var_51718_cast_fp16, var_52118_cast_fp16))[name = tensor("op_52204_cast_fp16")]; + tensor var_52206_equation_0 = const()[name = tensor("op_52206_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52206_cast_fp16 = einsum(equation = var_52206_equation_0, values = (var_51718_cast_fp16, var_52119_cast_fp16))[name = tensor("op_52206_cast_fp16")]; + tensor var_52208_equation_0 = const()[name = tensor("op_52208_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52208_cast_fp16 = einsum(equation = var_52208_equation_0, values = (var_51718_cast_fp16, var_52120_cast_fp16))[name = tensor("op_52208_cast_fp16")]; + tensor var_52210_equation_0 = const()[name = tensor("op_52210_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52210_cast_fp16 = einsum(equation = var_52210_equation_0, values = (var_51722_cast_fp16, var_52121_cast_fp16))[name = tensor("op_52210_cast_fp16")]; + tensor var_52212_equation_0 = const()[name = tensor("op_52212_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52212_cast_fp16 = einsum(equation = var_52212_equation_0, values = (var_51722_cast_fp16, var_52122_cast_fp16))[name = tensor("op_52212_cast_fp16")]; + tensor var_52214_equation_0 = const()[name = tensor("op_52214_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52214_cast_fp16 = einsum(equation = var_52214_equation_0, values = (var_51722_cast_fp16, var_52123_cast_fp16))[name = tensor("op_52214_cast_fp16")]; + tensor var_52216_equation_0 = const()[name = tensor("op_52216_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52216_cast_fp16 = einsum(equation = var_52216_equation_0, values = (var_51722_cast_fp16, var_52124_cast_fp16))[name = tensor("op_52216_cast_fp16")]; + tensor var_52218_equation_0 = const()[name = tensor("op_52218_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52218_cast_fp16 = einsum(equation = var_52218_equation_0, values = (var_51726_cast_fp16, var_52125_cast_fp16))[name = tensor("op_52218_cast_fp16")]; + tensor var_52220_equation_0 = const()[name = tensor("op_52220_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52220_cast_fp16 = einsum(equation = var_52220_equation_0, values = (var_51726_cast_fp16, var_52126_cast_fp16))[name = tensor("op_52220_cast_fp16")]; + tensor var_52222_equation_0 = const()[name = tensor("op_52222_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52222_cast_fp16 = einsum(equation = var_52222_equation_0, values = (var_51726_cast_fp16, var_52127_cast_fp16))[name = tensor("op_52222_cast_fp16")]; + tensor var_52224_equation_0 = const()[name = tensor("op_52224_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52224_cast_fp16 = einsum(equation = var_52224_equation_0, values = (var_51726_cast_fp16, var_52128_cast_fp16))[name = tensor("op_52224_cast_fp16")]; + tensor var_52226_equation_0 = const()[name = tensor("op_52226_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52226_cast_fp16 = einsum(equation = var_52226_equation_0, values = (var_51730_cast_fp16, var_52129_cast_fp16))[name = tensor("op_52226_cast_fp16")]; + tensor var_52228_equation_0 = const()[name = tensor("op_52228_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52228_cast_fp16 = einsum(equation = var_52228_equation_0, values = (var_51730_cast_fp16, var_52130_cast_fp16))[name = tensor("op_52228_cast_fp16")]; + tensor var_52230_equation_0 = const()[name = tensor("op_52230_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52230_cast_fp16 = einsum(equation = var_52230_equation_0, values = (var_51730_cast_fp16, var_52131_cast_fp16))[name = tensor("op_52230_cast_fp16")]; + tensor var_52232_equation_0 = const()[name = tensor("op_52232_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52232_cast_fp16 = einsum(equation = var_52232_equation_0, values = (var_51730_cast_fp16, var_52132_cast_fp16))[name = tensor("op_52232_cast_fp16")]; + tensor var_52234_equation_0 = const()[name = tensor("op_52234_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52234_cast_fp16 = einsum(equation = var_52234_equation_0, values = (var_51734_cast_fp16, var_52133_cast_fp16))[name = tensor("op_52234_cast_fp16")]; + tensor var_52236_equation_0 = const()[name = tensor("op_52236_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52236_cast_fp16 = einsum(equation = var_52236_equation_0, values = (var_51734_cast_fp16, var_52134_cast_fp16))[name = tensor("op_52236_cast_fp16")]; + tensor var_52238_equation_0 = const()[name = tensor("op_52238_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52238_cast_fp16 = einsum(equation = var_52238_equation_0, values = (var_51734_cast_fp16, var_52135_cast_fp16))[name = tensor("op_52238_cast_fp16")]; + tensor var_52240_equation_0 = const()[name = tensor("op_52240_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52240_cast_fp16 = einsum(equation = var_52240_equation_0, values = (var_51734_cast_fp16, var_52136_cast_fp16))[name = tensor("op_52240_cast_fp16")]; + tensor var_52242_equation_0 = const()[name = tensor("op_52242_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52242_cast_fp16 = einsum(equation = var_52242_equation_0, values = (var_51738_cast_fp16, var_52137_cast_fp16))[name = tensor("op_52242_cast_fp16")]; + tensor var_52244_equation_0 = const()[name = tensor("op_52244_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52244_cast_fp16 = einsum(equation = var_52244_equation_0, values = (var_51738_cast_fp16, var_52138_cast_fp16))[name = tensor("op_52244_cast_fp16")]; + tensor var_52246_equation_0 = const()[name = tensor("op_52246_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52246_cast_fp16 = einsum(equation = var_52246_equation_0, values = (var_51738_cast_fp16, var_52139_cast_fp16))[name = tensor("op_52246_cast_fp16")]; + tensor var_52248_equation_0 = const()[name = tensor("op_52248_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52248_cast_fp16 = einsum(equation = var_52248_equation_0, values = (var_51738_cast_fp16, var_52140_cast_fp16))[name = tensor("op_52248_cast_fp16")]; + tensor var_52250_equation_0 = const()[name = tensor("op_52250_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52250_cast_fp16 = einsum(equation = var_52250_equation_0, values = (var_51742_cast_fp16, var_52141_cast_fp16))[name = tensor("op_52250_cast_fp16")]; + tensor var_52252_equation_0 = const()[name = tensor("op_52252_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52252_cast_fp16 = einsum(equation = var_52252_equation_0, values = (var_51742_cast_fp16, var_52142_cast_fp16))[name = tensor("op_52252_cast_fp16")]; + tensor var_52254_equation_0 = const()[name = tensor("op_52254_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52254_cast_fp16 = einsum(equation = var_52254_equation_0, values = (var_51742_cast_fp16, var_52143_cast_fp16))[name = tensor("op_52254_cast_fp16")]; + tensor var_52256_equation_0 = const()[name = tensor("op_52256_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52256_cast_fp16 = einsum(equation = var_52256_equation_0, values = (var_51742_cast_fp16, var_52144_cast_fp16))[name = tensor("op_52256_cast_fp16")]; + tensor var_52258_equation_0 = const()[name = tensor("op_52258_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52258_cast_fp16 = einsum(equation = var_52258_equation_0, values = (var_51746_cast_fp16, var_52145_cast_fp16))[name = tensor("op_52258_cast_fp16")]; + tensor var_52260_equation_0 = const()[name = tensor("op_52260_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52260_cast_fp16 = einsum(equation = var_52260_equation_0, values = (var_51746_cast_fp16, var_52146_cast_fp16))[name = tensor("op_52260_cast_fp16")]; + tensor var_52262_equation_0 = const()[name = tensor("op_52262_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52262_cast_fp16 = einsum(equation = var_52262_equation_0, values = (var_51746_cast_fp16, var_52147_cast_fp16))[name = tensor("op_52262_cast_fp16")]; + tensor var_52264_equation_0 = const()[name = tensor("op_52264_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52264_cast_fp16 = einsum(equation = var_52264_equation_0, values = (var_51746_cast_fp16, var_52148_cast_fp16))[name = tensor("op_52264_cast_fp16")]; + tensor var_52266_equation_0 = const()[name = tensor("op_52266_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52266_cast_fp16 = einsum(equation = var_52266_equation_0, values = (var_51750_cast_fp16, var_52149_cast_fp16))[name = tensor("op_52266_cast_fp16")]; + tensor var_52268_equation_0 = const()[name = tensor("op_52268_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52268_cast_fp16 = einsum(equation = var_52268_equation_0, values = (var_51750_cast_fp16, var_52150_cast_fp16))[name = tensor("op_52268_cast_fp16")]; + tensor var_52270_equation_0 = const()[name = tensor("op_52270_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52270_cast_fp16 = einsum(equation = var_52270_equation_0, values = (var_51750_cast_fp16, var_52151_cast_fp16))[name = tensor("op_52270_cast_fp16")]; + tensor var_52272_equation_0 = const()[name = tensor("op_52272_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52272_cast_fp16 = einsum(equation = var_52272_equation_0, values = (var_51750_cast_fp16, var_52152_cast_fp16))[name = tensor("op_52272_cast_fp16")]; + tensor var_52274_equation_0 = const()[name = tensor("op_52274_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52274_cast_fp16 = einsum(equation = var_52274_equation_0, values = (var_51754_cast_fp16, var_52153_cast_fp16))[name = tensor("op_52274_cast_fp16")]; + tensor var_52276_equation_0 = const()[name = tensor("op_52276_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52276_cast_fp16 = einsum(equation = var_52276_equation_0, values = (var_51754_cast_fp16, var_52154_cast_fp16))[name = tensor("op_52276_cast_fp16")]; + tensor var_52278_equation_0 = const()[name = tensor("op_52278_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52278_cast_fp16 = einsum(equation = var_52278_equation_0, values = (var_51754_cast_fp16, var_52155_cast_fp16))[name = tensor("op_52278_cast_fp16")]; + tensor var_52280_equation_0 = const()[name = tensor("op_52280_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52280_cast_fp16 = einsum(equation = var_52280_equation_0, values = (var_51754_cast_fp16, var_52156_cast_fp16))[name = tensor("op_52280_cast_fp16")]; + tensor var_52282_equation_0 = const()[name = tensor("op_52282_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52282_cast_fp16 = einsum(equation = var_52282_equation_0, values = (var_51758_cast_fp16, var_52157_cast_fp16))[name = tensor("op_52282_cast_fp16")]; + tensor var_52284_equation_0 = const()[name = tensor("op_52284_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52284_cast_fp16 = einsum(equation = var_52284_equation_0, values = (var_51758_cast_fp16, var_52158_cast_fp16))[name = tensor("op_52284_cast_fp16")]; + tensor var_52286_equation_0 = const()[name = tensor("op_52286_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52286_cast_fp16 = einsum(equation = var_52286_equation_0, values = (var_51758_cast_fp16, var_52159_cast_fp16))[name = tensor("op_52286_cast_fp16")]; + tensor var_52288_equation_0 = const()[name = tensor("op_52288_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52288_cast_fp16 = einsum(equation = var_52288_equation_0, values = (var_51758_cast_fp16, var_52160_cast_fp16))[name = tensor("op_52288_cast_fp16")]; + tensor var_52290_equation_0 = const()[name = tensor("op_52290_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52290_cast_fp16 = einsum(equation = var_52290_equation_0, values = (var_51762_cast_fp16, var_52161_cast_fp16))[name = tensor("op_52290_cast_fp16")]; + tensor var_52292_equation_0 = const()[name = tensor("op_52292_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52292_cast_fp16 = einsum(equation = var_52292_equation_0, values = (var_51762_cast_fp16, var_52162_cast_fp16))[name = tensor("op_52292_cast_fp16")]; + tensor var_52294_equation_0 = const()[name = tensor("op_52294_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52294_cast_fp16 = einsum(equation = var_52294_equation_0, values = (var_51762_cast_fp16, var_52163_cast_fp16))[name = tensor("op_52294_cast_fp16")]; + tensor var_52296_equation_0 = const()[name = tensor("op_52296_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52296_cast_fp16 = einsum(equation = var_52296_equation_0, values = (var_51762_cast_fp16, var_52164_cast_fp16))[name = tensor("op_52296_cast_fp16")]; + tensor var_52298_equation_0 = const()[name = tensor("op_52298_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52298_cast_fp16 = einsum(equation = var_52298_equation_0, values = (var_51766_cast_fp16, var_52165_cast_fp16))[name = tensor("op_52298_cast_fp16")]; + tensor var_52300_equation_0 = const()[name = tensor("op_52300_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52300_cast_fp16 = einsum(equation = var_52300_equation_0, values = (var_51766_cast_fp16, var_52166_cast_fp16))[name = tensor("op_52300_cast_fp16")]; + tensor var_52302_equation_0 = const()[name = tensor("op_52302_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52302_cast_fp16 = einsum(equation = var_52302_equation_0, values = (var_51766_cast_fp16, var_52167_cast_fp16))[name = tensor("op_52302_cast_fp16")]; + tensor var_52304_equation_0 = const()[name = tensor("op_52304_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52304_cast_fp16 = einsum(equation = var_52304_equation_0, values = (var_51766_cast_fp16, var_52168_cast_fp16))[name = tensor("op_52304_cast_fp16")]; + tensor var_52306_equation_0 = const()[name = tensor("op_52306_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52306_cast_fp16 = einsum(equation = var_52306_equation_0, values = (var_51770_cast_fp16, var_52169_cast_fp16))[name = tensor("op_52306_cast_fp16")]; + tensor var_52308_equation_0 = const()[name = tensor("op_52308_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52308_cast_fp16 = einsum(equation = var_52308_equation_0, values = (var_51770_cast_fp16, var_52170_cast_fp16))[name = tensor("op_52308_cast_fp16")]; + tensor var_52310_equation_0 = const()[name = tensor("op_52310_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52310_cast_fp16 = einsum(equation = var_52310_equation_0, values = (var_51770_cast_fp16, var_52171_cast_fp16))[name = tensor("op_52310_cast_fp16")]; + tensor var_52312_equation_0 = const()[name = tensor("op_52312_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52312_cast_fp16 = einsum(equation = var_52312_equation_0, values = (var_51770_cast_fp16, var_52172_cast_fp16))[name = tensor("op_52312_cast_fp16")]; + tensor var_52314_equation_0 = const()[name = tensor("op_52314_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52314_cast_fp16 = einsum(equation = var_52314_equation_0, values = (var_51774_cast_fp16, var_52173_cast_fp16))[name = tensor("op_52314_cast_fp16")]; + tensor var_52316_equation_0 = const()[name = tensor("op_52316_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52316_cast_fp16 = einsum(equation = var_52316_equation_0, values = (var_51774_cast_fp16, var_52174_cast_fp16))[name = tensor("op_52316_cast_fp16")]; + tensor var_52318_equation_0 = const()[name = tensor("op_52318_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52318_cast_fp16 = einsum(equation = var_52318_equation_0, values = (var_51774_cast_fp16, var_52175_cast_fp16))[name = tensor("op_52318_cast_fp16")]; + tensor var_52320_equation_0 = const()[name = tensor("op_52320_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52320_cast_fp16 = einsum(equation = var_52320_equation_0, values = (var_51774_cast_fp16, var_52176_cast_fp16))[name = tensor("op_52320_cast_fp16")]; + tensor var_52322_equation_0 = const()[name = tensor("op_52322_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52322_cast_fp16 = einsum(equation = var_52322_equation_0, values = (var_51778_cast_fp16, var_52177_cast_fp16))[name = tensor("op_52322_cast_fp16")]; + tensor var_52324_equation_0 = const()[name = tensor("op_52324_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52324_cast_fp16 = einsum(equation = var_52324_equation_0, values = (var_51778_cast_fp16, var_52178_cast_fp16))[name = tensor("op_52324_cast_fp16")]; + tensor var_52326_equation_0 = const()[name = tensor("op_52326_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52326_cast_fp16 = einsum(equation = var_52326_equation_0, values = (var_51778_cast_fp16, var_52179_cast_fp16))[name = tensor("op_52326_cast_fp16")]; + tensor var_52328_equation_0 = const()[name = tensor("op_52328_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52328_cast_fp16 = einsum(equation = var_52328_equation_0, values = (var_51778_cast_fp16, var_52180_cast_fp16))[name = tensor("op_52328_cast_fp16")]; + tensor var_52330_equation_0 = const()[name = tensor("op_52330_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52330_cast_fp16 = einsum(equation = var_52330_equation_0, values = (var_51782_cast_fp16, var_52181_cast_fp16))[name = tensor("op_52330_cast_fp16")]; + tensor var_52332_equation_0 = const()[name = tensor("op_52332_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52332_cast_fp16 = einsum(equation = var_52332_equation_0, values = (var_51782_cast_fp16, var_52182_cast_fp16))[name = tensor("op_52332_cast_fp16")]; + tensor var_52334_equation_0 = const()[name = tensor("op_52334_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52334_cast_fp16 = einsum(equation = var_52334_equation_0, values = (var_51782_cast_fp16, var_52183_cast_fp16))[name = tensor("op_52334_cast_fp16")]; + tensor var_52336_equation_0 = const()[name = tensor("op_52336_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52336_cast_fp16 = einsum(equation = var_52336_equation_0, values = (var_51782_cast_fp16, var_52184_cast_fp16))[name = tensor("op_52336_cast_fp16")]; + tensor var_52338_equation_0 = const()[name = tensor("op_52338_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52338_cast_fp16 = einsum(equation = var_52338_equation_0, values = (var_51786_cast_fp16, var_52185_cast_fp16))[name = tensor("op_52338_cast_fp16")]; + tensor var_52340_equation_0 = const()[name = tensor("op_52340_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52340_cast_fp16 = einsum(equation = var_52340_equation_0, values = (var_51786_cast_fp16, var_52186_cast_fp16))[name = tensor("op_52340_cast_fp16")]; + tensor var_52342_equation_0 = const()[name = tensor("op_52342_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52342_cast_fp16 = einsum(equation = var_52342_equation_0, values = (var_51786_cast_fp16, var_52187_cast_fp16))[name = tensor("op_52342_cast_fp16")]; + tensor var_52344_equation_0 = const()[name = tensor("op_52344_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52344_cast_fp16 = einsum(equation = var_52344_equation_0, values = (var_51786_cast_fp16, var_52188_cast_fp16))[name = tensor("op_52344_cast_fp16")]; + tensor var_52346_equation_0 = const()[name = tensor("op_52346_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52346_cast_fp16 = einsum(equation = var_52346_equation_0, values = (var_51790_cast_fp16, var_52189_cast_fp16))[name = tensor("op_52346_cast_fp16")]; + tensor var_52348_equation_0 = const()[name = tensor("op_52348_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52348_cast_fp16 = einsum(equation = var_52348_equation_0, values = (var_51790_cast_fp16, var_52190_cast_fp16))[name = tensor("op_52348_cast_fp16")]; + tensor var_52350_equation_0 = const()[name = tensor("op_52350_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52350_cast_fp16 = einsum(equation = var_52350_equation_0, values = (var_51790_cast_fp16, var_52191_cast_fp16))[name = tensor("op_52350_cast_fp16")]; + tensor var_52352_equation_0 = const()[name = tensor("op_52352_equation_0"), val = tensor("bchk,bkhq->bchq")]; + tensor var_52352_cast_fp16 = einsum(equation = var_52352_equation_0, values = (var_51790_cast_fp16, var_52192_cast_fp16))[name = tensor("op_52352_cast_fp16")]; + tensor var_52354_interleave_0 = const()[name = tensor("op_52354_interleave_0"), val = tensor(false)]; + tensor var_52354_cast_fp16 = concat(axis = var_50859, interleave = var_52354_interleave_0, values = (var_52194_cast_fp16, var_52196_cast_fp16, var_52198_cast_fp16, var_52200_cast_fp16))[name = tensor("op_52354_cast_fp16")]; + tensor var_52356_interleave_0 = const()[name = tensor("op_52356_interleave_0"), val = tensor(false)]; + tensor var_52356_cast_fp16 = concat(axis = var_50859, interleave = var_52356_interleave_0, values = (var_52202_cast_fp16, var_52204_cast_fp16, var_52206_cast_fp16, var_52208_cast_fp16))[name = tensor("op_52356_cast_fp16")]; + tensor var_52358_interleave_0 = const()[name = tensor("op_52358_interleave_0"), val = tensor(false)]; + tensor var_52358_cast_fp16 = concat(axis = var_50859, interleave = var_52358_interleave_0, values = (var_52210_cast_fp16, var_52212_cast_fp16, var_52214_cast_fp16, var_52216_cast_fp16))[name = tensor("op_52358_cast_fp16")]; + tensor var_52360_interleave_0 = const()[name = tensor("op_52360_interleave_0"), val = tensor(false)]; + tensor var_52360_cast_fp16 = concat(axis = var_50859, interleave = var_52360_interleave_0, values = (var_52218_cast_fp16, var_52220_cast_fp16, var_52222_cast_fp16, var_52224_cast_fp16))[name = tensor("op_52360_cast_fp16")]; + tensor var_52362_interleave_0 = const()[name = tensor("op_52362_interleave_0"), val = tensor(false)]; + tensor var_52362_cast_fp16 = concat(axis = var_50859, interleave = var_52362_interleave_0, values = (var_52226_cast_fp16, var_52228_cast_fp16, var_52230_cast_fp16, var_52232_cast_fp16))[name = tensor("op_52362_cast_fp16")]; + tensor var_52364_interleave_0 = const()[name = tensor("op_52364_interleave_0"), val = tensor(false)]; + tensor var_52364_cast_fp16 = concat(axis = var_50859, interleave = var_52364_interleave_0, values = (var_52234_cast_fp16, var_52236_cast_fp16, var_52238_cast_fp16, var_52240_cast_fp16))[name = tensor("op_52364_cast_fp16")]; + tensor var_52366_interleave_0 = const()[name = tensor("op_52366_interleave_0"), val = tensor(false)]; + tensor var_52366_cast_fp16 = concat(axis = var_50859, interleave = var_52366_interleave_0, values = (var_52242_cast_fp16, var_52244_cast_fp16, var_52246_cast_fp16, var_52248_cast_fp16))[name = tensor("op_52366_cast_fp16")]; + tensor var_52368_interleave_0 = const()[name = tensor("op_52368_interleave_0"), val = tensor(false)]; + tensor var_52368_cast_fp16 = concat(axis = var_50859, interleave = var_52368_interleave_0, values = (var_52250_cast_fp16, var_52252_cast_fp16, var_52254_cast_fp16, var_52256_cast_fp16))[name = tensor("op_52368_cast_fp16")]; + tensor var_52370_interleave_0 = const()[name = tensor("op_52370_interleave_0"), val = tensor(false)]; + tensor var_52370_cast_fp16 = concat(axis = var_50859, interleave = var_52370_interleave_0, values = (var_52258_cast_fp16, var_52260_cast_fp16, var_52262_cast_fp16, var_52264_cast_fp16))[name = tensor("op_52370_cast_fp16")]; + tensor var_52372_interleave_0 = const()[name = tensor("op_52372_interleave_0"), val = tensor(false)]; + tensor var_52372_cast_fp16 = concat(axis = var_50859, interleave = var_52372_interleave_0, values = (var_52266_cast_fp16, var_52268_cast_fp16, var_52270_cast_fp16, var_52272_cast_fp16))[name = tensor("op_52372_cast_fp16")]; + tensor var_52374_interleave_0 = const()[name = tensor("op_52374_interleave_0"), val = tensor(false)]; + tensor var_52374_cast_fp16 = concat(axis = var_50859, interleave = var_52374_interleave_0, values = (var_52274_cast_fp16, var_52276_cast_fp16, var_52278_cast_fp16, var_52280_cast_fp16))[name = tensor("op_52374_cast_fp16")]; + tensor var_52376_interleave_0 = const()[name = tensor("op_52376_interleave_0"), val = tensor(false)]; + tensor var_52376_cast_fp16 = concat(axis = var_50859, interleave = var_52376_interleave_0, values = (var_52282_cast_fp16, var_52284_cast_fp16, var_52286_cast_fp16, var_52288_cast_fp16))[name = tensor("op_52376_cast_fp16")]; + tensor var_52378_interleave_0 = const()[name = tensor("op_52378_interleave_0"), val = tensor(false)]; + tensor var_52378_cast_fp16 = concat(axis = var_50859, interleave = var_52378_interleave_0, values = (var_52290_cast_fp16, var_52292_cast_fp16, var_52294_cast_fp16, var_52296_cast_fp16))[name = tensor("op_52378_cast_fp16")]; + tensor var_52380_interleave_0 = const()[name = tensor("op_52380_interleave_0"), val = tensor(false)]; + tensor var_52380_cast_fp16 = concat(axis = var_50859, interleave = var_52380_interleave_0, values = (var_52298_cast_fp16, var_52300_cast_fp16, var_52302_cast_fp16, var_52304_cast_fp16))[name = tensor("op_52380_cast_fp16")]; + tensor var_52382_interleave_0 = const()[name = tensor("op_52382_interleave_0"), val = tensor(false)]; + tensor var_52382_cast_fp16 = concat(axis = var_50859, interleave = var_52382_interleave_0, values = (var_52306_cast_fp16, var_52308_cast_fp16, var_52310_cast_fp16, var_52312_cast_fp16))[name = tensor("op_52382_cast_fp16")]; + tensor var_52384_interleave_0 = const()[name = tensor("op_52384_interleave_0"), val = tensor(false)]; + tensor var_52384_cast_fp16 = concat(axis = var_50859, interleave = var_52384_interleave_0, values = (var_52314_cast_fp16, var_52316_cast_fp16, var_52318_cast_fp16, var_52320_cast_fp16))[name = tensor("op_52384_cast_fp16")]; + tensor var_52386_interleave_0 = const()[name = tensor("op_52386_interleave_0"), val = tensor(false)]; + tensor var_52386_cast_fp16 = concat(axis = var_50859, interleave = var_52386_interleave_0, values = (var_52322_cast_fp16, var_52324_cast_fp16, var_52326_cast_fp16, var_52328_cast_fp16))[name = tensor("op_52386_cast_fp16")]; + tensor var_52388_interleave_0 = const()[name = tensor("op_52388_interleave_0"), val = tensor(false)]; + tensor var_52388_cast_fp16 = concat(axis = var_50859, interleave = var_52388_interleave_0, values = (var_52330_cast_fp16, var_52332_cast_fp16, var_52334_cast_fp16, var_52336_cast_fp16))[name = tensor("op_52388_cast_fp16")]; + tensor var_52390_interleave_0 = const()[name = tensor("op_52390_interleave_0"), val = tensor(false)]; + tensor var_52390_cast_fp16 = concat(axis = var_50859, interleave = var_52390_interleave_0, values = (var_52338_cast_fp16, var_52340_cast_fp16, var_52342_cast_fp16, var_52344_cast_fp16))[name = tensor("op_52390_cast_fp16")]; + tensor var_52392_interleave_0 = const()[name = tensor("op_52392_interleave_0"), val = tensor(false)]; + tensor var_52392_cast_fp16 = concat(axis = var_50859, interleave = var_52392_interleave_0, values = (var_52346_cast_fp16, var_52348_cast_fp16, var_52350_cast_fp16, var_52352_cast_fp16))[name = tensor("op_52392_cast_fp16")]; + tensor input_627_interleave_0 = const()[name = tensor("input_627_interleave_0"), val = tensor(false)]; + tensor input_627_cast_fp16 = concat(axis = var_50884, interleave = input_627_interleave_0, values = (var_52354_cast_fp16, var_52356_cast_fp16, var_52358_cast_fp16, var_52360_cast_fp16, var_52362_cast_fp16, var_52364_cast_fp16, var_52366_cast_fp16, var_52368_cast_fp16, var_52370_cast_fp16, var_52372_cast_fp16, var_52374_cast_fp16, var_52376_cast_fp16, var_52378_cast_fp16, var_52380_cast_fp16, var_52382_cast_fp16, var_52384_cast_fp16, var_52386_cast_fp16, var_52388_cast_fp16, var_52390_cast_fp16, var_52392_cast_fp16))[name = tensor("input_627_cast_fp16")]; + tensor var_52400 = const()[name = tensor("op_52400"), val = tensor([1, 1])]; + tensor var_52402 = const()[name = tensor("op_52402"), val = tensor([1, 1])]; + tensor pretrained_out_379_pad_type_0 = const()[name = tensor("pretrained_out_379_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_379_pad_0 = const()[name = tensor("pretrained_out_379_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_o_proj_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345648448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346467712))), name = tensor("layers_31_self_attn_o_proj_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 1280, 1, 1])]; + tensor layers_31_self_attn_o_proj_pretrained_bias_to_fp16 = const()[name = tensor("layers_31_self_attn_o_proj_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346467840)))]; + tensor pretrained_out_379_cast_fp16 = conv(bias = layers_31_self_attn_o_proj_pretrained_bias_to_fp16, dilations = var_52402, groups = var_50884, pad = pretrained_out_379_pad_0, pad_type = pretrained_out_379_pad_type_0, strides = var_52400, weight = layers_31_self_attn_o_proj_pretrained_weight_to_fp16_palettized, x = input_627_cast_fp16)[name = tensor("pretrained_out_379_cast_fp16")]; + tensor var_52406 = const()[name = tensor("op_52406"), val = tensor([1, 1])]; + tensor var_52408 = const()[name = tensor("op_52408"), val = tensor([1, 1])]; + tensor input_629_pad_type_0 = const()[name = tensor("input_629_pad_type_0"), val = tensor("custom")]; + tensor input_629_pad_0 = const()[name = tensor("input_629_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_self_attn_o_proj_loraA_weight_to_fp16 = const()[name = tensor("layers_31_self_attn_o_proj_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346470464)))]; + tensor input_629_cast_fp16 = conv(dilations = var_52408, groups = var_50884, pad = input_629_pad_0, pad_type = input_629_pad_type_0, strides = var_52406, weight = layers_31_self_attn_o_proj_loraA_weight_to_fp16, x = input_627_cast_fp16)[name = tensor("input_629_cast_fp16")]; + tensor var_52412 = const()[name = tensor("op_52412"), val = tensor([1, 1])]; + tensor var_52414 = const()[name = tensor("op_52414"), val = tensor([1, 1])]; + tensor lora_out_757_pad_type_0 = const()[name = tensor("lora_out_757_pad_type_0"), val = tensor("custom")]; + tensor lora_out_757_pad_0 = const()[name = tensor("lora_out_757_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_759_weight_0_to_fp16 = const()[name = tensor("lora_out_759_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346511488)))]; + tensor lora_out_759_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_52414, groups = var_50884, pad = lora_out_757_pad_0, pad_type = lora_out_757_pad_type_0, strides = var_52412, weight = lora_out_759_weight_0_to_fp16, x = input_629_cast_fp16)[name = tensor("lora_out_759_cast_fp16")]; + tensor obj_cast_fp16 = add(x = pretrained_out_379_cast_fp16, y = lora_out_759_cast_fp16)[name = tensor("obj_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = obj_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; + tensor var_52423 = const()[name = tensor("op_52423"), val = tensor([1])]; + tensor channels_mean_127_cast_fp16 = reduce_mean(axes = var_52423, keep_dims = var_50885, x = inputs_127_cast_fp16)[name = tensor("channels_mean_127_cast_fp16")]; + tensor zero_mean_127_cast_fp16 = sub(x = inputs_127_cast_fp16, y = channels_mean_127_cast_fp16)[name = tensor("zero_mean_127_cast_fp16")]; + tensor zero_mean_sq_127_cast_fp16 = mul(x = zero_mean_127_cast_fp16, y = zero_mean_127_cast_fp16)[name = tensor("zero_mean_sq_127_cast_fp16")]; + tensor var_52427 = const()[name = tensor("op_52427"), val = tensor([1])]; + tensor var_52428_cast_fp16 = reduce_mean(axes = var_52427, keep_dims = var_50885, x = zero_mean_sq_127_cast_fp16)[name = tensor("op_52428_cast_fp16")]; + tensor var_52429_to_fp16 = const()[name = tensor("op_52429_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_52430_cast_fp16 = add(x = var_52428_cast_fp16, y = var_52429_to_fp16)[name = tensor("op_52430_cast_fp16")]; + tensor denom_127_epsilon_0 = const()[name = tensor("denom_127_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_127_cast_fp16 = rsqrt(epsilon = denom_127_epsilon_0, x = var_52430_cast_fp16)[name = tensor("denom_127_cast_fp16")]; + tensor out_127_cast_fp16 = mul(x = zero_mean_127_cast_fp16, y = denom_127_cast_fp16)[name = tensor("out_127_cast_fp16")]; + tensor input_631_gamma_0_to_fp16 = const()[name = tensor("input_631_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346552512)))]; + tensor input_631_beta_0_to_fp16 = const()[name = tensor("input_631_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346555136)))]; + tensor input_631_epsilon_0_to_fp16 = const()[name = tensor("input_631_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_631_cast_fp16 = batch_norm(beta = input_631_beta_0_to_fp16, epsilon = input_631_epsilon_0_to_fp16, gamma = input_631_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_127_cast_fp16)[name = tensor("input_631_cast_fp16")]; + tensor var_52444 = const()[name = tensor("op_52444"), val = tensor([1, 1])]; + tensor var_52446 = const()[name = tensor("op_52446"), val = tensor([1, 1])]; + tensor pretrained_out_381_pad_type_0 = const()[name = tensor("pretrained_out_381_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_381_pad_0 = const()[name = tensor("pretrained_out_381_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_fc1_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346557760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349834624))), name = tensor("layers_31_fc1_pretrained_weight_to_fp16_palettized"), shape = tensor([5120, 1280, 1, 1])]; + tensor layers_31_fc1_pretrained_bias_to_fp16 = const()[name = tensor("layers_31_fc1_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349834752)))]; + tensor pretrained_out_381_cast_fp16 = conv(bias = layers_31_fc1_pretrained_bias_to_fp16, dilations = var_52446, groups = var_50884, pad = pretrained_out_381_pad_0, pad_type = pretrained_out_381_pad_type_0, strides = var_52444, weight = layers_31_fc1_pretrained_weight_to_fp16_palettized, x = input_631_cast_fp16)[name = tensor("pretrained_out_381_cast_fp16")]; + tensor var_52450 = const()[name = tensor("op_52450"), val = tensor([1, 1])]; + tensor var_52452 = const()[name = tensor("op_52452"), val = tensor([1, 1])]; + tensor input_633_pad_type_0 = const()[name = tensor("input_633_pad_type_0"), val = tensor("custom")]; + tensor input_633_pad_0 = const()[name = tensor("input_633_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_fc1_loraA_weight_to_fp16 = const()[name = tensor("layers_31_fc1_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349845056)))]; + tensor input_633_cast_fp16 = conv(dilations = var_52452, groups = var_50884, pad = input_633_pad_0, pad_type = input_633_pad_type_0, strides = var_52450, weight = layers_31_fc1_loraA_weight_to_fp16, x = input_631_cast_fp16)[name = tensor("input_633_cast_fp16")]; + tensor var_52456 = const()[name = tensor("op_52456"), val = tensor([1, 1])]; + tensor var_52458 = const()[name = tensor("op_52458"), val = tensor([1, 1])]; + tensor lora_out_761_pad_type_0 = const()[name = tensor("lora_out_761_pad_type_0"), val = tensor("custom")]; + tensor lora_out_761_pad_0 = const()[name = tensor("lora_out_761_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_763_weight_0_to_fp16 = const()[name = tensor("lora_out_763_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349886080)))]; + tensor lora_out_763_cast_fp16 = conv(bias = lora_out_19_bias_0_to_fp16, dilations = var_52458, groups = var_50884, pad = lora_out_761_pad_0, pad_type = lora_out_761_pad_type_0, strides = var_52456, weight = lora_out_763_weight_0_to_fp16, x = input_633_cast_fp16)[name = tensor("lora_out_763_cast_fp16")]; + tensor input_635_cast_fp16 = add(x = pretrained_out_381_cast_fp16, y = lora_out_763_cast_fp16)[name = tensor("input_635_cast_fp16")]; + tensor input_637_mode_0 = const()[name = tensor("input_637_mode_0"), val = tensor("EXACT")]; + tensor input_637_cast_fp16 = gelu(mode = input_637_mode_0, x = input_635_cast_fp16)[name = tensor("input_637_cast_fp16")]; + tensor var_52470 = const()[name = tensor("op_52470"), val = tensor([1, 1])]; + tensor var_52472 = const()[name = tensor("op_52472"), val = tensor([1, 1])]; + tensor pretrained_out_pad_type_0 = const()[name = tensor("pretrained_out_pad_type_0"), val = tensor("custom")]; + tensor pretrained_out_pad_0 = const()[name = tensor("pretrained_out_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_fc2_pretrained_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350049984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353326848))), name = tensor("layers_31_fc2_pretrained_weight_to_fp16_palettized"), shape = tensor([1280, 5120, 1, 1])]; + tensor layers_31_fc2_pretrained_bias_to_fp16 = const()[name = tensor("layers_31_fc2_pretrained_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353326976)))]; + tensor pretrained_out_cast_fp16 = conv(bias = layers_31_fc2_pretrained_bias_to_fp16, dilations = var_52472, groups = var_50884, pad = pretrained_out_pad_0, pad_type = pretrained_out_pad_type_0, strides = var_52470, weight = layers_31_fc2_pretrained_weight_to_fp16_palettized, x = input_637_cast_fp16)[name = tensor("pretrained_out_cast_fp16")]; + tensor var_52476 = const()[name = tensor("op_52476"), val = tensor([1, 1])]; + tensor var_52478 = const()[name = tensor("op_52478"), val = tensor([1, 1])]; + tensor input_pad_type_0 = const()[name = tensor("input_pad_type_0"), val = tensor("custom")]; + tensor input_pad_0 = const()[name = tensor("input_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor layers_31_fc2_loraA_weight_to_fp16 = const()[name = tensor("layers_31_fc2_loraA_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353329600)))]; + tensor input_cast_fp16 = conv(dilations = var_52478, groups = var_50884, pad = input_pad_0, pad_type = input_pad_type_0, strides = var_52476, weight = layers_31_fc2_loraA_weight_to_fp16, x = input_637_cast_fp16)[name = tensor("input_cast_fp16")]; + tensor var_52482 = const()[name = tensor("op_52482"), val = tensor([1, 1])]; + tensor var_52484 = const()[name = tensor("op_52484"), val = tensor([1, 1])]; + tensor lora_out_765_pad_type_0 = const()[name = tensor("lora_out_765_pad_type_0"), val = tensor("custom")]; + tensor lora_out_765_pad_0 = const()[name = tensor("lora_out_765_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor lora_out_weight_0_to_fp16 = const()[name = tensor("lora_out_weight_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353493504)))]; + tensor lora_out_cast_fp16 = conv(bias = obj_1_mean_0_to_fp16, dilations = var_52484, groups = var_50884, pad = lora_out_765_pad_0, pad_type = lora_out_765_pad_type_0, strides = var_52482, weight = lora_out_weight_0_to_fp16, x = input_cast_fp16)[name = tensor("lora_out_cast_fp16")]; + tensor hidden_states_cast_fp16 = add(x = pretrained_out_cast_fp16, y = lora_out_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_127_cast_fp16, y = hidden_states_cast_fp16)[name = tensor("inputs_cast_fp16")]; + tensor var_52493 = const()[name = tensor("op_52493"), val = tensor(true)]; + tensor var_52497 = const()[name = tensor("op_52497"), val = tensor([1])]; + tensor channels_mean_cast_fp16 = reduce_mean(axes = var_52497, keep_dims = var_52493, x = inputs_cast_fp16)[name = tensor("channels_mean_cast_fp16")]; + tensor zero_mean_cast_fp16 = sub(x = inputs_cast_fp16, y = channels_mean_cast_fp16)[name = tensor("zero_mean_cast_fp16")]; + tensor zero_mean_sq_cast_fp16 = mul(x = zero_mean_cast_fp16, y = zero_mean_cast_fp16)[name = tensor("zero_mean_sq_cast_fp16")]; + tensor var_52501 = const()[name = tensor("op_52501"), val = tensor([1])]; + tensor var_52502_cast_fp16 = reduce_mean(axes = var_52501, keep_dims = var_52493, x = zero_mean_sq_cast_fp16)[name = tensor("op_52502_cast_fp16")]; + tensor var_52503_to_fp16 = const()[name = tensor("op_52503_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_52504_cast_fp16 = add(x = var_52502_cast_fp16, y = var_52503_to_fp16)[name = tensor("op_52504_cast_fp16")]; + tensor denom_epsilon_0 = const()[name = tensor("denom_epsilon_0"), val = tensor(0x1.197998p-40)]; + tensor denom_cast_fp16 = rsqrt(epsilon = denom_epsilon_0, x = var_52504_cast_fp16)[name = tensor("denom_cast_fp16")]; + tensor out_cast_fp16 = mul(x = zero_mean_cast_fp16, y = denom_cast_fp16)[name = tensor("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353534528)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353537152)))]; + tensor encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor("encoder_output_embeds_type_fp32_cast_fp16")]; + } -> (encoder_output_embeds); +} \ No newline at end of file