program(1.0) [buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "2.2.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})] { func main(tensor melspectrogram_features) { tensor var_50 = const()[name = tensor("op_50"), val = tensor([1, 1])]; tensor var_56 = const()[name = tensor("op_56"), val = tensor([1, 1])]; tensor var_61 = const()[name = tensor("op_61"), val = tensor(1)]; tensor var_66_pad_type_0 = const()[name = tensor("op_66_pad_type_0"), val = tensor("custom")]; tensor var_66_pad_0 = const()[name = tensor("op_66_pad_0"), val = tensor([0, 0, 1, 1])]; tensor var_41_to_fp16 = const()[name = tensor("op_41_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; tensor var_47_to_fp16 = const()[name = tensor("op_47_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368768)))]; tensor var_66_cast_fp16 = conv(bias = var_47_to_fp16, dilations = var_56, groups = var_61, pad = var_66_pad_0, pad_type = var_66_pad_type_0, strides = var_50, weight = var_41_to_fp16, x = melspectrogram_features)[name = tensor("op_66_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_66_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; tensor var_90 = const()[name = tensor("op_90"), val = tensor([2, 2])]; 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(370368)))]; tensor var_87_to_fp16 = const()[name = tensor("op_87_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3909376)))]; 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 = hidden_states_1_cast_fp16)[name = tensor("op_106_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_106_cast_fp16)[name = tensor("hidden_states_3_cast_fp16")]; tensor var_124_to_fp16 = const()[name = tensor("op_124_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3910976)))]; tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_124_to_fp16)[name = tensor("inputs_1_cast_fp16")]; tensor var_134 = const()[name = tensor("op_134"), val = tensor(3)]; tensor var_151 = const()[name = tensor("op_151"), val = tensor(1)]; tensor var_152 = const()[name = tensor("op_152"), val = tensor(true)]; tensor var_162 = const()[name = tensor("op_162"), val = tensor([1])]; tensor channels_mean_1_cast_fp16 = reduce_mean(axes = var_162, keep_dims = var_152, 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_166 = const()[name = tensor("op_166"), val = tensor([1])]; tensor var_167_cast_fp16 = reduce_mean(axes = var_166, keep_dims = var_152, x = zero_mean_sq_1_cast_fp16)[name = tensor("op_167_cast_fp16")]; tensor var_168_to_fp16 = const()[name = tensor("op_168_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_169_cast_fp16 = add(x = var_167_cast_fp16, y = var_168_to_fp16)[name = tensor("op_169_cast_fp16")]; tensor denom_1_epsilon_0_to_fp16 = const()[name = tensor("denom_1_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_1_cast_fp16 = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_169_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(6215040)))]; 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(6216640)))]; 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(6218240)))]; 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(6219840)))]; 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_184 = const()[name = tensor("op_184"), val = tensor([1, 1])]; tensor var_186 = const()[name = tensor("op_186"), val = tensor([1, 1])]; tensor query_1_pad_type_0 = const()[name = tensor("query_1_pad_type_0"), val = tensor("custom")]; tensor query_1_pad_0 = const()[name = tensor("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6221440)))]; tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7401152)))]; tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = var_186, groups = var_151, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = var_184, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("query_1_cast_fp16")]; tensor var_190 = const()[name = tensor("op_190"), val = tensor([1, 1])]; tensor var_192 = const()[name = tensor("op_192"), val = tensor([1, 1])]; tensor key_1_pad_type_0 = const()[name = tensor("key_1_pad_type_0"), val = tensor("custom")]; tensor key_1_pad_0 = const()[name = tensor("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7402752)))]; tensor key_1_cast_fp16 = conv(dilations = var_192, groups = var_151, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = var_190, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("key_1_cast_fp16")]; tensor var_197 = const()[name = tensor("op_197"), val = tensor([1, 1])]; tensor var_199 = const()[name = tensor("op_199"), val = tensor([1, 1])]; tensor value_1_pad_type_0 = const()[name = tensor("value_1_pad_type_0"), val = tensor("custom")]; tensor value_1_pad_0 = const()[name = tensor("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8582464)))]; tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9762176)))]; tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = var_199, groups = var_151, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = var_197, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor("value_1_cast_fp16")]; tensor var_206_begin_0 = const()[name = tensor("op_206_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_206_end_0 = const()[name = tensor("op_206_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_206_end_mask_0 = const()[name = tensor("op_206_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_206_cast_fp16 = slice_by_index(begin = var_206_begin_0, end = var_206_end_0, end_mask = var_206_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_206_cast_fp16")]; tensor var_210_begin_0 = const()[name = tensor("op_210_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_210_end_0 = const()[name = tensor("op_210_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_210_end_mask_0 = const()[name = tensor("op_210_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_210_cast_fp16 = slice_by_index(begin = var_210_begin_0, end = var_210_end_0, end_mask = var_210_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_210_cast_fp16")]; tensor var_214_begin_0 = const()[name = tensor("op_214_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_214_end_0 = const()[name = tensor("op_214_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_214_end_mask_0 = const()[name = tensor("op_214_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_214_cast_fp16 = slice_by_index(begin = var_214_begin_0, end = var_214_end_0, end_mask = var_214_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_214_cast_fp16")]; tensor var_218_begin_0 = const()[name = tensor("op_218_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_218_end_0 = const()[name = tensor("op_218_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_218_end_mask_0 = const()[name = tensor("op_218_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_218_cast_fp16 = slice_by_index(begin = var_218_begin_0, end = var_218_end_0, end_mask = var_218_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_218_cast_fp16")]; tensor var_222_begin_0 = const()[name = tensor("op_222_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_222_end_0 = const()[name = tensor("op_222_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_222_end_mask_0 = const()[name = tensor("op_222_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_222_cast_fp16 = slice_by_index(begin = var_222_begin_0, end = var_222_end_0, end_mask = var_222_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_222_cast_fp16")]; tensor var_226_begin_0 = const()[name = tensor("op_226_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_226_end_0 = const()[name = tensor("op_226_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_226_end_mask_0 = const()[name = tensor("op_226_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_226_cast_fp16")]; tensor var_230_begin_0 = const()[name = tensor("op_230_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_230_end_0 = const()[name = tensor("op_230_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_230_end_mask_0 = const()[name = tensor("op_230_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_230_cast_fp16")]; tensor var_234_begin_0 = const()[name = tensor("op_234_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_234_end_0 = const()[name = tensor("op_234_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_234_end_mask_0 = const()[name = tensor("op_234_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_234_cast_fp16 = slice_by_index(begin = var_234_begin_0, end = var_234_end_0, end_mask = var_234_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_234_cast_fp16")]; tensor var_238_begin_0 = const()[name = tensor("op_238_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_238_end_0 = const()[name = tensor("op_238_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_238_end_mask_0 = const()[name = tensor("op_238_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_238_cast_fp16 = slice_by_index(begin = var_238_begin_0, end = var_238_end_0, end_mask = var_238_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_238_cast_fp16")]; tensor var_242_begin_0 = const()[name = tensor("op_242_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_242_end_0 = const()[name = tensor("op_242_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_242_end_mask_0 = const()[name = tensor("op_242_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_242_cast_fp16")]; tensor var_246_begin_0 = const()[name = tensor("op_246_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_246_end_0 = const()[name = tensor("op_246_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_246_end_mask_0 = const()[name = tensor("op_246_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_246_cast_fp16")]; tensor var_250_begin_0 = const()[name = tensor("op_250_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_250_end_0 = const()[name = tensor("op_250_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_250_end_mask_0 = const()[name = tensor("op_250_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_250_cast_fp16 = slice_by_index(begin = var_250_begin_0, end = var_250_end_0, end_mask = var_250_end_mask_0, x = query_1_cast_fp16)[name = tensor("op_250_cast_fp16")]; tensor var_259_begin_0 = const()[name = tensor("op_259_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_259_end_0 = const()[name = tensor("op_259_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_259_end_mask_0 = const()[name = tensor("op_259_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_259_cast_fp16 = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = var_206_cast_fp16)[name = tensor("op_259_cast_fp16")]; tensor var_266_begin_0 = const()[name = tensor("op_266_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_266_end_0 = const()[name = tensor("op_266_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_266_end_mask_0 = const()[name = tensor("op_266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_266_cast_fp16 = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = var_206_cast_fp16)[name = tensor("op_266_cast_fp16")]; tensor var_273_begin_0 = const()[name = tensor("op_273_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_273_end_0 = const()[name = tensor("op_273_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_273_end_mask_0 = const()[name = tensor("op_273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_273_cast_fp16 = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = var_206_cast_fp16)[name = tensor("op_273_cast_fp16")]; tensor var_280_begin_0 = const()[name = tensor("op_280_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_280_end_0 = const()[name = tensor("op_280_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_280_end_mask_0 = const()[name = tensor("op_280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_280_cast_fp16 = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = var_206_cast_fp16)[name = tensor("op_280_cast_fp16")]; tensor var_287_begin_0 = const()[name = tensor("op_287_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_287_end_0 = const()[name = tensor("op_287_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_287_end_mask_0 = const()[name = tensor("op_287_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_287_cast_fp16 = slice_by_index(begin = var_287_begin_0, end = var_287_end_0, end_mask = var_287_end_mask_0, x = var_210_cast_fp16)[name = tensor("op_287_cast_fp16")]; tensor var_294_begin_0 = const()[name = tensor("op_294_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_294_end_0 = const()[name = tensor("op_294_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_294_end_mask_0 = const()[name = tensor("op_294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_294_cast_fp16 = slice_by_index(begin = var_294_begin_0, end = var_294_end_0, end_mask = var_294_end_mask_0, x = var_210_cast_fp16)[name = tensor("op_294_cast_fp16")]; tensor var_301_begin_0 = const()[name = tensor("op_301_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_301_end_0 = const()[name = tensor("op_301_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_301_end_mask_0 = const()[name = tensor("op_301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_301_cast_fp16 = slice_by_index(begin = var_301_begin_0, end = var_301_end_0, end_mask = var_301_end_mask_0, x = var_210_cast_fp16)[name = tensor("op_301_cast_fp16")]; tensor var_308_begin_0 = const()[name = tensor("op_308_begin_0"), val = tensor([0, 0, 0, 1125])]; 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, true, true, false])]; 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 = var_210_cast_fp16)[name = tensor("op_308_cast_fp16")]; tensor var_315_begin_0 = const()[name = tensor("op_315_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_315_end_0 = const()[name = tensor("op_315_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_315_end_mask_0 = const()[name = tensor("op_315_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_315_cast_fp16 = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = var_214_cast_fp16)[name = tensor("op_315_cast_fp16")]; tensor var_322_begin_0 = const()[name = tensor("op_322_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_322_end_0 = const()[name = tensor("op_322_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_322_end_mask_0 = const()[name = tensor("op_322_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_322_cast_fp16 = slice_by_index(begin = var_322_begin_0, end = var_322_end_0, end_mask = var_322_end_mask_0, x = var_214_cast_fp16)[name = tensor("op_322_cast_fp16")]; tensor var_329_begin_0 = const()[name = tensor("op_329_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_329_end_0 = const()[name = tensor("op_329_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_329_end_mask_0 = const()[name = tensor("op_329_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_329_cast_fp16 = slice_by_index(begin = var_329_begin_0, end = var_329_end_0, end_mask = var_329_end_mask_0, x = var_214_cast_fp16)[name = tensor("op_329_cast_fp16")]; tensor var_336_begin_0 = const()[name = tensor("op_336_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_336_end_0 = const()[name = tensor("op_336_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_336_end_mask_0 = const()[name = tensor("op_336_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_214_cast_fp16)[name = tensor("op_336_cast_fp16")]; tensor var_343_begin_0 = const()[name = tensor("op_343_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_343_end_0 = const()[name = tensor("op_343_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_343_end_mask_0 = const()[name = tensor("op_343_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_343_cast_fp16 = slice_by_index(begin = var_343_begin_0, end = var_343_end_0, end_mask = var_343_end_mask_0, x = var_218_cast_fp16)[name = tensor("op_343_cast_fp16")]; tensor var_350_begin_0 = const()[name = tensor("op_350_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_350_end_0 = const()[name = tensor("op_350_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_350_end_mask_0 = const()[name = tensor("op_350_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_350_cast_fp16 = slice_by_index(begin = var_350_begin_0, end = var_350_end_0, end_mask = var_350_end_mask_0, x = var_218_cast_fp16)[name = tensor("op_350_cast_fp16")]; tensor var_357_begin_0 = const()[name = tensor("op_357_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_357_end_0 = const()[name = tensor("op_357_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_357_end_mask_0 = const()[name = tensor("op_357_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_357_cast_fp16 = slice_by_index(begin = var_357_begin_0, end = var_357_end_0, end_mask = var_357_end_mask_0, x = var_218_cast_fp16)[name = tensor("op_357_cast_fp16")]; tensor var_364_begin_0 = const()[name = tensor("op_364_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_364_end_0 = const()[name = tensor("op_364_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_364_end_mask_0 = const()[name = tensor("op_364_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_218_cast_fp16)[name = tensor("op_364_cast_fp16")]; tensor var_371_begin_0 = const()[name = tensor("op_371_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_371_end_0 = const()[name = tensor("op_371_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_371_end_mask_0 = const()[name = tensor("op_371_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_371_cast_fp16 = slice_by_index(begin = var_371_begin_0, end = var_371_end_0, end_mask = var_371_end_mask_0, x = var_222_cast_fp16)[name = tensor("op_371_cast_fp16")]; tensor var_378_begin_0 = const()[name = tensor("op_378_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_378_end_0 = const()[name = tensor("op_378_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_378_end_mask_0 = const()[name = tensor("op_378_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_378_cast_fp16 = slice_by_index(begin = var_378_begin_0, end = var_378_end_0, end_mask = var_378_end_mask_0, x = var_222_cast_fp16)[name = tensor("op_378_cast_fp16")]; tensor var_385_begin_0 = const()[name = tensor("op_385_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_385_end_0 = const()[name = tensor("op_385_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_385_end_mask_0 = const()[name = tensor("op_385_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_385_cast_fp16 = slice_by_index(begin = var_385_begin_0, end = var_385_end_0, end_mask = var_385_end_mask_0, x = var_222_cast_fp16)[name = tensor("op_385_cast_fp16")]; tensor var_392_begin_0 = const()[name = tensor("op_392_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_392_end_0 = const()[name = tensor("op_392_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_392_end_mask_0 = const()[name = tensor("op_392_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_392_cast_fp16 = slice_by_index(begin = var_392_begin_0, end = var_392_end_0, end_mask = var_392_end_mask_0, x = var_222_cast_fp16)[name = tensor("op_392_cast_fp16")]; tensor var_399_begin_0 = const()[name = tensor("op_399_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_399_end_0 = const()[name = tensor("op_399_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_399_end_mask_0 = const()[name = tensor("op_399_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_399_cast_fp16 = slice_by_index(begin = var_399_begin_0, end = var_399_end_0, end_mask = var_399_end_mask_0, x = var_226_cast_fp16)[name = tensor("op_399_cast_fp16")]; tensor var_406_begin_0 = const()[name = tensor("op_406_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_406_end_0 = const()[name = tensor("op_406_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_406_end_mask_0 = const()[name = tensor("op_406_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_406_cast_fp16 = slice_by_index(begin = var_406_begin_0, end = var_406_end_0, end_mask = var_406_end_mask_0, x = var_226_cast_fp16)[name = tensor("op_406_cast_fp16")]; tensor var_413_begin_0 = const()[name = tensor("op_413_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_413_end_0 = const()[name = tensor("op_413_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_413_end_mask_0 = const()[name = tensor("op_413_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = var_226_cast_fp16)[name = tensor("op_413_cast_fp16")]; tensor var_420_begin_0 = const()[name = tensor("op_420_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_420_end_0 = const()[name = tensor("op_420_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_420_end_mask_0 = const()[name = tensor("op_420_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_420_cast_fp16 = slice_by_index(begin = var_420_begin_0, end = var_420_end_0, end_mask = var_420_end_mask_0, x = var_226_cast_fp16)[name = tensor("op_420_cast_fp16")]; tensor var_427_begin_0 = const()[name = tensor("op_427_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_427_end_0 = const()[name = tensor("op_427_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_427_end_mask_0 = const()[name = tensor("op_427_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_427_cast_fp16 = slice_by_index(begin = var_427_begin_0, end = var_427_end_0, end_mask = var_427_end_mask_0, x = var_230_cast_fp16)[name = tensor("op_427_cast_fp16")]; tensor var_434_begin_0 = const()[name = tensor("op_434_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_434_end_0 = const()[name = tensor("op_434_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_434_end_mask_0 = const()[name = tensor("op_434_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, x = var_230_cast_fp16)[name = tensor("op_434_cast_fp16")]; tensor var_441_begin_0 = const()[name = tensor("op_441_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_441_end_0 = const()[name = tensor("op_441_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_441_end_mask_0 = const()[name = tensor("op_441_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_441_cast_fp16 = slice_by_index(begin = var_441_begin_0, end = var_441_end_0, end_mask = var_441_end_mask_0, x = var_230_cast_fp16)[name = tensor("op_441_cast_fp16")]; tensor var_448_begin_0 = const()[name = tensor("op_448_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_448_end_0 = const()[name = tensor("op_448_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_448_end_mask_0 = const()[name = tensor("op_448_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, x = var_230_cast_fp16)[name = tensor("op_448_cast_fp16")]; tensor var_455_begin_0 = const()[name = tensor("op_455_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_455_end_0 = const()[name = tensor("op_455_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_455_end_mask_0 = const()[name = tensor("op_455_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_455_cast_fp16 = slice_by_index(begin = var_455_begin_0, end = var_455_end_0, end_mask = var_455_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_455_cast_fp16")]; tensor var_462_begin_0 = const()[name = tensor("op_462_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_462_end_0 = const()[name = tensor("op_462_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_462_end_mask_0 = const()[name = tensor("op_462_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_462_cast_fp16 = slice_by_index(begin = var_462_begin_0, end = var_462_end_0, end_mask = var_462_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_462_cast_fp16")]; tensor var_469_begin_0 = const()[name = tensor("op_469_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_469_end_0 = const()[name = tensor("op_469_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_469_end_mask_0 = const()[name = tensor("op_469_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_469_cast_fp16")]; tensor var_476_begin_0 = const()[name = tensor("op_476_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_476_end_0 = const()[name = tensor("op_476_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_476_end_mask_0 = const()[name = tensor("op_476_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_476_cast_fp16 = slice_by_index(begin = var_476_begin_0, end = var_476_end_0, end_mask = var_476_end_mask_0, x = var_234_cast_fp16)[name = tensor("op_476_cast_fp16")]; tensor var_483_begin_0 = const()[name = tensor("op_483_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_483_end_0 = const()[name = tensor("op_483_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_483_end_mask_0 = const()[name = tensor("op_483_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_483_cast_fp16 = slice_by_index(begin = var_483_begin_0, end = var_483_end_0, end_mask = var_483_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_483_cast_fp16")]; tensor var_490_begin_0 = const()[name = tensor("op_490_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_490_end_0 = const()[name = tensor("op_490_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_490_end_mask_0 = const()[name = tensor("op_490_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_490_cast_fp16 = slice_by_index(begin = var_490_begin_0, end = var_490_end_0, end_mask = var_490_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_490_cast_fp16")]; tensor var_497_begin_0 = const()[name = tensor("op_497_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_497_end_0 = const()[name = tensor("op_497_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_497_end_mask_0 = const()[name = tensor("op_497_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_497_cast_fp16 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_497_cast_fp16")]; tensor var_504_begin_0 = const()[name = tensor("op_504_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_504_end_0 = const()[name = tensor("op_504_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_504_end_mask_0 = const()[name = tensor("op_504_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_504_cast_fp16 = slice_by_index(begin = var_504_begin_0, end = var_504_end_0, end_mask = var_504_end_mask_0, x = var_238_cast_fp16)[name = tensor("op_504_cast_fp16")]; tensor var_511_begin_0 = const()[name = tensor("op_511_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_511_end_0 = const()[name = tensor("op_511_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_511_end_mask_0 = const()[name = tensor("op_511_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_511_cast_fp16")]; tensor var_518_begin_0 = const()[name = tensor("op_518_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_518_end_0 = const()[name = tensor("op_518_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_518_end_mask_0 = const()[name = tensor("op_518_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_518_cast_fp16 = slice_by_index(begin = var_518_begin_0, end = var_518_end_0, end_mask = var_518_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_518_cast_fp16")]; tensor var_525_begin_0 = const()[name = tensor("op_525_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_525_end_0 = const()[name = tensor("op_525_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_525_end_mask_0 = const()[name = tensor("op_525_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_525_cast_fp16 = slice_by_index(begin = var_525_begin_0, end = var_525_end_0, end_mask = var_525_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_525_cast_fp16")]; tensor var_532_begin_0 = const()[name = tensor("op_532_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_532_end_0 = const()[name = tensor("op_532_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_532_end_mask_0 = const()[name = tensor("op_532_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_532_cast_fp16 = slice_by_index(begin = var_532_begin_0, end = var_532_end_0, end_mask = var_532_end_mask_0, x = var_242_cast_fp16)[name = tensor("op_532_cast_fp16")]; tensor var_539_begin_0 = const()[name = tensor("op_539_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_539_end_0 = const()[name = tensor("op_539_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_539_end_mask_0 = const()[name = tensor("op_539_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_539_cast_fp16 = slice_by_index(begin = var_539_begin_0, end = var_539_end_0, end_mask = var_539_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_539_cast_fp16")]; tensor var_546_begin_0 = const()[name = tensor("op_546_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_546_end_0 = const()[name = tensor("op_546_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_546_end_mask_0 = const()[name = tensor("op_546_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_546_cast_fp16 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_546_cast_fp16")]; tensor var_553_begin_0 = const()[name = tensor("op_553_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_553_end_0 = const()[name = tensor("op_553_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_553_end_mask_0 = const()[name = tensor("op_553_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_553_cast_fp16 = slice_by_index(begin = var_553_begin_0, end = var_553_end_0, end_mask = var_553_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_553_cast_fp16")]; tensor var_560_begin_0 = const()[name = tensor("op_560_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_560_end_0 = const()[name = tensor("op_560_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_560_end_mask_0 = const()[name = tensor("op_560_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_560_cast_fp16 = slice_by_index(begin = var_560_begin_0, end = var_560_end_0, end_mask = var_560_end_mask_0, x = var_246_cast_fp16)[name = tensor("op_560_cast_fp16")]; tensor var_567_begin_0 = const()[name = tensor("op_567_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_567_end_0 = const()[name = tensor("op_567_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_567_end_mask_0 = const()[name = tensor("op_567_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_567_cast_fp16 = slice_by_index(begin = var_567_begin_0, end = var_567_end_0, end_mask = var_567_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_567_cast_fp16")]; tensor var_574_begin_0 = const()[name = tensor("op_574_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_574_end_0 = const()[name = tensor("op_574_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_574_end_mask_0 = const()[name = tensor("op_574_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_574_cast_fp16 = slice_by_index(begin = var_574_begin_0, end = var_574_end_0, end_mask = var_574_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_574_cast_fp16")]; tensor var_581_begin_0 = const()[name = tensor("op_581_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_581_end_0 = const()[name = tensor("op_581_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_581_end_mask_0 = const()[name = tensor("op_581_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_581_cast_fp16 = slice_by_index(begin = var_581_begin_0, end = var_581_end_0, end_mask = var_581_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_581_cast_fp16")]; tensor var_588_begin_0 = const()[name = tensor("op_588_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_588_end_0 = const()[name = tensor("op_588_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_588_end_mask_0 = const()[name = tensor("op_588_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_588_cast_fp16 = slice_by_index(begin = var_588_begin_0, end = var_588_end_0, end_mask = var_588_end_mask_0, x = var_250_cast_fp16)[name = tensor("op_588_cast_fp16")]; tensor k_1_perm_0 = const()[name = tensor("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_593_begin_0 = const()[name = tensor("op_593_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_593_end_0 = const()[name = tensor("op_593_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_593_end_mask_0 = const()[name = tensor("op_593_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_11 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = tensor("transpose_11")]; tensor var_593_cast_fp16 = slice_by_index(begin = var_593_begin_0, end = var_593_end_0, end_mask = var_593_end_mask_0, x = transpose_11)[name = tensor("op_593_cast_fp16")]; tensor var_597_begin_0 = const()[name = tensor("op_597_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_597_end_0 = const()[name = tensor("op_597_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_597_end_mask_0 = const()[name = tensor("op_597_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_597_cast_fp16 = slice_by_index(begin = var_597_begin_0, end = var_597_end_0, end_mask = var_597_end_mask_0, x = transpose_11)[name = tensor("op_597_cast_fp16")]; tensor var_601_begin_0 = const()[name = tensor("op_601_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_601_end_0 = const()[name = tensor("op_601_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_601_end_mask_0 = const()[name = tensor("op_601_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_601_cast_fp16 = slice_by_index(begin = var_601_begin_0, end = var_601_end_0, end_mask = var_601_end_mask_0, x = transpose_11)[name = tensor("op_601_cast_fp16")]; tensor var_605_begin_0 = const()[name = tensor("op_605_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_605_end_0 = const()[name = tensor("op_605_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_605_end_mask_0 = const()[name = tensor("op_605_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_605_cast_fp16 = slice_by_index(begin = var_605_begin_0, end = var_605_end_0, end_mask = var_605_end_mask_0, x = transpose_11)[name = tensor("op_605_cast_fp16")]; tensor var_609_begin_0 = const()[name = tensor("op_609_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_609_end_0 = const()[name = tensor("op_609_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_609_end_mask_0 = const()[name = tensor("op_609_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, x = transpose_11)[name = tensor("op_609_cast_fp16")]; tensor var_613_begin_0 = const()[name = tensor("op_613_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_613_end_0 = const()[name = tensor("op_613_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_613_end_mask_0 = const()[name = tensor("op_613_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_613_cast_fp16 = slice_by_index(begin = var_613_begin_0, end = var_613_end_0, end_mask = var_613_end_mask_0, x = transpose_11)[name = tensor("op_613_cast_fp16")]; tensor var_617_begin_0 = const()[name = tensor("op_617_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_617_end_0 = const()[name = tensor("op_617_end_0"), val = tensor([1, 1500, 1, 448])]; 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 = transpose_11)[name = tensor("op_617_cast_fp16")]; tensor var_621_begin_0 = const()[name = tensor("op_621_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_621_end_0 = const()[name = tensor("op_621_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_621_end_mask_0 = const()[name = tensor("op_621_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_621_cast_fp16 = slice_by_index(begin = var_621_begin_0, end = var_621_end_0, end_mask = var_621_end_mask_0, x = transpose_11)[name = tensor("op_621_cast_fp16")]; tensor var_625_begin_0 = const()[name = tensor("op_625_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_625_end_0 = const()[name = tensor("op_625_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_625_end_mask_0 = const()[name = tensor("op_625_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_625_cast_fp16 = slice_by_index(begin = var_625_begin_0, end = var_625_end_0, end_mask = var_625_end_mask_0, x = transpose_11)[name = tensor("op_625_cast_fp16")]; tensor var_629_begin_0 = const()[name = tensor("op_629_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_629_end_0 = const()[name = tensor("op_629_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_629_end_mask_0 = const()[name = tensor("op_629_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_629_cast_fp16 = slice_by_index(begin = var_629_begin_0, end = var_629_end_0, end_mask = var_629_end_mask_0, x = transpose_11)[name = tensor("op_629_cast_fp16")]; tensor var_633_begin_0 = const()[name = tensor("op_633_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_633_end_0 = const()[name = tensor("op_633_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_633_end_mask_0 = const()[name = tensor("op_633_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_633_cast_fp16 = slice_by_index(begin = var_633_begin_0, end = var_633_end_0, end_mask = var_633_end_mask_0, x = transpose_11)[name = tensor("op_633_cast_fp16")]; tensor var_637_begin_0 = const()[name = tensor("op_637_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_637_end_0 = const()[name = tensor("op_637_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_637_end_mask_0 = const()[name = tensor("op_637_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, x = transpose_11)[name = tensor("op_637_cast_fp16")]; tensor var_639_begin_0 = const()[name = tensor("op_639_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_639_end_0 = const()[name = tensor("op_639_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_639_end_mask_0 = const()[name = tensor("op_639_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_639_cast_fp16 = slice_by_index(begin = var_639_begin_0, end = var_639_end_0, end_mask = var_639_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_639_cast_fp16")]; tensor var_643_begin_0 = const()[name = tensor("op_643_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_643_end_0 = const()[name = tensor("op_643_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_643_end_mask_0 = const()[name = tensor("op_643_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_643_cast_fp16 = slice_by_index(begin = var_643_begin_0, end = var_643_end_0, end_mask = var_643_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_643_cast_fp16")]; tensor var_647_begin_0 = const()[name = tensor("op_647_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_647_end_0 = const()[name = tensor("op_647_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_647_end_mask_0 = const()[name = tensor("op_647_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_647_cast_fp16 = slice_by_index(begin = var_647_begin_0, end = var_647_end_0, end_mask = var_647_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_647_cast_fp16")]; tensor var_651_begin_0 = const()[name = tensor("op_651_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_651_end_0 = const()[name = tensor("op_651_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_651_end_mask_0 = const()[name = tensor("op_651_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_651_cast_fp16 = slice_by_index(begin = var_651_begin_0, end = var_651_end_0, end_mask = var_651_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_651_cast_fp16")]; tensor var_655_begin_0 = const()[name = tensor("op_655_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_655_end_0 = const()[name = tensor("op_655_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_655_end_mask_0 = const()[name = tensor("op_655_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_655_cast_fp16 = slice_by_index(begin = var_655_begin_0, end = var_655_end_0, end_mask = var_655_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_655_cast_fp16")]; tensor var_659_begin_0 = const()[name = tensor("op_659_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_659_end_0 = const()[name = tensor("op_659_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_659_end_mask_0 = const()[name = tensor("op_659_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_1_cast_fp16)[name = tensor("op_659_cast_fp16")]; tensor var_663_begin_0 = const()[name = tensor("op_663_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_663_end_0 = const()[name = tensor("op_663_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_663_end_mask_0 = const()[name = tensor("op_663_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_663_cast_fp16 = slice_by_index(begin = var_663_begin_0, end = var_663_end_0, end_mask = var_663_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_663_cast_fp16")]; tensor var_667_begin_0 = const()[name = tensor("op_667_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_667_end_0 = const()[name = tensor("op_667_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_667_end_mask_0 = const()[name = tensor("op_667_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_667_cast_fp16 = slice_by_index(begin = var_667_begin_0, end = var_667_end_0, end_mask = var_667_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_667_cast_fp16")]; tensor var_671_begin_0 = const()[name = tensor("op_671_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_671_end_0 = const()[name = tensor("op_671_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_671_end_mask_0 = const()[name = tensor("op_671_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_671_cast_fp16 = slice_by_index(begin = var_671_begin_0, end = var_671_end_0, end_mask = var_671_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_671_cast_fp16")]; tensor var_675_begin_0 = const()[name = tensor("op_675_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_675_end_0 = const()[name = tensor("op_675_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_675_end_mask_0 = const()[name = tensor("op_675_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_675_cast_fp16 = slice_by_index(begin = var_675_begin_0, end = var_675_end_0, end_mask = var_675_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_675_cast_fp16")]; tensor var_679_begin_0 = const()[name = tensor("op_679_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_679_end_0 = const()[name = tensor("op_679_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_679_end_mask_0 = const()[name = tensor("op_679_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_679_cast_fp16")]; tensor var_683_begin_0 = const()[name = tensor("op_683_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_683_end_0 = const()[name = tensor("op_683_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_683_end_mask_0 = const()[name = tensor("op_683_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, x = value_1_cast_fp16)[name = tensor("op_683_cast_fp16")]; tensor var_687_equation_0 = const()[name = tensor("op_687_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_687_cast_fp16 = einsum(equation = var_687_equation_0, values = (var_593_cast_fp16, var_259_cast_fp16))[name = tensor("op_687_cast_fp16")]; tensor var_688_to_fp16 = const()[name = tensor("op_688_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1_cast_fp16 = mul(x = var_687_cast_fp16, y = var_688_to_fp16)[name = tensor("aw_chunk_1_cast_fp16")]; tensor var_691_equation_0 = const()[name = tensor("op_691_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_691_cast_fp16 = einsum(equation = var_691_equation_0, values = (var_593_cast_fp16, var_266_cast_fp16))[name = tensor("op_691_cast_fp16")]; tensor var_692_to_fp16 = const()[name = tensor("op_692_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_3_cast_fp16 = mul(x = var_691_cast_fp16, y = var_692_to_fp16)[name = tensor("aw_chunk_3_cast_fp16")]; tensor var_695_equation_0 = const()[name = tensor("op_695_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_695_cast_fp16 = einsum(equation = var_695_equation_0, values = (var_593_cast_fp16, var_273_cast_fp16))[name = tensor("op_695_cast_fp16")]; tensor var_696_to_fp16 = const()[name = tensor("op_696_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_5_cast_fp16 = mul(x = var_695_cast_fp16, y = var_696_to_fp16)[name = tensor("aw_chunk_5_cast_fp16")]; tensor var_699_equation_0 = const()[name = tensor("op_699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_699_cast_fp16 = einsum(equation = var_699_equation_0, values = (var_593_cast_fp16, var_280_cast_fp16))[name = tensor("op_699_cast_fp16")]; tensor var_700_to_fp16 = const()[name = tensor("op_700_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_7_cast_fp16 = mul(x = var_699_cast_fp16, y = var_700_to_fp16)[name = tensor("aw_chunk_7_cast_fp16")]; tensor var_703_equation_0 = const()[name = tensor("op_703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_703_cast_fp16 = einsum(equation = var_703_equation_0, values = (var_597_cast_fp16, var_287_cast_fp16))[name = tensor("op_703_cast_fp16")]; tensor var_704_to_fp16 = const()[name = tensor("op_704_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_9_cast_fp16 = mul(x = var_703_cast_fp16, y = var_704_to_fp16)[name = tensor("aw_chunk_9_cast_fp16")]; tensor var_707_equation_0 = const()[name = tensor("op_707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_707_cast_fp16 = einsum(equation = var_707_equation_0, values = (var_597_cast_fp16, var_294_cast_fp16))[name = tensor("op_707_cast_fp16")]; tensor var_708_to_fp16 = const()[name = tensor("op_708_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_11_cast_fp16 = mul(x = var_707_cast_fp16, y = var_708_to_fp16)[name = tensor("aw_chunk_11_cast_fp16")]; tensor var_711_equation_0 = const()[name = tensor("op_711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_711_cast_fp16 = einsum(equation = var_711_equation_0, values = (var_597_cast_fp16, var_301_cast_fp16))[name = tensor("op_711_cast_fp16")]; tensor var_712_to_fp16 = const()[name = tensor("op_712_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_13_cast_fp16 = mul(x = var_711_cast_fp16, y = var_712_to_fp16)[name = tensor("aw_chunk_13_cast_fp16")]; tensor var_715_equation_0 = const()[name = tensor("op_715_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_715_cast_fp16 = einsum(equation = var_715_equation_0, values = (var_597_cast_fp16, var_308_cast_fp16))[name = tensor("op_715_cast_fp16")]; tensor var_716_to_fp16 = const()[name = tensor("op_716_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_15_cast_fp16 = mul(x = var_715_cast_fp16, y = var_716_to_fp16)[name = tensor("aw_chunk_15_cast_fp16")]; tensor var_719_equation_0 = const()[name = tensor("op_719_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_719_cast_fp16 = einsum(equation = var_719_equation_0, values = (var_601_cast_fp16, var_315_cast_fp16))[name = tensor("op_719_cast_fp16")]; tensor var_720_to_fp16 = const()[name = tensor("op_720_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_17_cast_fp16 = mul(x = var_719_cast_fp16, y = var_720_to_fp16)[name = tensor("aw_chunk_17_cast_fp16")]; tensor var_723_equation_0 = const()[name = tensor("op_723_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_723_cast_fp16 = einsum(equation = var_723_equation_0, values = (var_601_cast_fp16, var_322_cast_fp16))[name = tensor("op_723_cast_fp16")]; tensor var_724_to_fp16 = const()[name = tensor("op_724_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_19_cast_fp16 = mul(x = var_723_cast_fp16, y = var_724_to_fp16)[name = tensor("aw_chunk_19_cast_fp16")]; tensor var_727_equation_0 = const()[name = tensor("op_727_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_727_cast_fp16 = einsum(equation = var_727_equation_0, values = (var_601_cast_fp16, var_329_cast_fp16))[name = tensor("op_727_cast_fp16")]; tensor var_728_to_fp16 = const()[name = tensor("op_728_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_21_cast_fp16 = mul(x = var_727_cast_fp16, y = var_728_to_fp16)[name = tensor("aw_chunk_21_cast_fp16")]; tensor var_731_equation_0 = const()[name = tensor("op_731_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_731_cast_fp16 = einsum(equation = var_731_equation_0, values = (var_601_cast_fp16, var_336_cast_fp16))[name = tensor("op_731_cast_fp16")]; tensor var_732_to_fp16 = const()[name = tensor("op_732_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_23_cast_fp16 = mul(x = var_731_cast_fp16, y = var_732_to_fp16)[name = tensor("aw_chunk_23_cast_fp16")]; tensor var_735_equation_0 = const()[name = tensor("op_735_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_735_cast_fp16 = einsum(equation = var_735_equation_0, values = (var_605_cast_fp16, var_343_cast_fp16))[name = tensor("op_735_cast_fp16")]; tensor var_736_to_fp16 = const()[name = tensor("op_736_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_25_cast_fp16 = mul(x = var_735_cast_fp16, y = var_736_to_fp16)[name = tensor("aw_chunk_25_cast_fp16")]; tensor var_739_equation_0 = const()[name = tensor("op_739_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_739_cast_fp16 = einsum(equation = var_739_equation_0, values = (var_605_cast_fp16, var_350_cast_fp16))[name = tensor("op_739_cast_fp16")]; tensor var_740_to_fp16 = const()[name = tensor("op_740_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_27_cast_fp16 = mul(x = var_739_cast_fp16, y = var_740_to_fp16)[name = tensor("aw_chunk_27_cast_fp16")]; tensor var_743_equation_0 = const()[name = tensor("op_743_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_743_cast_fp16 = einsum(equation = var_743_equation_0, values = (var_605_cast_fp16, var_357_cast_fp16))[name = tensor("op_743_cast_fp16")]; tensor var_744_to_fp16 = const()[name = tensor("op_744_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_29_cast_fp16 = mul(x = var_743_cast_fp16, y = var_744_to_fp16)[name = tensor("aw_chunk_29_cast_fp16")]; tensor var_747_equation_0 = const()[name = tensor("op_747_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_747_cast_fp16 = einsum(equation = var_747_equation_0, values = (var_605_cast_fp16, var_364_cast_fp16))[name = tensor("op_747_cast_fp16")]; tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_31_cast_fp16 = mul(x = var_747_cast_fp16, y = var_748_to_fp16)[name = tensor("aw_chunk_31_cast_fp16")]; tensor var_751_equation_0 = const()[name = tensor("op_751_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_751_cast_fp16 = einsum(equation = var_751_equation_0, values = (var_609_cast_fp16, var_371_cast_fp16))[name = tensor("op_751_cast_fp16")]; tensor var_752_to_fp16 = const()[name = tensor("op_752_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_33_cast_fp16 = mul(x = var_751_cast_fp16, y = var_752_to_fp16)[name = tensor("aw_chunk_33_cast_fp16")]; tensor var_755_equation_0 = const()[name = tensor("op_755_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_755_cast_fp16 = einsum(equation = var_755_equation_0, values = (var_609_cast_fp16, var_378_cast_fp16))[name = tensor("op_755_cast_fp16")]; tensor var_756_to_fp16 = const()[name = tensor("op_756_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_35_cast_fp16 = mul(x = var_755_cast_fp16, y = var_756_to_fp16)[name = tensor("aw_chunk_35_cast_fp16")]; tensor var_759_equation_0 = const()[name = tensor("op_759_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_759_cast_fp16 = einsum(equation = var_759_equation_0, values = (var_609_cast_fp16, var_385_cast_fp16))[name = tensor("op_759_cast_fp16")]; tensor var_760_to_fp16 = const()[name = tensor("op_760_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_37_cast_fp16 = mul(x = var_759_cast_fp16, y = var_760_to_fp16)[name = tensor("aw_chunk_37_cast_fp16")]; tensor var_763_equation_0 = const()[name = tensor("op_763_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_763_cast_fp16 = einsum(equation = var_763_equation_0, values = (var_609_cast_fp16, var_392_cast_fp16))[name = tensor("op_763_cast_fp16")]; tensor var_764_to_fp16 = const()[name = tensor("op_764_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_39_cast_fp16 = mul(x = var_763_cast_fp16, y = var_764_to_fp16)[name = tensor("aw_chunk_39_cast_fp16")]; tensor var_767_equation_0 = const()[name = tensor("op_767_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_767_cast_fp16 = einsum(equation = var_767_equation_0, values = (var_613_cast_fp16, var_399_cast_fp16))[name = tensor("op_767_cast_fp16")]; tensor var_768_to_fp16 = const()[name = tensor("op_768_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_41_cast_fp16 = mul(x = var_767_cast_fp16, y = var_768_to_fp16)[name = tensor("aw_chunk_41_cast_fp16")]; tensor var_771_equation_0 = const()[name = tensor("op_771_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_771_cast_fp16 = einsum(equation = var_771_equation_0, values = (var_613_cast_fp16, var_406_cast_fp16))[name = tensor("op_771_cast_fp16")]; tensor var_772_to_fp16 = const()[name = tensor("op_772_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_43_cast_fp16 = mul(x = var_771_cast_fp16, y = var_772_to_fp16)[name = tensor("aw_chunk_43_cast_fp16")]; tensor var_775_equation_0 = const()[name = tensor("op_775_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_775_cast_fp16 = einsum(equation = var_775_equation_0, values = (var_613_cast_fp16, var_413_cast_fp16))[name = tensor("op_775_cast_fp16")]; tensor var_776_to_fp16 = const()[name = tensor("op_776_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_45_cast_fp16 = mul(x = var_775_cast_fp16, y = var_776_to_fp16)[name = tensor("aw_chunk_45_cast_fp16")]; tensor var_779_equation_0 = const()[name = tensor("op_779_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_779_cast_fp16 = einsum(equation = var_779_equation_0, values = (var_613_cast_fp16, var_420_cast_fp16))[name = tensor("op_779_cast_fp16")]; tensor var_780_to_fp16 = const()[name = tensor("op_780_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_47_cast_fp16 = mul(x = var_779_cast_fp16, y = var_780_to_fp16)[name = tensor("aw_chunk_47_cast_fp16")]; tensor var_783_equation_0 = const()[name = tensor("op_783_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_783_cast_fp16 = einsum(equation = var_783_equation_0, values = (var_617_cast_fp16, var_427_cast_fp16))[name = tensor("op_783_cast_fp16")]; tensor var_784_to_fp16 = const()[name = tensor("op_784_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_49_cast_fp16 = mul(x = var_783_cast_fp16, y = var_784_to_fp16)[name = tensor("aw_chunk_49_cast_fp16")]; tensor var_787_equation_0 = const()[name = tensor("op_787_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_787_cast_fp16 = einsum(equation = var_787_equation_0, values = (var_617_cast_fp16, var_434_cast_fp16))[name = tensor("op_787_cast_fp16")]; tensor var_788_to_fp16 = const()[name = tensor("op_788_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_51_cast_fp16 = mul(x = var_787_cast_fp16, y = var_788_to_fp16)[name = tensor("aw_chunk_51_cast_fp16")]; tensor var_791_equation_0 = const()[name = tensor("op_791_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_791_cast_fp16 = einsum(equation = var_791_equation_0, values = (var_617_cast_fp16, var_441_cast_fp16))[name = tensor("op_791_cast_fp16")]; tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_53_cast_fp16 = mul(x = var_791_cast_fp16, y = var_792_to_fp16)[name = tensor("aw_chunk_53_cast_fp16")]; tensor var_795_equation_0 = const()[name = tensor("op_795_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_795_cast_fp16 = einsum(equation = var_795_equation_0, values = (var_617_cast_fp16, var_448_cast_fp16))[name = tensor("op_795_cast_fp16")]; tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_55_cast_fp16 = mul(x = var_795_cast_fp16, y = var_796_to_fp16)[name = tensor("aw_chunk_55_cast_fp16")]; tensor var_799_equation_0 = const()[name = tensor("op_799_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_799_cast_fp16 = einsum(equation = var_799_equation_0, values = (var_621_cast_fp16, var_455_cast_fp16))[name = tensor("op_799_cast_fp16")]; tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_57_cast_fp16 = mul(x = var_799_cast_fp16, y = var_800_to_fp16)[name = tensor("aw_chunk_57_cast_fp16")]; tensor var_803_equation_0 = const()[name = tensor("op_803_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_803_cast_fp16 = einsum(equation = var_803_equation_0, values = (var_621_cast_fp16, var_462_cast_fp16))[name = tensor("op_803_cast_fp16")]; tensor var_804_to_fp16 = const()[name = tensor("op_804_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_59_cast_fp16 = mul(x = var_803_cast_fp16, y = var_804_to_fp16)[name = tensor("aw_chunk_59_cast_fp16")]; tensor var_807_equation_0 = const()[name = tensor("op_807_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_807_cast_fp16 = einsum(equation = var_807_equation_0, values = (var_621_cast_fp16, var_469_cast_fp16))[name = tensor("op_807_cast_fp16")]; tensor var_808_to_fp16 = const()[name = tensor("op_808_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_61_cast_fp16 = mul(x = var_807_cast_fp16, y = var_808_to_fp16)[name = tensor("aw_chunk_61_cast_fp16")]; tensor var_811_equation_0 = const()[name = tensor("op_811_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_811_cast_fp16 = einsum(equation = var_811_equation_0, values = (var_621_cast_fp16, var_476_cast_fp16))[name = tensor("op_811_cast_fp16")]; tensor var_812_to_fp16 = const()[name = tensor("op_812_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_63_cast_fp16 = mul(x = var_811_cast_fp16, y = var_812_to_fp16)[name = tensor("aw_chunk_63_cast_fp16")]; tensor var_815_equation_0 = const()[name = tensor("op_815_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_815_cast_fp16 = einsum(equation = var_815_equation_0, values = (var_625_cast_fp16, var_483_cast_fp16))[name = tensor("op_815_cast_fp16")]; tensor var_816_to_fp16 = const()[name = tensor("op_816_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_65_cast_fp16 = mul(x = var_815_cast_fp16, y = var_816_to_fp16)[name = tensor("aw_chunk_65_cast_fp16")]; tensor var_819_equation_0 = const()[name = tensor("op_819_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_819_cast_fp16 = einsum(equation = var_819_equation_0, values = (var_625_cast_fp16, var_490_cast_fp16))[name = tensor("op_819_cast_fp16")]; tensor var_820_to_fp16 = const()[name = tensor("op_820_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_67_cast_fp16 = mul(x = var_819_cast_fp16, y = var_820_to_fp16)[name = tensor("aw_chunk_67_cast_fp16")]; tensor var_823_equation_0 = const()[name = tensor("op_823_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_823_cast_fp16 = einsum(equation = var_823_equation_0, values = (var_625_cast_fp16, var_497_cast_fp16))[name = tensor("op_823_cast_fp16")]; tensor var_824_to_fp16 = const()[name = tensor("op_824_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_69_cast_fp16 = mul(x = var_823_cast_fp16, y = var_824_to_fp16)[name = tensor("aw_chunk_69_cast_fp16")]; tensor var_827_equation_0 = const()[name = tensor("op_827_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_827_cast_fp16 = einsum(equation = var_827_equation_0, values = (var_625_cast_fp16, var_504_cast_fp16))[name = tensor("op_827_cast_fp16")]; tensor var_828_to_fp16 = const()[name = tensor("op_828_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_71_cast_fp16 = mul(x = var_827_cast_fp16, y = var_828_to_fp16)[name = tensor("aw_chunk_71_cast_fp16")]; tensor var_831_equation_0 = const()[name = tensor("op_831_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_831_cast_fp16 = einsum(equation = var_831_equation_0, values = (var_629_cast_fp16, var_511_cast_fp16))[name = tensor("op_831_cast_fp16")]; tensor var_832_to_fp16 = const()[name = tensor("op_832_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_73_cast_fp16 = mul(x = var_831_cast_fp16, y = var_832_to_fp16)[name = tensor("aw_chunk_73_cast_fp16")]; tensor var_835_equation_0 = const()[name = tensor("op_835_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_835_cast_fp16 = einsum(equation = var_835_equation_0, values = (var_629_cast_fp16, var_518_cast_fp16))[name = tensor("op_835_cast_fp16")]; tensor var_836_to_fp16 = const()[name = tensor("op_836_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_75_cast_fp16 = mul(x = var_835_cast_fp16, y = var_836_to_fp16)[name = tensor("aw_chunk_75_cast_fp16")]; tensor var_839_equation_0 = const()[name = tensor("op_839_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_839_cast_fp16 = einsum(equation = var_839_equation_0, values = (var_629_cast_fp16, var_525_cast_fp16))[name = tensor("op_839_cast_fp16")]; tensor var_840_to_fp16 = const()[name = tensor("op_840_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_77_cast_fp16 = mul(x = var_839_cast_fp16, y = var_840_to_fp16)[name = tensor("aw_chunk_77_cast_fp16")]; tensor var_843_equation_0 = const()[name = tensor("op_843_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_843_cast_fp16 = einsum(equation = var_843_equation_0, values = (var_629_cast_fp16, var_532_cast_fp16))[name = tensor("op_843_cast_fp16")]; tensor var_844_to_fp16 = const()[name = tensor("op_844_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_79_cast_fp16 = mul(x = var_843_cast_fp16, y = var_844_to_fp16)[name = tensor("aw_chunk_79_cast_fp16")]; tensor var_847_equation_0 = const()[name = tensor("op_847_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_847_cast_fp16 = einsum(equation = var_847_equation_0, values = (var_633_cast_fp16, var_539_cast_fp16))[name = tensor("op_847_cast_fp16")]; tensor var_848_to_fp16 = const()[name = tensor("op_848_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_81_cast_fp16 = mul(x = var_847_cast_fp16, y = var_848_to_fp16)[name = tensor("aw_chunk_81_cast_fp16")]; tensor var_851_equation_0 = const()[name = tensor("op_851_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_851_cast_fp16 = einsum(equation = var_851_equation_0, values = (var_633_cast_fp16, var_546_cast_fp16))[name = tensor("op_851_cast_fp16")]; tensor var_852_to_fp16 = const()[name = tensor("op_852_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_83_cast_fp16 = mul(x = var_851_cast_fp16, y = var_852_to_fp16)[name = tensor("aw_chunk_83_cast_fp16")]; tensor var_855_equation_0 = const()[name = tensor("op_855_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_855_cast_fp16 = einsum(equation = var_855_equation_0, values = (var_633_cast_fp16, var_553_cast_fp16))[name = tensor("op_855_cast_fp16")]; tensor var_856_to_fp16 = const()[name = tensor("op_856_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_85_cast_fp16 = mul(x = var_855_cast_fp16, y = var_856_to_fp16)[name = tensor("aw_chunk_85_cast_fp16")]; tensor var_859_equation_0 = const()[name = tensor("op_859_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_859_cast_fp16 = einsum(equation = var_859_equation_0, values = (var_633_cast_fp16, var_560_cast_fp16))[name = tensor("op_859_cast_fp16")]; tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_87_cast_fp16 = mul(x = var_859_cast_fp16, y = var_860_to_fp16)[name = tensor("aw_chunk_87_cast_fp16")]; tensor var_863_equation_0 = const()[name = tensor("op_863_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_863_cast_fp16 = einsum(equation = var_863_equation_0, values = (var_637_cast_fp16, var_567_cast_fp16))[name = tensor("op_863_cast_fp16")]; tensor var_864_to_fp16 = const()[name = tensor("op_864_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_89_cast_fp16 = mul(x = var_863_cast_fp16, y = var_864_to_fp16)[name = tensor("aw_chunk_89_cast_fp16")]; tensor var_867_equation_0 = const()[name = tensor("op_867_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_867_cast_fp16 = einsum(equation = var_867_equation_0, values = (var_637_cast_fp16, var_574_cast_fp16))[name = tensor("op_867_cast_fp16")]; tensor var_868_to_fp16 = const()[name = tensor("op_868_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_91_cast_fp16 = mul(x = var_867_cast_fp16, y = var_868_to_fp16)[name = tensor("aw_chunk_91_cast_fp16")]; tensor var_871_equation_0 = const()[name = tensor("op_871_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_871_cast_fp16 = einsum(equation = var_871_equation_0, values = (var_637_cast_fp16, var_581_cast_fp16))[name = tensor("op_871_cast_fp16")]; tensor var_872_to_fp16 = const()[name = tensor("op_872_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_93_cast_fp16 = mul(x = var_871_cast_fp16, y = var_872_to_fp16)[name = tensor("aw_chunk_93_cast_fp16")]; tensor var_875_equation_0 = const()[name = tensor("op_875_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_875_cast_fp16 = einsum(equation = var_875_equation_0, values = (var_637_cast_fp16, var_588_cast_fp16))[name = tensor("op_875_cast_fp16")]; tensor var_876_to_fp16 = const()[name = tensor("op_876_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_95_cast_fp16 = mul(x = var_875_cast_fp16, y = var_876_to_fp16)[name = tensor("aw_chunk_95_cast_fp16")]; tensor var_878_cast_fp16 = softmax(axis = var_151, x = aw_chunk_1_cast_fp16)[name = tensor("op_878_cast_fp16")]; tensor var_879_cast_fp16 = softmax(axis = var_151, x = aw_chunk_3_cast_fp16)[name = tensor("op_879_cast_fp16")]; tensor var_880_cast_fp16 = softmax(axis = var_151, x = aw_chunk_5_cast_fp16)[name = tensor("op_880_cast_fp16")]; tensor var_881_cast_fp16 = softmax(axis = var_151, x = aw_chunk_7_cast_fp16)[name = tensor("op_881_cast_fp16")]; tensor var_882_cast_fp16 = softmax(axis = var_151, x = aw_chunk_9_cast_fp16)[name = tensor("op_882_cast_fp16")]; tensor var_883_cast_fp16 = softmax(axis = var_151, x = aw_chunk_11_cast_fp16)[name = tensor("op_883_cast_fp16")]; tensor var_884_cast_fp16 = softmax(axis = var_151, x = aw_chunk_13_cast_fp16)[name = tensor("op_884_cast_fp16")]; tensor var_885_cast_fp16 = softmax(axis = var_151, x = aw_chunk_15_cast_fp16)[name = tensor("op_885_cast_fp16")]; tensor var_886_cast_fp16 = softmax(axis = var_151, x = aw_chunk_17_cast_fp16)[name = tensor("op_886_cast_fp16")]; tensor var_887_cast_fp16 = softmax(axis = var_151, x = aw_chunk_19_cast_fp16)[name = tensor("op_887_cast_fp16")]; tensor var_888_cast_fp16 = softmax(axis = var_151, x = aw_chunk_21_cast_fp16)[name = tensor("op_888_cast_fp16")]; tensor var_889_cast_fp16 = softmax(axis = var_151, x = aw_chunk_23_cast_fp16)[name = tensor("op_889_cast_fp16")]; tensor var_890_cast_fp16 = softmax(axis = var_151, x = aw_chunk_25_cast_fp16)[name = tensor("op_890_cast_fp16")]; tensor var_891_cast_fp16 = softmax(axis = var_151, x = aw_chunk_27_cast_fp16)[name = tensor("op_891_cast_fp16")]; tensor var_892_cast_fp16 = softmax(axis = var_151, x = aw_chunk_29_cast_fp16)[name = tensor("op_892_cast_fp16")]; tensor var_893_cast_fp16 = softmax(axis = var_151, x = aw_chunk_31_cast_fp16)[name = tensor("op_893_cast_fp16")]; tensor var_894_cast_fp16 = softmax(axis = var_151, x = aw_chunk_33_cast_fp16)[name = tensor("op_894_cast_fp16")]; tensor var_895_cast_fp16 = softmax(axis = var_151, x = aw_chunk_35_cast_fp16)[name = tensor("op_895_cast_fp16")]; tensor var_896_cast_fp16 = softmax(axis = var_151, x = aw_chunk_37_cast_fp16)[name = tensor("op_896_cast_fp16")]; tensor var_897_cast_fp16 = softmax(axis = var_151, x = aw_chunk_39_cast_fp16)[name = tensor("op_897_cast_fp16")]; tensor var_898_cast_fp16 = softmax(axis = var_151, x = aw_chunk_41_cast_fp16)[name = tensor("op_898_cast_fp16")]; tensor var_899_cast_fp16 = softmax(axis = var_151, x = aw_chunk_43_cast_fp16)[name = tensor("op_899_cast_fp16")]; tensor var_900_cast_fp16 = softmax(axis = var_151, x = aw_chunk_45_cast_fp16)[name = tensor("op_900_cast_fp16")]; tensor var_901_cast_fp16 = softmax(axis = var_151, x = aw_chunk_47_cast_fp16)[name = tensor("op_901_cast_fp16")]; tensor var_902_cast_fp16 = softmax(axis = var_151, x = aw_chunk_49_cast_fp16)[name = tensor("op_902_cast_fp16")]; tensor var_903_cast_fp16 = softmax(axis = var_151, x = aw_chunk_51_cast_fp16)[name = tensor("op_903_cast_fp16")]; tensor var_904_cast_fp16 = softmax(axis = var_151, x = aw_chunk_53_cast_fp16)[name = tensor("op_904_cast_fp16")]; tensor var_905_cast_fp16 = softmax(axis = var_151, x = aw_chunk_55_cast_fp16)[name = tensor("op_905_cast_fp16")]; tensor var_906_cast_fp16 = softmax(axis = var_151, x = aw_chunk_57_cast_fp16)[name = tensor("op_906_cast_fp16")]; tensor var_907_cast_fp16 = softmax(axis = var_151, x = aw_chunk_59_cast_fp16)[name = tensor("op_907_cast_fp16")]; tensor var_908_cast_fp16 = softmax(axis = var_151, x = aw_chunk_61_cast_fp16)[name = tensor("op_908_cast_fp16")]; tensor var_909_cast_fp16 = softmax(axis = var_151, x = aw_chunk_63_cast_fp16)[name = tensor("op_909_cast_fp16")]; tensor var_910_cast_fp16 = softmax(axis = var_151, x = aw_chunk_65_cast_fp16)[name = tensor("op_910_cast_fp16")]; tensor var_911_cast_fp16 = softmax(axis = var_151, x = aw_chunk_67_cast_fp16)[name = tensor("op_911_cast_fp16")]; tensor var_912_cast_fp16 = softmax(axis = var_151, x = aw_chunk_69_cast_fp16)[name = tensor("op_912_cast_fp16")]; tensor var_913_cast_fp16 = softmax(axis = var_151, x = aw_chunk_71_cast_fp16)[name = tensor("op_913_cast_fp16")]; tensor var_914_cast_fp16 = softmax(axis = var_151, x = aw_chunk_73_cast_fp16)[name = tensor("op_914_cast_fp16")]; tensor var_915_cast_fp16 = softmax(axis = var_151, x = aw_chunk_75_cast_fp16)[name = tensor("op_915_cast_fp16")]; tensor var_916_cast_fp16 = softmax(axis = var_151, x = aw_chunk_77_cast_fp16)[name = tensor("op_916_cast_fp16")]; tensor var_917_cast_fp16 = softmax(axis = var_151, x = aw_chunk_79_cast_fp16)[name = tensor("op_917_cast_fp16")]; tensor var_918_cast_fp16 = softmax(axis = var_151, x = aw_chunk_81_cast_fp16)[name = tensor("op_918_cast_fp16")]; tensor var_919_cast_fp16 = softmax(axis = var_151, x = aw_chunk_83_cast_fp16)[name = tensor("op_919_cast_fp16")]; tensor var_920_cast_fp16 = softmax(axis = var_151, x = aw_chunk_85_cast_fp16)[name = tensor("op_920_cast_fp16")]; tensor var_921_cast_fp16 = softmax(axis = var_151, x = aw_chunk_87_cast_fp16)[name = tensor("op_921_cast_fp16")]; tensor var_922_cast_fp16 = softmax(axis = var_151, x = aw_chunk_89_cast_fp16)[name = tensor("op_922_cast_fp16")]; tensor var_923_cast_fp16 = softmax(axis = var_151, x = aw_chunk_91_cast_fp16)[name = tensor("op_923_cast_fp16")]; tensor var_924_cast_fp16 = softmax(axis = var_151, x = aw_chunk_93_cast_fp16)[name = tensor("op_924_cast_fp16")]; tensor var_925_cast_fp16 = softmax(axis = var_151, x = aw_chunk_95_cast_fp16)[name = tensor("op_925_cast_fp16")]; tensor var_927_equation_0 = const()[name = tensor("op_927_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_927_cast_fp16 = einsum(equation = var_927_equation_0, values = (var_639_cast_fp16, var_878_cast_fp16))[name = tensor("op_927_cast_fp16")]; tensor var_929_equation_0 = const()[name = tensor("op_929_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_929_cast_fp16 = einsum(equation = var_929_equation_0, values = (var_639_cast_fp16, var_879_cast_fp16))[name = tensor("op_929_cast_fp16")]; tensor var_931_equation_0 = const()[name = tensor("op_931_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_931_cast_fp16 = einsum(equation = var_931_equation_0, values = (var_639_cast_fp16, var_880_cast_fp16))[name = tensor("op_931_cast_fp16")]; tensor var_933_equation_0 = const()[name = tensor("op_933_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_933_cast_fp16 = einsum(equation = var_933_equation_0, values = (var_639_cast_fp16, var_881_cast_fp16))[name = tensor("op_933_cast_fp16")]; tensor var_935_equation_0 = const()[name = tensor("op_935_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_935_cast_fp16 = einsum(equation = var_935_equation_0, values = (var_643_cast_fp16, var_882_cast_fp16))[name = tensor("op_935_cast_fp16")]; tensor var_937_equation_0 = const()[name = tensor("op_937_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_937_cast_fp16 = einsum(equation = var_937_equation_0, values = (var_643_cast_fp16, var_883_cast_fp16))[name = tensor("op_937_cast_fp16")]; tensor var_939_equation_0 = const()[name = tensor("op_939_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_939_cast_fp16 = einsum(equation = var_939_equation_0, values = (var_643_cast_fp16, var_884_cast_fp16))[name = tensor("op_939_cast_fp16")]; tensor var_941_equation_0 = const()[name = tensor("op_941_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_941_cast_fp16 = einsum(equation = var_941_equation_0, values = (var_643_cast_fp16, var_885_cast_fp16))[name = tensor("op_941_cast_fp16")]; tensor var_943_equation_0 = const()[name = tensor("op_943_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_943_cast_fp16 = einsum(equation = var_943_equation_0, values = (var_647_cast_fp16, var_886_cast_fp16))[name = tensor("op_943_cast_fp16")]; tensor var_945_equation_0 = const()[name = tensor("op_945_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_945_cast_fp16 = einsum(equation = var_945_equation_0, values = (var_647_cast_fp16, var_887_cast_fp16))[name = tensor("op_945_cast_fp16")]; tensor var_947_equation_0 = const()[name = tensor("op_947_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_947_cast_fp16 = einsum(equation = var_947_equation_0, values = (var_647_cast_fp16, var_888_cast_fp16))[name = tensor("op_947_cast_fp16")]; tensor var_949_equation_0 = const()[name = tensor("op_949_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_949_cast_fp16 = einsum(equation = var_949_equation_0, values = (var_647_cast_fp16, var_889_cast_fp16))[name = tensor("op_949_cast_fp16")]; tensor var_951_equation_0 = const()[name = tensor("op_951_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_951_cast_fp16 = einsum(equation = var_951_equation_0, values = (var_651_cast_fp16, var_890_cast_fp16))[name = tensor("op_951_cast_fp16")]; tensor var_953_equation_0 = const()[name = tensor("op_953_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_953_cast_fp16 = einsum(equation = var_953_equation_0, values = (var_651_cast_fp16, var_891_cast_fp16))[name = tensor("op_953_cast_fp16")]; tensor var_955_equation_0 = const()[name = tensor("op_955_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_955_cast_fp16 = einsum(equation = var_955_equation_0, values = (var_651_cast_fp16, var_892_cast_fp16))[name = tensor("op_955_cast_fp16")]; tensor var_957_equation_0 = const()[name = tensor("op_957_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_957_cast_fp16 = einsum(equation = var_957_equation_0, values = (var_651_cast_fp16, var_893_cast_fp16))[name = tensor("op_957_cast_fp16")]; tensor var_959_equation_0 = const()[name = tensor("op_959_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_959_cast_fp16 = einsum(equation = var_959_equation_0, values = (var_655_cast_fp16, var_894_cast_fp16))[name = tensor("op_959_cast_fp16")]; tensor var_961_equation_0 = const()[name = tensor("op_961_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_961_cast_fp16 = einsum(equation = var_961_equation_0, values = (var_655_cast_fp16, var_895_cast_fp16))[name = tensor("op_961_cast_fp16")]; tensor var_963_equation_0 = const()[name = tensor("op_963_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_963_cast_fp16 = einsum(equation = var_963_equation_0, values = (var_655_cast_fp16, var_896_cast_fp16))[name = tensor("op_963_cast_fp16")]; tensor var_965_equation_0 = const()[name = tensor("op_965_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_965_cast_fp16 = einsum(equation = var_965_equation_0, values = (var_655_cast_fp16, var_897_cast_fp16))[name = tensor("op_965_cast_fp16")]; tensor var_967_equation_0 = const()[name = tensor("op_967_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_967_cast_fp16 = einsum(equation = var_967_equation_0, values = (var_659_cast_fp16, var_898_cast_fp16))[name = tensor("op_967_cast_fp16")]; tensor var_969_equation_0 = const()[name = tensor("op_969_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_969_cast_fp16 = einsum(equation = var_969_equation_0, values = (var_659_cast_fp16, var_899_cast_fp16))[name = tensor("op_969_cast_fp16")]; tensor var_971_equation_0 = const()[name = tensor("op_971_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_971_cast_fp16 = einsum(equation = var_971_equation_0, values = (var_659_cast_fp16, var_900_cast_fp16))[name = tensor("op_971_cast_fp16")]; tensor var_973_equation_0 = const()[name = tensor("op_973_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_973_cast_fp16 = einsum(equation = var_973_equation_0, values = (var_659_cast_fp16, var_901_cast_fp16))[name = tensor("op_973_cast_fp16")]; tensor var_975_equation_0 = const()[name = tensor("op_975_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_975_cast_fp16 = einsum(equation = var_975_equation_0, values = (var_663_cast_fp16, var_902_cast_fp16))[name = tensor("op_975_cast_fp16")]; tensor var_977_equation_0 = const()[name = tensor("op_977_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_977_cast_fp16 = einsum(equation = var_977_equation_0, values = (var_663_cast_fp16, var_903_cast_fp16))[name = tensor("op_977_cast_fp16")]; tensor var_979_equation_0 = const()[name = tensor("op_979_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_979_cast_fp16 = einsum(equation = var_979_equation_0, values = (var_663_cast_fp16, var_904_cast_fp16))[name = tensor("op_979_cast_fp16")]; tensor var_981_equation_0 = const()[name = tensor("op_981_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_981_cast_fp16 = einsum(equation = var_981_equation_0, values = (var_663_cast_fp16, var_905_cast_fp16))[name = tensor("op_981_cast_fp16")]; tensor var_983_equation_0 = const()[name = tensor("op_983_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_983_cast_fp16 = einsum(equation = var_983_equation_0, values = (var_667_cast_fp16, var_906_cast_fp16))[name = tensor("op_983_cast_fp16")]; tensor var_985_equation_0 = const()[name = tensor("op_985_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_985_cast_fp16 = einsum(equation = var_985_equation_0, values = (var_667_cast_fp16, var_907_cast_fp16))[name = tensor("op_985_cast_fp16")]; tensor var_987_equation_0 = const()[name = tensor("op_987_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_987_cast_fp16 = einsum(equation = var_987_equation_0, values = (var_667_cast_fp16, var_908_cast_fp16))[name = tensor("op_987_cast_fp16")]; tensor var_989_equation_0 = const()[name = tensor("op_989_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_989_cast_fp16 = einsum(equation = var_989_equation_0, values = (var_667_cast_fp16, var_909_cast_fp16))[name = tensor("op_989_cast_fp16")]; tensor var_991_equation_0 = const()[name = tensor("op_991_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_991_cast_fp16 = einsum(equation = var_991_equation_0, values = (var_671_cast_fp16, var_910_cast_fp16))[name = tensor("op_991_cast_fp16")]; tensor var_993_equation_0 = const()[name = tensor("op_993_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_993_cast_fp16 = einsum(equation = var_993_equation_0, values = (var_671_cast_fp16, var_911_cast_fp16))[name = tensor("op_993_cast_fp16")]; tensor var_995_equation_0 = const()[name = tensor("op_995_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_995_cast_fp16 = einsum(equation = var_995_equation_0, values = (var_671_cast_fp16, var_912_cast_fp16))[name = tensor("op_995_cast_fp16")]; tensor var_997_equation_0 = const()[name = tensor("op_997_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_997_cast_fp16 = einsum(equation = var_997_equation_0, values = (var_671_cast_fp16, var_913_cast_fp16))[name = tensor("op_997_cast_fp16")]; tensor var_999_equation_0 = const()[name = tensor("op_999_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_999_cast_fp16 = einsum(equation = var_999_equation_0, values = (var_675_cast_fp16, var_914_cast_fp16))[name = tensor("op_999_cast_fp16")]; tensor var_1001_equation_0 = const()[name = tensor("op_1001_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1001_cast_fp16 = einsum(equation = var_1001_equation_0, values = (var_675_cast_fp16, var_915_cast_fp16))[name = tensor("op_1001_cast_fp16")]; tensor var_1003_equation_0 = const()[name = tensor("op_1003_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1003_cast_fp16 = einsum(equation = var_1003_equation_0, values = (var_675_cast_fp16, var_916_cast_fp16))[name = tensor("op_1003_cast_fp16")]; tensor var_1005_equation_0 = const()[name = tensor("op_1005_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1005_cast_fp16 = einsum(equation = var_1005_equation_0, values = (var_675_cast_fp16, var_917_cast_fp16))[name = tensor("op_1005_cast_fp16")]; tensor var_1007_equation_0 = const()[name = tensor("op_1007_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1007_cast_fp16 = einsum(equation = var_1007_equation_0, values = (var_679_cast_fp16, var_918_cast_fp16))[name = tensor("op_1007_cast_fp16")]; tensor var_1009_equation_0 = const()[name = tensor("op_1009_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1009_cast_fp16 = einsum(equation = var_1009_equation_0, values = (var_679_cast_fp16, var_919_cast_fp16))[name = tensor("op_1009_cast_fp16")]; tensor var_1011_equation_0 = const()[name = tensor("op_1011_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1011_cast_fp16 = einsum(equation = var_1011_equation_0, values = (var_679_cast_fp16, var_920_cast_fp16))[name = tensor("op_1011_cast_fp16")]; tensor var_1013_equation_0 = const()[name = tensor("op_1013_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1013_cast_fp16 = einsum(equation = var_1013_equation_0, values = (var_679_cast_fp16, var_921_cast_fp16))[name = tensor("op_1013_cast_fp16")]; tensor var_1015_equation_0 = const()[name = tensor("op_1015_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1015_cast_fp16 = einsum(equation = var_1015_equation_0, values = (var_683_cast_fp16, var_922_cast_fp16))[name = tensor("op_1015_cast_fp16")]; tensor var_1017_equation_0 = const()[name = tensor("op_1017_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1017_cast_fp16 = einsum(equation = var_1017_equation_0, values = (var_683_cast_fp16, var_923_cast_fp16))[name = tensor("op_1017_cast_fp16")]; tensor var_1019_equation_0 = const()[name = tensor("op_1019_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1019_cast_fp16 = einsum(equation = var_1019_equation_0, values = (var_683_cast_fp16, var_924_cast_fp16))[name = tensor("op_1019_cast_fp16")]; tensor var_1021_equation_0 = const()[name = tensor("op_1021_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1021_cast_fp16 = einsum(equation = var_1021_equation_0, values = (var_683_cast_fp16, var_925_cast_fp16))[name = tensor("op_1021_cast_fp16")]; tensor var_1023_interleave_0 = const()[name = tensor("op_1023_interleave_0"), val = tensor(false)]; tensor var_1023_cast_fp16 = concat(axis = var_134, interleave = var_1023_interleave_0, values = (var_927_cast_fp16, var_929_cast_fp16, var_931_cast_fp16, var_933_cast_fp16))[name = tensor("op_1023_cast_fp16")]; tensor var_1025_interleave_0 = const()[name = tensor("op_1025_interleave_0"), val = tensor(false)]; tensor var_1025_cast_fp16 = concat(axis = var_134, interleave = var_1025_interleave_0, values = (var_935_cast_fp16, var_937_cast_fp16, var_939_cast_fp16, var_941_cast_fp16))[name = tensor("op_1025_cast_fp16")]; tensor var_1027_interleave_0 = const()[name = tensor("op_1027_interleave_0"), val = tensor(false)]; tensor var_1027_cast_fp16 = concat(axis = var_134, interleave = var_1027_interleave_0, values = (var_943_cast_fp16, var_945_cast_fp16, var_947_cast_fp16, var_949_cast_fp16))[name = tensor("op_1027_cast_fp16")]; tensor var_1029_interleave_0 = const()[name = tensor("op_1029_interleave_0"), val = tensor(false)]; tensor var_1029_cast_fp16 = concat(axis = var_134, interleave = var_1029_interleave_0, values = (var_951_cast_fp16, var_953_cast_fp16, var_955_cast_fp16, var_957_cast_fp16))[name = tensor("op_1029_cast_fp16")]; tensor var_1031_interleave_0 = const()[name = tensor("op_1031_interleave_0"), val = tensor(false)]; tensor var_1031_cast_fp16 = concat(axis = var_134, interleave = var_1031_interleave_0, values = (var_959_cast_fp16, var_961_cast_fp16, var_963_cast_fp16, var_965_cast_fp16))[name = tensor("op_1031_cast_fp16")]; tensor var_1033_interleave_0 = const()[name = tensor("op_1033_interleave_0"), val = tensor(false)]; tensor var_1033_cast_fp16 = concat(axis = var_134, interleave = var_1033_interleave_0, values = (var_967_cast_fp16, var_969_cast_fp16, var_971_cast_fp16, var_973_cast_fp16))[name = tensor("op_1033_cast_fp16")]; tensor var_1035_interleave_0 = const()[name = tensor("op_1035_interleave_0"), val = tensor(false)]; tensor var_1035_cast_fp16 = concat(axis = var_134, interleave = var_1035_interleave_0, values = (var_975_cast_fp16, var_977_cast_fp16, var_979_cast_fp16, var_981_cast_fp16))[name = tensor("op_1035_cast_fp16")]; tensor var_1037_interleave_0 = const()[name = tensor("op_1037_interleave_0"), val = tensor(false)]; tensor var_1037_cast_fp16 = concat(axis = var_134, interleave = var_1037_interleave_0, values = (var_983_cast_fp16, var_985_cast_fp16, var_987_cast_fp16, var_989_cast_fp16))[name = tensor("op_1037_cast_fp16")]; tensor var_1039_interleave_0 = const()[name = tensor("op_1039_interleave_0"), val = tensor(false)]; tensor var_1039_cast_fp16 = concat(axis = var_134, interleave = var_1039_interleave_0, values = (var_991_cast_fp16, var_993_cast_fp16, var_995_cast_fp16, var_997_cast_fp16))[name = tensor("op_1039_cast_fp16")]; tensor var_1041_interleave_0 = const()[name = tensor("op_1041_interleave_0"), val = tensor(false)]; tensor var_1041_cast_fp16 = concat(axis = var_134, interleave = var_1041_interleave_0, values = (var_999_cast_fp16, var_1001_cast_fp16, var_1003_cast_fp16, var_1005_cast_fp16))[name = tensor("op_1041_cast_fp16")]; tensor var_1043_interleave_0 = const()[name = tensor("op_1043_interleave_0"), val = tensor(false)]; tensor var_1043_cast_fp16 = concat(axis = var_134, interleave = var_1043_interleave_0, values = (var_1007_cast_fp16, var_1009_cast_fp16, var_1011_cast_fp16, var_1013_cast_fp16))[name = tensor("op_1043_cast_fp16")]; tensor var_1045_interleave_0 = const()[name = tensor("op_1045_interleave_0"), val = tensor(false)]; tensor var_1045_cast_fp16 = concat(axis = var_134, interleave = var_1045_interleave_0, values = (var_1015_cast_fp16, var_1017_cast_fp16, var_1019_cast_fp16, var_1021_cast_fp16))[name = tensor("op_1045_cast_fp16")]; tensor input_1_interleave_0 = const()[name = tensor("input_1_interleave_0"), val = tensor(false)]; tensor input_1_cast_fp16 = concat(axis = var_151, interleave = input_1_interleave_0, values = (var_1023_cast_fp16, var_1025_cast_fp16, var_1027_cast_fp16, var_1029_cast_fp16, var_1031_cast_fp16, var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16, var_1039_cast_fp16, var_1041_cast_fp16, var_1043_cast_fp16, var_1045_cast_fp16))[name = tensor("input_1_cast_fp16")]; tensor var_1050 = const()[name = tensor("op_1050"), val = tensor([1, 1])]; tensor var_1052 = const()[name = tensor("op_1052"), val = tensor([1, 1])]; tensor obj_3_pad_type_0 = const()[name = tensor("obj_3_pad_type_0"), val = tensor("custom")]; tensor obj_3_pad_0 = const()[name = tensor("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9763776)))]; tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10943488)))]; tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = var_1052, groups = var_151, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = var_1050, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_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_1058 = const()[name = tensor("op_1058"), val = tensor([1])]; tensor channels_mean_3_cast_fp16 = reduce_mean(axes = var_1058, keep_dims = var_152, 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_1062 = const()[name = tensor("op_1062"), val = tensor([1])]; tensor var_1063_cast_fp16 = reduce_mean(axes = var_1062, keep_dims = var_152, x = zero_mean_sq_3_cast_fp16)[name = tensor("op_1063_cast_fp16")]; tensor var_1064_to_fp16 = const()[name = tensor("op_1064_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_1065_cast_fp16 = add(x = var_1063_cast_fp16, y = var_1064_to_fp16)[name = tensor("op_1065_cast_fp16")]; tensor denom_3_epsilon_0_to_fp16 = const()[name = tensor("denom_3_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_3_cast_fp16 = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_1065_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_3_gamma_0_to_fp16 = const()[name = tensor("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10945088)))]; tensor input_3_beta_0_to_fp16 = const()[name = tensor("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10946688)))]; tensor input_3_epsilon_0_to_fp16 = const()[name = tensor("input_3_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_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_3_cast_fp16")]; tensor var_1076 = const()[name = tensor("op_1076"), val = tensor([1, 1])]; tensor var_1078 = const()[name = tensor("op_1078"), 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_fc1_weight_to_fp16 = const()[name = tensor("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10948288)))]; tensor layers_0_fc1_bias_to_fp16 = const()[name = tensor("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15666944)))]; tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = var_1078, groups = var_151, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = var_1076, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; tensor input_7_mode_0 = const()[name = tensor("input_7_mode_0"), val = tensor("EXACT")]; tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; tensor var_1084 = const()[name = tensor("op_1084"), val = tensor([1, 1])]; tensor var_1086 = const()[name = tensor("op_1086"), val = tensor([1, 1])]; tensor hidden_states_5_pad_type_0 = const()[name = tensor("hidden_states_5_pad_type_0"), val = tensor("custom")]; tensor hidden_states_5_pad_0 = const()[name = tensor("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_0_fc2_weight_to_fp16 = const()[name = tensor("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15673152)))]; tensor layers_0_fc2_bias_to_fp16 = const()[name = tensor("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20391808)))]; tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = var_1086, groups = var_151, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = var_1084, weight = layers_0_fc2_weight_to_fp16, x = input_7_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_1093 = const()[name = tensor("op_1093"), val = tensor(3)]; tensor var_1110 = const()[name = tensor("op_1110"), val = tensor(1)]; tensor var_1111 = const()[name = tensor("op_1111"), val = tensor(true)]; tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([1])]; tensor channels_mean_5_cast_fp16 = reduce_mean(axes = var_1121, keep_dims = var_1111, 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_1125 = const()[name = tensor("op_1125"), val = tensor([1])]; tensor var_1126_cast_fp16 = reduce_mean(axes = var_1125, keep_dims = var_1111, x = zero_mean_sq_5_cast_fp16)[name = tensor("op_1126_cast_fp16")]; tensor var_1127_to_fp16 = const()[name = tensor("op_1127_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_1128_cast_fp16 = add(x = var_1126_cast_fp16, y = var_1127_to_fp16)[name = tensor("op_1128_cast_fp16")]; tensor denom_5_epsilon_0_to_fp16 = const()[name = tensor("denom_5_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_5_cast_fp16 = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_1128_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(20393408)))]; 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(20395008)))]; 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_1143 = const()[name = tensor("op_1143"), val = tensor([1, 1])]; tensor var_1145 = const()[name = tensor("op_1145"), val = tensor([1, 1])]; tensor query_3_pad_type_0 = const()[name = tensor("query_3_pad_type_0"), val = tensor("custom")]; tensor query_3_pad_0 = const()[name = tensor("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20396608)))]; tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21576320)))]; tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = var_1145, groups = var_1110, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = var_1143, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("query_3_cast_fp16")]; tensor var_1149 = const()[name = tensor("op_1149"), val = tensor([1, 1])]; tensor var_1151 = const()[name = tensor("op_1151"), val = tensor([1, 1])]; tensor key_3_pad_type_0 = const()[name = tensor("key_3_pad_type_0"), val = tensor("custom")]; tensor key_3_pad_0 = const()[name = tensor("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21577920)))]; tensor key_3_cast_fp16 = conv(dilations = var_1151, groups = var_1110, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = var_1149, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("key_3_cast_fp16")]; tensor var_1156 = const()[name = tensor("op_1156"), val = tensor([1, 1])]; tensor var_1158 = const()[name = tensor("op_1158"), val = tensor([1, 1])]; tensor value_3_pad_type_0 = const()[name = tensor("value_3_pad_type_0"), val = tensor("custom")]; tensor value_3_pad_0 = const()[name = tensor("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22757632)))]; tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23937344)))]; tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = var_1158, groups = var_1110, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = var_1156, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = tensor("value_3_cast_fp16")]; tensor var_1165_begin_0 = const()[name = tensor("op_1165_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1165_end_0 = const()[name = tensor("op_1165_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1165_end_mask_0 = const()[name = tensor("op_1165_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1165_cast_fp16 = slice_by_index(begin = var_1165_begin_0, end = var_1165_end_0, end_mask = var_1165_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1165_cast_fp16")]; tensor var_1169_begin_0 = const()[name = tensor("op_1169_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_1169_end_0 = const()[name = tensor("op_1169_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_1169_end_mask_0 = const()[name = tensor("op_1169_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1169_cast_fp16 = slice_by_index(begin = var_1169_begin_0, end = var_1169_end_0, end_mask = var_1169_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1169_cast_fp16")]; tensor var_1173_begin_0 = const()[name = tensor("op_1173_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_1173_end_0 = const()[name = tensor("op_1173_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_1173_end_mask_0 = const()[name = tensor("op_1173_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1173_cast_fp16 = slice_by_index(begin = var_1173_begin_0, end = var_1173_end_0, end_mask = var_1173_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1173_cast_fp16")]; tensor var_1177_begin_0 = const()[name = tensor("op_1177_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_1177_end_0 = const()[name = tensor("op_1177_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_1177_end_mask_0 = const()[name = tensor("op_1177_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1177_cast_fp16 = slice_by_index(begin = var_1177_begin_0, end = var_1177_end_0, end_mask = var_1177_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1177_cast_fp16")]; tensor var_1181_begin_0 = const()[name = tensor("op_1181_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_1181_end_0 = const()[name = tensor("op_1181_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_1181_end_mask_0 = const()[name = tensor("op_1181_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1181_cast_fp16 = slice_by_index(begin = var_1181_begin_0, end = var_1181_end_0, end_mask = var_1181_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1181_cast_fp16")]; tensor var_1185_begin_0 = const()[name = tensor("op_1185_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1185_end_0 = const()[name = tensor("op_1185_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_1185_end_mask_0 = const()[name = tensor("op_1185_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1185_cast_fp16 = slice_by_index(begin = var_1185_begin_0, end = var_1185_end_0, end_mask = var_1185_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1185_cast_fp16")]; tensor var_1189_begin_0 = const()[name = tensor("op_1189_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_1189_end_0 = const()[name = tensor("op_1189_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_1189_end_mask_0 = const()[name = tensor("op_1189_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1189_cast_fp16 = slice_by_index(begin = var_1189_begin_0, end = var_1189_end_0, end_mask = var_1189_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1189_cast_fp16")]; tensor var_1193_begin_0 = const()[name = tensor("op_1193_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_1193_end_0 = const()[name = tensor("op_1193_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_1193_end_mask_0 = const()[name = tensor("op_1193_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1193_cast_fp16 = slice_by_index(begin = var_1193_begin_0, end = var_1193_end_0, end_mask = var_1193_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1193_cast_fp16")]; tensor var_1197_begin_0 = const()[name = tensor("op_1197_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_1197_end_0 = const()[name = tensor("op_1197_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_1197_end_mask_0 = const()[name = tensor("op_1197_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1197_cast_fp16 = slice_by_index(begin = var_1197_begin_0, end = var_1197_end_0, end_mask = var_1197_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1197_cast_fp16")]; tensor var_1201_begin_0 = const()[name = tensor("op_1201_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_1201_end_0 = const()[name = tensor("op_1201_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_1201_end_mask_0 = const()[name = tensor("op_1201_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1201_cast_fp16")]; tensor var_1205_begin_0 = const()[name = tensor("op_1205_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_1205_end_0 = const()[name = tensor("op_1205_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_1205_end_mask_0 = const()[name = tensor("op_1205_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1205_cast_fp16 = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1205_cast_fp16")]; tensor var_1209_begin_0 = const()[name = tensor("op_1209_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_1209_end_0 = const()[name = tensor("op_1209_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_1209_end_mask_0 = const()[name = tensor("op_1209_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1209_cast_fp16 = slice_by_index(begin = var_1209_begin_0, end = var_1209_end_0, end_mask = var_1209_end_mask_0, x = query_3_cast_fp16)[name = tensor("op_1209_cast_fp16")]; tensor var_1218_begin_0 = const()[name = tensor("op_1218_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1218_end_0 = const()[name = tensor("op_1218_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1218_end_mask_0 = const()[name = tensor("op_1218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1218_cast_fp16 = slice_by_index(begin = var_1218_begin_0, end = var_1218_end_0, end_mask = var_1218_end_mask_0, x = var_1165_cast_fp16)[name = tensor("op_1218_cast_fp16")]; tensor var_1225_begin_0 = const()[name = tensor("op_1225_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1225_end_0 = const()[name = tensor("op_1225_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1225_end_mask_0 = const()[name = tensor("op_1225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1225_cast_fp16 = slice_by_index(begin = var_1225_begin_0, end = var_1225_end_0, end_mask = var_1225_end_mask_0, x = var_1165_cast_fp16)[name = tensor("op_1225_cast_fp16")]; tensor var_1232_begin_0 = const()[name = tensor("op_1232_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1232_end_0 = const()[name = tensor("op_1232_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1232_end_mask_0 = const()[name = tensor("op_1232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1232_cast_fp16 = slice_by_index(begin = var_1232_begin_0, end = var_1232_end_0, end_mask = var_1232_end_mask_0, x = var_1165_cast_fp16)[name = tensor("op_1232_cast_fp16")]; tensor var_1239_begin_0 = const()[name = tensor("op_1239_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1239_end_0 = const()[name = tensor("op_1239_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1239_end_mask_0 = const()[name = tensor("op_1239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1239_cast_fp16 = slice_by_index(begin = var_1239_begin_0, end = var_1239_end_0, end_mask = var_1239_end_mask_0, x = var_1165_cast_fp16)[name = tensor("op_1239_cast_fp16")]; tensor var_1246_begin_0 = const()[name = tensor("op_1246_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1246_end_0 = const()[name = tensor("op_1246_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1246_end_mask_0 = const()[name = tensor("op_1246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1246_cast_fp16 = slice_by_index(begin = var_1246_begin_0, end = var_1246_end_0, end_mask = var_1246_end_mask_0, x = var_1169_cast_fp16)[name = tensor("op_1246_cast_fp16")]; tensor var_1253_begin_0 = const()[name = tensor("op_1253_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1253_end_0 = const()[name = tensor("op_1253_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1253_end_mask_0 = const()[name = tensor("op_1253_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1253_cast_fp16 = slice_by_index(begin = var_1253_begin_0, end = var_1253_end_0, end_mask = var_1253_end_mask_0, x = var_1169_cast_fp16)[name = tensor("op_1253_cast_fp16")]; tensor var_1260_begin_0 = const()[name = tensor("op_1260_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1260_end_0 = const()[name = tensor("op_1260_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1260_end_mask_0 = const()[name = tensor("op_1260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1260_cast_fp16 = slice_by_index(begin = var_1260_begin_0, end = var_1260_end_0, end_mask = var_1260_end_mask_0, x = var_1169_cast_fp16)[name = tensor("op_1260_cast_fp16")]; tensor var_1267_begin_0 = const()[name = tensor("op_1267_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1267_end_0 = const()[name = tensor("op_1267_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1267_end_mask_0 = const()[name = tensor("op_1267_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1267_cast_fp16 = slice_by_index(begin = var_1267_begin_0, end = var_1267_end_0, end_mask = var_1267_end_mask_0, x = var_1169_cast_fp16)[name = tensor("op_1267_cast_fp16")]; tensor var_1274_begin_0 = const()[name = tensor("op_1274_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1274_end_0 = const()[name = tensor("op_1274_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1274_end_mask_0 = const()[name = tensor("op_1274_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1274_cast_fp16 = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = var_1173_cast_fp16)[name = tensor("op_1274_cast_fp16")]; tensor var_1281_begin_0 = const()[name = tensor("op_1281_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1281_end_0 = const()[name = tensor("op_1281_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1281_end_mask_0 = const()[name = tensor("op_1281_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1281_cast_fp16 = slice_by_index(begin = var_1281_begin_0, end = var_1281_end_0, end_mask = var_1281_end_mask_0, x = var_1173_cast_fp16)[name = tensor("op_1281_cast_fp16")]; tensor var_1288_begin_0 = const()[name = tensor("op_1288_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1288_end_0 = const()[name = tensor("op_1288_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1288_end_mask_0 = const()[name = tensor("op_1288_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1288_cast_fp16 = slice_by_index(begin = var_1288_begin_0, end = var_1288_end_0, end_mask = var_1288_end_mask_0, x = var_1173_cast_fp16)[name = tensor("op_1288_cast_fp16")]; tensor var_1295_begin_0 = const()[name = tensor("op_1295_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1295_end_0 = const()[name = tensor("op_1295_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1295_end_mask_0 = const()[name = tensor("op_1295_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1295_cast_fp16 = slice_by_index(begin = var_1295_begin_0, end = var_1295_end_0, end_mask = var_1295_end_mask_0, x = var_1173_cast_fp16)[name = tensor("op_1295_cast_fp16")]; tensor var_1302_begin_0 = const()[name = tensor("op_1302_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1302_end_0 = const()[name = tensor("op_1302_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1302_end_mask_0 = const()[name = tensor("op_1302_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1302_cast_fp16 = slice_by_index(begin = var_1302_begin_0, end = var_1302_end_0, end_mask = var_1302_end_mask_0, x = var_1177_cast_fp16)[name = tensor("op_1302_cast_fp16")]; tensor var_1309_begin_0 = const()[name = tensor("op_1309_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1309_end_0 = const()[name = tensor("op_1309_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1309_end_mask_0 = const()[name = tensor("op_1309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1309_cast_fp16 = slice_by_index(begin = var_1309_begin_0, end = var_1309_end_0, end_mask = var_1309_end_mask_0, x = var_1177_cast_fp16)[name = tensor("op_1309_cast_fp16")]; tensor var_1316_begin_0 = const()[name = tensor("op_1316_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1316_end_0 = const()[name = tensor("op_1316_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1316_end_mask_0 = const()[name = tensor("op_1316_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1316_cast_fp16 = slice_by_index(begin = var_1316_begin_0, end = var_1316_end_0, end_mask = var_1316_end_mask_0, x = var_1177_cast_fp16)[name = tensor("op_1316_cast_fp16")]; tensor var_1323_begin_0 = const()[name = tensor("op_1323_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1323_end_0 = const()[name = tensor("op_1323_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1323_end_mask_0 = const()[name = tensor("op_1323_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1323_cast_fp16 = slice_by_index(begin = var_1323_begin_0, end = var_1323_end_0, end_mask = var_1323_end_mask_0, x = var_1177_cast_fp16)[name = tensor("op_1323_cast_fp16")]; tensor var_1330_begin_0 = const()[name = tensor("op_1330_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1330_end_0 = const()[name = tensor("op_1330_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1330_end_mask_0 = const()[name = tensor("op_1330_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1330_cast_fp16 = slice_by_index(begin = var_1330_begin_0, end = var_1330_end_0, end_mask = var_1330_end_mask_0, x = var_1181_cast_fp16)[name = tensor("op_1330_cast_fp16")]; tensor var_1337_begin_0 = const()[name = tensor("op_1337_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1337_end_0 = const()[name = tensor("op_1337_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1337_end_mask_0 = const()[name = tensor("op_1337_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1337_cast_fp16 = slice_by_index(begin = var_1337_begin_0, end = var_1337_end_0, end_mask = var_1337_end_mask_0, x = var_1181_cast_fp16)[name = tensor("op_1337_cast_fp16")]; tensor var_1344_begin_0 = const()[name = tensor("op_1344_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1344_end_0 = const()[name = tensor("op_1344_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1344_end_mask_0 = const()[name = tensor("op_1344_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1344_cast_fp16 = slice_by_index(begin = var_1344_begin_0, end = var_1344_end_0, end_mask = var_1344_end_mask_0, x = var_1181_cast_fp16)[name = tensor("op_1344_cast_fp16")]; tensor var_1351_begin_0 = const()[name = tensor("op_1351_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1351_end_0 = const()[name = tensor("op_1351_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1351_end_mask_0 = const()[name = tensor("op_1351_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1351_cast_fp16 = slice_by_index(begin = var_1351_begin_0, end = var_1351_end_0, end_mask = var_1351_end_mask_0, x = var_1181_cast_fp16)[name = tensor("op_1351_cast_fp16")]; tensor var_1358_begin_0 = const()[name = tensor("op_1358_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1358_end_0 = const()[name = tensor("op_1358_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1358_end_mask_0 = const()[name = tensor("op_1358_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1358_cast_fp16 = slice_by_index(begin = var_1358_begin_0, end = var_1358_end_0, end_mask = var_1358_end_mask_0, x = var_1185_cast_fp16)[name = tensor("op_1358_cast_fp16")]; tensor var_1365_begin_0 = const()[name = tensor("op_1365_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1365_end_0 = const()[name = tensor("op_1365_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1365_end_mask_0 = const()[name = tensor("op_1365_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1365_cast_fp16 = slice_by_index(begin = var_1365_begin_0, end = var_1365_end_0, end_mask = var_1365_end_mask_0, x = var_1185_cast_fp16)[name = tensor("op_1365_cast_fp16")]; tensor var_1372_begin_0 = const()[name = tensor("op_1372_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1372_end_0 = const()[name = tensor("op_1372_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1372_end_mask_0 = const()[name = tensor("op_1372_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1372_cast_fp16 = slice_by_index(begin = var_1372_begin_0, end = var_1372_end_0, end_mask = var_1372_end_mask_0, x = var_1185_cast_fp16)[name = tensor("op_1372_cast_fp16")]; tensor var_1379_begin_0 = const()[name = tensor("op_1379_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1379_end_0 = const()[name = tensor("op_1379_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1379_end_mask_0 = const()[name = tensor("op_1379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1379_cast_fp16 = slice_by_index(begin = var_1379_begin_0, end = var_1379_end_0, end_mask = var_1379_end_mask_0, x = var_1185_cast_fp16)[name = tensor("op_1379_cast_fp16")]; tensor var_1386_begin_0 = const()[name = tensor("op_1386_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1386_end_0 = const()[name = tensor("op_1386_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1386_end_mask_0 = const()[name = tensor("op_1386_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1386_cast_fp16 = slice_by_index(begin = var_1386_begin_0, end = var_1386_end_0, end_mask = var_1386_end_mask_0, x = var_1189_cast_fp16)[name = tensor("op_1386_cast_fp16")]; tensor var_1393_begin_0 = const()[name = tensor("op_1393_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1393_end_0 = const()[name = tensor("op_1393_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1393_end_mask_0 = const()[name = tensor("op_1393_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1393_cast_fp16 = slice_by_index(begin = var_1393_begin_0, end = var_1393_end_0, end_mask = var_1393_end_mask_0, x = var_1189_cast_fp16)[name = tensor("op_1393_cast_fp16")]; tensor var_1400_begin_0 = const()[name = tensor("op_1400_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1400_end_0 = const()[name = tensor("op_1400_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1400_end_mask_0 = const()[name = tensor("op_1400_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1400_cast_fp16 = slice_by_index(begin = var_1400_begin_0, end = var_1400_end_0, end_mask = var_1400_end_mask_0, x = var_1189_cast_fp16)[name = tensor("op_1400_cast_fp16")]; tensor var_1407_begin_0 = const()[name = tensor("op_1407_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1407_end_0 = const()[name = tensor("op_1407_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1407_end_mask_0 = const()[name = tensor("op_1407_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1407_cast_fp16 = slice_by_index(begin = var_1407_begin_0, end = var_1407_end_0, end_mask = var_1407_end_mask_0, x = var_1189_cast_fp16)[name = tensor("op_1407_cast_fp16")]; tensor var_1414_begin_0 = const()[name = tensor("op_1414_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1414_end_0 = const()[name = tensor("op_1414_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1414_end_mask_0 = const()[name = tensor("op_1414_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1414_cast_fp16 = slice_by_index(begin = var_1414_begin_0, end = var_1414_end_0, end_mask = var_1414_end_mask_0, x = var_1193_cast_fp16)[name = tensor("op_1414_cast_fp16")]; tensor var_1421_begin_0 = const()[name = tensor("op_1421_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1421_end_0 = const()[name = tensor("op_1421_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1421_end_mask_0 = const()[name = tensor("op_1421_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1421_cast_fp16 = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = var_1193_cast_fp16)[name = tensor("op_1421_cast_fp16")]; tensor var_1428_begin_0 = const()[name = tensor("op_1428_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1428_end_0 = const()[name = tensor("op_1428_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1428_end_mask_0 = const()[name = tensor("op_1428_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1428_cast_fp16 = slice_by_index(begin = var_1428_begin_0, end = var_1428_end_0, end_mask = var_1428_end_mask_0, x = var_1193_cast_fp16)[name = tensor("op_1428_cast_fp16")]; tensor var_1435_begin_0 = const()[name = tensor("op_1435_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1435_end_0 = const()[name = tensor("op_1435_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1435_end_mask_0 = const()[name = tensor("op_1435_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1435_cast_fp16 = slice_by_index(begin = var_1435_begin_0, end = var_1435_end_0, end_mask = var_1435_end_mask_0, x = var_1193_cast_fp16)[name = tensor("op_1435_cast_fp16")]; tensor var_1442_begin_0 = const()[name = tensor("op_1442_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1442_end_0 = const()[name = tensor("op_1442_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1442_end_mask_0 = const()[name = tensor("op_1442_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1442_cast_fp16 = slice_by_index(begin = var_1442_begin_0, end = var_1442_end_0, end_mask = var_1442_end_mask_0, x = var_1197_cast_fp16)[name = tensor("op_1442_cast_fp16")]; tensor var_1449_begin_0 = const()[name = tensor("op_1449_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1449_end_0 = const()[name = tensor("op_1449_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1449_end_mask_0 = const()[name = tensor("op_1449_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1449_cast_fp16 = slice_by_index(begin = var_1449_begin_0, end = var_1449_end_0, end_mask = var_1449_end_mask_0, x = var_1197_cast_fp16)[name = tensor("op_1449_cast_fp16")]; tensor var_1456_begin_0 = const()[name = tensor("op_1456_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1456_end_0 = const()[name = tensor("op_1456_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1456_end_mask_0 = const()[name = tensor("op_1456_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1456_cast_fp16 = slice_by_index(begin = var_1456_begin_0, end = var_1456_end_0, end_mask = var_1456_end_mask_0, x = var_1197_cast_fp16)[name = tensor("op_1456_cast_fp16")]; tensor var_1463_begin_0 = const()[name = tensor("op_1463_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1463_end_0 = const()[name = tensor("op_1463_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1463_end_mask_0 = const()[name = tensor("op_1463_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1463_cast_fp16 = slice_by_index(begin = var_1463_begin_0, end = var_1463_end_0, end_mask = var_1463_end_mask_0, x = var_1197_cast_fp16)[name = tensor("op_1463_cast_fp16")]; tensor var_1470_begin_0 = const()[name = tensor("op_1470_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1470_end_0 = const()[name = tensor("op_1470_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1470_end_mask_0 = const()[name = tensor("op_1470_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1470_cast_fp16 = slice_by_index(begin = var_1470_begin_0, end = var_1470_end_0, end_mask = var_1470_end_mask_0, x = var_1201_cast_fp16)[name = tensor("op_1470_cast_fp16")]; tensor var_1477_begin_0 = const()[name = tensor("op_1477_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1477_end_0 = const()[name = tensor("op_1477_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1477_end_mask_0 = const()[name = tensor("op_1477_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1477_cast_fp16 = slice_by_index(begin = var_1477_begin_0, end = var_1477_end_0, end_mask = var_1477_end_mask_0, x = var_1201_cast_fp16)[name = tensor("op_1477_cast_fp16")]; tensor var_1484_begin_0 = const()[name = tensor("op_1484_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1484_end_0 = const()[name = tensor("op_1484_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1484_end_mask_0 = const()[name = tensor("op_1484_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1484_cast_fp16 = slice_by_index(begin = var_1484_begin_0, end = var_1484_end_0, end_mask = var_1484_end_mask_0, x = var_1201_cast_fp16)[name = tensor("op_1484_cast_fp16")]; tensor var_1491_begin_0 = const()[name = tensor("op_1491_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1491_end_0 = const()[name = tensor("op_1491_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1491_end_mask_0 = const()[name = tensor("op_1491_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1491_cast_fp16 = slice_by_index(begin = var_1491_begin_0, end = var_1491_end_0, end_mask = var_1491_end_mask_0, x = var_1201_cast_fp16)[name = tensor("op_1491_cast_fp16")]; tensor var_1498_begin_0 = const()[name = tensor("op_1498_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1498_end_0 = const()[name = tensor("op_1498_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1498_end_mask_0 = const()[name = tensor("op_1498_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1498_cast_fp16 = slice_by_index(begin = var_1498_begin_0, end = var_1498_end_0, end_mask = var_1498_end_mask_0, x = var_1205_cast_fp16)[name = tensor("op_1498_cast_fp16")]; tensor var_1505_begin_0 = const()[name = tensor("op_1505_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1505_end_0 = const()[name = tensor("op_1505_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1505_end_mask_0 = const()[name = tensor("op_1505_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1505_cast_fp16 = slice_by_index(begin = var_1505_begin_0, end = var_1505_end_0, end_mask = var_1505_end_mask_0, x = var_1205_cast_fp16)[name = tensor("op_1505_cast_fp16")]; tensor var_1512_begin_0 = const()[name = tensor("op_1512_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1512_end_0 = const()[name = tensor("op_1512_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1512_end_mask_0 = const()[name = tensor("op_1512_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1512_cast_fp16 = slice_by_index(begin = var_1512_begin_0, end = var_1512_end_0, end_mask = var_1512_end_mask_0, x = var_1205_cast_fp16)[name = tensor("op_1512_cast_fp16")]; tensor var_1519_begin_0 = const()[name = tensor("op_1519_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1519_end_0 = const()[name = tensor("op_1519_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1519_end_mask_0 = const()[name = tensor("op_1519_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1519_cast_fp16 = slice_by_index(begin = var_1519_begin_0, end = var_1519_end_0, end_mask = var_1519_end_mask_0, x = var_1205_cast_fp16)[name = tensor("op_1519_cast_fp16")]; tensor var_1526_begin_0 = const()[name = tensor("op_1526_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1526_end_0 = const()[name = tensor("op_1526_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_1526_end_mask_0 = const()[name = tensor("op_1526_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1526_cast_fp16 = slice_by_index(begin = var_1526_begin_0, end = var_1526_end_0, end_mask = var_1526_end_mask_0, x = var_1209_cast_fp16)[name = tensor("op_1526_cast_fp16")]; tensor var_1533_begin_0 = const()[name = tensor("op_1533_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_1533_end_0 = const()[name = tensor("op_1533_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_1533_end_mask_0 = const()[name = tensor("op_1533_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1533_cast_fp16 = slice_by_index(begin = var_1533_begin_0, end = var_1533_end_0, end_mask = var_1533_end_mask_0, x = var_1209_cast_fp16)[name = tensor("op_1533_cast_fp16")]; tensor var_1540_begin_0 = const()[name = tensor("op_1540_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_1540_end_0 = const()[name = tensor("op_1540_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_1540_end_mask_0 = const()[name = tensor("op_1540_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1540_cast_fp16 = slice_by_index(begin = var_1540_begin_0, end = var_1540_end_0, end_mask = var_1540_end_mask_0, x = var_1209_cast_fp16)[name = tensor("op_1540_cast_fp16")]; tensor var_1547_begin_0 = const()[name = tensor("op_1547_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_1547_end_0 = const()[name = tensor("op_1547_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1547_end_mask_0 = const()[name = tensor("op_1547_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1547_cast_fp16 = slice_by_index(begin = var_1547_begin_0, end = var_1547_end_0, end_mask = var_1547_end_mask_0, x = var_1209_cast_fp16)[name = tensor("op_1547_cast_fp16")]; tensor k_3_perm_0 = const()[name = tensor("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_1552_begin_0 = const()[name = tensor("op_1552_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1552_end_0 = const()[name = tensor("op_1552_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_1552_end_mask_0 = const()[name = tensor("op_1552_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_10 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = tensor("transpose_10")]; tensor var_1552_cast_fp16 = slice_by_index(begin = var_1552_begin_0, end = var_1552_end_0, end_mask = var_1552_end_mask_0, x = transpose_10)[name = tensor("op_1552_cast_fp16")]; tensor var_1556_begin_0 = const()[name = tensor("op_1556_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_1556_end_0 = const()[name = tensor("op_1556_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_1556_end_mask_0 = const()[name = tensor("op_1556_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1556_cast_fp16 = slice_by_index(begin = var_1556_begin_0, end = var_1556_end_0, end_mask = var_1556_end_mask_0, x = transpose_10)[name = tensor("op_1556_cast_fp16")]; tensor var_1560_begin_0 = const()[name = tensor("op_1560_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_1560_end_0 = const()[name = tensor("op_1560_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_1560_end_mask_0 = const()[name = tensor("op_1560_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1560_cast_fp16 = slice_by_index(begin = var_1560_begin_0, end = var_1560_end_0, end_mask = var_1560_end_mask_0, x = transpose_10)[name = tensor("op_1560_cast_fp16")]; tensor var_1564_begin_0 = const()[name = tensor("op_1564_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_1564_end_0 = const()[name = tensor("op_1564_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_1564_end_mask_0 = const()[name = tensor("op_1564_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = transpose_10)[name = tensor("op_1564_cast_fp16")]; tensor var_1568_begin_0 = const()[name = tensor("op_1568_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_1568_end_0 = const()[name = tensor("op_1568_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_1568_end_mask_0 = const()[name = tensor("op_1568_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1568_cast_fp16 = slice_by_index(begin = var_1568_begin_0, end = var_1568_end_0, end_mask = var_1568_end_mask_0, x = transpose_10)[name = tensor("op_1568_cast_fp16")]; tensor var_1572_begin_0 = const()[name = tensor("op_1572_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_1572_end_0 = const()[name = tensor("op_1572_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_1572_end_mask_0 = const()[name = tensor("op_1572_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1572_cast_fp16 = slice_by_index(begin = var_1572_begin_0, end = var_1572_end_0, end_mask = var_1572_end_mask_0, x = transpose_10)[name = tensor("op_1572_cast_fp16")]; tensor var_1576_begin_0 = const()[name = tensor("op_1576_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_1576_end_0 = const()[name = tensor("op_1576_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_1576_end_mask_0 = const()[name = tensor("op_1576_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = transpose_10)[name = tensor("op_1576_cast_fp16")]; tensor var_1580_begin_0 = const()[name = tensor("op_1580_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_1580_end_0 = const()[name = tensor("op_1580_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_1580_end_mask_0 = const()[name = tensor("op_1580_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = transpose_10)[name = tensor("op_1580_cast_fp16")]; tensor var_1584_begin_0 = const()[name = tensor("op_1584_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_1584_end_0 = const()[name = tensor("op_1584_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_1584_end_mask_0 = const()[name = tensor("op_1584_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = transpose_10)[name = tensor("op_1584_cast_fp16")]; tensor var_1588_begin_0 = const()[name = tensor("op_1588_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_1588_end_0 = const()[name = tensor("op_1588_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_1588_end_mask_0 = const()[name = tensor("op_1588_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1588_cast_fp16 = slice_by_index(begin = var_1588_begin_0, end = var_1588_end_0, end_mask = var_1588_end_mask_0, x = transpose_10)[name = tensor("op_1588_cast_fp16")]; tensor var_1592_begin_0 = const()[name = tensor("op_1592_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_1592_end_0 = const()[name = tensor("op_1592_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_1592_end_mask_0 = const()[name = tensor("op_1592_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1592_cast_fp16 = slice_by_index(begin = var_1592_begin_0, end = var_1592_end_0, end_mask = var_1592_end_mask_0, x = transpose_10)[name = tensor("op_1592_cast_fp16")]; tensor var_1596_begin_0 = const()[name = tensor("op_1596_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_1596_end_0 = const()[name = tensor("op_1596_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_1596_end_mask_0 = const()[name = tensor("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = transpose_10)[name = tensor("op_1596_cast_fp16")]; tensor var_1598_begin_0 = const()[name = tensor("op_1598_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_1598_end_0 = const()[name = tensor("op_1598_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_1598_end_mask_0 = const()[name = tensor("op_1598_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1598_cast_fp16 = slice_by_index(begin = var_1598_begin_0, end = var_1598_end_0, end_mask = var_1598_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1598_cast_fp16")]; tensor var_1602_begin_0 = const()[name = tensor("op_1602_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_1602_end_0 = const()[name = tensor("op_1602_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_1602_end_mask_0 = const()[name = tensor("op_1602_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1602_cast_fp16 = slice_by_index(begin = var_1602_begin_0, end = var_1602_end_0, end_mask = var_1602_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1602_cast_fp16")]; tensor var_1606_begin_0 = const()[name = tensor("op_1606_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_1606_end_0 = const()[name = tensor("op_1606_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_1606_end_mask_0 = const()[name = tensor("op_1606_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1606_cast_fp16 = slice_by_index(begin = var_1606_begin_0, end = var_1606_end_0, end_mask = var_1606_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1606_cast_fp16")]; tensor var_1610_begin_0 = const()[name = tensor("op_1610_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_1610_end_0 = const()[name = tensor("op_1610_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_1610_end_mask_0 = const()[name = tensor("op_1610_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1610_cast_fp16 = slice_by_index(begin = var_1610_begin_0, end = var_1610_end_0, end_mask = var_1610_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1610_cast_fp16")]; tensor var_1614_begin_0 = const()[name = tensor("op_1614_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_1614_end_0 = const()[name = tensor("op_1614_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_1614_end_mask_0 = const()[name = tensor("op_1614_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1614_cast_fp16 = slice_by_index(begin = var_1614_begin_0, end = var_1614_end_0, end_mask = var_1614_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1614_cast_fp16")]; tensor var_1618_begin_0 = const()[name = tensor("op_1618_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_1618_end_0 = const()[name = tensor("op_1618_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_1618_end_mask_0 = const()[name = tensor("op_1618_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1618_cast_fp16 = slice_by_index(begin = var_1618_begin_0, end = var_1618_end_0, end_mask = var_1618_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1618_cast_fp16")]; tensor var_1622_begin_0 = const()[name = tensor("op_1622_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_1622_end_0 = const()[name = tensor("op_1622_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_1622_end_mask_0 = const()[name = tensor("op_1622_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1622_cast_fp16 = slice_by_index(begin = var_1622_begin_0, end = var_1622_end_0, end_mask = var_1622_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1622_cast_fp16")]; tensor var_1626_begin_0 = const()[name = tensor("op_1626_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_1626_end_0 = const()[name = tensor("op_1626_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_1626_end_mask_0 = const()[name = tensor("op_1626_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1626_cast_fp16 = slice_by_index(begin = var_1626_begin_0, end = var_1626_end_0, end_mask = var_1626_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1626_cast_fp16")]; tensor var_1630_begin_0 = const()[name = tensor("op_1630_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_1630_end_0 = const()[name = tensor("op_1630_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_1630_end_mask_0 = const()[name = tensor("op_1630_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1630_cast_fp16 = slice_by_index(begin = var_1630_begin_0, end = var_1630_end_0, end_mask = var_1630_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1630_cast_fp16")]; tensor var_1634_begin_0 = const()[name = tensor("op_1634_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_1634_end_0 = const()[name = tensor("op_1634_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_1634_end_mask_0 = const()[name = tensor("op_1634_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1634_cast_fp16 = slice_by_index(begin = var_1634_begin_0, end = var_1634_end_0, end_mask = var_1634_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1634_cast_fp16")]; tensor var_1638_begin_0 = const()[name = tensor("op_1638_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_1638_end_0 = const()[name = tensor("op_1638_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_1638_end_mask_0 = const()[name = tensor("op_1638_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1638_cast_fp16")]; tensor var_1642_begin_0 = const()[name = tensor("op_1642_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_1642_end_0 = const()[name = tensor("op_1642_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_1642_end_mask_0 = const()[name = tensor("op_1642_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_1642_cast_fp16 = slice_by_index(begin = var_1642_begin_0, end = var_1642_end_0, end_mask = var_1642_end_mask_0, x = value_3_cast_fp16)[name = tensor("op_1642_cast_fp16")]; tensor var_1646_equation_0 = const()[name = tensor("op_1646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1646_cast_fp16 = einsum(equation = var_1646_equation_0, values = (var_1552_cast_fp16, var_1218_cast_fp16))[name = tensor("op_1646_cast_fp16")]; tensor var_1647_to_fp16 = const()[name = tensor("op_1647_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_97_cast_fp16 = mul(x = var_1646_cast_fp16, y = var_1647_to_fp16)[name = tensor("aw_chunk_97_cast_fp16")]; tensor var_1650_equation_0 = const()[name = tensor("op_1650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1650_cast_fp16 = einsum(equation = var_1650_equation_0, values = (var_1552_cast_fp16, var_1225_cast_fp16))[name = tensor("op_1650_cast_fp16")]; tensor var_1651_to_fp16 = const()[name = tensor("op_1651_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_99_cast_fp16 = mul(x = var_1650_cast_fp16, y = var_1651_to_fp16)[name = tensor("aw_chunk_99_cast_fp16")]; tensor var_1654_equation_0 = const()[name = tensor("op_1654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1654_cast_fp16 = einsum(equation = var_1654_equation_0, values = (var_1552_cast_fp16, var_1232_cast_fp16))[name = tensor("op_1654_cast_fp16")]; tensor var_1655_to_fp16 = const()[name = tensor("op_1655_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_101_cast_fp16 = mul(x = var_1654_cast_fp16, y = var_1655_to_fp16)[name = tensor("aw_chunk_101_cast_fp16")]; tensor var_1658_equation_0 = const()[name = tensor("op_1658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1658_cast_fp16 = einsum(equation = var_1658_equation_0, values = (var_1552_cast_fp16, var_1239_cast_fp16))[name = tensor("op_1658_cast_fp16")]; tensor var_1659_to_fp16 = const()[name = tensor("op_1659_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_103_cast_fp16 = mul(x = var_1658_cast_fp16, y = var_1659_to_fp16)[name = tensor("aw_chunk_103_cast_fp16")]; tensor var_1662_equation_0 = const()[name = tensor("op_1662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1662_cast_fp16 = einsum(equation = var_1662_equation_0, values = (var_1556_cast_fp16, var_1246_cast_fp16))[name = tensor("op_1662_cast_fp16")]; tensor var_1663_to_fp16 = const()[name = tensor("op_1663_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_105_cast_fp16 = mul(x = var_1662_cast_fp16, y = var_1663_to_fp16)[name = tensor("aw_chunk_105_cast_fp16")]; tensor var_1666_equation_0 = const()[name = tensor("op_1666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1666_cast_fp16 = einsum(equation = var_1666_equation_0, values = (var_1556_cast_fp16, var_1253_cast_fp16))[name = tensor("op_1666_cast_fp16")]; tensor var_1667_to_fp16 = const()[name = tensor("op_1667_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_107_cast_fp16 = mul(x = var_1666_cast_fp16, y = var_1667_to_fp16)[name = tensor("aw_chunk_107_cast_fp16")]; tensor var_1670_equation_0 = const()[name = tensor("op_1670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1670_cast_fp16 = einsum(equation = var_1670_equation_0, values = (var_1556_cast_fp16, var_1260_cast_fp16))[name = tensor("op_1670_cast_fp16")]; tensor var_1671_to_fp16 = const()[name = tensor("op_1671_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_109_cast_fp16 = mul(x = var_1670_cast_fp16, y = var_1671_to_fp16)[name = tensor("aw_chunk_109_cast_fp16")]; tensor var_1674_equation_0 = const()[name = tensor("op_1674_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1674_cast_fp16 = einsum(equation = var_1674_equation_0, values = (var_1556_cast_fp16, var_1267_cast_fp16))[name = tensor("op_1674_cast_fp16")]; tensor var_1675_to_fp16 = const()[name = tensor("op_1675_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_111_cast_fp16 = mul(x = var_1674_cast_fp16, y = var_1675_to_fp16)[name = tensor("aw_chunk_111_cast_fp16")]; tensor var_1678_equation_0 = const()[name = tensor("op_1678_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1678_cast_fp16 = einsum(equation = var_1678_equation_0, values = (var_1560_cast_fp16, var_1274_cast_fp16))[name = tensor("op_1678_cast_fp16")]; tensor var_1679_to_fp16 = const()[name = tensor("op_1679_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_113_cast_fp16 = mul(x = var_1678_cast_fp16, y = var_1679_to_fp16)[name = tensor("aw_chunk_113_cast_fp16")]; tensor var_1682_equation_0 = const()[name = tensor("op_1682_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1682_cast_fp16 = einsum(equation = var_1682_equation_0, values = (var_1560_cast_fp16, var_1281_cast_fp16))[name = tensor("op_1682_cast_fp16")]; tensor var_1683_to_fp16 = const()[name = tensor("op_1683_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_115_cast_fp16 = mul(x = var_1682_cast_fp16, y = var_1683_to_fp16)[name = tensor("aw_chunk_115_cast_fp16")]; tensor var_1686_equation_0 = const()[name = tensor("op_1686_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1686_cast_fp16 = einsum(equation = var_1686_equation_0, values = (var_1560_cast_fp16, var_1288_cast_fp16))[name = tensor("op_1686_cast_fp16")]; tensor var_1687_to_fp16 = const()[name = tensor("op_1687_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_117_cast_fp16 = mul(x = var_1686_cast_fp16, y = var_1687_to_fp16)[name = tensor("aw_chunk_117_cast_fp16")]; tensor var_1690_equation_0 = const()[name = tensor("op_1690_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1690_cast_fp16 = einsum(equation = var_1690_equation_0, values = (var_1560_cast_fp16, var_1295_cast_fp16))[name = tensor("op_1690_cast_fp16")]; tensor var_1691_to_fp16 = const()[name = tensor("op_1691_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_119_cast_fp16 = mul(x = var_1690_cast_fp16, y = var_1691_to_fp16)[name = tensor("aw_chunk_119_cast_fp16")]; tensor var_1694_equation_0 = const()[name = tensor("op_1694_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1694_cast_fp16 = einsum(equation = var_1694_equation_0, values = (var_1564_cast_fp16, var_1302_cast_fp16))[name = tensor("op_1694_cast_fp16")]; tensor var_1695_to_fp16 = const()[name = tensor("op_1695_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_121_cast_fp16 = mul(x = var_1694_cast_fp16, y = var_1695_to_fp16)[name = tensor("aw_chunk_121_cast_fp16")]; tensor var_1698_equation_0 = const()[name = tensor("op_1698_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1698_cast_fp16 = einsum(equation = var_1698_equation_0, values = (var_1564_cast_fp16, var_1309_cast_fp16))[name = tensor("op_1698_cast_fp16")]; tensor var_1699_to_fp16 = const()[name = tensor("op_1699_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_123_cast_fp16 = mul(x = var_1698_cast_fp16, y = var_1699_to_fp16)[name = tensor("aw_chunk_123_cast_fp16")]; tensor var_1702_equation_0 = const()[name = tensor("op_1702_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1702_cast_fp16 = einsum(equation = var_1702_equation_0, values = (var_1564_cast_fp16, var_1316_cast_fp16))[name = tensor("op_1702_cast_fp16")]; tensor var_1703_to_fp16 = const()[name = tensor("op_1703_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_125_cast_fp16 = mul(x = var_1702_cast_fp16, y = var_1703_to_fp16)[name = tensor("aw_chunk_125_cast_fp16")]; tensor var_1706_equation_0 = const()[name = tensor("op_1706_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1706_cast_fp16 = einsum(equation = var_1706_equation_0, values = (var_1564_cast_fp16, var_1323_cast_fp16))[name = tensor("op_1706_cast_fp16")]; tensor var_1707_to_fp16 = const()[name = tensor("op_1707_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_127_cast_fp16 = mul(x = var_1706_cast_fp16, y = var_1707_to_fp16)[name = tensor("aw_chunk_127_cast_fp16")]; tensor var_1710_equation_0 = const()[name = tensor("op_1710_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1710_cast_fp16 = einsum(equation = var_1710_equation_0, values = (var_1568_cast_fp16, var_1330_cast_fp16))[name = tensor("op_1710_cast_fp16")]; tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_129_cast_fp16 = mul(x = var_1710_cast_fp16, y = var_1711_to_fp16)[name = tensor("aw_chunk_129_cast_fp16")]; tensor var_1714_equation_0 = const()[name = tensor("op_1714_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1714_cast_fp16 = einsum(equation = var_1714_equation_0, values = (var_1568_cast_fp16, var_1337_cast_fp16))[name = tensor("op_1714_cast_fp16")]; tensor var_1715_to_fp16 = const()[name = tensor("op_1715_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_131_cast_fp16 = mul(x = var_1714_cast_fp16, y = var_1715_to_fp16)[name = tensor("aw_chunk_131_cast_fp16")]; tensor var_1718_equation_0 = const()[name = tensor("op_1718_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1718_cast_fp16 = einsum(equation = var_1718_equation_0, values = (var_1568_cast_fp16, var_1344_cast_fp16))[name = tensor("op_1718_cast_fp16")]; tensor var_1719_to_fp16 = const()[name = tensor("op_1719_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_133_cast_fp16 = mul(x = var_1718_cast_fp16, y = var_1719_to_fp16)[name = tensor("aw_chunk_133_cast_fp16")]; tensor var_1722_equation_0 = const()[name = tensor("op_1722_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1722_cast_fp16 = einsum(equation = var_1722_equation_0, values = (var_1568_cast_fp16, var_1351_cast_fp16))[name = tensor("op_1722_cast_fp16")]; tensor var_1723_to_fp16 = const()[name = tensor("op_1723_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_135_cast_fp16 = mul(x = var_1722_cast_fp16, y = var_1723_to_fp16)[name = tensor("aw_chunk_135_cast_fp16")]; tensor var_1726_equation_0 = const()[name = tensor("op_1726_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1726_cast_fp16 = einsum(equation = var_1726_equation_0, values = (var_1572_cast_fp16, var_1358_cast_fp16))[name = tensor("op_1726_cast_fp16")]; tensor var_1727_to_fp16 = const()[name = tensor("op_1727_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_137_cast_fp16 = mul(x = var_1726_cast_fp16, y = var_1727_to_fp16)[name = tensor("aw_chunk_137_cast_fp16")]; tensor var_1730_equation_0 = const()[name = tensor("op_1730_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1730_cast_fp16 = einsum(equation = var_1730_equation_0, values = (var_1572_cast_fp16, var_1365_cast_fp16))[name = tensor("op_1730_cast_fp16")]; tensor var_1731_to_fp16 = const()[name = tensor("op_1731_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_139_cast_fp16 = mul(x = var_1730_cast_fp16, y = var_1731_to_fp16)[name = tensor("aw_chunk_139_cast_fp16")]; tensor var_1734_equation_0 = const()[name = tensor("op_1734_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1734_cast_fp16 = einsum(equation = var_1734_equation_0, values = (var_1572_cast_fp16, var_1372_cast_fp16))[name = tensor("op_1734_cast_fp16")]; tensor var_1735_to_fp16 = const()[name = tensor("op_1735_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_141_cast_fp16 = mul(x = var_1734_cast_fp16, y = var_1735_to_fp16)[name = tensor("aw_chunk_141_cast_fp16")]; tensor var_1738_equation_0 = const()[name = tensor("op_1738_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1738_cast_fp16 = einsum(equation = var_1738_equation_0, values = (var_1572_cast_fp16, var_1379_cast_fp16))[name = tensor("op_1738_cast_fp16")]; tensor var_1739_to_fp16 = const()[name = tensor("op_1739_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_143_cast_fp16 = mul(x = var_1738_cast_fp16, y = var_1739_to_fp16)[name = tensor("aw_chunk_143_cast_fp16")]; tensor var_1742_equation_0 = const()[name = tensor("op_1742_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1742_cast_fp16 = einsum(equation = var_1742_equation_0, values = (var_1576_cast_fp16, var_1386_cast_fp16))[name = tensor("op_1742_cast_fp16")]; tensor var_1743_to_fp16 = const()[name = tensor("op_1743_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_145_cast_fp16 = mul(x = var_1742_cast_fp16, y = var_1743_to_fp16)[name = tensor("aw_chunk_145_cast_fp16")]; tensor var_1746_equation_0 = const()[name = tensor("op_1746_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1746_cast_fp16 = einsum(equation = var_1746_equation_0, values = (var_1576_cast_fp16, var_1393_cast_fp16))[name = tensor("op_1746_cast_fp16")]; tensor var_1747_to_fp16 = const()[name = tensor("op_1747_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_147_cast_fp16 = mul(x = var_1746_cast_fp16, y = var_1747_to_fp16)[name = tensor("aw_chunk_147_cast_fp16")]; tensor var_1750_equation_0 = const()[name = tensor("op_1750_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1750_cast_fp16 = einsum(equation = var_1750_equation_0, values = (var_1576_cast_fp16, var_1400_cast_fp16))[name = tensor("op_1750_cast_fp16")]; tensor var_1751_to_fp16 = const()[name = tensor("op_1751_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_149_cast_fp16 = mul(x = var_1750_cast_fp16, y = var_1751_to_fp16)[name = tensor("aw_chunk_149_cast_fp16")]; tensor var_1754_equation_0 = const()[name = tensor("op_1754_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1754_cast_fp16 = einsum(equation = var_1754_equation_0, values = (var_1576_cast_fp16, var_1407_cast_fp16))[name = tensor("op_1754_cast_fp16")]; tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_151_cast_fp16 = mul(x = var_1754_cast_fp16, y = var_1755_to_fp16)[name = tensor("aw_chunk_151_cast_fp16")]; tensor var_1758_equation_0 = const()[name = tensor("op_1758_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1758_cast_fp16 = einsum(equation = var_1758_equation_0, values = (var_1580_cast_fp16, var_1414_cast_fp16))[name = tensor("op_1758_cast_fp16")]; tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_153_cast_fp16 = mul(x = var_1758_cast_fp16, y = var_1759_to_fp16)[name = tensor("aw_chunk_153_cast_fp16")]; tensor var_1762_equation_0 = const()[name = tensor("op_1762_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1762_cast_fp16 = einsum(equation = var_1762_equation_0, values = (var_1580_cast_fp16, var_1421_cast_fp16))[name = tensor("op_1762_cast_fp16")]; tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_155_cast_fp16 = mul(x = var_1762_cast_fp16, y = var_1763_to_fp16)[name = tensor("aw_chunk_155_cast_fp16")]; tensor var_1766_equation_0 = const()[name = tensor("op_1766_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1766_cast_fp16 = einsum(equation = var_1766_equation_0, values = (var_1580_cast_fp16, var_1428_cast_fp16))[name = tensor("op_1766_cast_fp16")]; tensor var_1767_to_fp16 = const()[name = tensor("op_1767_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_157_cast_fp16 = mul(x = var_1766_cast_fp16, y = var_1767_to_fp16)[name = tensor("aw_chunk_157_cast_fp16")]; tensor var_1770_equation_0 = const()[name = tensor("op_1770_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1770_cast_fp16 = einsum(equation = var_1770_equation_0, values = (var_1580_cast_fp16, var_1435_cast_fp16))[name = tensor("op_1770_cast_fp16")]; tensor var_1771_to_fp16 = const()[name = tensor("op_1771_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_159_cast_fp16 = mul(x = var_1770_cast_fp16, y = var_1771_to_fp16)[name = tensor("aw_chunk_159_cast_fp16")]; tensor var_1774_equation_0 = const()[name = tensor("op_1774_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1774_cast_fp16 = einsum(equation = var_1774_equation_0, values = (var_1584_cast_fp16, var_1442_cast_fp16))[name = tensor("op_1774_cast_fp16")]; tensor var_1775_to_fp16 = const()[name = tensor("op_1775_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_161_cast_fp16 = mul(x = var_1774_cast_fp16, y = var_1775_to_fp16)[name = tensor("aw_chunk_161_cast_fp16")]; tensor var_1778_equation_0 = const()[name = tensor("op_1778_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1778_cast_fp16 = einsum(equation = var_1778_equation_0, values = (var_1584_cast_fp16, var_1449_cast_fp16))[name = tensor("op_1778_cast_fp16")]; tensor var_1779_to_fp16 = const()[name = tensor("op_1779_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_163_cast_fp16 = mul(x = var_1778_cast_fp16, y = var_1779_to_fp16)[name = tensor("aw_chunk_163_cast_fp16")]; tensor var_1782_equation_0 = const()[name = tensor("op_1782_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1782_cast_fp16 = einsum(equation = var_1782_equation_0, values = (var_1584_cast_fp16, var_1456_cast_fp16))[name = tensor("op_1782_cast_fp16")]; tensor var_1783_to_fp16 = const()[name = tensor("op_1783_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_165_cast_fp16 = mul(x = var_1782_cast_fp16, y = var_1783_to_fp16)[name = tensor("aw_chunk_165_cast_fp16")]; tensor var_1786_equation_0 = const()[name = tensor("op_1786_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1786_cast_fp16 = einsum(equation = var_1786_equation_0, values = (var_1584_cast_fp16, var_1463_cast_fp16))[name = tensor("op_1786_cast_fp16")]; tensor var_1787_to_fp16 = const()[name = tensor("op_1787_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_167_cast_fp16 = mul(x = var_1786_cast_fp16, y = var_1787_to_fp16)[name = tensor("aw_chunk_167_cast_fp16")]; tensor var_1790_equation_0 = const()[name = tensor("op_1790_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1790_cast_fp16 = einsum(equation = var_1790_equation_0, values = (var_1588_cast_fp16, var_1470_cast_fp16))[name = tensor("op_1790_cast_fp16")]; tensor var_1791_to_fp16 = const()[name = tensor("op_1791_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_169_cast_fp16 = mul(x = var_1790_cast_fp16, y = var_1791_to_fp16)[name = tensor("aw_chunk_169_cast_fp16")]; tensor var_1794_equation_0 = const()[name = tensor("op_1794_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1794_cast_fp16 = einsum(equation = var_1794_equation_0, values = (var_1588_cast_fp16, var_1477_cast_fp16))[name = tensor("op_1794_cast_fp16")]; tensor var_1795_to_fp16 = const()[name = tensor("op_1795_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_171_cast_fp16 = mul(x = var_1794_cast_fp16, y = var_1795_to_fp16)[name = tensor("aw_chunk_171_cast_fp16")]; tensor var_1798_equation_0 = const()[name = tensor("op_1798_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1798_cast_fp16 = einsum(equation = var_1798_equation_0, values = (var_1588_cast_fp16, var_1484_cast_fp16))[name = tensor("op_1798_cast_fp16")]; tensor var_1799_to_fp16 = const()[name = tensor("op_1799_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_173_cast_fp16 = mul(x = var_1798_cast_fp16, y = var_1799_to_fp16)[name = tensor("aw_chunk_173_cast_fp16")]; tensor var_1802_equation_0 = const()[name = tensor("op_1802_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1802_cast_fp16 = einsum(equation = var_1802_equation_0, values = (var_1588_cast_fp16, var_1491_cast_fp16))[name = tensor("op_1802_cast_fp16")]; tensor var_1803_to_fp16 = const()[name = tensor("op_1803_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_175_cast_fp16 = mul(x = var_1802_cast_fp16, y = var_1803_to_fp16)[name = tensor("aw_chunk_175_cast_fp16")]; tensor var_1806_equation_0 = const()[name = tensor("op_1806_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1806_cast_fp16 = einsum(equation = var_1806_equation_0, values = (var_1592_cast_fp16, var_1498_cast_fp16))[name = tensor("op_1806_cast_fp16")]; tensor var_1807_to_fp16 = const()[name = tensor("op_1807_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_177_cast_fp16 = mul(x = var_1806_cast_fp16, y = var_1807_to_fp16)[name = tensor("aw_chunk_177_cast_fp16")]; tensor var_1810_equation_0 = const()[name = tensor("op_1810_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1810_cast_fp16 = einsum(equation = var_1810_equation_0, values = (var_1592_cast_fp16, var_1505_cast_fp16))[name = tensor("op_1810_cast_fp16")]; tensor var_1811_to_fp16 = const()[name = tensor("op_1811_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_179_cast_fp16 = mul(x = var_1810_cast_fp16, y = var_1811_to_fp16)[name = tensor("aw_chunk_179_cast_fp16")]; tensor var_1814_equation_0 = const()[name = tensor("op_1814_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1814_cast_fp16 = einsum(equation = var_1814_equation_0, values = (var_1592_cast_fp16, var_1512_cast_fp16))[name = tensor("op_1814_cast_fp16")]; tensor var_1815_to_fp16 = const()[name = tensor("op_1815_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_181_cast_fp16 = mul(x = var_1814_cast_fp16, y = var_1815_to_fp16)[name = tensor("aw_chunk_181_cast_fp16")]; tensor var_1818_equation_0 = const()[name = tensor("op_1818_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1818_cast_fp16 = einsum(equation = var_1818_equation_0, values = (var_1592_cast_fp16, var_1519_cast_fp16))[name = tensor("op_1818_cast_fp16")]; tensor var_1819_to_fp16 = const()[name = tensor("op_1819_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_183_cast_fp16 = mul(x = var_1818_cast_fp16, y = var_1819_to_fp16)[name = tensor("aw_chunk_183_cast_fp16")]; tensor var_1822_equation_0 = const()[name = tensor("op_1822_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1822_cast_fp16 = einsum(equation = var_1822_equation_0, values = (var_1596_cast_fp16, var_1526_cast_fp16))[name = tensor("op_1822_cast_fp16")]; tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_185_cast_fp16 = mul(x = var_1822_cast_fp16, y = var_1823_to_fp16)[name = tensor("aw_chunk_185_cast_fp16")]; tensor var_1826_equation_0 = const()[name = tensor("op_1826_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1826_cast_fp16 = einsum(equation = var_1826_equation_0, values = (var_1596_cast_fp16, var_1533_cast_fp16))[name = tensor("op_1826_cast_fp16")]; tensor var_1827_to_fp16 = const()[name = tensor("op_1827_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_187_cast_fp16 = mul(x = var_1826_cast_fp16, y = var_1827_to_fp16)[name = tensor("aw_chunk_187_cast_fp16")]; tensor var_1830_equation_0 = const()[name = tensor("op_1830_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1830_cast_fp16 = einsum(equation = var_1830_equation_0, values = (var_1596_cast_fp16, var_1540_cast_fp16))[name = tensor("op_1830_cast_fp16")]; tensor var_1831_to_fp16 = const()[name = tensor("op_1831_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_189_cast_fp16 = mul(x = var_1830_cast_fp16, y = var_1831_to_fp16)[name = tensor("aw_chunk_189_cast_fp16")]; tensor var_1834_equation_0 = const()[name = tensor("op_1834_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_1834_cast_fp16 = einsum(equation = var_1834_equation_0, values = (var_1596_cast_fp16, var_1547_cast_fp16))[name = tensor("op_1834_cast_fp16")]; tensor var_1835_to_fp16 = const()[name = tensor("op_1835_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_191_cast_fp16 = mul(x = var_1834_cast_fp16, y = var_1835_to_fp16)[name = tensor("aw_chunk_191_cast_fp16")]; tensor var_1837_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_97_cast_fp16)[name = tensor("op_1837_cast_fp16")]; tensor var_1838_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_99_cast_fp16)[name = tensor("op_1838_cast_fp16")]; tensor var_1839_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_101_cast_fp16)[name = tensor("op_1839_cast_fp16")]; tensor var_1840_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_103_cast_fp16)[name = tensor("op_1840_cast_fp16")]; tensor var_1841_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_105_cast_fp16)[name = tensor("op_1841_cast_fp16")]; tensor var_1842_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_107_cast_fp16)[name = tensor("op_1842_cast_fp16")]; tensor var_1843_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_109_cast_fp16)[name = tensor("op_1843_cast_fp16")]; tensor var_1844_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_111_cast_fp16)[name = tensor("op_1844_cast_fp16")]; tensor var_1845_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_113_cast_fp16)[name = tensor("op_1845_cast_fp16")]; tensor var_1846_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_115_cast_fp16)[name = tensor("op_1846_cast_fp16")]; tensor var_1847_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_117_cast_fp16)[name = tensor("op_1847_cast_fp16")]; tensor var_1848_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_119_cast_fp16)[name = tensor("op_1848_cast_fp16")]; tensor var_1849_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_121_cast_fp16)[name = tensor("op_1849_cast_fp16")]; tensor var_1850_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_123_cast_fp16)[name = tensor("op_1850_cast_fp16")]; tensor var_1851_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_125_cast_fp16)[name = tensor("op_1851_cast_fp16")]; tensor var_1852_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_127_cast_fp16)[name = tensor("op_1852_cast_fp16")]; tensor var_1853_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_129_cast_fp16)[name = tensor("op_1853_cast_fp16")]; tensor var_1854_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_131_cast_fp16)[name = tensor("op_1854_cast_fp16")]; tensor var_1855_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_133_cast_fp16)[name = tensor("op_1855_cast_fp16")]; tensor var_1856_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_135_cast_fp16)[name = tensor("op_1856_cast_fp16")]; tensor var_1857_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_137_cast_fp16)[name = tensor("op_1857_cast_fp16")]; tensor var_1858_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_139_cast_fp16)[name = tensor("op_1858_cast_fp16")]; tensor var_1859_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_141_cast_fp16)[name = tensor("op_1859_cast_fp16")]; tensor var_1860_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_143_cast_fp16)[name = tensor("op_1860_cast_fp16")]; tensor var_1861_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_145_cast_fp16)[name = tensor("op_1861_cast_fp16")]; tensor var_1862_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_147_cast_fp16)[name = tensor("op_1862_cast_fp16")]; tensor var_1863_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_149_cast_fp16)[name = tensor("op_1863_cast_fp16")]; tensor var_1864_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_151_cast_fp16)[name = tensor("op_1864_cast_fp16")]; tensor var_1865_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_153_cast_fp16)[name = tensor("op_1865_cast_fp16")]; tensor var_1866_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_155_cast_fp16)[name = tensor("op_1866_cast_fp16")]; tensor var_1867_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_157_cast_fp16)[name = tensor("op_1867_cast_fp16")]; tensor var_1868_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_159_cast_fp16)[name = tensor("op_1868_cast_fp16")]; tensor var_1869_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_161_cast_fp16)[name = tensor("op_1869_cast_fp16")]; tensor var_1870_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_163_cast_fp16)[name = tensor("op_1870_cast_fp16")]; tensor var_1871_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_165_cast_fp16)[name = tensor("op_1871_cast_fp16")]; tensor var_1872_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_167_cast_fp16)[name = tensor("op_1872_cast_fp16")]; tensor var_1873_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_169_cast_fp16)[name = tensor("op_1873_cast_fp16")]; tensor var_1874_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_171_cast_fp16)[name = tensor("op_1874_cast_fp16")]; tensor var_1875_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_173_cast_fp16)[name = tensor("op_1875_cast_fp16")]; tensor var_1876_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_175_cast_fp16)[name = tensor("op_1876_cast_fp16")]; tensor var_1877_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_177_cast_fp16)[name = tensor("op_1877_cast_fp16")]; tensor var_1878_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_179_cast_fp16)[name = tensor("op_1878_cast_fp16")]; tensor var_1879_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_181_cast_fp16)[name = tensor("op_1879_cast_fp16")]; tensor var_1880_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_183_cast_fp16)[name = tensor("op_1880_cast_fp16")]; tensor var_1881_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_185_cast_fp16)[name = tensor("op_1881_cast_fp16")]; tensor var_1882_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_187_cast_fp16)[name = tensor("op_1882_cast_fp16")]; tensor var_1883_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_189_cast_fp16)[name = tensor("op_1883_cast_fp16")]; tensor var_1884_cast_fp16 = softmax(axis = var_1110, x = aw_chunk_191_cast_fp16)[name = tensor("op_1884_cast_fp16")]; tensor var_1886_equation_0 = const()[name = tensor("op_1886_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1886_cast_fp16 = einsum(equation = var_1886_equation_0, values = (var_1598_cast_fp16, var_1837_cast_fp16))[name = tensor("op_1886_cast_fp16")]; tensor var_1888_equation_0 = const()[name = tensor("op_1888_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1888_cast_fp16 = einsum(equation = var_1888_equation_0, values = (var_1598_cast_fp16, var_1838_cast_fp16))[name = tensor("op_1888_cast_fp16")]; tensor var_1890_equation_0 = const()[name = tensor("op_1890_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1890_cast_fp16 = einsum(equation = var_1890_equation_0, values = (var_1598_cast_fp16, var_1839_cast_fp16))[name = tensor("op_1890_cast_fp16")]; tensor var_1892_equation_0 = const()[name = tensor("op_1892_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1892_cast_fp16 = einsum(equation = var_1892_equation_0, values = (var_1598_cast_fp16, var_1840_cast_fp16))[name = tensor("op_1892_cast_fp16")]; tensor var_1894_equation_0 = const()[name = tensor("op_1894_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1894_cast_fp16 = einsum(equation = var_1894_equation_0, values = (var_1602_cast_fp16, var_1841_cast_fp16))[name = tensor("op_1894_cast_fp16")]; tensor var_1896_equation_0 = const()[name = tensor("op_1896_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1896_cast_fp16 = einsum(equation = var_1896_equation_0, values = (var_1602_cast_fp16, var_1842_cast_fp16))[name = tensor("op_1896_cast_fp16")]; tensor var_1898_equation_0 = const()[name = tensor("op_1898_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1898_cast_fp16 = einsum(equation = var_1898_equation_0, values = (var_1602_cast_fp16, var_1843_cast_fp16))[name = tensor("op_1898_cast_fp16")]; tensor var_1900_equation_0 = const()[name = tensor("op_1900_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1900_cast_fp16 = einsum(equation = var_1900_equation_0, values = (var_1602_cast_fp16, var_1844_cast_fp16))[name = tensor("op_1900_cast_fp16")]; tensor var_1902_equation_0 = const()[name = tensor("op_1902_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1902_cast_fp16 = einsum(equation = var_1902_equation_0, values = (var_1606_cast_fp16, var_1845_cast_fp16))[name = tensor("op_1902_cast_fp16")]; tensor var_1904_equation_0 = const()[name = tensor("op_1904_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1904_cast_fp16 = einsum(equation = var_1904_equation_0, values = (var_1606_cast_fp16, var_1846_cast_fp16))[name = tensor("op_1904_cast_fp16")]; tensor var_1906_equation_0 = const()[name = tensor("op_1906_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1906_cast_fp16 = einsum(equation = var_1906_equation_0, values = (var_1606_cast_fp16, var_1847_cast_fp16))[name = tensor("op_1906_cast_fp16")]; tensor var_1908_equation_0 = const()[name = tensor("op_1908_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1908_cast_fp16 = einsum(equation = var_1908_equation_0, values = (var_1606_cast_fp16, var_1848_cast_fp16))[name = tensor("op_1908_cast_fp16")]; tensor var_1910_equation_0 = const()[name = tensor("op_1910_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1910_cast_fp16 = einsum(equation = var_1910_equation_0, values = (var_1610_cast_fp16, var_1849_cast_fp16))[name = tensor("op_1910_cast_fp16")]; tensor var_1912_equation_0 = const()[name = tensor("op_1912_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1912_cast_fp16 = einsum(equation = var_1912_equation_0, values = (var_1610_cast_fp16, var_1850_cast_fp16))[name = tensor("op_1912_cast_fp16")]; tensor var_1914_equation_0 = const()[name = tensor("op_1914_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1914_cast_fp16 = einsum(equation = var_1914_equation_0, values = (var_1610_cast_fp16, var_1851_cast_fp16))[name = tensor("op_1914_cast_fp16")]; tensor var_1916_equation_0 = const()[name = tensor("op_1916_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1916_cast_fp16 = einsum(equation = var_1916_equation_0, values = (var_1610_cast_fp16, var_1852_cast_fp16))[name = tensor("op_1916_cast_fp16")]; tensor var_1918_equation_0 = const()[name = tensor("op_1918_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1918_cast_fp16 = einsum(equation = var_1918_equation_0, values = (var_1614_cast_fp16, var_1853_cast_fp16))[name = tensor("op_1918_cast_fp16")]; tensor var_1920_equation_0 = const()[name = tensor("op_1920_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1920_cast_fp16 = einsum(equation = var_1920_equation_0, values = (var_1614_cast_fp16, var_1854_cast_fp16))[name = tensor("op_1920_cast_fp16")]; tensor var_1922_equation_0 = const()[name = tensor("op_1922_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1922_cast_fp16 = einsum(equation = var_1922_equation_0, values = (var_1614_cast_fp16, var_1855_cast_fp16))[name = tensor("op_1922_cast_fp16")]; tensor var_1924_equation_0 = const()[name = tensor("op_1924_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1924_cast_fp16 = einsum(equation = var_1924_equation_0, values = (var_1614_cast_fp16, var_1856_cast_fp16))[name = tensor("op_1924_cast_fp16")]; tensor var_1926_equation_0 = const()[name = tensor("op_1926_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1926_cast_fp16 = einsum(equation = var_1926_equation_0, values = (var_1618_cast_fp16, var_1857_cast_fp16))[name = tensor("op_1926_cast_fp16")]; tensor var_1928_equation_0 = const()[name = tensor("op_1928_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1928_cast_fp16 = einsum(equation = var_1928_equation_0, values = (var_1618_cast_fp16, var_1858_cast_fp16))[name = tensor("op_1928_cast_fp16")]; tensor var_1930_equation_0 = const()[name = tensor("op_1930_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1930_cast_fp16 = einsum(equation = var_1930_equation_0, values = (var_1618_cast_fp16, var_1859_cast_fp16))[name = tensor("op_1930_cast_fp16")]; tensor var_1932_equation_0 = const()[name = tensor("op_1932_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1932_cast_fp16 = einsum(equation = var_1932_equation_0, values = (var_1618_cast_fp16, var_1860_cast_fp16))[name = tensor("op_1932_cast_fp16")]; tensor var_1934_equation_0 = const()[name = tensor("op_1934_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1934_cast_fp16 = einsum(equation = var_1934_equation_0, values = (var_1622_cast_fp16, var_1861_cast_fp16))[name = tensor("op_1934_cast_fp16")]; tensor var_1936_equation_0 = const()[name = tensor("op_1936_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1936_cast_fp16 = einsum(equation = var_1936_equation_0, values = (var_1622_cast_fp16, var_1862_cast_fp16))[name = tensor("op_1936_cast_fp16")]; tensor var_1938_equation_0 = const()[name = tensor("op_1938_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1938_cast_fp16 = einsum(equation = var_1938_equation_0, values = (var_1622_cast_fp16, var_1863_cast_fp16))[name = tensor("op_1938_cast_fp16")]; tensor var_1940_equation_0 = const()[name = tensor("op_1940_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1940_cast_fp16 = einsum(equation = var_1940_equation_0, values = (var_1622_cast_fp16, var_1864_cast_fp16))[name = tensor("op_1940_cast_fp16")]; tensor var_1942_equation_0 = const()[name = tensor("op_1942_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1942_cast_fp16 = einsum(equation = var_1942_equation_0, values = (var_1626_cast_fp16, var_1865_cast_fp16))[name = tensor("op_1942_cast_fp16")]; tensor var_1944_equation_0 = const()[name = tensor("op_1944_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1944_cast_fp16 = einsum(equation = var_1944_equation_0, values = (var_1626_cast_fp16, var_1866_cast_fp16))[name = tensor("op_1944_cast_fp16")]; tensor var_1946_equation_0 = const()[name = tensor("op_1946_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1946_cast_fp16 = einsum(equation = var_1946_equation_0, values = (var_1626_cast_fp16, var_1867_cast_fp16))[name = tensor("op_1946_cast_fp16")]; tensor var_1948_equation_0 = const()[name = tensor("op_1948_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1948_cast_fp16 = einsum(equation = var_1948_equation_0, values = (var_1626_cast_fp16, var_1868_cast_fp16))[name = tensor("op_1948_cast_fp16")]; tensor var_1950_equation_0 = const()[name = tensor("op_1950_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1950_cast_fp16 = einsum(equation = var_1950_equation_0, values = (var_1630_cast_fp16, var_1869_cast_fp16))[name = tensor("op_1950_cast_fp16")]; tensor var_1952_equation_0 = const()[name = tensor("op_1952_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1952_cast_fp16 = einsum(equation = var_1952_equation_0, values = (var_1630_cast_fp16, var_1870_cast_fp16))[name = tensor("op_1952_cast_fp16")]; tensor var_1954_equation_0 = const()[name = tensor("op_1954_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1954_cast_fp16 = einsum(equation = var_1954_equation_0, values = (var_1630_cast_fp16, var_1871_cast_fp16))[name = tensor("op_1954_cast_fp16")]; tensor var_1956_equation_0 = const()[name = tensor("op_1956_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1956_cast_fp16 = einsum(equation = var_1956_equation_0, values = (var_1630_cast_fp16, var_1872_cast_fp16))[name = tensor("op_1956_cast_fp16")]; tensor var_1958_equation_0 = const()[name = tensor("op_1958_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1958_cast_fp16 = einsum(equation = var_1958_equation_0, values = (var_1634_cast_fp16, var_1873_cast_fp16))[name = tensor("op_1958_cast_fp16")]; tensor var_1960_equation_0 = const()[name = tensor("op_1960_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1960_cast_fp16 = einsum(equation = var_1960_equation_0, values = (var_1634_cast_fp16, var_1874_cast_fp16))[name = tensor("op_1960_cast_fp16")]; tensor var_1962_equation_0 = const()[name = tensor("op_1962_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1962_cast_fp16 = einsum(equation = var_1962_equation_0, values = (var_1634_cast_fp16, var_1875_cast_fp16))[name = tensor("op_1962_cast_fp16")]; tensor var_1964_equation_0 = const()[name = tensor("op_1964_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1964_cast_fp16 = einsum(equation = var_1964_equation_0, values = (var_1634_cast_fp16, var_1876_cast_fp16))[name = tensor("op_1964_cast_fp16")]; tensor var_1966_equation_0 = const()[name = tensor("op_1966_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1966_cast_fp16 = einsum(equation = var_1966_equation_0, values = (var_1638_cast_fp16, var_1877_cast_fp16))[name = tensor("op_1966_cast_fp16")]; tensor var_1968_equation_0 = const()[name = tensor("op_1968_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1968_cast_fp16 = einsum(equation = var_1968_equation_0, values = (var_1638_cast_fp16, var_1878_cast_fp16))[name = tensor("op_1968_cast_fp16")]; tensor var_1970_equation_0 = const()[name = tensor("op_1970_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1970_cast_fp16 = einsum(equation = var_1970_equation_0, values = (var_1638_cast_fp16, var_1879_cast_fp16))[name = tensor("op_1970_cast_fp16")]; tensor var_1972_equation_0 = const()[name = tensor("op_1972_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1972_cast_fp16 = einsum(equation = var_1972_equation_0, values = (var_1638_cast_fp16, var_1880_cast_fp16))[name = tensor("op_1972_cast_fp16")]; tensor var_1974_equation_0 = const()[name = tensor("op_1974_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1974_cast_fp16 = einsum(equation = var_1974_equation_0, values = (var_1642_cast_fp16, var_1881_cast_fp16))[name = tensor("op_1974_cast_fp16")]; tensor var_1976_equation_0 = const()[name = tensor("op_1976_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1976_cast_fp16 = einsum(equation = var_1976_equation_0, values = (var_1642_cast_fp16, var_1882_cast_fp16))[name = tensor("op_1976_cast_fp16")]; tensor var_1978_equation_0 = const()[name = tensor("op_1978_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1978_cast_fp16 = einsum(equation = var_1978_equation_0, values = (var_1642_cast_fp16, var_1883_cast_fp16))[name = tensor("op_1978_cast_fp16")]; tensor var_1980_equation_0 = const()[name = tensor("op_1980_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_1980_cast_fp16 = einsum(equation = var_1980_equation_0, values = (var_1642_cast_fp16, var_1884_cast_fp16))[name = tensor("op_1980_cast_fp16")]; tensor var_1982_interleave_0 = const()[name = tensor("op_1982_interleave_0"), val = tensor(false)]; tensor var_1982_cast_fp16 = concat(axis = var_1093, interleave = var_1982_interleave_0, values = (var_1886_cast_fp16, var_1888_cast_fp16, var_1890_cast_fp16, var_1892_cast_fp16))[name = tensor("op_1982_cast_fp16")]; tensor var_1984_interleave_0 = const()[name = tensor("op_1984_interleave_0"), val = tensor(false)]; tensor var_1984_cast_fp16 = concat(axis = var_1093, interleave = var_1984_interleave_0, values = (var_1894_cast_fp16, var_1896_cast_fp16, var_1898_cast_fp16, var_1900_cast_fp16))[name = tensor("op_1984_cast_fp16")]; tensor var_1986_interleave_0 = const()[name = tensor("op_1986_interleave_0"), val = tensor(false)]; tensor var_1986_cast_fp16 = concat(axis = var_1093, interleave = var_1986_interleave_0, values = (var_1902_cast_fp16, var_1904_cast_fp16, var_1906_cast_fp16, var_1908_cast_fp16))[name = tensor("op_1986_cast_fp16")]; tensor var_1988_interleave_0 = const()[name = tensor("op_1988_interleave_0"), val = tensor(false)]; tensor var_1988_cast_fp16 = concat(axis = var_1093, interleave = var_1988_interleave_0, values = (var_1910_cast_fp16, var_1912_cast_fp16, var_1914_cast_fp16, var_1916_cast_fp16))[name = tensor("op_1988_cast_fp16")]; tensor var_1990_interleave_0 = const()[name = tensor("op_1990_interleave_0"), val = tensor(false)]; tensor var_1990_cast_fp16 = concat(axis = var_1093, interleave = var_1990_interleave_0, values = (var_1918_cast_fp16, var_1920_cast_fp16, var_1922_cast_fp16, var_1924_cast_fp16))[name = tensor("op_1990_cast_fp16")]; tensor var_1992_interleave_0 = const()[name = tensor("op_1992_interleave_0"), val = tensor(false)]; tensor var_1992_cast_fp16 = concat(axis = var_1093, interleave = var_1992_interleave_0, values = (var_1926_cast_fp16, var_1928_cast_fp16, var_1930_cast_fp16, var_1932_cast_fp16))[name = tensor("op_1992_cast_fp16")]; tensor var_1994_interleave_0 = const()[name = tensor("op_1994_interleave_0"), val = tensor(false)]; tensor var_1994_cast_fp16 = concat(axis = var_1093, interleave = var_1994_interleave_0, values = (var_1934_cast_fp16, var_1936_cast_fp16, var_1938_cast_fp16, var_1940_cast_fp16))[name = tensor("op_1994_cast_fp16")]; tensor var_1996_interleave_0 = const()[name = tensor("op_1996_interleave_0"), val = tensor(false)]; tensor var_1996_cast_fp16 = concat(axis = var_1093, interleave = var_1996_interleave_0, values = (var_1942_cast_fp16, var_1944_cast_fp16, var_1946_cast_fp16, var_1948_cast_fp16))[name = tensor("op_1996_cast_fp16")]; tensor var_1998_interleave_0 = const()[name = tensor("op_1998_interleave_0"), val = tensor(false)]; tensor var_1998_cast_fp16 = concat(axis = var_1093, interleave = var_1998_interleave_0, values = (var_1950_cast_fp16, var_1952_cast_fp16, var_1954_cast_fp16, var_1956_cast_fp16))[name = tensor("op_1998_cast_fp16")]; tensor var_2000_interleave_0 = const()[name = tensor("op_2000_interleave_0"), val = tensor(false)]; tensor var_2000_cast_fp16 = concat(axis = var_1093, interleave = var_2000_interleave_0, values = (var_1958_cast_fp16, var_1960_cast_fp16, var_1962_cast_fp16, var_1964_cast_fp16))[name = tensor("op_2000_cast_fp16")]; tensor var_2002_interleave_0 = const()[name = tensor("op_2002_interleave_0"), val = tensor(false)]; tensor var_2002_cast_fp16 = concat(axis = var_1093, interleave = var_2002_interleave_0, values = (var_1966_cast_fp16, var_1968_cast_fp16, var_1970_cast_fp16, var_1972_cast_fp16))[name = tensor("op_2002_cast_fp16")]; tensor var_2004_interleave_0 = const()[name = tensor("op_2004_interleave_0"), val = tensor(false)]; tensor var_2004_cast_fp16 = concat(axis = var_1093, interleave = var_2004_interleave_0, values = (var_1974_cast_fp16, var_1976_cast_fp16, var_1978_cast_fp16, var_1980_cast_fp16))[name = tensor("op_2004_cast_fp16")]; tensor input_9_interleave_0 = const()[name = tensor("input_9_interleave_0"), val = tensor(false)]; tensor input_9_cast_fp16 = concat(axis = var_1110, interleave = input_9_interleave_0, values = (var_1982_cast_fp16, var_1984_cast_fp16, var_1986_cast_fp16, var_1988_cast_fp16, var_1990_cast_fp16, var_1992_cast_fp16, var_1994_cast_fp16, var_1996_cast_fp16, var_1998_cast_fp16, var_2000_cast_fp16, var_2002_cast_fp16, var_2004_cast_fp16))[name = tensor("input_9_cast_fp16")]; tensor var_2009 = const()[name = tensor("op_2009"), val = tensor([1, 1])]; tensor var_2011 = const()[name = tensor("op_2011"), val = tensor([1, 1])]; tensor obj_7_pad_type_0 = const()[name = tensor("obj_7_pad_type_0"), val = tensor("custom")]; tensor obj_7_pad_0 = const()[name = tensor("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23938944)))]; tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25118656)))]; tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = var_2011, groups = var_1110, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = var_2009, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_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_2017 = const()[name = tensor("op_2017"), val = tensor([1])]; tensor channels_mean_7_cast_fp16 = reduce_mean(axes = var_2017, keep_dims = var_1111, 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_2021 = const()[name = tensor("op_2021"), val = tensor([1])]; tensor var_2022_cast_fp16 = reduce_mean(axes = var_2021, keep_dims = var_1111, x = zero_mean_sq_7_cast_fp16)[name = tensor("op_2022_cast_fp16")]; tensor var_2023_to_fp16 = const()[name = tensor("op_2023_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2024_cast_fp16 = add(x = var_2022_cast_fp16, y = var_2023_to_fp16)[name = tensor("op_2024_cast_fp16")]; tensor denom_7_epsilon_0_to_fp16 = const()[name = tensor("denom_7_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_7_cast_fp16 = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_2024_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_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(25120256)))]; 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(25121856)))]; 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_7_cast_fp16)[name = tensor("input_11_cast_fp16")]; tensor var_2035 = const()[name = tensor("op_2035"), val = tensor([1, 1])]; tensor var_2037 = const()[name = tensor("op_2037"), 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_1_fc1_weight_to_fp16 = const()[name = tensor("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25123456)))]; tensor layers_1_fc1_bias_to_fp16 = const()[name = tensor("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29842112)))]; tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = var_2037, groups = var_1110, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_2035, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; tensor input_15_mode_0 = const()[name = tensor("input_15_mode_0"), val = tensor("EXACT")]; tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = tensor("input_15_cast_fp16")]; tensor var_2043 = const()[name = tensor("op_2043"), val = tensor([1, 1])]; tensor var_2045 = const()[name = tensor("op_2045"), val = tensor([1, 1])]; tensor hidden_states_7_pad_type_0 = const()[name = tensor("hidden_states_7_pad_type_0"), val = tensor("custom")]; tensor hidden_states_7_pad_0 = const()[name = tensor("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_1_fc2_weight_to_fp16 = const()[name = tensor("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29848320)))]; tensor layers_1_fc2_bias_to_fp16 = const()[name = tensor("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34566976)))]; tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = var_2045, groups = var_1110, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_2043, weight = layers_1_fc2_weight_to_fp16, x = input_15_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_2052 = const()[name = tensor("op_2052"), val = tensor(3)]; tensor var_2069 = const()[name = tensor("op_2069"), val = tensor(1)]; tensor var_2070 = const()[name = tensor("op_2070"), val = tensor(true)]; tensor var_2080 = const()[name = tensor("op_2080"), val = tensor([1])]; tensor channels_mean_9_cast_fp16 = reduce_mean(axes = var_2080, keep_dims = var_2070, 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_2084 = const()[name = tensor("op_2084"), val = tensor([1])]; tensor var_2085_cast_fp16 = reduce_mean(axes = var_2084, keep_dims = var_2070, x = zero_mean_sq_9_cast_fp16)[name = tensor("op_2085_cast_fp16")]; tensor var_2086_to_fp16 = const()[name = tensor("op_2086_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2087_cast_fp16 = add(x = var_2085_cast_fp16, y = var_2086_to_fp16)[name = tensor("op_2087_cast_fp16")]; tensor denom_9_epsilon_0_to_fp16 = const()[name = tensor("denom_9_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_9_cast_fp16 = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_2087_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(34568576)))]; 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(34570176)))]; 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_2102 = const()[name = tensor("op_2102"), val = tensor([1, 1])]; tensor var_2104 = const()[name = tensor("op_2104"), val = tensor([1, 1])]; tensor query_5_pad_type_0 = const()[name = tensor("query_5_pad_type_0"), val = tensor("custom")]; tensor query_5_pad_0 = const()[name = tensor("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34571776)))]; tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35751488)))]; tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = var_2104, groups = var_2069, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = var_2102, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("query_5_cast_fp16")]; tensor var_2108 = const()[name = tensor("op_2108"), val = tensor([1, 1])]; tensor var_2110 = const()[name = tensor("op_2110"), val = tensor([1, 1])]; tensor key_5_pad_type_0 = const()[name = tensor("key_5_pad_type_0"), val = tensor("custom")]; tensor key_5_pad_0 = const()[name = tensor("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35753088)))]; tensor key_5_cast_fp16 = conv(dilations = var_2110, groups = var_2069, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = var_2108, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("key_5_cast_fp16")]; tensor var_2115 = const()[name = tensor("op_2115"), val = tensor([1, 1])]; tensor var_2117 = const()[name = tensor("op_2117"), val = tensor([1, 1])]; tensor value_5_pad_type_0 = const()[name = tensor("value_5_pad_type_0"), val = tensor("custom")]; tensor value_5_pad_0 = const()[name = tensor("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36932800)))]; tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38112512)))]; tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = var_2117, groups = var_2069, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = var_2115, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor("value_5_cast_fp16")]; tensor var_2124_begin_0 = const()[name = tensor("op_2124_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2124_end_0 = const()[name = tensor("op_2124_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2124_end_mask_0 = const()[name = tensor("op_2124_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2124_cast_fp16 = slice_by_index(begin = var_2124_begin_0, end = var_2124_end_0, end_mask = var_2124_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2124_cast_fp16")]; tensor var_2128_begin_0 = const()[name = tensor("op_2128_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_2128_end_0 = const()[name = tensor("op_2128_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_2128_end_mask_0 = const()[name = tensor("op_2128_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2128_cast_fp16 = slice_by_index(begin = var_2128_begin_0, end = var_2128_end_0, end_mask = var_2128_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2128_cast_fp16")]; tensor var_2132_begin_0 = const()[name = tensor("op_2132_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_2132_end_0 = const()[name = tensor("op_2132_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_2132_end_mask_0 = const()[name = tensor("op_2132_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2132_cast_fp16 = slice_by_index(begin = var_2132_begin_0, end = var_2132_end_0, end_mask = var_2132_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2132_cast_fp16")]; tensor var_2136_begin_0 = const()[name = tensor("op_2136_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_2136_end_0 = const()[name = tensor("op_2136_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_2136_end_mask_0 = const()[name = tensor("op_2136_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2136_cast_fp16 = slice_by_index(begin = var_2136_begin_0, end = var_2136_end_0, end_mask = var_2136_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2136_cast_fp16")]; tensor var_2140_begin_0 = const()[name = tensor("op_2140_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_2140_end_0 = const()[name = tensor("op_2140_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_2140_end_mask_0 = const()[name = tensor("op_2140_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_5_cast_fp16)[name = tensor("op_2140_cast_fp16")]; tensor var_2144_begin_0 = const()[name = tensor("op_2144_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2144_end_0 = const()[name = tensor("op_2144_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_2144_end_mask_0 = const()[name = tensor("op_2144_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2144_cast_fp16 = slice_by_index(begin = var_2144_begin_0, end = var_2144_end_0, end_mask = var_2144_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2144_cast_fp16")]; tensor var_2148_begin_0 = const()[name = tensor("op_2148_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_2148_end_0 = const()[name = tensor("op_2148_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_2148_end_mask_0 = const()[name = tensor("op_2148_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2148_cast_fp16 = slice_by_index(begin = var_2148_begin_0, end = var_2148_end_0, end_mask = var_2148_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2148_cast_fp16")]; tensor var_2152_begin_0 = const()[name = tensor("op_2152_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_2152_end_0 = const()[name = tensor("op_2152_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_2152_end_mask_0 = const()[name = tensor("op_2152_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2152_cast_fp16 = slice_by_index(begin = var_2152_begin_0, end = var_2152_end_0, end_mask = var_2152_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2152_cast_fp16")]; tensor var_2156_begin_0 = const()[name = tensor("op_2156_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_2156_end_0 = const()[name = tensor("op_2156_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_2156_end_mask_0 = const()[name = tensor("op_2156_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2156_cast_fp16 = slice_by_index(begin = var_2156_begin_0, end = var_2156_end_0, end_mask = var_2156_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2156_cast_fp16")]; tensor var_2160_begin_0 = const()[name = tensor("op_2160_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_2160_end_0 = const()[name = tensor("op_2160_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_2160_end_mask_0 = const()[name = tensor("op_2160_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2160_cast_fp16")]; tensor var_2164_begin_0 = const()[name = tensor("op_2164_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2164_end_0 = const()[name = tensor("op_2164_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_2164_end_mask_0 = const()[name = tensor("op_2164_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, x = query_5_cast_fp16)[name = tensor("op_2164_cast_fp16")]; tensor var_2168_begin_0 = const()[name = tensor("op_2168_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_2168_end_0 = const()[name = tensor("op_2168_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_2168_end_mask_0 = const()[name = tensor("op_2168_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_5_cast_fp16)[name = tensor("op_2168_cast_fp16")]; tensor var_2177_begin_0 = const()[name = tensor("op_2177_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2177_end_0 = const()[name = tensor("op_2177_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2177_end_mask_0 = const()[name = tensor("op_2177_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2177_cast_fp16 = slice_by_index(begin = var_2177_begin_0, end = var_2177_end_0, end_mask = var_2177_end_mask_0, x = var_2124_cast_fp16)[name = tensor("op_2177_cast_fp16")]; tensor var_2184_begin_0 = const()[name = tensor("op_2184_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2184_end_0 = const()[name = tensor("op_2184_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2184_end_mask_0 = const()[name = tensor("op_2184_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2184_cast_fp16 = slice_by_index(begin = var_2184_begin_0, end = var_2184_end_0, end_mask = var_2184_end_mask_0, x = var_2124_cast_fp16)[name = tensor("op_2184_cast_fp16")]; tensor var_2191_begin_0 = const()[name = tensor("op_2191_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2191_end_0 = const()[name = tensor("op_2191_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2191_end_mask_0 = const()[name = tensor("op_2191_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2191_cast_fp16 = slice_by_index(begin = var_2191_begin_0, end = var_2191_end_0, end_mask = var_2191_end_mask_0, x = var_2124_cast_fp16)[name = tensor("op_2191_cast_fp16")]; tensor var_2198_begin_0 = const()[name = tensor("op_2198_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2198_end_0 = const()[name = tensor("op_2198_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2198_end_mask_0 = const()[name = tensor("op_2198_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2198_cast_fp16 = slice_by_index(begin = var_2198_begin_0, end = var_2198_end_0, end_mask = var_2198_end_mask_0, x = var_2124_cast_fp16)[name = tensor("op_2198_cast_fp16")]; tensor var_2205_begin_0 = const()[name = tensor("op_2205_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2205_end_0 = const()[name = tensor("op_2205_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2205_end_mask_0 = const()[name = tensor("op_2205_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2205_cast_fp16 = slice_by_index(begin = var_2205_begin_0, end = var_2205_end_0, end_mask = var_2205_end_mask_0, x = var_2128_cast_fp16)[name = tensor("op_2205_cast_fp16")]; tensor var_2212_begin_0 = const()[name = tensor("op_2212_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2212_end_0 = const()[name = tensor("op_2212_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2212_end_mask_0 = const()[name = tensor("op_2212_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2212_cast_fp16 = slice_by_index(begin = var_2212_begin_0, end = var_2212_end_0, end_mask = var_2212_end_mask_0, x = var_2128_cast_fp16)[name = tensor("op_2212_cast_fp16")]; tensor var_2219_begin_0 = const()[name = tensor("op_2219_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2219_end_0 = const()[name = tensor("op_2219_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2219_end_mask_0 = const()[name = tensor("op_2219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2219_cast_fp16 = slice_by_index(begin = var_2219_begin_0, end = var_2219_end_0, end_mask = var_2219_end_mask_0, x = var_2128_cast_fp16)[name = tensor("op_2219_cast_fp16")]; tensor var_2226_begin_0 = const()[name = tensor("op_2226_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2226_end_0 = const()[name = tensor("op_2226_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2226_end_mask_0 = const()[name = tensor("op_2226_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2226_cast_fp16 = slice_by_index(begin = var_2226_begin_0, end = var_2226_end_0, end_mask = var_2226_end_mask_0, x = var_2128_cast_fp16)[name = tensor("op_2226_cast_fp16")]; tensor var_2233_begin_0 = const()[name = tensor("op_2233_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2233_end_0 = const()[name = tensor("op_2233_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2233_end_mask_0 = const()[name = tensor("op_2233_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2233_cast_fp16 = slice_by_index(begin = var_2233_begin_0, end = var_2233_end_0, end_mask = var_2233_end_mask_0, x = var_2132_cast_fp16)[name = tensor("op_2233_cast_fp16")]; tensor var_2240_begin_0 = const()[name = tensor("op_2240_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2240_end_0 = const()[name = tensor("op_2240_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2240_end_mask_0 = const()[name = tensor("op_2240_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2240_cast_fp16 = slice_by_index(begin = var_2240_begin_0, end = var_2240_end_0, end_mask = var_2240_end_mask_0, x = var_2132_cast_fp16)[name = tensor("op_2240_cast_fp16")]; tensor var_2247_begin_0 = const()[name = tensor("op_2247_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2247_end_0 = const()[name = tensor("op_2247_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2247_end_mask_0 = const()[name = tensor("op_2247_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2247_cast_fp16 = slice_by_index(begin = var_2247_begin_0, end = var_2247_end_0, end_mask = var_2247_end_mask_0, x = var_2132_cast_fp16)[name = tensor("op_2247_cast_fp16")]; tensor var_2254_begin_0 = const()[name = tensor("op_2254_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2254_end_0 = const()[name = tensor("op_2254_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2254_end_mask_0 = const()[name = tensor("op_2254_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2254_cast_fp16 = slice_by_index(begin = var_2254_begin_0, end = var_2254_end_0, end_mask = var_2254_end_mask_0, x = var_2132_cast_fp16)[name = tensor("op_2254_cast_fp16")]; tensor var_2261_begin_0 = const()[name = tensor("op_2261_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2261_end_0 = const()[name = tensor("op_2261_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2261_end_mask_0 = const()[name = tensor("op_2261_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2261_cast_fp16 = slice_by_index(begin = var_2261_begin_0, end = var_2261_end_0, end_mask = var_2261_end_mask_0, x = var_2136_cast_fp16)[name = tensor("op_2261_cast_fp16")]; tensor var_2268_begin_0 = const()[name = tensor("op_2268_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2268_end_0 = const()[name = tensor("op_2268_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2268_end_mask_0 = const()[name = tensor("op_2268_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2268_cast_fp16 = slice_by_index(begin = var_2268_begin_0, end = var_2268_end_0, end_mask = var_2268_end_mask_0, x = var_2136_cast_fp16)[name = tensor("op_2268_cast_fp16")]; tensor var_2275_begin_0 = const()[name = tensor("op_2275_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2275_end_0 = const()[name = tensor("op_2275_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2275_end_mask_0 = const()[name = tensor("op_2275_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2275_cast_fp16 = slice_by_index(begin = var_2275_begin_0, end = var_2275_end_0, end_mask = var_2275_end_mask_0, x = var_2136_cast_fp16)[name = tensor("op_2275_cast_fp16")]; tensor var_2282_begin_0 = const()[name = tensor("op_2282_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2282_end_0 = const()[name = tensor("op_2282_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2282_end_mask_0 = const()[name = tensor("op_2282_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2282_cast_fp16 = slice_by_index(begin = var_2282_begin_0, end = var_2282_end_0, end_mask = var_2282_end_mask_0, x = var_2136_cast_fp16)[name = tensor("op_2282_cast_fp16")]; tensor var_2289_begin_0 = const()[name = tensor("op_2289_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2289_end_0 = const()[name = tensor("op_2289_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2289_end_mask_0 = const()[name = tensor("op_2289_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2289_cast_fp16 = slice_by_index(begin = var_2289_begin_0, end = var_2289_end_0, end_mask = var_2289_end_mask_0, x = var_2140_cast_fp16)[name = tensor("op_2289_cast_fp16")]; tensor var_2296_begin_0 = const()[name = tensor("op_2296_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2296_end_0 = const()[name = tensor("op_2296_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2296_end_mask_0 = const()[name = tensor("op_2296_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2296_cast_fp16 = slice_by_index(begin = var_2296_begin_0, end = var_2296_end_0, end_mask = var_2296_end_mask_0, x = var_2140_cast_fp16)[name = tensor("op_2296_cast_fp16")]; tensor var_2303_begin_0 = const()[name = tensor("op_2303_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2303_end_0 = const()[name = tensor("op_2303_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2303_end_mask_0 = const()[name = tensor("op_2303_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2303_cast_fp16 = slice_by_index(begin = var_2303_begin_0, end = var_2303_end_0, end_mask = var_2303_end_mask_0, x = var_2140_cast_fp16)[name = tensor("op_2303_cast_fp16")]; tensor var_2310_begin_0 = const()[name = tensor("op_2310_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2310_end_0 = const()[name = tensor("op_2310_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2310_end_mask_0 = const()[name = tensor("op_2310_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2310_cast_fp16 = slice_by_index(begin = var_2310_begin_0, end = var_2310_end_0, end_mask = var_2310_end_mask_0, x = var_2140_cast_fp16)[name = tensor("op_2310_cast_fp16")]; tensor var_2317_begin_0 = const()[name = tensor("op_2317_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2317_end_0 = const()[name = tensor("op_2317_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2317_end_mask_0 = const()[name = tensor("op_2317_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2317_cast_fp16 = slice_by_index(begin = var_2317_begin_0, end = var_2317_end_0, end_mask = var_2317_end_mask_0, x = var_2144_cast_fp16)[name = tensor("op_2317_cast_fp16")]; tensor var_2324_begin_0 = const()[name = tensor("op_2324_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2324_end_0 = const()[name = tensor("op_2324_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2324_end_mask_0 = const()[name = tensor("op_2324_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2324_cast_fp16 = slice_by_index(begin = var_2324_begin_0, end = var_2324_end_0, end_mask = var_2324_end_mask_0, x = var_2144_cast_fp16)[name = tensor("op_2324_cast_fp16")]; tensor var_2331_begin_0 = const()[name = tensor("op_2331_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2331_end_0 = const()[name = tensor("op_2331_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2331_end_mask_0 = const()[name = tensor("op_2331_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2331_cast_fp16 = slice_by_index(begin = var_2331_begin_0, end = var_2331_end_0, end_mask = var_2331_end_mask_0, x = var_2144_cast_fp16)[name = tensor("op_2331_cast_fp16")]; tensor var_2338_begin_0 = const()[name = tensor("op_2338_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2338_end_0 = const()[name = tensor("op_2338_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2338_end_mask_0 = const()[name = tensor("op_2338_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2338_cast_fp16 = slice_by_index(begin = var_2338_begin_0, end = var_2338_end_0, end_mask = var_2338_end_mask_0, x = var_2144_cast_fp16)[name = tensor("op_2338_cast_fp16")]; tensor var_2345_begin_0 = const()[name = tensor("op_2345_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2345_end_0 = const()[name = tensor("op_2345_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2345_end_mask_0 = const()[name = tensor("op_2345_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2345_cast_fp16 = slice_by_index(begin = var_2345_begin_0, end = var_2345_end_0, end_mask = var_2345_end_mask_0, x = var_2148_cast_fp16)[name = tensor("op_2345_cast_fp16")]; tensor var_2352_begin_0 = const()[name = tensor("op_2352_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2352_end_0 = const()[name = tensor("op_2352_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2352_end_mask_0 = const()[name = tensor("op_2352_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2352_cast_fp16 = slice_by_index(begin = var_2352_begin_0, end = var_2352_end_0, end_mask = var_2352_end_mask_0, x = var_2148_cast_fp16)[name = tensor("op_2352_cast_fp16")]; tensor var_2359_begin_0 = const()[name = tensor("op_2359_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2359_end_0 = const()[name = tensor("op_2359_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2359_end_mask_0 = const()[name = tensor("op_2359_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2359_cast_fp16 = slice_by_index(begin = var_2359_begin_0, end = var_2359_end_0, end_mask = var_2359_end_mask_0, x = var_2148_cast_fp16)[name = tensor("op_2359_cast_fp16")]; tensor var_2366_begin_0 = const()[name = tensor("op_2366_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2366_end_0 = const()[name = tensor("op_2366_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2366_end_mask_0 = const()[name = tensor("op_2366_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2366_cast_fp16 = slice_by_index(begin = var_2366_begin_0, end = var_2366_end_0, end_mask = var_2366_end_mask_0, x = var_2148_cast_fp16)[name = tensor("op_2366_cast_fp16")]; tensor var_2373_begin_0 = const()[name = tensor("op_2373_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2373_end_0 = const()[name = tensor("op_2373_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2373_end_mask_0 = const()[name = tensor("op_2373_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2373_cast_fp16 = slice_by_index(begin = var_2373_begin_0, end = var_2373_end_0, end_mask = var_2373_end_mask_0, x = var_2152_cast_fp16)[name = tensor("op_2373_cast_fp16")]; tensor var_2380_begin_0 = const()[name = tensor("op_2380_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2380_end_0 = const()[name = tensor("op_2380_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2380_end_mask_0 = const()[name = tensor("op_2380_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2380_cast_fp16 = slice_by_index(begin = var_2380_begin_0, end = var_2380_end_0, end_mask = var_2380_end_mask_0, x = var_2152_cast_fp16)[name = tensor("op_2380_cast_fp16")]; tensor var_2387_begin_0 = const()[name = tensor("op_2387_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2387_end_0 = const()[name = tensor("op_2387_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2387_end_mask_0 = const()[name = tensor("op_2387_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2387_cast_fp16 = slice_by_index(begin = var_2387_begin_0, end = var_2387_end_0, end_mask = var_2387_end_mask_0, x = var_2152_cast_fp16)[name = tensor("op_2387_cast_fp16")]; tensor var_2394_begin_0 = const()[name = tensor("op_2394_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2394_end_0 = const()[name = tensor("op_2394_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2394_end_mask_0 = const()[name = tensor("op_2394_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2394_cast_fp16 = slice_by_index(begin = var_2394_begin_0, end = var_2394_end_0, end_mask = var_2394_end_mask_0, x = var_2152_cast_fp16)[name = tensor("op_2394_cast_fp16")]; tensor var_2401_begin_0 = const()[name = tensor("op_2401_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2401_end_0 = const()[name = tensor("op_2401_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2401_end_mask_0 = const()[name = tensor("op_2401_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2401_cast_fp16 = slice_by_index(begin = var_2401_begin_0, end = var_2401_end_0, end_mask = var_2401_end_mask_0, x = var_2156_cast_fp16)[name = tensor("op_2401_cast_fp16")]; tensor var_2408_begin_0 = const()[name = tensor("op_2408_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2408_end_0 = const()[name = tensor("op_2408_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2408_end_mask_0 = const()[name = tensor("op_2408_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2408_cast_fp16 = slice_by_index(begin = var_2408_begin_0, end = var_2408_end_0, end_mask = var_2408_end_mask_0, x = var_2156_cast_fp16)[name = tensor("op_2408_cast_fp16")]; tensor var_2415_begin_0 = const()[name = tensor("op_2415_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2415_end_0 = const()[name = tensor("op_2415_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2415_end_mask_0 = const()[name = tensor("op_2415_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2415_cast_fp16 = slice_by_index(begin = var_2415_begin_0, end = var_2415_end_0, end_mask = var_2415_end_mask_0, x = var_2156_cast_fp16)[name = tensor("op_2415_cast_fp16")]; tensor var_2422_begin_0 = const()[name = tensor("op_2422_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2422_end_0 = const()[name = tensor("op_2422_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2422_end_mask_0 = const()[name = tensor("op_2422_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2422_cast_fp16 = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = var_2156_cast_fp16)[name = tensor("op_2422_cast_fp16")]; tensor var_2429_begin_0 = const()[name = tensor("op_2429_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2429_end_0 = const()[name = tensor("op_2429_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2429_end_mask_0 = const()[name = tensor("op_2429_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2429_cast_fp16 = slice_by_index(begin = var_2429_begin_0, end = var_2429_end_0, end_mask = var_2429_end_mask_0, x = var_2160_cast_fp16)[name = tensor("op_2429_cast_fp16")]; tensor var_2436_begin_0 = const()[name = tensor("op_2436_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2436_end_0 = const()[name = tensor("op_2436_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2436_end_mask_0 = const()[name = tensor("op_2436_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2436_cast_fp16 = slice_by_index(begin = var_2436_begin_0, end = var_2436_end_0, end_mask = var_2436_end_mask_0, x = var_2160_cast_fp16)[name = tensor("op_2436_cast_fp16")]; tensor var_2443_begin_0 = const()[name = tensor("op_2443_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2443_end_0 = const()[name = tensor("op_2443_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2443_end_mask_0 = const()[name = tensor("op_2443_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2443_cast_fp16 = slice_by_index(begin = var_2443_begin_0, end = var_2443_end_0, end_mask = var_2443_end_mask_0, x = var_2160_cast_fp16)[name = tensor("op_2443_cast_fp16")]; tensor var_2450_begin_0 = const()[name = tensor("op_2450_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2450_end_0 = const()[name = tensor("op_2450_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2450_end_mask_0 = const()[name = tensor("op_2450_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2450_cast_fp16 = slice_by_index(begin = var_2450_begin_0, end = var_2450_end_0, end_mask = var_2450_end_mask_0, x = var_2160_cast_fp16)[name = tensor("op_2450_cast_fp16")]; tensor var_2457_begin_0 = const()[name = tensor("op_2457_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2457_end_0 = const()[name = tensor("op_2457_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2457_end_mask_0 = const()[name = tensor("op_2457_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2457_cast_fp16 = slice_by_index(begin = var_2457_begin_0, end = var_2457_end_0, end_mask = var_2457_end_mask_0, x = var_2164_cast_fp16)[name = tensor("op_2457_cast_fp16")]; tensor var_2464_begin_0 = const()[name = tensor("op_2464_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2464_end_0 = const()[name = tensor("op_2464_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2464_end_mask_0 = const()[name = tensor("op_2464_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2464_cast_fp16 = slice_by_index(begin = var_2464_begin_0, end = var_2464_end_0, end_mask = var_2464_end_mask_0, x = var_2164_cast_fp16)[name = tensor("op_2464_cast_fp16")]; tensor var_2471_begin_0 = const()[name = tensor("op_2471_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2471_end_0 = const()[name = tensor("op_2471_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2471_end_mask_0 = const()[name = tensor("op_2471_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2471_cast_fp16 = slice_by_index(begin = var_2471_begin_0, end = var_2471_end_0, end_mask = var_2471_end_mask_0, x = var_2164_cast_fp16)[name = tensor("op_2471_cast_fp16")]; tensor var_2478_begin_0 = const()[name = tensor("op_2478_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2478_end_0 = const()[name = tensor("op_2478_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2478_end_mask_0 = const()[name = tensor("op_2478_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2478_cast_fp16 = slice_by_index(begin = var_2478_begin_0, end = var_2478_end_0, end_mask = var_2478_end_mask_0, x = var_2164_cast_fp16)[name = tensor("op_2478_cast_fp16")]; tensor var_2485_begin_0 = const()[name = tensor("op_2485_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2485_end_0 = const()[name = tensor("op_2485_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_2485_end_mask_0 = const()[name = tensor("op_2485_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2485_cast_fp16 = slice_by_index(begin = var_2485_begin_0, end = var_2485_end_0, end_mask = var_2485_end_mask_0, x = var_2168_cast_fp16)[name = tensor("op_2485_cast_fp16")]; tensor var_2492_begin_0 = const()[name = tensor("op_2492_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_2492_end_0 = const()[name = tensor("op_2492_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_2492_end_mask_0 = const()[name = tensor("op_2492_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2492_cast_fp16 = slice_by_index(begin = var_2492_begin_0, end = var_2492_end_0, end_mask = var_2492_end_mask_0, x = var_2168_cast_fp16)[name = tensor("op_2492_cast_fp16")]; tensor var_2499_begin_0 = const()[name = tensor("op_2499_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_2499_end_0 = const()[name = tensor("op_2499_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_2499_end_mask_0 = const()[name = tensor("op_2499_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2499_cast_fp16 = slice_by_index(begin = var_2499_begin_0, end = var_2499_end_0, end_mask = var_2499_end_mask_0, x = var_2168_cast_fp16)[name = tensor("op_2499_cast_fp16")]; tensor var_2506_begin_0 = const()[name = tensor("op_2506_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_2506_end_0 = const()[name = tensor("op_2506_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2506_end_mask_0 = const()[name = tensor("op_2506_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = var_2168_cast_fp16)[name = tensor("op_2506_cast_fp16")]; tensor k_5_perm_0 = const()[name = tensor("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_2511_begin_0 = const()[name = tensor("op_2511_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2511_end_0 = const()[name = tensor("op_2511_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_2511_end_mask_0 = const()[name = tensor("op_2511_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_9 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = tensor("transpose_9")]; 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 = transpose_9)[name = tensor("op_2511_cast_fp16")]; tensor var_2515_begin_0 = const()[name = tensor("op_2515_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_2515_end_0 = const()[name = tensor("op_2515_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_2515_end_mask_0 = const()[name = tensor("op_2515_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2515_cast_fp16 = slice_by_index(begin = var_2515_begin_0, end = var_2515_end_0, end_mask = var_2515_end_mask_0, x = transpose_9)[name = tensor("op_2515_cast_fp16")]; tensor var_2519_begin_0 = const()[name = tensor("op_2519_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_2519_end_0 = const()[name = tensor("op_2519_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_2519_end_mask_0 = const()[name = tensor("op_2519_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2519_cast_fp16 = slice_by_index(begin = var_2519_begin_0, end = var_2519_end_0, end_mask = var_2519_end_mask_0, x = transpose_9)[name = tensor("op_2519_cast_fp16")]; tensor var_2523_begin_0 = const()[name = tensor("op_2523_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_2523_end_0 = const()[name = tensor("op_2523_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_2523_end_mask_0 = const()[name = tensor("op_2523_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2523_cast_fp16 = slice_by_index(begin = var_2523_begin_0, end = var_2523_end_0, end_mask = var_2523_end_mask_0, x = transpose_9)[name = tensor("op_2523_cast_fp16")]; tensor var_2527_begin_0 = const()[name = tensor("op_2527_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_2527_end_0 = const()[name = tensor("op_2527_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_2527_end_mask_0 = const()[name = tensor("op_2527_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2527_cast_fp16 = slice_by_index(begin = var_2527_begin_0, end = var_2527_end_0, end_mask = var_2527_end_mask_0, x = transpose_9)[name = tensor("op_2527_cast_fp16")]; tensor var_2531_begin_0 = const()[name = tensor("op_2531_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_2531_end_0 = const()[name = tensor("op_2531_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_2531_end_mask_0 = const()[name = tensor("op_2531_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2531_cast_fp16 = slice_by_index(begin = var_2531_begin_0, end = var_2531_end_0, end_mask = var_2531_end_mask_0, x = transpose_9)[name = tensor("op_2531_cast_fp16")]; tensor var_2535_begin_0 = const()[name = tensor("op_2535_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_2535_end_0 = const()[name = tensor("op_2535_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_2535_end_mask_0 = const()[name = tensor("op_2535_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2535_cast_fp16 = slice_by_index(begin = var_2535_begin_0, end = var_2535_end_0, end_mask = var_2535_end_mask_0, x = transpose_9)[name = tensor("op_2535_cast_fp16")]; tensor var_2539_begin_0 = const()[name = tensor("op_2539_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_2539_end_0 = const()[name = tensor("op_2539_end_0"), val = tensor([1, 1500, 1, 512])]; 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 = transpose_9)[name = tensor("op_2539_cast_fp16")]; tensor var_2543_begin_0 = const()[name = tensor("op_2543_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_2543_end_0 = const()[name = tensor("op_2543_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_2543_end_mask_0 = const()[name = tensor("op_2543_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = transpose_9)[name = tensor("op_2543_cast_fp16")]; tensor var_2547_begin_0 = const()[name = tensor("op_2547_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_2547_end_0 = const()[name = tensor("op_2547_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_2547_end_mask_0 = const()[name = tensor("op_2547_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2547_cast_fp16 = slice_by_index(begin = var_2547_begin_0, end = var_2547_end_0, end_mask = var_2547_end_mask_0, x = transpose_9)[name = tensor("op_2547_cast_fp16")]; tensor var_2551_begin_0 = const()[name = tensor("op_2551_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_2551_end_0 = const()[name = tensor("op_2551_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_2551_end_mask_0 = const()[name = tensor("op_2551_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2551_cast_fp16 = slice_by_index(begin = var_2551_begin_0, end = var_2551_end_0, end_mask = var_2551_end_mask_0, x = transpose_9)[name = tensor("op_2551_cast_fp16")]; tensor var_2555_begin_0 = const()[name = tensor("op_2555_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_2555_end_0 = const()[name = tensor("op_2555_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_2555_end_mask_0 = const()[name = tensor("op_2555_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_2555_cast_fp16 = slice_by_index(begin = var_2555_begin_0, end = var_2555_end_0, end_mask = var_2555_end_mask_0, x = transpose_9)[name = tensor("op_2555_cast_fp16")]; tensor var_2557_begin_0 = const()[name = tensor("op_2557_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_2557_end_0 = const()[name = tensor("op_2557_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_2557_end_mask_0 = const()[name = tensor("op_2557_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2557_cast_fp16 = slice_by_index(begin = var_2557_begin_0, end = var_2557_end_0, end_mask = var_2557_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2557_cast_fp16")]; tensor var_2561_begin_0 = const()[name = tensor("op_2561_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_2561_end_0 = const()[name = tensor("op_2561_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_2561_end_mask_0 = const()[name = tensor("op_2561_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2561_cast_fp16 = slice_by_index(begin = var_2561_begin_0, end = var_2561_end_0, end_mask = var_2561_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2561_cast_fp16")]; tensor var_2565_begin_0 = const()[name = tensor("op_2565_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_2565_end_0 = const()[name = tensor("op_2565_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_2565_end_mask_0 = const()[name = tensor("op_2565_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2565_cast_fp16 = slice_by_index(begin = var_2565_begin_0, end = var_2565_end_0, end_mask = var_2565_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2565_cast_fp16")]; tensor var_2569_begin_0 = const()[name = tensor("op_2569_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_2569_end_0 = const()[name = tensor("op_2569_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_2569_end_mask_0 = const()[name = tensor("op_2569_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2569_cast_fp16 = slice_by_index(begin = var_2569_begin_0, end = var_2569_end_0, end_mask = var_2569_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2569_cast_fp16")]; tensor var_2573_begin_0 = const()[name = tensor("op_2573_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_2573_end_0 = const()[name = tensor("op_2573_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_2573_end_mask_0 = const()[name = tensor("op_2573_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2573_cast_fp16 = slice_by_index(begin = var_2573_begin_0, end = var_2573_end_0, end_mask = var_2573_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2573_cast_fp16")]; tensor var_2577_begin_0 = const()[name = tensor("op_2577_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_2577_end_0 = const()[name = tensor("op_2577_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_2577_end_mask_0 = const()[name = tensor("op_2577_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2577_cast_fp16 = slice_by_index(begin = var_2577_begin_0, end = var_2577_end_0, end_mask = var_2577_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2577_cast_fp16")]; tensor var_2581_begin_0 = const()[name = tensor("op_2581_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_2581_end_0 = const()[name = tensor("op_2581_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_2581_end_mask_0 = const()[name = tensor("op_2581_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_5_cast_fp16)[name = tensor("op_2581_cast_fp16")]; tensor var_2585_begin_0 = const()[name = tensor("op_2585_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_2585_end_0 = const()[name = tensor("op_2585_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_2585_end_mask_0 = const()[name = tensor("op_2585_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2585_cast_fp16 = slice_by_index(begin = var_2585_begin_0, end = var_2585_end_0, end_mask = var_2585_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2585_cast_fp16")]; tensor var_2589_begin_0 = const()[name = tensor("op_2589_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_2589_end_0 = const()[name = tensor("op_2589_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_2589_end_mask_0 = const()[name = tensor("op_2589_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2589_cast_fp16 = slice_by_index(begin = var_2589_begin_0, end = var_2589_end_0, end_mask = var_2589_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2589_cast_fp16")]; tensor var_2593_begin_0 = const()[name = tensor("op_2593_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_2593_end_0 = const()[name = tensor("op_2593_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_2593_end_mask_0 = const()[name = tensor("op_2593_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2593_cast_fp16 = slice_by_index(begin = var_2593_begin_0, end = var_2593_end_0, end_mask = var_2593_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2593_cast_fp16")]; tensor var_2597_begin_0 = const()[name = tensor("op_2597_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_2597_end_0 = const()[name = tensor("op_2597_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_2597_end_mask_0 = const()[name = tensor("op_2597_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2597_cast_fp16 = slice_by_index(begin = var_2597_begin_0, end = var_2597_end_0, end_mask = var_2597_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2597_cast_fp16")]; tensor var_2601_begin_0 = const()[name = tensor("op_2601_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_2601_end_0 = const()[name = tensor("op_2601_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_2601_end_mask_0 = const()[name = tensor("op_2601_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_2601_cast_fp16 = slice_by_index(begin = var_2601_begin_0, end = var_2601_end_0, end_mask = var_2601_end_mask_0, x = value_5_cast_fp16)[name = tensor("op_2601_cast_fp16")]; tensor var_2605_equation_0 = const()[name = tensor("op_2605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2605_cast_fp16 = einsum(equation = var_2605_equation_0, values = (var_2511_cast_fp16, var_2177_cast_fp16))[name = tensor("op_2605_cast_fp16")]; tensor var_2606_to_fp16 = const()[name = tensor("op_2606_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_193_cast_fp16 = mul(x = var_2605_cast_fp16, y = var_2606_to_fp16)[name = tensor("aw_chunk_193_cast_fp16")]; tensor var_2609_equation_0 = const()[name = tensor("op_2609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2609_cast_fp16 = einsum(equation = var_2609_equation_0, values = (var_2511_cast_fp16, var_2184_cast_fp16))[name = tensor("op_2609_cast_fp16")]; tensor var_2610_to_fp16 = const()[name = tensor("op_2610_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_195_cast_fp16 = mul(x = var_2609_cast_fp16, y = var_2610_to_fp16)[name = tensor("aw_chunk_195_cast_fp16")]; tensor var_2613_equation_0 = const()[name = tensor("op_2613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2613_cast_fp16 = einsum(equation = var_2613_equation_0, values = (var_2511_cast_fp16, var_2191_cast_fp16))[name = tensor("op_2613_cast_fp16")]; tensor var_2614_to_fp16 = const()[name = tensor("op_2614_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_197_cast_fp16 = mul(x = var_2613_cast_fp16, y = var_2614_to_fp16)[name = tensor("aw_chunk_197_cast_fp16")]; tensor var_2617_equation_0 = const()[name = tensor("op_2617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2617_cast_fp16 = einsum(equation = var_2617_equation_0, values = (var_2511_cast_fp16, var_2198_cast_fp16))[name = tensor("op_2617_cast_fp16")]; tensor var_2618_to_fp16 = const()[name = tensor("op_2618_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_199_cast_fp16 = mul(x = var_2617_cast_fp16, y = var_2618_to_fp16)[name = tensor("aw_chunk_199_cast_fp16")]; tensor var_2621_equation_0 = const()[name = tensor("op_2621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2621_cast_fp16 = einsum(equation = var_2621_equation_0, values = (var_2515_cast_fp16, var_2205_cast_fp16))[name = tensor("op_2621_cast_fp16")]; tensor var_2622_to_fp16 = const()[name = tensor("op_2622_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_201_cast_fp16 = mul(x = var_2621_cast_fp16, y = var_2622_to_fp16)[name = tensor("aw_chunk_201_cast_fp16")]; tensor var_2625_equation_0 = const()[name = tensor("op_2625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2625_cast_fp16 = einsum(equation = var_2625_equation_0, values = (var_2515_cast_fp16, var_2212_cast_fp16))[name = tensor("op_2625_cast_fp16")]; tensor var_2626_to_fp16 = const()[name = tensor("op_2626_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_203_cast_fp16 = mul(x = var_2625_cast_fp16, y = var_2626_to_fp16)[name = tensor("aw_chunk_203_cast_fp16")]; tensor var_2629_equation_0 = const()[name = tensor("op_2629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2629_cast_fp16 = einsum(equation = var_2629_equation_0, values = (var_2515_cast_fp16, var_2219_cast_fp16))[name = tensor("op_2629_cast_fp16")]; tensor var_2630_to_fp16 = const()[name = tensor("op_2630_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_205_cast_fp16 = mul(x = var_2629_cast_fp16, y = var_2630_to_fp16)[name = tensor("aw_chunk_205_cast_fp16")]; tensor var_2633_equation_0 = const()[name = tensor("op_2633_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2633_cast_fp16 = einsum(equation = var_2633_equation_0, values = (var_2515_cast_fp16, var_2226_cast_fp16))[name = tensor("op_2633_cast_fp16")]; tensor var_2634_to_fp16 = const()[name = tensor("op_2634_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_207_cast_fp16 = mul(x = var_2633_cast_fp16, y = var_2634_to_fp16)[name = tensor("aw_chunk_207_cast_fp16")]; tensor var_2637_equation_0 = const()[name = tensor("op_2637_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2637_cast_fp16 = einsum(equation = var_2637_equation_0, values = (var_2519_cast_fp16, var_2233_cast_fp16))[name = tensor("op_2637_cast_fp16")]; tensor var_2638_to_fp16 = const()[name = tensor("op_2638_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_209_cast_fp16 = mul(x = var_2637_cast_fp16, y = var_2638_to_fp16)[name = tensor("aw_chunk_209_cast_fp16")]; tensor var_2641_equation_0 = const()[name = tensor("op_2641_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2641_cast_fp16 = einsum(equation = var_2641_equation_0, values = (var_2519_cast_fp16, var_2240_cast_fp16))[name = tensor("op_2641_cast_fp16")]; tensor var_2642_to_fp16 = const()[name = tensor("op_2642_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_211_cast_fp16 = mul(x = var_2641_cast_fp16, y = var_2642_to_fp16)[name = tensor("aw_chunk_211_cast_fp16")]; tensor var_2645_equation_0 = const()[name = tensor("op_2645_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2645_cast_fp16 = einsum(equation = var_2645_equation_0, values = (var_2519_cast_fp16, var_2247_cast_fp16))[name = tensor("op_2645_cast_fp16")]; tensor var_2646_to_fp16 = const()[name = tensor("op_2646_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_213_cast_fp16 = mul(x = var_2645_cast_fp16, y = var_2646_to_fp16)[name = tensor("aw_chunk_213_cast_fp16")]; tensor var_2649_equation_0 = const()[name = tensor("op_2649_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2649_cast_fp16 = einsum(equation = var_2649_equation_0, values = (var_2519_cast_fp16, var_2254_cast_fp16))[name = tensor("op_2649_cast_fp16")]; tensor var_2650_to_fp16 = const()[name = tensor("op_2650_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_215_cast_fp16 = mul(x = var_2649_cast_fp16, y = var_2650_to_fp16)[name = tensor("aw_chunk_215_cast_fp16")]; tensor var_2653_equation_0 = const()[name = tensor("op_2653_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2653_cast_fp16 = einsum(equation = var_2653_equation_0, values = (var_2523_cast_fp16, var_2261_cast_fp16))[name = tensor("op_2653_cast_fp16")]; tensor var_2654_to_fp16 = const()[name = tensor("op_2654_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_217_cast_fp16 = mul(x = var_2653_cast_fp16, y = var_2654_to_fp16)[name = tensor("aw_chunk_217_cast_fp16")]; tensor var_2657_equation_0 = const()[name = tensor("op_2657_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2657_cast_fp16 = einsum(equation = var_2657_equation_0, values = (var_2523_cast_fp16, var_2268_cast_fp16))[name = tensor("op_2657_cast_fp16")]; tensor var_2658_to_fp16 = const()[name = tensor("op_2658_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_219_cast_fp16 = mul(x = var_2657_cast_fp16, y = var_2658_to_fp16)[name = tensor("aw_chunk_219_cast_fp16")]; tensor var_2661_equation_0 = const()[name = tensor("op_2661_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2661_cast_fp16 = einsum(equation = var_2661_equation_0, values = (var_2523_cast_fp16, var_2275_cast_fp16))[name = tensor("op_2661_cast_fp16")]; tensor var_2662_to_fp16 = const()[name = tensor("op_2662_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_221_cast_fp16 = mul(x = var_2661_cast_fp16, y = var_2662_to_fp16)[name = tensor("aw_chunk_221_cast_fp16")]; tensor var_2665_equation_0 = const()[name = tensor("op_2665_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2665_cast_fp16 = einsum(equation = var_2665_equation_0, values = (var_2523_cast_fp16, var_2282_cast_fp16))[name = tensor("op_2665_cast_fp16")]; tensor var_2666_to_fp16 = const()[name = tensor("op_2666_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_223_cast_fp16 = mul(x = var_2665_cast_fp16, y = var_2666_to_fp16)[name = tensor("aw_chunk_223_cast_fp16")]; tensor var_2669_equation_0 = const()[name = tensor("op_2669_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2669_cast_fp16 = einsum(equation = var_2669_equation_0, values = (var_2527_cast_fp16, var_2289_cast_fp16))[name = tensor("op_2669_cast_fp16")]; tensor var_2670_to_fp16 = const()[name = tensor("op_2670_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_225_cast_fp16 = mul(x = var_2669_cast_fp16, y = var_2670_to_fp16)[name = tensor("aw_chunk_225_cast_fp16")]; tensor var_2673_equation_0 = const()[name = tensor("op_2673_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2673_cast_fp16 = einsum(equation = var_2673_equation_0, values = (var_2527_cast_fp16, var_2296_cast_fp16))[name = tensor("op_2673_cast_fp16")]; tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_227_cast_fp16 = mul(x = var_2673_cast_fp16, y = var_2674_to_fp16)[name = tensor("aw_chunk_227_cast_fp16")]; tensor var_2677_equation_0 = const()[name = tensor("op_2677_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2677_cast_fp16 = einsum(equation = var_2677_equation_0, values = (var_2527_cast_fp16, var_2303_cast_fp16))[name = tensor("op_2677_cast_fp16")]; tensor var_2678_to_fp16 = const()[name = tensor("op_2678_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_229_cast_fp16 = mul(x = var_2677_cast_fp16, y = var_2678_to_fp16)[name = tensor("aw_chunk_229_cast_fp16")]; tensor var_2681_equation_0 = const()[name = tensor("op_2681_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2681_cast_fp16 = einsum(equation = var_2681_equation_0, values = (var_2527_cast_fp16, var_2310_cast_fp16))[name = tensor("op_2681_cast_fp16")]; tensor var_2682_to_fp16 = const()[name = tensor("op_2682_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_231_cast_fp16 = mul(x = var_2681_cast_fp16, y = var_2682_to_fp16)[name = tensor("aw_chunk_231_cast_fp16")]; tensor var_2685_equation_0 = const()[name = tensor("op_2685_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2685_cast_fp16 = einsum(equation = var_2685_equation_0, values = (var_2531_cast_fp16, var_2317_cast_fp16))[name = tensor("op_2685_cast_fp16")]; tensor var_2686_to_fp16 = const()[name = tensor("op_2686_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_233_cast_fp16 = mul(x = var_2685_cast_fp16, y = var_2686_to_fp16)[name = tensor("aw_chunk_233_cast_fp16")]; tensor var_2689_equation_0 = const()[name = tensor("op_2689_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2689_cast_fp16 = einsum(equation = var_2689_equation_0, values = (var_2531_cast_fp16, var_2324_cast_fp16))[name = tensor("op_2689_cast_fp16")]; tensor var_2690_to_fp16 = const()[name = tensor("op_2690_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_235_cast_fp16 = mul(x = var_2689_cast_fp16, y = var_2690_to_fp16)[name = tensor("aw_chunk_235_cast_fp16")]; tensor var_2693_equation_0 = const()[name = tensor("op_2693_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2693_cast_fp16 = einsum(equation = var_2693_equation_0, values = (var_2531_cast_fp16, var_2331_cast_fp16))[name = tensor("op_2693_cast_fp16")]; tensor var_2694_to_fp16 = const()[name = tensor("op_2694_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_237_cast_fp16 = mul(x = var_2693_cast_fp16, y = var_2694_to_fp16)[name = tensor("aw_chunk_237_cast_fp16")]; tensor var_2697_equation_0 = const()[name = tensor("op_2697_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2697_cast_fp16 = einsum(equation = var_2697_equation_0, values = (var_2531_cast_fp16, var_2338_cast_fp16))[name = tensor("op_2697_cast_fp16")]; tensor var_2698_to_fp16 = const()[name = tensor("op_2698_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_239_cast_fp16 = mul(x = var_2697_cast_fp16, y = var_2698_to_fp16)[name = tensor("aw_chunk_239_cast_fp16")]; tensor var_2701_equation_0 = const()[name = tensor("op_2701_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2701_cast_fp16 = einsum(equation = var_2701_equation_0, values = (var_2535_cast_fp16, var_2345_cast_fp16))[name = tensor("op_2701_cast_fp16")]; tensor var_2702_to_fp16 = const()[name = tensor("op_2702_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_241_cast_fp16 = mul(x = var_2701_cast_fp16, y = var_2702_to_fp16)[name = tensor("aw_chunk_241_cast_fp16")]; tensor var_2705_equation_0 = const()[name = tensor("op_2705_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2705_cast_fp16 = einsum(equation = var_2705_equation_0, values = (var_2535_cast_fp16, var_2352_cast_fp16))[name = tensor("op_2705_cast_fp16")]; tensor var_2706_to_fp16 = const()[name = tensor("op_2706_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_243_cast_fp16 = mul(x = var_2705_cast_fp16, y = var_2706_to_fp16)[name = tensor("aw_chunk_243_cast_fp16")]; tensor var_2709_equation_0 = const()[name = tensor("op_2709_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2709_cast_fp16 = einsum(equation = var_2709_equation_0, values = (var_2535_cast_fp16, var_2359_cast_fp16))[name = tensor("op_2709_cast_fp16")]; tensor var_2710_to_fp16 = const()[name = tensor("op_2710_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_245_cast_fp16 = mul(x = var_2709_cast_fp16, y = var_2710_to_fp16)[name = tensor("aw_chunk_245_cast_fp16")]; tensor var_2713_equation_0 = const()[name = tensor("op_2713_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2713_cast_fp16 = einsum(equation = var_2713_equation_0, values = (var_2535_cast_fp16, var_2366_cast_fp16))[name = tensor("op_2713_cast_fp16")]; tensor var_2714_to_fp16 = const()[name = tensor("op_2714_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_247_cast_fp16 = mul(x = var_2713_cast_fp16, y = var_2714_to_fp16)[name = tensor("aw_chunk_247_cast_fp16")]; tensor var_2717_equation_0 = const()[name = tensor("op_2717_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2717_cast_fp16 = einsum(equation = var_2717_equation_0, values = (var_2539_cast_fp16, var_2373_cast_fp16))[name = tensor("op_2717_cast_fp16")]; tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_249_cast_fp16 = mul(x = var_2717_cast_fp16, y = var_2718_to_fp16)[name = tensor("aw_chunk_249_cast_fp16")]; tensor var_2721_equation_0 = const()[name = tensor("op_2721_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2721_cast_fp16 = einsum(equation = var_2721_equation_0, values = (var_2539_cast_fp16, var_2380_cast_fp16))[name = tensor("op_2721_cast_fp16")]; tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_251_cast_fp16 = mul(x = var_2721_cast_fp16, y = var_2722_to_fp16)[name = tensor("aw_chunk_251_cast_fp16")]; tensor var_2725_equation_0 = const()[name = tensor("op_2725_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2725_cast_fp16 = einsum(equation = var_2725_equation_0, values = (var_2539_cast_fp16, var_2387_cast_fp16))[name = tensor("op_2725_cast_fp16")]; tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_253_cast_fp16 = mul(x = var_2725_cast_fp16, y = var_2726_to_fp16)[name = tensor("aw_chunk_253_cast_fp16")]; tensor var_2729_equation_0 = const()[name = tensor("op_2729_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2729_cast_fp16 = einsum(equation = var_2729_equation_0, values = (var_2539_cast_fp16, var_2394_cast_fp16))[name = tensor("op_2729_cast_fp16")]; tensor var_2730_to_fp16 = const()[name = tensor("op_2730_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_255_cast_fp16 = mul(x = var_2729_cast_fp16, y = var_2730_to_fp16)[name = tensor("aw_chunk_255_cast_fp16")]; tensor var_2733_equation_0 = const()[name = tensor("op_2733_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2733_cast_fp16 = einsum(equation = var_2733_equation_0, values = (var_2543_cast_fp16, var_2401_cast_fp16))[name = tensor("op_2733_cast_fp16")]; tensor var_2734_to_fp16 = const()[name = tensor("op_2734_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_257_cast_fp16 = mul(x = var_2733_cast_fp16, y = var_2734_to_fp16)[name = tensor("aw_chunk_257_cast_fp16")]; tensor var_2737_equation_0 = const()[name = tensor("op_2737_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2737_cast_fp16 = einsum(equation = var_2737_equation_0, values = (var_2543_cast_fp16, var_2408_cast_fp16))[name = tensor("op_2737_cast_fp16")]; tensor var_2738_to_fp16 = const()[name = tensor("op_2738_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_259_cast_fp16 = mul(x = var_2737_cast_fp16, y = var_2738_to_fp16)[name = tensor("aw_chunk_259_cast_fp16")]; tensor var_2741_equation_0 = const()[name = tensor("op_2741_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2741_cast_fp16 = einsum(equation = var_2741_equation_0, values = (var_2543_cast_fp16, var_2415_cast_fp16))[name = tensor("op_2741_cast_fp16")]; tensor var_2742_to_fp16 = const()[name = tensor("op_2742_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_261_cast_fp16 = mul(x = var_2741_cast_fp16, y = var_2742_to_fp16)[name = tensor("aw_chunk_261_cast_fp16")]; tensor var_2745_equation_0 = const()[name = tensor("op_2745_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2745_cast_fp16 = einsum(equation = var_2745_equation_0, values = (var_2543_cast_fp16, var_2422_cast_fp16))[name = tensor("op_2745_cast_fp16")]; tensor var_2746_to_fp16 = const()[name = tensor("op_2746_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_263_cast_fp16 = mul(x = var_2745_cast_fp16, y = var_2746_to_fp16)[name = tensor("aw_chunk_263_cast_fp16")]; tensor var_2749_equation_0 = const()[name = tensor("op_2749_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2749_cast_fp16 = einsum(equation = var_2749_equation_0, values = (var_2547_cast_fp16, var_2429_cast_fp16))[name = tensor("op_2749_cast_fp16")]; tensor var_2750_to_fp16 = const()[name = tensor("op_2750_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_265_cast_fp16 = mul(x = var_2749_cast_fp16, y = var_2750_to_fp16)[name = tensor("aw_chunk_265_cast_fp16")]; tensor var_2753_equation_0 = const()[name = tensor("op_2753_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2753_cast_fp16 = einsum(equation = var_2753_equation_0, values = (var_2547_cast_fp16, var_2436_cast_fp16))[name = tensor("op_2753_cast_fp16")]; tensor var_2754_to_fp16 = const()[name = tensor("op_2754_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_267_cast_fp16 = mul(x = var_2753_cast_fp16, y = var_2754_to_fp16)[name = tensor("aw_chunk_267_cast_fp16")]; tensor var_2757_equation_0 = const()[name = tensor("op_2757_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2757_cast_fp16 = einsum(equation = var_2757_equation_0, values = (var_2547_cast_fp16, var_2443_cast_fp16))[name = tensor("op_2757_cast_fp16")]; tensor var_2758_to_fp16 = const()[name = tensor("op_2758_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_269_cast_fp16 = mul(x = var_2757_cast_fp16, y = var_2758_to_fp16)[name = tensor("aw_chunk_269_cast_fp16")]; tensor var_2761_equation_0 = const()[name = tensor("op_2761_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2761_cast_fp16 = einsum(equation = var_2761_equation_0, values = (var_2547_cast_fp16, var_2450_cast_fp16))[name = tensor("op_2761_cast_fp16")]; tensor var_2762_to_fp16 = const()[name = tensor("op_2762_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_271_cast_fp16 = mul(x = var_2761_cast_fp16, y = var_2762_to_fp16)[name = tensor("aw_chunk_271_cast_fp16")]; tensor var_2765_equation_0 = const()[name = tensor("op_2765_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2765_cast_fp16 = einsum(equation = var_2765_equation_0, values = (var_2551_cast_fp16, var_2457_cast_fp16))[name = tensor("op_2765_cast_fp16")]; tensor var_2766_to_fp16 = const()[name = tensor("op_2766_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_273_cast_fp16 = mul(x = var_2765_cast_fp16, y = var_2766_to_fp16)[name = tensor("aw_chunk_273_cast_fp16")]; tensor var_2769_equation_0 = const()[name = tensor("op_2769_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2769_cast_fp16 = einsum(equation = var_2769_equation_0, values = (var_2551_cast_fp16, var_2464_cast_fp16))[name = tensor("op_2769_cast_fp16")]; tensor var_2770_to_fp16 = const()[name = tensor("op_2770_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_275_cast_fp16 = mul(x = var_2769_cast_fp16, y = var_2770_to_fp16)[name = tensor("aw_chunk_275_cast_fp16")]; tensor var_2773_equation_0 = const()[name = tensor("op_2773_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2773_cast_fp16 = einsum(equation = var_2773_equation_0, values = (var_2551_cast_fp16, var_2471_cast_fp16))[name = tensor("op_2773_cast_fp16")]; tensor var_2774_to_fp16 = const()[name = tensor("op_2774_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_277_cast_fp16 = mul(x = var_2773_cast_fp16, y = var_2774_to_fp16)[name = tensor("aw_chunk_277_cast_fp16")]; tensor var_2777_equation_0 = const()[name = tensor("op_2777_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2777_cast_fp16 = einsum(equation = var_2777_equation_0, values = (var_2551_cast_fp16, var_2478_cast_fp16))[name = tensor("op_2777_cast_fp16")]; tensor var_2778_to_fp16 = const()[name = tensor("op_2778_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_279_cast_fp16 = mul(x = var_2777_cast_fp16, y = var_2778_to_fp16)[name = tensor("aw_chunk_279_cast_fp16")]; tensor var_2781_equation_0 = const()[name = tensor("op_2781_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2781_cast_fp16 = einsum(equation = var_2781_equation_0, values = (var_2555_cast_fp16, var_2485_cast_fp16))[name = tensor("op_2781_cast_fp16")]; tensor var_2782_to_fp16 = const()[name = tensor("op_2782_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_281_cast_fp16 = mul(x = var_2781_cast_fp16, y = var_2782_to_fp16)[name = tensor("aw_chunk_281_cast_fp16")]; tensor var_2785_equation_0 = const()[name = tensor("op_2785_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2785_cast_fp16 = einsum(equation = var_2785_equation_0, values = (var_2555_cast_fp16, var_2492_cast_fp16))[name = tensor("op_2785_cast_fp16")]; tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_283_cast_fp16 = mul(x = var_2785_cast_fp16, y = var_2786_to_fp16)[name = tensor("aw_chunk_283_cast_fp16")]; tensor var_2789_equation_0 = const()[name = tensor("op_2789_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2789_cast_fp16 = einsum(equation = var_2789_equation_0, values = (var_2555_cast_fp16, var_2499_cast_fp16))[name = tensor("op_2789_cast_fp16")]; tensor var_2790_to_fp16 = const()[name = tensor("op_2790_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_285_cast_fp16 = mul(x = var_2789_cast_fp16, y = var_2790_to_fp16)[name = tensor("aw_chunk_285_cast_fp16")]; tensor var_2793_equation_0 = const()[name = tensor("op_2793_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_2793_cast_fp16 = einsum(equation = var_2793_equation_0, values = (var_2555_cast_fp16, var_2506_cast_fp16))[name = tensor("op_2793_cast_fp16")]; tensor var_2794_to_fp16 = const()[name = tensor("op_2794_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_287_cast_fp16 = mul(x = var_2793_cast_fp16, y = var_2794_to_fp16)[name = tensor("aw_chunk_287_cast_fp16")]; tensor var_2796_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_193_cast_fp16)[name = tensor("op_2796_cast_fp16")]; tensor var_2797_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_195_cast_fp16)[name = tensor("op_2797_cast_fp16")]; tensor var_2798_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_197_cast_fp16)[name = tensor("op_2798_cast_fp16")]; tensor var_2799_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_199_cast_fp16)[name = tensor("op_2799_cast_fp16")]; tensor var_2800_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_201_cast_fp16)[name = tensor("op_2800_cast_fp16")]; tensor var_2801_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_203_cast_fp16)[name = tensor("op_2801_cast_fp16")]; tensor var_2802_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_205_cast_fp16)[name = tensor("op_2802_cast_fp16")]; tensor var_2803_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_207_cast_fp16)[name = tensor("op_2803_cast_fp16")]; tensor var_2804_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_209_cast_fp16)[name = tensor("op_2804_cast_fp16")]; tensor var_2805_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_211_cast_fp16)[name = tensor("op_2805_cast_fp16")]; tensor var_2806_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_213_cast_fp16)[name = tensor("op_2806_cast_fp16")]; tensor var_2807_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_215_cast_fp16)[name = tensor("op_2807_cast_fp16")]; tensor var_2808_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_217_cast_fp16)[name = tensor("op_2808_cast_fp16")]; tensor var_2809_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_219_cast_fp16)[name = tensor("op_2809_cast_fp16")]; tensor var_2810_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_221_cast_fp16)[name = tensor("op_2810_cast_fp16")]; tensor var_2811_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_223_cast_fp16)[name = tensor("op_2811_cast_fp16")]; tensor var_2812_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_225_cast_fp16)[name = tensor("op_2812_cast_fp16")]; tensor var_2813_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_227_cast_fp16)[name = tensor("op_2813_cast_fp16")]; tensor var_2814_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_229_cast_fp16)[name = tensor("op_2814_cast_fp16")]; tensor var_2815_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_231_cast_fp16)[name = tensor("op_2815_cast_fp16")]; tensor var_2816_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_233_cast_fp16)[name = tensor("op_2816_cast_fp16")]; tensor var_2817_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_235_cast_fp16)[name = tensor("op_2817_cast_fp16")]; tensor var_2818_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_237_cast_fp16)[name = tensor("op_2818_cast_fp16")]; tensor var_2819_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_239_cast_fp16)[name = tensor("op_2819_cast_fp16")]; tensor var_2820_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_241_cast_fp16)[name = tensor("op_2820_cast_fp16")]; tensor var_2821_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_243_cast_fp16)[name = tensor("op_2821_cast_fp16")]; tensor var_2822_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_245_cast_fp16)[name = tensor("op_2822_cast_fp16")]; tensor var_2823_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_247_cast_fp16)[name = tensor("op_2823_cast_fp16")]; tensor var_2824_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_249_cast_fp16)[name = tensor("op_2824_cast_fp16")]; tensor var_2825_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_251_cast_fp16)[name = tensor("op_2825_cast_fp16")]; tensor var_2826_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_253_cast_fp16)[name = tensor("op_2826_cast_fp16")]; tensor var_2827_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_255_cast_fp16)[name = tensor("op_2827_cast_fp16")]; tensor var_2828_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_257_cast_fp16)[name = tensor("op_2828_cast_fp16")]; tensor var_2829_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_259_cast_fp16)[name = tensor("op_2829_cast_fp16")]; tensor var_2830_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_261_cast_fp16)[name = tensor("op_2830_cast_fp16")]; tensor var_2831_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_263_cast_fp16)[name = tensor("op_2831_cast_fp16")]; tensor var_2832_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_265_cast_fp16)[name = tensor("op_2832_cast_fp16")]; tensor var_2833_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_267_cast_fp16)[name = tensor("op_2833_cast_fp16")]; tensor var_2834_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_269_cast_fp16)[name = tensor("op_2834_cast_fp16")]; tensor var_2835_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_271_cast_fp16)[name = tensor("op_2835_cast_fp16")]; tensor var_2836_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_273_cast_fp16)[name = tensor("op_2836_cast_fp16")]; tensor var_2837_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_275_cast_fp16)[name = tensor("op_2837_cast_fp16")]; tensor var_2838_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_277_cast_fp16)[name = tensor("op_2838_cast_fp16")]; tensor var_2839_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_279_cast_fp16)[name = tensor("op_2839_cast_fp16")]; tensor var_2840_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_281_cast_fp16)[name = tensor("op_2840_cast_fp16")]; tensor var_2841_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_283_cast_fp16)[name = tensor("op_2841_cast_fp16")]; tensor var_2842_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_285_cast_fp16)[name = tensor("op_2842_cast_fp16")]; tensor var_2843_cast_fp16 = softmax(axis = var_2069, x = aw_chunk_287_cast_fp16)[name = tensor("op_2843_cast_fp16")]; tensor var_2845_equation_0 = const()[name = tensor("op_2845_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2845_cast_fp16 = einsum(equation = var_2845_equation_0, values = (var_2557_cast_fp16, var_2796_cast_fp16))[name = tensor("op_2845_cast_fp16")]; tensor var_2847_equation_0 = const()[name = tensor("op_2847_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2847_cast_fp16 = einsum(equation = var_2847_equation_0, values = (var_2557_cast_fp16, var_2797_cast_fp16))[name = tensor("op_2847_cast_fp16")]; tensor var_2849_equation_0 = const()[name = tensor("op_2849_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2849_cast_fp16 = einsum(equation = var_2849_equation_0, values = (var_2557_cast_fp16, var_2798_cast_fp16))[name = tensor("op_2849_cast_fp16")]; tensor var_2851_equation_0 = const()[name = tensor("op_2851_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2851_cast_fp16 = einsum(equation = var_2851_equation_0, values = (var_2557_cast_fp16, var_2799_cast_fp16))[name = tensor("op_2851_cast_fp16")]; tensor var_2853_equation_0 = const()[name = tensor("op_2853_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2853_cast_fp16 = einsum(equation = var_2853_equation_0, values = (var_2561_cast_fp16, var_2800_cast_fp16))[name = tensor("op_2853_cast_fp16")]; tensor var_2855_equation_0 = const()[name = tensor("op_2855_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2855_cast_fp16 = einsum(equation = var_2855_equation_0, values = (var_2561_cast_fp16, var_2801_cast_fp16))[name = tensor("op_2855_cast_fp16")]; tensor var_2857_equation_0 = const()[name = tensor("op_2857_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2857_cast_fp16 = einsum(equation = var_2857_equation_0, values = (var_2561_cast_fp16, var_2802_cast_fp16))[name = tensor("op_2857_cast_fp16")]; tensor var_2859_equation_0 = const()[name = tensor("op_2859_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2859_cast_fp16 = einsum(equation = var_2859_equation_0, values = (var_2561_cast_fp16, var_2803_cast_fp16))[name = tensor("op_2859_cast_fp16")]; tensor var_2861_equation_0 = const()[name = tensor("op_2861_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2861_cast_fp16 = einsum(equation = var_2861_equation_0, values = (var_2565_cast_fp16, var_2804_cast_fp16))[name = tensor("op_2861_cast_fp16")]; tensor var_2863_equation_0 = const()[name = tensor("op_2863_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2863_cast_fp16 = einsum(equation = var_2863_equation_0, values = (var_2565_cast_fp16, var_2805_cast_fp16))[name = tensor("op_2863_cast_fp16")]; tensor var_2865_equation_0 = const()[name = tensor("op_2865_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2865_cast_fp16 = einsum(equation = var_2865_equation_0, values = (var_2565_cast_fp16, var_2806_cast_fp16))[name = tensor("op_2865_cast_fp16")]; tensor var_2867_equation_0 = const()[name = tensor("op_2867_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2867_cast_fp16 = einsum(equation = var_2867_equation_0, values = (var_2565_cast_fp16, var_2807_cast_fp16))[name = tensor("op_2867_cast_fp16")]; tensor var_2869_equation_0 = const()[name = tensor("op_2869_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2869_cast_fp16 = einsum(equation = var_2869_equation_0, values = (var_2569_cast_fp16, var_2808_cast_fp16))[name = tensor("op_2869_cast_fp16")]; tensor var_2871_equation_0 = const()[name = tensor("op_2871_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2871_cast_fp16 = einsum(equation = var_2871_equation_0, values = (var_2569_cast_fp16, var_2809_cast_fp16))[name = tensor("op_2871_cast_fp16")]; tensor var_2873_equation_0 = const()[name = tensor("op_2873_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2873_cast_fp16 = einsum(equation = var_2873_equation_0, values = (var_2569_cast_fp16, var_2810_cast_fp16))[name = tensor("op_2873_cast_fp16")]; tensor var_2875_equation_0 = const()[name = tensor("op_2875_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2875_cast_fp16 = einsum(equation = var_2875_equation_0, values = (var_2569_cast_fp16, var_2811_cast_fp16))[name = tensor("op_2875_cast_fp16")]; tensor var_2877_equation_0 = const()[name = tensor("op_2877_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2877_cast_fp16 = einsum(equation = var_2877_equation_0, values = (var_2573_cast_fp16, var_2812_cast_fp16))[name = tensor("op_2877_cast_fp16")]; tensor var_2879_equation_0 = const()[name = tensor("op_2879_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2879_cast_fp16 = einsum(equation = var_2879_equation_0, values = (var_2573_cast_fp16, var_2813_cast_fp16))[name = tensor("op_2879_cast_fp16")]; tensor var_2881_equation_0 = const()[name = tensor("op_2881_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2881_cast_fp16 = einsum(equation = var_2881_equation_0, values = (var_2573_cast_fp16, var_2814_cast_fp16))[name = tensor("op_2881_cast_fp16")]; tensor var_2883_equation_0 = const()[name = tensor("op_2883_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2883_cast_fp16 = einsum(equation = var_2883_equation_0, values = (var_2573_cast_fp16, var_2815_cast_fp16))[name = tensor("op_2883_cast_fp16")]; tensor var_2885_equation_0 = const()[name = tensor("op_2885_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2885_cast_fp16 = einsum(equation = var_2885_equation_0, values = (var_2577_cast_fp16, var_2816_cast_fp16))[name = tensor("op_2885_cast_fp16")]; tensor var_2887_equation_0 = const()[name = tensor("op_2887_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2887_cast_fp16 = einsum(equation = var_2887_equation_0, values = (var_2577_cast_fp16, var_2817_cast_fp16))[name = tensor("op_2887_cast_fp16")]; tensor var_2889_equation_0 = const()[name = tensor("op_2889_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2889_cast_fp16 = einsum(equation = var_2889_equation_0, values = (var_2577_cast_fp16, var_2818_cast_fp16))[name = tensor("op_2889_cast_fp16")]; tensor var_2891_equation_0 = const()[name = tensor("op_2891_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2891_cast_fp16 = einsum(equation = var_2891_equation_0, values = (var_2577_cast_fp16, var_2819_cast_fp16))[name = tensor("op_2891_cast_fp16")]; tensor var_2893_equation_0 = const()[name = tensor("op_2893_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2893_cast_fp16 = einsum(equation = var_2893_equation_0, values = (var_2581_cast_fp16, var_2820_cast_fp16))[name = tensor("op_2893_cast_fp16")]; tensor var_2895_equation_0 = const()[name = tensor("op_2895_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2895_cast_fp16 = einsum(equation = var_2895_equation_0, values = (var_2581_cast_fp16, var_2821_cast_fp16))[name = tensor("op_2895_cast_fp16")]; tensor var_2897_equation_0 = const()[name = tensor("op_2897_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2897_cast_fp16 = einsum(equation = var_2897_equation_0, values = (var_2581_cast_fp16, var_2822_cast_fp16))[name = tensor("op_2897_cast_fp16")]; tensor var_2899_equation_0 = const()[name = tensor("op_2899_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2899_cast_fp16 = einsum(equation = var_2899_equation_0, values = (var_2581_cast_fp16, var_2823_cast_fp16))[name = tensor("op_2899_cast_fp16")]; tensor var_2901_equation_0 = const()[name = tensor("op_2901_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2901_cast_fp16 = einsum(equation = var_2901_equation_0, values = (var_2585_cast_fp16, var_2824_cast_fp16))[name = tensor("op_2901_cast_fp16")]; tensor var_2903_equation_0 = const()[name = tensor("op_2903_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2903_cast_fp16 = einsum(equation = var_2903_equation_0, values = (var_2585_cast_fp16, var_2825_cast_fp16))[name = tensor("op_2903_cast_fp16")]; tensor var_2905_equation_0 = const()[name = tensor("op_2905_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2905_cast_fp16 = einsum(equation = var_2905_equation_0, values = (var_2585_cast_fp16, var_2826_cast_fp16))[name = tensor("op_2905_cast_fp16")]; tensor var_2907_equation_0 = const()[name = tensor("op_2907_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2907_cast_fp16 = einsum(equation = var_2907_equation_0, values = (var_2585_cast_fp16, var_2827_cast_fp16))[name = tensor("op_2907_cast_fp16")]; tensor var_2909_equation_0 = const()[name = tensor("op_2909_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2909_cast_fp16 = einsum(equation = var_2909_equation_0, values = (var_2589_cast_fp16, var_2828_cast_fp16))[name = tensor("op_2909_cast_fp16")]; tensor var_2911_equation_0 = const()[name = tensor("op_2911_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2911_cast_fp16 = einsum(equation = var_2911_equation_0, values = (var_2589_cast_fp16, var_2829_cast_fp16))[name = tensor("op_2911_cast_fp16")]; tensor var_2913_equation_0 = const()[name = tensor("op_2913_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2913_cast_fp16 = einsum(equation = var_2913_equation_0, values = (var_2589_cast_fp16, var_2830_cast_fp16))[name = tensor("op_2913_cast_fp16")]; tensor var_2915_equation_0 = const()[name = tensor("op_2915_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2915_cast_fp16 = einsum(equation = var_2915_equation_0, values = (var_2589_cast_fp16, var_2831_cast_fp16))[name = tensor("op_2915_cast_fp16")]; tensor var_2917_equation_0 = const()[name = tensor("op_2917_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2917_cast_fp16 = einsum(equation = var_2917_equation_0, values = (var_2593_cast_fp16, var_2832_cast_fp16))[name = tensor("op_2917_cast_fp16")]; tensor var_2919_equation_0 = const()[name = tensor("op_2919_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2919_cast_fp16 = einsum(equation = var_2919_equation_0, values = (var_2593_cast_fp16, var_2833_cast_fp16))[name = tensor("op_2919_cast_fp16")]; tensor var_2921_equation_0 = const()[name = tensor("op_2921_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2921_cast_fp16 = einsum(equation = var_2921_equation_0, values = (var_2593_cast_fp16, var_2834_cast_fp16))[name = tensor("op_2921_cast_fp16")]; tensor var_2923_equation_0 = const()[name = tensor("op_2923_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2923_cast_fp16 = einsum(equation = var_2923_equation_0, values = (var_2593_cast_fp16, var_2835_cast_fp16))[name = tensor("op_2923_cast_fp16")]; tensor var_2925_equation_0 = const()[name = tensor("op_2925_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2925_cast_fp16 = einsum(equation = var_2925_equation_0, values = (var_2597_cast_fp16, var_2836_cast_fp16))[name = tensor("op_2925_cast_fp16")]; tensor var_2927_equation_0 = const()[name = tensor("op_2927_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2927_cast_fp16 = einsum(equation = var_2927_equation_0, values = (var_2597_cast_fp16, var_2837_cast_fp16))[name = tensor("op_2927_cast_fp16")]; tensor var_2929_equation_0 = const()[name = tensor("op_2929_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2929_cast_fp16 = einsum(equation = var_2929_equation_0, values = (var_2597_cast_fp16, var_2838_cast_fp16))[name = tensor("op_2929_cast_fp16")]; tensor var_2931_equation_0 = const()[name = tensor("op_2931_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2931_cast_fp16 = einsum(equation = var_2931_equation_0, values = (var_2597_cast_fp16, var_2839_cast_fp16))[name = tensor("op_2931_cast_fp16")]; tensor var_2933_equation_0 = const()[name = tensor("op_2933_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2933_cast_fp16 = einsum(equation = var_2933_equation_0, values = (var_2601_cast_fp16, var_2840_cast_fp16))[name = tensor("op_2933_cast_fp16")]; tensor var_2935_equation_0 = const()[name = tensor("op_2935_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2935_cast_fp16 = einsum(equation = var_2935_equation_0, values = (var_2601_cast_fp16, var_2841_cast_fp16))[name = tensor("op_2935_cast_fp16")]; tensor var_2937_equation_0 = const()[name = tensor("op_2937_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2937_cast_fp16 = einsum(equation = var_2937_equation_0, values = (var_2601_cast_fp16, var_2842_cast_fp16))[name = tensor("op_2937_cast_fp16")]; tensor var_2939_equation_0 = const()[name = tensor("op_2939_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_2939_cast_fp16 = einsum(equation = var_2939_equation_0, values = (var_2601_cast_fp16, var_2843_cast_fp16))[name = tensor("op_2939_cast_fp16")]; tensor var_2941_interleave_0 = const()[name = tensor("op_2941_interleave_0"), val = tensor(false)]; tensor var_2941_cast_fp16 = concat(axis = var_2052, interleave = var_2941_interleave_0, values = (var_2845_cast_fp16, var_2847_cast_fp16, var_2849_cast_fp16, var_2851_cast_fp16))[name = tensor("op_2941_cast_fp16")]; tensor var_2943_interleave_0 = const()[name = tensor("op_2943_interleave_0"), val = tensor(false)]; tensor var_2943_cast_fp16 = concat(axis = var_2052, interleave = var_2943_interleave_0, values = (var_2853_cast_fp16, var_2855_cast_fp16, var_2857_cast_fp16, var_2859_cast_fp16))[name = tensor("op_2943_cast_fp16")]; tensor var_2945_interleave_0 = const()[name = tensor("op_2945_interleave_0"), val = tensor(false)]; tensor var_2945_cast_fp16 = concat(axis = var_2052, interleave = var_2945_interleave_0, values = (var_2861_cast_fp16, var_2863_cast_fp16, var_2865_cast_fp16, var_2867_cast_fp16))[name = tensor("op_2945_cast_fp16")]; tensor var_2947_interleave_0 = const()[name = tensor("op_2947_interleave_0"), val = tensor(false)]; tensor var_2947_cast_fp16 = concat(axis = var_2052, interleave = var_2947_interleave_0, values = (var_2869_cast_fp16, var_2871_cast_fp16, var_2873_cast_fp16, var_2875_cast_fp16))[name = tensor("op_2947_cast_fp16")]; tensor var_2949_interleave_0 = const()[name = tensor("op_2949_interleave_0"), val = tensor(false)]; tensor var_2949_cast_fp16 = concat(axis = var_2052, interleave = var_2949_interleave_0, values = (var_2877_cast_fp16, var_2879_cast_fp16, var_2881_cast_fp16, var_2883_cast_fp16))[name = tensor("op_2949_cast_fp16")]; tensor var_2951_interleave_0 = const()[name = tensor("op_2951_interleave_0"), val = tensor(false)]; tensor var_2951_cast_fp16 = concat(axis = var_2052, interleave = var_2951_interleave_0, values = (var_2885_cast_fp16, var_2887_cast_fp16, var_2889_cast_fp16, var_2891_cast_fp16))[name = tensor("op_2951_cast_fp16")]; tensor var_2953_interleave_0 = const()[name = tensor("op_2953_interleave_0"), val = tensor(false)]; tensor var_2953_cast_fp16 = concat(axis = var_2052, interleave = var_2953_interleave_0, values = (var_2893_cast_fp16, var_2895_cast_fp16, var_2897_cast_fp16, var_2899_cast_fp16))[name = tensor("op_2953_cast_fp16")]; tensor var_2955_interleave_0 = const()[name = tensor("op_2955_interleave_0"), val = tensor(false)]; tensor var_2955_cast_fp16 = concat(axis = var_2052, interleave = var_2955_interleave_0, values = (var_2901_cast_fp16, var_2903_cast_fp16, var_2905_cast_fp16, var_2907_cast_fp16))[name = tensor("op_2955_cast_fp16")]; tensor var_2957_interleave_0 = const()[name = tensor("op_2957_interleave_0"), val = tensor(false)]; tensor var_2957_cast_fp16 = concat(axis = var_2052, interleave = var_2957_interleave_0, values = (var_2909_cast_fp16, var_2911_cast_fp16, var_2913_cast_fp16, var_2915_cast_fp16))[name = tensor("op_2957_cast_fp16")]; tensor var_2959_interleave_0 = const()[name = tensor("op_2959_interleave_0"), val = tensor(false)]; tensor var_2959_cast_fp16 = concat(axis = var_2052, interleave = var_2959_interleave_0, values = (var_2917_cast_fp16, var_2919_cast_fp16, var_2921_cast_fp16, var_2923_cast_fp16))[name = tensor("op_2959_cast_fp16")]; tensor var_2961_interleave_0 = const()[name = tensor("op_2961_interleave_0"), val = tensor(false)]; tensor var_2961_cast_fp16 = concat(axis = var_2052, interleave = var_2961_interleave_0, values = (var_2925_cast_fp16, var_2927_cast_fp16, var_2929_cast_fp16, var_2931_cast_fp16))[name = tensor("op_2961_cast_fp16")]; tensor var_2963_interleave_0 = const()[name = tensor("op_2963_interleave_0"), val = tensor(false)]; tensor var_2963_cast_fp16 = concat(axis = var_2052, interleave = var_2963_interleave_0, values = (var_2933_cast_fp16, var_2935_cast_fp16, var_2937_cast_fp16, var_2939_cast_fp16))[name = tensor("op_2963_cast_fp16")]; tensor input_17_interleave_0 = const()[name = tensor("input_17_interleave_0"), val = tensor(false)]; tensor input_17_cast_fp16 = concat(axis = var_2069, interleave = input_17_interleave_0, values = (var_2941_cast_fp16, var_2943_cast_fp16, var_2945_cast_fp16, var_2947_cast_fp16, var_2949_cast_fp16, var_2951_cast_fp16, var_2953_cast_fp16, var_2955_cast_fp16, var_2957_cast_fp16, var_2959_cast_fp16, var_2961_cast_fp16, var_2963_cast_fp16))[name = tensor("input_17_cast_fp16")]; tensor var_2968 = const()[name = tensor("op_2968"), val = tensor([1, 1])]; tensor var_2970 = const()[name = tensor("op_2970"), val = tensor([1, 1])]; tensor obj_11_pad_type_0 = const()[name = tensor("obj_11_pad_type_0"), val = tensor("custom")]; tensor obj_11_pad_0 = const()[name = tensor("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38114112)))]; tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39293824)))]; tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = var_2970, groups = var_2069, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = var_2968, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_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_2976 = const()[name = tensor("op_2976"), val = tensor([1])]; tensor channels_mean_11_cast_fp16 = reduce_mean(axes = var_2976, keep_dims = var_2070, 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_2980 = const()[name = tensor("op_2980"), val = tensor([1])]; tensor var_2981_cast_fp16 = reduce_mean(axes = var_2980, keep_dims = var_2070, x = zero_mean_sq_11_cast_fp16)[name = tensor("op_2981_cast_fp16")]; tensor var_2982_to_fp16 = const()[name = tensor("op_2982_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_2983_cast_fp16 = add(x = var_2981_cast_fp16, y = var_2982_to_fp16)[name = tensor("op_2983_cast_fp16")]; tensor denom_11_epsilon_0_to_fp16 = const()[name = tensor("denom_11_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_11_cast_fp16 = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_2983_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_19_gamma_0_to_fp16 = const()[name = tensor("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39295424)))]; tensor input_19_beta_0_to_fp16 = const()[name = tensor("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39297024)))]; tensor input_19_epsilon_0_to_fp16 = const()[name = tensor("input_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_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_19_cast_fp16")]; tensor var_2994 = const()[name = tensor("op_2994"), val = tensor([1, 1])]; tensor var_2996 = const()[name = tensor("op_2996"), 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_2_fc1_weight_to_fp16 = const()[name = tensor("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39298624)))]; tensor layers_2_fc1_bias_to_fp16 = const()[name = tensor("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44017280)))]; tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = var_2996, groups = var_2069, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = var_2994, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; tensor input_23_mode_0 = const()[name = tensor("input_23_mode_0"), val = tensor("EXACT")]; tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = tensor("input_23_cast_fp16")]; tensor var_3002 = const()[name = tensor("op_3002"), val = tensor([1, 1])]; tensor var_3004 = const()[name = tensor("op_3004"), val = tensor([1, 1])]; tensor hidden_states_9_pad_type_0 = const()[name = tensor("hidden_states_9_pad_type_0"), val = tensor("custom")]; tensor hidden_states_9_pad_0 = const()[name = tensor("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_2_fc2_weight_to_fp16 = const()[name = tensor("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44023488)))]; tensor layers_2_fc2_bias_to_fp16 = const()[name = tensor("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48742144)))]; tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = var_3004, groups = var_2069, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = var_3002, weight = layers_2_fc2_weight_to_fp16, x = input_23_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_3011 = const()[name = tensor("op_3011"), val = tensor(3)]; tensor var_3028 = const()[name = tensor("op_3028"), val = tensor(1)]; tensor var_3029 = const()[name = tensor("op_3029"), val = tensor(true)]; tensor var_3039 = const()[name = tensor("op_3039"), val = tensor([1])]; tensor channels_mean_13_cast_fp16 = reduce_mean(axes = var_3039, keep_dims = var_3029, 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_3043 = const()[name = tensor("op_3043"), val = tensor([1])]; tensor var_3044_cast_fp16 = reduce_mean(axes = var_3043, keep_dims = var_3029, x = zero_mean_sq_13_cast_fp16)[name = tensor("op_3044_cast_fp16")]; tensor var_3045_to_fp16 = const()[name = tensor("op_3045_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3046_cast_fp16 = add(x = var_3044_cast_fp16, y = var_3045_to_fp16)[name = tensor("op_3046_cast_fp16")]; tensor denom_13_epsilon_0_to_fp16 = const()[name = tensor("denom_13_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_13_cast_fp16 = rsqrt(epsilon = denom_13_epsilon_0_to_fp16, x = var_3046_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(48743744)))]; 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(48745344)))]; 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_3061 = const()[name = tensor("op_3061"), val = tensor([1, 1])]; tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1])]; tensor query_7_pad_type_0 = const()[name = tensor("query_7_pad_type_0"), val = tensor("custom")]; tensor query_7_pad_0 = const()[name = tensor("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48746944)))]; tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49926656)))]; tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = var_3063, groups = var_3028, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = var_3061, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("query_7_cast_fp16")]; tensor var_3067 = const()[name = tensor("op_3067"), val = tensor([1, 1])]; tensor var_3069 = const()[name = tensor("op_3069"), val = tensor([1, 1])]; tensor key_7_pad_type_0 = const()[name = tensor("key_7_pad_type_0"), val = tensor("custom")]; tensor key_7_pad_0 = const()[name = tensor("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49928256)))]; tensor key_7_cast_fp16 = conv(dilations = var_3069, groups = var_3028, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = var_3067, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("key_7_cast_fp16")]; tensor var_3074 = const()[name = tensor("op_3074"), val = tensor([1, 1])]; tensor var_3076 = const()[name = tensor("op_3076"), val = tensor([1, 1])]; tensor value_7_pad_type_0 = const()[name = tensor("value_7_pad_type_0"), val = tensor("custom")]; tensor value_7_pad_0 = const()[name = tensor("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51107968)))]; tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52287680)))]; tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = var_3076, groups = var_3028, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = var_3074, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor("value_7_cast_fp16")]; tensor var_3083_begin_0 = const()[name = tensor("op_3083_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3083_end_0 = const()[name = tensor("op_3083_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3083_end_mask_0 = const()[name = tensor("op_3083_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3083_cast_fp16 = slice_by_index(begin = var_3083_begin_0, end = var_3083_end_0, end_mask = var_3083_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3083_cast_fp16")]; tensor var_3087_begin_0 = const()[name = tensor("op_3087_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_3087_end_0 = const()[name = tensor("op_3087_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_3087_end_mask_0 = const()[name = tensor("op_3087_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3087_cast_fp16 = slice_by_index(begin = var_3087_begin_0, end = var_3087_end_0, end_mask = var_3087_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3087_cast_fp16")]; tensor var_3091_begin_0 = const()[name = tensor("op_3091_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_3091_end_0 = const()[name = tensor("op_3091_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_3091_end_mask_0 = const()[name = tensor("op_3091_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3091_cast_fp16 = slice_by_index(begin = var_3091_begin_0, end = var_3091_end_0, end_mask = var_3091_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3091_cast_fp16")]; tensor var_3095_begin_0 = const()[name = tensor("op_3095_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_3095_end_0 = const()[name = tensor("op_3095_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_3095_end_mask_0 = const()[name = tensor("op_3095_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3095_cast_fp16 = slice_by_index(begin = var_3095_begin_0, end = var_3095_end_0, end_mask = var_3095_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3095_cast_fp16")]; tensor var_3099_begin_0 = const()[name = tensor("op_3099_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_3099_end_0 = const()[name = tensor("op_3099_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_3099_end_mask_0 = const()[name = tensor("op_3099_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3099_cast_fp16 = slice_by_index(begin = var_3099_begin_0, end = var_3099_end_0, end_mask = var_3099_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3099_cast_fp16")]; tensor var_3103_begin_0 = const()[name = tensor("op_3103_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3103_end_0 = const()[name = tensor("op_3103_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_3103_end_mask_0 = const()[name = tensor("op_3103_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3103_cast_fp16 = slice_by_index(begin = var_3103_begin_0, end = var_3103_end_0, end_mask = var_3103_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3103_cast_fp16")]; tensor var_3107_begin_0 = const()[name = tensor("op_3107_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_3107_end_0 = const()[name = tensor("op_3107_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_3107_end_mask_0 = const()[name = tensor("op_3107_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3107_cast_fp16")]; tensor var_3111_begin_0 = const()[name = tensor("op_3111_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_3111_end_0 = const()[name = tensor("op_3111_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_3111_end_mask_0 = const()[name = tensor("op_3111_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3111_cast_fp16 = slice_by_index(begin = var_3111_begin_0, end = var_3111_end_0, end_mask = var_3111_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3111_cast_fp16")]; tensor var_3115_begin_0 = const()[name = tensor("op_3115_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_3115_end_0 = const()[name = tensor("op_3115_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_3115_end_mask_0 = const()[name = tensor("op_3115_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3115_cast_fp16 = slice_by_index(begin = var_3115_begin_0, end = var_3115_end_0, end_mask = var_3115_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3115_cast_fp16")]; tensor var_3119_begin_0 = const()[name = tensor("op_3119_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_3119_end_0 = const()[name = tensor("op_3119_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_3119_end_mask_0 = const()[name = tensor("op_3119_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3119_cast_fp16 = slice_by_index(begin = var_3119_begin_0, end = var_3119_end_0, end_mask = var_3119_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3119_cast_fp16")]; tensor var_3123_begin_0 = const()[name = tensor("op_3123_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3123_end_0 = const()[name = tensor("op_3123_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_3123_end_mask_0 = const()[name = tensor("op_3123_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3123_cast_fp16 = slice_by_index(begin = var_3123_begin_0, end = var_3123_end_0, end_mask = var_3123_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3123_cast_fp16")]; tensor var_3127_begin_0 = const()[name = tensor("op_3127_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_3127_end_0 = const()[name = tensor("op_3127_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_3127_end_mask_0 = const()[name = tensor("op_3127_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3127_cast_fp16 = slice_by_index(begin = var_3127_begin_0, end = var_3127_end_0, end_mask = var_3127_end_mask_0, x = query_7_cast_fp16)[name = tensor("op_3127_cast_fp16")]; tensor var_3136_begin_0 = const()[name = tensor("op_3136_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3136_end_0 = const()[name = tensor("op_3136_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3136_end_mask_0 = const()[name = tensor("op_3136_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3136_cast_fp16 = slice_by_index(begin = var_3136_begin_0, end = var_3136_end_0, end_mask = var_3136_end_mask_0, x = var_3083_cast_fp16)[name = tensor("op_3136_cast_fp16")]; tensor var_3143_begin_0 = const()[name = tensor("op_3143_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3143_end_0 = const()[name = tensor("op_3143_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3143_end_mask_0 = const()[name = tensor("op_3143_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3143_cast_fp16 = slice_by_index(begin = var_3143_begin_0, end = var_3143_end_0, end_mask = var_3143_end_mask_0, x = var_3083_cast_fp16)[name = tensor("op_3143_cast_fp16")]; tensor var_3150_begin_0 = const()[name = tensor("op_3150_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3150_end_0 = const()[name = tensor("op_3150_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3150_end_mask_0 = const()[name = tensor("op_3150_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3150_cast_fp16 = slice_by_index(begin = var_3150_begin_0, end = var_3150_end_0, end_mask = var_3150_end_mask_0, x = var_3083_cast_fp16)[name = tensor("op_3150_cast_fp16")]; tensor var_3157_begin_0 = const()[name = tensor("op_3157_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3157_end_0 = const()[name = tensor("op_3157_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3157_end_mask_0 = const()[name = tensor("op_3157_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3157_cast_fp16 = slice_by_index(begin = var_3157_begin_0, end = var_3157_end_0, end_mask = var_3157_end_mask_0, x = var_3083_cast_fp16)[name = tensor("op_3157_cast_fp16")]; tensor var_3164_begin_0 = const()[name = tensor("op_3164_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3164_end_0 = const()[name = tensor("op_3164_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3164_end_mask_0 = const()[name = tensor("op_3164_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3164_cast_fp16 = slice_by_index(begin = var_3164_begin_0, end = var_3164_end_0, end_mask = var_3164_end_mask_0, x = var_3087_cast_fp16)[name = tensor("op_3164_cast_fp16")]; tensor var_3171_begin_0 = const()[name = tensor("op_3171_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3171_end_0 = const()[name = tensor("op_3171_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3171_end_mask_0 = const()[name = tensor("op_3171_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3171_cast_fp16 = slice_by_index(begin = var_3171_begin_0, end = var_3171_end_0, end_mask = var_3171_end_mask_0, x = var_3087_cast_fp16)[name = tensor("op_3171_cast_fp16")]; tensor var_3178_begin_0 = const()[name = tensor("op_3178_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3178_end_0 = const()[name = tensor("op_3178_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3178_end_mask_0 = const()[name = tensor("op_3178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3178_cast_fp16 = slice_by_index(begin = var_3178_begin_0, end = var_3178_end_0, end_mask = var_3178_end_mask_0, x = var_3087_cast_fp16)[name = tensor("op_3178_cast_fp16")]; tensor var_3185_begin_0 = const()[name = tensor("op_3185_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3185_end_0 = const()[name = tensor("op_3185_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3185_end_mask_0 = const()[name = tensor("op_3185_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3185_cast_fp16 = slice_by_index(begin = var_3185_begin_0, end = var_3185_end_0, end_mask = var_3185_end_mask_0, x = var_3087_cast_fp16)[name = tensor("op_3185_cast_fp16")]; tensor var_3192_begin_0 = const()[name = tensor("op_3192_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3192_end_0 = const()[name = tensor("op_3192_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3192_end_mask_0 = const()[name = tensor("op_3192_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3192_cast_fp16 = slice_by_index(begin = var_3192_begin_0, end = var_3192_end_0, end_mask = var_3192_end_mask_0, x = var_3091_cast_fp16)[name = tensor("op_3192_cast_fp16")]; tensor var_3199_begin_0 = const()[name = tensor("op_3199_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3199_end_0 = const()[name = tensor("op_3199_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3199_end_mask_0 = const()[name = tensor("op_3199_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3199_cast_fp16 = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = var_3091_cast_fp16)[name = tensor("op_3199_cast_fp16")]; tensor var_3206_begin_0 = const()[name = tensor("op_3206_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3206_end_0 = const()[name = tensor("op_3206_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3206_end_mask_0 = const()[name = tensor("op_3206_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3206_cast_fp16 = slice_by_index(begin = var_3206_begin_0, end = var_3206_end_0, end_mask = var_3206_end_mask_0, x = var_3091_cast_fp16)[name = tensor("op_3206_cast_fp16")]; tensor var_3213_begin_0 = const()[name = tensor("op_3213_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3213_end_0 = const()[name = tensor("op_3213_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3213_end_mask_0 = const()[name = tensor("op_3213_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3213_cast_fp16 = slice_by_index(begin = var_3213_begin_0, end = var_3213_end_0, end_mask = var_3213_end_mask_0, x = var_3091_cast_fp16)[name = tensor("op_3213_cast_fp16")]; tensor var_3220_begin_0 = const()[name = tensor("op_3220_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3220_end_0 = const()[name = tensor("op_3220_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3220_end_mask_0 = const()[name = tensor("op_3220_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3220_cast_fp16 = slice_by_index(begin = var_3220_begin_0, end = var_3220_end_0, end_mask = var_3220_end_mask_0, x = var_3095_cast_fp16)[name = tensor("op_3220_cast_fp16")]; tensor var_3227_begin_0 = const()[name = tensor("op_3227_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3227_end_0 = const()[name = tensor("op_3227_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3227_end_mask_0 = const()[name = tensor("op_3227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3227_cast_fp16 = slice_by_index(begin = var_3227_begin_0, end = var_3227_end_0, end_mask = var_3227_end_mask_0, x = var_3095_cast_fp16)[name = tensor("op_3227_cast_fp16")]; tensor var_3234_begin_0 = const()[name = tensor("op_3234_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3234_end_0 = const()[name = tensor("op_3234_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3234_end_mask_0 = const()[name = tensor("op_3234_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3234_cast_fp16 = slice_by_index(begin = var_3234_begin_0, end = var_3234_end_0, end_mask = var_3234_end_mask_0, x = var_3095_cast_fp16)[name = tensor("op_3234_cast_fp16")]; tensor var_3241_begin_0 = const()[name = tensor("op_3241_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3241_end_0 = const()[name = tensor("op_3241_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3241_end_mask_0 = const()[name = tensor("op_3241_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3241_cast_fp16 = slice_by_index(begin = var_3241_begin_0, end = var_3241_end_0, end_mask = var_3241_end_mask_0, x = var_3095_cast_fp16)[name = tensor("op_3241_cast_fp16")]; tensor var_3248_begin_0 = const()[name = tensor("op_3248_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3248_end_0 = const()[name = tensor("op_3248_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3248_end_mask_0 = const()[name = tensor("op_3248_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3248_cast_fp16 = slice_by_index(begin = var_3248_begin_0, end = var_3248_end_0, end_mask = var_3248_end_mask_0, x = var_3099_cast_fp16)[name = tensor("op_3248_cast_fp16")]; tensor var_3255_begin_0 = const()[name = tensor("op_3255_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3255_end_0 = const()[name = tensor("op_3255_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3255_end_mask_0 = const()[name = tensor("op_3255_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3255_cast_fp16 = slice_by_index(begin = var_3255_begin_0, end = var_3255_end_0, end_mask = var_3255_end_mask_0, x = var_3099_cast_fp16)[name = tensor("op_3255_cast_fp16")]; tensor var_3262_begin_0 = const()[name = tensor("op_3262_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3262_end_0 = const()[name = tensor("op_3262_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3262_end_mask_0 = const()[name = tensor("op_3262_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3262_cast_fp16 = slice_by_index(begin = var_3262_begin_0, end = var_3262_end_0, end_mask = var_3262_end_mask_0, x = var_3099_cast_fp16)[name = tensor("op_3262_cast_fp16")]; tensor var_3269_begin_0 = const()[name = tensor("op_3269_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3269_end_0 = const()[name = tensor("op_3269_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3269_end_mask_0 = const()[name = tensor("op_3269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3269_cast_fp16 = slice_by_index(begin = var_3269_begin_0, end = var_3269_end_0, end_mask = var_3269_end_mask_0, x = var_3099_cast_fp16)[name = tensor("op_3269_cast_fp16")]; tensor var_3276_begin_0 = const()[name = tensor("op_3276_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3276_end_0 = const()[name = tensor("op_3276_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3276_end_mask_0 = const()[name = tensor("op_3276_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3276_cast_fp16 = slice_by_index(begin = var_3276_begin_0, end = var_3276_end_0, end_mask = var_3276_end_mask_0, x = var_3103_cast_fp16)[name = tensor("op_3276_cast_fp16")]; tensor var_3283_begin_0 = const()[name = tensor("op_3283_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3283_end_0 = const()[name = tensor("op_3283_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3283_end_mask_0 = const()[name = tensor("op_3283_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3283_cast_fp16 = slice_by_index(begin = var_3283_begin_0, end = var_3283_end_0, end_mask = var_3283_end_mask_0, x = var_3103_cast_fp16)[name = tensor("op_3283_cast_fp16")]; tensor var_3290_begin_0 = const()[name = tensor("op_3290_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3290_end_0 = const()[name = tensor("op_3290_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3290_end_mask_0 = const()[name = tensor("op_3290_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3290_cast_fp16 = slice_by_index(begin = var_3290_begin_0, end = var_3290_end_0, end_mask = var_3290_end_mask_0, x = var_3103_cast_fp16)[name = tensor("op_3290_cast_fp16")]; tensor var_3297_begin_0 = const()[name = tensor("op_3297_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3297_end_0 = const()[name = tensor("op_3297_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3297_end_mask_0 = const()[name = tensor("op_3297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3297_cast_fp16 = slice_by_index(begin = var_3297_begin_0, end = var_3297_end_0, end_mask = var_3297_end_mask_0, x = var_3103_cast_fp16)[name = tensor("op_3297_cast_fp16")]; tensor var_3304_begin_0 = const()[name = tensor("op_3304_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3304_end_0 = const()[name = tensor("op_3304_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3304_end_mask_0 = const()[name = tensor("op_3304_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3304_cast_fp16 = slice_by_index(begin = var_3304_begin_0, end = var_3304_end_0, end_mask = var_3304_end_mask_0, x = var_3107_cast_fp16)[name = tensor("op_3304_cast_fp16")]; tensor var_3311_begin_0 = const()[name = tensor("op_3311_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3311_end_0 = const()[name = tensor("op_3311_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3311_end_mask_0 = const()[name = tensor("op_3311_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3311_cast_fp16 = slice_by_index(begin = var_3311_begin_0, end = var_3311_end_0, end_mask = var_3311_end_mask_0, x = var_3107_cast_fp16)[name = tensor("op_3311_cast_fp16")]; tensor var_3318_begin_0 = const()[name = tensor("op_3318_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3318_end_0 = const()[name = tensor("op_3318_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3318_end_mask_0 = const()[name = tensor("op_3318_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3318_cast_fp16 = slice_by_index(begin = var_3318_begin_0, end = var_3318_end_0, end_mask = var_3318_end_mask_0, x = var_3107_cast_fp16)[name = tensor("op_3318_cast_fp16")]; tensor var_3325_begin_0 = const()[name = tensor("op_3325_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3325_end_0 = const()[name = tensor("op_3325_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3325_end_mask_0 = const()[name = tensor("op_3325_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3325_cast_fp16 = slice_by_index(begin = var_3325_begin_0, end = var_3325_end_0, end_mask = var_3325_end_mask_0, x = var_3107_cast_fp16)[name = tensor("op_3325_cast_fp16")]; tensor var_3332_begin_0 = const()[name = tensor("op_3332_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3332_end_0 = const()[name = tensor("op_3332_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3332_end_mask_0 = const()[name = tensor("op_3332_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3332_cast_fp16 = slice_by_index(begin = var_3332_begin_0, end = var_3332_end_0, end_mask = var_3332_end_mask_0, x = var_3111_cast_fp16)[name = tensor("op_3332_cast_fp16")]; tensor var_3339_begin_0 = const()[name = tensor("op_3339_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3339_end_0 = const()[name = tensor("op_3339_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3339_end_mask_0 = const()[name = tensor("op_3339_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3339_cast_fp16 = slice_by_index(begin = var_3339_begin_0, end = var_3339_end_0, end_mask = var_3339_end_mask_0, x = var_3111_cast_fp16)[name = tensor("op_3339_cast_fp16")]; tensor var_3346_begin_0 = const()[name = tensor("op_3346_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3346_end_0 = const()[name = tensor("op_3346_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3346_end_mask_0 = const()[name = tensor("op_3346_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3346_cast_fp16 = slice_by_index(begin = var_3346_begin_0, end = var_3346_end_0, end_mask = var_3346_end_mask_0, x = var_3111_cast_fp16)[name = tensor("op_3346_cast_fp16")]; tensor var_3353_begin_0 = const()[name = tensor("op_3353_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3353_end_0 = const()[name = tensor("op_3353_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3353_end_mask_0 = const()[name = tensor("op_3353_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3353_cast_fp16 = slice_by_index(begin = var_3353_begin_0, end = var_3353_end_0, end_mask = var_3353_end_mask_0, x = var_3111_cast_fp16)[name = tensor("op_3353_cast_fp16")]; tensor var_3360_begin_0 = const()[name = tensor("op_3360_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3360_end_0 = const()[name = tensor("op_3360_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3360_end_mask_0 = const()[name = tensor("op_3360_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3360_cast_fp16 = slice_by_index(begin = var_3360_begin_0, end = var_3360_end_0, end_mask = var_3360_end_mask_0, x = var_3115_cast_fp16)[name = tensor("op_3360_cast_fp16")]; tensor var_3367_begin_0 = const()[name = tensor("op_3367_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3367_end_0 = const()[name = tensor("op_3367_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3367_end_mask_0 = const()[name = tensor("op_3367_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3367_cast_fp16 = slice_by_index(begin = var_3367_begin_0, end = var_3367_end_0, end_mask = var_3367_end_mask_0, x = var_3115_cast_fp16)[name = tensor("op_3367_cast_fp16")]; tensor var_3374_begin_0 = const()[name = tensor("op_3374_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3374_end_0 = const()[name = tensor("op_3374_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3374_end_mask_0 = const()[name = tensor("op_3374_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3374_cast_fp16 = slice_by_index(begin = var_3374_begin_0, end = var_3374_end_0, end_mask = var_3374_end_mask_0, x = var_3115_cast_fp16)[name = tensor("op_3374_cast_fp16")]; tensor var_3381_begin_0 = const()[name = tensor("op_3381_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3381_end_0 = const()[name = tensor("op_3381_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3381_end_mask_0 = const()[name = tensor("op_3381_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3381_cast_fp16 = slice_by_index(begin = var_3381_begin_0, end = var_3381_end_0, end_mask = var_3381_end_mask_0, x = var_3115_cast_fp16)[name = tensor("op_3381_cast_fp16")]; tensor var_3388_begin_0 = const()[name = tensor("op_3388_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3388_end_0 = const()[name = tensor("op_3388_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3388_end_mask_0 = const()[name = tensor("op_3388_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3388_cast_fp16 = slice_by_index(begin = var_3388_begin_0, end = var_3388_end_0, end_mask = var_3388_end_mask_0, x = var_3119_cast_fp16)[name = tensor("op_3388_cast_fp16")]; tensor var_3395_begin_0 = const()[name = tensor("op_3395_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3395_end_0 = const()[name = tensor("op_3395_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3395_end_mask_0 = const()[name = tensor("op_3395_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3395_cast_fp16 = slice_by_index(begin = var_3395_begin_0, end = var_3395_end_0, end_mask = var_3395_end_mask_0, x = var_3119_cast_fp16)[name = tensor("op_3395_cast_fp16")]; tensor var_3402_begin_0 = const()[name = tensor("op_3402_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3402_end_0 = const()[name = tensor("op_3402_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3402_end_mask_0 = const()[name = tensor("op_3402_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3402_cast_fp16 = slice_by_index(begin = var_3402_begin_0, end = var_3402_end_0, end_mask = var_3402_end_mask_0, x = var_3119_cast_fp16)[name = tensor("op_3402_cast_fp16")]; tensor var_3409_begin_0 = const()[name = tensor("op_3409_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3409_end_0 = const()[name = tensor("op_3409_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3409_end_mask_0 = const()[name = tensor("op_3409_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3409_cast_fp16 = slice_by_index(begin = var_3409_begin_0, end = var_3409_end_0, end_mask = var_3409_end_mask_0, x = var_3119_cast_fp16)[name = tensor("op_3409_cast_fp16")]; tensor var_3416_begin_0 = const()[name = tensor("op_3416_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3416_end_0 = const()[name = tensor("op_3416_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3416_end_mask_0 = const()[name = tensor("op_3416_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3416_cast_fp16 = slice_by_index(begin = var_3416_begin_0, end = var_3416_end_0, end_mask = var_3416_end_mask_0, x = var_3123_cast_fp16)[name = tensor("op_3416_cast_fp16")]; tensor var_3423_begin_0 = const()[name = tensor("op_3423_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3423_end_0 = const()[name = tensor("op_3423_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3423_end_mask_0 = const()[name = tensor("op_3423_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3423_cast_fp16 = slice_by_index(begin = var_3423_begin_0, end = var_3423_end_0, end_mask = var_3423_end_mask_0, x = var_3123_cast_fp16)[name = tensor("op_3423_cast_fp16")]; tensor var_3430_begin_0 = const()[name = tensor("op_3430_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3430_end_0 = const()[name = tensor("op_3430_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3430_end_mask_0 = const()[name = tensor("op_3430_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3430_cast_fp16 = slice_by_index(begin = var_3430_begin_0, end = var_3430_end_0, end_mask = var_3430_end_mask_0, x = var_3123_cast_fp16)[name = tensor("op_3430_cast_fp16")]; tensor var_3437_begin_0 = const()[name = tensor("op_3437_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3437_end_0 = const()[name = tensor("op_3437_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3437_end_mask_0 = const()[name = tensor("op_3437_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3437_cast_fp16 = slice_by_index(begin = var_3437_begin_0, end = var_3437_end_0, end_mask = var_3437_end_mask_0, x = var_3123_cast_fp16)[name = tensor("op_3437_cast_fp16")]; tensor var_3444_begin_0 = const()[name = tensor("op_3444_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3444_end_0 = const()[name = tensor("op_3444_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_3444_end_mask_0 = const()[name = tensor("op_3444_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3444_cast_fp16 = slice_by_index(begin = var_3444_begin_0, end = var_3444_end_0, end_mask = var_3444_end_mask_0, x = var_3127_cast_fp16)[name = tensor("op_3444_cast_fp16")]; tensor var_3451_begin_0 = const()[name = tensor("op_3451_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_3451_end_0 = const()[name = tensor("op_3451_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_3451_end_mask_0 = const()[name = tensor("op_3451_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3451_cast_fp16 = slice_by_index(begin = var_3451_begin_0, end = var_3451_end_0, end_mask = var_3451_end_mask_0, x = var_3127_cast_fp16)[name = tensor("op_3451_cast_fp16")]; tensor var_3458_begin_0 = const()[name = tensor("op_3458_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_3458_end_0 = const()[name = tensor("op_3458_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_3458_end_mask_0 = const()[name = tensor("op_3458_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3458_cast_fp16 = slice_by_index(begin = var_3458_begin_0, end = var_3458_end_0, end_mask = var_3458_end_mask_0, x = var_3127_cast_fp16)[name = tensor("op_3458_cast_fp16")]; tensor var_3465_begin_0 = const()[name = tensor("op_3465_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_3465_end_0 = const()[name = tensor("op_3465_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3465_end_mask_0 = const()[name = tensor("op_3465_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3465_cast_fp16 = slice_by_index(begin = var_3465_begin_0, end = var_3465_end_0, end_mask = var_3465_end_mask_0, x = var_3127_cast_fp16)[name = tensor("op_3465_cast_fp16")]; tensor k_7_perm_0 = const()[name = tensor("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_3470_begin_0 = const()[name = tensor("op_3470_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3470_end_0 = const()[name = tensor("op_3470_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_3470_end_mask_0 = const()[name = tensor("op_3470_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_8 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = tensor("transpose_8")]; tensor var_3470_cast_fp16 = slice_by_index(begin = var_3470_begin_0, end = var_3470_end_0, end_mask = var_3470_end_mask_0, x = transpose_8)[name = tensor("op_3470_cast_fp16")]; tensor var_3474_begin_0 = const()[name = tensor("op_3474_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_3474_end_0 = const()[name = tensor("op_3474_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_3474_end_mask_0 = const()[name = tensor("op_3474_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3474_cast_fp16 = slice_by_index(begin = var_3474_begin_0, end = var_3474_end_0, end_mask = var_3474_end_mask_0, x = transpose_8)[name = tensor("op_3474_cast_fp16")]; tensor var_3478_begin_0 = const()[name = tensor("op_3478_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_3478_end_0 = const()[name = tensor("op_3478_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_3478_end_mask_0 = const()[name = tensor("op_3478_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3478_cast_fp16 = slice_by_index(begin = var_3478_begin_0, end = var_3478_end_0, end_mask = var_3478_end_mask_0, x = transpose_8)[name = tensor("op_3478_cast_fp16")]; tensor var_3482_begin_0 = const()[name = tensor("op_3482_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_3482_end_0 = const()[name = tensor("op_3482_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_3482_end_mask_0 = const()[name = tensor("op_3482_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3482_cast_fp16 = slice_by_index(begin = var_3482_begin_0, end = var_3482_end_0, end_mask = var_3482_end_mask_0, x = transpose_8)[name = tensor("op_3482_cast_fp16")]; tensor var_3486_begin_0 = const()[name = tensor("op_3486_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_3486_end_0 = const()[name = tensor("op_3486_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_3486_end_mask_0 = const()[name = tensor("op_3486_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3486_cast_fp16 = slice_by_index(begin = var_3486_begin_0, end = var_3486_end_0, end_mask = var_3486_end_mask_0, x = transpose_8)[name = tensor("op_3486_cast_fp16")]; tensor var_3490_begin_0 = const()[name = tensor("op_3490_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_3490_end_0 = const()[name = tensor("op_3490_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_3490_end_mask_0 = const()[name = tensor("op_3490_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3490_cast_fp16 = slice_by_index(begin = var_3490_begin_0, end = var_3490_end_0, end_mask = var_3490_end_mask_0, x = transpose_8)[name = tensor("op_3490_cast_fp16")]; tensor var_3494_begin_0 = const()[name = tensor("op_3494_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_3494_end_0 = const()[name = tensor("op_3494_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_3494_end_mask_0 = const()[name = tensor("op_3494_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3494_cast_fp16 = slice_by_index(begin = var_3494_begin_0, end = var_3494_end_0, end_mask = var_3494_end_mask_0, x = transpose_8)[name = tensor("op_3494_cast_fp16")]; tensor var_3498_begin_0 = const()[name = tensor("op_3498_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_3498_end_0 = const()[name = tensor("op_3498_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_3498_end_mask_0 = const()[name = tensor("op_3498_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3498_cast_fp16 = slice_by_index(begin = var_3498_begin_0, end = var_3498_end_0, end_mask = var_3498_end_mask_0, x = transpose_8)[name = tensor("op_3498_cast_fp16")]; tensor var_3502_begin_0 = const()[name = tensor("op_3502_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_3502_end_0 = const()[name = tensor("op_3502_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_3502_end_mask_0 = const()[name = tensor("op_3502_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3502_cast_fp16 = slice_by_index(begin = var_3502_begin_0, end = var_3502_end_0, end_mask = var_3502_end_mask_0, x = transpose_8)[name = tensor("op_3502_cast_fp16")]; tensor var_3506_begin_0 = const()[name = tensor("op_3506_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_3506_end_0 = const()[name = tensor("op_3506_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_3506_end_mask_0 = const()[name = tensor("op_3506_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3506_cast_fp16 = slice_by_index(begin = var_3506_begin_0, end = var_3506_end_0, end_mask = var_3506_end_mask_0, x = transpose_8)[name = tensor("op_3506_cast_fp16")]; tensor var_3510_begin_0 = const()[name = tensor("op_3510_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_3510_end_0 = const()[name = tensor("op_3510_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_3510_end_mask_0 = const()[name = tensor("op_3510_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3510_cast_fp16 = slice_by_index(begin = var_3510_begin_0, end = var_3510_end_0, end_mask = var_3510_end_mask_0, x = transpose_8)[name = tensor("op_3510_cast_fp16")]; tensor var_3514_begin_0 = const()[name = tensor("op_3514_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_3514_end_0 = const()[name = tensor("op_3514_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_3514_end_mask_0 = const()[name = tensor("op_3514_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_3514_cast_fp16 = slice_by_index(begin = var_3514_begin_0, end = var_3514_end_0, end_mask = var_3514_end_mask_0, x = transpose_8)[name = tensor("op_3514_cast_fp16")]; tensor var_3516_begin_0 = const()[name = tensor("op_3516_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_3516_end_0 = const()[name = tensor("op_3516_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_3516_end_mask_0 = const()[name = tensor("op_3516_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3516_cast_fp16 = slice_by_index(begin = var_3516_begin_0, end = var_3516_end_0, end_mask = var_3516_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3516_cast_fp16")]; tensor var_3520_begin_0 = const()[name = tensor("op_3520_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_3520_end_0 = const()[name = tensor("op_3520_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_3520_end_mask_0 = const()[name = tensor("op_3520_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3520_cast_fp16 = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3520_cast_fp16")]; tensor var_3524_begin_0 = const()[name = tensor("op_3524_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_3524_end_0 = const()[name = tensor("op_3524_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_3524_end_mask_0 = const()[name = tensor("op_3524_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3524_cast_fp16 = slice_by_index(begin = var_3524_begin_0, end = var_3524_end_0, end_mask = var_3524_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3524_cast_fp16")]; tensor var_3528_begin_0 = const()[name = tensor("op_3528_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_3528_end_0 = const()[name = tensor("op_3528_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_3528_end_mask_0 = const()[name = tensor("op_3528_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3528_cast_fp16 = slice_by_index(begin = var_3528_begin_0, end = var_3528_end_0, end_mask = var_3528_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3528_cast_fp16")]; tensor var_3532_begin_0 = const()[name = tensor("op_3532_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_3532_end_0 = const()[name = tensor("op_3532_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_3532_end_mask_0 = const()[name = tensor("op_3532_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3532_cast_fp16 = slice_by_index(begin = var_3532_begin_0, end = var_3532_end_0, end_mask = var_3532_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3532_cast_fp16")]; tensor var_3536_begin_0 = const()[name = tensor("op_3536_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_3536_end_0 = const()[name = tensor("op_3536_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_3536_end_mask_0 = const()[name = tensor("op_3536_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3536_cast_fp16 = slice_by_index(begin = var_3536_begin_0, end = var_3536_end_0, end_mask = var_3536_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3536_cast_fp16")]; tensor var_3540_begin_0 = const()[name = tensor("op_3540_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_3540_end_0 = const()[name = tensor("op_3540_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_3540_end_mask_0 = const()[name = tensor("op_3540_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3540_cast_fp16")]; tensor var_3544_begin_0 = const()[name = tensor("op_3544_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_3544_end_0 = const()[name = tensor("op_3544_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_3544_end_mask_0 = const()[name = tensor("op_3544_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3544_cast_fp16 = slice_by_index(begin = var_3544_begin_0, end = var_3544_end_0, end_mask = var_3544_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3544_cast_fp16")]; tensor var_3548_begin_0 = const()[name = tensor("op_3548_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_3548_end_0 = const()[name = tensor("op_3548_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_3548_end_mask_0 = const()[name = tensor("op_3548_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3548_cast_fp16 = slice_by_index(begin = var_3548_begin_0, end = var_3548_end_0, end_mask = var_3548_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3548_cast_fp16")]; tensor var_3552_begin_0 = const()[name = tensor("op_3552_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_3552_end_0 = const()[name = tensor("op_3552_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_3552_end_mask_0 = const()[name = tensor("op_3552_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3552_cast_fp16 = slice_by_index(begin = var_3552_begin_0, end = var_3552_end_0, end_mask = var_3552_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3552_cast_fp16")]; tensor var_3556_begin_0 = const()[name = tensor("op_3556_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_3556_end_0 = const()[name = tensor("op_3556_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_3556_end_mask_0 = const()[name = tensor("op_3556_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3556_cast_fp16 = slice_by_index(begin = var_3556_begin_0, end = var_3556_end_0, end_mask = var_3556_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3556_cast_fp16")]; tensor var_3560_begin_0 = const()[name = tensor("op_3560_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_3560_end_0 = const()[name = tensor("op_3560_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_3560_end_mask_0 = const()[name = tensor("op_3560_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_3560_cast_fp16 = slice_by_index(begin = var_3560_begin_0, end = var_3560_end_0, end_mask = var_3560_end_mask_0, x = value_7_cast_fp16)[name = tensor("op_3560_cast_fp16")]; tensor var_3564_equation_0 = const()[name = tensor("op_3564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3564_cast_fp16 = einsum(equation = var_3564_equation_0, values = (var_3470_cast_fp16, var_3136_cast_fp16))[name = tensor("op_3564_cast_fp16")]; tensor var_3565_to_fp16 = const()[name = tensor("op_3565_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_289_cast_fp16 = mul(x = var_3564_cast_fp16, y = var_3565_to_fp16)[name = tensor("aw_chunk_289_cast_fp16")]; tensor var_3568_equation_0 = const()[name = tensor("op_3568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3568_cast_fp16 = einsum(equation = var_3568_equation_0, values = (var_3470_cast_fp16, var_3143_cast_fp16))[name = tensor("op_3568_cast_fp16")]; tensor var_3569_to_fp16 = const()[name = tensor("op_3569_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_291_cast_fp16 = mul(x = var_3568_cast_fp16, y = var_3569_to_fp16)[name = tensor("aw_chunk_291_cast_fp16")]; tensor var_3572_equation_0 = const()[name = tensor("op_3572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3572_cast_fp16 = einsum(equation = var_3572_equation_0, values = (var_3470_cast_fp16, var_3150_cast_fp16))[name = tensor("op_3572_cast_fp16")]; tensor var_3573_to_fp16 = const()[name = tensor("op_3573_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_293_cast_fp16 = mul(x = var_3572_cast_fp16, y = var_3573_to_fp16)[name = tensor("aw_chunk_293_cast_fp16")]; tensor var_3576_equation_0 = const()[name = tensor("op_3576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3576_cast_fp16 = einsum(equation = var_3576_equation_0, values = (var_3470_cast_fp16, var_3157_cast_fp16))[name = tensor("op_3576_cast_fp16")]; tensor var_3577_to_fp16 = const()[name = tensor("op_3577_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_295_cast_fp16 = mul(x = var_3576_cast_fp16, y = var_3577_to_fp16)[name = tensor("aw_chunk_295_cast_fp16")]; tensor var_3580_equation_0 = const()[name = tensor("op_3580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3580_cast_fp16 = einsum(equation = var_3580_equation_0, values = (var_3474_cast_fp16, var_3164_cast_fp16))[name = tensor("op_3580_cast_fp16")]; tensor var_3581_to_fp16 = const()[name = tensor("op_3581_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_297_cast_fp16 = mul(x = var_3580_cast_fp16, y = var_3581_to_fp16)[name = tensor("aw_chunk_297_cast_fp16")]; tensor var_3584_equation_0 = const()[name = tensor("op_3584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3584_cast_fp16 = einsum(equation = var_3584_equation_0, values = (var_3474_cast_fp16, var_3171_cast_fp16))[name = tensor("op_3584_cast_fp16")]; tensor var_3585_to_fp16 = const()[name = tensor("op_3585_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_299_cast_fp16 = mul(x = var_3584_cast_fp16, y = var_3585_to_fp16)[name = tensor("aw_chunk_299_cast_fp16")]; tensor var_3588_equation_0 = const()[name = tensor("op_3588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3588_cast_fp16 = einsum(equation = var_3588_equation_0, values = (var_3474_cast_fp16, var_3178_cast_fp16))[name = tensor("op_3588_cast_fp16")]; tensor var_3589_to_fp16 = const()[name = tensor("op_3589_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_301_cast_fp16 = mul(x = var_3588_cast_fp16, y = var_3589_to_fp16)[name = tensor("aw_chunk_301_cast_fp16")]; tensor var_3592_equation_0 = const()[name = tensor("op_3592_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3592_cast_fp16 = einsum(equation = var_3592_equation_0, values = (var_3474_cast_fp16, var_3185_cast_fp16))[name = tensor("op_3592_cast_fp16")]; tensor var_3593_to_fp16 = const()[name = tensor("op_3593_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_303_cast_fp16 = mul(x = var_3592_cast_fp16, y = var_3593_to_fp16)[name = tensor("aw_chunk_303_cast_fp16")]; tensor var_3596_equation_0 = const()[name = tensor("op_3596_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3596_cast_fp16 = einsum(equation = var_3596_equation_0, values = (var_3478_cast_fp16, var_3192_cast_fp16))[name = tensor("op_3596_cast_fp16")]; tensor var_3597_to_fp16 = const()[name = tensor("op_3597_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_305_cast_fp16 = mul(x = var_3596_cast_fp16, y = var_3597_to_fp16)[name = tensor("aw_chunk_305_cast_fp16")]; tensor var_3600_equation_0 = const()[name = tensor("op_3600_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3600_cast_fp16 = einsum(equation = var_3600_equation_0, values = (var_3478_cast_fp16, var_3199_cast_fp16))[name = tensor("op_3600_cast_fp16")]; tensor var_3601_to_fp16 = const()[name = tensor("op_3601_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_307_cast_fp16 = mul(x = var_3600_cast_fp16, y = var_3601_to_fp16)[name = tensor("aw_chunk_307_cast_fp16")]; tensor var_3604_equation_0 = const()[name = tensor("op_3604_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3604_cast_fp16 = einsum(equation = var_3604_equation_0, values = (var_3478_cast_fp16, var_3206_cast_fp16))[name = tensor("op_3604_cast_fp16")]; tensor var_3605_to_fp16 = const()[name = tensor("op_3605_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_309_cast_fp16 = mul(x = var_3604_cast_fp16, y = var_3605_to_fp16)[name = tensor("aw_chunk_309_cast_fp16")]; tensor var_3608_equation_0 = const()[name = tensor("op_3608_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3608_cast_fp16 = einsum(equation = var_3608_equation_0, values = (var_3478_cast_fp16, var_3213_cast_fp16))[name = tensor("op_3608_cast_fp16")]; tensor var_3609_to_fp16 = const()[name = tensor("op_3609_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_311_cast_fp16 = mul(x = var_3608_cast_fp16, y = var_3609_to_fp16)[name = tensor("aw_chunk_311_cast_fp16")]; tensor var_3612_equation_0 = const()[name = tensor("op_3612_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3612_cast_fp16 = einsum(equation = var_3612_equation_0, values = (var_3482_cast_fp16, var_3220_cast_fp16))[name = tensor("op_3612_cast_fp16")]; tensor var_3613_to_fp16 = const()[name = tensor("op_3613_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_313_cast_fp16 = mul(x = var_3612_cast_fp16, y = var_3613_to_fp16)[name = tensor("aw_chunk_313_cast_fp16")]; tensor var_3616_equation_0 = const()[name = tensor("op_3616_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3616_cast_fp16 = einsum(equation = var_3616_equation_0, values = (var_3482_cast_fp16, var_3227_cast_fp16))[name = tensor("op_3616_cast_fp16")]; tensor var_3617_to_fp16 = const()[name = tensor("op_3617_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_315_cast_fp16 = mul(x = var_3616_cast_fp16, y = var_3617_to_fp16)[name = tensor("aw_chunk_315_cast_fp16")]; tensor var_3620_equation_0 = const()[name = tensor("op_3620_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3620_cast_fp16 = einsum(equation = var_3620_equation_0, values = (var_3482_cast_fp16, var_3234_cast_fp16))[name = tensor("op_3620_cast_fp16")]; tensor var_3621_to_fp16 = const()[name = tensor("op_3621_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_317_cast_fp16 = mul(x = var_3620_cast_fp16, y = var_3621_to_fp16)[name = tensor("aw_chunk_317_cast_fp16")]; tensor var_3624_equation_0 = const()[name = tensor("op_3624_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3624_cast_fp16 = einsum(equation = var_3624_equation_0, values = (var_3482_cast_fp16, var_3241_cast_fp16))[name = tensor("op_3624_cast_fp16")]; tensor var_3625_to_fp16 = const()[name = tensor("op_3625_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_319_cast_fp16 = mul(x = var_3624_cast_fp16, y = var_3625_to_fp16)[name = tensor("aw_chunk_319_cast_fp16")]; tensor var_3628_equation_0 = const()[name = tensor("op_3628_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3628_cast_fp16 = einsum(equation = var_3628_equation_0, values = (var_3486_cast_fp16, var_3248_cast_fp16))[name = tensor("op_3628_cast_fp16")]; tensor var_3629_to_fp16 = const()[name = tensor("op_3629_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_321_cast_fp16 = mul(x = var_3628_cast_fp16, y = var_3629_to_fp16)[name = tensor("aw_chunk_321_cast_fp16")]; tensor var_3632_equation_0 = const()[name = tensor("op_3632_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3632_cast_fp16 = einsum(equation = var_3632_equation_0, values = (var_3486_cast_fp16, var_3255_cast_fp16))[name = tensor("op_3632_cast_fp16")]; tensor var_3633_to_fp16 = const()[name = tensor("op_3633_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_323_cast_fp16 = mul(x = var_3632_cast_fp16, y = var_3633_to_fp16)[name = tensor("aw_chunk_323_cast_fp16")]; tensor var_3636_equation_0 = const()[name = tensor("op_3636_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3636_cast_fp16 = einsum(equation = var_3636_equation_0, values = (var_3486_cast_fp16, var_3262_cast_fp16))[name = tensor("op_3636_cast_fp16")]; tensor var_3637_to_fp16 = const()[name = tensor("op_3637_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_325_cast_fp16 = mul(x = var_3636_cast_fp16, y = var_3637_to_fp16)[name = tensor("aw_chunk_325_cast_fp16")]; tensor var_3640_equation_0 = const()[name = tensor("op_3640_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3640_cast_fp16 = einsum(equation = var_3640_equation_0, values = (var_3486_cast_fp16, var_3269_cast_fp16))[name = tensor("op_3640_cast_fp16")]; tensor var_3641_to_fp16 = const()[name = tensor("op_3641_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_327_cast_fp16 = mul(x = var_3640_cast_fp16, y = var_3641_to_fp16)[name = tensor("aw_chunk_327_cast_fp16")]; tensor var_3644_equation_0 = const()[name = tensor("op_3644_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3644_cast_fp16 = einsum(equation = var_3644_equation_0, values = (var_3490_cast_fp16, var_3276_cast_fp16))[name = tensor("op_3644_cast_fp16")]; tensor var_3645_to_fp16 = const()[name = tensor("op_3645_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_329_cast_fp16 = mul(x = var_3644_cast_fp16, y = var_3645_to_fp16)[name = tensor("aw_chunk_329_cast_fp16")]; tensor var_3648_equation_0 = const()[name = tensor("op_3648_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3648_cast_fp16 = einsum(equation = var_3648_equation_0, values = (var_3490_cast_fp16, var_3283_cast_fp16))[name = tensor("op_3648_cast_fp16")]; tensor var_3649_to_fp16 = const()[name = tensor("op_3649_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_331_cast_fp16 = mul(x = var_3648_cast_fp16, y = var_3649_to_fp16)[name = tensor("aw_chunk_331_cast_fp16")]; tensor var_3652_equation_0 = const()[name = tensor("op_3652_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3652_cast_fp16 = einsum(equation = var_3652_equation_0, values = (var_3490_cast_fp16, var_3290_cast_fp16))[name = tensor("op_3652_cast_fp16")]; tensor var_3653_to_fp16 = const()[name = tensor("op_3653_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_333_cast_fp16 = mul(x = var_3652_cast_fp16, y = var_3653_to_fp16)[name = tensor("aw_chunk_333_cast_fp16")]; tensor var_3656_equation_0 = const()[name = tensor("op_3656_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3656_cast_fp16 = einsum(equation = var_3656_equation_0, values = (var_3490_cast_fp16, var_3297_cast_fp16))[name = tensor("op_3656_cast_fp16")]; tensor var_3657_to_fp16 = const()[name = tensor("op_3657_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_335_cast_fp16 = mul(x = var_3656_cast_fp16, y = var_3657_to_fp16)[name = tensor("aw_chunk_335_cast_fp16")]; tensor var_3660_equation_0 = const()[name = tensor("op_3660_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3660_cast_fp16 = einsum(equation = var_3660_equation_0, values = (var_3494_cast_fp16, var_3304_cast_fp16))[name = tensor("op_3660_cast_fp16")]; tensor var_3661_to_fp16 = const()[name = tensor("op_3661_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_337_cast_fp16 = mul(x = var_3660_cast_fp16, y = var_3661_to_fp16)[name = tensor("aw_chunk_337_cast_fp16")]; tensor var_3664_equation_0 = const()[name = tensor("op_3664_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3664_cast_fp16 = einsum(equation = var_3664_equation_0, values = (var_3494_cast_fp16, var_3311_cast_fp16))[name = tensor("op_3664_cast_fp16")]; tensor var_3665_to_fp16 = const()[name = tensor("op_3665_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_339_cast_fp16 = mul(x = var_3664_cast_fp16, y = var_3665_to_fp16)[name = tensor("aw_chunk_339_cast_fp16")]; tensor var_3668_equation_0 = const()[name = tensor("op_3668_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3668_cast_fp16 = einsum(equation = var_3668_equation_0, values = (var_3494_cast_fp16, var_3318_cast_fp16))[name = tensor("op_3668_cast_fp16")]; tensor var_3669_to_fp16 = const()[name = tensor("op_3669_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_341_cast_fp16 = mul(x = var_3668_cast_fp16, y = var_3669_to_fp16)[name = tensor("aw_chunk_341_cast_fp16")]; tensor var_3672_equation_0 = const()[name = tensor("op_3672_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3672_cast_fp16 = einsum(equation = var_3672_equation_0, values = (var_3494_cast_fp16, var_3325_cast_fp16))[name = tensor("op_3672_cast_fp16")]; tensor var_3673_to_fp16 = const()[name = tensor("op_3673_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_343_cast_fp16 = mul(x = var_3672_cast_fp16, y = var_3673_to_fp16)[name = tensor("aw_chunk_343_cast_fp16")]; tensor var_3676_equation_0 = const()[name = tensor("op_3676_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3676_cast_fp16 = einsum(equation = var_3676_equation_0, values = (var_3498_cast_fp16, var_3332_cast_fp16))[name = tensor("op_3676_cast_fp16")]; tensor var_3677_to_fp16 = const()[name = tensor("op_3677_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_345_cast_fp16 = mul(x = var_3676_cast_fp16, y = var_3677_to_fp16)[name = tensor("aw_chunk_345_cast_fp16")]; tensor var_3680_equation_0 = const()[name = tensor("op_3680_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3680_cast_fp16 = einsum(equation = var_3680_equation_0, values = (var_3498_cast_fp16, var_3339_cast_fp16))[name = tensor("op_3680_cast_fp16")]; tensor var_3681_to_fp16 = const()[name = tensor("op_3681_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_347_cast_fp16 = mul(x = var_3680_cast_fp16, y = var_3681_to_fp16)[name = tensor("aw_chunk_347_cast_fp16")]; tensor var_3684_equation_0 = const()[name = tensor("op_3684_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3684_cast_fp16 = einsum(equation = var_3684_equation_0, values = (var_3498_cast_fp16, var_3346_cast_fp16))[name = tensor("op_3684_cast_fp16")]; tensor var_3685_to_fp16 = const()[name = tensor("op_3685_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_349_cast_fp16 = mul(x = var_3684_cast_fp16, y = var_3685_to_fp16)[name = tensor("aw_chunk_349_cast_fp16")]; tensor var_3688_equation_0 = const()[name = tensor("op_3688_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3688_cast_fp16 = einsum(equation = var_3688_equation_0, values = (var_3498_cast_fp16, var_3353_cast_fp16))[name = tensor("op_3688_cast_fp16")]; tensor var_3689_to_fp16 = const()[name = tensor("op_3689_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_351_cast_fp16 = mul(x = var_3688_cast_fp16, y = var_3689_to_fp16)[name = tensor("aw_chunk_351_cast_fp16")]; tensor var_3692_equation_0 = const()[name = tensor("op_3692_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3692_cast_fp16 = einsum(equation = var_3692_equation_0, values = (var_3502_cast_fp16, var_3360_cast_fp16))[name = tensor("op_3692_cast_fp16")]; tensor var_3693_to_fp16 = const()[name = tensor("op_3693_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_353_cast_fp16 = mul(x = var_3692_cast_fp16, y = var_3693_to_fp16)[name = tensor("aw_chunk_353_cast_fp16")]; tensor var_3696_equation_0 = const()[name = tensor("op_3696_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3696_cast_fp16 = einsum(equation = var_3696_equation_0, values = (var_3502_cast_fp16, var_3367_cast_fp16))[name = tensor("op_3696_cast_fp16")]; tensor var_3697_to_fp16 = const()[name = tensor("op_3697_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_355_cast_fp16 = mul(x = var_3696_cast_fp16, y = var_3697_to_fp16)[name = tensor("aw_chunk_355_cast_fp16")]; tensor var_3700_equation_0 = const()[name = tensor("op_3700_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3700_cast_fp16 = einsum(equation = var_3700_equation_0, values = (var_3502_cast_fp16, var_3374_cast_fp16))[name = tensor("op_3700_cast_fp16")]; tensor var_3701_to_fp16 = const()[name = tensor("op_3701_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_357_cast_fp16 = mul(x = var_3700_cast_fp16, y = var_3701_to_fp16)[name = tensor("aw_chunk_357_cast_fp16")]; tensor var_3704_equation_0 = const()[name = tensor("op_3704_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3704_cast_fp16 = einsum(equation = var_3704_equation_0, values = (var_3502_cast_fp16, var_3381_cast_fp16))[name = tensor("op_3704_cast_fp16")]; tensor var_3705_to_fp16 = const()[name = tensor("op_3705_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_359_cast_fp16 = mul(x = var_3704_cast_fp16, y = var_3705_to_fp16)[name = tensor("aw_chunk_359_cast_fp16")]; tensor var_3708_equation_0 = const()[name = tensor("op_3708_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3708_cast_fp16 = einsum(equation = var_3708_equation_0, values = (var_3506_cast_fp16, var_3388_cast_fp16))[name = tensor("op_3708_cast_fp16")]; tensor var_3709_to_fp16 = const()[name = tensor("op_3709_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_361_cast_fp16 = mul(x = var_3708_cast_fp16, y = var_3709_to_fp16)[name = tensor("aw_chunk_361_cast_fp16")]; tensor var_3712_equation_0 = const()[name = tensor("op_3712_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3712_cast_fp16 = einsum(equation = var_3712_equation_0, values = (var_3506_cast_fp16, var_3395_cast_fp16))[name = tensor("op_3712_cast_fp16")]; tensor var_3713_to_fp16 = const()[name = tensor("op_3713_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_363_cast_fp16 = mul(x = var_3712_cast_fp16, y = var_3713_to_fp16)[name = tensor("aw_chunk_363_cast_fp16")]; tensor var_3716_equation_0 = const()[name = tensor("op_3716_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3716_cast_fp16 = einsum(equation = var_3716_equation_0, values = (var_3506_cast_fp16, var_3402_cast_fp16))[name = tensor("op_3716_cast_fp16")]; tensor var_3717_to_fp16 = const()[name = tensor("op_3717_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_365_cast_fp16 = mul(x = var_3716_cast_fp16, y = var_3717_to_fp16)[name = tensor("aw_chunk_365_cast_fp16")]; tensor var_3720_equation_0 = const()[name = tensor("op_3720_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3720_cast_fp16 = einsum(equation = var_3720_equation_0, values = (var_3506_cast_fp16, var_3409_cast_fp16))[name = tensor("op_3720_cast_fp16")]; tensor var_3721_to_fp16 = const()[name = tensor("op_3721_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_367_cast_fp16 = mul(x = var_3720_cast_fp16, y = var_3721_to_fp16)[name = tensor("aw_chunk_367_cast_fp16")]; tensor var_3724_equation_0 = const()[name = tensor("op_3724_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3724_cast_fp16 = einsum(equation = var_3724_equation_0, values = (var_3510_cast_fp16, var_3416_cast_fp16))[name = tensor("op_3724_cast_fp16")]; tensor var_3725_to_fp16 = const()[name = tensor("op_3725_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_369_cast_fp16 = mul(x = var_3724_cast_fp16, y = var_3725_to_fp16)[name = tensor("aw_chunk_369_cast_fp16")]; tensor var_3728_equation_0 = const()[name = tensor("op_3728_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3728_cast_fp16 = einsum(equation = var_3728_equation_0, values = (var_3510_cast_fp16, var_3423_cast_fp16))[name = tensor("op_3728_cast_fp16")]; tensor var_3729_to_fp16 = const()[name = tensor("op_3729_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_371_cast_fp16 = mul(x = var_3728_cast_fp16, y = var_3729_to_fp16)[name = tensor("aw_chunk_371_cast_fp16")]; tensor var_3732_equation_0 = const()[name = tensor("op_3732_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3732_cast_fp16 = einsum(equation = var_3732_equation_0, values = (var_3510_cast_fp16, var_3430_cast_fp16))[name = tensor("op_3732_cast_fp16")]; tensor var_3733_to_fp16 = const()[name = tensor("op_3733_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_373_cast_fp16 = mul(x = var_3732_cast_fp16, y = var_3733_to_fp16)[name = tensor("aw_chunk_373_cast_fp16")]; tensor var_3736_equation_0 = const()[name = tensor("op_3736_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3736_cast_fp16 = einsum(equation = var_3736_equation_0, values = (var_3510_cast_fp16, var_3437_cast_fp16))[name = tensor("op_3736_cast_fp16")]; tensor var_3737_to_fp16 = const()[name = tensor("op_3737_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_375_cast_fp16 = mul(x = var_3736_cast_fp16, y = var_3737_to_fp16)[name = tensor("aw_chunk_375_cast_fp16")]; tensor var_3740_equation_0 = const()[name = tensor("op_3740_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3740_cast_fp16 = einsum(equation = var_3740_equation_0, values = (var_3514_cast_fp16, var_3444_cast_fp16))[name = tensor("op_3740_cast_fp16")]; tensor var_3741_to_fp16 = const()[name = tensor("op_3741_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_377_cast_fp16 = mul(x = var_3740_cast_fp16, y = var_3741_to_fp16)[name = tensor("aw_chunk_377_cast_fp16")]; tensor var_3744_equation_0 = const()[name = tensor("op_3744_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3744_cast_fp16 = einsum(equation = var_3744_equation_0, values = (var_3514_cast_fp16, var_3451_cast_fp16))[name = tensor("op_3744_cast_fp16")]; tensor var_3745_to_fp16 = const()[name = tensor("op_3745_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_379_cast_fp16 = mul(x = var_3744_cast_fp16, y = var_3745_to_fp16)[name = tensor("aw_chunk_379_cast_fp16")]; tensor var_3748_equation_0 = const()[name = tensor("op_3748_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3748_cast_fp16 = einsum(equation = var_3748_equation_0, values = (var_3514_cast_fp16, var_3458_cast_fp16))[name = tensor("op_3748_cast_fp16")]; tensor var_3749_to_fp16 = const()[name = tensor("op_3749_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_381_cast_fp16 = mul(x = var_3748_cast_fp16, y = var_3749_to_fp16)[name = tensor("aw_chunk_381_cast_fp16")]; tensor var_3752_equation_0 = const()[name = tensor("op_3752_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_3752_cast_fp16 = einsum(equation = var_3752_equation_0, values = (var_3514_cast_fp16, var_3465_cast_fp16))[name = tensor("op_3752_cast_fp16")]; tensor var_3753_to_fp16 = const()[name = tensor("op_3753_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_383_cast_fp16 = mul(x = var_3752_cast_fp16, y = var_3753_to_fp16)[name = tensor("aw_chunk_383_cast_fp16")]; tensor var_3755_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_289_cast_fp16)[name = tensor("op_3755_cast_fp16")]; tensor var_3756_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_291_cast_fp16)[name = tensor("op_3756_cast_fp16")]; tensor var_3757_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_293_cast_fp16)[name = tensor("op_3757_cast_fp16")]; tensor var_3758_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_295_cast_fp16)[name = tensor("op_3758_cast_fp16")]; tensor var_3759_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_297_cast_fp16)[name = tensor("op_3759_cast_fp16")]; tensor var_3760_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_299_cast_fp16)[name = tensor("op_3760_cast_fp16")]; tensor var_3761_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_301_cast_fp16)[name = tensor("op_3761_cast_fp16")]; tensor var_3762_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_303_cast_fp16)[name = tensor("op_3762_cast_fp16")]; tensor var_3763_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_305_cast_fp16)[name = tensor("op_3763_cast_fp16")]; tensor var_3764_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_307_cast_fp16)[name = tensor("op_3764_cast_fp16")]; tensor var_3765_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_309_cast_fp16)[name = tensor("op_3765_cast_fp16")]; tensor var_3766_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_311_cast_fp16)[name = tensor("op_3766_cast_fp16")]; tensor var_3767_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_313_cast_fp16)[name = tensor("op_3767_cast_fp16")]; tensor var_3768_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_315_cast_fp16)[name = tensor("op_3768_cast_fp16")]; tensor var_3769_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_317_cast_fp16)[name = tensor("op_3769_cast_fp16")]; tensor var_3770_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_319_cast_fp16)[name = tensor("op_3770_cast_fp16")]; tensor var_3771_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_321_cast_fp16)[name = tensor("op_3771_cast_fp16")]; tensor var_3772_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_323_cast_fp16)[name = tensor("op_3772_cast_fp16")]; tensor var_3773_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_325_cast_fp16)[name = tensor("op_3773_cast_fp16")]; tensor var_3774_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_327_cast_fp16)[name = tensor("op_3774_cast_fp16")]; tensor var_3775_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_329_cast_fp16)[name = tensor("op_3775_cast_fp16")]; tensor var_3776_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_331_cast_fp16)[name = tensor("op_3776_cast_fp16")]; tensor var_3777_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_333_cast_fp16)[name = tensor("op_3777_cast_fp16")]; tensor var_3778_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_335_cast_fp16)[name = tensor("op_3778_cast_fp16")]; tensor var_3779_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_337_cast_fp16)[name = tensor("op_3779_cast_fp16")]; tensor var_3780_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_339_cast_fp16)[name = tensor("op_3780_cast_fp16")]; tensor var_3781_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_341_cast_fp16)[name = tensor("op_3781_cast_fp16")]; tensor var_3782_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_343_cast_fp16)[name = tensor("op_3782_cast_fp16")]; tensor var_3783_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_345_cast_fp16)[name = tensor("op_3783_cast_fp16")]; tensor var_3784_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_347_cast_fp16)[name = tensor("op_3784_cast_fp16")]; tensor var_3785_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_349_cast_fp16)[name = tensor("op_3785_cast_fp16")]; tensor var_3786_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_351_cast_fp16)[name = tensor("op_3786_cast_fp16")]; tensor var_3787_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_353_cast_fp16)[name = tensor("op_3787_cast_fp16")]; tensor var_3788_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_355_cast_fp16)[name = tensor("op_3788_cast_fp16")]; tensor var_3789_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_357_cast_fp16)[name = tensor("op_3789_cast_fp16")]; tensor var_3790_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_359_cast_fp16)[name = tensor("op_3790_cast_fp16")]; tensor var_3791_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_361_cast_fp16)[name = tensor("op_3791_cast_fp16")]; tensor var_3792_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_363_cast_fp16)[name = tensor("op_3792_cast_fp16")]; tensor var_3793_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_365_cast_fp16)[name = tensor("op_3793_cast_fp16")]; tensor var_3794_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_367_cast_fp16)[name = tensor("op_3794_cast_fp16")]; tensor var_3795_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_369_cast_fp16)[name = tensor("op_3795_cast_fp16")]; tensor var_3796_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_371_cast_fp16)[name = tensor("op_3796_cast_fp16")]; tensor var_3797_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_373_cast_fp16)[name = tensor("op_3797_cast_fp16")]; tensor var_3798_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_375_cast_fp16)[name = tensor("op_3798_cast_fp16")]; tensor var_3799_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_377_cast_fp16)[name = tensor("op_3799_cast_fp16")]; tensor var_3800_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_379_cast_fp16)[name = tensor("op_3800_cast_fp16")]; tensor var_3801_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_381_cast_fp16)[name = tensor("op_3801_cast_fp16")]; tensor var_3802_cast_fp16 = softmax(axis = var_3028, x = aw_chunk_383_cast_fp16)[name = tensor("op_3802_cast_fp16")]; tensor var_3804_equation_0 = const()[name = tensor("op_3804_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3804_cast_fp16 = einsum(equation = var_3804_equation_0, values = (var_3516_cast_fp16, var_3755_cast_fp16))[name = tensor("op_3804_cast_fp16")]; tensor var_3806_equation_0 = const()[name = tensor("op_3806_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3806_cast_fp16 = einsum(equation = var_3806_equation_0, values = (var_3516_cast_fp16, var_3756_cast_fp16))[name = tensor("op_3806_cast_fp16")]; tensor var_3808_equation_0 = const()[name = tensor("op_3808_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3808_cast_fp16 = einsum(equation = var_3808_equation_0, values = (var_3516_cast_fp16, var_3757_cast_fp16))[name = tensor("op_3808_cast_fp16")]; tensor var_3810_equation_0 = const()[name = tensor("op_3810_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3810_cast_fp16 = einsum(equation = var_3810_equation_0, values = (var_3516_cast_fp16, var_3758_cast_fp16))[name = tensor("op_3810_cast_fp16")]; tensor var_3812_equation_0 = const()[name = tensor("op_3812_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3812_cast_fp16 = einsum(equation = var_3812_equation_0, values = (var_3520_cast_fp16, var_3759_cast_fp16))[name = tensor("op_3812_cast_fp16")]; tensor var_3814_equation_0 = const()[name = tensor("op_3814_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3814_cast_fp16 = einsum(equation = var_3814_equation_0, values = (var_3520_cast_fp16, var_3760_cast_fp16))[name = tensor("op_3814_cast_fp16")]; tensor var_3816_equation_0 = const()[name = tensor("op_3816_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3816_cast_fp16 = einsum(equation = var_3816_equation_0, values = (var_3520_cast_fp16, var_3761_cast_fp16))[name = tensor("op_3816_cast_fp16")]; tensor var_3818_equation_0 = const()[name = tensor("op_3818_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3818_cast_fp16 = einsum(equation = var_3818_equation_0, values = (var_3520_cast_fp16, var_3762_cast_fp16))[name = tensor("op_3818_cast_fp16")]; tensor var_3820_equation_0 = const()[name = tensor("op_3820_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3820_cast_fp16 = einsum(equation = var_3820_equation_0, values = (var_3524_cast_fp16, var_3763_cast_fp16))[name = tensor("op_3820_cast_fp16")]; tensor var_3822_equation_0 = const()[name = tensor("op_3822_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3822_cast_fp16 = einsum(equation = var_3822_equation_0, values = (var_3524_cast_fp16, var_3764_cast_fp16))[name = tensor("op_3822_cast_fp16")]; tensor var_3824_equation_0 = const()[name = tensor("op_3824_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3824_cast_fp16 = einsum(equation = var_3824_equation_0, values = (var_3524_cast_fp16, var_3765_cast_fp16))[name = tensor("op_3824_cast_fp16")]; tensor var_3826_equation_0 = const()[name = tensor("op_3826_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3826_cast_fp16 = einsum(equation = var_3826_equation_0, values = (var_3524_cast_fp16, var_3766_cast_fp16))[name = tensor("op_3826_cast_fp16")]; tensor var_3828_equation_0 = const()[name = tensor("op_3828_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3828_cast_fp16 = einsum(equation = var_3828_equation_0, values = (var_3528_cast_fp16, var_3767_cast_fp16))[name = tensor("op_3828_cast_fp16")]; tensor var_3830_equation_0 = const()[name = tensor("op_3830_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3830_cast_fp16 = einsum(equation = var_3830_equation_0, values = (var_3528_cast_fp16, var_3768_cast_fp16))[name = tensor("op_3830_cast_fp16")]; tensor var_3832_equation_0 = const()[name = tensor("op_3832_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3832_cast_fp16 = einsum(equation = var_3832_equation_0, values = (var_3528_cast_fp16, var_3769_cast_fp16))[name = tensor("op_3832_cast_fp16")]; tensor var_3834_equation_0 = const()[name = tensor("op_3834_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3834_cast_fp16 = einsum(equation = var_3834_equation_0, values = (var_3528_cast_fp16, var_3770_cast_fp16))[name = tensor("op_3834_cast_fp16")]; tensor var_3836_equation_0 = const()[name = tensor("op_3836_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3836_cast_fp16 = einsum(equation = var_3836_equation_0, values = (var_3532_cast_fp16, var_3771_cast_fp16))[name = tensor("op_3836_cast_fp16")]; tensor var_3838_equation_0 = const()[name = tensor("op_3838_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3838_cast_fp16 = einsum(equation = var_3838_equation_0, values = (var_3532_cast_fp16, var_3772_cast_fp16))[name = tensor("op_3838_cast_fp16")]; tensor var_3840_equation_0 = const()[name = tensor("op_3840_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3840_cast_fp16 = einsum(equation = var_3840_equation_0, values = (var_3532_cast_fp16, var_3773_cast_fp16))[name = tensor("op_3840_cast_fp16")]; tensor var_3842_equation_0 = const()[name = tensor("op_3842_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3842_cast_fp16 = einsum(equation = var_3842_equation_0, values = (var_3532_cast_fp16, var_3774_cast_fp16))[name = tensor("op_3842_cast_fp16")]; tensor var_3844_equation_0 = const()[name = tensor("op_3844_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3844_cast_fp16 = einsum(equation = var_3844_equation_0, values = (var_3536_cast_fp16, var_3775_cast_fp16))[name = tensor("op_3844_cast_fp16")]; tensor var_3846_equation_0 = const()[name = tensor("op_3846_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3846_cast_fp16 = einsum(equation = var_3846_equation_0, values = (var_3536_cast_fp16, var_3776_cast_fp16))[name = tensor("op_3846_cast_fp16")]; tensor var_3848_equation_0 = const()[name = tensor("op_3848_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3848_cast_fp16 = einsum(equation = var_3848_equation_0, values = (var_3536_cast_fp16, var_3777_cast_fp16))[name = tensor("op_3848_cast_fp16")]; tensor var_3850_equation_0 = const()[name = tensor("op_3850_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3850_cast_fp16 = einsum(equation = var_3850_equation_0, values = (var_3536_cast_fp16, var_3778_cast_fp16))[name = tensor("op_3850_cast_fp16")]; tensor var_3852_equation_0 = const()[name = tensor("op_3852_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3852_cast_fp16 = einsum(equation = var_3852_equation_0, values = (var_3540_cast_fp16, var_3779_cast_fp16))[name = tensor("op_3852_cast_fp16")]; tensor var_3854_equation_0 = const()[name = tensor("op_3854_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3854_cast_fp16 = einsum(equation = var_3854_equation_0, values = (var_3540_cast_fp16, var_3780_cast_fp16))[name = tensor("op_3854_cast_fp16")]; tensor var_3856_equation_0 = const()[name = tensor("op_3856_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3856_cast_fp16 = einsum(equation = var_3856_equation_0, values = (var_3540_cast_fp16, var_3781_cast_fp16))[name = tensor("op_3856_cast_fp16")]; tensor var_3858_equation_0 = const()[name = tensor("op_3858_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3858_cast_fp16 = einsum(equation = var_3858_equation_0, values = (var_3540_cast_fp16, var_3782_cast_fp16))[name = tensor("op_3858_cast_fp16")]; tensor var_3860_equation_0 = const()[name = tensor("op_3860_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3860_cast_fp16 = einsum(equation = var_3860_equation_0, values = (var_3544_cast_fp16, var_3783_cast_fp16))[name = tensor("op_3860_cast_fp16")]; tensor var_3862_equation_0 = const()[name = tensor("op_3862_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3862_cast_fp16 = einsum(equation = var_3862_equation_0, values = (var_3544_cast_fp16, var_3784_cast_fp16))[name = tensor("op_3862_cast_fp16")]; tensor var_3864_equation_0 = const()[name = tensor("op_3864_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3864_cast_fp16 = einsum(equation = var_3864_equation_0, values = (var_3544_cast_fp16, var_3785_cast_fp16))[name = tensor("op_3864_cast_fp16")]; tensor var_3866_equation_0 = const()[name = tensor("op_3866_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3866_cast_fp16 = einsum(equation = var_3866_equation_0, values = (var_3544_cast_fp16, var_3786_cast_fp16))[name = tensor("op_3866_cast_fp16")]; tensor var_3868_equation_0 = const()[name = tensor("op_3868_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3868_cast_fp16 = einsum(equation = var_3868_equation_0, values = (var_3548_cast_fp16, var_3787_cast_fp16))[name = tensor("op_3868_cast_fp16")]; tensor var_3870_equation_0 = const()[name = tensor("op_3870_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3870_cast_fp16 = einsum(equation = var_3870_equation_0, values = (var_3548_cast_fp16, var_3788_cast_fp16))[name = tensor("op_3870_cast_fp16")]; tensor var_3872_equation_0 = const()[name = tensor("op_3872_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3872_cast_fp16 = einsum(equation = var_3872_equation_0, values = (var_3548_cast_fp16, var_3789_cast_fp16))[name = tensor("op_3872_cast_fp16")]; tensor var_3874_equation_0 = const()[name = tensor("op_3874_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3874_cast_fp16 = einsum(equation = var_3874_equation_0, values = (var_3548_cast_fp16, var_3790_cast_fp16))[name = tensor("op_3874_cast_fp16")]; tensor var_3876_equation_0 = const()[name = tensor("op_3876_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3876_cast_fp16 = einsum(equation = var_3876_equation_0, values = (var_3552_cast_fp16, var_3791_cast_fp16))[name = tensor("op_3876_cast_fp16")]; tensor var_3878_equation_0 = const()[name = tensor("op_3878_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3878_cast_fp16 = einsum(equation = var_3878_equation_0, values = (var_3552_cast_fp16, var_3792_cast_fp16))[name = tensor("op_3878_cast_fp16")]; tensor var_3880_equation_0 = const()[name = tensor("op_3880_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3880_cast_fp16 = einsum(equation = var_3880_equation_0, values = (var_3552_cast_fp16, var_3793_cast_fp16))[name = tensor("op_3880_cast_fp16")]; tensor var_3882_equation_0 = const()[name = tensor("op_3882_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3882_cast_fp16 = einsum(equation = var_3882_equation_0, values = (var_3552_cast_fp16, var_3794_cast_fp16))[name = tensor("op_3882_cast_fp16")]; tensor var_3884_equation_0 = const()[name = tensor("op_3884_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3884_cast_fp16 = einsum(equation = var_3884_equation_0, values = (var_3556_cast_fp16, var_3795_cast_fp16))[name = tensor("op_3884_cast_fp16")]; tensor var_3886_equation_0 = const()[name = tensor("op_3886_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3886_cast_fp16 = einsum(equation = var_3886_equation_0, values = (var_3556_cast_fp16, var_3796_cast_fp16))[name = tensor("op_3886_cast_fp16")]; tensor var_3888_equation_0 = const()[name = tensor("op_3888_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3888_cast_fp16 = einsum(equation = var_3888_equation_0, values = (var_3556_cast_fp16, var_3797_cast_fp16))[name = tensor("op_3888_cast_fp16")]; tensor var_3890_equation_0 = const()[name = tensor("op_3890_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3890_cast_fp16 = einsum(equation = var_3890_equation_0, values = (var_3556_cast_fp16, var_3798_cast_fp16))[name = tensor("op_3890_cast_fp16")]; tensor var_3892_equation_0 = const()[name = tensor("op_3892_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3892_cast_fp16 = einsum(equation = var_3892_equation_0, values = (var_3560_cast_fp16, var_3799_cast_fp16))[name = tensor("op_3892_cast_fp16")]; tensor var_3894_equation_0 = const()[name = tensor("op_3894_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3894_cast_fp16 = einsum(equation = var_3894_equation_0, values = (var_3560_cast_fp16, var_3800_cast_fp16))[name = tensor("op_3894_cast_fp16")]; tensor var_3896_equation_0 = const()[name = tensor("op_3896_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3896_cast_fp16 = einsum(equation = var_3896_equation_0, values = (var_3560_cast_fp16, var_3801_cast_fp16))[name = tensor("op_3896_cast_fp16")]; tensor var_3898_equation_0 = const()[name = tensor("op_3898_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_3898_cast_fp16 = einsum(equation = var_3898_equation_0, values = (var_3560_cast_fp16, var_3802_cast_fp16))[name = tensor("op_3898_cast_fp16")]; tensor var_3900_interleave_0 = const()[name = tensor("op_3900_interleave_0"), val = tensor(false)]; tensor var_3900_cast_fp16 = concat(axis = var_3011, interleave = var_3900_interleave_0, values = (var_3804_cast_fp16, var_3806_cast_fp16, var_3808_cast_fp16, var_3810_cast_fp16))[name = tensor("op_3900_cast_fp16")]; tensor var_3902_interleave_0 = const()[name = tensor("op_3902_interleave_0"), val = tensor(false)]; tensor var_3902_cast_fp16 = concat(axis = var_3011, interleave = var_3902_interleave_0, values = (var_3812_cast_fp16, var_3814_cast_fp16, var_3816_cast_fp16, var_3818_cast_fp16))[name = tensor("op_3902_cast_fp16")]; tensor var_3904_interleave_0 = const()[name = tensor("op_3904_interleave_0"), val = tensor(false)]; tensor var_3904_cast_fp16 = concat(axis = var_3011, interleave = var_3904_interleave_0, values = (var_3820_cast_fp16, var_3822_cast_fp16, var_3824_cast_fp16, var_3826_cast_fp16))[name = tensor("op_3904_cast_fp16")]; tensor var_3906_interleave_0 = const()[name = tensor("op_3906_interleave_0"), val = tensor(false)]; tensor var_3906_cast_fp16 = concat(axis = var_3011, interleave = var_3906_interleave_0, values = (var_3828_cast_fp16, var_3830_cast_fp16, var_3832_cast_fp16, var_3834_cast_fp16))[name = tensor("op_3906_cast_fp16")]; tensor var_3908_interleave_0 = const()[name = tensor("op_3908_interleave_0"), val = tensor(false)]; tensor var_3908_cast_fp16 = concat(axis = var_3011, interleave = var_3908_interleave_0, values = (var_3836_cast_fp16, var_3838_cast_fp16, var_3840_cast_fp16, var_3842_cast_fp16))[name = tensor("op_3908_cast_fp16")]; tensor var_3910_interleave_0 = const()[name = tensor("op_3910_interleave_0"), val = tensor(false)]; tensor var_3910_cast_fp16 = concat(axis = var_3011, interleave = var_3910_interleave_0, values = (var_3844_cast_fp16, var_3846_cast_fp16, var_3848_cast_fp16, var_3850_cast_fp16))[name = tensor("op_3910_cast_fp16")]; tensor var_3912_interleave_0 = const()[name = tensor("op_3912_interleave_0"), val = tensor(false)]; tensor var_3912_cast_fp16 = concat(axis = var_3011, interleave = var_3912_interleave_0, values = (var_3852_cast_fp16, var_3854_cast_fp16, var_3856_cast_fp16, var_3858_cast_fp16))[name = tensor("op_3912_cast_fp16")]; tensor var_3914_interleave_0 = const()[name = tensor("op_3914_interleave_0"), val = tensor(false)]; tensor var_3914_cast_fp16 = concat(axis = var_3011, interleave = var_3914_interleave_0, values = (var_3860_cast_fp16, var_3862_cast_fp16, var_3864_cast_fp16, var_3866_cast_fp16))[name = tensor("op_3914_cast_fp16")]; tensor var_3916_interleave_0 = const()[name = tensor("op_3916_interleave_0"), val = tensor(false)]; tensor var_3916_cast_fp16 = concat(axis = var_3011, interleave = var_3916_interleave_0, values = (var_3868_cast_fp16, var_3870_cast_fp16, var_3872_cast_fp16, var_3874_cast_fp16))[name = tensor("op_3916_cast_fp16")]; tensor var_3918_interleave_0 = const()[name = tensor("op_3918_interleave_0"), val = tensor(false)]; tensor var_3918_cast_fp16 = concat(axis = var_3011, interleave = var_3918_interleave_0, values = (var_3876_cast_fp16, var_3878_cast_fp16, var_3880_cast_fp16, var_3882_cast_fp16))[name = tensor("op_3918_cast_fp16")]; tensor var_3920_interleave_0 = const()[name = tensor("op_3920_interleave_0"), val = tensor(false)]; tensor var_3920_cast_fp16 = concat(axis = var_3011, interleave = var_3920_interleave_0, values = (var_3884_cast_fp16, var_3886_cast_fp16, var_3888_cast_fp16, var_3890_cast_fp16))[name = tensor("op_3920_cast_fp16")]; tensor var_3922_interleave_0 = const()[name = tensor("op_3922_interleave_0"), val = tensor(false)]; tensor var_3922_cast_fp16 = concat(axis = var_3011, interleave = var_3922_interleave_0, values = (var_3892_cast_fp16, var_3894_cast_fp16, var_3896_cast_fp16, var_3898_cast_fp16))[name = tensor("op_3922_cast_fp16")]; tensor input_25_interleave_0 = const()[name = tensor("input_25_interleave_0"), val = tensor(false)]; tensor input_25_cast_fp16 = concat(axis = var_3028, interleave = input_25_interleave_0, values = (var_3900_cast_fp16, var_3902_cast_fp16, var_3904_cast_fp16, var_3906_cast_fp16, var_3908_cast_fp16, var_3910_cast_fp16, var_3912_cast_fp16, var_3914_cast_fp16, var_3916_cast_fp16, var_3918_cast_fp16, var_3920_cast_fp16, var_3922_cast_fp16))[name = tensor("input_25_cast_fp16")]; tensor var_3927 = const()[name = tensor("op_3927"), val = tensor([1, 1])]; tensor var_3929 = const()[name = tensor("op_3929"), val = tensor([1, 1])]; tensor obj_15_pad_type_0 = const()[name = tensor("obj_15_pad_type_0"), val = tensor("custom")]; tensor obj_15_pad_0 = const()[name = tensor("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52289280)))]; tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53468992)))]; tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = var_3929, groups = var_3028, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = var_3927, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_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_3935 = const()[name = tensor("op_3935"), val = tensor([1])]; tensor channels_mean_15_cast_fp16 = reduce_mean(axes = var_3935, keep_dims = var_3029, 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_3939 = const()[name = tensor("op_3939"), val = tensor([1])]; tensor var_3940_cast_fp16 = reduce_mean(axes = var_3939, keep_dims = var_3029, x = zero_mean_sq_15_cast_fp16)[name = tensor("op_3940_cast_fp16")]; tensor var_3941_to_fp16 = const()[name = tensor("op_3941_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_3942_cast_fp16 = add(x = var_3940_cast_fp16, y = var_3941_to_fp16)[name = tensor("op_3942_cast_fp16")]; tensor denom_15_epsilon_0_to_fp16 = const()[name = tensor("denom_15_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_15_cast_fp16 = rsqrt(epsilon = denom_15_epsilon_0_to_fp16, x = var_3942_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_27_gamma_0_to_fp16 = const()[name = tensor("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53470592)))]; tensor input_27_beta_0_to_fp16 = const()[name = tensor("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53472192)))]; tensor input_27_epsilon_0_to_fp16 = const()[name = tensor("input_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_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_27_cast_fp16")]; tensor var_3953 = const()[name = tensor("op_3953"), val = tensor([1, 1])]; tensor var_3955 = const()[name = tensor("op_3955"), 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_3_fc1_weight_to_fp16 = const()[name = tensor("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53473792)))]; tensor layers_3_fc1_bias_to_fp16 = const()[name = tensor("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58192448)))]; tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = var_3955, groups = var_3028, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = var_3953, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; tensor input_31_mode_0 = const()[name = tensor("input_31_mode_0"), val = tensor("EXACT")]; tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = tensor("input_31_cast_fp16")]; tensor var_3961 = const()[name = tensor("op_3961"), val = tensor([1, 1])]; tensor var_3963 = const()[name = tensor("op_3963"), val = tensor([1, 1])]; tensor hidden_states_11_pad_type_0 = const()[name = tensor("hidden_states_11_pad_type_0"), val = tensor("custom")]; tensor hidden_states_11_pad_0 = const()[name = tensor("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_3_fc2_weight_to_fp16 = const()[name = tensor("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58198656)))]; tensor layers_3_fc2_bias_to_fp16 = const()[name = tensor("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62917312)))]; tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = var_3963, groups = var_3028, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = var_3961, weight = layers_3_fc2_weight_to_fp16, x = input_31_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_3970 = const()[name = tensor("op_3970"), val = tensor(3)]; tensor var_3987 = const()[name = tensor("op_3987"), val = tensor(1)]; tensor var_3988 = const()[name = tensor("op_3988"), val = tensor(true)]; tensor var_3998 = const()[name = tensor("op_3998"), val = tensor([1])]; tensor channels_mean_17_cast_fp16 = reduce_mean(axes = var_3998, keep_dims = var_3988, 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_4002 = const()[name = tensor("op_4002"), val = tensor([1])]; tensor var_4003_cast_fp16 = reduce_mean(axes = var_4002, keep_dims = var_3988, x = zero_mean_sq_17_cast_fp16)[name = tensor("op_4003_cast_fp16")]; tensor var_4004_to_fp16 = const()[name = tensor("op_4004_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4005_cast_fp16 = add(x = var_4003_cast_fp16, y = var_4004_to_fp16)[name = tensor("op_4005_cast_fp16")]; tensor denom_17_epsilon_0_to_fp16 = const()[name = tensor("denom_17_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_17_cast_fp16 = rsqrt(epsilon = denom_17_epsilon_0_to_fp16, x = var_4005_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(62918912)))]; 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(62920512)))]; 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_4020 = const()[name = tensor("op_4020"), val = tensor([1, 1])]; tensor var_4022 = const()[name = tensor("op_4022"), val = tensor([1, 1])]; tensor query_9_pad_type_0 = const()[name = tensor("query_9_pad_type_0"), val = tensor("custom")]; tensor query_9_pad_0 = const()[name = tensor("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62922112)))]; tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64101824)))]; tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = var_4022, groups = var_3987, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = var_4020, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("query_9_cast_fp16")]; tensor var_4026 = const()[name = tensor("op_4026"), val = tensor([1, 1])]; tensor var_4028 = const()[name = tensor("op_4028"), val = tensor([1, 1])]; tensor key_9_pad_type_0 = const()[name = tensor("key_9_pad_type_0"), val = tensor("custom")]; tensor key_9_pad_0 = const()[name = tensor("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64103424)))]; tensor key_9_cast_fp16 = conv(dilations = var_4028, groups = var_3987, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = var_4026, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("key_9_cast_fp16")]; tensor var_4033 = const()[name = tensor("op_4033"), val = tensor([1, 1])]; tensor var_4035 = const()[name = tensor("op_4035"), val = tensor([1, 1])]; tensor value_9_pad_type_0 = const()[name = tensor("value_9_pad_type_0"), val = tensor("custom")]; tensor value_9_pad_0 = const()[name = tensor("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65283136)))]; tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66462848)))]; tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = var_4035, groups = var_3987, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = var_4033, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = tensor("value_9_cast_fp16")]; tensor var_4042_begin_0 = const()[name = tensor("op_4042_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4042_end_0 = const()[name = tensor("op_4042_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4042_end_mask_0 = const()[name = tensor("op_4042_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4042_cast_fp16 = slice_by_index(begin = var_4042_begin_0, end = var_4042_end_0, end_mask = var_4042_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4042_cast_fp16")]; tensor var_4046_begin_0 = const()[name = tensor("op_4046_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_4046_end_0 = const()[name = tensor("op_4046_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_4046_end_mask_0 = const()[name = tensor("op_4046_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4046_cast_fp16 = slice_by_index(begin = var_4046_begin_0, end = var_4046_end_0, end_mask = var_4046_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4046_cast_fp16")]; tensor var_4050_begin_0 = const()[name = tensor("op_4050_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_4050_end_0 = const()[name = tensor("op_4050_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_4050_end_mask_0 = const()[name = tensor("op_4050_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4050_cast_fp16 = slice_by_index(begin = var_4050_begin_0, end = var_4050_end_0, end_mask = var_4050_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4050_cast_fp16")]; tensor var_4054_begin_0 = const()[name = tensor("op_4054_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_4054_end_0 = const()[name = tensor("op_4054_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_4054_end_mask_0 = const()[name = tensor("op_4054_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4054_cast_fp16 = slice_by_index(begin = var_4054_begin_0, end = var_4054_end_0, end_mask = var_4054_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4054_cast_fp16")]; tensor var_4058_begin_0 = const()[name = tensor("op_4058_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_4058_end_0 = const()[name = tensor("op_4058_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_4058_end_mask_0 = const()[name = tensor("op_4058_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4058_cast_fp16 = slice_by_index(begin = var_4058_begin_0, end = var_4058_end_0, end_mask = var_4058_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4058_cast_fp16")]; tensor var_4062_begin_0 = const()[name = tensor("op_4062_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4062_end_0 = const()[name = tensor("op_4062_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_4062_end_mask_0 = const()[name = tensor("op_4062_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_9_cast_fp16)[name = tensor("op_4062_cast_fp16")]; tensor var_4066_begin_0 = const()[name = tensor("op_4066_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_4066_end_0 = const()[name = tensor("op_4066_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_4066_end_mask_0 = const()[name = tensor("op_4066_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4066_cast_fp16 = slice_by_index(begin = var_4066_begin_0, end = var_4066_end_0, end_mask = var_4066_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4066_cast_fp16")]; tensor var_4070_begin_0 = const()[name = tensor("op_4070_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_4070_end_0 = const()[name = tensor("op_4070_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_4070_end_mask_0 = const()[name = tensor("op_4070_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4070_cast_fp16 = slice_by_index(begin = var_4070_begin_0, end = var_4070_end_0, end_mask = var_4070_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4070_cast_fp16")]; tensor var_4074_begin_0 = const()[name = tensor("op_4074_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_4074_end_0 = const()[name = tensor("op_4074_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_4074_end_mask_0 = const()[name = tensor("op_4074_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4074_cast_fp16 = slice_by_index(begin = var_4074_begin_0, end = var_4074_end_0, end_mask = var_4074_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4074_cast_fp16")]; tensor var_4078_begin_0 = const()[name = tensor("op_4078_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_4078_end_0 = const()[name = tensor("op_4078_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_4078_end_mask_0 = const()[name = tensor("op_4078_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4078_cast_fp16 = slice_by_index(begin = var_4078_begin_0, end = var_4078_end_0, end_mask = var_4078_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4078_cast_fp16")]; tensor var_4082_begin_0 = const()[name = tensor("op_4082_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_4082_end_0 = const()[name = tensor("op_4082_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_4082_end_mask_0 = const()[name = tensor("op_4082_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4082_cast_fp16 = slice_by_index(begin = var_4082_begin_0, end = var_4082_end_0, end_mask = var_4082_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4082_cast_fp16")]; tensor var_4086_begin_0 = const()[name = tensor("op_4086_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_4086_end_0 = const()[name = tensor("op_4086_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_4086_end_mask_0 = const()[name = tensor("op_4086_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4086_cast_fp16 = slice_by_index(begin = var_4086_begin_0, end = var_4086_end_0, end_mask = var_4086_end_mask_0, x = query_9_cast_fp16)[name = tensor("op_4086_cast_fp16")]; tensor var_4095_begin_0 = const()[name = tensor("op_4095_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4095_end_0 = const()[name = tensor("op_4095_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4095_end_mask_0 = const()[name = tensor("op_4095_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4095_cast_fp16 = slice_by_index(begin = var_4095_begin_0, end = var_4095_end_0, end_mask = var_4095_end_mask_0, x = var_4042_cast_fp16)[name = tensor("op_4095_cast_fp16")]; tensor var_4102_begin_0 = const()[name = tensor("op_4102_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4102_end_0 = const()[name = tensor("op_4102_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4102_end_mask_0 = const()[name = tensor("op_4102_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4102_cast_fp16 = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = var_4042_cast_fp16)[name = tensor("op_4102_cast_fp16")]; tensor var_4109_begin_0 = const()[name = tensor("op_4109_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4109_end_0 = const()[name = tensor("op_4109_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4109_end_mask_0 = const()[name = tensor("op_4109_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4109_cast_fp16 = slice_by_index(begin = var_4109_begin_0, end = var_4109_end_0, end_mask = var_4109_end_mask_0, x = var_4042_cast_fp16)[name = tensor("op_4109_cast_fp16")]; tensor var_4116_begin_0 = const()[name = tensor("op_4116_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4116_end_0 = const()[name = tensor("op_4116_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4116_end_mask_0 = const()[name = tensor("op_4116_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4116_cast_fp16 = slice_by_index(begin = var_4116_begin_0, end = var_4116_end_0, end_mask = var_4116_end_mask_0, x = var_4042_cast_fp16)[name = tensor("op_4116_cast_fp16")]; tensor var_4123_begin_0 = const()[name = tensor("op_4123_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4123_end_0 = const()[name = tensor("op_4123_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4123_end_mask_0 = const()[name = tensor("op_4123_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4123_cast_fp16 = slice_by_index(begin = var_4123_begin_0, end = var_4123_end_0, end_mask = var_4123_end_mask_0, x = var_4046_cast_fp16)[name = tensor("op_4123_cast_fp16")]; tensor var_4130_begin_0 = const()[name = tensor("op_4130_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4130_end_0 = const()[name = tensor("op_4130_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4130_end_mask_0 = const()[name = tensor("op_4130_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4130_cast_fp16 = slice_by_index(begin = var_4130_begin_0, end = var_4130_end_0, end_mask = var_4130_end_mask_0, x = var_4046_cast_fp16)[name = tensor("op_4130_cast_fp16")]; tensor var_4137_begin_0 = const()[name = tensor("op_4137_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4137_end_0 = const()[name = tensor("op_4137_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4137_end_mask_0 = const()[name = tensor("op_4137_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4137_cast_fp16 = slice_by_index(begin = var_4137_begin_0, end = var_4137_end_0, end_mask = var_4137_end_mask_0, x = var_4046_cast_fp16)[name = tensor("op_4137_cast_fp16")]; tensor var_4144_begin_0 = const()[name = tensor("op_4144_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4144_end_0 = const()[name = tensor("op_4144_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4144_end_mask_0 = const()[name = tensor("op_4144_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4144_cast_fp16 = slice_by_index(begin = var_4144_begin_0, end = var_4144_end_0, end_mask = var_4144_end_mask_0, x = var_4046_cast_fp16)[name = tensor("op_4144_cast_fp16")]; tensor var_4151_begin_0 = const()[name = tensor("op_4151_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4151_end_0 = const()[name = tensor("op_4151_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4151_end_mask_0 = const()[name = tensor("op_4151_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4151_cast_fp16 = slice_by_index(begin = var_4151_begin_0, end = var_4151_end_0, end_mask = var_4151_end_mask_0, x = var_4050_cast_fp16)[name = tensor("op_4151_cast_fp16")]; tensor var_4158_begin_0 = const()[name = tensor("op_4158_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4158_end_0 = const()[name = tensor("op_4158_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4158_end_mask_0 = const()[name = tensor("op_4158_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4158_cast_fp16 = slice_by_index(begin = var_4158_begin_0, end = var_4158_end_0, end_mask = var_4158_end_mask_0, x = var_4050_cast_fp16)[name = tensor("op_4158_cast_fp16")]; tensor var_4165_begin_0 = const()[name = tensor("op_4165_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4165_end_0 = const()[name = tensor("op_4165_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4165_end_mask_0 = const()[name = tensor("op_4165_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4165_cast_fp16 = slice_by_index(begin = var_4165_begin_0, end = var_4165_end_0, end_mask = var_4165_end_mask_0, x = var_4050_cast_fp16)[name = tensor("op_4165_cast_fp16")]; tensor var_4172_begin_0 = const()[name = tensor("op_4172_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4172_end_0 = const()[name = tensor("op_4172_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4172_end_mask_0 = const()[name = tensor("op_4172_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4172_cast_fp16 = slice_by_index(begin = var_4172_begin_0, end = var_4172_end_0, end_mask = var_4172_end_mask_0, x = var_4050_cast_fp16)[name = tensor("op_4172_cast_fp16")]; tensor var_4179_begin_0 = const()[name = tensor("op_4179_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4179_end_0 = const()[name = tensor("op_4179_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4179_end_mask_0 = const()[name = tensor("op_4179_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4179_cast_fp16 = slice_by_index(begin = var_4179_begin_0, end = var_4179_end_0, end_mask = var_4179_end_mask_0, x = var_4054_cast_fp16)[name = tensor("op_4179_cast_fp16")]; tensor var_4186_begin_0 = const()[name = tensor("op_4186_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4186_end_0 = const()[name = tensor("op_4186_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4186_end_mask_0 = const()[name = tensor("op_4186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4186_cast_fp16 = slice_by_index(begin = var_4186_begin_0, end = var_4186_end_0, end_mask = var_4186_end_mask_0, x = var_4054_cast_fp16)[name = tensor("op_4186_cast_fp16")]; tensor var_4193_begin_0 = const()[name = tensor("op_4193_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4193_end_0 = const()[name = tensor("op_4193_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4193_end_mask_0 = const()[name = tensor("op_4193_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4193_cast_fp16 = slice_by_index(begin = var_4193_begin_0, end = var_4193_end_0, end_mask = var_4193_end_mask_0, x = var_4054_cast_fp16)[name = tensor("op_4193_cast_fp16")]; tensor var_4200_begin_0 = const()[name = tensor("op_4200_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4200_end_0 = const()[name = tensor("op_4200_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4200_end_mask_0 = const()[name = tensor("op_4200_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4200_cast_fp16 = slice_by_index(begin = var_4200_begin_0, end = var_4200_end_0, end_mask = var_4200_end_mask_0, x = var_4054_cast_fp16)[name = tensor("op_4200_cast_fp16")]; tensor var_4207_begin_0 = const()[name = tensor("op_4207_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4207_end_0 = const()[name = tensor("op_4207_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4207_end_mask_0 = const()[name = tensor("op_4207_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4207_cast_fp16 = slice_by_index(begin = var_4207_begin_0, end = var_4207_end_0, end_mask = var_4207_end_mask_0, x = var_4058_cast_fp16)[name = tensor("op_4207_cast_fp16")]; tensor var_4214_begin_0 = const()[name = tensor("op_4214_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4214_end_0 = const()[name = tensor("op_4214_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4214_end_mask_0 = const()[name = tensor("op_4214_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4214_cast_fp16 = slice_by_index(begin = var_4214_begin_0, end = var_4214_end_0, end_mask = var_4214_end_mask_0, x = var_4058_cast_fp16)[name = tensor("op_4214_cast_fp16")]; tensor var_4221_begin_0 = const()[name = tensor("op_4221_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4221_end_0 = const()[name = tensor("op_4221_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4221_end_mask_0 = const()[name = tensor("op_4221_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_4058_cast_fp16)[name = tensor("op_4221_cast_fp16")]; tensor var_4228_begin_0 = const()[name = tensor("op_4228_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4228_end_0 = const()[name = tensor("op_4228_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4228_end_mask_0 = const()[name = tensor("op_4228_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4228_cast_fp16 = slice_by_index(begin = var_4228_begin_0, end = var_4228_end_0, end_mask = var_4228_end_mask_0, x = var_4058_cast_fp16)[name = tensor("op_4228_cast_fp16")]; tensor var_4235_begin_0 = const()[name = tensor("op_4235_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4235_end_0 = const()[name = tensor("op_4235_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4235_end_mask_0 = const()[name = tensor("op_4235_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4235_cast_fp16 = slice_by_index(begin = var_4235_begin_0, end = var_4235_end_0, end_mask = var_4235_end_mask_0, x = var_4062_cast_fp16)[name = tensor("op_4235_cast_fp16")]; tensor var_4242_begin_0 = const()[name = tensor("op_4242_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4242_end_0 = const()[name = tensor("op_4242_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4242_end_mask_0 = const()[name = tensor("op_4242_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4242_cast_fp16 = slice_by_index(begin = var_4242_begin_0, end = var_4242_end_0, end_mask = var_4242_end_mask_0, x = var_4062_cast_fp16)[name = tensor("op_4242_cast_fp16")]; tensor var_4249_begin_0 = const()[name = tensor("op_4249_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4249_end_0 = const()[name = tensor("op_4249_end_0"), val = tensor([1, 64, 1, 1125])]; 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 = var_4062_cast_fp16)[name = tensor("op_4249_cast_fp16")]; tensor var_4256_begin_0 = const()[name = tensor("op_4256_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4256_end_0 = const()[name = tensor("op_4256_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4256_end_mask_0 = const()[name = tensor("op_4256_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4256_cast_fp16 = slice_by_index(begin = var_4256_begin_0, end = var_4256_end_0, end_mask = var_4256_end_mask_0, x = var_4062_cast_fp16)[name = tensor("op_4256_cast_fp16")]; tensor var_4263_begin_0 = const()[name = tensor("op_4263_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4263_end_0 = const()[name = tensor("op_4263_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4263_end_mask_0 = const()[name = tensor("op_4263_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4263_cast_fp16 = slice_by_index(begin = var_4263_begin_0, end = var_4263_end_0, end_mask = var_4263_end_mask_0, x = var_4066_cast_fp16)[name = tensor("op_4263_cast_fp16")]; tensor var_4270_begin_0 = const()[name = tensor("op_4270_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4270_end_0 = const()[name = tensor("op_4270_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4270_end_mask_0 = const()[name = tensor("op_4270_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4270_cast_fp16 = slice_by_index(begin = var_4270_begin_0, end = var_4270_end_0, end_mask = var_4270_end_mask_0, x = var_4066_cast_fp16)[name = tensor("op_4270_cast_fp16")]; tensor var_4277_begin_0 = const()[name = tensor("op_4277_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4277_end_0 = const()[name = tensor("op_4277_end_0"), val = tensor([1, 64, 1, 1125])]; 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 = var_4066_cast_fp16)[name = tensor("op_4277_cast_fp16")]; tensor var_4284_begin_0 = const()[name = tensor("op_4284_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4284_end_0 = const()[name = tensor("op_4284_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4284_end_mask_0 = const()[name = tensor("op_4284_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4284_cast_fp16 = slice_by_index(begin = var_4284_begin_0, end = var_4284_end_0, end_mask = var_4284_end_mask_0, x = var_4066_cast_fp16)[name = tensor("op_4284_cast_fp16")]; tensor var_4291_begin_0 = const()[name = tensor("op_4291_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4291_end_0 = const()[name = tensor("op_4291_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4291_end_mask_0 = const()[name = tensor("op_4291_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4291_cast_fp16 = slice_by_index(begin = var_4291_begin_0, end = var_4291_end_0, end_mask = var_4291_end_mask_0, x = var_4070_cast_fp16)[name = tensor("op_4291_cast_fp16")]; tensor var_4298_begin_0 = const()[name = tensor("op_4298_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4298_end_0 = const()[name = tensor("op_4298_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4298_end_mask_0 = const()[name = tensor("op_4298_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4298_cast_fp16 = slice_by_index(begin = var_4298_begin_0, end = var_4298_end_0, end_mask = var_4298_end_mask_0, x = var_4070_cast_fp16)[name = tensor("op_4298_cast_fp16")]; tensor var_4305_begin_0 = const()[name = tensor("op_4305_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4305_end_0 = const()[name = tensor("op_4305_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4305_end_mask_0 = const()[name = tensor("op_4305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4305_cast_fp16 = slice_by_index(begin = var_4305_begin_0, end = var_4305_end_0, end_mask = var_4305_end_mask_0, x = var_4070_cast_fp16)[name = tensor("op_4305_cast_fp16")]; tensor var_4312_begin_0 = const()[name = tensor("op_4312_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4312_end_0 = const()[name = tensor("op_4312_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4312_end_mask_0 = const()[name = tensor("op_4312_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4312_cast_fp16 = slice_by_index(begin = var_4312_begin_0, end = var_4312_end_0, end_mask = var_4312_end_mask_0, x = var_4070_cast_fp16)[name = tensor("op_4312_cast_fp16")]; tensor var_4319_begin_0 = const()[name = tensor("op_4319_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4319_end_0 = const()[name = tensor("op_4319_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4319_end_mask_0 = const()[name = tensor("op_4319_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_4074_cast_fp16)[name = tensor("op_4319_cast_fp16")]; tensor var_4326_begin_0 = const()[name = tensor("op_4326_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4326_end_0 = const()[name = tensor("op_4326_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4326_end_mask_0 = const()[name = tensor("op_4326_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4326_cast_fp16 = slice_by_index(begin = var_4326_begin_0, end = var_4326_end_0, end_mask = var_4326_end_mask_0, x = var_4074_cast_fp16)[name = tensor("op_4326_cast_fp16")]; tensor var_4333_begin_0 = const()[name = tensor("op_4333_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4333_end_0 = const()[name = tensor("op_4333_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4333_end_mask_0 = const()[name = tensor("op_4333_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4333_cast_fp16 = slice_by_index(begin = var_4333_begin_0, end = var_4333_end_0, end_mask = var_4333_end_mask_0, x = var_4074_cast_fp16)[name = tensor("op_4333_cast_fp16")]; tensor var_4340_begin_0 = const()[name = tensor("op_4340_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4340_end_0 = const()[name = tensor("op_4340_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4340_end_mask_0 = const()[name = tensor("op_4340_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4340_cast_fp16 = slice_by_index(begin = var_4340_begin_0, end = var_4340_end_0, end_mask = var_4340_end_mask_0, x = var_4074_cast_fp16)[name = tensor("op_4340_cast_fp16")]; tensor var_4347_begin_0 = const()[name = tensor("op_4347_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4347_end_0 = const()[name = tensor("op_4347_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4347_end_mask_0 = const()[name = tensor("op_4347_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_4078_cast_fp16)[name = tensor("op_4347_cast_fp16")]; tensor var_4354_begin_0 = const()[name = tensor("op_4354_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4354_end_0 = const()[name = tensor("op_4354_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4354_end_mask_0 = const()[name = tensor("op_4354_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4354_cast_fp16 = slice_by_index(begin = var_4354_begin_0, end = var_4354_end_0, end_mask = var_4354_end_mask_0, x = var_4078_cast_fp16)[name = tensor("op_4354_cast_fp16")]; tensor var_4361_begin_0 = const()[name = tensor("op_4361_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4361_end_0 = const()[name = tensor("op_4361_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4361_end_mask_0 = const()[name = tensor("op_4361_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4361_cast_fp16 = slice_by_index(begin = var_4361_begin_0, end = var_4361_end_0, end_mask = var_4361_end_mask_0, x = var_4078_cast_fp16)[name = tensor("op_4361_cast_fp16")]; tensor var_4368_begin_0 = const()[name = tensor("op_4368_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4368_end_0 = const()[name = tensor("op_4368_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4368_end_mask_0 = const()[name = tensor("op_4368_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4368_cast_fp16 = slice_by_index(begin = var_4368_begin_0, end = var_4368_end_0, end_mask = var_4368_end_mask_0, x = var_4078_cast_fp16)[name = tensor("op_4368_cast_fp16")]; tensor var_4375_begin_0 = const()[name = tensor("op_4375_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4375_end_0 = const()[name = tensor("op_4375_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4375_end_mask_0 = const()[name = tensor("op_4375_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_4082_cast_fp16)[name = tensor("op_4375_cast_fp16")]; tensor var_4382_begin_0 = const()[name = tensor("op_4382_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4382_end_0 = const()[name = tensor("op_4382_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4382_end_mask_0 = const()[name = tensor("op_4382_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4382_cast_fp16 = slice_by_index(begin = var_4382_begin_0, end = var_4382_end_0, end_mask = var_4382_end_mask_0, x = var_4082_cast_fp16)[name = tensor("op_4382_cast_fp16")]; tensor var_4389_begin_0 = const()[name = tensor("op_4389_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4389_end_0 = const()[name = tensor("op_4389_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4389_end_mask_0 = const()[name = tensor("op_4389_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4389_cast_fp16 = slice_by_index(begin = var_4389_begin_0, end = var_4389_end_0, end_mask = var_4389_end_mask_0, x = var_4082_cast_fp16)[name = tensor("op_4389_cast_fp16")]; tensor var_4396_begin_0 = const()[name = tensor("op_4396_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4396_end_0 = const()[name = tensor("op_4396_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4396_end_mask_0 = const()[name = tensor("op_4396_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4396_cast_fp16 = slice_by_index(begin = var_4396_begin_0, end = var_4396_end_0, end_mask = var_4396_end_mask_0, x = var_4082_cast_fp16)[name = tensor("op_4396_cast_fp16")]; tensor var_4403_begin_0 = const()[name = tensor("op_4403_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4403_end_0 = const()[name = tensor("op_4403_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_4403_end_mask_0 = const()[name = tensor("op_4403_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4403_cast_fp16 = slice_by_index(begin = var_4403_begin_0, end = var_4403_end_0, end_mask = var_4403_end_mask_0, x = var_4086_cast_fp16)[name = tensor("op_4403_cast_fp16")]; tensor var_4410_begin_0 = const()[name = tensor("op_4410_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_4410_end_0 = const()[name = tensor("op_4410_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_4410_end_mask_0 = const()[name = tensor("op_4410_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4410_cast_fp16 = slice_by_index(begin = var_4410_begin_0, end = var_4410_end_0, end_mask = var_4410_end_mask_0, x = var_4086_cast_fp16)[name = tensor("op_4410_cast_fp16")]; tensor var_4417_begin_0 = const()[name = tensor("op_4417_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_4417_end_0 = const()[name = tensor("op_4417_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_4417_end_mask_0 = const()[name = tensor("op_4417_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4417_cast_fp16 = slice_by_index(begin = var_4417_begin_0, end = var_4417_end_0, end_mask = var_4417_end_mask_0, x = var_4086_cast_fp16)[name = tensor("op_4417_cast_fp16")]; tensor var_4424_begin_0 = const()[name = tensor("op_4424_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_4424_end_0 = const()[name = tensor("op_4424_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4424_end_mask_0 = const()[name = tensor("op_4424_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4424_cast_fp16 = slice_by_index(begin = var_4424_begin_0, end = var_4424_end_0, end_mask = var_4424_end_mask_0, x = var_4086_cast_fp16)[name = tensor("op_4424_cast_fp16")]; tensor k_9_perm_0 = const()[name = tensor("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_4429_begin_0 = const()[name = tensor("op_4429_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4429_end_0 = const()[name = tensor("op_4429_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_4429_end_mask_0 = const()[name = tensor("op_4429_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_7 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = tensor("transpose_7")]; tensor var_4429_cast_fp16 = slice_by_index(begin = var_4429_begin_0, end = var_4429_end_0, end_mask = var_4429_end_mask_0, x = transpose_7)[name = tensor("op_4429_cast_fp16")]; tensor var_4433_begin_0 = const()[name = tensor("op_4433_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_4433_end_0 = const()[name = tensor("op_4433_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_4433_end_mask_0 = const()[name = tensor("op_4433_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4433_cast_fp16 = slice_by_index(begin = var_4433_begin_0, end = var_4433_end_0, end_mask = var_4433_end_mask_0, x = transpose_7)[name = tensor("op_4433_cast_fp16")]; tensor var_4437_begin_0 = const()[name = tensor("op_4437_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_4437_end_0 = const()[name = tensor("op_4437_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_4437_end_mask_0 = const()[name = tensor("op_4437_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4437_cast_fp16 = slice_by_index(begin = var_4437_begin_0, end = var_4437_end_0, end_mask = var_4437_end_mask_0, x = transpose_7)[name = tensor("op_4437_cast_fp16")]; tensor var_4441_begin_0 = const()[name = tensor("op_4441_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_4441_end_0 = const()[name = tensor("op_4441_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_4441_end_mask_0 = const()[name = tensor("op_4441_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4441_cast_fp16 = slice_by_index(begin = var_4441_begin_0, end = var_4441_end_0, end_mask = var_4441_end_mask_0, x = transpose_7)[name = tensor("op_4441_cast_fp16")]; tensor var_4445_begin_0 = const()[name = tensor("op_4445_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_4445_end_0 = const()[name = tensor("op_4445_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_4445_end_mask_0 = const()[name = tensor("op_4445_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4445_cast_fp16 = slice_by_index(begin = var_4445_begin_0, end = var_4445_end_0, end_mask = var_4445_end_mask_0, x = transpose_7)[name = tensor("op_4445_cast_fp16")]; tensor var_4449_begin_0 = const()[name = tensor("op_4449_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_4449_end_0 = const()[name = tensor("op_4449_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_4449_end_mask_0 = const()[name = tensor("op_4449_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4449_cast_fp16 = slice_by_index(begin = var_4449_begin_0, end = var_4449_end_0, end_mask = var_4449_end_mask_0, x = transpose_7)[name = tensor("op_4449_cast_fp16")]; tensor var_4453_begin_0 = const()[name = tensor("op_4453_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_4453_end_0 = const()[name = tensor("op_4453_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_4453_end_mask_0 = const()[name = tensor("op_4453_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4453_cast_fp16 = slice_by_index(begin = var_4453_begin_0, end = var_4453_end_0, end_mask = var_4453_end_mask_0, x = transpose_7)[name = tensor("op_4453_cast_fp16")]; tensor var_4457_begin_0 = const()[name = tensor("op_4457_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_4457_end_0 = const()[name = tensor("op_4457_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_4457_end_mask_0 = const()[name = tensor("op_4457_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4457_cast_fp16 = slice_by_index(begin = var_4457_begin_0, end = var_4457_end_0, end_mask = var_4457_end_mask_0, x = transpose_7)[name = tensor("op_4457_cast_fp16")]; tensor var_4461_begin_0 = const()[name = tensor("op_4461_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_4461_end_0 = const()[name = tensor("op_4461_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_4461_end_mask_0 = const()[name = tensor("op_4461_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4461_cast_fp16 = slice_by_index(begin = var_4461_begin_0, end = var_4461_end_0, end_mask = var_4461_end_mask_0, x = transpose_7)[name = tensor("op_4461_cast_fp16")]; tensor var_4465_begin_0 = const()[name = tensor("op_4465_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_4465_end_0 = const()[name = tensor("op_4465_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_4465_end_mask_0 = const()[name = tensor("op_4465_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4465_cast_fp16 = slice_by_index(begin = var_4465_begin_0, end = var_4465_end_0, end_mask = var_4465_end_mask_0, x = transpose_7)[name = tensor("op_4465_cast_fp16")]; tensor var_4469_begin_0 = const()[name = tensor("op_4469_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_4469_end_0 = const()[name = tensor("op_4469_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_4469_end_mask_0 = const()[name = tensor("op_4469_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4469_cast_fp16 = slice_by_index(begin = var_4469_begin_0, end = var_4469_end_0, end_mask = var_4469_end_mask_0, x = transpose_7)[name = tensor("op_4469_cast_fp16")]; tensor var_4473_begin_0 = const()[name = tensor("op_4473_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_4473_end_0 = const()[name = tensor("op_4473_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_4473_end_mask_0 = const()[name = tensor("op_4473_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_4473_cast_fp16 = slice_by_index(begin = var_4473_begin_0, end = var_4473_end_0, end_mask = var_4473_end_mask_0, x = transpose_7)[name = tensor("op_4473_cast_fp16")]; tensor var_4475_begin_0 = const()[name = tensor("op_4475_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_4475_end_0 = const()[name = tensor("op_4475_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_4475_end_mask_0 = const()[name = tensor("op_4475_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4475_cast_fp16 = slice_by_index(begin = var_4475_begin_0, end = var_4475_end_0, end_mask = var_4475_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4475_cast_fp16")]; tensor var_4479_begin_0 = const()[name = tensor("op_4479_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_4479_end_0 = const()[name = tensor("op_4479_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_4479_end_mask_0 = const()[name = tensor("op_4479_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4479_cast_fp16 = slice_by_index(begin = var_4479_begin_0, end = var_4479_end_0, end_mask = var_4479_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4479_cast_fp16")]; tensor var_4483_begin_0 = const()[name = tensor("op_4483_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_4483_end_0 = const()[name = tensor("op_4483_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_4483_end_mask_0 = const()[name = tensor("op_4483_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4483_cast_fp16 = slice_by_index(begin = var_4483_begin_0, end = var_4483_end_0, end_mask = var_4483_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4483_cast_fp16")]; tensor var_4487_begin_0 = const()[name = tensor("op_4487_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_4487_end_0 = const()[name = tensor("op_4487_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_4487_end_mask_0 = const()[name = tensor("op_4487_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4487_cast_fp16 = slice_by_index(begin = var_4487_begin_0, end = var_4487_end_0, end_mask = var_4487_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4487_cast_fp16")]; tensor var_4491_begin_0 = const()[name = tensor("op_4491_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_4491_end_0 = const()[name = tensor("op_4491_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_4491_end_mask_0 = const()[name = tensor("op_4491_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4491_cast_fp16 = slice_by_index(begin = var_4491_begin_0, end = var_4491_end_0, end_mask = var_4491_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4491_cast_fp16")]; tensor var_4495_begin_0 = const()[name = tensor("op_4495_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_4495_end_0 = const()[name = tensor("op_4495_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_4495_end_mask_0 = const()[name = tensor("op_4495_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4495_cast_fp16 = slice_by_index(begin = var_4495_begin_0, end = var_4495_end_0, end_mask = var_4495_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4495_cast_fp16")]; tensor var_4499_begin_0 = const()[name = tensor("op_4499_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_4499_end_0 = const()[name = tensor("op_4499_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_4499_end_mask_0 = const()[name = tensor("op_4499_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4499_cast_fp16 = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4499_cast_fp16")]; tensor var_4503_begin_0 = const()[name = tensor("op_4503_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_4503_end_0 = const()[name = tensor("op_4503_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_4503_end_mask_0 = const()[name = tensor("op_4503_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4503_cast_fp16 = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4503_cast_fp16")]; tensor var_4507_begin_0 = const()[name = tensor("op_4507_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_4507_end_0 = const()[name = tensor("op_4507_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_4507_end_mask_0 = const()[name = tensor("op_4507_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4507_cast_fp16 = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4507_cast_fp16")]; tensor var_4511_begin_0 = const()[name = tensor("op_4511_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_4511_end_0 = const()[name = tensor("op_4511_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_4511_end_mask_0 = const()[name = tensor("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4511_cast_fp16")]; tensor var_4515_begin_0 = const()[name = tensor("op_4515_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_4515_end_0 = const()[name = tensor("op_4515_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_4515_end_mask_0 = const()[name = tensor("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4515_cast_fp16")]; tensor var_4519_begin_0 = const()[name = tensor("op_4519_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_4519_end_0 = const()[name = tensor("op_4519_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_4519_end_mask_0 = const()[name = tensor("op_4519_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_4519_cast_fp16 = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = value_9_cast_fp16)[name = tensor("op_4519_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_4429_cast_fp16, var_4095_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_385_cast_fp16 = mul(x = var_4523_cast_fp16, y = var_4524_to_fp16)[name = tensor("aw_chunk_385_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_4429_cast_fp16, var_4102_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_387_cast_fp16 = mul(x = var_4527_cast_fp16, y = var_4528_to_fp16)[name = tensor("aw_chunk_387_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_4429_cast_fp16, var_4109_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_389_cast_fp16 = mul(x = var_4531_cast_fp16, y = var_4532_to_fp16)[name = tensor("aw_chunk_389_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_4429_cast_fp16, var_4116_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_391_cast_fp16 = mul(x = var_4535_cast_fp16, y = var_4536_to_fp16)[name = tensor("aw_chunk_391_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_4433_cast_fp16, var_4123_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_393_cast_fp16 = mul(x = var_4539_cast_fp16, y = var_4540_to_fp16)[name = tensor("aw_chunk_393_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_4433_cast_fp16, var_4130_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_395_cast_fp16 = mul(x = var_4543_cast_fp16, y = var_4544_to_fp16)[name = tensor("aw_chunk_395_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_4433_cast_fp16, var_4137_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_397_cast_fp16 = mul(x = var_4547_cast_fp16, y = var_4548_to_fp16)[name = tensor("aw_chunk_397_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_4433_cast_fp16, var_4144_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_399_cast_fp16 = mul(x = var_4551_cast_fp16, y = var_4552_to_fp16)[name = tensor("aw_chunk_399_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_4437_cast_fp16, var_4151_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_401_cast_fp16 = mul(x = var_4555_cast_fp16, y = var_4556_to_fp16)[name = tensor("aw_chunk_401_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_4437_cast_fp16, var_4158_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_403_cast_fp16 = mul(x = var_4559_cast_fp16, y = var_4560_to_fp16)[name = tensor("aw_chunk_403_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_4437_cast_fp16, var_4165_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_405_cast_fp16 = mul(x = var_4563_cast_fp16, y = var_4564_to_fp16)[name = tensor("aw_chunk_405_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_4437_cast_fp16, var_4172_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_407_cast_fp16 = mul(x = var_4567_cast_fp16, y = var_4568_to_fp16)[name = tensor("aw_chunk_407_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_4441_cast_fp16, var_4179_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_409_cast_fp16 = mul(x = var_4571_cast_fp16, y = var_4572_to_fp16)[name = tensor("aw_chunk_409_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_4441_cast_fp16, var_4186_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_411_cast_fp16 = mul(x = var_4575_cast_fp16, y = var_4576_to_fp16)[name = tensor("aw_chunk_411_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_4441_cast_fp16, var_4193_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_413_cast_fp16 = mul(x = var_4579_cast_fp16, y = var_4580_to_fp16)[name = tensor("aw_chunk_413_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_4441_cast_fp16, var_4200_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_415_cast_fp16 = mul(x = var_4583_cast_fp16, y = var_4584_to_fp16)[name = tensor("aw_chunk_415_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_4445_cast_fp16, var_4207_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_417_cast_fp16 = mul(x = var_4587_cast_fp16, y = var_4588_to_fp16)[name = tensor("aw_chunk_417_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_4445_cast_fp16, var_4214_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_419_cast_fp16 = mul(x = var_4591_cast_fp16, y = var_4592_to_fp16)[name = tensor("aw_chunk_419_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_4445_cast_fp16, var_4221_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_421_cast_fp16 = mul(x = var_4595_cast_fp16, y = var_4596_to_fp16)[name = tensor("aw_chunk_421_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_4445_cast_fp16, var_4228_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_423_cast_fp16 = mul(x = var_4599_cast_fp16, y = var_4600_to_fp16)[name = tensor("aw_chunk_423_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_4449_cast_fp16, var_4235_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_425_cast_fp16 = mul(x = var_4603_cast_fp16, y = var_4604_to_fp16)[name = tensor("aw_chunk_425_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_4449_cast_fp16, var_4242_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_427_cast_fp16 = mul(x = var_4607_cast_fp16, y = var_4608_to_fp16)[name = tensor("aw_chunk_427_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_4449_cast_fp16, var_4249_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_429_cast_fp16 = mul(x = var_4611_cast_fp16, y = var_4612_to_fp16)[name = tensor("aw_chunk_429_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_4449_cast_fp16, var_4256_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_431_cast_fp16 = mul(x = var_4615_cast_fp16, y = var_4616_to_fp16)[name = tensor("aw_chunk_431_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_4453_cast_fp16, var_4263_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_433_cast_fp16 = mul(x = var_4619_cast_fp16, y = var_4620_to_fp16)[name = tensor("aw_chunk_433_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_4453_cast_fp16, var_4270_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_435_cast_fp16 = mul(x = var_4623_cast_fp16, y = var_4624_to_fp16)[name = tensor("aw_chunk_435_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_4453_cast_fp16, var_4277_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_437_cast_fp16 = mul(x = var_4627_cast_fp16, y = var_4628_to_fp16)[name = tensor("aw_chunk_437_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_4453_cast_fp16, var_4284_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_439_cast_fp16 = mul(x = var_4631_cast_fp16, y = var_4632_to_fp16)[name = tensor("aw_chunk_439_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_4457_cast_fp16, var_4291_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_441_cast_fp16 = mul(x = var_4635_cast_fp16, y = var_4636_to_fp16)[name = tensor("aw_chunk_441_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_4457_cast_fp16, var_4298_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_443_cast_fp16 = mul(x = var_4639_cast_fp16, y = var_4640_to_fp16)[name = tensor("aw_chunk_443_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_4457_cast_fp16, var_4305_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_445_cast_fp16 = mul(x = var_4643_cast_fp16, y = var_4644_to_fp16)[name = tensor("aw_chunk_445_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_4457_cast_fp16, var_4312_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_447_cast_fp16 = mul(x = var_4647_cast_fp16, y = var_4648_to_fp16)[name = tensor("aw_chunk_447_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_4461_cast_fp16, var_4319_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_449_cast_fp16 = mul(x = var_4651_cast_fp16, y = var_4652_to_fp16)[name = tensor("aw_chunk_449_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_4461_cast_fp16, var_4326_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_451_cast_fp16 = mul(x = var_4655_cast_fp16, y = var_4656_to_fp16)[name = tensor("aw_chunk_451_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_4461_cast_fp16, var_4333_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_453_cast_fp16 = mul(x = var_4659_cast_fp16, y = var_4660_to_fp16)[name = tensor("aw_chunk_453_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_4461_cast_fp16, var_4340_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_455_cast_fp16 = mul(x = var_4663_cast_fp16, y = var_4664_to_fp16)[name = tensor("aw_chunk_455_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_4465_cast_fp16, var_4347_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_457_cast_fp16 = mul(x = var_4667_cast_fp16, y = var_4668_to_fp16)[name = tensor("aw_chunk_457_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_4465_cast_fp16, var_4354_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_459_cast_fp16 = mul(x = var_4671_cast_fp16, y = var_4672_to_fp16)[name = tensor("aw_chunk_459_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_4465_cast_fp16, var_4361_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_461_cast_fp16 = mul(x = var_4675_cast_fp16, y = var_4676_to_fp16)[name = tensor("aw_chunk_461_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_4465_cast_fp16, var_4368_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_463_cast_fp16 = mul(x = var_4679_cast_fp16, y = var_4680_to_fp16)[name = tensor("aw_chunk_463_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_4469_cast_fp16, var_4375_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_465_cast_fp16 = mul(x = var_4683_cast_fp16, y = var_4684_to_fp16)[name = tensor("aw_chunk_465_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_4469_cast_fp16, var_4382_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_467_cast_fp16 = mul(x = var_4687_cast_fp16, y = var_4688_to_fp16)[name = tensor("aw_chunk_467_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_4469_cast_fp16, var_4389_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_469_cast_fp16 = mul(x = var_4691_cast_fp16, y = var_4692_to_fp16)[name = tensor("aw_chunk_469_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_4469_cast_fp16, var_4396_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_471_cast_fp16 = mul(x = var_4695_cast_fp16, y = var_4696_to_fp16)[name = tensor("aw_chunk_471_cast_fp16")]; tensor var_4699_equation_0 = const()[name = tensor("op_4699_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4699_cast_fp16 = einsum(equation = var_4699_equation_0, values = (var_4473_cast_fp16, var_4403_cast_fp16))[name = tensor("op_4699_cast_fp16")]; tensor var_4700_to_fp16 = const()[name = tensor("op_4700_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_473_cast_fp16 = mul(x = var_4699_cast_fp16, y = var_4700_to_fp16)[name = tensor("aw_chunk_473_cast_fp16")]; tensor var_4703_equation_0 = const()[name = tensor("op_4703_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4703_cast_fp16 = einsum(equation = var_4703_equation_0, values = (var_4473_cast_fp16, var_4410_cast_fp16))[name = tensor("op_4703_cast_fp16")]; tensor var_4704_to_fp16 = const()[name = tensor("op_4704_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_475_cast_fp16 = mul(x = var_4703_cast_fp16, y = var_4704_to_fp16)[name = tensor("aw_chunk_475_cast_fp16")]; tensor var_4707_equation_0 = const()[name = tensor("op_4707_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4707_cast_fp16 = einsum(equation = var_4707_equation_0, values = (var_4473_cast_fp16, var_4417_cast_fp16))[name = tensor("op_4707_cast_fp16")]; tensor var_4708_to_fp16 = const()[name = tensor("op_4708_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_477_cast_fp16 = mul(x = var_4707_cast_fp16, y = var_4708_to_fp16)[name = tensor("aw_chunk_477_cast_fp16")]; tensor var_4711_equation_0 = const()[name = tensor("op_4711_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_4711_cast_fp16 = einsum(equation = var_4711_equation_0, values = (var_4473_cast_fp16, var_4424_cast_fp16))[name = tensor("op_4711_cast_fp16")]; tensor var_4712_to_fp16 = const()[name = tensor("op_4712_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_479_cast_fp16 = mul(x = var_4711_cast_fp16, y = var_4712_to_fp16)[name = tensor("aw_chunk_479_cast_fp16")]; tensor var_4714_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_385_cast_fp16)[name = tensor("op_4714_cast_fp16")]; tensor var_4715_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_387_cast_fp16)[name = tensor("op_4715_cast_fp16")]; tensor var_4716_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_389_cast_fp16)[name = tensor("op_4716_cast_fp16")]; tensor var_4717_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_391_cast_fp16)[name = tensor("op_4717_cast_fp16")]; tensor var_4718_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_393_cast_fp16)[name = tensor("op_4718_cast_fp16")]; tensor var_4719_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_395_cast_fp16)[name = tensor("op_4719_cast_fp16")]; tensor var_4720_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_397_cast_fp16)[name = tensor("op_4720_cast_fp16")]; tensor var_4721_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_399_cast_fp16)[name = tensor("op_4721_cast_fp16")]; tensor var_4722_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_401_cast_fp16)[name = tensor("op_4722_cast_fp16")]; tensor var_4723_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_403_cast_fp16)[name = tensor("op_4723_cast_fp16")]; tensor var_4724_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_405_cast_fp16)[name = tensor("op_4724_cast_fp16")]; tensor var_4725_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_407_cast_fp16)[name = tensor("op_4725_cast_fp16")]; tensor var_4726_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_409_cast_fp16)[name = tensor("op_4726_cast_fp16")]; tensor var_4727_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_411_cast_fp16)[name = tensor("op_4727_cast_fp16")]; tensor var_4728_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_413_cast_fp16)[name = tensor("op_4728_cast_fp16")]; tensor var_4729_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_415_cast_fp16)[name = tensor("op_4729_cast_fp16")]; tensor var_4730_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_417_cast_fp16)[name = tensor("op_4730_cast_fp16")]; tensor var_4731_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_419_cast_fp16)[name = tensor("op_4731_cast_fp16")]; tensor var_4732_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_421_cast_fp16)[name = tensor("op_4732_cast_fp16")]; tensor var_4733_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_423_cast_fp16)[name = tensor("op_4733_cast_fp16")]; tensor var_4734_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_425_cast_fp16)[name = tensor("op_4734_cast_fp16")]; tensor var_4735_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_427_cast_fp16)[name = tensor("op_4735_cast_fp16")]; tensor var_4736_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_429_cast_fp16)[name = tensor("op_4736_cast_fp16")]; tensor var_4737_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_431_cast_fp16)[name = tensor("op_4737_cast_fp16")]; tensor var_4738_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_433_cast_fp16)[name = tensor("op_4738_cast_fp16")]; tensor var_4739_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_435_cast_fp16)[name = tensor("op_4739_cast_fp16")]; tensor var_4740_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_437_cast_fp16)[name = tensor("op_4740_cast_fp16")]; tensor var_4741_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_439_cast_fp16)[name = tensor("op_4741_cast_fp16")]; tensor var_4742_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_441_cast_fp16)[name = tensor("op_4742_cast_fp16")]; tensor var_4743_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_443_cast_fp16)[name = tensor("op_4743_cast_fp16")]; tensor var_4744_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_445_cast_fp16)[name = tensor("op_4744_cast_fp16")]; tensor var_4745_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_447_cast_fp16)[name = tensor("op_4745_cast_fp16")]; tensor var_4746_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_449_cast_fp16)[name = tensor("op_4746_cast_fp16")]; tensor var_4747_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_451_cast_fp16)[name = tensor("op_4747_cast_fp16")]; tensor var_4748_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_453_cast_fp16)[name = tensor("op_4748_cast_fp16")]; tensor var_4749_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_455_cast_fp16)[name = tensor("op_4749_cast_fp16")]; tensor var_4750_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_457_cast_fp16)[name = tensor("op_4750_cast_fp16")]; tensor var_4751_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_459_cast_fp16)[name = tensor("op_4751_cast_fp16")]; tensor var_4752_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_461_cast_fp16)[name = tensor("op_4752_cast_fp16")]; tensor var_4753_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_463_cast_fp16)[name = tensor("op_4753_cast_fp16")]; tensor var_4754_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_465_cast_fp16)[name = tensor("op_4754_cast_fp16")]; tensor var_4755_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_467_cast_fp16)[name = tensor("op_4755_cast_fp16")]; tensor var_4756_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_469_cast_fp16)[name = tensor("op_4756_cast_fp16")]; tensor var_4757_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_471_cast_fp16)[name = tensor("op_4757_cast_fp16")]; tensor var_4758_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_473_cast_fp16)[name = tensor("op_4758_cast_fp16")]; tensor var_4759_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_475_cast_fp16)[name = tensor("op_4759_cast_fp16")]; tensor var_4760_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_477_cast_fp16)[name = tensor("op_4760_cast_fp16")]; tensor var_4761_cast_fp16 = softmax(axis = var_3987, x = aw_chunk_479_cast_fp16)[name = tensor("op_4761_cast_fp16")]; tensor var_4763_equation_0 = const()[name = tensor("op_4763_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4763_cast_fp16 = einsum(equation = var_4763_equation_0, values = (var_4475_cast_fp16, var_4714_cast_fp16))[name = tensor("op_4763_cast_fp16")]; tensor var_4765_equation_0 = const()[name = tensor("op_4765_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4765_cast_fp16 = einsum(equation = var_4765_equation_0, values = (var_4475_cast_fp16, var_4715_cast_fp16))[name = tensor("op_4765_cast_fp16")]; tensor var_4767_equation_0 = const()[name = tensor("op_4767_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4767_cast_fp16 = einsum(equation = var_4767_equation_0, values = (var_4475_cast_fp16, var_4716_cast_fp16))[name = tensor("op_4767_cast_fp16")]; tensor var_4769_equation_0 = const()[name = tensor("op_4769_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4769_cast_fp16 = einsum(equation = var_4769_equation_0, values = (var_4475_cast_fp16, var_4717_cast_fp16))[name = tensor("op_4769_cast_fp16")]; tensor var_4771_equation_0 = const()[name = tensor("op_4771_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4771_cast_fp16 = einsum(equation = var_4771_equation_0, values = (var_4479_cast_fp16, var_4718_cast_fp16))[name = tensor("op_4771_cast_fp16")]; tensor var_4773_equation_0 = const()[name = tensor("op_4773_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4773_cast_fp16 = einsum(equation = var_4773_equation_0, values = (var_4479_cast_fp16, var_4719_cast_fp16))[name = tensor("op_4773_cast_fp16")]; tensor var_4775_equation_0 = const()[name = tensor("op_4775_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4775_cast_fp16 = einsum(equation = var_4775_equation_0, values = (var_4479_cast_fp16, var_4720_cast_fp16))[name = tensor("op_4775_cast_fp16")]; tensor var_4777_equation_0 = const()[name = tensor("op_4777_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_4777_cast_fp16 = einsum(equation = var_4777_equation_0, values = (var_4479_cast_fp16, var_4721_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_4483_cast_fp16, var_4722_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_4483_cast_fp16, var_4723_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_4483_cast_fp16, var_4724_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_4483_cast_fp16, var_4725_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_4487_cast_fp16, var_4726_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_4487_cast_fp16, var_4727_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_4487_cast_fp16, var_4728_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_4487_cast_fp16, var_4729_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_4491_cast_fp16, var_4730_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_4491_cast_fp16, var_4731_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_4491_cast_fp16, var_4732_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_4491_cast_fp16, var_4733_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_4495_cast_fp16, var_4734_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_4495_cast_fp16, var_4735_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_4495_cast_fp16, var_4736_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_4495_cast_fp16, var_4737_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_4499_cast_fp16, var_4738_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_4499_cast_fp16, var_4739_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_4499_cast_fp16, var_4740_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_4499_cast_fp16, var_4741_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_4503_cast_fp16, var_4742_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_4503_cast_fp16, var_4743_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_4503_cast_fp16, var_4744_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_4503_cast_fp16, var_4745_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_4507_cast_fp16, var_4746_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_4507_cast_fp16, var_4747_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_4507_cast_fp16, var_4748_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_4507_cast_fp16, var_4749_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_4511_cast_fp16, var_4750_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_4511_cast_fp16, var_4751_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_4511_cast_fp16, var_4752_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_4511_cast_fp16, var_4753_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_4515_cast_fp16, var_4754_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_4515_cast_fp16, var_4755_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_4515_cast_fp16, var_4756_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_4515_cast_fp16, var_4757_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_4519_cast_fp16, var_4758_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_4519_cast_fp16, var_4759_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_4519_cast_fp16, var_4760_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_4519_cast_fp16, var_4761_cast_fp16))[name = tensor("op_4857_cast_fp16")]; tensor var_4859_interleave_0 = const()[name = tensor("op_4859_interleave_0"), val = tensor(false)]; tensor var_4859_cast_fp16 = concat(axis = var_3970, interleave = var_4859_interleave_0, values = (var_4763_cast_fp16, var_4765_cast_fp16, var_4767_cast_fp16, var_4769_cast_fp16))[name = tensor("op_4859_cast_fp16")]; tensor var_4861_interleave_0 = const()[name = tensor("op_4861_interleave_0"), val = tensor(false)]; tensor var_4861_cast_fp16 = concat(axis = var_3970, interleave = var_4861_interleave_0, values = (var_4771_cast_fp16, var_4773_cast_fp16, var_4775_cast_fp16, var_4777_cast_fp16))[name = tensor("op_4861_cast_fp16")]; tensor var_4863_interleave_0 = const()[name = tensor("op_4863_interleave_0"), val = tensor(false)]; tensor var_4863_cast_fp16 = concat(axis = var_3970, interleave = var_4863_interleave_0, values = (var_4779_cast_fp16, var_4781_cast_fp16, var_4783_cast_fp16, var_4785_cast_fp16))[name = tensor("op_4863_cast_fp16")]; tensor var_4865_interleave_0 = const()[name = tensor("op_4865_interleave_0"), val = tensor(false)]; tensor var_4865_cast_fp16 = concat(axis = var_3970, interleave = var_4865_interleave_0, values = (var_4787_cast_fp16, var_4789_cast_fp16, var_4791_cast_fp16, var_4793_cast_fp16))[name = tensor("op_4865_cast_fp16")]; tensor var_4867_interleave_0 = const()[name = tensor("op_4867_interleave_0"), val = tensor(false)]; tensor var_4867_cast_fp16 = concat(axis = var_3970, interleave = var_4867_interleave_0, values = (var_4795_cast_fp16, var_4797_cast_fp16, var_4799_cast_fp16, var_4801_cast_fp16))[name = tensor("op_4867_cast_fp16")]; tensor var_4869_interleave_0 = const()[name = tensor("op_4869_interleave_0"), val = tensor(false)]; tensor var_4869_cast_fp16 = concat(axis = var_3970, interleave = var_4869_interleave_0, values = (var_4803_cast_fp16, var_4805_cast_fp16, var_4807_cast_fp16, var_4809_cast_fp16))[name = tensor("op_4869_cast_fp16")]; tensor var_4871_interleave_0 = const()[name = tensor("op_4871_interleave_0"), val = tensor(false)]; tensor var_4871_cast_fp16 = concat(axis = var_3970, interleave = var_4871_interleave_0, values = (var_4811_cast_fp16, var_4813_cast_fp16, var_4815_cast_fp16, var_4817_cast_fp16))[name = tensor("op_4871_cast_fp16")]; tensor var_4873_interleave_0 = const()[name = tensor("op_4873_interleave_0"), val = tensor(false)]; tensor var_4873_cast_fp16 = concat(axis = var_3970, interleave = var_4873_interleave_0, values = (var_4819_cast_fp16, var_4821_cast_fp16, var_4823_cast_fp16, var_4825_cast_fp16))[name = tensor("op_4873_cast_fp16")]; tensor var_4875_interleave_0 = const()[name = tensor("op_4875_interleave_0"), val = tensor(false)]; tensor var_4875_cast_fp16 = concat(axis = var_3970, interleave = var_4875_interleave_0, values = (var_4827_cast_fp16, var_4829_cast_fp16, var_4831_cast_fp16, var_4833_cast_fp16))[name = tensor("op_4875_cast_fp16")]; tensor var_4877_interleave_0 = const()[name = tensor("op_4877_interleave_0"), val = tensor(false)]; tensor var_4877_cast_fp16 = concat(axis = var_3970, interleave = var_4877_interleave_0, values = (var_4835_cast_fp16, var_4837_cast_fp16, var_4839_cast_fp16, var_4841_cast_fp16))[name = tensor("op_4877_cast_fp16")]; tensor var_4879_interleave_0 = const()[name = tensor("op_4879_interleave_0"), val = tensor(false)]; tensor var_4879_cast_fp16 = concat(axis = var_3970, interleave = var_4879_interleave_0, values = (var_4843_cast_fp16, var_4845_cast_fp16, var_4847_cast_fp16, var_4849_cast_fp16))[name = tensor("op_4879_cast_fp16")]; tensor var_4881_interleave_0 = const()[name = tensor("op_4881_interleave_0"), val = tensor(false)]; tensor var_4881_cast_fp16 = concat(axis = var_3970, interleave = var_4881_interleave_0, values = (var_4851_cast_fp16, var_4853_cast_fp16, var_4855_cast_fp16, var_4857_cast_fp16))[name = tensor("op_4881_cast_fp16")]; tensor input_33_interleave_0 = const()[name = tensor("input_33_interleave_0"), val = tensor(false)]; tensor input_33_cast_fp16 = concat(axis = var_3987, interleave = input_33_interleave_0, values = (var_4859_cast_fp16, var_4861_cast_fp16, var_4863_cast_fp16, var_4865_cast_fp16, var_4867_cast_fp16, var_4869_cast_fp16, var_4871_cast_fp16, var_4873_cast_fp16, var_4875_cast_fp16, var_4877_cast_fp16, var_4879_cast_fp16, var_4881_cast_fp16))[name = tensor("input_33_cast_fp16")]; tensor var_4886 = const()[name = tensor("op_4886"), val = tensor([1, 1])]; tensor var_4888 = const()[name = tensor("op_4888"), val = tensor([1, 1])]; tensor obj_19_pad_type_0 = const()[name = tensor("obj_19_pad_type_0"), val = tensor("custom")]; tensor obj_19_pad_0 = const()[name = tensor("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66464448)))]; tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67644160)))]; tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = var_4888, groups = var_3987, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = var_4886, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_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_4894 = const()[name = tensor("op_4894"), val = tensor([1])]; tensor channels_mean_19_cast_fp16 = reduce_mean(axes = var_4894, keep_dims = var_3988, 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_4898 = const()[name = tensor("op_4898"), val = tensor([1])]; tensor var_4899_cast_fp16 = reduce_mean(axes = var_4898, keep_dims = var_3988, x = zero_mean_sq_19_cast_fp16)[name = tensor("op_4899_cast_fp16")]; tensor var_4900_to_fp16 = const()[name = tensor("op_4900_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4901_cast_fp16 = add(x = var_4899_cast_fp16, y = var_4900_to_fp16)[name = tensor("op_4901_cast_fp16")]; tensor denom_19_epsilon_0_to_fp16 = const()[name = tensor("denom_19_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_19_cast_fp16 = rsqrt(epsilon = denom_19_epsilon_0_to_fp16, x = var_4901_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_35_gamma_0_to_fp16 = const()[name = tensor("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67645760)))]; tensor input_35_beta_0_to_fp16 = const()[name = tensor("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67647360)))]; tensor input_35_epsilon_0_to_fp16 = const()[name = tensor("input_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_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_35_cast_fp16")]; tensor var_4912 = const()[name = tensor("op_4912"), val = tensor([1, 1])]; tensor var_4914 = const()[name = tensor("op_4914"), val = tensor([1, 1])]; tensor input_37_pad_type_0 = const()[name = tensor("input_37_pad_type_0"), val = tensor("custom")]; tensor input_37_pad_0 = const()[name = tensor("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_fc1_weight_to_fp16 = const()[name = tensor("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67648960)))]; tensor layers_4_fc1_bias_to_fp16 = const()[name = tensor("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72367616)))]; tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = var_4914, groups = var_3987, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = var_4912, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor("input_37_cast_fp16")]; tensor input_39_mode_0 = const()[name = tensor("input_39_mode_0"), val = tensor("EXACT")]; tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; tensor var_4920 = const()[name = tensor("op_4920"), val = tensor([1, 1])]; tensor var_4922 = const()[name = tensor("op_4922"), val = tensor([1, 1])]; tensor hidden_states_13_pad_type_0 = const()[name = tensor("hidden_states_13_pad_type_0"), val = tensor("custom")]; tensor hidden_states_13_pad_0 = const()[name = tensor("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_4_fc2_weight_to_fp16 = const()[name = tensor("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72373824)))]; tensor layers_4_fc2_bias_to_fp16 = const()[name = tensor("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77092480)))]; tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = var_4922, groups = var_3987, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = var_4920, weight = layers_4_fc2_weight_to_fp16, x = input_39_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_4929 = const()[name = tensor("op_4929"), val = tensor(3)]; tensor var_4946 = const()[name = tensor("op_4946"), val = tensor(1)]; tensor var_4947 = const()[name = tensor("op_4947"), val = tensor(true)]; tensor var_4957 = const()[name = tensor("op_4957"), val = tensor([1])]; tensor channels_mean_21_cast_fp16 = reduce_mean(axes = var_4957, keep_dims = var_4947, 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_4961 = const()[name = tensor("op_4961"), val = tensor([1])]; tensor var_4962_cast_fp16 = reduce_mean(axes = var_4961, keep_dims = var_4947, x = zero_mean_sq_21_cast_fp16)[name = tensor("op_4962_cast_fp16")]; tensor var_4963_to_fp16 = const()[name = tensor("op_4963_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_4964_cast_fp16 = add(x = var_4962_cast_fp16, y = var_4963_to_fp16)[name = tensor("op_4964_cast_fp16")]; tensor denom_21_epsilon_0_to_fp16 = const()[name = tensor("denom_21_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_21_cast_fp16 = rsqrt(epsilon = denom_21_epsilon_0_to_fp16, x = var_4964_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(77094080)))]; 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(77095680)))]; 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_4979 = const()[name = tensor("op_4979"), val = tensor([1, 1])]; tensor var_4981 = const()[name = tensor("op_4981"), val = tensor([1, 1])]; tensor query_11_pad_type_0 = const()[name = tensor("query_11_pad_type_0"), val = tensor("custom")]; tensor query_11_pad_0 = const()[name = tensor("query_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77097280)))]; tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78276992)))]; tensor query_11_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = var_4981, groups = var_4946, pad = query_11_pad_0, pad_type = query_11_pad_type_0, strides = var_4979, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("query_11_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 key_11_pad_type_0 = const()[name = tensor("key_11_pad_type_0"), val = tensor("custom")]; tensor key_11_pad_0 = const()[name = tensor("key_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78278592)))]; tensor key_11_cast_fp16 = conv(dilations = var_4987, groups = var_4946, pad = key_11_pad_0, pad_type = key_11_pad_type_0, strides = var_4985, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("key_11_cast_fp16")]; tensor var_4992 = const()[name = tensor("op_4992"), val = tensor([1, 1])]; tensor var_4994 = const()[name = tensor("op_4994"), val = tensor([1, 1])]; tensor value_11_pad_type_0 = const()[name = tensor("value_11_pad_type_0"), val = tensor("custom")]; tensor value_11_pad_0 = const()[name = tensor("value_11_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79458304)))]; tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80638016)))]; tensor value_11_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = var_4994, groups = var_4946, pad = value_11_pad_0, pad_type = value_11_pad_type_0, strides = var_4992, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = tensor("value_11_cast_fp16")]; tensor var_5001_begin_0 = const()[name = tensor("op_5001_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5001_end_0 = const()[name = tensor("op_5001_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5001_end_mask_0 = const()[name = tensor("op_5001_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5001_cast_fp16 = slice_by_index(begin = var_5001_begin_0, end = var_5001_end_0, end_mask = var_5001_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5001_cast_fp16")]; tensor var_5005_begin_0 = const()[name = tensor("op_5005_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5005_end_0 = const()[name = tensor("op_5005_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5005_end_mask_0 = const()[name = tensor("op_5005_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5005_cast_fp16 = slice_by_index(begin = var_5005_begin_0, end = var_5005_end_0, end_mask = var_5005_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5005_cast_fp16")]; tensor var_5009_begin_0 = const()[name = tensor("op_5009_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5009_end_0 = const()[name = tensor("op_5009_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5009_end_mask_0 = const()[name = tensor("op_5009_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5009_cast_fp16 = slice_by_index(begin = var_5009_begin_0, end = var_5009_end_0, end_mask = var_5009_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5009_cast_fp16")]; tensor var_5013_begin_0 = const()[name = tensor("op_5013_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5013_end_0 = const()[name = tensor("op_5013_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5013_end_mask_0 = const()[name = tensor("op_5013_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5013_cast_fp16 = slice_by_index(begin = var_5013_begin_0, end = var_5013_end_0, end_mask = var_5013_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5013_cast_fp16")]; tensor var_5017_begin_0 = const()[name = tensor("op_5017_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5017_end_0 = const()[name = tensor("op_5017_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_5017_end_mask_0 = const()[name = tensor("op_5017_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5017_cast_fp16 = slice_by_index(begin = var_5017_begin_0, end = var_5017_end_0, end_mask = var_5017_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5017_cast_fp16")]; tensor var_5021_begin_0 = const()[name = tensor("op_5021_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5021_end_0 = const()[name = tensor("op_5021_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_5021_end_mask_0 = const()[name = tensor("op_5021_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5021_cast_fp16 = slice_by_index(begin = var_5021_begin_0, end = var_5021_end_0, end_mask = var_5021_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5021_cast_fp16")]; tensor var_5025_begin_0 = const()[name = tensor("op_5025_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5025_end_0 = const()[name = tensor("op_5025_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_5025_end_mask_0 = const()[name = tensor("op_5025_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5025_cast_fp16 = slice_by_index(begin = var_5025_begin_0, end = var_5025_end_0, end_mask = var_5025_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5025_cast_fp16")]; tensor var_5029_begin_0 = const()[name = tensor("op_5029_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5029_end_0 = const()[name = tensor("op_5029_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_5029_end_mask_0 = const()[name = tensor("op_5029_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5029_cast_fp16 = slice_by_index(begin = var_5029_begin_0, end = var_5029_end_0, end_mask = var_5029_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5029_cast_fp16")]; tensor var_5033_begin_0 = const()[name = tensor("op_5033_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5033_end_0 = const()[name = tensor("op_5033_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_5033_end_mask_0 = const()[name = tensor("op_5033_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5033_cast_fp16 = slice_by_index(begin = var_5033_begin_0, end = var_5033_end_0, end_mask = var_5033_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5033_cast_fp16")]; tensor var_5037_begin_0 = const()[name = tensor("op_5037_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5037_end_0 = const()[name = tensor("op_5037_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_5037_end_mask_0 = const()[name = tensor("op_5037_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5037_cast_fp16 = slice_by_index(begin = var_5037_begin_0, end = var_5037_end_0, end_mask = var_5037_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5037_cast_fp16")]; tensor var_5041_begin_0 = const()[name = tensor("op_5041_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5041_end_0 = const()[name = tensor("op_5041_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_5041_end_mask_0 = const()[name = tensor("op_5041_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5041_cast_fp16 = slice_by_index(begin = var_5041_begin_0, end = var_5041_end_0, end_mask = var_5041_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5041_cast_fp16")]; tensor var_5045_begin_0 = const()[name = tensor("op_5045_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5045_end_0 = const()[name = tensor("op_5045_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_5045_end_mask_0 = const()[name = tensor("op_5045_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5045_cast_fp16 = slice_by_index(begin = var_5045_begin_0, end = var_5045_end_0, end_mask = var_5045_end_mask_0, x = query_11_cast_fp16)[name = tensor("op_5045_cast_fp16")]; tensor var_5054_begin_0 = const()[name = tensor("op_5054_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5054_end_0 = const()[name = tensor("op_5054_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5054_end_mask_0 = const()[name = tensor("op_5054_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5054_cast_fp16 = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = var_5001_cast_fp16)[name = tensor("op_5054_cast_fp16")]; tensor var_5061_begin_0 = const()[name = tensor("op_5061_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5061_end_0 = const()[name = tensor("op_5061_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5061_end_mask_0 = const()[name = tensor("op_5061_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5061_cast_fp16 = slice_by_index(begin = var_5061_begin_0, end = var_5061_end_0, end_mask = var_5061_end_mask_0, x = var_5001_cast_fp16)[name = tensor("op_5061_cast_fp16")]; tensor var_5068_begin_0 = const()[name = tensor("op_5068_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5068_end_0 = const()[name = tensor("op_5068_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5068_end_mask_0 = const()[name = tensor("op_5068_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5068_cast_fp16 = slice_by_index(begin = var_5068_begin_0, end = var_5068_end_0, end_mask = var_5068_end_mask_0, x = var_5001_cast_fp16)[name = tensor("op_5068_cast_fp16")]; tensor var_5075_begin_0 = const()[name = tensor("op_5075_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5075_end_0 = const()[name = tensor("op_5075_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5075_end_mask_0 = const()[name = tensor("op_5075_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5075_cast_fp16 = slice_by_index(begin = var_5075_begin_0, end = var_5075_end_0, end_mask = var_5075_end_mask_0, x = var_5001_cast_fp16)[name = tensor("op_5075_cast_fp16")]; tensor var_5082_begin_0 = const()[name = tensor("op_5082_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5082_end_0 = const()[name = tensor("op_5082_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5082_end_mask_0 = const()[name = tensor("op_5082_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5082_cast_fp16 = slice_by_index(begin = var_5082_begin_0, end = var_5082_end_0, end_mask = var_5082_end_mask_0, x = var_5005_cast_fp16)[name = tensor("op_5082_cast_fp16")]; tensor var_5089_begin_0 = const()[name = tensor("op_5089_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5089_end_0 = const()[name = tensor("op_5089_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5089_end_mask_0 = const()[name = tensor("op_5089_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5089_cast_fp16 = slice_by_index(begin = var_5089_begin_0, end = var_5089_end_0, end_mask = var_5089_end_mask_0, x = var_5005_cast_fp16)[name = tensor("op_5089_cast_fp16")]; tensor var_5096_begin_0 = const()[name = tensor("op_5096_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5096_end_0 = const()[name = tensor("op_5096_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5096_end_mask_0 = const()[name = tensor("op_5096_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5096_cast_fp16 = slice_by_index(begin = var_5096_begin_0, end = var_5096_end_0, end_mask = var_5096_end_mask_0, x = var_5005_cast_fp16)[name = tensor("op_5096_cast_fp16")]; tensor var_5103_begin_0 = const()[name = tensor("op_5103_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5103_end_0 = const()[name = tensor("op_5103_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5103_end_mask_0 = const()[name = tensor("op_5103_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5103_cast_fp16 = slice_by_index(begin = var_5103_begin_0, end = var_5103_end_0, end_mask = var_5103_end_mask_0, x = var_5005_cast_fp16)[name = tensor("op_5103_cast_fp16")]; tensor var_5110_begin_0 = const()[name = tensor("op_5110_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5110_end_0 = const()[name = tensor("op_5110_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5110_end_mask_0 = const()[name = tensor("op_5110_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5110_cast_fp16 = slice_by_index(begin = var_5110_begin_0, end = var_5110_end_0, end_mask = var_5110_end_mask_0, x = var_5009_cast_fp16)[name = tensor("op_5110_cast_fp16")]; tensor var_5117_begin_0 = const()[name = tensor("op_5117_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5117_end_0 = const()[name = tensor("op_5117_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5117_end_mask_0 = const()[name = tensor("op_5117_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5117_cast_fp16 = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = var_5009_cast_fp16)[name = tensor("op_5117_cast_fp16")]; tensor var_5124_begin_0 = const()[name = tensor("op_5124_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5124_end_0 = const()[name = tensor("op_5124_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5124_end_mask_0 = const()[name = tensor("op_5124_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5124_cast_fp16 = slice_by_index(begin = var_5124_begin_0, end = var_5124_end_0, end_mask = var_5124_end_mask_0, x = var_5009_cast_fp16)[name = tensor("op_5124_cast_fp16")]; tensor var_5131_begin_0 = const()[name = tensor("op_5131_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5131_end_0 = const()[name = tensor("op_5131_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5131_end_mask_0 = const()[name = tensor("op_5131_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5131_cast_fp16 = slice_by_index(begin = var_5131_begin_0, end = var_5131_end_0, end_mask = var_5131_end_mask_0, x = var_5009_cast_fp16)[name = tensor("op_5131_cast_fp16")]; tensor var_5138_begin_0 = const()[name = tensor("op_5138_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5138_end_0 = const()[name = tensor("op_5138_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5138_end_mask_0 = const()[name = tensor("op_5138_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5138_cast_fp16 = slice_by_index(begin = var_5138_begin_0, end = var_5138_end_0, end_mask = var_5138_end_mask_0, x = var_5013_cast_fp16)[name = tensor("op_5138_cast_fp16")]; tensor var_5145_begin_0 = const()[name = tensor("op_5145_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5145_end_0 = const()[name = tensor("op_5145_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5145_end_mask_0 = const()[name = tensor("op_5145_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5145_cast_fp16 = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = var_5013_cast_fp16)[name = tensor("op_5145_cast_fp16")]; tensor var_5152_begin_0 = const()[name = tensor("op_5152_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5152_end_0 = const()[name = tensor("op_5152_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5152_end_mask_0 = const()[name = tensor("op_5152_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5152_cast_fp16 = slice_by_index(begin = var_5152_begin_0, end = var_5152_end_0, end_mask = var_5152_end_mask_0, x = var_5013_cast_fp16)[name = tensor("op_5152_cast_fp16")]; tensor var_5159_begin_0 = const()[name = tensor("op_5159_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5159_end_0 = const()[name = tensor("op_5159_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5159_end_mask_0 = const()[name = tensor("op_5159_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5159_cast_fp16 = slice_by_index(begin = var_5159_begin_0, end = var_5159_end_0, end_mask = var_5159_end_mask_0, x = var_5013_cast_fp16)[name = tensor("op_5159_cast_fp16")]; tensor var_5166_begin_0 = const()[name = tensor("op_5166_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5166_end_0 = const()[name = tensor("op_5166_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5166_end_mask_0 = const()[name = tensor("op_5166_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5166_cast_fp16 = slice_by_index(begin = var_5166_begin_0, end = var_5166_end_0, end_mask = var_5166_end_mask_0, x = var_5017_cast_fp16)[name = tensor("op_5166_cast_fp16")]; tensor var_5173_begin_0 = const()[name = tensor("op_5173_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5173_end_0 = const()[name = tensor("op_5173_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5173_end_mask_0 = const()[name = tensor("op_5173_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5173_cast_fp16 = slice_by_index(begin = var_5173_begin_0, end = var_5173_end_0, end_mask = var_5173_end_mask_0, x = var_5017_cast_fp16)[name = tensor("op_5173_cast_fp16")]; tensor var_5180_begin_0 = const()[name = tensor("op_5180_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5180_end_0 = const()[name = tensor("op_5180_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5180_end_mask_0 = const()[name = tensor("op_5180_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5180_cast_fp16 = slice_by_index(begin = var_5180_begin_0, end = var_5180_end_0, end_mask = var_5180_end_mask_0, x = var_5017_cast_fp16)[name = tensor("op_5180_cast_fp16")]; tensor var_5187_begin_0 = const()[name = tensor("op_5187_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5187_end_0 = const()[name = tensor("op_5187_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5187_end_mask_0 = const()[name = tensor("op_5187_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5187_cast_fp16 = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = var_5017_cast_fp16)[name = tensor("op_5187_cast_fp16")]; tensor var_5194_begin_0 = const()[name = tensor("op_5194_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5194_end_0 = const()[name = tensor("op_5194_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5194_end_mask_0 = const()[name = tensor("op_5194_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5194_cast_fp16 = slice_by_index(begin = var_5194_begin_0, end = var_5194_end_0, end_mask = var_5194_end_mask_0, x = var_5021_cast_fp16)[name = tensor("op_5194_cast_fp16")]; tensor var_5201_begin_0 = const()[name = tensor("op_5201_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5201_end_0 = const()[name = tensor("op_5201_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5201_end_mask_0 = const()[name = tensor("op_5201_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5201_cast_fp16 = slice_by_index(begin = var_5201_begin_0, end = var_5201_end_0, end_mask = var_5201_end_mask_0, x = var_5021_cast_fp16)[name = tensor("op_5201_cast_fp16")]; tensor var_5208_begin_0 = const()[name = tensor("op_5208_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5208_end_0 = const()[name = tensor("op_5208_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5208_end_mask_0 = const()[name = tensor("op_5208_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5208_cast_fp16 = slice_by_index(begin = var_5208_begin_0, end = var_5208_end_0, end_mask = var_5208_end_mask_0, x = var_5021_cast_fp16)[name = tensor("op_5208_cast_fp16")]; tensor var_5215_begin_0 = const()[name = tensor("op_5215_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5215_end_0 = const()[name = tensor("op_5215_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5215_end_mask_0 = const()[name = tensor("op_5215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5215_cast_fp16 = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = var_5021_cast_fp16)[name = tensor("op_5215_cast_fp16")]; tensor var_5222_begin_0 = const()[name = tensor("op_5222_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5222_end_0 = const()[name = tensor("op_5222_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5222_end_mask_0 = const()[name = tensor("op_5222_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5222_cast_fp16 = slice_by_index(begin = var_5222_begin_0, end = var_5222_end_0, end_mask = var_5222_end_mask_0, x = var_5025_cast_fp16)[name = tensor("op_5222_cast_fp16")]; tensor var_5229_begin_0 = const()[name = tensor("op_5229_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5229_end_0 = const()[name = tensor("op_5229_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5229_end_mask_0 = const()[name = tensor("op_5229_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_5025_cast_fp16)[name = tensor("op_5229_cast_fp16")]; tensor var_5236_begin_0 = const()[name = tensor("op_5236_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5236_end_0 = const()[name = tensor("op_5236_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5236_end_mask_0 = const()[name = tensor("op_5236_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5236_cast_fp16 = slice_by_index(begin = var_5236_begin_0, end = var_5236_end_0, end_mask = var_5236_end_mask_0, x = var_5025_cast_fp16)[name = tensor("op_5236_cast_fp16")]; tensor var_5243_begin_0 = const()[name = tensor("op_5243_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5243_end_0 = const()[name = tensor("op_5243_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5243_end_mask_0 = const()[name = tensor("op_5243_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5243_cast_fp16 = slice_by_index(begin = var_5243_begin_0, end = var_5243_end_0, end_mask = var_5243_end_mask_0, x = var_5025_cast_fp16)[name = tensor("op_5243_cast_fp16")]; tensor var_5250_begin_0 = const()[name = tensor("op_5250_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5250_end_0 = const()[name = tensor("op_5250_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5250_end_mask_0 = const()[name = tensor("op_5250_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5250_cast_fp16 = slice_by_index(begin = var_5250_begin_0, end = var_5250_end_0, end_mask = var_5250_end_mask_0, x = var_5029_cast_fp16)[name = tensor("op_5250_cast_fp16")]; tensor var_5257_begin_0 = const()[name = tensor("op_5257_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5257_end_0 = const()[name = tensor("op_5257_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5257_end_mask_0 = const()[name = tensor("op_5257_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_5029_cast_fp16)[name = tensor("op_5257_cast_fp16")]; tensor var_5264_begin_0 = const()[name = tensor("op_5264_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5264_end_0 = const()[name = tensor("op_5264_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5264_end_mask_0 = const()[name = tensor("op_5264_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5264_cast_fp16 = slice_by_index(begin = var_5264_begin_0, end = var_5264_end_0, end_mask = var_5264_end_mask_0, x = var_5029_cast_fp16)[name = tensor("op_5264_cast_fp16")]; tensor var_5271_begin_0 = const()[name = tensor("op_5271_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5271_end_0 = const()[name = tensor("op_5271_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5271_end_mask_0 = const()[name = tensor("op_5271_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5271_cast_fp16 = slice_by_index(begin = var_5271_begin_0, end = var_5271_end_0, end_mask = var_5271_end_mask_0, x = var_5029_cast_fp16)[name = tensor("op_5271_cast_fp16")]; tensor var_5278_begin_0 = const()[name = tensor("op_5278_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5278_end_0 = const()[name = tensor("op_5278_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5278_end_mask_0 = const()[name = tensor("op_5278_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5278_cast_fp16 = slice_by_index(begin = var_5278_begin_0, end = var_5278_end_0, end_mask = var_5278_end_mask_0, x = var_5033_cast_fp16)[name = tensor("op_5278_cast_fp16")]; tensor var_5285_begin_0 = const()[name = tensor("op_5285_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5285_end_0 = const()[name = tensor("op_5285_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5285_end_mask_0 = const()[name = tensor("op_5285_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_5033_cast_fp16)[name = tensor("op_5285_cast_fp16")]; tensor var_5292_begin_0 = const()[name = tensor("op_5292_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5292_end_0 = const()[name = tensor("op_5292_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5292_end_mask_0 = const()[name = tensor("op_5292_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5292_cast_fp16 = slice_by_index(begin = var_5292_begin_0, end = var_5292_end_0, end_mask = var_5292_end_mask_0, x = var_5033_cast_fp16)[name = tensor("op_5292_cast_fp16")]; tensor var_5299_begin_0 = const()[name = tensor("op_5299_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5299_end_0 = const()[name = tensor("op_5299_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5299_end_mask_0 = const()[name = tensor("op_5299_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5299_cast_fp16 = slice_by_index(begin = var_5299_begin_0, end = var_5299_end_0, end_mask = var_5299_end_mask_0, x = var_5033_cast_fp16)[name = tensor("op_5299_cast_fp16")]; tensor var_5306_begin_0 = const()[name = tensor("op_5306_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5306_end_0 = const()[name = tensor("op_5306_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5306_end_mask_0 = const()[name = tensor("op_5306_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5306_cast_fp16 = slice_by_index(begin = var_5306_begin_0, end = var_5306_end_0, end_mask = var_5306_end_mask_0, x = var_5037_cast_fp16)[name = tensor("op_5306_cast_fp16")]; tensor var_5313_begin_0 = const()[name = tensor("op_5313_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5313_end_0 = const()[name = tensor("op_5313_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5313_end_mask_0 = const()[name = tensor("op_5313_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5313_cast_fp16 = slice_by_index(begin = var_5313_begin_0, end = var_5313_end_0, end_mask = var_5313_end_mask_0, x = var_5037_cast_fp16)[name = tensor("op_5313_cast_fp16")]; tensor var_5320_begin_0 = const()[name = tensor("op_5320_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5320_end_0 = const()[name = tensor("op_5320_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5320_end_mask_0 = const()[name = tensor("op_5320_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5320_cast_fp16 = slice_by_index(begin = var_5320_begin_0, end = var_5320_end_0, end_mask = var_5320_end_mask_0, x = var_5037_cast_fp16)[name = tensor("op_5320_cast_fp16")]; tensor var_5327_begin_0 = const()[name = tensor("op_5327_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5327_end_0 = const()[name = tensor("op_5327_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5327_end_mask_0 = const()[name = tensor("op_5327_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5327_cast_fp16 = slice_by_index(begin = var_5327_begin_0, end = var_5327_end_0, end_mask = var_5327_end_mask_0, x = var_5037_cast_fp16)[name = tensor("op_5327_cast_fp16")]; tensor var_5334_begin_0 = const()[name = tensor("op_5334_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5334_end_0 = const()[name = tensor("op_5334_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5334_end_mask_0 = const()[name = tensor("op_5334_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5334_cast_fp16 = slice_by_index(begin = var_5334_begin_0, end = var_5334_end_0, end_mask = var_5334_end_mask_0, x = var_5041_cast_fp16)[name = tensor("op_5334_cast_fp16")]; tensor var_5341_begin_0 = const()[name = tensor("op_5341_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5341_end_0 = const()[name = tensor("op_5341_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5341_end_mask_0 = const()[name = tensor("op_5341_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5341_cast_fp16 = slice_by_index(begin = var_5341_begin_0, end = var_5341_end_0, end_mask = var_5341_end_mask_0, x = var_5041_cast_fp16)[name = tensor("op_5341_cast_fp16")]; tensor var_5348_begin_0 = const()[name = tensor("op_5348_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5348_end_0 = const()[name = tensor("op_5348_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5348_end_mask_0 = const()[name = tensor("op_5348_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5348_cast_fp16 = slice_by_index(begin = var_5348_begin_0, end = var_5348_end_0, end_mask = var_5348_end_mask_0, x = var_5041_cast_fp16)[name = tensor("op_5348_cast_fp16")]; tensor var_5355_begin_0 = const()[name = tensor("op_5355_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5355_end_0 = const()[name = tensor("op_5355_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5355_end_mask_0 = const()[name = tensor("op_5355_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5355_cast_fp16 = slice_by_index(begin = var_5355_begin_0, end = var_5355_end_0, end_mask = var_5355_end_mask_0, x = var_5041_cast_fp16)[name = tensor("op_5355_cast_fp16")]; tensor var_5362_begin_0 = const()[name = tensor("op_5362_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5362_end_0 = const()[name = tensor("op_5362_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_5362_end_mask_0 = const()[name = tensor("op_5362_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5362_cast_fp16 = slice_by_index(begin = var_5362_begin_0, end = var_5362_end_0, end_mask = var_5362_end_mask_0, x = var_5045_cast_fp16)[name = tensor("op_5362_cast_fp16")]; tensor var_5369_begin_0 = const()[name = tensor("op_5369_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_5369_end_0 = const()[name = tensor("op_5369_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_5369_end_mask_0 = const()[name = tensor("op_5369_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5369_cast_fp16 = slice_by_index(begin = var_5369_begin_0, end = var_5369_end_0, end_mask = var_5369_end_mask_0, x = var_5045_cast_fp16)[name = tensor("op_5369_cast_fp16")]; tensor var_5376_begin_0 = const()[name = tensor("op_5376_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_5376_end_0 = const()[name = tensor("op_5376_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_5376_end_mask_0 = const()[name = tensor("op_5376_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5376_cast_fp16 = slice_by_index(begin = var_5376_begin_0, end = var_5376_end_0, end_mask = var_5376_end_mask_0, x = var_5045_cast_fp16)[name = tensor("op_5376_cast_fp16")]; tensor var_5383_begin_0 = const()[name = tensor("op_5383_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_5383_end_0 = const()[name = tensor("op_5383_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5383_end_mask_0 = const()[name = tensor("op_5383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5383_cast_fp16 = slice_by_index(begin = var_5383_begin_0, end = var_5383_end_0, end_mask = var_5383_end_mask_0, x = var_5045_cast_fp16)[name = tensor("op_5383_cast_fp16")]; tensor k_11_perm_0 = const()[name = tensor("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_5388_begin_0 = const()[name = tensor("op_5388_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5388_end_0 = const()[name = tensor("op_5388_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_5388_end_mask_0 = const()[name = tensor("op_5388_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_6 = transpose(perm = k_11_perm_0, x = key_11_cast_fp16)[name = tensor("transpose_6")]; tensor var_5388_cast_fp16 = slice_by_index(begin = var_5388_begin_0, end = var_5388_end_0, end_mask = var_5388_end_mask_0, x = transpose_6)[name = tensor("op_5388_cast_fp16")]; tensor var_5392_begin_0 = const()[name = tensor("op_5392_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_5392_end_0 = const()[name = tensor("op_5392_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_5392_end_mask_0 = const()[name = tensor("op_5392_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5392_cast_fp16 = slice_by_index(begin = var_5392_begin_0, end = var_5392_end_0, end_mask = var_5392_end_mask_0, x = transpose_6)[name = tensor("op_5392_cast_fp16")]; tensor var_5396_begin_0 = const()[name = tensor("op_5396_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_5396_end_0 = const()[name = tensor("op_5396_end_0"), val = tensor([1, 1500, 1, 192])]; 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 = transpose_6)[name = tensor("op_5396_cast_fp16")]; tensor var_5400_begin_0 = const()[name = tensor("op_5400_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_5400_end_0 = const()[name = tensor("op_5400_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_5400_end_mask_0 = const()[name = tensor("op_5400_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5400_cast_fp16 = slice_by_index(begin = var_5400_begin_0, end = var_5400_end_0, end_mask = var_5400_end_mask_0, x = transpose_6)[name = tensor("op_5400_cast_fp16")]; tensor var_5404_begin_0 = const()[name = tensor("op_5404_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_5404_end_0 = const()[name = tensor("op_5404_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_5404_end_mask_0 = const()[name = tensor("op_5404_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5404_cast_fp16 = slice_by_index(begin = var_5404_begin_0, end = var_5404_end_0, end_mask = var_5404_end_mask_0, x = transpose_6)[name = tensor("op_5404_cast_fp16")]; tensor var_5408_begin_0 = const()[name = tensor("op_5408_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_5408_end_0 = const()[name = tensor("op_5408_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_5408_end_mask_0 = const()[name = tensor("op_5408_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5408_cast_fp16 = slice_by_index(begin = var_5408_begin_0, end = var_5408_end_0, end_mask = var_5408_end_mask_0, x = transpose_6)[name = tensor("op_5408_cast_fp16")]; tensor var_5412_begin_0 = const()[name = tensor("op_5412_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_5412_end_0 = const()[name = tensor("op_5412_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_5412_end_mask_0 = const()[name = tensor("op_5412_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5412_cast_fp16 = slice_by_index(begin = var_5412_begin_0, end = var_5412_end_0, end_mask = var_5412_end_mask_0, x = transpose_6)[name = tensor("op_5412_cast_fp16")]; tensor var_5416_begin_0 = const()[name = tensor("op_5416_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_5416_end_0 = const()[name = tensor("op_5416_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_5416_end_mask_0 = const()[name = tensor("op_5416_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5416_cast_fp16 = slice_by_index(begin = var_5416_begin_0, end = var_5416_end_0, end_mask = var_5416_end_mask_0, x = transpose_6)[name = tensor("op_5416_cast_fp16")]; tensor var_5420_begin_0 = const()[name = tensor("op_5420_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_5420_end_0 = const()[name = tensor("op_5420_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_5420_end_mask_0 = const()[name = tensor("op_5420_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5420_cast_fp16 = slice_by_index(begin = var_5420_begin_0, end = var_5420_end_0, end_mask = var_5420_end_mask_0, x = transpose_6)[name = tensor("op_5420_cast_fp16")]; tensor var_5424_begin_0 = const()[name = tensor("op_5424_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_5424_end_0 = const()[name = tensor("op_5424_end_0"), val = tensor([1, 1500, 1, 640])]; 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 = transpose_6)[name = tensor("op_5424_cast_fp16")]; tensor var_5428_begin_0 = const()[name = tensor("op_5428_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_5428_end_0 = const()[name = tensor("op_5428_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_5428_end_mask_0 = const()[name = tensor("op_5428_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5428_cast_fp16 = slice_by_index(begin = var_5428_begin_0, end = var_5428_end_0, end_mask = var_5428_end_mask_0, x = transpose_6)[name = tensor("op_5428_cast_fp16")]; tensor var_5432_begin_0 = const()[name = tensor("op_5432_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_5432_end_0 = const()[name = tensor("op_5432_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_5432_end_mask_0 = const()[name = tensor("op_5432_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_5432_cast_fp16 = slice_by_index(begin = var_5432_begin_0, end = var_5432_end_0, end_mask = var_5432_end_mask_0, x = transpose_6)[name = tensor("op_5432_cast_fp16")]; tensor var_5434_begin_0 = const()[name = tensor("op_5434_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5434_end_0 = const()[name = tensor("op_5434_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5434_end_mask_0 = const()[name = tensor("op_5434_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5434_cast_fp16 = slice_by_index(begin = var_5434_begin_0, end = var_5434_end_0, end_mask = var_5434_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5434_cast_fp16")]; tensor var_5438_begin_0 = const()[name = tensor("op_5438_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5438_end_0 = const()[name = tensor("op_5438_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5438_end_mask_0 = const()[name = tensor("op_5438_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_11_cast_fp16)[name = tensor("op_5438_cast_fp16")]; tensor var_5442_begin_0 = const()[name = tensor("op_5442_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5442_end_0 = const()[name = tensor("op_5442_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5442_end_mask_0 = const()[name = tensor("op_5442_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5442_cast_fp16 = slice_by_index(begin = var_5442_begin_0, end = var_5442_end_0, end_mask = var_5442_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5442_cast_fp16")]; tensor var_5446_begin_0 = const()[name = tensor("op_5446_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5446_end_0 = const()[name = tensor("op_5446_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5446_end_mask_0 = const()[name = tensor("op_5446_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5446_cast_fp16 = slice_by_index(begin = var_5446_begin_0, end = var_5446_end_0, end_mask = var_5446_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5446_cast_fp16")]; tensor var_5450_begin_0 = const()[name = tensor("op_5450_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5450_end_0 = const()[name = tensor("op_5450_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_5450_end_mask_0 = const()[name = tensor("op_5450_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5450_cast_fp16 = slice_by_index(begin = var_5450_begin_0, end = var_5450_end_0, end_mask = var_5450_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5450_cast_fp16")]; tensor var_5454_begin_0 = const()[name = tensor("op_5454_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5454_end_0 = const()[name = tensor("op_5454_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_5454_end_mask_0 = const()[name = tensor("op_5454_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5454_cast_fp16 = slice_by_index(begin = var_5454_begin_0, end = var_5454_end_0, end_mask = var_5454_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5454_cast_fp16")]; tensor var_5458_begin_0 = const()[name = tensor("op_5458_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5458_end_0 = const()[name = tensor("op_5458_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_5458_end_mask_0 = const()[name = tensor("op_5458_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5458_cast_fp16 = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5458_cast_fp16")]; tensor var_5462_begin_0 = const()[name = tensor("op_5462_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5462_end_0 = const()[name = tensor("op_5462_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_5462_end_mask_0 = const()[name = tensor("op_5462_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5462_cast_fp16 = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5462_cast_fp16")]; tensor var_5466_begin_0 = const()[name = tensor("op_5466_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5466_end_0 = const()[name = tensor("op_5466_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_5466_end_mask_0 = const()[name = tensor("op_5466_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_11_cast_fp16)[name = tensor("op_5466_cast_fp16")]; tensor var_5470_begin_0 = const()[name = tensor("op_5470_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5470_end_0 = const()[name = tensor("op_5470_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_5470_end_mask_0 = const()[name = tensor("op_5470_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5470_cast_fp16 = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5470_cast_fp16")]; tensor var_5474_begin_0 = const()[name = tensor("op_5474_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_5474_end_0 = const()[name = tensor("op_5474_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_5474_end_mask_0 = const()[name = tensor("op_5474_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5474_cast_fp16 = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5474_cast_fp16")]; tensor var_5478_begin_0 = const()[name = tensor("op_5478_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_5478_end_0 = const()[name = tensor("op_5478_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_5478_end_mask_0 = const()[name = tensor("op_5478_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5478_cast_fp16 = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = value_11_cast_fp16)[name = tensor("op_5478_cast_fp16")]; tensor var_5482_equation_0 = const()[name = tensor("op_5482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5482_cast_fp16 = einsum(equation = var_5482_equation_0, values = (var_5388_cast_fp16, var_5054_cast_fp16))[name = tensor("op_5482_cast_fp16")]; tensor var_5483_to_fp16 = const()[name = tensor("op_5483_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_481_cast_fp16 = mul(x = var_5482_cast_fp16, y = var_5483_to_fp16)[name = tensor("aw_chunk_481_cast_fp16")]; tensor var_5486_equation_0 = const()[name = tensor("op_5486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5486_cast_fp16 = einsum(equation = var_5486_equation_0, values = (var_5388_cast_fp16, var_5061_cast_fp16))[name = tensor("op_5486_cast_fp16")]; tensor var_5487_to_fp16 = const()[name = tensor("op_5487_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_483_cast_fp16 = mul(x = var_5486_cast_fp16, y = var_5487_to_fp16)[name = tensor("aw_chunk_483_cast_fp16")]; tensor var_5490_equation_0 = const()[name = tensor("op_5490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5490_cast_fp16 = einsum(equation = var_5490_equation_0, values = (var_5388_cast_fp16, var_5068_cast_fp16))[name = tensor("op_5490_cast_fp16")]; tensor var_5491_to_fp16 = const()[name = tensor("op_5491_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_485_cast_fp16 = mul(x = var_5490_cast_fp16, y = var_5491_to_fp16)[name = tensor("aw_chunk_485_cast_fp16")]; tensor var_5494_equation_0 = const()[name = tensor("op_5494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5494_cast_fp16 = einsum(equation = var_5494_equation_0, values = (var_5388_cast_fp16, var_5075_cast_fp16))[name = tensor("op_5494_cast_fp16")]; tensor var_5495_to_fp16 = const()[name = tensor("op_5495_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_487_cast_fp16 = mul(x = var_5494_cast_fp16, y = var_5495_to_fp16)[name = tensor("aw_chunk_487_cast_fp16")]; tensor var_5498_equation_0 = const()[name = tensor("op_5498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5498_cast_fp16 = einsum(equation = var_5498_equation_0, values = (var_5392_cast_fp16, var_5082_cast_fp16))[name = tensor("op_5498_cast_fp16")]; tensor var_5499_to_fp16 = const()[name = tensor("op_5499_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_489_cast_fp16 = mul(x = var_5498_cast_fp16, y = var_5499_to_fp16)[name = tensor("aw_chunk_489_cast_fp16")]; tensor var_5502_equation_0 = const()[name = tensor("op_5502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5502_cast_fp16 = einsum(equation = var_5502_equation_0, values = (var_5392_cast_fp16, var_5089_cast_fp16))[name = tensor("op_5502_cast_fp16")]; tensor var_5503_to_fp16 = const()[name = tensor("op_5503_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_491_cast_fp16 = mul(x = var_5502_cast_fp16, y = var_5503_to_fp16)[name = tensor("aw_chunk_491_cast_fp16")]; tensor var_5506_equation_0 = const()[name = tensor("op_5506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5506_cast_fp16 = einsum(equation = var_5506_equation_0, values = (var_5392_cast_fp16, var_5096_cast_fp16))[name = tensor("op_5506_cast_fp16")]; tensor var_5507_to_fp16 = const()[name = tensor("op_5507_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_493_cast_fp16 = mul(x = var_5506_cast_fp16, y = var_5507_to_fp16)[name = tensor("aw_chunk_493_cast_fp16")]; tensor var_5510_equation_0 = const()[name = tensor("op_5510_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5510_cast_fp16 = einsum(equation = var_5510_equation_0, values = (var_5392_cast_fp16, var_5103_cast_fp16))[name = tensor("op_5510_cast_fp16")]; tensor var_5511_to_fp16 = const()[name = tensor("op_5511_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_495_cast_fp16 = mul(x = var_5510_cast_fp16, y = var_5511_to_fp16)[name = tensor("aw_chunk_495_cast_fp16")]; tensor var_5514_equation_0 = const()[name = tensor("op_5514_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5514_cast_fp16 = einsum(equation = var_5514_equation_0, values = (var_5396_cast_fp16, var_5110_cast_fp16))[name = tensor("op_5514_cast_fp16")]; tensor var_5515_to_fp16 = const()[name = tensor("op_5515_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_497_cast_fp16 = mul(x = var_5514_cast_fp16, y = var_5515_to_fp16)[name = tensor("aw_chunk_497_cast_fp16")]; tensor var_5518_equation_0 = const()[name = tensor("op_5518_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5518_cast_fp16 = einsum(equation = var_5518_equation_0, values = (var_5396_cast_fp16, var_5117_cast_fp16))[name = tensor("op_5518_cast_fp16")]; tensor var_5519_to_fp16 = const()[name = tensor("op_5519_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_499_cast_fp16 = mul(x = var_5518_cast_fp16, y = var_5519_to_fp16)[name = tensor("aw_chunk_499_cast_fp16")]; tensor var_5522_equation_0 = const()[name = tensor("op_5522_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5522_cast_fp16 = einsum(equation = var_5522_equation_0, values = (var_5396_cast_fp16, var_5124_cast_fp16))[name = tensor("op_5522_cast_fp16")]; tensor var_5523_to_fp16 = const()[name = tensor("op_5523_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_501_cast_fp16 = mul(x = var_5522_cast_fp16, y = var_5523_to_fp16)[name = tensor("aw_chunk_501_cast_fp16")]; tensor var_5526_equation_0 = const()[name = tensor("op_5526_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5526_cast_fp16 = einsum(equation = var_5526_equation_0, values = (var_5396_cast_fp16, var_5131_cast_fp16))[name = tensor("op_5526_cast_fp16")]; tensor var_5527_to_fp16 = const()[name = tensor("op_5527_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_503_cast_fp16 = mul(x = var_5526_cast_fp16, y = var_5527_to_fp16)[name = tensor("aw_chunk_503_cast_fp16")]; tensor var_5530_equation_0 = const()[name = tensor("op_5530_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5530_cast_fp16 = einsum(equation = var_5530_equation_0, values = (var_5400_cast_fp16, var_5138_cast_fp16))[name = tensor("op_5530_cast_fp16")]; tensor var_5531_to_fp16 = const()[name = tensor("op_5531_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_505_cast_fp16 = mul(x = var_5530_cast_fp16, y = var_5531_to_fp16)[name = tensor("aw_chunk_505_cast_fp16")]; tensor var_5534_equation_0 = const()[name = tensor("op_5534_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5534_cast_fp16 = einsum(equation = var_5534_equation_0, values = (var_5400_cast_fp16, var_5145_cast_fp16))[name = tensor("op_5534_cast_fp16")]; tensor var_5535_to_fp16 = const()[name = tensor("op_5535_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_507_cast_fp16 = mul(x = var_5534_cast_fp16, y = var_5535_to_fp16)[name = tensor("aw_chunk_507_cast_fp16")]; tensor var_5538_equation_0 = const()[name = tensor("op_5538_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5538_cast_fp16 = einsum(equation = var_5538_equation_0, values = (var_5400_cast_fp16, var_5152_cast_fp16))[name = tensor("op_5538_cast_fp16")]; tensor var_5539_to_fp16 = const()[name = tensor("op_5539_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_509_cast_fp16 = mul(x = var_5538_cast_fp16, y = var_5539_to_fp16)[name = tensor("aw_chunk_509_cast_fp16")]; tensor var_5542_equation_0 = const()[name = tensor("op_5542_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5542_cast_fp16 = einsum(equation = var_5542_equation_0, values = (var_5400_cast_fp16, var_5159_cast_fp16))[name = tensor("op_5542_cast_fp16")]; tensor var_5543_to_fp16 = const()[name = tensor("op_5543_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_511_cast_fp16 = mul(x = var_5542_cast_fp16, y = var_5543_to_fp16)[name = tensor("aw_chunk_511_cast_fp16")]; tensor var_5546_equation_0 = const()[name = tensor("op_5546_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5546_cast_fp16 = einsum(equation = var_5546_equation_0, values = (var_5404_cast_fp16, var_5166_cast_fp16))[name = tensor("op_5546_cast_fp16")]; tensor var_5547_to_fp16 = const()[name = tensor("op_5547_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_513_cast_fp16 = mul(x = var_5546_cast_fp16, y = var_5547_to_fp16)[name = tensor("aw_chunk_513_cast_fp16")]; tensor var_5550_equation_0 = const()[name = tensor("op_5550_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5550_cast_fp16 = einsum(equation = var_5550_equation_0, values = (var_5404_cast_fp16, var_5173_cast_fp16))[name = tensor("op_5550_cast_fp16")]; tensor var_5551_to_fp16 = const()[name = tensor("op_5551_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_515_cast_fp16 = mul(x = var_5550_cast_fp16, y = var_5551_to_fp16)[name = tensor("aw_chunk_515_cast_fp16")]; tensor var_5554_equation_0 = const()[name = tensor("op_5554_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5554_cast_fp16 = einsum(equation = var_5554_equation_0, values = (var_5404_cast_fp16, var_5180_cast_fp16))[name = tensor("op_5554_cast_fp16")]; tensor var_5555_to_fp16 = const()[name = tensor("op_5555_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_517_cast_fp16 = mul(x = var_5554_cast_fp16, y = var_5555_to_fp16)[name = tensor("aw_chunk_517_cast_fp16")]; tensor var_5558_equation_0 = const()[name = tensor("op_5558_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5558_cast_fp16 = einsum(equation = var_5558_equation_0, values = (var_5404_cast_fp16, var_5187_cast_fp16))[name = tensor("op_5558_cast_fp16")]; tensor var_5559_to_fp16 = const()[name = tensor("op_5559_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_519_cast_fp16 = mul(x = var_5558_cast_fp16, y = var_5559_to_fp16)[name = tensor("aw_chunk_519_cast_fp16")]; tensor var_5562_equation_0 = const()[name = tensor("op_5562_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5562_cast_fp16 = einsum(equation = var_5562_equation_0, values = (var_5408_cast_fp16, var_5194_cast_fp16))[name = tensor("op_5562_cast_fp16")]; tensor var_5563_to_fp16 = const()[name = tensor("op_5563_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_521_cast_fp16 = mul(x = var_5562_cast_fp16, y = var_5563_to_fp16)[name = tensor("aw_chunk_521_cast_fp16")]; tensor var_5566_equation_0 = const()[name = tensor("op_5566_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5566_cast_fp16 = einsum(equation = var_5566_equation_0, values = (var_5408_cast_fp16, var_5201_cast_fp16))[name = tensor("op_5566_cast_fp16")]; tensor var_5567_to_fp16 = const()[name = tensor("op_5567_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_523_cast_fp16 = mul(x = var_5566_cast_fp16, y = var_5567_to_fp16)[name = tensor("aw_chunk_523_cast_fp16")]; tensor var_5570_equation_0 = const()[name = tensor("op_5570_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5570_cast_fp16 = einsum(equation = var_5570_equation_0, values = (var_5408_cast_fp16, var_5208_cast_fp16))[name = tensor("op_5570_cast_fp16")]; tensor var_5571_to_fp16 = const()[name = tensor("op_5571_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_525_cast_fp16 = mul(x = var_5570_cast_fp16, y = var_5571_to_fp16)[name = tensor("aw_chunk_525_cast_fp16")]; tensor var_5574_equation_0 = const()[name = tensor("op_5574_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5574_cast_fp16 = einsum(equation = var_5574_equation_0, values = (var_5408_cast_fp16, var_5215_cast_fp16))[name = tensor("op_5574_cast_fp16")]; tensor var_5575_to_fp16 = const()[name = tensor("op_5575_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_527_cast_fp16 = mul(x = var_5574_cast_fp16, y = var_5575_to_fp16)[name = tensor("aw_chunk_527_cast_fp16")]; tensor var_5578_equation_0 = const()[name = tensor("op_5578_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5578_cast_fp16 = einsum(equation = var_5578_equation_0, values = (var_5412_cast_fp16, var_5222_cast_fp16))[name = tensor("op_5578_cast_fp16")]; tensor var_5579_to_fp16 = const()[name = tensor("op_5579_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_529_cast_fp16 = mul(x = var_5578_cast_fp16, y = var_5579_to_fp16)[name = tensor("aw_chunk_529_cast_fp16")]; tensor var_5582_equation_0 = const()[name = tensor("op_5582_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5582_cast_fp16 = einsum(equation = var_5582_equation_0, values = (var_5412_cast_fp16, var_5229_cast_fp16))[name = tensor("op_5582_cast_fp16")]; tensor var_5583_to_fp16 = const()[name = tensor("op_5583_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_531_cast_fp16 = mul(x = var_5582_cast_fp16, y = var_5583_to_fp16)[name = tensor("aw_chunk_531_cast_fp16")]; tensor var_5586_equation_0 = const()[name = tensor("op_5586_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5586_cast_fp16 = einsum(equation = var_5586_equation_0, values = (var_5412_cast_fp16, var_5236_cast_fp16))[name = tensor("op_5586_cast_fp16")]; tensor var_5587_to_fp16 = const()[name = tensor("op_5587_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_533_cast_fp16 = mul(x = var_5586_cast_fp16, y = var_5587_to_fp16)[name = tensor("aw_chunk_533_cast_fp16")]; tensor var_5590_equation_0 = const()[name = tensor("op_5590_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5590_cast_fp16 = einsum(equation = var_5590_equation_0, values = (var_5412_cast_fp16, var_5243_cast_fp16))[name = tensor("op_5590_cast_fp16")]; tensor var_5591_to_fp16 = const()[name = tensor("op_5591_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_535_cast_fp16 = mul(x = var_5590_cast_fp16, y = var_5591_to_fp16)[name = tensor("aw_chunk_535_cast_fp16")]; tensor var_5594_equation_0 = const()[name = tensor("op_5594_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5594_cast_fp16 = einsum(equation = var_5594_equation_0, values = (var_5416_cast_fp16, var_5250_cast_fp16))[name = tensor("op_5594_cast_fp16")]; tensor var_5595_to_fp16 = const()[name = tensor("op_5595_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_537_cast_fp16 = mul(x = var_5594_cast_fp16, y = var_5595_to_fp16)[name = tensor("aw_chunk_537_cast_fp16")]; tensor var_5598_equation_0 = const()[name = tensor("op_5598_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5598_cast_fp16 = einsum(equation = var_5598_equation_0, values = (var_5416_cast_fp16, var_5257_cast_fp16))[name = tensor("op_5598_cast_fp16")]; tensor var_5599_to_fp16 = const()[name = tensor("op_5599_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_539_cast_fp16 = mul(x = var_5598_cast_fp16, y = var_5599_to_fp16)[name = tensor("aw_chunk_539_cast_fp16")]; tensor var_5602_equation_0 = const()[name = tensor("op_5602_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5602_cast_fp16 = einsum(equation = var_5602_equation_0, values = (var_5416_cast_fp16, var_5264_cast_fp16))[name = tensor("op_5602_cast_fp16")]; tensor var_5603_to_fp16 = const()[name = tensor("op_5603_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_541_cast_fp16 = mul(x = var_5602_cast_fp16, y = var_5603_to_fp16)[name = tensor("aw_chunk_541_cast_fp16")]; tensor var_5606_equation_0 = const()[name = tensor("op_5606_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5606_cast_fp16 = einsum(equation = var_5606_equation_0, values = (var_5416_cast_fp16, var_5271_cast_fp16))[name = tensor("op_5606_cast_fp16")]; tensor var_5607_to_fp16 = const()[name = tensor("op_5607_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_543_cast_fp16 = mul(x = var_5606_cast_fp16, y = var_5607_to_fp16)[name = tensor("aw_chunk_543_cast_fp16")]; tensor var_5610_equation_0 = const()[name = tensor("op_5610_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5610_cast_fp16 = einsum(equation = var_5610_equation_0, values = (var_5420_cast_fp16, var_5278_cast_fp16))[name = tensor("op_5610_cast_fp16")]; tensor var_5611_to_fp16 = const()[name = tensor("op_5611_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_545_cast_fp16 = mul(x = var_5610_cast_fp16, y = var_5611_to_fp16)[name = tensor("aw_chunk_545_cast_fp16")]; tensor var_5614_equation_0 = const()[name = tensor("op_5614_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5614_cast_fp16 = einsum(equation = var_5614_equation_0, values = (var_5420_cast_fp16, var_5285_cast_fp16))[name = tensor("op_5614_cast_fp16")]; tensor var_5615_to_fp16 = const()[name = tensor("op_5615_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_547_cast_fp16 = mul(x = var_5614_cast_fp16, y = var_5615_to_fp16)[name = tensor("aw_chunk_547_cast_fp16")]; tensor var_5618_equation_0 = const()[name = tensor("op_5618_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5618_cast_fp16 = einsum(equation = var_5618_equation_0, values = (var_5420_cast_fp16, var_5292_cast_fp16))[name = tensor("op_5618_cast_fp16")]; tensor var_5619_to_fp16 = const()[name = tensor("op_5619_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_549_cast_fp16 = mul(x = var_5618_cast_fp16, y = var_5619_to_fp16)[name = tensor("aw_chunk_549_cast_fp16")]; tensor var_5622_equation_0 = const()[name = tensor("op_5622_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5622_cast_fp16 = einsum(equation = var_5622_equation_0, values = (var_5420_cast_fp16, var_5299_cast_fp16))[name = tensor("op_5622_cast_fp16")]; tensor var_5623_to_fp16 = const()[name = tensor("op_5623_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_551_cast_fp16 = mul(x = var_5622_cast_fp16, y = var_5623_to_fp16)[name = tensor("aw_chunk_551_cast_fp16")]; tensor var_5626_equation_0 = const()[name = tensor("op_5626_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5626_cast_fp16 = einsum(equation = var_5626_equation_0, values = (var_5424_cast_fp16, var_5306_cast_fp16))[name = tensor("op_5626_cast_fp16")]; tensor var_5627_to_fp16 = const()[name = tensor("op_5627_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_553_cast_fp16 = mul(x = var_5626_cast_fp16, y = var_5627_to_fp16)[name = tensor("aw_chunk_553_cast_fp16")]; tensor var_5630_equation_0 = const()[name = tensor("op_5630_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5630_cast_fp16 = einsum(equation = var_5630_equation_0, values = (var_5424_cast_fp16, var_5313_cast_fp16))[name = tensor("op_5630_cast_fp16")]; tensor var_5631_to_fp16 = const()[name = tensor("op_5631_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_555_cast_fp16 = mul(x = var_5630_cast_fp16, y = var_5631_to_fp16)[name = tensor("aw_chunk_555_cast_fp16")]; tensor var_5634_equation_0 = const()[name = tensor("op_5634_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5634_cast_fp16 = einsum(equation = var_5634_equation_0, values = (var_5424_cast_fp16, var_5320_cast_fp16))[name = tensor("op_5634_cast_fp16")]; tensor var_5635_to_fp16 = const()[name = tensor("op_5635_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_557_cast_fp16 = mul(x = var_5634_cast_fp16, y = var_5635_to_fp16)[name = tensor("aw_chunk_557_cast_fp16")]; tensor var_5638_equation_0 = const()[name = tensor("op_5638_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5638_cast_fp16 = einsum(equation = var_5638_equation_0, values = (var_5424_cast_fp16, var_5327_cast_fp16))[name = tensor("op_5638_cast_fp16")]; tensor var_5639_to_fp16 = const()[name = tensor("op_5639_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_559_cast_fp16 = mul(x = var_5638_cast_fp16, y = var_5639_to_fp16)[name = tensor("aw_chunk_559_cast_fp16")]; tensor var_5642_equation_0 = const()[name = tensor("op_5642_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5642_cast_fp16 = einsum(equation = var_5642_equation_0, values = (var_5428_cast_fp16, var_5334_cast_fp16))[name = tensor("op_5642_cast_fp16")]; tensor var_5643_to_fp16 = const()[name = tensor("op_5643_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_561_cast_fp16 = mul(x = var_5642_cast_fp16, y = var_5643_to_fp16)[name = tensor("aw_chunk_561_cast_fp16")]; tensor var_5646_equation_0 = const()[name = tensor("op_5646_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5646_cast_fp16 = einsum(equation = var_5646_equation_0, values = (var_5428_cast_fp16, var_5341_cast_fp16))[name = tensor("op_5646_cast_fp16")]; tensor var_5647_to_fp16 = const()[name = tensor("op_5647_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_563_cast_fp16 = mul(x = var_5646_cast_fp16, y = var_5647_to_fp16)[name = tensor("aw_chunk_563_cast_fp16")]; tensor var_5650_equation_0 = const()[name = tensor("op_5650_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5650_cast_fp16 = einsum(equation = var_5650_equation_0, values = (var_5428_cast_fp16, var_5348_cast_fp16))[name = tensor("op_5650_cast_fp16")]; tensor var_5651_to_fp16 = const()[name = tensor("op_5651_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_565_cast_fp16 = mul(x = var_5650_cast_fp16, y = var_5651_to_fp16)[name = tensor("aw_chunk_565_cast_fp16")]; tensor var_5654_equation_0 = const()[name = tensor("op_5654_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5654_cast_fp16 = einsum(equation = var_5654_equation_0, values = (var_5428_cast_fp16, var_5355_cast_fp16))[name = tensor("op_5654_cast_fp16")]; tensor var_5655_to_fp16 = const()[name = tensor("op_5655_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_567_cast_fp16 = mul(x = var_5654_cast_fp16, y = var_5655_to_fp16)[name = tensor("aw_chunk_567_cast_fp16")]; tensor var_5658_equation_0 = const()[name = tensor("op_5658_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5658_cast_fp16 = einsum(equation = var_5658_equation_0, values = (var_5432_cast_fp16, var_5362_cast_fp16))[name = tensor("op_5658_cast_fp16")]; tensor var_5659_to_fp16 = const()[name = tensor("op_5659_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_569_cast_fp16 = mul(x = var_5658_cast_fp16, y = var_5659_to_fp16)[name = tensor("aw_chunk_569_cast_fp16")]; tensor var_5662_equation_0 = const()[name = tensor("op_5662_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5662_cast_fp16 = einsum(equation = var_5662_equation_0, values = (var_5432_cast_fp16, var_5369_cast_fp16))[name = tensor("op_5662_cast_fp16")]; tensor var_5663_to_fp16 = const()[name = tensor("op_5663_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_571_cast_fp16 = mul(x = var_5662_cast_fp16, y = var_5663_to_fp16)[name = tensor("aw_chunk_571_cast_fp16")]; tensor var_5666_equation_0 = const()[name = tensor("op_5666_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5666_cast_fp16 = einsum(equation = var_5666_equation_0, values = (var_5432_cast_fp16, var_5376_cast_fp16))[name = tensor("op_5666_cast_fp16")]; tensor var_5667_to_fp16 = const()[name = tensor("op_5667_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_573_cast_fp16 = mul(x = var_5666_cast_fp16, y = var_5667_to_fp16)[name = tensor("aw_chunk_573_cast_fp16")]; tensor var_5670_equation_0 = const()[name = tensor("op_5670_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_5670_cast_fp16 = einsum(equation = var_5670_equation_0, values = (var_5432_cast_fp16, var_5383_cast_fp16))[name = tensor("op_5670_cast_fp16")]; tensor var_5671_to_fp16 = const()[name = tensor("op_5671_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_575_cast_fp16 = mul(x = var_5670_cast_fp16, y = var_5671_to_fp16)[name = tensor("aw_chunk_575_cast_fp16")]; tensor var_5673_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_481_cast_fp16)[name = tensor("op_5673_cast_fp16")]; tensor var_5674_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_483_cast_fp16)[name = tensor("op_5674_cast_fp16")]; tensor var_5675_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_485_cast_fp16)[name = tensor("op_5675_cast_fp16")]; tensor var_5676_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_487_cast_fp16)[name = tensor("op_5676_cast_fp16")]; tensor var_5677_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_489_cast_fp16)[name = tensor("op_5677_cast_fp16")]; tensor var_5678_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_491_cast_fp16)[name = tensor("op_5678_cast_fp16")]; tensor var_5679_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_493_cast_fp16)[name = tensor("op_5679_cast_fp16")]; tensor var_5680_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_495_cast_fp16)[name = tensor("op_5680_cast_fp16")]; tensor var_5681_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_497_cast_fp16)[name = tensor("op_5681_cast_fp16")]; tensor var_5682_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_499_cast_fp16)[name = tensor("op_5682_cast_fp16")]; tensor var_5683_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_501_cast_fp16)[name = tensor("op_5683_cast_fp16")]; tensor var_5684_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_503_cast_fp16)[name = tensor("op_5684_cast_fp16")]; tensor var_5685_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_505_cast_fp16)[name = tensor("op_5685_cast_fp16")]; tensor var_5686_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_507_cast_fp16)[name = tensor("op_5686_cast_fp16")]; tensor var_5687_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_509_cast_fp16)[name = tensor("op_5687_cast_fp16")]; tensor var_5688_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_511_cast_fp16)[name = tensor("op_5688_cast_fp16")]; tensor var_5689_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_513_cast_fp16)[name = tensor("op_5689_cast_fp16")]; tensor var_5690_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_515_cast_fp16)[name = tensor("op_5690_cast_fp16")]; tensor var_5691_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_517_cast_fp16)[name = tensor("op_5691_cast_fp16")]; tensor var_5692_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_519_cast_fp16)[name = tensor("op_5692_cast_fp16")]; tensor var_5693_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_521_cast_fp16)[name = tensor("op_5693_cast_fp16")]; tensor var_5694_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_523_cast_fp16)[name = tensor("op_5694_cast_fp16")]; tensor var_5695_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_525_cast_fp16)[name = tensor("op_5695_cast_fp16")]; tensor var_5696_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_527_cast_fp16)[name = tensor("op_5696_cast_fp16")]; tensor var_5697_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_529_cast_fp16)[name = tensor("op_5697_cast_fp16")]; tensor var_5698_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_531_cast_fp16)[name = tensor("op_5698_cast_fp16")]; tensor var_5699_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_533_cast_fp16)[name = tensor("op_5699_cast_fp16")]; tensor var_5700_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_535_cast_fp16)[name = tensor("op_5700_cast_fp16")]; tensor var_5701_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_537_cast_fp16)[name = tensor("op_5701_cast_fp16")]; tensor var_5702_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_539_cast_fp16)[name = tensor("op_5702_cast_fp16")]; tensor var_5703_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_541_cast_fp16)[name = tensor("op_5703_cast_fp16")]; tensor var_5704_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_543_cast_fp16)[name = tensor("op_5704_cast_fp16")]; tensor var_5705_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_545_cast_fp16)[name = tensor("op_5705_cast_fp16")]; tensor var_5706_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_547_cast_fp16)[name = tensor("op_5706_cast_fp16")]; tensor var_5707_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_549_cast_fp16)[name = tensor("op_5707_cast_fp16")]; tensor var_5708_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_551_cast_fp16)[name = tensor("op_5708_cast_fp16")]; tensor var_5709_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_553_cast_fp16)[name = tensor("op_5709_cast_fp16")]; tensor var_5710_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_555_cast_fp16)[name = tensor("op_5710_cast_fp16")]; tensor var_5711_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_557_cast_fp16)[name = tensor("op_5711_cast_fp16")]; tensor var_5712_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_559_cast_fp16)[name = tensor("op_5712_cast_fp16")]; tensor var_5713_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_561_cast_fp16)[name = tensor("op_5713_cast_fp16")]; tensor var_5714_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_563_cast_fp16)[name = tensor("op_5714_cast_fp16")]; tensor var_5715_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_565_cast_fp16)[name = tensor("op_5715_cast_fp16")]; tensor var_5716_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_567_cast_fp16)[name = tensor("op_5716_cast_fp16")]; tensor var_5717_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_569_cast_fp16)[name = tensor("op_5717_cast_fp16")]; tensor var_5718_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_571_cast_fp16)[name = tensor("op_5718_cast_fp16")]; tensor var_5719_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_573_cast_fp16)[name = tensor("op_5719_cast_fp16")]; tensor var_5720_cast_fp16 = softmax(axis = var_4946, x = aw_chunk_575_cast_fp16)[name = tensor("op_5720_cast_fp16")]; tensor var_5722_equation_0 = const()[name = tensor("op_5722_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5722_cast_fp16 = einsum(equation = var_5722_equation_0, values = (var_5434_cast_fp16, var_5673_cast_fp16))[name = tensor("op_5722_cast_fp16")]; tensor var_5724_equation_0 = const()[name = tensor("op_5724_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5724_cast_fp16 = einsum(equation = var_5724_equation_0, values = (var_5434_cast_fp16, var_5674_cast_fp16))[name = tensor("op_5724_cast_fp16")]; tensor var_5726_equation_0 = const()[name = tensor("op_5726_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5726_cast_fp16 = einsum(equation = var_5726_equation_0, values = (var_5434_cast_fp16, var_5675_cast_fp16))[name = tensor("op_5726_cast_fp16")]; tensor var_5728_equation_0 = const()[name = tensor("op_5728_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5728_cast_fp16 = einsum(equation = var_5728_equation_0, values = (var_5434_cast_fp16, var_5676_cast_fp16))[name = tensor("op_5728_cast_fp16")]; tensor var_5730_equation_0 = const()[name = tensor("op_5730_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5730_cast_fp16 = einsum(equation = var_5730_equation_0, values = (var_5438_cast_fp16, var_5677_cast_fp16))[name = tensor("op_5730_cast_fp16")]; tensor var_5732_equation_0 = const()[name = tensor("op_5732_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5732_cast_fp16 = einsum(equation = var_5732_equation_0, values = (var_5438_cast_fp16, var_5678_cast_fp16))[name = tensor("op_5732_cast_fp16")]; tensor var_5734_equation_0 = const()[name = tensor("op_5734_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5734_cast_fp16 = einsum(equation = var_5734_equation_0, values = (var_5438_cast_fp16, var_5679_cast_fp16))[name = tensor("op_5734_cast_fp16")]; tensor var_5736_equation_0 = const()[name = tensor("op_5736_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5736_cast_fp16 = einsum(equation = var_5736_equation_0, values = (var_5438_cast_fp16, var_5680_cast_fp16))[name = tensor("op_5736_cast_fp16")]; tensor var_5738_equation_0 = const()[name = tensor("op_5738_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5738_cast_fp16 = einsum(equation = var_5738_equation_0, values = (var_5442_cast_fp16, var_5681_cast_fp16))[name = tensor("op_5738_cast_fp16")]; tensor var_5740_equation_0 = const()[name = tensor("op_5740_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5740_cast_fp16 = einsum(equation = var_5740_equation_0, values = (var_5442_cast_fp16, var_5682_cast_fp16))[name = tensor("op_5740_cast_fp16")]; tensor var_5742_equation_0 = const()[name = tensor("op_5742_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5742_cast_fp16 = einsum(equation = var_5742_equation_0, values = (var_5442_cast_fp16, var_5683_cast_fp16))[name = tensor("op_5742_cast_fp16")]; tensor var_5744_equation_0 = const()[name = tensor("op_5744_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5744_cast_fp16 = einsum(equation = var_5744_equation_0, values = (var_5442_cast_fp16, var_5684_cast_fp16))[name = tensor("op_5744_cast_fp16")]; tensor var_5746_equation_0 = const()[name = tensor("op_5746_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5746_cast_fp16 = einsum(equation = var_5746_equation_0, values = (var_5446_cast_fp16, var_5685_cast_fp16))[name = tensor("op_5746_cast_fp16")]; tensor var_5748_equation_0 = const()[name = tensor("op_5748_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5748_cast_fp16 = einsum(equation = var_5748_equation_0, values = (var_5446_cast_fp16, var_5686_cast_fp16))[name = tensor("op_5748_cast_fp16")]; tensor var_5750_equation_0 = const()[name = tensor("op_5750_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5750_cast_fp16 = einsum(equation = var_5750_equation_0, values = (var_5446_cast_fp16, var_5687_cast_fp16))[name = tensor("op_5750_cast_fp16")]; tensor var_5752_equation_0 = const()[name = tensor("op_5752_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5752_cast_fp16 = einsum(equation = var_5752_equation_0, values = (var_5446_cast_fp16, var_5688_cast_fp16))[name = tensor("op_5752_cast_fp16")]; tensor var_5754_equation_0 = const()[name = tensor("op_5754_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5754_cast_fp16 = einsum(equation = var_5754_equation_0, values = (var_5450_cast_fp16, var_5689_cast_fp16))[name = tensor("op_5754_cast_fp16")]; tensor var_5756_equation_0 = const()[name = tensor("op_5756_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5756_cast_fp16 = einsum(equation = var_5756_equation_0, values = (var_5450_cast_fp16, var_5690_cast_fp16))[name = tensor("op_5756_cast_fp16")]; tensor var_5758_equation_0 = const()[name = tensor("op_5758_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5758_cast_fp16 = einsum(equation = var_5758_equation_0, values = (var_5450_cast_fp16, var_5691_cast_fp16))[name = tensor("op_5758_cast_fp16")]; tensor var_5760_equation_0 = const()[name = tensor("op_5760_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5760_cast_fp16 = einsum(equation = var_5760_equation_0, values = (var_5450_cast_fp16, var_5692_cast_fp16))[name = tensor("op_5760_cast_fp16")]; tensor var_5762_equation_0 = const()[name = tensor("op_5762_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5762_cast_fp16 = einsum(equation = var_5762_equation_0, values = (var_5454_cast_fp16, var_5693_cast_fp16))[name = tensor("op_5762_cast_fp16")]; tensor var_5764_equation_0 = const()[name = tensor("op_5764_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5764_cast_fp16 = einsum(equation = var_5764_equation_0, values = (var_5454_cast_fp16, var_5694_cast_fp16))[name = tensor("op_5764_cast_fp16")]; tensor var_5766_equation_0 = const()[name = tensor("op_5766_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5766_cast_fp16 = einsum(equation = var_5766_equation_0, values = (var_5454_cast_fp16, var_5695_cast_fp16))[name = tensor("op_5766_cast_fp16")]; tensor var_5768_equation_0 = const()[name = tensor("op_5768_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5768_cast_fp16 = einsum(equation = var_5768_equation_0, values = (var_5454_cast_fp16, var_5696_cast_fp16))[name = tensor("op_5768_cast_fp16")]; tensor var_5770_equation_0 = const()[name = tensor("op_5770_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5770_cast_fp16 = einsum(equation = var_5770_equation_0, values = (var_5458_cast_fp16, var_5697_cast_fp16))[name = tensor("op_5770_cast_fp16")]; tensor var_5772_equation_0 = const()[name = tensor("op_5772_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5772_cast_fp16 = einsum(equation = var_5772_equation_0, values = (var_5458_cast_fp16, var_5698_cast_fp16))[name = tensor("op_5772_cast_fp16")]; tensor var_5774_equation_0 = const()[name = tensor("op_5774_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5774_cast_fp16 = einsum(equation = var_5774_equation_0, values = (var_5458_cast_fp16, var_5699_cast_fp16))[name = tensor("op_5774_cast_fp16")]; tensor var_5776_equation_0 = const()[name = tensor("op_5776_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5776_cast_fp16 = einsum(equation = var_5776_equation_0, values = (var_5458_cast_fp16, var_5700_cast_fp16))[name = tensor("op_5776_cast_fp16")]; tensor var_5778_equation_0 = const()[name = tensor("op_5778_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5778_cast_fp16 = einsum(equation = var_5778_equation_0, values = (var_5462_cast_fp16, var_5701_cast_fp16))[name = tensor("op_5778_cast_fp16")]; tensor var_5780_equation_0 = const()[name = tensor("op_5780_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5780_cast_fp16 = einsum(equation = var_5780_equation_0, values = (var_5462_cast_fp16, var_5702_cast_fp16))[name = tensor("op_5780_cast_fp16")]; tensor var_5782_equation_0 = const()[name = tensor("op_5782_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5782_cast_fp16 = einsum(equation = var_5782_equation_0, values = (var_5462_cast_fp16, var_5703_cast_fp16))[name = tensor("op_5782_cast_fp16")]; tensor var_5784_equation_0 = const()[name = tensor("op_5784_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5784_cast_fp16 = einsum(equation = var_5784_equation_0, values = (var_5462_cast_fp16, var_5704_cast_fp16))[name = tensor("op_5784_cast_fp16")]; tensor var_5786_equation_0 = const()[name = tensor("op_5786_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5786_cast_fp16 = einsum(equation = var_5786_equation_0, values = (var_5466_cast_fp16, var_5705_cast_fp16))[name = tensor("op_5786_cast_fp16")]; tensor var_5788_equation_0 = const()[name = tensor("op_5788_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5788_cast_fp16 = einsum(equation = var_5788_equation_0, values = (var_5466_cast_fp16, var_5706_cast_fp16))[name = tensor("op_5788_cast_fp16")]; tensor var_5790_equation_0 = const()[name = tensor("op_5790_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5790_cast_fp16 = einsum(equation = var_5790_equation_0, values = (var_5466_cast_fp16, var_5707_cast_fp16))[name = tensor("op_5790_cast_fp16")]; tensor var_5792_equation_0 = const()[name = tensor("op_5792_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5792_cast_fp16 = einsum(equation = var_5792_equation_0, values = (var_5466_cast_fp16, var_5708_cast_fp16))[name = tensor("op_5792_cast_fp16")]; tensor var_5794_equation_0 = const()[name = tensor("op_5794_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5794_cast_fp16 = einsum(equation = var_5794_equation_0, values = (var_5470_cast_fp16, var_5709_cast_fp16))[name = tensor("op_5794_cast_fp16")]; tensor var_5796_equation_0 = const()[name = tensor("op_5796_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5796_cast_fp16 = einsum(equation = var_5796_equation_0, values = (var_5470_cast_fp16, var_5710_cast_fp16))[name = tensor("op_5796_cast_fp16")]; tensor var_5798_equation_0 = const()[name = tensor("op_5798_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5798_cast_fp16 = einsum(equation = var_5798_equation_0, values = (var_5470_cast_fp16, var_5711_cast_fp16))[name = tensor("op_5798_cast_fp16")]; tensor var_5800_equation_0 = const()[name = tensor("op_5800_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5800_cast_fp16 = einsum(equation = var_5800_equation_0, values = (var_5470_cast_fp16, var_5712_cast_fp16))[name = tensor("op_5800_cast_fp16")]; tensor var_5802_equation_0 = const()[name = tensor("op_5802_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5802_cast_fp16 = einsum(equation = var_5802_equation_0, values = (var_5474_cast_fp16, var_5713_cast_fp16))[name = tensor("op_5802_cast_fp16")]; tensor var_5804_equation_0 = const()[name = tensor("op_5804_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5804_cast_fp16 = einsum(equation = var_5804_equation_0, values = (var_5474_cast_fp16, var_5714_cast_fp16))[name = tensor("op_5804_cast_fp16")]; tensor var_5806_equation_0 = const()[name = tensor("op_5806_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5806_cast_fp16 = einsum(equation = var_5806_equation_0, values = (var_5474_cast_fp16, var_5715_cast_fp16))[name = tensor("op_5806_cast_fp16")]; tensor var_5808_equation_0 = const()[name = tensor("op_5808_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5808_cast_fp16 = einsum(equation = var_5808_equation_0, values = (var_5474_cast_fp16, var_5716_cast_fp16))[name = tensor("op_5808_cast_fp16")]; tensor var_5810_equation_0 = const()[name = tensor("op_5810_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5810_cast_fp16 = einsum(equation = var_5810_equation_0, values = (var_5478_cast_fp16, var_5717_cast_fp16))[name = tensor("op_5810_cast_fp16")]; tensor var_5812_equation_0 = const()[name = tensor("op_5812_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5812_cast_fp16 = einsum(equation = var_5812_equation_0, values = (var_5478_cast_fp16, var_5718_cast_fp16))[name = tensor("op_5812_cast_fp16")]; tensor var_5814_equation_0 = const()[name = tensor("op_5814_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5814_cast_fp16 = einsum(equation = var_5814_equation_0, values = (var_5478_cast_fp16, var_5719_cast_fp16))[name = tensor("op_5814_cast_fp16")]; tensor var_5816_equation_0 = const()[name = tensor("op_5816_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_5816_cast_fp16 = einsum(equation = var_5816_equation_0, values = (var_5478_cast_fp16, var_5720_cast_fp16))[name = tensor("op_5816_cast_fp16")]; tensor var_5818_interleave_0 = const()[name = tensor("op_5818_interleave_0"), val = tensor(false)]; tensor var_5818_cast_fp16 = concat(axis = var_4929, interleave = var_5818_interleave_0, values = (var_5722_cast_fp16, var_5724_cast_fp16, var_5726_cast_fp16, var_5728_cast_fp16))[name = tensor("op_5818_cast_fp16")]; tensor var_5820_interleave_0 = const()[name = tensor("op_5820_interleave_0"), val = tensor(false)]; tensor var_5820_cast_fp16 = concat(axis = var_4929, interleave = var_5820_interleave_0, values = (var_5730_cast_fp16, var_5732_cast_fp16, var_5734_cast_fp16, var_5736_cast_fp16))[name = tensor("op_5820_cast_fp16")]; tensor var_5822_interleave_0 = const()[name = tensor("op_5822_interleave_0"), val = tensor(false)]; tensor var_5822_cast_fp16 = concat(axis = var_4929, interleave = var_5822_interleave_0, values = (var_5738_cast_fp16, var_5740_cast_fp16, var_5742_cast_fp16, var_5744_cast_fp16))[name = tensor("op_5822_cast_fp16")]; tensor var_5824_interleave_0 = const()[name = tensor("op_5824_interleave_0"), val = tensor(false)]; tensor var_5824_cast_fp16 = concat(axis = var_4929, interleave = var_5824_interleave_0, values = (var_5746_cast_fp16, var_5748_cast_fp16, var_5750_cast_fp16, var_5752_cast_fp16))[name = tensor("op_5824_cast_fp16")]; tensor var_5826_interleave_0 = const()[name = tensor("op_5826_interleave_0"), val = tensor(false)]; tensor var_5826_cast_fp16 = concat(axis = var_4929, interleave = var_5826_interleave_0, values = (var_5754_cast_fp16, var_5756_cast_fp16, var_5758_cast_fp16, var_5760_cast_fp16))[name = tensor("op_5826_cast_fp16")]; tensor var_5828_interleave_0 = const()[name = tensor("op_5828_interleave_0"), val = tensor(false)]; tensor var_5828_cast_fp16 = concat(axis = var_4929, interleave = var_5828_interleave_0, values = (var_5762_cast_fp16, var_5764_cast_fp16, var_5766_cast_fp16, var_5768_cast_fp16))[name = tensor("op_5828_cast_fp16")]; tensor var_5830_interleave_0 = const()[name = tensor("op_5830_interleave_0"), val = tensor(false)]; tensor var_5830_cast_fp16 = concat(axis = var_4929, interleave = var_5830_interleave_0, values = (var_5770_cast_fp16, var_5772_cast_fp16, var_5774_cast_fp16, var_5776_cast_fp16))[name = tensor("op_5830_cast_fp16")]; tensor var_5832_interleave_0 = const()[name = tensor("op_5832_interleave_0"), val = tensor(false)]; tensor var_5832_cast_fp16 = concat(axis = var_4929, interleave = var_5832_interleave_0, values = (var_5778_cast_fp16, var_5780_cast_fp16, var_5782_cast_fp16, var_5784_cast_fp16))[name = tensor("op_5832_cast_fp16")]; tensor var_5834_interleave_0 = const()[name = tensor("op_5834_interleave_0"), val = tensor(false)]; tensor var_5834_cast_fp16 = concat(axis = var_4929, interleave = var_5834_interleave_0, values = (var_5786_cast_fp16, var_5788_cast_fp16, var_5790_cast_fp16, var_5792_cast_fp16))[name = tensor("op_5834_cast_fp16")]; tensor var_5836_interleave_0 = const()[name = tensor("op_5836_interleave_0"), val = tensor(false)]; tensor var_5836_cast_fp16 = concat(axis = var_4929, interleave = var_5836_interleave_0, values = (var_5794_cast_fp16, var_5796_cast_fp16, var_5798_cast_fp16, var_5800_cast_fp16))[name = tensor("op_5836_cast_fp16")]; tensor var_5838_interleave_0 = const()[name = tensor("op_5838_interleave_0"), val = tensor(false)]; tensor var_5838_cast_fp16 = concat(axis = var_4929, interleave = var_5838_interleave_0, values = (var_5802_cast_fp16, var_5804_cast_fp16, var_5806_cast_fp16, var_5808_cast_fp16))[name = tensor("op_5838_cast_fp16")]; tensor var_5840_interleave_0 = const()[name = tensor("op_5840_interleave_0"), val = tensor(false)]; tensor var_5840_cast_fp16 = concat(axis = var_4929, interleave = var_5840_interleave_0, values = (var_5810_cast_fp16, var_5812_cast_fp16, var_5814_cast_fp16, var_5816_cast_fp16))[name = tensor("op_5840_cast_fp16")]; tensor input_41_interleave_0 = const()[name = tensor("input_41_interleave_0"), val = tensor(false)]; tensor input_41_cast_fp16 = concat(axis = var_4946, interleave = input_41_interleave_0, values = (var_5818_cast_fp16, var_5820_cast_fp16, var_5822_cast_fp16, var_5824_cast_fp16, var_5826_cast_fp16, var_5828_cast_fp16, var_5830_cast_fp16, var_5832_cast_fp16, var_5834_cast_fp16, var_5836_cast_fp16, var_5838_cast_fp16, var_5840_cast_fp16))[name = tensor("input_41_cast_fp16")]; tensor var_5845 = const()[name = tensor("op_5845"), val = tensor([1, 1])]; tensor var_5847 = const()[name = tensor("op_5847"), val = tensor([1, 1])]; tensor obj_23_pad_type_0 = const()[name = tensor("obj_23_pad_type_0"), val = tensor("custom")]; tensor obj_23_pad_0 = const()[name = tensor("obj_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80639616)))]; tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81819328)))]; tensor obj_23_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = var_5847, groups = var_4946, pad = obj_23_pad_0, pad_type = obj_23_pad_type_0, strides = var_5845, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_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_5853 = const()[name = tensor("op_5853"), val = tensor([1])]; tensor channels_mean_23_cast_fp16 = reduce_mean(axes = var_5853, keep_dims = var_4947, 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_5857 = const()[name = tensor("op_5857"), val = tensor([1])]; tensor var_5858_cast_fp16 = reduce_mean(axes = var_5857, keep_dims = var_4947, x = zero_mean_sq_23_cast_fp16)[name = tensor("op_5858_cast_fp16")]; tensor var_5859_to_fp16 = const()[name = tensor("op_5859_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5860_cast_fp16 = add(x = var_5858_cast_fp16, y = var_5859_to_fp16)[name = tensor("op_5860_cast_fp16")]; tensor denom_23_epsilon_0_to_fp16 = const()[name = tensor("denom_23_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_23_cast_fp16 = rsqrt(epsilon = denom_23_epsilon_0_to_fp16, x = var_5860_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_43_gamma_0_to_fp16 = const()[name = tensor("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81820928)))]; tensor input_43_beta_0_to_fp16 = const()[name = tensor("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81822528)))]; tensor input_43_epsilon_0_to_fp16 = const()[name = tensor("input_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_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_43_cast_fp16")]; tensor var_5871 = const()[name = tensor("op_5871"), val = tensor([1, 1])]; tensor var_5873 = const()[name = tensor("op_5873"), 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_5_fc1_weight_to_fp16 = const()[name = tensor("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81824128)))]; tensor layers_5_fc1_bias_to_fp16 = const()[name = tensor("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86542784)))]; tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = var_5873, groups = var_4946, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = var_5871, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = tensor("input_45_cast_fp16")]; tensor input_47_mode_0 = const()[name = tensor("input_47_mode_0"), val = tensor("EXACT")]; tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = tensor("input_47_cast_fp16")]; tensor var_5879 = const()[name = tensor("op_5879"), val = tensor([1, 1])]; tensor var_5881 = const()[name = tensor("op_5881"), val = tensor([1, 1])]; tensor hidden_states_15_pad_type_0 = const()[name = tensor("hidden_states_15_pad_type_0"), val = tensor("custom")]; tensor hidden_states_15_pad_0 = const()[name = tensor("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_5_fc2_weight_to_fp16 = const()[name = tensor("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86548992)))]; tensor layers_5_fc2_bias_to_fp16 = const()[name = tensor("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91267648)))]; tensor hidden_states_15_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = var_5881, groups = var_4946, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = var_5879, weight = layers_5_fc2_weight_to_fp16, x = input_47_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_5888 = const()[name = tensor("op_5888"), val = tensor(3)]; tensor var_5905 = const()[name = tensor("op_5905"), val = tensor(1)]; tensor var_5906 = const()[name = tensor("op_5906"), val = tensor(true)]; tensor var_5916 = const()[name = tensor("op_5916"), val = tensor([1])]; tensor channels_mean_25_cast_fp16 = reduce_mean(axes = var_5916, keep_dims = var_5906, 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_5920 = const()[name = tensor("op_5920"), val = tensor([1])]; tensor var_5921_cast_fp16 = reduce_mean(axes = var_5920, keep_dims = var_5906, x = zero_mean_sq_25_cast_fp16)[name = tensor("op_5921_cast_fp16")]; tensor var_5922_to_fp16 = const()[name = tensor("op_5922_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_5923_cast_fp16 = add(x = var_5921_cast_fp16, y = var_5922_to_fp16)[name = tensor("op_5923_cast_fp16")]; tensor denom_25_epsilon_0_to_fp16 = const()[name = tensor("denom_25_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_25_cast_fp16 = rsqrt(epsilon = denom_25_epsilon_0_to_fp16, x = var_5923_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(91269248)))]; 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(91270848)))]; 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_5938 = const()[name = tensor("op_5938"), val = tensor([1, 1])]; tensor var_5940 = const()[name = tensor("op_5940"), val = tensor([1, 1])]; tensor query_13_pad_type_0 = const()[name = tensor("query_13_pad_type_0"), val = tensor("custom")]; tensor query_13_pad_0 = const()[name = tensor("query_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91272448)))]; tensor layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92452160)))]; tensor query_13_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_bias_to_fp16, dilations = var_5940, groups = var_5905, pad = query_13_pad_0, pad_type = query_13_pad_type_0, strides = var_5938, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("query_13_cast_fp16")]; tensor var_5944 = const()[name = tensor("op_5944"), val = tensor([1, 1])]; tensor var_5946 = const()[name = tensor("op_5946"), val = tensor([1, 1])]; tensor key_13_pad_type_0 = const()[name = tensor("key_13_pad_type_0"), val = tensor("custom")]; tensor key_13_pad_0 = const()[name = tensor("key_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92453760)))]; tensor key_13_cast_fp16 = conv(dilations = var_5946, groups = var_5905, pad = key_13_pad_0, pad_type = key_13_pad_type_0, strides = var_5944, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("key_13_cast_fp16")]; tensor var_5951 = const()[name = tensor("op_5951"), val = tensor([1, 1])]; tensor var_5953 = const()[name = tensor("op_5953"), val = tensor([1, 1])]; tensor value_13_pad_type_0 = const()[name = tensor("value_13_pad_type_0"), val = tensor("custom")]; tensor value_13_pad_0 = const()[name = tensor("value_13_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93633472)))]; tensor layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94813184)))]; tensor value_13_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_bias_to_fp16, dilations = var_5953, groups = var_5905, pad = value_13_pad_0, pad_type = value_13_pad_type_0, strides = var_5951, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor("value_13_cast_fp16")]; tensor var_5960_begin_0 = const()[name = tensor("op_5960_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_5960_end_0 = const()[name = tensor("op_5960_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_5960_end_mask_0 = const()[name = tensor("op_5960_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5960_cast_fp16 = slice_by_index(begin = var_5960_begin_0, end = var_5960_end_0, end_mask = var_5960_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5960_cast_fp16")]; tensor var_5964_begin_0 = const()[name = tensor("op_5964_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_5964_end_0 = const()[name = tensor("op_5964_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_5964_end_mask_0 = const()[name = tensor("op_5964_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5964_cast_fp16 = slice_by_index(begin = var_5964_begin_0, end = var_5964_end_0, end_mask = var_5964_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5964_cast_fp16")]; tensor var_5968_begin_0 = const()[name = tensor("op_5968_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_5968_end_0 = const()[name = tensor("op_5968_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_5968_end_mask_0 = const()[name = tensor("op_5968_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5968_cast_fp16 = slice_by_index(begin = var_5968_begin_0, end = var_5968_end_0, end_mask = var_5968_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5968_cast_fp16")]; tensor var_5972_begin_0 = const()[name = tensor("op_5972_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_5972_end_0 = const()[name = tensor("op_5972_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_5972_end_mask_0 = const()[name = tensor("op_5972_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5972_cast_fp16 = slice_by_index(begin = var_5972_begin_0, end = var_5972_end_0, end_mask = var_5972_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5972_cast_fp16")]; tensor var_5976_begin_0 = const()[name = tensor("op_5976_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_5976_end_0 = const()[name = tensor("op_5976_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_5976_end_mask_0 = const()[name = tensor("op_5976_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5976_cast_fp16 = slice_by_index(begin = var_5976_begin_0, end = var_5976_end_0, end_mask = var_5976_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5976_cast_fp16")]; tensor var_5980_begin_0 = const()[name = tensor("op_5980_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_5980_end_0 = const()[name = tensor("op_5980_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_5980_end_mask_0 = const()[name = tensor("op_5980_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5980_cast_fp16 = slice_by_index(begin = var_5980_begin_0, end = var_5980_end_0, end_mask = var_5980_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5980_cast_fp16")]; tensor var_5984_begin_0 = const()[name = tensor("op_5984_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_5984_end_0 = const()[name = tensor("op_5984_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_5984_end_mask_0 = const()[name = tensor("op_5984_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5984_cast_fp16 = slice_by_index(begin = var_5984_begin_0, end = var_5984_end_0, end_mask = var_5984_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5984_cast_fp16")]; tensor var_5988_begin_0 = const()[name = tensor("op_5988_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_5988_end_0 = const()[name = tensor("op_5988_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_5988_end_mask_0 = const()[name = tensor("op_5988_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5988_cast_fp16 = slice_by_index(begin = var_5988_begin_0, end = var_5988_end_0, end_mask = var_5988_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5988_cast_fp16")]; tensor var_5992_begin_0 = const()[name = tensor("op_5992_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_5992_end_0 = const()[name = tensor("op_5992_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_5992_end_mask_0 = const()[name = tensor("op_5992_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5992_cast_fp16 = slice_by_index(begin = var_5992_begin_0, end = var_5992_end_0, end_mask = var_5992_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5992_cast_fp16")]; tensor var_5996_begin_0 = const()[name = tensor("op_5996_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_5996_end_0 = const()[name = tensor("op_5996_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_5996_end_mask_0 = const()[name = tensor("op_5996_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_5996_cast_fp16 = slice_by_index(begin = var_5996_begin_0, end = var_5996_end_0, end_mask = var_5996_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_5996_cast_fp16")]; tensor var_6000_begin_0 = const()[name = tensor("op_6000_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6000_end_0 = const()[name = tensor("op_6000_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6000_end_mask_0 = const()[name = tensor("op_6000_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6000_cast_fp16 = slice_by_index(begin = var_6000_begin_0, end = var_6000_end_0, end_mask = var_6000_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_6000_cast_fp16")]; tensor var_6004_begin_0 = const()[name = tensor("op_6004_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6004_end_0 = const()[name = tensor("op_6004_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6004_end_mask_0 = const()[name = tensor("op_6004_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6004_cast_fp16 = slice_by_index(begin = var_6004_begin_0, end = var_6004_end_0, end_mask = var_6004_end_mask_0, x = query_13_cast_fp16)[name = tensor("op_6004_cast_fp16")]; tensor var_6013_begin_0 = const()[name = tensor("op_6013_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6013_end_0 = const()[name = tensor("op_6013_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6013_end_mask_0 = const()[name = tensor("op_6013_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6013_cast_fp16 = slice_by_index(begin = var_6013_begin_0, end = var_6013_end_0, end_mask = var_6013_end_mask_0, x = var_5960_cast_fp16)[name = tensor("op_6013_cast_fp16")]; tensor var_6020_begin_0 = const()[name = tensor("op_6020_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6020_end_0 = const()[name = tensor("op_6020_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6020_end_mask_0 = const()[name = tensor("op_6020_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6020_cast_fp16 = slice_by_index(begin = var_6020_begin_0, end = var_6020_end_0, end_mask = var_6020_end_mask_0, x = var_5960_cast_fp16)[name = tensor("op_6020_cast_fp16")]; tensor var_6027_begin_0 = const()[name = tensor("op_6027_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6027_end_0 = const()[name = tensor("op_6027_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6027_end_mask_0 = const()[name = tensor("op_6027_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6027_cast_fp16 = slice_by_index(begin = var_6027_begin_0, end = var_6027_end_0, end_mask = var_6027_end_mask_0, x = var_5960_cast_fp16)[name = tensor("op_6027_cast_fp16")]; tensor var_6034_begin_0 = const()[name = tensor("op_6034_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6034_end_0 = const()[name = tensor("op_6034_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6034_end_mask_0 = const()[name = tensor("op_6034_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6034_cast_fp16 = slice_by_index(begin = var_6034_begin_0, end = var_6034_end_0, end_mask = var_6034_end_mask_0, x = var_5960_cast_fp16)[name = tensor("op_6034_cast_fp16")]; tensor var_6041_begin_0 = const()[name = tensor("op_6041_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6041_end_0 = const()[name = tensor("op_6041_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6041_end_mask_0 = const()[name = tensor("op_6041_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6041_cast_fp16 = slice_by_index(begin = var_6041_begin_0, end = var_6041_end_0, end_mask = var_6041_end_mask_0, x = var_5964_cast_fp16)[name = tensor("op_6041_cast_fp16")]; tensor var_6048_begin_0 = const()[name = tensor("op_6048_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6048_end_0 = const()[name = tensor("op_6048_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6048_end_mask_0 = const()[name = tensor("op_6048_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6048_cast_fp16 = slice_by_index(begin = var_6048_begin_0, end = var_6048_end_0, end_mask = var_6048_end_mask_0, x = var_5964_cast_fp16)[name = tensor("op_6048_cast_fp16")]; tensor var_6055_begin_0 = const()[name = tensor("op_6055_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6055_end_0 = const()[name = tensor("op_6055_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6055_end_mask_0 = const()[name = tensor("op_6055_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6055_cast_fp16 = slice_by_index(begin = var_6055_begin_0, end = var_6055_end_0, end_mask = var_6055_end_mask_0, x = var_5964_cast_fp16)[name = tensor("op_6055_cast_fp16")]; tensor var_6062_begin_0 = const()[name = tensor("op_6062_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6062_end_0 = const()[name = tensor("op_6062_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6062_end_mask_0 = const()[name = tensor("op_6062_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6062_cast_fp16 = slice_by_index(begin = var_6062_begin_0, end = var_6062_end_0, end_mask = var_6062_end_mask_0, x = var_5964_cast_fp16)[name = tensor("op_6062_cast_fp16")]; tensor var_6069_begin_0 = const()[name = tensor("op_6069_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6069_end_0 = const()[name = tensor("op_6069_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6069_end_mask_0 = const()[name = tensor("op_6069_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6069_cast_fp16 = slice_by_index(begin = var_6069_begin_0, end = var_6069_end_0, end_mask = var_6069_end_mask_0, x = var_5968_cast_fp16)[name = tensor("op_6069_cast_fp16")]; tensor var_6076_begin_0 = const()[name = tensor("op_6076_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6076_end_0 = const()[name = tensor("op_6076_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6076_end_mask_0 = const()[name = tensor("op_6076_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6076_cast_fp16 = slice_by_index(begin = var_6076_begin_0, end = var_6076_end_0, end_mask = var_6076_end_mask_0, x = var_5968_cast_fp16)[name = tensor("op_6076_cast_fp16")]; tensor var_6083_begin_0 = const()[name = tensor("op_6083_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6083_end_0 = const()[name = tensor("op_6083_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6083_end_mask_0 = const()[name = tensor("op_6083_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6083_cast_fp16 = slice_by_index(begin = var_6083_begin_0, end = var_6083_end_0, end_mask = var_6083_end_mask_0, x = var_5968_cast_fp16)[name = tensor("op_6083_cast_fp16")]; tensor var_6090_begin_0 = const()[name = tensor("op_6090_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6090_end_0 = const()[name = tensor("op_6090_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6090_end_mask_0 = const()[name = tensor("op_6090_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6090_cast_fp16 = slice_by_index(begin = var_6090_begin_0, end = var_6090_end_0, end_mask = var_6090_end_mask_0, x = var_5968_cast_fp16)[name = tensor("op_6090_cast_fp16")]; tensor var_6097_begin_0 = const()[name = tensor("op_6097_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6097_end_0 = const()[name = tensor("op_6097_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6097_end_mask_0 = const()[name = tensor("op_6097_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6097_cast_fp16 = slice_by_index(begin = var_6097_begin_0, end = var_6097_end_0, end_mask = var_6097_end_mask_0, x = var_5972_cast_fp16)[name = tensor("op_6097_cast_fp16")]; tensor var_6104_begin_0 = const()[name = tensor("op_6104_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6104_end_0 = const()[name = tensor("op_6104_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6104_end_mask_0 = const()[name = tensor("op_6104_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6104_cast_fp16 = slice_by_index(begin = var_6104_begin_0, end = var_6104_end_0, end_mask = var_6104_end_mask_0, x = var_5972_cast_fp16)[name = tensor("op_6104_cast_fp16")]; tensor var_6111_begin_0 = const()[name = tensor("op_6111_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6111_end_0 = const()[name = tensor("op_6111_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6111_end_mask_0 = const()[name = tensor("op_6111_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6111_cast_fp16 = slice_by_index(begin = var_6111_begin_0, end = var_6111_end_0, end_mask = var_6111_end_mask_0, x = var_5972_cast_fp16)[name = tensor("op_6111_cast_fp16")]; tensor var_6118_begin_0 = const()[name = tensor("op_6118_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6118_end_0 = const()[name = tensor("op_6118_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6118_end_mask_0 = const()[name = tensor("op_6118_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6118_cast_fp16 = slice_by_index(begin = var_6118_begin_0, end = var_6118_end_0, end_mask = var_6118_end_mask_0, x = var_5972_cast_fp16)[name = tensor("op_6118_cast_fp16")]; tensor var_6125_begin_0 = const()[name = tensor("op_6125_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6125_end_0 = const()[name = tensor("op_6125_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6125_end_mask_0 = const()[name = tensor("op_6125_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6125_cast_fp16 = slice_by_index(begin = var_6125_begin_0, end = var_6125_end_0, end_mask = var_6125_end_mask_0, x = var_5976_cast_fp16)[name = tensor("op_6125_cast_fp16")]; tensor var_6132_begin_0 = const()[name = tensor("op_6132_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6132_end_0 = const()[name = tensor("op_6132_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6132_end_mask_0 = const()[name = tensor("op_6132_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6132_cast_fp16 = slice_by_index(begin = var_6132_begin_0, end = var_6132_end_0, end_mask = var_6132_end_mask_0, x = var_5976_cast_fp16)[name = tensor("op_6132_cast_fp16")]; tensor var_6139_begin_0 = const()[name = tensor("op_6139_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6139_end_0 = const()[name = tensor("op_6139_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6139_end_mask_0 = const()[name = tensor("op_6139_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6139_cast_fp16 = slice_by_index(begin = var_6139_begin_0, end = var_6139_end_0, end_mask = var_6139_end_mask_0, x = var_5976_cast_fp16)[name = tensor("op_6139_cast_fp16")]; tensor var_6146_begin_0 = const()[name = tensor("op_6146_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6146_end_0 = const()[name = tensor("op_6146_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6146_end_mask_0 = const()[name = tensor("op_6146_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6146_cast_fp16 = slice_by_index(begin = var_6146_begin_0, end = var_6146_end_0, end_mask = var_6146_end_mask_0, x = var_5976_cast_fp16)[name = tensor("op_6146_cast_fp16")]; tensor var_6153_begin_0 = const()[name = tensor("op_6153_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6153_end_0 = const()[name = tensor("op_6153_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6153_end_mask_0 = const()[name = tensor("op_6153_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6153_cast_fp16 = slice_by_index(begin = var_6153_begin_0, end = var_6153_end_0, end_mask = var_6153_end_mask_0, x = var_5980_cast_fp16)[name = tensor("op_6153_cast_fp16")]; tensor var_6160_begin_0 = const()[name = tensor("op_6160_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6160_end_0 = const()[name = tensor("op_6160_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6160_end_mask_0 = const()[name = tensor("op_6160_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6160_cast_fp16 = slice_by_index(begin = var_6160_begin_0, end = var_6160_end_0, end_mask = var_6160_end_mask_0, x = var_5980_cast_fp16)[name = tensor("op_6160_cast_fp16")]; tensor var_6167_begin_0 = const()[name = tensor("op_6167_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6167_end_0 = const()[name = tensor("op_6167_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6167_end_mask_0 = const()[name = tensor("op_6167_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6167_cast_fp16 = slice_by_index(begin = var_6167_begin_0, end = var_6167_end_0, end_mask = var_6167_end_mask_0, x = var_5980_cast_fp16)[name = tensor("op_6167_cast_fp16")]; tensor var_6174_begin_0 = const()[name = tensor("op_6174_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6174_end_0 = const()[name = tensor("op_6174_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6174_end_mask_0 = const()[name = tensor("op_6174_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6174_cast_fp16 = slice_by_index(begin = var_6174_begin_0, end = var_6174_end_0, end_mask = var_6174_end_mask_0, x = var_5980_cast_fp16)[name = tensor("op_6174_cast_fp16")]; tensor var_6181_begin_0 = const()[name = tensor("op_6181_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6181_end_0 = const()[name = tensor("op_6181_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6181_end_mask_0 = const()[name = tensor("op_6181_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6181_cast_fp16 = slice_by_index(begin = var_6181_begin_0, end = var_6181_end_0, end_mask = var_6181_end_mask_0, x = var_5984_cast_fp16)[name = tensor("op_6181_cast_fp16")]; tensor var_6188_begin_0 = const()[name = tensor("op_6188_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6188_end_0 = const()[name = tensor("op_6188_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6188_end_mask_0 = const()[name = tensor("op_6188_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6188_cast_fp16 = slice_by_index(begin = var_6188_begin_0, end = var_6188_end_0, end_mask = var_6188_end_mask_0, x = var_5984_cast_fp16)[name = tensor("op_6188_cast_fp16")]; tensor var_6195_begin_0 = const()[name = tensor("op_6195_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6195_end_0 = const()[name = tensor("op_6195_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6195_end_mask_0 = const()[name = tensor("op_6195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6195_cast_fp16 = slice_by_index(begin = var_6195_begin_0, end = var_6195_end_0, end_mask = var_6195_end_mask_0, x = var_5984_cast_fp16)[name = tensor("op_6195_cast_fp16")]; tensor var_6202_begin_0 = const()[name = tensor("op_6202_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6202_end_0 = const()[name = tensor("op_6202_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6202_end_mask_0 = const()[name = tensor("op_6202_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6202_cast_fp16 = slice_by_index(begin = var_6202_begin_0, end = var_6202_end_0, end_mask = var_6202_end_mask_0, x = var_5984_cast_fp16)[name = tensor("op_6202_cast_fp16")]; tensor var_6209_begin_0 = const()[name = tensor("op_6209_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6209_end_0 = const()[name = tensor("op_6209_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6209_end_mask_0 = const()[name = tensor("op_6209_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6209_cast_fp16 = slice_by_index(begin = var_6209_begin_0, end = var_6209_end_0, end_mask = var_6209_end_mask_0, x = var_5988_cast_fp16)[name = tensor("op_6209_cast_fp16")]; tensor var_6216_begin_0 = const()[name = tensor("op_6216_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6216_end_0 = const()[name = tensor("op_6216_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6216_end_mask_0 = const()[name = tensor("op_6216_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6216_cast_fp16 = slice_by_index(begin = var_6216_begin_0, end = var_6216_end_0, end_mask = var_6216_end_mask_0, x = var_5988_cast_fp16)[name = tensor("op_6216_cast_fp16")]; tensor var_6223_begin_0 = const()[name = tensor("op_6223_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6223_end_0 = const()[name = tensor("op_6223_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6223_end_mask_0 = const()[name = tensor("op_6223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6223_cast_fp16 = slice_by_index(begin = var_6223_begin_0, end = var_6223_end_0, end_mask = var_6223_end_mask_0, x = var_5988_cast_fp16)[name = tensor("op_6223_cast_fp16")]; tensor var_6230_begin_0 = const()[name = tensor("op_6230_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6230_end_0 = const()[name = tensor("op_6230_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6230_end_mask_0 = const()[name = tensor("op_6230_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6230_cast_fp16 = slice_by_index(begin = var_6230_begin_0, end = var_6230_end_0, end_mask = var_6230_end_mask_0, x = var_5988_cast_fp16)[name = tensor("op_6230_cast_fp16")]; tensor var_6237_begin_0 = const()[name = tensor("op_6237_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6237_end_0 = const()[name = tensor("op_6237_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6237_end_mask_0 = const()[name = tensor("op_6237_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6237_cast_fp16 = slice_by_index(begin = var_6237_begin_0, end = var_6237_end_0, end_mask = var_6237_end_mask_0, x = var_5992_cast_fp16)[name = tensor("op_6237_cast_fp16")]; tensor var_6244_begin_0 = const()[name = tensor("op_6244_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6244_end_0 = const()[name = tensor("op_6244_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6244_end_mask_0 = const()[name = tensor("op_6244_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6244_cast_fp16 = slice_by_index(begin = var_6244_begin_0, end = var_6244_end_0, end_mask = var_6244_end_mask_0, x = var_5992_cast_fp16)[name = tensor("op_6244_cast_fp16")]; tensor var_6251_begin_0 = const()[name = tensor("op_6251_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6251_end_0 = const()[name = tensor("op_6251_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6251_end_mask_0 = const()[name = tensor("op_6251_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6251_cast_fp16 = slice_by_index(begin = var_6251_begin_0, end = var_6251_end_0, end_mask = var_6251_end_mask_0, x = var_5992_cast_fp16)[name = tensor("op_6251_cast_fp16")]; tensor var_6258_begin_0 = const()[name = tensor("op_6258_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6258_end_0 = const()[name = tensor("op_6258_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6258_end_mask_0 = const()[name = tensor("op_6258_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6258_cast_fp16 = slice_by_index(begin = var_6258_begin_0, end = var_6258_end_0, end_mask = var_6258_end_mask_0, x = var_5992_cast_fp16)[name = tensor("op_6258_cast_fp16")]; tensor var_6265_begin_0 = const()[name = tensor("op_6265_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6265_end_0 = const()[name = tensor("op_6265_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6265_end_mask_0 = const()[name = tensor("op_6265_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6265_cast_fp16 = slice_by_index(begin = var_6265_begin_0, end = var_6265_end_0, end_mask = var_6265_end_mask_0, x = var_5996_cast_fp16)[name = tensor("op_6265_cast_fp16")]; tensor var_6272_begin_0 = const()[name = tensor("op_6272_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6272_end_0 = const()[name = tensor("op_6272_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6272_end_mask_0 = const()[name = tensor("op_6272_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6272_cast_fp16 = slice_by_index(begin = var_6272_begin_0, end = var_6272_end_0, end_mask = var_6272_end_mask_0, x = var_5996_cast_fp16)[name = tensor("op_6272_cast_fp16")]; tensor var_6279_begin_0 = const()[name = tensor("op_6279_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6279_end_0 = const()[name = tensor("op_6279_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6279_end_mask_0 = const()[name = tensor("op_6279_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6279_cast_fp16 = slice_by_index(begin = var_6279_begin_0, end = var_6279_end_0, end_mask = var_6279_end_mask_0, x = var_5996_cast_fp16)[name = tensor("op_6279_cast_fp16")]; tensor var_6286_begin_0 = const()[name = tensor("op_6286_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6286_end_0 = const()[name = tensor("op_6286_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6286_end_mask_0 = const()[name = tensor("op_6286_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6286_cast_fp16 = slice_by_index(begin = var_6286_begin_0, end = var_6286_end_0, end_mask = var_6286_end_mask_0, x = var_5996_cast_fp16)[name = tensor("op_6286_cast_fp16")]; tensor var_6293_begin_0 = const()[name = tensor("op_6293_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6293_end_0 = const()[name = tensor("op_6293_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6293_end_mask_0 = const()[name = tensor("op_6293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6293_cast_fp16 = slice_by_index(begin = var_6293_begin_0, end = var_6293_end_0, end_mask = var_6293_end_mask_0, x = var_6000_cast_fp16)[name = tensor("op_6293_cast_fp16")]; tensor var_6300_begin_0 = const()[name = tensor("op_6300_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6300_end_0 = const()[name = tensor("op_6300_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6300_end_mask_0 = const()[name = tensor("op_6300_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6300_cast_fp16 = slice_by_index(begin = var_6300_begin_0, end = var_6300_end_0, end_mask = var_6300_end_mask_0, x = var_6000_cast_fp16)[name = tensor("op_6300_cast_fp16")]; tensor var_6307_begin_0 = const()[name = tensor("op_6307_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6307_end_0 = const()[name = tensor("op_6307_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6307_end_mask_0 = const()[name = tensor("op_6307_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6307_cast_fp16 = slice_by_index(begin = var_6307_begin_0, end = var_6307_end_0, end_mask = var_6307_end_mask_0, x = var_6000_cast_fp16)[name = tensor("op_6307_cast_fp16")]; tensor var_6314_begin_0 = const()[name = tensor("op_6314_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6314_end_0 = const()[name = tensor("op_6314_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6314_end_mask_0 = const()[name = tensor("op_6314_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6314_cast_fp16 = slice_by_index(begin = var_6314_begin_0, end = var_6314_end_0, end_mask = var_6314_end_mask_0, x = var_6000_cast_fp16)[name = tensor("op_6314_cast_fp16")]; tensor var_6321_begin_0 = const()[name = tensor("op_6321_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6321_end_0 = const()[name = tensor("op_6321_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6321_end_mask_0 = const()[name = tensor("op_6321_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6321_cast_fp16 = slice_by_index(begin = var_6321_begin_0, end = var_6321_end_0, end_mask = var_6321_end_mask_0, x = var_6004_cast_fp16)[name = tensor("op_6321_cast_fp16")]; tensor var_6328_begin_0 = const()[name = tensor("op_6328_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6328_end_0 = const()[name = tensor("op_6328_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6328_end_mask_0 = const()[name = tensor("op_6328_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6328_cast_fp16 = slice_by_index(begin = var_6328_begin_0, end = var_6328_end_0, end_mask = var_6328_end_mask_0, x = var_6004_cast_fp16)[name = tensor("op_6328_cast_fp16")]; tensor var_6335_begin_0 = const()[name = tensor("op_6335_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6335_end_0 = const()[name = tensor("op_6335_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6335_end_mask_0 = const()[name = tensor("op_6335_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6335_cast_fp16 = slice_by_index(begin = var_6335_begin_0, end = var_6335_end_0, end_mask = var_6335_end_mask_0, x = var_6004_cast_fp16)[name = tensor("op_6335_cast_fp16")]; tensor var_6342_begin_0 = const()[name = tensor("op_6342_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6342_end_0 = const()[name = tensor("op_6342_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6342_end_mask_0 = const()[name = tensor("op_6342_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6342_cast_fp16 = slice_by_index(begin = var_6342_begin_0, end = var_6342_end_0, end_mask = var_6342_end_mask_0, x = var_6004_cast_fp16)[name = tensor("op_6342_cast_fp16")]; tensor k_13_perm_0 = const()[name = tensor("k_13_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_6347_begin_0 = const()[name = tensor("op_6347_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6347_end_0 = const()[name = tensor("op_6347_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_6347_end_mask_0 = const()[name = tensor("op_6347_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_5 = transpose(perm = k_13_perm_0, x = key_13_cast_fp16)[name = tensor("transpose_5")]; tensor var_6347_cast_fp16 = slice_by_index(begin = var_6347_begin_0, end = var_6347_end_0, end_mask = var_6347_end_mask_0, x = transpose_5)[name = tensor("op_6347_cast_fp16")]; tensor var_6351_begin_0 = const()[name = tensor("op_6351_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_6351_end_0 = const()[name = tensor("op_6351_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_6351_end_mask_0 = const()[name = tensor("op_6351_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6351_cast_fp16 = slice_by_index(begin = var_6351_begin_0, end = var_6351_end_0, end_mask = var_6351_end_mask_0, x = transpose_5)[name = tensor("op_6351_cast_fp16")]; tensor var_6355_begin_0 = const()[name = tensor("op_6355_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_6355_end_0 = const()[name = tensor("op_6355_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_6355_end_mask_0 = const()[name = tensor("op_6355_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6355_cast_fp16 = slice_by_index(begin = var_6355_begin_0, end = var_6355_end_0, end_mask = var_6355_end_mask_0, x = transpose_5)[name = tensor("op_6355_cast_fp16")]; tensor var_6359_begin_0 = const()[name = tensor("op_6359_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_6359_end_0 = const()[name = tensor("op_6359_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_6359_end_mask_0 = const()[name = tensor("op_6359_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6359_cast_fp16 = slice_by_index(begin = var_6359_begin_0, end = var_6359_end_0, end_mask = var_6359_end_mask_0, x = transpose_5)[name = tensor("op_6359_cast_fp16")]; tensor var_6363_begin_0 = const()[name = tensor("op_6363_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_6363_end_0 = const()[name = tensor("op_6363_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_6363_end_mask_0 = const()[name = tensor("op_6363_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6363_cast_fp16 = slice_by_index(begin = var_6363_begin_0, end = var_6363_end_0, end_mask = var_6363_end_mask_0, x = transpose_5)[name = tensor("op_6363_cast_fp16")]; tensor var_6367_begin_0 = const()[name = tensor("op_6367_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_6367_end_0 = const()[name = tensor("op_6367_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_6367_end_mask_0 = const()[name = tensor("op_6367_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6367_cast_fp16 = slice_by_index(begin = var_6367_begin_0, end = var_6367_end_0, end_mask = var_6367_end_mask_0, x = transpose_5)[name = tensor("op_6367_cast_fp16")]; tensor var_6371_begin_0 = const()[name = tensor("op_6371_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_6371_end_0 = const()[name = tensor("op_6371_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_6371_end_mask_0 = const()[name = tensor("op_6371_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6371_cast_fp16 = slice_by_index(begin = var_6371_begin_0, end = var_6371_end_0, end_mask = var_6371_end_mask_0, x = transpose_5)[name = tensor("op_6371_cast_fp16")]; tensor var_6375_begin_0 = const()[name = tensor("op_6375_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_6375_end_0 = const()[name = tensor("op_6375_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_6375_end_mask_0 = const()[name = tensor("op_6375_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6375_cast_fp16 = slice_by_index(begin = var_6375_begin_0, end = var_6375_end_0, end_mask = var_6375_end_mask_0, x = transpose_5)[name = tensor("op_6375_cast_fp16")]; tensor var_6379_begin_0 = const()[name = tensor("op_6379_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_6379_end_0 = const()[name = tensor("op_6379_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_6379_end_mask_0 = const()[name = tensor("op_6379_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6379_cast_fp16 = slice_by_index(begin = var_6379_begin_0, end = var_6379_end_0, end_mask = var_6379_end_mask_0, x = transpose_5)[name = tensor("op_6379_cast_fp16")]; tensor var_6383_begin_0 = const()[name = tensor("op_6383_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_6383_end_0 = const()[name = tensor("op_6383_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_6383_end_mask_0 = const()[name = tensor("op_6383_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6383_cast_fp16 = slice_by_index(begin = var_6383_begin_0, end = var_6383_end_0, end_mask = var_6383_end_mask_0, x = transpose_5)[name = tensor("op_6383_cast_fp16")]; tensor var_6387_begin_0 = const()[name = tensor("op_6387_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_6387_end_0 = const()[name = tensor("op_6387_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_6387_end_mask_0 = const()[name = tensor("op_6387_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6387_cast_fp16 = slice_by_index(begin = var_6387_begin_0, end = var_6387_end_0, end_mask = var_6387_end_mask_0, x = transpose_5)[name = tensor("op_6387_cast_fp16")]; tensor var_6391_begin_0 = const()[name = tensor("op_6391_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_6391_end_0 = const()[name = tensor("op_6391_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_6391_end_mask_0 = const()[name = tensor("op_6391_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6391_cast_fp16 = slice_by_index(begin = var_6391_begin_0, end = var_6391_end_0, end_mask = var_6391_end_mask_0, x = transpose_5)[name = tensor("op_6391_cast_fp16")]; tensor var_6393_begin_0 = const()[name = tensor("op_6393_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6393_end_0 = const()[name = tensor("op_6393_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6393_end_mask_0 = const()[name = tensor("op_6393_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6393_cast_fp16 = slice_by_index(begin = var_6393_begin_0, end = var_6393_end_0, end_mask = var_6393_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6393_cast_fp16")]; tensor var_6397_begin_0 = const()[name = tensor("op_6397_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6397_end_0 = const()[name = tensor("op_6397_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_6397_end_mask_0 = const()[name = tensor("op_6397_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6397_cast_fp16 = slice_by_index(begin = var_6397_begin_0, end = var_6397_end_0, end_mask = var_6397_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6397_cast_fp16")]; tensor var_6401_begin_0 = const()[name = tensor("op_6401_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6401_end_0 = const()[name = tensor("op_6401_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_6401_end_mask_0 = const()[name = tensor("op_6401_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6401_cast_fp16 = slice_by_index(begin = var_6401_begin_0, end = var_6401_end_0, end_mask = var_6401_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6401_cast_fp16")]; tensor var_6405_begin_0 = const()[name = tensor("op_6405_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6405_end_0 = const()[name = tensor("op_6405_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_6405_end_mask_0 = const()[name = tensor("op_6405_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6405_cast_fp16 = slice_by_index(begin = var_6405_begin_0, end = var_6405_end_0, end_mask = var_6405_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6405_cast_fp16")]; tensor var_6409_begin_0 = const()[name = tensor("op_6409_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6409_end_0 = const()[name = tensor("op_6409_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_6409_end_mask_0 = const()[name = tensor("op_6409_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6409_cast_fp16 = slice_by_index(begin = var_6409_begin_0, end = var_6409_end_0, end_mask = var_6409_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6409_cast_fp16")]; tensor var_6413_begin_0 = const()[name = tensor("op_6413_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6413_end_0 = const()[name = tensor("op_6413_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_6413_end_mask_0 = const()[name = tensor("op_6413_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6413_cast_fp16 = slice_by_index(begin = var_6413_begin_0, end = var_6413_end_0, end_mask = var_6413_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6413_cast_fp16")]; tensor var_6417_begin_0 = const()[name = tensor("op_6417_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6417_end_0 = const()[name = tensor("op_6417_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_6417_end_mask_0 = const()[name = tensor("op_6417_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6417_cast_fp16 = slice_by_index(begin = var_6417_begin_0, end = var_6417_end_0, end_mask = var_6417_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6417_cast_fp16")]; tensor var_6421_begin_0 = const()[name = tensor("op_6421_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6421_end_0 = const()[name = tensor("op_6421_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_6421_end_mask_0 = const()[name = tensor("op_6421_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6421_cast_fp16 = slice_by_index(begin = var_6421_begin_0, end = var_6421_end_0, end_mask = var_6421_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6421_cast_fp16")]; tensor var_6425_begin_0 = const()[name = tensor("op_6425_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6425_end_0 = const()[name = tensor("op_6425_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_6425_end_mask_0 = const()[name = tensor("op_6425_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6425_cast_fp16 = slice_by_index(begin = var_6425_begin_0, end = var_6425_end_0, end_mask = var_6425_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6425_cast_fp16")]; tensor var_6429_begin_0 = const()[name = tensor("op_6429_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6429_end_0 = const()[name = tensor("op_6429_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_6429_end_mask_0 = const()[name = tensor("op_6429_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6429_cast_fp16 = slice_by_index(begin = var_6429_begin_0, end = var_6429_end_0, end_mask = var_6429_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6429_cast_fp16")]; tensor var_6433_begin_0 = const()[name = tensor("op_6433_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6433_end_0 = const()[name = tensor("op_6433_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6433_end_mask_0 = const()[name = tensor("op_6433_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6433_cast_fp16 = slice_by_index(begin = var_6433_begin_0, end = var_6433_end_0, end_mask = var_6433_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6433_cast_fp16")]; tensor var_6437_begin_0 = const()[name = tensor("op_6437_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6437_end_0 = const()[name = tensor("op_6437_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6437_end_mask_0 = const()[name = tensor("op_6437_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6437_cast_fp16 = slice_by_index(begin = var_6437_begin_0, end = var_6437_end_0, end_mask = var_6437_end_mask_0, x = value_13_cast_fp16)[name = tensor("op_6437_cast_fp16")]; tensor var_6441_equation_0 = const()[name = tensor("op_6441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6441_cast_fp16 = einsum(equation = var_6441_equation_0, values = (var_6347_cast_fp16, var_6013_cast_fp16))[name = tensor("op_6441_cast_fp16")]; tensor var_6442_to_fp16 = const()[name = tensor("op_6442_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_577_cast_fp16 = mul(x = var_6441_cast_fp16, y = var_6442_to_fp16)[name = tensor("aw_chunk_577_cast_fp16")]; tensor var_6445_equation_0 = const()[name = tensor("op_6445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6445_cast_fp16 = einsum(equation = var_6445_equation_0, values = (var_6347_cast_fp16, var_6020_cast_fp16))[name = tensor("op_6445_cast_fp16")]; tensor var_6446_to_fp16 = const()[name = tensor("op_6446_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_579_cast_fp16 = mul(x = var_6445_cast_fp16, y = var_6446_to_fp16)[name = tensor("aw_chunk_579_cast_fp16")]; tensor var_6449_equation_0 = const()[name = tensor("op_6449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6449_cast_fp16 = einsum(equation = var_6449_equation_0, values = (var_6347_cast_fp16, var_6027_cast_fp16))[name = tensor("op_6449_cast_fp16")]; tensor var_6450_to_fp16 = const()[name = tensor("op_6450_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_581_cast_fp16 = mul(x = var_6449_cast_fp16, y = var_6450_to_fp16)[name = tensor("aw_chunk_581_cast_fp16")]; tensor var_6453_equation_0 = const()[name = tensor("op_6453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6453_cast_fp16 = einsum(equation = var_6453_equation_0, values = (var_6347_cast_fp16, var_6034_cast_fp16))[name = tensor("op_6453_cast_fp16")]; tensor var_6454_to_fp16 = const()[name = tensor("op_6454_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_583_cast_fp16 = mul(x = var_6453_cast_fp16, y = var_6454_to_fp16)[name = tensor("aw_chunk_583_cast_fp16")]; tensor var_6457_equation_0 = const()[name = tensor("op_6457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6457_cast_fp16 = einsum(equation = var_6457_equation_0, values = (var_6351_cast_fp16, var_6041_cast_fp16))[name = tensor("op_6457_cast_fp16")]; tensor var_6458_to_fp16 = const()[name = tensor("op_6458_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_585_cast_fp16 = mul(x = var_6457_cast_fp16, y = var_6458_to_fp16)[name = tensor("aw_chunk_585_cast_fp16")]; tensor var_6461_equation_0 = const()[name = tensor("op_6461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6461_cast_fp16 = einsum(equation = var_6461_equation_0, values = (var_6351_cast_fp16, var_6048_cast_fp16))[name = tensor("op_6461_cast_fp16")]; tensor var_6462_to_fp16 = const()[name = tensor("op_6462_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_587_cast_fp16 = mul(x = var_6461_cast_fp16, y = var_6462_to_fp16)[name = tensor("aw_chunk_587_cast_fp16")]; tensor var_6465_equation_0 = const()[name = tensor("op_6465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6465_cast_fp16 = einsum(equation = var_6465_equation_0, values = (var_6351_cast_fp16, var_6055_cast_fp16))[name = tensor("op_6465_cast_fp16")]; tensor var_6466_to_fp16 = const()[name = tensor("op_6466_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_589_cast_fp16 = mul(x = var_6465_cast_fp16, y = var_6466_to_fp16)[name = tensor("aw_chunk_589_cast_fp16")]; tensor var_6469_equation_0 = const()[name = tensor("op_6469_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6469_cast_fp16 = einsum(equation = var_6469_equation_0, values = (var_6351_cast_fp16, var_6062_cast_fp16))[name = tensor("op_6469_cast_fp16")]; tensor var_6470_to_fp16 = const()[name = tensor("op_6470_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_591_cast_fp16 = mul(x = var_6469_cast_fp16, y = var_6470_to_fp16)[name = tensor("aw_chunk_591_cast_fp16")]; tensor var_6473_equation_0 = const()[name = tensor("op_6473_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6473_cast_fp16 = einsum(equation = var_6473_equation_0, values = (var_6355_cast_fp16, var_6069_cast_fp16))[name = tensor("op_6473_cast_fp16")]; tensor var_6474_to_fp16 = const()[name = tensor("op_6474_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_593_cast_fp16 = mul(x = var_6473_cast_fp16, y = var_6474_to_fp16)[name = tensor("aw_chunk_593_cast_fp16")]; tensor var_6477_equation_0 = const()[name = tensor("op_6477_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6477_cast_fp16 = einsum(equation = var_6477_equation_0, values = (var_6355_cast_fp16, var_6076_cast_fp16))[name = tensor("op_6477_cast_fp16")]; tensor var_6478_to_fp16 = const()[name = tensor("op_6478_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_595_cast_fp16 = mul(x = var_6477_cast_fp16, y = var_6478_to_fp16)[name = tensor("aw_chunk_595_cast_fp16")]; tensor var_6481_equation_0 = const()[name = tensor("op_6481_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6481_cast_fp16 = einsum(equation = var_6481_equation_0, values = (var_6355_cast_fp16, var_6083_cast_fp16))[name = tensor("op_6481_cast_fp16")]; tensor var_6482_to_fp16 = const()[name = tensor("op_6482_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_597_cast_fp16 = mul(x = var_6481_cast_fp16, y = var_6482_to_fp16)[name = tensor("aw_chunk_597_cast_fp16")]; tensor var_6485_equation_0 = const()[name = tensor("op_6485_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6485_cast_fp16 = einsum(equation = var_6485_equation_0, values = (var_6355_cast_fp16, var_6090_cast_fp16))[name = tensor("op_6485_cast_fp16")]; tensor var_6486_to_fp16 = const()[name = tensor("op_6486_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_599_cast_fp16 = mul(x = var_6485_cast_fp16, y = var_6486_to_fp16)[name = tensor("aw_chunk_599_cast_fp16")]; tensor var_6489_equation_0 = const()[name = tensor("op_6489_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6489_cast_fp16 = einsum(equation = var_6489_equation_0, values = (var_6359_cast_fp16, var_6097_cast_fp16))[name = tensor("op_6489_cast_fp16")]; tensor var_6490_to_fp16 = const()[name = tensor("op_6490_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_601_cast_fp16 = mul(x = var_6489_cast_fp16, y = var_6490_to_fp16)[name = tensor("aw_chunk_601_cast_fp16")]; tensor var_6493_equation_0 = const()[name = tensor("op_6493_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6493_cast_fp16 = einsum(equation = var_6493_equation_0, values = (var_6359_cast_fp16, var_6104_cast_fp16))[name = tensor("op_6493_cast_fp16")]; tensor var_6494_to_fp16 = const()[name = tensor("op_6494_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_603_cast_fp16 = mul(x = var_6493_cast_fp16, y = var_6494_to_fp16)[name = tensor("aw_chunk_603_cast_fp16")]; tensor var_6497_equation_0 = const()[name = tensor("op_6497_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6497_cast_fp16 = einsum(equation = var_6497_equation_0, values = (var_6359_cast_fp16, var_6111_cast_fp16))[name = tensor("op_6497_cast_fp16")]; tensor var_6498_to_fp16 = const()[name = tensor("op_6498_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_605_cast_fp16 = mul(x = var_6497_cast_fp16, y = var_6498_to_fp16)[name = tensor("aw_chunk_605_cast_fp16")]; tensor var_6501_equation_0 = const()[name = tensor("op_6501_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6501_cast_fp16 = einsum(equation = var_6501_equation_0, values = (var_6359_cast_fp16, var_6118_cast_fp16))[name = tensor("op_6501_cast_fp16")]; tensor var_6502_to_fp16 = const()[name = tensor("op_6502_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_607_cast_fp16 = mul(x = var_6501_cast_fp16, y = var_6502_to_fp16)[name = tensor("aw_chunk_607_cast_fp16")]; tensor var_6505_equation_0 = const()[name = tensor("op_6505_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6505_cast_fp16 = einsum(equation = var_6505_equation_0, values = (var_6363_cast_fp16, var_6125_cast_fp16))[name = tensor("op_6505_cast_fp16")]; tensor var_6506_to_fp16 = const()[name = tensor("op_6506_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_609_cast_fp16 = mul(x = var_6505_cast_fp16, y = var_6506_to_fp16)[name = tensor("aw_chunk_609_cast_fp16")]; tensor var_6509_equation_0 = const()[name = tensor("op_6509_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6509_cast_fp16 = einsum(equation = var_6509_equation_0, values = (var_6363_cast_fp16, var_6132_cast_fp16))[name = tensor("op_6509_cast_fp16")]; tensor var_6510_to_fp16 = const()[name = tensor("op_6510_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_611_cast_fp16 = mul(x = var_6509_cast_fp16, y = var_6510_to_fp16)[name = tensor("aw_chunk_611_cast_fp16")]; tensor var_6513_equation_0 = const()[name = tensor("op_6513_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6513_cast_fp16 = einsum(equation = var_6513_equation_0, values = (var_6363_cast_fp16, var_6139_cast_fp16))[name = tensor("op_6513_cast_fp16")]; tensor var_6514_to_fp16 = const()[name = tensor("op_6514_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_613_cast_fp16 = mul(x = var_6513_cast_fp16, y = var_6514_to_fp16)[name = tensor("aw_chunk_613_cast_fp16")]; tensor var_6517_equation_0 = const()[name = tensor("op_6517_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6517_cast_fp16 = einsum(equation = var_6517_equation_0, values = (var_6363_cast_fp16, var_6146_cast_fp16))[name = tensor("op_6517_cast_fp16")]; tensor var_6518_to_fp16 = const()[name = tensor("op_6518_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_615_cast_fp16 = mul(x = var_6517_cast_fp16, y = var_6518_to_fp16)[name = tensor("aw_chunk_615_cast_fp16")]; tensor var_6521_equation_0 = const()[name = tensor("op_6521_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6521_cast_fp16 = einsum(equation = var_6521_equation_0, values = (var_6367_cast_fp16, var_6153_cast_fp16))[name = tensor("op_6521_cast_fp16")]; tensor var_6522_to_fp16 = const()[name = tensor("op_6522_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_617_cast_fp16 = mul(x = var_6521_cast_fp16, y = var_6522_to_fp16)[name = tensor("aw_chunk_617_cast_fp16")]; tensor var_6525_equation_0 = const()[name = tensor("op_6525_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6525_cast_fp16 = einsum(equation = var_6525_equation_0, values = (var_6367_cast_fp16, var_6160_cast_fp16))[name = tensor("op_6525_cast_fp16")]; tensor var_6526_to_fp16 = const()[name = tensor("op_6526_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_619_cast_fp16 = mul(x = var_6525_cast_fp16, y = var_6526_to_fp16)[name = tensor("aw_chunk_619_cast_fp16")]; tensor var_6529_equation_0 = const()[name = tensor("op_6529_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6529_cast_fp16 = einsum(equation = var_6529_equation_0, values = (var_6367_cast_fp16, var_6167_cast_fp16))[name = tensor("op_6529_cast_fp16")]; tensor var_6530_to_fp16 = const()[name = tensor("op_6530_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_621_cast_fp16 = mul(x = var_6529_cast_fp16, y = var_6530_to_fp16)[name = tensor("aw_chunk_621_cast_fp16")]; tensor var_6533_equation_0 = const()[name = tensor("op_6533_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6533_cast_fp16 = einsum(equation = var_6533_equation_0, values = (var_6367_cast_fp16, var_6174_cast_fp16))[name = tensor("op_6533_cast_fp16")]; tensor var_6534_to_fp16 = const()[name = tensor("op_6534_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_623_cast_fp16 = mul(x = var_6533_cast_fp16, y = var_6534_to_fp16)[name = tensor("aw_chunk_623_cast_fp16")]; tensor var_6537_equation_0 = const()[name = tensor("op_6537_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6537_cast_fp16 = einsum(equation = var_6537_equation_0, values = (var_6371_cast_fp16, var_6181_cast_fp16))[name = tensor("op_6537_cast_fp16")]; tensor var_6538_to_fp16 = const()[name = tensor("op_6538_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_625_cast_fp16 = mul(x = var_6537_cast_fp16, y = var_6538_to_fp16)[name = tensor("aw_chunk_625_cast_fp16")]; tensor var_6541_equation_0 = const()[name = tensor("op_6541_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6541_cast_fp16 = einsum(equation = var_6541_equation_0, values = (var_6371_cast_fp16, var_6188_cast_fp16))[name = tensor("op_6541_cast_fp16")]; tensor var_6542_to_fp16 = const()[name = tensor("op_6542_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_627_cast_fp16 = mul(x = var_6541_cast_fp16, y = var_6542_to_fp16)[name = tensor("aw_chunk_627_cast_fp16")]; tensor var_6545_equation_0 = const()[name = tensor("op_6545_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6545_cast_fp16 = einsum(equation = var_6545_equation_0, values = (var_6371_cast_fp16, var_6195_cast_fp16))[name = tensor("op_6545_cast_fp16")]; tensor var_6546_to_fp16 = const()[name = tensor("op_6546_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_629_cast_fp16 = mul(x = var_6545_cast_fp16, y = var_6546_to_fp16)[name = tensor("aw_chunk_629_cast_fp16")]; tensor var_6549_equation_0 = const()[name = tensor("op_6549_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6549_cast_fp16 = einsum(equation = var_6549_equation_0, values = (var_6371_cast_fp16, var_6202_cast_fp16))[name = tensor("op_6549_cast_fp16")]; tensor var_6550_to_fp16 = const()[name = tensor("op_6550_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_631_cast_fp16 = mul(x = var_6549_cast_fp16, y = var_6550_to_fp16)[name = tensor("aw_chunk_631_cast_fp16")]; tensor var_6553_equation_0 = const()[name = tensor("op_6553_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6553_cast_fp16 = einsum(equation = var_6553_equation_0, values = (var_6375_cast_fp16, var_6209_cast_fp16))[name = tensor("op_6553_cast_fp16")]; tensor var_6554_to_fp16 = const()[name = tensor("op_6554_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_633_cast_fp16 = mul(x = var_6553_cast_fp16, y = var_6554_to_fp16)[name = tensor("aw_chunk_633_cast_fp16")]; tensor var_6557_equation_0 = const()[name = tensor("op_6557_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6557_cast_fp16 = einsum(equation = var_6557_equation_0, values = (var_6375_cast_fp16, var_6216_cast_fp16))[name = tensor("op_6557_cast_fp16")]; tensor var_6558_to_fp16 = const()[name = tensor("op_6558_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_635_cast_fp16 = mul(x = var_6557_cast_fp16, y = var_6558_to_fp16)[name = tensor("aw_chunk_635_cast_fp16")]; tensor var_6561_equation_0 = const()[name = tensor("op_6561_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6561_cast_fp16 = einsum(equation = var_6561_equation_0, values = (var_6375_cast_fp16, var_6223_cast_fp16))[name = tensor("op_6561_cast_fp16")]; tensor var_6562_to_fp16 = const()[name = tensor("op_6562_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_637_cast_fp16 = mul(x = var_6561_cast_fp16, y = var_6562_to_fp16)[name = tensor("aw_chunk_637_cast_fp16")]; tensor var_6565_equation_0 = const()[name = tensor("op_6565_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6565_cast_fp16 = einsum(equation = var_6565_equation_0, values = (var_6375_cast_fp16, var_6230_cast_fp16))[name = tensor("op_6565_cast_fp16")]; tensor var_6566_to_fp16 = const()[name = tensor("op_6566_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_639_cast_fp16 = mul(x = var_6565_cast_fp16, y = var_6566_to_fp16)[name = tensor("aw_chunk_639_cast_fp16")]; tensor var_6569_equation_0 = const()[name = tensor("op_6569_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6569_cast_fp16 = einsum(equation = var_6569_equation_0, values = (var_6379_cast_fp16, var_6237_cast_fp16))[name = tensor("op_6569_cast_fp16")]; tensor var_6570_to_fp16 = const()[name = tensor("op_6570_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_641_cast_fp16 = mul(x = var_6569_cast_fp16, y = var_6570_to_fp16)[name = tensor("aw_chunk_641_cast_fp16")]; tensor var_6573_equation_0 = const()[name = tensor("op_6573_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6573_cast_fp16 = einsum(equation = var_6573_equation_0, values = (var_6379_cast_fp16, var_6244_cast_fp16))[name = tensor("op_6573_cast_fp16")]; tensor var_6574_to_fp16 = const()[name = tensor("op_6574_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_643_cast_fp16 = mul(x = var_6573_cast_fp16, y = var_6574_to_fp16)[name = tensor("aw_chunk_643_cast_fp16")]; tensor var_6577_equation_0 = const()[name = tensor("op_6577_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6577_cast_fp16 = einsum(equation = var_6577_equation_0, values = (var_6379_cast_fp16, var_6251_cast_fp16))[name = tensor("op_6577_cast_fp16")]; tensor var_6578_to_fp16 = const()[name = tensor("op_6578_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_645_cast_fp16 = mul(x = var_6577_cast_fp16, y = var_6578_to_fp16)[name = tensor("aw_chunk_645_cast_fp16")]; tensor var_6581_equation_0 = const()[name = tensor("op_6581_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6581_cast_fp16 = einsum(equation = var_6581_equation_0, values = (var_6379_cast_fp16, var_6258_cast_fp16))[name = tensor("op_6581_cast_fp16")]; tensor var_6582_to_fp16 = const()[name = tensor("op_6582_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_647_cast_fp16 = mul(x = var_6581_cast_fp16, y = var_6582_to_fp16)[name = tensor("aw_chunk_647_cast_fp16")]; tensor var_6585_equation_0 = const()[name = tensor("op_6585_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6585_cast_fp16 = einsum(equation = var_6585_equation_0, values = (var_6383_cast_fp16, var_6265_cast_fp16))[name = tensor("op_6585_cast_fp16")]; tensor var_6586_to_fp16 = const()[name = tensor("op_6586_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_649_cast_fp16 = mul(x = var_6585_cast_fp16, y = var_6586_to_fp16)[name = tensor("aw_chunk_649_cast_fp16")]; tensor var_6589_equation_0 = const()[name = tensor("op_6589_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6589_cast_fp16 = einsum(equation = var_6589_equation_0, values = (var_6383_cast_fp16, var_6272_cast_fp16))[name = tensor("op_6589_cast_fp16")]; tensor var_6590_to_fp16 = const()[name = tensor("op_6590_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_651_cast_fp16 = mul(x = var_6589_cast_fp16, y = var_6590_to_fp16)[name = tensor("aw_chunk_651_cast_fp16")]; tensor var_6593_equation_0 = const()[name = tensor("op_6593_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6593_cast_fp16 = einsum(equation = var_6593_equation_0, values = (var_6383_cast_fp16, var_6279_cast_fp16))[name = tensor("op_6593_cast_fp16")]; tensor var_6594_to_fp16 = const()[name = tensor("op_6594_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_653_cast_fp16 = mul(x = var_6593_cast_fp16, y = var_6594_to_fp16)[name = tensor("aw_chunk_653_cast_fp16")]; tensor var_6597_equation_0 = const()[name = tensor("op_6597_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6597_cast_fp16 = einsum(equation = var_6597_equation_0, values = (var_6383_cast_fp16, var_6286_cast_fp16))[name = tensor("op_6597_cast_fp16")]; tensor var_6598_to_fp16 = const()[name = tensor("op_6598_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_655_cast_fp16 = mul(x = var_6597_cast_fp16, y = var_6598_to_fp16)[name = tensor("aw_chunk_655_cast_fp16")]; tensor var_6601_equation_0 = const()[name = tensor("op_6601_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6601_cast_fp16 = einsum(equation = var_6601_equation_0, values = (var_6387_cast_fp16, var_6293_cast_fp16))[name = tensor("op_6601_cast_fp16")]; tensor var_6602_to_fp16 = const()[name = tensor("op_6602_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_657_cast_fp16 = mul(x = var_6601_cast_fp16, y = var_6602_to_fp16)[name = tensor("aw_chunk_657_cast_fp16")]; tensor var_6605_equation_0 = const()[name = tensor("op_6605_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6605_cast_fp16 = einsum(equation = var_6605_equation_0, values = (var_6387_cast_fp16, var_6300_cast_fp16))[name = tensor("op_6605_cast_fp16")]; tensor var_6606_to_fp16 = const()[name = tensor("op_6606_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_659_cast_fp16 = mul(x = var_6605_cast_fp16, y = var_6606_to_fp16)[name = tensor("aw_chunk_659_cast_fp16")]; tensor var_6609_equation_0 = const()[name = tensor("op_6609_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6609_cast_fp16 = einsum(equation = var_6609_equation_0, values = (var_6387_cast_fp16, var_6307_cast_fp16))[name = tensor("op_6609_cast_fp16")]; tensor var_6610_to_fp16 = const()[name = tensor("op_6610_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_661_cast_fp16 = mul(x = var_6609_cast_fp16, y = var_6610_to_fp16)[name = tensor("aw_chunk_661_cast_fp16")]; tensor var_6613_equation_0 = const()[name = tensor("op_6613_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6613_cast_fp16 = einsum(equation = var_6613_equation_0, values = (var_6387_cast_fp16, var_6314_cast_fp16))[name = tensor("op_6613_cast_fp16")]; tensor var_6614_to_fp16 = const()[name = tensor("op_6614_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_663_cast_fp16 = mul(x = var_6613_cast_fp16, y = var_6614_to_fp16)[name = tensor("aw_chunk_663_cast_fp16")]; tensor var_6617_equation_0 = const()[name = tensor("op_6617_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6617_cast_fp16 = einsum(equation = var_6617_equation_0, values = (var_6391_cast_fp16, var_6321_cast_fp16))[name = tensor("op_6617_cast_fp16")]; tensor var_6618_to_fp16 = const()[name = tensor("op_6618_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_665_cast_fp16 = mul(x = var_6617_cast_fp16, y = var_6618_to_fp16)[name = tensor("aw_chunk_665_cast_fp16")]; tensor var_6621_equation_0 = const()[name = tensor("op_6621_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6621_cast_fp16 = einsum(equation = var_6621_equation_0, values = (var_6391_cast_fp16, var_6328_cast_fp16))[name = tensor("op_6621_cast_fp16")]; tensor var_6622_to_fp16 = const()[name = tensor("op_6622_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_667_cast_fp16 = mul(x = var_6621_cast_fp16, y = var_6622_to_fp16)[name = tensor("aw_chunk_667_cast_fp16")]; tensor var_6625_equation_0 = const()[name = tensor("op_6625_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6625_cast_fp16 = einsum(equation = var_6625_equation_0, values = (var_6391_cast_fp16, var_6335_cast_fp16))[name = tensor("op_6625_cast_fp16")]; tensor var_6626_to_fp16 = const()[name = tensor("op_6626_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_669_cast_fp16 = mul(x = var_6625_cast_fp16, y = var_6626_to_fp16)[name = tensor("aw_chunk_669_cast_fp16")]; tensor var_6629_equation_0 = const()[name = tensor("op_6629_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_6629_cast_fp16 = einsum(equation = var_6629_equation_0, values = (var_6391_cast_fp16, var_6342_cast_fp16))[name = tensor("op_6629_cast_fp16")]; tensor var_6630_to_fp16 = const()[name = tensor("op_6630_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_671_cast_fp16 = mul(x = var_6629_cast_fp16, y = var_6630_to_fp16)[name = tensor("aw_chunk_671_cast_fp16")]; tensor var_6632_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_577_cast_fp16)[name = tensor("op_6632_cast_fp16")]; tensor var_6633_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_579_cast_fp16)[name = tensor("op_6633_cast_fp16")]; tensor var_6634_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_581_cast_fp16)[name = tensor("op_6634_cast_fp16")]; tensor var_6635_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_583_cast_fp16)[name = tensor("op_6635_cast_fp16")]; tensor var_6636_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_585_cast_fp16)[name = tensor("op_6636_cast_fp16")]; tensor var_6637_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_587_cast_fp16)[name = tensor("op_6637_cast_fp16")]; tensor var_6638_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_589_cast_fp16)[name = tensor("op_6638_cast_fp16")]; tensor var_6639_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_591_cast_fp16)[name = tensor("op_6639_cast_fp16")]; tensor var_6640_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_593_cast_fp16)[name = tensor("op_6640_cast_fp16")]; tensor var_6641_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_595_cast_fp16)[name = tensor("op_6641_cast_fp16")]; tensor var_6642_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_597_cast_fp16)[name = tensor("op_6642_cast_fp16")]; tensor var_6643_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_599_cast_fp16)[name = tensor("op_6643_cast_fp16")]; tensor var_6644_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_601_cast_fp16)[name = tensor("op_6644_cast_fp16")]; tensor var_6645_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_603_cast_fp16)[name = tensor("op_6645_cast_fp16")]; tensor var_6646_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_605_cast_fp16)[name = tensor("op_6646_cast_fp16")]; tensor var_6647_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_607_cast_fp16)[name = tensor("op_6647_cast_fp16")]; tensor var_6648_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_609_cast_fp16)[name = tensor("op_6648_cast_fp16")]; tensor var_6649_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_611_cast_fp16)[name = tensor("op_6649_cast_fp16")]; tensor var_6650_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_613_cast_fp16)[name = tensor("op_6650_cast_fp16")]; tensor var_6651_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_615_cast_fp16)[name = tensor("op_6651_cast_fp16")]; tensor var_6652_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_617_cast_fp16)[name = tensor("op_6652_cast_fp16")]; tensor var_6653_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_619_cast_fp16)[name = tensor("op_6653_cast_fp16")]; tensor var_6654_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_621_cast_fp16)[name = tensor("op_6654_cast_fp16")]; tensor var_6655_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_623_cast_fp16)[name = tensor("op_6655_cast_fp16")]; tensor var_6656_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_625_cast_fp16)[name = tensor("op_6656_cast_fp16")]; tensor var_6657_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_627_cast_fp16)[name = tensor("op_6657_cast_fp16")]; tensor var_6658_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_629_cast_fp16)[name = tensor("op_6658_cast_fp16")]; tensor var_6659_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_631_cast_fp16)[name = tensor("op_6659_cast_fp16")]; tensor var_6660_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_633_cast_fp16)[name = tensor("op_6660_cast_fp16")]; tensor var_6661_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_635_cast_fp16)[name = tensor("op_6661_cast_fp16")]; tensor var_6662_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_637_cast_fp16)[name = tensor("op_6662_cast_fp16")]; tensor var_6663_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_639_cast_fp16)[name = tensor("op_6663_cast_fp16")]; tensor var_6664_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_641_cast_fp16)[name = tensor("op_6664_cast_fp16")]; tensor var_6665_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_643_cast_fp16)[name = tensor("op_6665_cast_fp16")]; tensor var_6666_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_645_cast_fp16)[name = tensor("op_6666_cast_fp16")]; tensor var_6667_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_647_cast_fp16)[name = tensor("op_6667_cast_fp16")]; tensor var_6668_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_649_cast_fp16)[name = tensor("op_6668_cast_fp16")]; tensor var_6669_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_651_cast_fp16)[name = tensor("op_6669_cast_fp16")]; tensor var_6670_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_653_cast_fp16)[name = tensor("op_6670_cast_fp16")]; tensor var_6671_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_655_cast_fp16)[name = tensor("op_6671_cast_fp16")]; tensor var_6672_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_657_cast_fp16)[name = tensor("op_6672_cast_fp16")]; tensor var_6673_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_659_cast_fp16)[name = tensor("op_6673_cast_fp16")]; tensor var_6674_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_661_cast_fp16)[name = tensor("op_6674_cast_fp16")]; tensor var_6675_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_663_cast_fp16)[name = tensor("op_6675_cast_fp16")]; tensor var_6676_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_665_cast_fp16)[name = tensor("op_6676_cast_fp16")]; tensor var_6677_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_667_cast_fp16)[name = tensor("op_6677_cast_fp16")]; tensor var_6678_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_669_cast_fp16)[name = tensor("op_6678_cast_fp16")]; tensor var_6679_cast_fp16 = softmax(axis = var_5905, x = aw_chunk_671_cast_fp16)[name = tensor("op_6679_cast_fp16")]; tensor var_6681_equation_0 = const()[name = tensor("op_6681_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6681_cast_fp16 = einsum(equation = var_6681_equation_0, values = (var_6393_cast_fp16, var_6632_cast_fp16))[name = tensor("op_6681_cast_fp16")]; tensor var_6683_equation_0 = const()[name = tensor("op_6683_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6683_cast_fp16 = einsum(equation = var_6683_equation_0, values = (var_6393_cast_fp16, var_6633_cast_fp16))[name = tensor("op_6683_cast_fp16")]; tensor var_6685_equation_0 = const()[name = tensor("op_6685_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6685_cast_fp16 = einsum(equation = var_6685_equation_0, values = (var_6393_cast_fp16, var_6634_cast_fp16))[name = tensor("op_6685_cast_fp16")]; tensor var_6687_equation_0 = const()[name = tensor("op_6687_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6687_cast_fp16 = einsum(equation = var_6687_equation_0, values = (var_6393_cast_fp16, var_6635_cast_fp16))[name = tensor("op_6687_cast_fp16")]; tensor var_6689_equation_0 = const()[name = tensor("op_6689_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6689_cast_fp16 = einsum(equation = var_6689_equation_0, values = (var_6397_cast_fp16, var_6636_cast_fp16))[name = tensor("op_6689_cast_fp16")]; tensor var_6691_equation_0 = const()[name = tensor("op_6691_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6691_cast_fp16 = einsum(equation = var_6691_equation_0, values = (var_6397_cast_fp16, var_6637_cast_fp16))[name = tensor("op_6691_cast_fp16")]; tensor var_6693_equation_0 = const()[name = tensor("op_6693_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6693_cast_fp16 = einsum(equation = var_6693_equation_0, values = (var_6397_cast_fp16, var_6638_cast_fp16))[name = tensor("op_6693_cast_fp16")]; tensor var_6695_equation_0 = const()[name = tensor("op_6695_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6695_cast_fp16 = einsum(equation = var_6695_equation_0, values = (var_6397_cast_fp16, var_6639_cast_fp16))[name = tensor("op_6695_cast_fp16")]; tensor var_6697_equation_0 = const()[name = tensor("op_6697_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6697_cast_fp16 = einsum(equation = var_6697_equation_0, values = (var_6401_cast_fp16, var_6640_cast_fp16))[name = tensor("op_6697_cast_fp16")]; tensor var_6699_equation_0 = const()[name = tensor("op_6699_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6699_cast_fp16 = einsum(equation = var_6699_equation_0, values = (var_6401_cast_fp16, var_6641_cast_fp16))[name = tensor("op_6699_cast_fp16")]; tensor var_6701_equation_0 = const()[name = tensor("op_6701_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6701_cast_fp16 = einsum(equation = var_6701_equation_0, values = (var_6401_cast_fp16, var_6642_cast_fp16))[name = tensor("op_6701_cast_fp16")]; tensor var_6703_equation_0 = const()[name = tensor("op_6703_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6703_cast_fp16 = einsum(equation = var_6703_equation_0, values = (var_6401_cast_fp16, var_6643_cast_fp16))[name = tensor("op_6703_cast_fp16")]; tensor var_6705_equation_0 = const()[name = tensor("op_6705_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6705_cast_fp16 = einsum(equation = var_6705_equation_0, values = (var_6405_cast_fp16, var_6644_cast_fp16))[name = tensor("op_6705_cast_fp16")]; tensor var_6707_equation_0 = const()[name = tensor("op_6707_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6707_cast_fp16 = einsum(equation = var_6707_equation_0, values = (var_6405_cast_fp16, var_6645_cast_fp16))[name = tensor("op_6707_cast_fp16")]; tensor var_6709_equation_0 = const()[name = tensor("op_6709_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6709_cast_fp16 = einsum(equation = var_6709_equation_0, values = (var_6405_cast_fp16, var_6646_cast_fp16))[name = tensor("op_6709_cast_fp16")]; tensor var_6711_equation_0 = const()[name = tensor("op_6711_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6711_cast_fp16 = einsum(equation = var_6711_equation_0, values = (var_6405_cast_fp16, var_6647_cast_fp16))[name = tensor("op_6711_cast_fp16")]; tensor var_6713_equation_0 = const()[name = tensor("op_6713_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6713_cast_fp16 = einsum(equation = var_6713_equation_0, values = (var_6409_cast_fp16, var_6648_cast_fp16))[name = tensor("op_6713_cast_fp16")]; tensor var_6715_equation_0 = const()[name = tensor("op_6715_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6715_cast_fp16 = einsum(equation = var_6715_equation_0, values = (var_6409_cast_fp16, var_6649_cast_fp16))[name = tensor("op_6715_cast_fp16")]; tensor var_6717_equation_0 = const()[name = tensor("op_6717_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6717_cast_fp16 = einsum(equation = var_6717_equation_0, values = (var_6409_cast_fp16, var_6650_cast_fp16))[name = tensor("op_6717_cast_fp16")]; tensor var_6719_equation_0 = const()[name = tensor("op_6719_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6719_cast_fp16 = einsum(equation = var_6719_equation_0, values = (var_6409_cast_fp16, var_6651_cast_fp16))[name = tensor("op_6719_cast_fp16")]; tensor var_6721_equation_0 = const()[name = tensor("op_6721_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6721_cast_fp16 = einsum(equation = var_6721_equation_0, values = (var_6413_cast_fp16, var_6652_cast_fp16))[name = tensor("op_6721_cast_fp16")]; tensor var_6723_equation_0 = const()[name = tensor("op_6723_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6723_cast_fp16 = einsum(equation = var_6723_equation_0, values = (var_6413_cast_fp16, var_6653_cast_fp16))[name = tensor("op_6723_cast_fp16")]; tensor var_6725_equation_0 = const()[name = tensor("op_6725_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6725_cast_fp16 = einsum(equation = var_6725_equation_0, values = (var_6413_cast_fp16, var_6654_cast_fp16))[name = tensor("op_6725_cast_fp16")]; tensor var_6727_equation_0 = const()[name = tensor("op_6727_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6727_cast_fp16 = einsum(equation = var_6727_equation_0, values = (var_6413_cast_fp16, var_6655_cast_fp16))[name = tensor("op_6727_cast_fp16")]; tensor var_6729_equation_0 = const()[name = tensor("op_6729_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6729_cast_fp16 = einsum(equation = var_6729_equation_0, values = (var_6417_cast_fp16, var_6656_cast_fp16))[name = tensor("op_6729_cast_fp16")]; tensor var_6731_equation_0 = const()[name = tensor("op_6731_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6731_cast_fp16 = einsum(equation = var_6731_equation_0, values = (var_6417_cast_fp16, var_6657_cast_fp16))[name = tensor("op_6731_cast_fp16")]; tensor var_6733_equation_0 = const()[name = tensor("op_6733_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6733_cast_fp16 = einsum(equation = var_6733_equation_0, values = (var_6417_cast_fp16, var_6658_cast_fp16))[name = tensor("op_6733_cast_fp16")]; tensor var_6735_equation_0 = const()[name = tensor("op_6735_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6735_cast_fp16 = einsum(equation = var_6735_equation_0, values = (var_6417_cast_fp16, var_6659_cast_fp16))[name = tensor("op_6735_cast_fp16")]; tensor var_6737_equation_0 = const()[name = tensor("op_6737_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6737_cast_fp16 = einsum(equation = var_6737_equation_0, values = (var_6421_cast_fp16, var_6660_cast_fp16))[name = tensor("op_6737_cast_fp16")]; tensor var_6739_equation_0 = const()[name = tensor("op_6739_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6739_cast_fp16 = einsum(equation = var_6739_equation_0, values = (var_6421_cast_fp16, var_6661_cast_fp16))[name = tensor("op_6739_cast_fp16")]; tensor var_6741_equation_0 = const()[name = tensor("op_6741_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6741_cast_fp16 = einsum(equation = var_6741_equation_0, values = (var_6421_cast_fp16, var_6662_cast_fp16))[name = tensor("op_6741_cast_fp16")]; tensor var_6743_equation_0 = const()[name = tensor("op_6743_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6743_cast_fp16 = einsum(equation = var_6743_equation_0, values = (var_6421_cast_fp16, var_6663_cast_fp16))[name = tensor("op_6743_cast_fp16")]; tensor var_6745_equation_0 = const()[name = tensor("op_6745_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6745_cast_fp16 = einsum(equation = var_6745_equation_0, values = (var_6425_cast_fp16, var_6664_cast_fp16))[name = tensor("op_6745_cast_fp16")]; tensor var_6747_equation_0 = const()[name = tensor("op_6747_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6747_cast_fp16 = einsum(equation = var_6747_equation_0, values = (var_6425_cast_fp16, var_6665_cast_fp16))[name = tensor("op_6747_cast_fp16")]; tensor var_6749_equation_0 = const()[name = tensor("op_6749_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6749_cast_fp16 = einsum(equation = var_6749_equation_0, values = (var_6425_cast_fp16, var_6666_cast_fp16))[name = tensor("op_6749_cast_fp16")]; tensor var_6751_equation_0 = const()[name = tensor("op_6751_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6751_cast_fp16 = einsum(equation = var_6751_equation_0, values = (var_6425_cast_fp16, var_6667_cast_fp16))[name = tensor("op_6751_cast_fp16")]; tensor var_6753_equation_0 = const()[name = tensor("op_6753_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6753_cast_fp16 = einsum(equation = var_6753_equation_0, values = (var_6429_cast_fp16, var_6668_cast_fp16))[name = tensor("op_6753_cast_fp16")]; tensor var_6755_equation_0 = const()[name = tensor("op_6755_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6755_cast_fp16 = einsum(equation = var_6755_equation_0, values = (var_6429_cast_fp16, var_6669_cast_fp16))[name = tensor("op_6755_cast_fp16")]; tensor var_6757_equation_0 = const()[name = tensor("op_6757_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6757_cast_fp16 = einsum(equation = var_6757_equation_0, values = (var_6429_cast_fp16, var_6670_cast_fp16))[name = tensor("op_6757_cast_fp16")]; tensor var_6759_equation_0 = const()[name = tensor("op_6759_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6759_cast_fp16 = einsum(equation = var_6759_equation_0, values = (var_6429_cast_fp16, var_6671_cast_fp16))[name = tensor("op_6759_cast_fp16")]; tensor var_6761_equation_0 = const()[name = tensor("op_6761_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6761_cast_fp16 = einsum(equation = var_6761_equation_0, values = (var_6433_cast_fp16, var_6672_cast_fp16))[name = tensor("op_6761_cast_fp16")]; tensor var_6763_equation_0 = const()[name = tensor("op_6763_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6763_cast_fp16 = einsum(equation = var_6763_equation_0, values = (var_6433_cast_fp16, var_6673_cast_fp16))[name = tensor("op_6763_cast_fp16")]; tensor var_6765_equation_0 = const()[name = tensor("op_6765_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6765_cast_fp16 = einsum(equation = var_6765_equation_0, values = (var_6433_cast_fp16, var_6674_cast_fp16))[name = tensor("op_6765_cast_fp16")]; tensor var_6767_equation_0 = const()[name = tensor("op_6767_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6767_cast_fp16 = einsum(equation = var_6767_equation_0, values = (var_6433_cast_fp16, var_6675_cast_fp16))[name = tensor("op_6767_cast_fp16")]; tensor var_6769_equation_0 = const()[name = tensor("op_6769_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6769_cast_fp16 = einsum(equation = var_6769_equation_0, values = (var_6437_cast_fp16, var_6676_cast_fp16))[name = tensor("op_6769_cast_fp16")]; tensor var_6771_equation_0 = const()[name = tensor("op_6771_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6771_cast_fp16 = einsum(equation = var_6771_equation_0, values = (var_6437_cast_fp16, var_6677_cast_fp16))[name = tensor("op_6771_cast_fp16")]; tensor var_6773_equation_0 = const()[name = tensor("op_6773_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6773_cast_fp16 = einsum(equation = var_6773_equation_0, values = (var_6437_cast_fp16, var_6678_cast_fp16))[name = tensor("op_6773_cast_fp16")]; tensor var_6775_equation_0 = const()[name = tensor("op_6775_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_6775_cast_fp16 = einsum(equation = var_6775_equation_0, values = (var_6437_cast_fp16, var_6679_cast_fp16))[name = tensor("op_6775_cast_fp16")]; tensor var_6777_interleave_0 = const()[name = tensor("op_6777_interleave_0"), val = tensor(false)]; tensor var_6777_cast_fp16 = concat(axis = var_5888, interleave = var_6777_interleave_0, values = (var_6681_cast_fp16, var_6683_cast_fp16, var_6685_cast_fp16, var_6687_cast_fp16))[name = tensor("op_6777_cast_fp16")]; tensor var_6779_interleave_0 = const()[name = tensor("op_6779_interleave_0"), val = tensor(false)]; tensor var_6779_cast_fp16 = concat(axis = var_5888, interleave = var_6779_interleave_0, values = (var_6689_cast_fp16, var_6691_cast_fp16, var_6693_cast_fp16, var_6695_cast_fp16))[name = tensor("op_6779_cast_fp16")]; tensor var_6781_interleave_0 = const()[name = tensor("op_6781_interleave_0"), val = tensor(false)]; tensor var_6781_cast_fp16 = concat(axis = var_5888, interleave = var_6781_interleave_0, values = (var_6697_cast_fp16, var_6699_cast_fp16, var_6701_cast_fp16, var_6703_cast_fp16))[name = tensor("op_6781_cast_fp16")]; tensor var_6783_interleave_0 = const()[name = tensor("op_6783_interleave_0"), val = tensor(false)]; tensor var_6783_cast_fp16 = concat(axis = var_5888, interleave = var_6783_interleave_0, values = (var_6705_cast_fp16, var_6707_cast_fp16, var_6709_cast_fp16, var_6711_cast_fp16))[name = tensor("op_6783_cast_fp16")]; tensor var_6785_interleave_0 = const()[name = tensor("op_6785_interleave_0"), val = tensor(false)]; tensor var_6785_cast_fp16 = concat(axis = var_5888, interleave = var_6785_interleave_0, values = (var_6713_cast_fp16, var_6715_cast_fp16, var_6717_cast_fp16, var_6719_cast_fp16))[name = tensor("op_6785_cast_fp16")]; tensor var_6787_interleave_0 = const()[name = tensor("op_6787_interleave_0"), val = tensor(false)]; tensor var_6787_cast_fp16 = concat(axis = var_5888, interleave = var_6787_interleave_0, values = (var_6721_cast_fp16, var_6723_cast_fp16, var_6725_cast_fp16, var_6727_cast_fp16))[name = tensor("op_6787_cast_fp16")]; tensor var_6789_interleave_0 = const()[name = tensor("op_6789_interleave_0"), val = tensor(false)]; tensor var_6789_cast_fp16 = concat(axis = var_5888, interleave = var_6789_interleave_0, values = (var_6729_cast_fp16, var_6731_cast_fp16, var_6733_cast_fp16, var_6735_cast_fp16))[name = tensor("op_6789_cast_fp16")]; tensor var_6791_interleave_0 = const()[name = tensor("op_6791_interleave_0"), val = tensor(false)]; tensor var_6791_cast_fp16 = concat(axis = var_5888, interleave = var_6791_interleave_0, values = (var_6737_cast_fp16, var_6739_cast_fp16, var_6741_cast_fp16, var_6743_cast_fp16))[name = tensor("op_6791_cast_fp16")]; tensor var_6793_interleave_0 = const()[name = tensor("op_6793_interleave_0"), val = tensor(false)]; tensor var_6793_cast_fp16 = concat(axis = var_5888, interleave = var_6793_interleave_0, values = (var_6745_cast_fp16, var_6747_cast_fp16, var_6749_cast_fp16, var_6751_cast_fp16))[name = tensor("op_6793_cast_fp16")]; tensor var_6795_interleave_0 = const()[name = tensor("op_6795_interleave_0"), val = tensor(false)]; tensor var_6795_cast_fp16 = concat(axis = var_5888, interleave = var_6795_interleave_0, values = (var_6753_cast_fp16, var_6755_cast_fp16, var_6757_cast_fp16, var_6759_cast_fp16))[name = tensor("op_6795_cast_fp16")]; tensor var_6797_interleave_0 = const()[name = tensor("op_6797_interleave_0"), val = tensor(false)]; tensor var_6797_cast_fp16 = concat(axis = var_5888, interleave = var_6797_interleave_0, values = (var_6761_cast_fp16, var_6763_cast_fp16, var_6765_cast_fp16, var_6767_cast_fp16))[name = tensor("op_6797_cast_fp16")]; tensor var_6799_interleave_0 = const()[name = tensor("op_6799_interleave_0"), val = tensor(false)]; tensor var_6799_cast_fp16 = concat(axis = var_5888, interleave = var_6799_interleave_0, values = (var_6769_cast_fp16, var_6771_cast_fp16, var_6773_cast_fp16, var_6775_cast_fp16))[name = tensor("op_6799_cast_fp16")]; tensor input_49_interleave_0 = const()[name = tensor("input_49_interleave_0"), val = tensor(false)]; tensor input_49_cast_fp16 = concat(axis = var_5905, interleave = input_49_interleave_0, values = (var_6777_cast_fp16, var_6779_cast_fp16, var_6781_cast_fp16, var_6783_cast_fp16, var_6785_cast_fp16, var_6787_cast_fp16, var_6789_cast_fp16, var_6791_cast_fp16, var_6793_cast_fp16, var_6795_cast_fp16, var_6797_cast_fp16, var_6799_cast_fp16))[name = tensor("input_49_cast_fp16")]; tensor var_6804 = const()[name = tensor("op_6804"), val = tensor([1, 1])]; tensor var_6806 = const()[name = tensor("op_6806"), val = tensor([1, 1])]; tensor obj_27_pad_type_0 = const()[name = tensor("obj_27_pad_type_0"), val = tensor("custom")]; tensor obj_27_pad_0 = const()[name = tensor("obj_27_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94814784)))]; tensor layers_6_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95994496)))]; tensor obj_27_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_bias_to_fp16, dilations = var_6806, groups = var_5905, pad = obj_27_pad_0, pad_type = obj_27_pad_type_0, strides = var_6804, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_49_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_6812 = const()[name = tensor("op_6812"), val = tensor([1])]; tensor channels_mean_27_cast_fp16 = reduce_mean(axes = var_6812, keep_dims = var_5906, 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_6816 = const()[name = tensor("op_6816"), val = tensor([1])]; tensor var_6817_cast_fp16 = reduce_mean(axes = var_6816, keep_dims = var_5906, x = zero_mean_sq_27_cast_fp16)[name = tensor("op_6817_cast_fp16")]; tensor var_6818_to_fp16 = const()[name = tensor("op_6818_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6819_cast_fp16 = add(x = var_6817_cast_fp16, y = var_6818_to_fp16)[name = tensor("op_6819_cast_fp16")]; tensor denom_27_epsilon_0_to_fp16 = const()[name = tensor("denom_27_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_27_cast_fp16 = rsqrt(epsilon = denom_27_epsilon_0_to_fp16, x = var_6819_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_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(95996096)))]; 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(95997696)))]; 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_27_cast_fp16)[name = tensor("input_51_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_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_6_fc1_weight_to_fp16 = const()[name = tensor("layers_6_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95999296)))]; tensor layers_6_fc1_bias_to_fp16 = const()[name = tensor("layers_6_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100717952)))]; tensor input_53_cast_fp16 = conv(bias = layers_6_fc1_bias_to_fp16, dilations = var_6832, groups = var_5905, pad = input_53_pad_0, pad_type = input_53_pad_type_0, strides = var_6830, weight = layers_6_fc1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("input_53_cast_fp16")]; tensor input_55_mode_0 = const()[name = tensor("input_55_mode_0"), val = tensor("EXACT")]; tensor input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = tensor("input_55_cast_fp16")]; tensor var_6838 = const()[name = tensor("op_6838"), val = tensor([1, 1])]; tensor var_6840 = const()[name = tensor("op_6840"), val = tensor([1, 1])]; tensor hidden_states_17_pad_type_0 = const()[name = tensor("hidden_states_17_pad_type_0"), val = tensor("custom")]; tensor hidden_states_17_pad_0 = const()[name = tensor("hidden_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_6_fc2_weight_to_fp16 = const()[name = tensor("layers_6_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100724160)))]; tensor layers_6_fc2_bias_to_fp16 = const()[name = tensor("layers_6_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105442816)))]; tensor hidden_states_17_cast_fp16 = conv(bias = layers_6_fc2_bias_to_fp16, dilations = var_6840, groups = var_5905, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_6838, weight = layers_6_fc2_weight_to_fp16, x = input_55_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_6847 = const()[name = tensor("op_6847"), val = tensor(3)]; tensor var_6864 = const()[name = tensor("op_6864"), val = tensor(1)]; tensor var_6865 = const()[name = tensor("op_6865"), val = tensor(true)]; tensor var_6875 = const()[name = tensor("op_6875"), val = tensor([1])]; tensor channels_mean_29_cast_fp16 = reduce_mean(axes = var_6875, keep_dims = var_6865, 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_6879 = const()[name = tensor("op_6879"), val = tensor([1])]; tensor var_6880_cast_fp16 = reduce_mean(axes = var_6879, keep_dims = var_6865, x = zero_mean_sq_29_cast_fp16)[name = tensor("op_6880_cast_fp16")]; tensor var_6881_to_fp16 = const()[name = tensor("op_6881_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_6882_cast_fp16 = add(x = var_6880_cast_fp16, y = var_6881_to_fp16)[name = tensor("op_6882_cast_fp16")]; tensor denom_29_epsilon_0_to_fp16 = const()[name = tensor("denom_29_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_29_cast_fp16 = rsqrt(epsilon = denom_29_epsilon_0_to_fp16, x = var_6882_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(105444416)))]; 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(105446016)))]; 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_6897 = const()[name = tensor("op_6897"), val = tensor([1, 1])]; tensor var_6899 = const()[name = tensor("op_6899"), val = tensor([1, 1])]; tensor query_15_pad_type_0 = const()[name = tensor("query_15_pad_type_0"), val = tensor("custom")]; tensor query_15_pad_0 = const()[name = tensor("query_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105447616)))]; tensor layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106627328)))]; tensor query_15_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_bias_to_fp16, dilations = var_6899, groups = var_6864, pad = query_15_pad_0, pad_type = query_15_pad_type_0, strides = var_6897, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("query_15_cast_fp16")]; tensor var_6903 = const()[name = tensor("op_6903"), val = tensor([1, 1])]; tensor var_6905 = const()[name = tensor("op_6905"), val = tensor([1, 1])]; tensor key_15_pad_type_0 = const()[name = tensor("key_15_pad_type_0"), val = tensor("custom")]; tensor key_15_pad_0 = const()[name = tensor("key_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106628928)))]; tensor key_15_cast_fp16 = conv(dilations = var_6905, groups = var_6864, pad = key_15_pad_0, pad_type = key_15_pad_type_0, strides = var_6903, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("key_15_cast_fp16")]; tensor var_6910 = const()[name = tensor("op_6910"), val = tensor([1, 1])]; tensor var_6912 = const()[name = tensor("op_6912"), val = tensor([1, 1])]; tensor value_15_pad_type_0 = const()[name = tensor("value_15_pad_type_0"), val = tensor("custom")]; tensor value_15_pad_0 = const()[name = tensor("value_15_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107808640)))]; tensor layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108988352)))]; tensor value_15_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_bias_to_fp16, dilations = var_6912, groups = var_6864, pad = value_15_pad_0, pad_type = value_15_pad_type_0, strides = var_6910, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = obj_29_cast_fp16)[name = tensor("value_15_cast_fp16")]; tensor var_6919_begin_0 = const()[name = tensor("op_6919_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6919_end_0 = const()[name = tensor("op_6919_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6919_end_mask_0 = const()[name = tensor("op_6919_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6919_cast_fp16 = slice_by_index(begin = var_6919_begin_0, end = var_6919_end_0, end_mask = var_6919_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6919_cast_fp16")]; tensor var_6923_begin_0 = const()[name = tensor("op_6923_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_6923_end_0 = const()[name = tensor("op_6923_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_6923_end_mask_0 = const()[name = tensor("op_6923_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6923_cast_fp16 = slice_by_index(begin = var_6923_begin_0, end = var_6923_end_0, end_mask = var_6923_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6923_cast_fp16")]; tensor var_6927_begin_0 = const()[name = tensor("op_6927_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_6927_end_0 = const()[name = tensor("op_6927_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_6927_end_mask_0 = const()[name = tensor("op_6927_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6927_cast_fp16 = slice_by_index(begin = var_6927_begin_0, end = var_6927_end_0, end_mask = var_6927_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6927_cast_fp16")]; tensor var_6931_begin_0 = const()[name = tensor("op_6931_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_6931_end_0 = const()[name = tensor("op_6931_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_6931_end_mask_0 = const()[name = tensor("op_6931_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6931_cast_fp16 = slice_by_index(begin = var_6931_begin_0, end = var_6931_end_0, end_mask = var_6931_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6931_cast_fp16")]; tensor var_6935_begin_0 = const()[name = tensor("op_6935_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_6935_end_0 = const()[name = tensor("op_6935_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_6935_end_mask_0 = const()[name = tensor("op_6935_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6935_cast_fp16 = slice_by_index(begin = var_6935_begin_0, end = var_6935_end_0, end_mask = var_6935_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6935_cast_fp16")]; tensor var_6939_begin_0 = const()[name = tensor("op_6939_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_6939_end_0 = const()[name = tensor("op_6939_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_6939_end_mask_0 = const()[name = tensor("op_6939_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6939_cast_fp16 = slice_by_index(begin = var_6939_begin_0, end = var_6939_end_0, end_mask = var_6939_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6939_cast_fp16")]; tensor var_6943_begin_0 = const()[name = tensor("op_6943_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_6943_end_0 = const()[name = tensor("op_6943_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_6943_end_mask_0 = const()[name = tensor("op_6943_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6943_cast_fp16 = slice_by_index(begin = var_6943_begin_0, end = var_6943_end_0, end_mask = var_6943_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6943_cast_fp16")]; tensor var_6947_begin_0 = const()[name = tensor("op_6947_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_6947_end_0 = const()[name = tensor("op_6947_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_6947_end_mask_0 = const()[name = tensor("op_6947_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_15_cast_fp16)[name = tensor("op_6947_cast_fp16")]; tensor var_6951_begin_0 = const()[name = tensor("op_6951_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_6951_end_0 = const()[name = tensor("op_6951_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_6951_end_mask_0 = const()[name = tensor("op_6951_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6951_cast_fp16 = slice_by_index(begin = var_6951_begin_0, end = var_6951_end_0, end_mask = var_6951_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6951_cast_fp16")]; tensor var_6955_begin_0 = const()[name = tensor("op_6955_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_6955_end_0 = const()[name = tensor("op_6955_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_6955_end_mask_0 = const()[name = tensor("op_6955_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6955_cast_fp16 = slice_by_index(begin = var_6955_begin_0, end = var_6955_end_0, end_mask = var_6955_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6955_cast_fp16")]; tensor var_6959_begin_0 = const()[name = tensor("op_6959_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_6959_end_0 = const()[name = tensor("op_6959_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_6959_end_mask_0 = const()[name = tensor("op_6959_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6959_cast_fp16 = slice_by_index(begin = var_6959_begin_0, end = var_6959_end_0, end_mask = var_6959_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6959_cast_fp16")]; tensor var_6963_begin_0 = const()[name = tensor("op_6963_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_6963_end_0 = const()[name = tensor("op_6963_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_6963_end_mask_0 = const()[name = tensor("op_6963_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_6963_cast_fp16 = slice_by_index(begin = var_6963_begin_0, end = var_6963_end_0, end_mask = var_6963_end_mask_0, x = query_15_cast_fp16)[name = tensor("op_6963_cast_fp16")]; tensor var_6972_begin_0 = const()[name = tensor("op_6972_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_6972_end_0 = const()[name = tensor("op_6972_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_6972_end_mask_0 = const()[name = tensor("op_6972_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6972_cast_fp16 = slice_by_index(begin = var_6972_begin_0, end = var_6972_end_0, end_mask = var_6972_end_mask_0, x = var_6919_cast_fp16)[name = tensor("op_6972_cast_fp16")]; tensor var_6979_begin_0 = const()[name = tensor("op_6979_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_6979_end_0 = const()[name = tensor("op_6979_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_6979_end_mask_0 = const()[name = tensor("op_6979_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6979_cast_fp16 = slice_by_index(begin = var_6979_begin_0, end = var_6979_end_0, end_mask = var_6979_end_mask_0, x = var_6919_cast_fp16)[name = tensor("op_6979_cast_fp16")]; tensor var_6986_begin_0 = const()[name = tensor("op_6986_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_6986_end_0 = const()[name = tensor("op_6986_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_6986_end_mask_0 = const()[name = tensor("op_6986_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6986_cast_fp16 = slice_by_index(begin = var_6986_begin_0, end = var_6986_end_0, end_mask = var_6986_end_mask_0, x = var_6919_cast_fp16)[name = tensor("op_6986_cast_fp16")]; tensor var_6993_begin_0 = const()[name = tensor("op_6993_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_6993_end_0 = const()[name = tensor("op_6993_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_6993_end_mask_0 = const()[name = tensor("op_6993_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_6993_cast_fp16 = slice_by_index(begin = var_6993_begin_0, end = var_6993_end_0, end_mask = var_6993_end_mask_0, x = var_6919_cast_fp16)[name = tensor("op_6993_cast_fp16")]; tensor var_7000_begin_0 = const()[name = tensor("op_7000_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7000_end_0 = const()[name = tensor("op_7000_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7000_end_mask_0 = const()[name = tensor("op_7000_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7000_cast_fp16 = slice_by_index(begin = var_7000_begin_0, end = var_7000_end_0, end_mask = var_7000_end_mask_0, x = var_6923_cast_fp16)[name = tensor("op_7000_cast_fp16")]; tensor var_7007_begin_0 = const()[name = tensor("op_7007_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7007_end_0 = const()[name = tensor("op_7007_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7007_end_mask_0 = const()[name = tensor("op_7007_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7007_cast_fp16 = slice_by_index(begin = var_7007_begin_0, end = var_7007_end_0, end_mask = var_7007_end_mask_0, x = var_6923_cast_fp16)[name = tensor("op_7007_cast_fp16")]; tensor var_7014_begin_0 = const()[name = tensor("op_7014_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7014_end_0 = const()[name = tensor("op_7014_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7014_end_mask_0 = const()[name = tensor("op_7014_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7014_cast_fp16 = slice_by_index(begin = var_7014_begin_0, end = var_7014_end_0, end_mask = var_7014_end_mask_0, x = var_6923_cast_fp16)[name = tensor("op_7014_cast_fp16")]; tensor var_7021_begin_0 = const()[name = tensor("op_7021_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7021_end_0 = const()[name = tensor("op_7021_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7021_end_mask_0 = const()[name = tensor("op_7021_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7021_cast_fp16 = slice_by_index(begin = var_7021_begin_0, end = var_7021_end_0, end_mask = var_7021_end_mask_0, x = var_6923_cast_fp16)[name = tensor("op_7021_cast_fp16")]; tensor var_7028_begin_0 = const()[name = tensor("op_7028_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7028_end_0 = const()[name = tensor("op_7028_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7028_end_mask_0 = const()[name = tensor("op_7028_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7028_cast_fp16 = slice_by_index(begin = var_7028_begin_0, end = var_7028_end_0, end_mask = var_7028_end_mask_0, x = var_6927_cast_fp16)[name = tensor("op_7028_cast_fp16")]; tensor var_7035_begin_0 = const()[name = tensor("op_7035_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7035_end_0 = const()[name = tensor("op_7035_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7035_end_mask_0 = const()[name = tensor("op_7035_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7035_cast_fp16 = slice_by_index(begin = var_7035_begin_0, end = var_7035_end_0, end_mask = var_7035_end_mask_0, x = var_6927_cast_fp16)[name = tensor("op_7035_cast_fp16")]; tensor var_7042_begin_0 = const()[name = tensor("op_7042_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7042_end_0 = const()[name = tensor("op_7042_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7042_end_mask_0 = const()[name = tensor("op_7042_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7042_cast_fp16 = slice_by_index(begin = var_7042_begin_0, end = var_7042_end_0, end_mask = var_7042_end_mask_0, x = var_6927_cast_fp16)[name = tensor("op_7042_cast_fp16")]; tensor var_7049_begin_0 = const()[name = tensor("op_7049_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7049_end_0 = const()[name = tensor("op_7049_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7049_end_mask_0 = const()[name = tensor("op_7049_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7049_cast_fp16 = slice_by_index(begin = var_7049_begin_0, end = var_7049_end_0, end_mask = var_7049_end_mask_0, x = var_6927_cast_fp16)[name = tensor("op_7049_cast_fp16")]; tensor var_7056_begin_0 = const()[name = tensor("op_7056_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7056_end_0 = const()[name = tensor("op_7056_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7056_end_mask_0 = const()[name = tensor("op_7056_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7056_cast_fp16 = slice_by_index(begin = var_7056_begin_0, end = var_7056_end_0, end_mask = var_7056_end_mask_0, x = var_6931_cast_fp16)[name = tensor("op_7056_cast_fp16")]; tensor var_7063_begin_0 = const()[name = tensor("op_7063_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7063_end_0 = const()[name = tensor("op_7063_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7063_end_mask_0 = const()[name = tensor("op_7063_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7063_cast_fp16 = slice_by_index(begin = var_7063_begin_0, end = var_7063_end_0, end_mask = var_7063_end_mask_0, x = var_6931_cast_fp16)[name = tensor("op_7063_cast_fp16")]; tensor var_7070_begin_0 = const()[name = tensor("op_7070_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7070_end_0 = const()[name = tensor("op_7070_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7070_end_mask_0 = const()[name = tensor("op_7070_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7070_cast_fp16 = slice_by_index(begin = var_7070_begin_0, end = var_7070_end_0, end_mask = var_7070_end_mask_0, x = var_6931_cast_fp16)[name = tensor("op_7070_cast_fp16")]; tensor var_7077_begin_0 = const()[name = tensor("op_7077_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7077_end_0 = const()[name = tensor("op_7077_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7077_end_mask_0 = const()[name = tensor("op_7077_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7077_cast_fp16 = slice_by_index(begin = var_7077_begin_0, end = var_7077_end_0, end_mask = var_7077_end_mask_0, x = var_6931_cast_fp16)[name = tensor("op_7077_cast_fp16")]; tensor var_7084_begin_0 = const()[name = tensor("op_7084_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7084_end_0 = const()[name = tensor("op_7084_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7084_end_mask_0 = const()[name = tensor("op_7084_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7084_cast_fp16 = slice_by_index(begin = var_7084_begin_0, end = var_7084_end_0, end_mask = var_7084_end_mask_0, x = var_6935_cast_fp16)[name = tensor("op_7084_cast_fp16")]; tensor var_7091_begin_0 = const()[name = tensor("op_7091_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7091_end_0 = const()[name = tensor("op_7091_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7091_end_mask_0 = const()[name = tensor("op_7091_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7091_cast_fp16 = slice_by_index(begin = var_7091_begin_0, end = var_7091_end_0, end_mask = var_7091_end_mask_0, x = var_6935_cast_fp16)[name = tensor("op_7091_cast_fp16")]; tensor var_7098_begin_0 = const()[name = tensor("op_7098_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7098_end_0 = const()[name = tensor("op_7098_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7098_end_mask_0 = const()[name = tensor("op_7098_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7098_cast_fp16 = slice_by_index(begin = var_7098_begin_0, end = var_7098_end_0, end_mask = var_7098_end_mask_0, x = var_6935_cast_fp16)[name = tensor("op_7098_cast_fp16")]; tensor var_7105_begin_0 = const()[name = tensor("op_7105_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7105_end_0 = const()[name = tensor("op_7105_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7105_end_mask_0 = const()[name = tensor("op_7105_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7105_cast_fp16 = slice_by_index(begin = var_7105_begin_0, end = var_7105_end_0, end_mask = var_7105_end_mask_0, x = var_6935_cast_fp16)[name = tensor("op_7105_cast_fp16")]; tensor var_7112_begin_0 = const()[name = tensor("op_7112_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7112_end_0 = const()[name = tensor("op_7112_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7112_end_mask_0 = const()[name = tensor("op_7112_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7112_cast_fp16 = slice_by_index(begin = var_7112_begin_0, end = var_7112_end_0, end_mask = var_7112_end_mask_0, x = var_6939_cast_fp16)[name = tensor("op_7112_cast_fp16")]; tensor var_7119_begin_0 = const()[name = tensor("op_7119_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7119_end_0 = const()[name = tensor("op_7119_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7119_end_mask_0 = const()[name = tensor("op_7119_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7119_cast_fp16 = slice_by_index(begin = var_7119_begin_0, end = var_7119_end_0, end_mask = var_7119_end_mask_0, x = var_6939_cast_fp16)[name = tensor("op_7119_cast_fp16")]; tensor var_7126_begin_0 = const()[name = tensor("op_7126_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7126_end_0 = const()[name = tensor("op_7126_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7126_end_mask_0 = const()[name = tensor("op_7126_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7126_cast_fp16 = slice_by_index(begin = var_7126_begin_0, end = var_7126_end_0, end_mask = var_7126_end_mask_0, x = var_6939_cast_fp16)[name = tensor("op_7126_cast_fp16")]; tensor var_7133_begin_0 = const()[name = tensor("op_7133_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7133_end_0 = const()[name = tensor("op_7133_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7133_end_mask_0 = const()[name = tensor("op_7133_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7133_cast_fp16 = slice_by_index(begin = var_7133_begin_0, end = var_7133_end_0, end_mask = var_7133_end_mask_0, x = var_6939_cast_fp16)[name = tensor("op_7133_cast_fp16")]; tensor var_7140_begin_0 = const()[name = tensor("op_7140_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7140_end_0 = const()[name = tensor("op_7140_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7140_end_mask_0 = const()[name = tensor("op_7140_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7140_cast_fp16 = slice_by_index(begin = var_7140_begin_0, end = var_7140_end_0, end_mask = var_7140_end_mask_0, x = var_6943_cast_fp16)[name = tensor("op_7140_cast_fp16")]; tensor var_7147_begin_0 = const()[name = tensor("op_7147_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7147_end_0 = const()[name = tensor("op_7147_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7147_end_mask_0 = const()[name = tensor("op_7147_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7147_cast_fp16 = slice_by_index(begin = var_7147_begin_0, end = var_7147_end_0, end_mask = var_7147_end_mask_0, x = var_6943_cast_fp16)[name = tensor("op_7147_cast_fp16")]; tensor var_7154_begin_0 = const()[name = tensor("op_7154_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7154_end_0 = const()[name = tensor("op_7154_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7154_end_mask_0 = const()[name = tensor("op_7154_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7154_cast_fp16 = slice_by_index(begin = var_7154_begin_0, end = var_7154_end_0, end_mask = var_7154_end_mask_0, x = var_6943_cast_fp16)[name = tensor("op_7154_cast_fp16")]; tensor var_7161_begin_0 = const()[name = tensor("op_7161_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7161_end_0 = const()[name = tensor("op_7161_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7161_end_mask_0 = const()[name = tensor("op_7161_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7161_cast_fp16 = slice_by_index(begin = var_7161_begin_0, end = var_7161_end_0, end_mask = var_7161_end_mask_0, x = var_6943_cast_fp16)[name = tensor("op_7161_cast_fp16")]; tensor var_7168_begin_0 = const()[name = tensor("op_7168_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7168_end_0 = const()[name = tensor("op_7168_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7168_end_mask_0 = const()[name = tensor("op_7168_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7168_cast_fp16 = slice_by_index(begin = var_7168_begin_0, end = var_7168_end_0, end_mask = var_7168_end_mask_0, x = var_6947_cast_fp16)[name = tensor("op_7168_cast_fp16")]; tensor var_7175_begin_0 = const()[name = tensor("op_7175_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7175_end_0 = const()[name = tensor("op_7175_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7175_end_mask_0 = const()[name = tensor("op_7175_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7175_cast_fp16 = slice_by_index(begin = var_7175_begin_0, end = var_7175_end_0, end_mask = var_7175_end_mask_0, x = var_6947_cast_fp16)[name = tensor("op_7175_cast_fp16")]; tensor var_7182_begin_0 = const()[name = tensor("op_7182_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7182_end_0 = const()[name = tensor("op_7182_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7182_end_mask_0 = const()[name = tensor("op_7182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7182_cast_fp16 = slice_by_index(begin = var_7182_begin_0, end = var_7182_end_0, end_mask = var_7182_end_mask_0, x = var_6947_cast_fp16)[name = tensor("op_7182_cast_fp16")]; tensor var_7189_begin_0 = const()[name = tensor("op_7189_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7189_end_0 = const()[name = tensor("op_7189_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7189_end_mask_0 = const()[name = tensor("op_7189_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7189_cast_fp16 = slice_by_index(begin = var_7189_begin_0, end = var_7189_end_0, end_mask = var_7189_end_mask_0, x = var_6947_cast_fp16)[name = tensor("op_7189_cast_fp16")]; tensor var_7196_begin_0 = const()[name = tensor("op_7196_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7196_end_0 = const()[name = tensor("op_7196_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7196_end_mask_0 = const()[name = tensor("op_7196_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7196_cast_fp16 = slice_by_index(begin = var_7196_begin_0, end = var_7196_end_0, end_mask = var_7196_end_mask_0, x = var_6951_cast_fp16)[name = tensor("op_7196_cast_fp16")]; tensor var_7203_begin_0 = const()[name = tensor("op_7203_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7203_end_0 = const()[name = tensor("op_7203_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7203_end_mask_0 = const()[name = tensor("op_7203_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7203_cast_fp16 = slice_by_index(begin = var_7203_begin_0, end = var_7203_end_0, end_mask = var_7203_end_mask_0, x = var_6951_cast_fp16)[name = tensor("op_7203_cast_fp16")]; tensor var_7210_begin_0 = const()[name = tensor("op_7210_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7210_end_0 = const()[name = tensor("op_7210_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7210_end_mask_0 = const()[name = tensor("op_7210_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7210_cast_fp16 = slice_by_index(begin = var_7210_begin_0, end = var_7210_end_0, end_mask = var_7210_end_mask_0, x = var_6951_cast_fp16)[name = tensor("op_7210_cast_fp16")]; tensor var_7217_begin_0 = const()[name = tensor("op_7217_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7217_end_0 = const()[name = tensor("op_7217_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7217_end_mask_0 = const()[name = tensor("op_7217_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7217_cast_fp16 = slice_by_index(begin = var_7217_begin_0, end = var_7217_end_0, end_mask = var_7217_end_mask_0, x = var_6951_cast_fp16)[name = tensor("op_7217_cast_fp16")]; tensor var_7224_begin_0 = const()[name = tensor("op_7224_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7224_end_0 = const()[name = tensor("op_7224_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7224_end_mask_0 = const()[name = tensor("op_7224_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7224_cast_fp16 = slice_by_index(begin = var_7224_begin_0, end = var_7224_end_0, end_mask = var_7224_end_mask_0, x = var_6955_cast_fp16)[name = tensor("op_7224_cast_fp16")]; tensor var_7231_begin_0 = const()[name = tensor("op_7231_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7231_end_0 = const()[name = tensor("op_7231_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7231_end_mask_0 = const()[name = tensor("op_7231_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7231_cast_fp16 = slice_by_index(begin = var_7231_begin_0, end = var_7231_end_0, end_mask = var_7231_end_mask_0, x = var_6955_cast_fp16)[name = tensor("op_7231_cast_fp16")]; tensor var_7238_begin_0 = const()[name = tensor("op_7238_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7238_end_0 = const()[name = tensor("op_7238_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7238_end_mask_0 = const()[name = tensor("op_7238_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7238_cast_fp16 = slice_by_index(begin = var_7238_begin_0, end = var_7238_end_0, end_mask = var_7238_end_mask_0, x = var_6955_cast_fp16)[name = tensor("op_7238_cast_fp16")]; tensor var_7245_begin_0 = const()[name = tensor("op_7245_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7245_end_0 = const()[name = tensor("op_7245_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7245_end_mask_0 = const()[name = tensor("op_7245_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7245_cast_fp16 = slice_by_index(begin = var_7245_begin_0, end = var_7245_end_0, end_mask = var_7245_end_mask_0, x = var_6955_cast_fp16)[name = tensor("op_7245_cast_fp16")]; tensor var_7252_begin_0 = const()[name = tensor("op_7252_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7252_end_0 = const()[name = tensor("op_7252_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7252_end_mask_0 = const()[name = tensor("op_7252_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7252_cast_fp16 = slice_by_index(begin = var_7252_begin_0, end = var_7252_end_0, end_mask = var_7252_end_mask_0, x = var_6959_cast_fp16)[name = tensor("op_7252_cast_fp16")]; tensor var_7259_begin_0 = const()[name = tensor("op_7259_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7259_end_0 = const()[name = tensor("op_7259_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7259_end_mask_0 = const()[name = tensor("op_7259_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7259_cast_fp16 = slice_by_index(begin = var_7259_begin_0, end = var_7259_end_0, end_mask = var_7259_end_mask_0, x = var_6959_cast_fp16)[name = tensor("op_7259_cast_fp16")]; tensor var_7266_begin_0 = const()[name = tensor("op_7266_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7266_end_0 = const()[name = tensor("op_7266_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7266_end_mask_0 = const()[name = tensor("op_7266_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7266_cast_fp16 = slice_by_index(begin = var_7266_begin_0, end = var_7266_end_0, end_mask = var_7266_end_mask_0, x = var_6959_cast_fp16)[name = tensor("op_7266_cast_fp16")]; tensor var_7273_begin_0 = const()[name = tensor("op_7273_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7273_end_0 = const()[name = tensor("op_7273_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7273_end_mask_0 = const()[name = tensor("op_7273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7273_cast_fp16 = slice_by_index(begin = var_7273_begin_0, end = var_7273_end_0, end_mask = var_7273_end_mask_0, x = var_6959_cast_fp16)[name = tensor("op_7273_cast_fp16")]; tensor var_7280_begin_0 = const()[name = tensor("op_7280_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7280_end_0 = const()[name = tensor("op_7280_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7280_end_mask_0 = const()[name = tensor("op_7280_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7280_cast_fp16 = slice_by_index(begin = var_7280_begin_0, end = var_7280_end_0, end_mask = var_7280_end_mask_0, x = var_6963_cast_fp16)[name = tensor("op_7280_cast_fp16")]; tensor var_7287_begin_0 = const()[name = tensor("op_7287_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7287_end_0 = const()[name = tensor("op_7287_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7287_end_mask_0 = const()[name = tensor("op_7287_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7287_cast_fp16 = slice_by_index(begin = var_7287_begin_0, end = var_7287_end_0, end_mask = var_7287_end_mask_0, x = var_6963_cast_fp16)[name = tensor("op_7287_cast_fp16")]; tensor var_7294_begin_0 = const()[name = tensor("op_7294_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7294_end_0 = const()[name = tensor("op_7294_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7294_end_mask_0 = const()[name = tensor("op_7294_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7294_cast_fp16 = slice_by_index(begin = var_7294_begin_0, end = var_7294_end_0, end_mask = var_7294_end_mask_0, x = var_6963_cast_fp16)[name = tensor("op_7294_cast_fp16")]; tensor var_7301_begin_0 = const()[name = tensor("op_7301_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7301_end_0 = const()[name = tensor("op_7301_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7301_end_mask_0 = const()[name = tensor("op_7301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7301_cast_fp16 = slice_by_index(begin = var_7301_begin_0, end = var_7301_end_0, end_mask = var_7301_end_mask_0, x = var_6963_cast_fp16)[name = tensor("op_7301_cast_fp16")]; tensor k_15_perm_0 = const()[name = tensor("k_15_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_7306_begin_0 = const()[name = tensor("op_7306_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7306_end_0 = const()[name = tensor("op_7306_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_7306_end_mask_0 = const()[name = tensor("op_7306_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_4 = transpose(perm = k_15_perm_0, x = key_15_cast_fp16)[name = tensor("transpose_4")]; tensor var_7306_cast_fp16 = slice_by_index(begin = var_7306_begin_0, end = var_7306_end_0, end_mask = var_7306_end_mask_0, x = transpose_4)[name = tensor("op_7306_cast_fp16")]; tensor var_7310_begin_0 = const()[name = tensor("op_7310_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_7310_end_0 = const()[name = tensor("op_7310_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_7310_end_mask_0 = const()[name = tensor("op_7310_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7310_cast_fp16 = slice_by_index(begin = var_7310_begin_0, end = var_7310_end_0, end_mask = var_7310_end_mask_0, x = transpose_4)[name = tensor("op_7310_cast_fp16")]; tensor var_7314_begin_0 = const()[name = tensor("op_7314_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_7314_end_0 = const()[name = tensor("op_7314_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_7314_end_mask_0 = const()[name = tensor("op_7314_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7314_cast_fp16 = slice_by_index(begin = var_7314_begin_0, end = var_7314_end_0, end_mask = var_7314_end_mask_0, x = transpose_4)[name = tensor("op_7314_cast_fp16")]; tensor var_7318_begin_0 = const()[name = tensor("op_7318_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_7318_end_0 = const()[name = tensor("op_7318_end_0"), val = tensor([1, 1500, 1, 256])]; 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 = transpose_4)[name = tensor("op_7318_cast_fp16")]; tensor var_7322_begin_0 = const()[name = tensor("op_7322_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_7322_end_0 = const()[name = tensor("op_7322_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_7322_end_mask_0 = const()[name = tensor("op_7322_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7322_cast_fp16 = slice_by_index(begin = var_7322_begin_0, end = var_7322_end_0, end_mask = var_7322_end_mask_0, x = transpose_4)[name = tensor("op_7322_cast_fp16")]; tensor var_7326_begin_0 = const()[name = tensor("op_7326_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_7326_end_0 = const()[name = tensor("op_7326_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_7326_end_mask_0 = const()[name = tensor("op_7326_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7326_cast_fp16 = slice_by_index(begin = var_7326_begin_0, end = var_7326_end_0, end_mask = var_7326_end_mask_0, x = transpose_4)[name = tensor("op_7326_cast_fp16")]; tensor var_7330_begin_0 = const()[name = tensor("op_7330_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_7330_end_0 = const()[name = tensor("op_7330_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_7330_end_mask_0 = const()[name = tensor("op_7330_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7330_cast_fp16 = slice_by_index(begin = var_7330_begin_0, end = var_7330_end_0, end_mask = var_7330_end_mask_0, x = transpose_4)[name = tensor("op_7330_cast_fp16")]; tensor var_7334_begin_0 = const()[name = tensor("op_7334_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_7334_end_0 = const()[name = tensor("op_7334_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_7334_end_mask_0 = const()[name = tensor("op_7334_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7334_cast_fp16 = slice_by_index(begin = var_7334_begin_0, end = var_7334_end_0, end_mask = var_7334_end_mask_0, x = transpose_4)[name = tensor("op_7334_cast_fp16")]; tensor var_7338_begin_0 = const()[name = tensor("op_7338_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_7338_end_0 = const()[name = tensor("op_7338_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_7338_end_mask_0 = const()[name = tensor("op_7338_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7338_cast_fp16 = slice_by_index(begin = var_7338_begin_0, end = var_7338_end_0, end_mask = var_7338_end_mask_0, x = transpose_4)[name = tensor("op_7338_cast_fp16")]; tensor var_7342_begin_0 = const()[name = tensor("op_7342_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_7342_end_0 = const()[name = tensor("op_7342_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_7342_end_mask_0 = const()[name = tensor("op_7342_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7342_cast_fp16 = slice_by_index(begin = var_7342_begin_0, end = var_7342_end_0, end_mask = var_7342_end_mask_0, x = transpose_4)[name = tensor("op_7342_cast_fp16")]; tensor var_7346_begin_0 = const()[name = tensor("op_7346_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_7346_end_0 = const()[name = tensor("op_7346_end_0"), val = tensor([1, 1500, 1, 704])]; 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 = transpose_4)[name = tensor("op_7346_cast_fp16")]; tensor var_7350_begin_0 = const()[name = tensor("op_7350_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_7350_end_0 = const()[name = tensor("op_7350_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_7350_end_mask_0 = const()[name = tensor("op_7350_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7350_cast_fp16 = slice_by_index(begin = var_7350_begin_0, end = var_7350_end_0, end_mask = var_7350_end_mask_0, x = transpose_4)[name = tensor("op_7350_cast_fp16")]; tensor var_7352_begin_0 = const()[name = tensor("op_7352_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7352_end_0 = const()[name = tensor("op_7352_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7352_end_mask_0 = const()[name = tensor("op_7352_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7352_cast_fp16 = slice_by_index(begin = var_7352_begin_0, end = var_7352_end_0, end_mask = var_7352_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7352_cast_fp16")]; tensor var_7356_begin_0 = const()[name = tensor("op_7356_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7356_end_0 = const()[name = tensor("op_7356_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_7356_end_mask_0 = const()[name = tensor("op_7356_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7356_cast_fp16 = slice_by_index(begin = var_7356_begin_0, end = var_7356_end_0, end_mask = var_7356_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7356_cast_fp16")]; tensor var_7360_begin_0 = const()[name = tensor("op_7360_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7360_end_0 = const()[name = tensor("op_7360_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_7360_end_mask_0 = const()[name = tensor("op_7360_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_15_cast_fp16)[name = tensor("op_7360_cast_fp16")]; tensor var_7364_begin_0 = const()[name = tensor("op_7364_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7364_end_0 = const()[name = tensor("op_7364_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_7364_end_mask_0 = const()[name = tensor("op_7364_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7364_cast_fp16 = slice_by_index(begin = var_7364_begin_0, end = var_7364_end_0, end_mask = var_7364_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7364_cast_fp16")]; tensor var_7368_begin_0 = const()[name = tensor("op_7368_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7368_end_0 = const()[name = tensor("op_7368_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_7368_end_mask_0 = const()[name = tensor("op_7368_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7368_cast_fp16 = slice_by_index(begin = var_7368_begin_0, end = var_7368_end_0, end_mask = var_7368_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7368_cast_fp16")]; tensor var_7372_begin_0 = const()[name = tensor("op_7372_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7372_end_0 = const()[name = tensor("op_7372_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_7372_end_mask_0 = const()[name = tensor("op_7372_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7372_cast_fp16 = slice_by_index(begin = var_7372_begin_0, end = var_7372_end_0, end_mask = var_7372_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7372_cast_fp16")]; tensor var_7376_begin_0 = const()[name = tensor("op_7376_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7376_end_0 = const()[name = tensor("op_7376_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_7376_end_mask_0 = const()[name = tensor("op_7376_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7376_cast_fp16 = slice_by_index(begin = var_7376_begin_0, end = var_7376_end_0, end_mask = var_7376_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7376_cast_fp16")]; tensor var_7380_begin_0 = const()[name = tensor("op_7380_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7380_end_0 = const()[name = tensor("op_7380_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_7380_end_mask_0 = const()[name = tensor("op_7380_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7380_cast_fp16 = slice_by_index(begin = var_7380_begin_0, end = var_7380_end_0, end_mask = var_7380_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7380_cast_fp16")]; tensor var_7384_begin_0 = const()[name = tensor("op_7384_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7384_end_0 = const()[name = tensor("op_7384_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_7384_end_mask_0 = const()[name = tensor("op_7384_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7384_cast_fp16 = slice_by_index(begin = var_7384_begin_0, end = var_7384_end_0, end_mask = var_7384_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7384_cast_fp16")]; tensor var_7388_begin_0 = const()[name = tensor("op_7388_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7388_end_0 = const()[name = tensor("op_7388_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_7388_end_mask_0 = const()[name = tensor("op_7388_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_15_cast_fp16)[name = tensor("op_7388_cast_fp16")]; tensor var_7392_begin_0 = const()[name = tensor("op_7392_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7392_end_0 = const()[name = tensor("op_7392_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_7392_end_mask_0 = const()[name = tensor("op_7392_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7392_cast_fp16 = slice_by_index(begin = var_7392_begin_0, end = var_7392_end_0, end_mask = var_7392_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7392_cast_fp16")]; tensor var_7396_begin_0 = const()[name = tensor("op_7396_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7396_end_0 = const()[name = tensor("op_7396_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_7396_end_mask_0 = const()[name = tensor("op_7396_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7396_cast_fp16 = slice_by_index(begin = var_7396_begin_0, end = var_7396_end_0, end_mask = var_7396_end_mask_0, x = value_15_cast_fp16)[name = tensor("op_7396_cast_fp16")]; tensor var_7400_equation_0 = const()[name = tensor("op_7400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7400_cast_fp16 = einsum(equation = var_7400_equation_0, values = (var_7306_cast_fp16, var_6972_cast_fp16))[name = tensor("op_7400_cast_fp16")]; tensor var_7401_to_fp16 = const()[name = tensor("op_7401_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_673_cast_fp16 = mul(x = var_7400_cast_fp16, y = var_7401_to_fp16)[name = tensor("aw_chunk_673_cast_fp16")]; tensor var_7404_equation_0 = const()[name = tensor("op_7404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7404_cast_fp16 = einsum(equation = var_7404_equation_0, values = (var_7306_cast_fp16, var_6979_cast_fp16))[name = tensor("op_7404_cast_fp16")]; tensor var_7405_to_fp16 = const()[name = tensor("op_7405_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_675_cast_fp16 = mul(x = var_7404_cast_fp16, y = var_7405_to_fp16)[name = tensor("aw_chunk_675_cast_fp16")]; tensor var_7408_equation_0 = const()[name = tensor("op_7408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7408_cast_fp16 = einsum(equation = var_7408_equation_0, values = (var_7306_cast_fp16, var_6986_cast_fp16))[name = tensor("op_7408_cast_fp16")]; tensor var_7409_to_fp16 = const()[name = tensor("op_7409_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_677_cast_fp16 = mul(x = var_7408_cast_fp16, y = var_7409_to_fp16)[name = tensor("aw_chunk_677_cast_fp16")]; tensor var_7412_equation_0 = const()[name = tensor("op_7412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7412_cast_fp16 = einsum(equation = var_7412_equation_0, values = (var_7306_cast_fp16, var_6993_cast_fp16))[name = tensor("op_7412_cast_fp16")]; tensor var_7413_to_fp16 = const()[name = tensor("op_7413_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_679_cast_fp16 = mul(x = var_7412_cast_fp16, y = var_7413_to_fp16)[name = tensor("aw_chunk_679_cast_fp16")]; tensor var_7416_equation_0 = const()[name = tensor("op_7416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7416_cast_fp16 = einsum(equation = var_7416_equation_0, values = (var_7310_cast_fp16, var_7000_cast_fp16))[name = tensor("op_7416_cast_fp16")]; tensor var_7417_to_fp16 = const()[name = tensor("op_7417_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_681_cast_fp16 = mul(x = var_7416_cast_fp16, y = var_7417_to_fp16)[name = tensor("aw_chunk_681_cast_fp16")]; tensor var_7420_equation_0 = const()[name = tensor("op_7420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7420_cast_fp16 = einsum(equation = var_7420_equation_0, values = (var_7310_cast_fp16, var_7007_cast_fp16))[name = tensor("op_7420_cast_fp16")]; tensor var_7421_to_fp16 = const()[name = tensor("op_7421_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_683_cast_fp16 = mul(x = var_7420_cast_fp16, y = var_7421_to_fp16)[name = tensor("aw_chunk_683_cast_fp16")]; tensor var_7424_equation_0 = const()[name = tensor("op_7424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7424_cast_fp16 = einsum(equation = var_7424_equation_0, values = (var_7310_cast_fp16, var_7014_cast_fp16))[name = tensor("op_7424_cast_fp16")]; tensor var_7425_to_fp16 = const()[name = tensor("op_7425_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_685_cast_fp16 = mul(x = var_7424_cast_fp16, y = var_7425_to_fp16)[name = tensor("aw_chunk_685_cast_fp16")]; tensor var_7428_equation_0 = const()[name = tensor("op_7428_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7428_cast_fp16 = einsum(equation = var_7428_equation_0, values = (var_7310_cast_fp16, var_7021_cast_fp16))[name = tensor("op_7428_cast_fp16")]; tensor var_7429_to_fp16 = const()[name = tensor("op_7429_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_687_cast_fp16 = mul(x = var_7428_cast_fp16, y = var_7429_to_fp16)[name = tensor("aw_chunk_687_cast_fp16")]; tensor var_7432_equation_0 = const()[name = tensor("op_7432_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7432_cast_fp16 = einsum(equation = var_7432_equation_0, values = (var_7314_cast_fp16, var_7028_cast_fp16))[name = tensor("op_7432_cast_fp16")]; tensor var_7433_to_fp16 = const()[name = tensor("op_7433_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_689_cast_fp16 = mul(x = var_7432_cast_fp16, y = var_7433_to_fp16)[name = tensor("aw_chunk_689_cast_fp16")]; tensor var_7436_equation_0 = const()[name = tensor("op_7436_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7436_cast_fp16 = einsum(equation = var_7436_equation_0, values = (var_7314_cast_fp16, var_7035_cast_fp16))[name = tensor("op_7436_cast_fp16")]; tensor var_7437_to_fp16 = const()[name = tensor("op_7437_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_691_cast_fp16 = mul(x = var_7436_cast_fp16, y = var_7437_to_fp16)[name = tensor("aw_chunk_691_cast_fp16")]; tensor var_7440_equation_0 = const()[name = tensor("op_7440_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7440_cast_fp16 = einsum(equation = var_7440_equation_0, values = (var_7314_cast_fp16, var_7042_cast_fp16))[name = tensor("op_7440_cast_fp16")]; tensor var_7441_to_fp16 = const()[name = tensor("op_7441_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_693_cast_fp16 = mul(x = var_7440_cast_fp16, y = var_7441_to_fp16)[name = tensor("aw_chunk_693_cast_fp16")]; tensor var_7444_equation_0 = const()[name = tensor("op_7444_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7444_cast_fp16 = einsum(equation = var_7444_equation_0, values = (var_7314_cast_fp16, var_7049_cast_fp16))[name = tensor("op_7444_cast_fp16")]; tensor var_7445_to_fp16 = const()[name = tensor("op_7445_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_695_cast_fp16 = mul(x = var_7444_cast_fp16, y = var_7445_to_fp16)[name = tensor("aw_chunk_695_cast_fp16")]; tensor var_7448_equation_0 = const()[name = tensor("op_7448_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7448_cast_fp16 = einsum(equation = var_7448_equation_0, values = (var_7318_cast_fp16, var_7056_cast_fp16))[name = tensor("op_7448_cast_fp16")]; tensor var_7449_to_fp16 = const()[name = tensor("op_7449_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_697_cast_fp16 = mul(x = var_7448_cast_fp16, y = var_7449_to_fp16)[name = tensor("aw_chunk_697_cast_fp16")]; tensor var_7452_equation_0 = const()[name = tensor("op_7452_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7452_cast_fp16 = einsum(equation = var_7452_equation_0, values = (var_7318_cast_fp16, var_7063_cast_fp16))[name = tensor("op_7452_cast_fp16")]; tensor var_7453_to_fp16 = const()[name = tensor("op_7453_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_699_cast_fp16 = mul(x = var_7452_cast_fp16, y = var_7453_to_fp16)[name = tensor("aw_chunk_699_cast_fp16")]; tensor var_7456_equation_0 = const()[name = tensor("op_7456_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7456_cast_fp16 = einsum(equation = var_7456_equation_0, values = (var_7318_cast_fp16, var_7070_cast_fp16))[name = tensor("op_7456_cast_fp16")]; tensor var_7457_to_fp16 = const()[name = tensor("op_7457_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_701_cast_fp16 = mul(x = var_7456_cast_fp16, y = var_7457_to_fp16)[name = tensor("aw_chunk_701_cast_fp16")]; tensor var_7460_equation_0 = const()[name = tensor("op_7460_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7460_cast_fp16 = einsum(equation = var_7460_equation_0, values = (var_7318_cast_fp16, var_7077_cast_fp16))[name = tensor("op_7460_cast_fp16")]; tensor var_7461_to_fp16 = const()[name = tensor("op_7461_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_703_cast_fp16 = mul(x = var_7460_cast_fp16, y = var_7461_to_fp16)[name = tensor("aw_chunk_703_cast_fp16")]; tensor var_7464_equation_0 = const()[name = tensor("op_7464_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7464_cast_fp16 = einsum(equation = var_7464_equation_0, values = (var_7322_cast_fp16, var_7084_cast_fp16))[name = tensor("op_7464_cast_fp16")]; tensor var_7465_to_fp16 = const()[name = tensor("op_7465_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_705_cast_fp16 = mul(x = var_7464_cast_fp16, y = var_7465_to_fp16)[name = tensor("aw_chunk_705_cast_fp16")]; tensor var_7468_equation_0 = const()[name = tensor("op_7468_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7468_cast_fp16 = einsum(equation = var_7468_equation_0, values = (var_7322_cast_fp16, var_7091_cast_fp16))[name = tensor("op_7468_cast_fp16")]; tensor var_7469_to_fp16 = const()[name = tensor("op_7469_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_707_cast_fp16 = mul(x = var_7468_cast_fp16, y = var_7469_to_fp16)[name = tensor("aw_chunk_707_cast_fp16")]; tensor var_7472_equation_0 = const()[name = tensor("op_7472_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7472_cast_fp16 = einsum(equation = var_7472_equation_0, values = (var_7322_cast_fp16, var_7098_cast_fp16))[name = tensor("op_7472_cast_fp16")]; tensor var_7473_to_fp16 = const()[name = tensor("op_7473_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_709_cast_fp16 = mul(x = var_7472_cast_fp16, y = var_7473_to_fp16)[name = tensor("aw_chunk_709_cast_fp16")]; tensor var_7476_equation_0 = const()[name = tensor("op_7476_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7476_cast_fp16 = einsum(equation = var_7476_equation_0, values = (var_7322_cast_fp16, var_7105_cast_fp16))[name = tensor("op_7476_cast_fp16")]; tensor var_7477_to_fp16 = const()[name = tensor("op_7477_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_711_cast_fp16 = mul(x = var_7476_cast_fp16, y = var_7477_to_fp16)[name = tensor("aw_chunk_711_cast_fp16")]; tensor var_7480_equation_0 = const()[name = tensor("op_7480_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7480_cast_fp16 = einsum(equation = var_7480_equation_0, values = (var_7326_cast_fp16, var_7112_cast_fp16))[name = tensor("op_7480_cast_fp16")]; tensor var_7481_to_fp16 = const()[name = tensor("op_7481_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_713_cast_fp16 = mul(x = var_7480_cast_fp16, y = var_7481_to_fp16)[name = tensor("aw_chunk_713_cast_fp16")]; tensor var_7484_equation_0 = const()[name = tensor("op_7484_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7484_cast_fp16 = einsum(equation = var_7484_equation_0, values = (var_7326_cast_fp16, var_7119_cast_fp16))[name = tensor("op_7484_cast_fp16")]; tensor var_7485_to_fp16 = const()[name = tensor("op_7485_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_715_cast_fp16 = mul(x = var_7484_cast_fp16, y = var_7485_to_fp16)[name = tensor("aw_chunk_715_cast_fp16")]; tensor var_7488_equation_0 = const()[name = tensor("op_7488_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7488_cast_fp16 = einsum(equation = var_7488_equation_0, values = (var_7326_cast_fp16, var_7126_cast_fp16))[name = tensor("op_7488_cast_fp16")]; tensor var_7489_to_fp16 = const()[name = tensor("op_7489_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_717_cast_fp16 = mul(x = var_7488_cast_fp16, y = var_7489_to_fp16)[name = tensor("aw_chunk_717_cast_fp16")]; tensor var_7492_equation_0 = const()[name = tensor("op_7492_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7492_cast_fp16 = einsum(equation = var_7492_equation_0, values = (var_7326_cast_fp16, var_7133_cast_fp16))[name = tensor("op_7492_cast_fp16")]; tensor var_7493_to_fp16 = const()[name = tensor("op_7493_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_719_cast_fp16 = mul(x = var_7492_cast_fp16, y = var_7493_to_fp16)[name = tensor("aw_chunk_719_cast_fp16")]; tensor var_7496_equation_0 = const()[name = tensor("op_7496_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7496_cast_fp16 = einsum(equation = var_7496_equation_0, values = (var_7330_cast_fp16, var_7140_cast_fp16))[name = tensor("op_7496_cast_fp16")]; tensor var_7497_to_fp16 = const()[name = tensor("op_7497_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_721_cast_fp16 = mul(x = var_7496_cast_fp16, y = var_7497_to_fp16)[name = tensor("aw_chunk_721_cast_fp16")]; tensor var_7500_equation_0 = const()[name = tensor("op_7500_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7500_cast_fp16 = einsum(equation = var_7500_equation_0, values = (var_7330_cast_fp16, var_7147_cast_fp16))[name = tensor("op_7500_cast_fp16")]; tensor var_7501_to_fp16 = const()[name = tensor("op_7501_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_723_cast_fp16 = mul(x = var_7500_cast_fp16, y = var_7501_to_fp16)[name = tensor("aw_chunk_723_cast_fp16")]; tensor var_7504_equation_0 = const()[name = tensor("op_7504_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7504_cast_fp16 = einsum(equation = var_7504_equation_0, values = (var_7330_cast_fp16, var_7154_cast_fp16))[name = tensor("op_7504_cast_fp16")]; tensor var_7505_to_fp16 = const()[name = tensor("op_7505_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_725_cast_fp16 = mul(x = var_7504_cast_fp16, y = var_7505_to_fp16)[name = tensor("aw_chunk_725_cast_fp16")]; tensor var_7508_equation_0 = const()[name = tensor("op_7508_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7508_cast_fp16 = einsum(equation = var_7508_equation_0, values = (var_7330_cast_fp16, var_7161_cast_fp16))[name = tensor("op_7508_cast_fp16")]; tensor var_7509_to_fp16 = const()[name = tensor("op_7509_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_727_cast_fp16 = mul(x = var_7508_cast_fp16, y = var_7509_to_fp16)[name = tensor("aw_chunk_727_cast_fp16")]; tensor var_7512_equation_0 = const()[name = tensor("op_7512_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7512_cast_fp16 = einsum(equation = var_7512_equation_0, values = (var_7334_cast_fp16, var_7168_cast_fp16))[name = tensor("op_7512_cast_fp16")]; tensor var_7513_to_fp16 = const()[name = tensor("op_7513_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_729_cast_fp16 = mul(x = var_7512_cast_fp16, y = var_7513_to_fp16)[name = tensor("aw_chunk_729_cast_fp16")]; tensor var_7516_equation_0 = const()[name = tensor("op_7516_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7516_cast_fp16 = einsum(equation = var_7516_equation_0, values = (var_7334_cast_fp16, var_7175_cast_fp16))[name = tensor("op_7516_cast_fp16")]; tensor var_7517_to_fp16 = const()[name = tensor("op_7517_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_731_cast_fp16 = mul(x = var_7516_cast_fp16, y = var_7517_to_fp16)[name = tensor("aw_chunk_731_cast_fp16")]; tensor var_7520_equation_0 = const()[name = tensor("op_7520_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7520_cast_fp16 = einsum(equation = var_7520_equation_0, values = (var_7334_cast_fp16, var_7182_cast_fp16))[name = tensor("op_7520_cast_fp16")]; tensor var_7521_to_fp16 = const()[name = tensor("op_7521_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_733_cast_fp16 = mul(x = var_7520_cast_fp16, y = var_7521_to_fp16)[name = tensor("aw_chunk_733_cast_fp16")]; tensor var_7524_equation_0 = const()[name = tensor("op_7524_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7524_cast_fp16 = einsum(equation = var_7524_equation_0, values = (var_7334_cast_fp16, var_7189_cast_fp16))[name = tensor("op_7524_cast_fp16")]; tensor var_7525_to_fp16 = const()[name = tensor("op_7525_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_735_cast_fp16 = mul(x = var_7524_cast_fp16, y = var_7525_to_fp16)[name = tensor("aw_chunk_735_cast_fp16")]; tensor var_7528_equation_0 = const()[name = tensor("op_7528_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7528_cast_fp16 = einsum(equation = var_7528_equation_0, values = (var_7338_cast_fp16, var_7196_cast_fp16))[name = tensor("op_7528_cast_fp16")]; tensor var_7529_to_fp16 = const()[name = tensor("op_7529_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_737_cast_fp16 = mul(x = var_7528_cast_fp16, y = var_7529_to_fp16)[name = tensor("aw_chunk_737_cast_fp16")]; tensor var_7532_equation_0 = const()[name = tensor("op_7532_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7532_cast_fp16 = einsum(equation = var_7532_equation_0, values = (var_7338_cast_fp16, var_7203_cast_fp16))[name = tensor("op_7532_cast_fp16")]; tensor var_7533_to_fp16 = const()[name = tensor("op_7533_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_739_cast_fp16 = mul(x = var_7532_cast_fp16, y = var_7533_to_fp16)[name = tensor("aw_chunk_739_cast_fp16")]; tensor var_7536_equation_0 = const()[name = tensor("op_7536_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7536_cast_fp16 = einsum(equation = var_7536_equation_0, values = (var_7338_cast_fp16, var_7210_cast_fp16))[name = tensor("op_7536_cast_fp16")]; tensor var_7537_to_fp16 = const()[name = tensor("op_7537_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_741_cast_fp16 = mul(x = var_7536_cast_fp16, y = var_7537_to_fp16)[name = tensor("aw_chunk_741_cast_fp16")]; tensor var_7540_equation_0 = const()[name = tensor("op_7540_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7540_cast_fp16 = einsum(equation = var_7540_equation_0, values = (var_7338_cast_fp16, var_7217_cast_fp16))[name = tensor("op_7540_cast_fp16")]; tensor var_7541_to_fp16 = const()[name = tensor("op_7541_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_743_cast_fp16 = mul(x = var_7540_cast_fp16, y = var_7541_to_fp16)[name = tensor("aw_chunk_743_cast_fp16")]; tensor var_7544_equation_0 = const()[name = tensor("op_7544_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7544_cast_fp16 = einsum(equation = var_7544_equation_0, values = (var_7342_cast_fp16, var_7224_cast_fp16))[name = tensor("op_7544_cast_fp16")]; tensor var_7545_to_fp16 = const()[name = tensor("op_7545_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_745_cast_fp16 = mul(x = var_7544_cast_fp16, y = var_7545_to_fp16)[name = tensor("aw_chunk_745_cast_fp16")]; tensor var_7548_equation_0 = const()[name = tensor("op_7548_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7548_cast_fp16 = einsum(equation = var_7548_equation_0, values = (var_7342_cast_fp16, var_7231_cast_fp16))[name = tensor("op_7548_cast_fp16")]; tensor var_7549_to_fp16 = const()[name = tensor("op_7549_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_747_cast_fp16 = mul(x = var_7548_cast_fp16, y = var_7549_to_fp16)[name = tensor("aw_chunk_747_cast_fp16")]; tensor var_7552_equation_0 = const()[name = tensor("op_7552_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7552_cast_fp16 = einsum(equation = var_7552_equation_0, values = (var_7342_cast_fp16, var_7238_cast_fp16))[name = tensor("op_7552_cast_fp16")]; tensor var_7553_to_fp16 = const()[name = tensor("op_7553_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_749_cast_fp16 = mul(x = var_7552_cast_fp16, y = var_7553_to_fp16)[name = tensor("aw_chunk_749_cast_fp16")]; tensor var_7556_equation_0 = const()[name = tensor("op_7556_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7556_cast_fp16 = einsum(equation = var_7556_equation_0, values = (var_7342_cast_fp16, var_7245_cast_fp16))[name = tensor("op_7556_cast_fp16")]; tensor var_7557_to_fp16 = const()[name = tensor("op_7557_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_751_cast_fp16 = mul(x = var_7556_cast_fp16, y = var_7557_to_fp16)[name = tensor("aw_chunk_751_cast_fp16")]; tensor var_7560_equation_0 = const()[name = tensor("op_7560_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7560_cast_fp16 = einsum(equation = var_7560_equation_0, values = (var_7346_cast_fp16, var_7252_cast_fp16))[name = tensor("op_7560_cast_fp16")]; tensor var_7561_to_fp16 = const()[name = tensor("op_7561_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_753_cast_fp16 = mul(x = var_7560_cast_fp16, y = var_7561_to_fp16)[name = tensor("aw_chunk_753_cast_fp16")]; tensor var_7564_equation_0 = const()[name = tensor("op_7564_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7564_cast_fp16 = einsum(equation = var_7564_equation_0, values = (var_7346_cast_fp16, var_7259_cast_fp16))[name = tensor("op_7564_cast_fp16")]; tensor var_7565_to_fp16 = const()[name = tensor("op_7565_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_755_cast_fp16 = mul(x = var_7564_cast_fp16, y = var_7565_to_fp16)[name = tensor("aw_chunk_755_cast_fp16")]; tensor var_7568_equation_0 = const()[name = tensor("op_7568_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7568_cast_fp16 = einsum(equation = var_7568_equation_0, values = (var_7346_cast_fp16, var_7266_cast_fp16))[name = tensor("op_7568_cast_fp16")]; tensor var_7569_to_fp16 = const()[name = tensor("op_7569_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_757_cast_fp16 = mul(x = var_7568_cast_fp16, y = var_7569_to_fp16)[name = tensor("aw_chunk_757_cast_fp16")]; tensor var_7572_equation_0 = const()[name = tensor("op_7572_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7572_cast_fp16 = einsum(equation = var_7572_equation_0, values = (var_7346_cast_fp16, var_7273_cast_fp16))[name = tensor("op_7572_cast_fp16")]; tensor var_7573_to_fp16 = const()[name = tensor("op_7573_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_759_cast_fp16 = mul(x = var_7572_cast_fp16, y = var_7573_to_fp16)[name = tensor("aw_chunk_759_cast_fp16")]; tensor var_7576_equation_0 = const()[name = tensor("op_7576_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7576_cast_fp16 = einsum(equation = var_7576_equation_0, values = (var_7350_cast_fp16, var_7280_cast_fp16))[name = tensor("op_7576_cast_fp16")]; tensor var_7577_to_fp16 = const()[name = tensor("op_7577_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_761_cast_fp16 = mul(x = var_7576_cast_fp16, y = var_7577_to_fp16)[name = tensor("aw_chunk_761_cast_fp16")]; tensor var_7580_equation_0 = const()[name = tensor("op_7580_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7580_cast_fp16 = einsum(equation = var_7580_equation_0, values = (var_7350_cast_fp16, var_7287_cast_fp16))[name = tensor("op_7580_cast_fp16")]; tensor var_7581_to_fp16 = const()[name = tensor("op_7581_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_763_cast_fp16 = mul(x = var_7580_cast_fp16, y = var_7581_to_fp16)[name = tensor("aw_chunk_763_cast_fp16")]; tensor var_7584_equation_0 = const()[name = tensor("op_7584_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7584_cast_fp16 = einsum(equation = var_7584_equation_0, values = (var_7350_cast_fp16, var_7294_cast_fp16))[name = tensor("op_7584_cast_fp16")]; tensor var_7585_to_fp16 = const()[name = tensor("op_7585_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_765_cast_fp16 = mul(x = var_7584_cast_fp16, y = var_7585_to_fp16)[name = tensor("aw_chunk_765_cast_fp16")]; tensor var_7588_equation_0 = const()[name = tensor("op_7588_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_7588_cast_fp16 = einsum(equation = var_7588_equation_0, values = (var_7350_cast_fp16, var_7301_cast_fp16))[name = tensor("op_7588_cast_fp16")]; tensor var_7589_to_fp16 = const()[name = tensor("op_7589_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_767_cast_fp16 = mul(x = var_7588_cast_fp16, y = var_7589_to_fp16)[name = tensor("aw_chunk_767_cast_fp16")]; tensor var_7591_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_673_cast_fp16)[name = tensor("op_7591_cast_fp16")]; tensor var_7592_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_675_cast_fp16)[name = tensor("op_7592_cast_fp16")]; tensor var_7593_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_677_cast_fp16)[name = tensor("op_7593_cast_fp16")]; tensor var_7594_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_679_cast_fp16)[name = tensor("op_7594_cast_fp16")]; tensor var_7595_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_681_cast_fp16)[name = tensor("op_7595_cast_fp16")]; tensor var_7596_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_683_cast_fp16)[name = tensor("op_7596_cast_fp16")]; tensor var_7597_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_685_cast_fp16)[name = tensor("op_7597_cast_fp16")]; tensor var_7598_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_687_cast_fp16)[name = tensor("op_7598_cast_fp16")]; tensor var_7599_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_689_cast_fp16)[name = tensor("op_7599_cast_fp16")]; tensor var_7600_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_691_cast_fp16)[name = tensor("op_7600_cast_fp16")]; tensor var_7601_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_693_cast_fp16)[name = tensor("op_7601_cast_fp16")]; tensor var_7602_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_695_cast_fp16)[name = tensor("op_7602_cast_fp16")]; tensor var_7603_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_697_cast_fp16)[name = tensor("op_7603_cast_fp16")]; tensor var_7604_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_699_cast_fp16)[name = tensor("op_7604_cast_fp16")]; tensor var_7605_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_701_cast_fp16)[name = tensor("op_7605_cast_fp16")]; tensor var_7606_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_703_cast_fp16)[name = tensor("op_7606_cast_fp16")]; tensor var_7607_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_705_cast_fp16)[name = tensor("op_7607_cast_fp16")]; tensor var_7608_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_707_cast_fp16)[name = tensor("op_7608_cast_fp16")]; tensor var_7609_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_709_cast_fp16)[name = tensor("op_7609_cast_fp16")]; tensor var_7610_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_711_cast_fp16)[name = tensor("op_7610_cast_fp16")]; tensor var_7611_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_713_cast_fp16)[name = tensor("op_7611_cast_fp16")]; tensor var_7612_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_715_cast_fp16)[name = tensor("op_7612_cast_fp16")]; tensor var_7613_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_717_cast_fp16)[name = tensor("op_7613_cast_fp16")]; tensor var_7614_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_719_cast_fp16)[name = tensor("op_7614_cast_fp16")]; tensor var_7615_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_721_cast_fp16)[name = tensor("op_7615_cast_fp16")]; tensor var_7616_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_723_cast_fp16)[name = tensor("op_7616_cast_fp16")]; tensor var_7617_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_725_cast_fp16)[name = tensor("op_7617_cast_fp16")]; tensor var_7618_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_727_cast_fp16)[name = tensor("op_7618_cast_fp16")]; tensor var_7619_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_729_cast_fp16)[name = tensor("op_7619_cast_fp16")]; tensor var_7620_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_731_cast_fp16)[name = tensor("op_7620_cast_fp16")]; tensor var_7621_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_733_cast_fp16)[name = tensor("op_7621_cast_fp16")]; tensor var_7622_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_735_cast_fp16)[name = tensor("op_7622_cast_fp16")]; tensor var_7623_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_737_cast_fp16)[name = tensor("op_7623_cast_fp16")]; tensor var_7624_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_739_cast_fp16)[name = tensor("op_7624_cast_fp16")]; tensor var_7625_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_741_cast_fp16)[name = tensor("op_7625_cast_fp16")]; tensor var_7626_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_743_cast_fp16)[name = tensor("op_7626_cast_fp16")]; tensor var_7627_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_745_cast_fp16)[name = tensor("op_7627_cast_fp16")]; tensor var_7628_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_747_cast_fp16)[name = tensor("op_7628_cast_fp16")]; tensor var_7629_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_749_cast_fp16)[name = tensor("op_7629_cast_fp16")]; tensor var_7630_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_751_cast_fp16)[name = tensor("op_7630_cast_fp16")]; tensor var_7631_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_753_cast_fp16)[name = tensor("op_7631_cast_fp16")]; tensor var_7632_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_755_cast_fp16)[name = tensor("op_7632_cast_fp16")]; tensor var_7633_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_757_cast_fp16)[name = tensor("op_7633_cast_fp16")]; tensor var_7634_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_759_cast_fp16)[name = tensor("op_7634_cast_fp16")]; tensor var_7635_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_761_cast_fp16)[name = tensor("op_7635_cast_fp16")]; tensor var_7636_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_763_cast_fp16)[name = tensor("op_7636_cast_fp16")]; tensor var_7637_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_765_cast_fp16)[name = tensor("op_7637_cast_fp16")]; tensor var_7638_cast_fp16 = softmax(axis = var_6864, x = aw_chunk_767_cast_fp16)[name = tensor("op_7638_cast_fp16")]; tensor var_7640_equation_0 = const()[name = tensor("op_7640_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7640_cast_fp16 = einsum(equation = var_7640_equation_0, values = (var_7352_cast_fp16, var_7591_cast_fp16))[name = tensor("op_7640_cast_fp16")]; tensor var_7642_equation_0 = const()[name = tensor("op_7642_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7642_cast_fp16 = einsum(equation = var_7642_equation_0, values = (var_7352_cast_fp16, var_7592_cast_fp16))[name = tensor("op_7642_cast_fp16")]; tensor var_7644_equation_0 = const()[name = tensor("op_7644_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7644_cast_fp16 = einsum(equation = var_7644_equation_0, values = (var_7352_cast_fp16, var_7593_cast_fp16))[name = tensor("op_7644_cast_fp16")]; tensor var_7646_equation_0 = const()[name = tensor("op_7646_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7646_cast_fp16 = einsum(equation = var_7646_equation_0, values = (var_7352_cast_fp16, var_7594_cast_fp16))[name = tensor("op_7646_cast_fp16")]; tensor var_7648_equation_0 = const()[name = tensor("op_7648_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7648_cast_fp16 = einsum(equation = var_7648_equation_0, values = (var_7356_cast_fp16, var_7595_cast_fp16))[name = tensor("op_7648_cast_fp16")]; tensor var_7650_equation_0 = const()[name = tensor("op_7650_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7650_cast_fp16 = einsum(equation = var_7650_equation_0, values = (var_7356_cast_fp16, var_7596_cast_fp16))[name = tensor("op_7650_cast_fp16")]; tensor var_7652_equation_0 = const()[name = tensor("op_7652_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7652_cast_fp16 = einsum(equation = var_7652_equation_0, values = (var_7356_cast_fp16, var_7597_cast_fp16))[name = tensor("op_7652_cast_fp16")]; tensor var_7654_equation_0 = const()[name = tensor("op_7654_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7654_cast_fp16 = einsum(equation = var_7654_equation_0, values = (var_7356_cast_fp16, var_7598_cast_fp16))[name = tensor("op_7654_cast_fp16")]; tensor var_7656_equation_0 = const()[name = tensor("op_7656_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7656_cast_fp16 = einsum(equation = var_7656_equation_0, values = (var_7360_cast_fp16, var_7599_cast_fp16))[name = tensor("op_7656_cast_fp16")]; tensor var_7658_equation_0 = const()[name = tensor("op_7658_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7658_cast_fp16 = einsum(equation = var_7658_equation_0, values = (var_7360_cast_fp16, var_7600_cast_fp16))[name = tensor("op_7658_cast_fp16")]; tensor var_7660_equation_0 = const()[name = tensor("op_7660_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7660_cast_fp16 = einsum(equation = var_7660_equation_0, values = (var_7360_cast_fp16, var_7601_cast_fp16))[name = tensor("op_7660_cast_fp16")]; tensor var_7662_equation_0 = const()[name = tensor("op_7662_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7662_cast_fp16 = einsum(equation = var_7662_equation_0, values = (var_7360_cast_fp16, var_7602_cast_fp16))[name = tensor("op_7662_cast_fp16")]; tensor var_7664_equation_0 = const()[name = tensor("op_7664_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7664_cast_fp16 = einsum(equation = var_7664_equation_0, values = (var_7364_cast_fp16, var_7603_cast_fp16))[name = tensor("op_7664_cast_fp16")]; tensor var_7666_equation_0 = const()[name = tensor("op_7666_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7666_cast_fp16 = einsum(equation = var_7666_equation_0, values = (var_7364_cast_fp16, var_7604_cast_fp16))[name = tensor("op_7666_cast_fp16")]; tensor var_7668_equation_0 = const()[name = tensor("op_7668_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7668_cast_fp16 = einsum(equation = var_7668_equation_0, values = (var_7364_cast_fp16, var_7605_cast_fp16))[name = tensor("op_7668_cast_fp16")]; tensor var_7670_equation_0 = const()[name = tensor("op_7670_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7670_cast_fp16 = einsum(equation = var_7670_equation_0, values = (var_7364_cast_fp16, var_7606_cast_fp16))[name = tensor("op_7670_cast_fp16")]; tensor var_7672_equation_0 = const()[name = tensor("op_7672_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7672_cast_fp16 = einsum(equation = var_7672_equation_0, values = (var_7368_cast_fp16, var_7607_cast_fp16))[name = tensor("op_7672_cast_fp16")]; tensor var_7674_equation_0 = const()[name = tensor("op_7674_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7674_cast_fp16 = einsum(equation = var_7674_equation_0, values = (var_7368_cast_fp16, var_7608_cast_fp16))[name = tensor("op_7674_cast_fp16")]; tensor var_7676_equation_0 = const()[name = tensor("op_7676_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7676_cast_fp16 = einsum(equation = var_7676_equation_0, values = (var_7368_cast_fp16, var_7609_cast_fp16))[name = tensor("op_7676_cast_fp16")]; tensor var_7678_equation_0 = const()[name = tensor("op_7678_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7678_cast_fp16 = einsum(equation = var_7678_equation_0, values = (var_7368_cast_fp16, var_7610_cast_fp16))[name = tensor("op_7678_cast_fp16")]; tensor var_7680_equation_0 = const()[name = tensor("op_7680_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7680_cast_fp16 = einsum(equation = var_7680_equation_0, values = (var_7372_cast_fp16, var_7611_cast_fp16))[name = tensor("op_7680_cast_fp16")]; tensor var_7682_equation_0 = const()[name = tensor("op_7682_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7682_cast_fp16 = einsum(equation = var_7682_equation_0, values = (var_7372_cast_fp16, var_7612_cast_fp16))[name = tensor("op_7682_cast_fp16")]; tensor var_7684_equation_0 = const()[name = tensor("op_7684_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7684_cast_fp16 = einsum(equation = var_7684_equation_0, values = (var_7372_cast_fp16, var_7613_cast_fp16))[name = tensor("op_7684_cast_fp16")]; tensor var_7686_equation_0 = const()[name = tensor("op_7686_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7686_cast_fp16 = einsum(equation = var_7686_equation_0, values = (var_7372_cast_fp16, var_7614_cast_fp16))[name = tensor("op_7686_cast_fp16")]; tensor var_7688_equation_0 = const()[name = tensor("op_7688_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7688_cast_fp16 = einsum(equation = var_7688_equation_0, values = (var_7376_cast_fp16, var_7615_cast_fp16))[name = tensor("op_7688_cast_fp16")]; tensor var_7690_equation_0 = const()[name = tensor("op_7690_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7690_cast_fp16 = einsum(equation = var_7690_equation_0, values = (var_7376_cast_fp16, var_7616_cast_fp16))[name = tensor("op_7690_cast_fp16")]; tensor var_7692_equation_0 = const()[name = tensor("op_7692_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7692_cast_fp16 = einsum(equation = var_7692_equation_0, values = (var_7376_cast_fp16, var_7617_cast_fp16))[name = tensor("op_7692_cast_fp16")]; tensor var_7694_equation_0 = const()[name = tensor("op_7694_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7694_cast_fp16 = einsum(equation = var_7694_equation_0, values = (var_7376_cast_fp16, var_7618_cast_fp16))[name = tensor("op_7694_cast_fp16")]; tensor var_7696_equation_0 = const()[name = tensor("op_7696_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7696_cast_fp16 = einsum(equation = var_7696_equation_0, values = (var_7380_cast_fp16, var_7619_cast_fp16))[name = tensor("op_7696_cast_fp16")]; tensor var_7698_equation_0 = const()[name = tensor("op_7698_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7698_cast_fp16 = einsum(equation = var_7698_equation_0, values = (var_7380_cast_fp16, var_7620_cast_fp16))[name = tensor("op_7698_cast_fp16")]; tensor var_7700_equation_0 = const()[name = tensor("op_7700_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7700_cast_fp16 = einsum(equation = var_7700_equation_0, values = (var_7380_cast_fp16, var_7621_cast_fp16))[name = tensor("op_7700_cast_fp16")]; tensor var_7702_equation_0 = const()[name = tensor("op_7702_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7702_cast_fp16 = einsum(equation = var_7702_equation_0, values = (var_7380_cast_fp16, var_7622_cast_fp16))[name = tensor("op_7702_cast_fp16")]; tensor var_7704_equation_0 = const()[name = tensor("op_7704_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7704_cast_fp16 = einsum(equation = var_7704_equation_0, values = (var_7384_cast_fp16, var_7623_cast_fp16))[name = tensor("op_7704_cast_fp16")]; tensor var_7706_equation_0 = const()[name = tensor("op_7706_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7706_cast_fp16 = einsum(equation = var_7706_equation_0, values = (var_7384_cast_fp16, var_7624_cast_fp16))[name = tensor("op_7706_cast_fp16")]; tensor var_7708_equation_0 = const()[name = tensor("op_7708_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7708_cast_fp16 = einsum(equation = var_7708_equation_0, values = (var_7384_cast_fp16, var_7625_cast_fp16))[name = tensor("op_7708_cast_fp16")]; tensor var_7710_equation_0 = const()[name = tensor("op_7710_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7710_cast_fp16 = einsum(equation = var_7710_equation_0, values = (var_7384_cast_fp16, var_7626_cast_fp16))[name = tensor("op_7710_cast_fp16")]; tensor var_7712_equation_0 = const()[name = tensor("op_7712_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7712_cast_fp16 = einsum(equation = var_7712_equation_0, values = (var_7388_cast_fp16, var_7627_cast_fp16))[name = tensor("op_7712_cast_fp16")]; tensor var_7714_equation_0 = const()[name = tensor("op_7714_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7714_cast_fp16 = einsum(equation = var_7714_equation_0, values = (var_7388_cast_fp16, var_7628_cast_fp16))[name = tensor("op_7714_cast_fp16")]; tensor var_7716_equation_0 = const()[name = tensor("op_7716_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7716_cast_fp16 = einsum(equation = var_7716_equation_0, values = (var_7388_cast_fp16, var_7629_cast_fp16))[name = tensor("op_7716_cast_fp16")]; tensor var_7718_equation_0 = const()[name = tensor("op_7718_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7718_cast_fp16 = einsum(equation = var_7718_equation_0, values = (var_7388_cast_fp16, var_7630_cast_fp16))[name = tensor("op_7718_cast_fp16")]; tensor var_7720_equation_0 = const()[name = tensor("op_7720_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7720_cast_fp16 = einsum(equation = var_7720_equation_0, values = (var_7392_cast_fp16, var_7631_cast_fp16))[name = tensor("op_7720_cast_fp16")]; tensor var_7722_equation_0 = const()[name = tensor("op_7722_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7722_cast_fp16 = einsum(equation = var_7722_equation_0, values = (var_7392_cast_fp16, var_7632_cast_fp16))[name = tensor("op_7722_cast_fp16")]; tensor var_7724_equation_0 = const()[name = tensor("op_7724_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7724_cast_fp16 = einsum(equation = var_7724_equation_0, values = (var_7392_cast_fp16, var_7633_cast_fp16))[name = tensor("op_7724_cast_fp16")]; tensor var_7726_equation_0 = const()[name = tensor("op_7726_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7726_cast_fp16 = einsum(equation = var_7726_equation_0, values = (var_7392_cast_fp16, var_7634_cast_fp16))[name = tensor("op_7726_cast_fp16")]; tensor var_7728_equation_0 = const()[name = tensor("op_7728_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7728_cast_fp16 = einsum(equation = var_7728_equation_0, values = (var_7396_cast_fp16, var_7635_cast_fp16))[name = tensor("op_7728_cast_fp16")]; tensor var_7730_equation_0 = const()[name = tensor("op_7730_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7730_cast_fp16 = einsum(equation = var_7730_equation_0, values = (var_7396_cast_fp16, var_7636_cast_fp16))[name = tensor("op_7730_cast_fp16")]; tensor var_7732_equation_0 = const()[name = tensor("op_7732_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7732_cast_fp16 = einsum(equation = var_7732_equation_0, values = (var_7396_cast_fp16, var_7637_cast_fp16))[name = tensor("op_7732_cast_fp16")]; tensor var_7734_equation_0 = const()[name = tensor("op_7734_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_7734_cast_fp16 = einsum(equation = var_7734_equation_0, values = (var_7396_cast_fp16, var_7638_cast_fp16))[name = tensor("op_7734_cast_fp16")]; tensor var_7736_interleave_0 = const()[name = tensor("op_7736_interleave_0"), val = tensor(false)]; tensor var_7736_cast_fp16 = concat(axis = var_6847, interleave = var_7736_interleave_0, values = (var_7640_cast_fp16, var_7642_cast_fp16, var_7644_cast_fp16, var_7646_cast_fp16))[name = tensor("op_7736_cast_fp16")]; tensor var_7738_interleave_0 = const()[name = tensor("op_7738_interleave_0"), val = tensor(false)]; tensor var_7738_cast_fp16 = concat(axis = var_6847, interleave = var_7738_interleave_0, values = (var_7648_cast_fp16, var_7650_cast_fp16, var_7652_cast_fp16, var_7654_cast_fp16))[name = tensor("op_7738_cast_fp16")]; tensor var_7740_interleave_0 = const()[name = tensor("op_7740_interleave_0"), val = tensor(false)]; tensor var_7740_cast_fp16 = concat(axis = var_6847, interleave = var_7740_interleave_0, values = (var_7656_cast_fp16, var_7658_cast_fp16, var_7660_cast_fp16, var_7662_cast_fp16))[name = tensor("op_7740_cast_fp16")]; tensor var_7742_interleave_0 = const()[name = tensor("op_7742_interleave_0"), val = tensor(false)]; tensor var_7742_cast_fp16 = concat(axis = var_6847, interleave = var_7742_interleave_0, values = (var_7664_cast_fp16, var_7666_cast_fp16, var_7668_cast_fp16, var_7670_cast_fp16))[name = tensor("op_7742_cast_fp16")]; tensor var_7744_interleave_0 = const()[name = tensor("op_7744_interleave_0"), val = tensor(false)]; tensor var_7744_cast_fp16 = concat(axis = var_6847, interleave = var_7744_interleave_0, values = (var_7672_cast_fp16, var_7674_cast_fp16, var_7676_cast_fp16, var_7678_cast_fp16))[name = tensor("op_7744_cast_fp16")]; tensor var_7746_interleave_0 = const()[name = tensor("op_7746_interleave_0"), val = tensor(false)]; tensor var_7746_cast_fp16 = concat(axis = var_6847, interleave = var_7746_interleave_0, values = (var_7680_cast_fp16, var_7682_cast_fp16, var_7684_cast_fp16, var_7686_cast_fp16))[name = tensor("op_7746_cast_fp16")]; tensor var_7748_interleave_0 = const()[name = tensor("op_7748_interleave_0"), val = tensor(false)]; tensor var_7748_cast_fp16 = concat(axis = var_6847, interleave = var_7748_interleave_0, values = (var_7688_cast_fp16, var_7690_cast_fp16, var_7692_cast_fp16, var_7694_cast_fp16))[name = tensor("op_7748_cast_fp16")]; tensor var_7750_interleave_0 = const()[name = tensor("op_7750_interleave_0"), val = tensor(false)]; tensor var_7750_cast_fp16 = concat(axis = var_6847, interleave = var_7750_interleave_0, values = (var_7696_cast_fp16, var_7698_cast_fp16, var_7700_cast_fp16, var_7702_cast_fp16))[name = tensor("op_7750_cast_fp16")]; tensor var_7752_interleave_0 = const()[name = tensor("op_7752_interleave_0"), val = tensor(false)]; tensor var_7752_cast_fp16 = concat(axis = var_6847, interleave = var_7752_interleave_0, values = (var_7704_cast_fp16, var_7706_cast_fp16, var_7708_cast_fp16, var_7710_cast_fp16))[name = tensor("op_7752_cast_fp16")]; tensor var_7754_interleave_0 = const()[name = tensor("op_7754_interleave_0"), val = tensor(false)]; tensor var_7754_cast_fp16 = concat(axis = var_6847, interleave = var_7754_interleave_0, values = (var_7712_cast_fp16, var_7714_cast_fp16, var_7716_cast_fp16, var_7718_cast_fp16))[name = tensor("op_7754_cast_fp16")]; tensor var_7756_interleave_0 = const()[name = tensor("op_7756_interleave_0"), val = tensor(false)]; tensor var_7756_cast_fp16 = concat(axis = var_6847, interleave = var_7756_interleave_0, values = (var_7720_cast_fp16, var_7722_cast_fp16, var_7724_cast_fp16, var_7726_cast_fp16))[name = tensor("op_7756_cast_fp16")]; tensor var_7758_interleave_0 = const()[name = tensor("op_7758_interleave_0"), val = tensor(false)]; tensor var_7758_cast_fp16 = concat(axis = var_6847, interleave = var_7758_interleave_0, values = (var_7728_cast_fp16, var_7730_cast_fp16, var_7732_cast_fp16, var_7734_cast_fp16))[name = tensor("op_7758_cast_fp16")]; tensor input_57_interleave_0 = const()[name = tensor("input_57_interleave_0"), val = tensor(false)]; tensor input_57_cast_fp16 = concat(axis = var_6864, interleave = input_57_interleave_0, values = (var_7736_cast_fp16, var_7738_cast_fp16, var_7740_cast_fp16, var_7742_cast_fp16, var_7744_cast_fp16, var_7746_cast_fp16, var_7748_cast_fp16, var_7750_cast_fp16, var_7752_cast_fp16, var_7754_cast_fp16, var_7756_cast_fp16, var_7758_cast_fp16))[name = tensor("input_57_cast_fp16")]; tensor var_7763 = const()[name = tensor("op_7763"), val = tensor([1, 1])]; tensor var_7765 = const()[name = tensor("op_7765"), val = tensor([1, 1])]; tensor obj_31_pad_type_0 = const()[name = tensor("obj_31_pad_type_0"), val = tensor("custom")]; tensor obj_31_pad_0 = const()[name = tensor("obj_31_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108989952)))]; tensor layers_7_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110169664)))]; tensor obj_31_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_bias_to_fp16, dilations = var_7765, groups = var_6864, pad = obj_31_pad_0, pad_type = obj_31_pad_type_0, strides = var_7763, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_57_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_7771 = const()[name = tensor("op_7771"), val = tensor([1])]; tensor channels_mean_31_cast_fp16 = reduce_mean(axes = var_7771, keep_dims = var_6865, 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_7775 = const()[name = tensor("op_7775"), val = tensor([1])]; tensor var_7776_cast_fp16 = reduce_mean(axes = var_7775, keep_dims = var_6865, x = zero_mean_sq_31_cast_fp16)[name = tensor("op_7776_cast_fp16")]; tensor var_7777_to_fp16 = const()[name = tensor("op_7777_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7778_cast_fp16 = add(x = var_7776_cast_fp16, y = var_7777_to_fp16)[name = tensor("op_7778_cast_fp16")]; tensor denom_31_epsilon_0_to_fp16 = const()[name = tensor("denom_31_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_31_cast_fp16 = rsqrt(epsilon = denom_31_epsilon_0_to_fp16, x = var_7778_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_59_gamma_0_to_fp16 = const()[name = tensor("input_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110171264)))]; tensor input_59_beta_0_to_fp16 = const()[name = tensor("input_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110172864)))]; tensor input_59_epsilon_0_to_fp16 = const()[name = tensor("input_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_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_59_cast_fp16")]; tensor var_7789 = const()[name = tensor("op_7789"), val = tensor([1, 1])]; tensor var_7791 = const()[name = tensor("op_7791"), 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_7_fc1_weight_to_fp16 = const()[name = tensor("layers_7_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110174464)))]; tensor layers_7_fc1_bias_to_fp16 = const()[name = tensor("layers_7_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114893120)))]; tensor input_61_cast_fp16 = conv(bias = layers_7_fc1_bias_to_fp16, dilations = var_7791, groups = var_6864, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = var_7789, weight = layers_7_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = tensor("input_61_cast_fp16")]; tensor input_63_mode_0 = const()[name = tensor("input_63_mode_0"), val = tensor("EXACT")]; tensor input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = tensor("input_63_cast_fp16")]; tensor var_7797 = const()[name = tensor("op_7797"), val = tensor([1, 1])]; tensor var_7799 = const()[name = tensor("op_7799"), val = tensor([1, 1])]; tensor hidden_states_19_pad_type_0 = const()[name = tensor("hidden_states_19_pad_type_0"), val = tensor("custom")]; tensor hidden_states_19_pad_0 = const()[name = tensor("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_7_fc2_weight_to_fp16 = const()[name = tensor("layers_7_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114899328)))]; tensor layers_7_fc2_bias_to_fp16 = const()[name = tensor("layers_7_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119617984)))]; tensor hidden_states_19_cast_fp16 = conv(bias = layers_7_fc2_bias_to_fp16, dilations = var_7799, groups = var_6864, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_7797, weight = layers_7_fc2_weight_to_fp16, x = input_63_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_7806 = const()[name = tensor("op_7806"), val = tensor(3)]; tensor var_7823 = const()[name = tensor("op_7823"), val = tensor(1)]; tensor var_7824 = const()[name = tensor("op_7824"), val = tensor(true)]; tensor var_7834 = const()[name = tensor("op_7834"), val = tensor([1])]; tensor channels_mean_33_cast_fp16 = reduce_mean(axes = var_7834, keep_dims = var_7824, 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_7838 = const()[name = tensor("op_7838"), val = tensor([1])]; tensor var_7839_cast_fp16 = reduce_mean(axes = var_7838, keep_dims = var_7824, x = zero_mean_sq_33_cast_fp16)[name = tensor("op_7839_cast_fp16")]; tensor var_7840_to_fp16 = const()[name = tensor("op_7840_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_7841_cast_fp16 = add(x = var_7839_cast_fp16, y = var_7840_to_fp16)[name = tensor("op_7841_cast_fp16")]; tensor denom_33_epsilon_0_to_fp16 = const()[name = tensor("denom_33_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_33_cast_fp16 = rsqrt(epsilon = denom_33_epsilon_0_to_fp16, x = var_7841_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(119619584)))]; 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(119621184)))]; 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_7856 = const()[name = tensor("op_7856"), val = tensor([1, 1])]; tensor var_7858 = const()[name = tensor("op_7858"), val = tensor([1, 1])]; tensor query_17_pad_type_0 = const()[name = tensor("query_17_pad_type_0"), val = tensor("custom")]; tensor query_17_pad_0 = const()[name = tensor("query_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119622784)))]; tensor layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120802496)))]; tensor query_17_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_bias_to_fp16, dilations = var_7858, groups = var_7823, pad = query_17_pad_0, pad_type = query_17_pad_type_0, strides = var_7856, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("query_17_cast_fp16")]; tensor var_7862 = const()[name = tensor("op_7862"), val = tensor([1, 1])]; tensor var_7864 = const()[name = tensor("op_7864"), val = tensor([1, 1])]; tensor key_17_pad_type_0 = const()[name = tensor("key_17_pad_type_0"), val = tensor("custom")]; tensor key_17_pad_0 = const()[name = tensor("key_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120804096)))]; tensor key_17_cast_fp16 = conv(dilations = var_7864, groups = var_7823, pad = key_17_pad_0, pad_type = key_17_pad_type_0, strides = var_7862, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("key_17_cast_fp16")]; tensor var_7869 = const()[name = tensor("op_7869"), val = tensor([1, 1])]; tensor var_7871 = const()[name = tensor("op_7871"), val = tensor([1, 1])]; tensor value_17_pad_type_0 = const()[name = tensor("value_17_pad_type_0"), val = tensor("custom")]; tensor value_17_pad_0 = const()[name = tensor("value_17_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121983808)))]; tensor layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123163520)))]; tensor value_17_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_bias_to_fp16, dilations = var_7871, groups = var_7823, pad = value_17_pad_0, pad_type = value_17_pad_type_0, strides = var_7869, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = obj_33_cast_fp16)[name = tensor("value_17_cast_fp16")]; tensor var_7878_begin_0 = const()[name = tensor("op_7878_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7878_end_0 = const()[name = tensor("op_7878_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7878_end_mask_0 = const()[name = tensor("op_7878_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7878_cast_fp16 = slice_by_index(begin = var_7878_begin_0, end = var_7878_end_0, end_mask = var_7878_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7878_cast_fp16")]; tensor var_7882_begin_0 = const()[name = tensor("op_7882_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_7882_end_0 = const()[name = tensor("op_7882_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_7882_end_mask_0 = const()[name = tensor("op_7882_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7882_cast_fp16 = slice_by_index(begin = var_7882_begin_0, end = var_7882_end_0, end_mask = var_7882_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7882_cast_fp16")]; tensor var_7886_begin_0 = const()[name = tensor("op_7886_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_7886_end_0 = const()[name = tensor("op_7886_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_7886_end_mask_0 = const()[name = tensor("op_7886_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7886_cast_fp16 = slice_by_index(begin = var_7886_begin_0, end = var_7886_end_0, end_mask = var_7886_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7886_cast_fp16")]; tensor var_7890_begin_0 = const()[name = tensor("op_7890_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_7890_end_0 = const()[name = tensor("op_7890_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_7890_end_mask_0 = const()[name = tensor("op_7890_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7890_cast_fp16 = slice_by_index(begin = var_7890_begin_0, end = var_7890_end_0, end_mask = var_7890_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7890_cast_fp16")]; tensor var_7894_begin_0 = const()[name = tensor("op_7894_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_7894_end_0 = const()[name = tensor("op_7894_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_7894_end_mask_0 = const()[name = tensor("op_7894_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7894_cast_fp16 = slice_by_index(begin = var_7894_begin_0, end = var_7894_end_0, end_mask = var_7894_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7894_cast_fp16")]; tensor var_7898_begin_0 = const()[name = tensor("op_7898_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_7898_end_0 = const()[name = tensor("op_7898_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_7898_end_mask_0 = const()[name = tensor("op_7898_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7898_cast_fp16 = slice_by_index(begin = var_7898_begin_0, end = var_7898_end_0, end_mask = var_7898_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7898_cast_fp16")]; tensor var_7902_begin_0 = const()[name = tensor("op_7902_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_7902_end_0 = const()[name = tensor("op_7902_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_7902_end_mask_0 = const()[name = tensor("op_7902_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7902_cast_fp16 = slice_by_index(begin = var_7902_begin_0, end = var_7902_end_0, end_mask = var_7902_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7902_cast_fp16")]; tensor var_7906_begin_0 = const()[name = tensor("op_7906_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_7906_end_0 = const()[name = tensor("op_7906_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_7906_end_mask_0 = const()[name = tensor("op_7906_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7906_cast_fp16 = slice_by_index(begin = var_7906_begin_0, end = var_7906_end_0, end_mask = var_7906_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7906_cast_fp16")]; tensor var_7910_begin_0 = const()[name = tensor("op_7910_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_7910_end_0 = const()[name = tensor("op_7910_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_7910_end_mask_0 = const()[name = tensor("op_7910_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7910_cast_fp16 = slice_by_index(begin = var_7910_begin_0, end = var_7910_end_0, end_mask = var_7910_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7910_cast_fp16")]; tensor var_7914_begin_0 = const()[name = tensor("op_7914_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_7914_end_0 = const()[name = tensor("op_7914_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_7914_end_mask_0 = const()[name = tensor("op_7914_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7914_cast_fp16 = slice_by_index(begin = var_7914_begin_0, end = var_7914_end_0, end_mask = var_7914_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7914_cast_fp16")]; tensor var_7918_begin_0 = const()[name = tensor("op_7918_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_7918_end_0 = const()[name = tensor("op_7918_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_7918_end_mask_0 = const()[name = tensor("op_7918_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7918_cast_fp16 = slice_by_index(begin = var_7918_begin_0, end = var_7918_end_0, end_mask = var_7918_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7918_cast_fp16")]; tensor var_7922_begin_0 = const()[name = tensor("op_7922_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_7922_end_0 = const()[name = tensor("op_7922_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_7922_end_mask_0 = const()[name = tensor("op_7922_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_7922_cast_fp16 = slice_by_index(begin = var_7922_begin_0, end = var_7922_end_0, end_mask = var_7922_end_mask_0, x = query_17_cast_fp16)[name = tensor("op_7922_cast_fp16")]; tensor var_7931_begin_0 = const()[name = tensor("op_7931_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7931_end_0 = const()[name = tensor("op_7931_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7931_end_mask_0 = const()[name = tensor("op_7931_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7931_cast_fp16 = slice_by_index(begin = var_7931_begin_0, end = var_7931_end_0, end_mask = var_7931_end_mask_0, x = var_7878_cast_fp16)[name = tensor("op_7931_cast_fp16")]; tensor var_7938_begin_0 = const()[name = tensor("op_7938_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7938_end_0 = const()[name = tensor("op_7938_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7938_end_mask_0 = const()[name = tensor("op_7938_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7938_cast_fp16 = slice_by_index(begin = var_7938_begin_0, end = var_7938_end_0, end_mask = var_7938_end_mask_0, x = var_7878_cast_fp16)[name = tensor("op_7938_cast_fp16")]; tensor var_7945_begin_0 = const()[name = tensor("op_7945_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7945_end_0 = const()[name = tensor("op_7945_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7945_end_mask_0 = const()[name = tensor("op_7945_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7945_cast_fp16 = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = var_7878_cast_fp16)[name = tensor("op_7945_cast_fp16")]; tensor var_7952_begin_0 = const()[name = tensor("op_7952_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7952_end_0 = const()[name = tensor("op_7952_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7952_end_mask_0 = const()[name = tensor("op_7952_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7952_cast_fp16 = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = var_7878_cast_fp16)[name = tensor("op_7952_cast_fp16")]; tensor var_7959_begin_0 = const()[name = tensor("op_7959_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7959_end_0 = const()[name = tensor("op_7959_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7959_end_mask_0 = const()[name = tensor("op_7959_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7959_cast_fp16 = slice_by_index(begin = var_7959_begin_0, end = var_7959_end_0, end_mask = var_7959_end_mask_0, x = var_7882_cast_fp16)[name = tensor("op_7959_cast_fp16")]; tensor var_7966_begin_0 = const()[name = tensor("op_7966_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7966_end_0 = const()[name = tensor("op_7966_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7966_end_mask_0 = const()[name = tensor("op_7966_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7966_cast_fp16 = slice_by_index(begin = var_7966_begin_0, end = var_7966_end_0, end_mask = var_7966_end_mask_0, x = var_7882_cast_fp16)[name = tensor("op_7966_cast_fp16")]; tensor var_7973_begin_0 = const()[name = tensor("op_7973_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_7973_end_0 = const()[name = tensor("op_7973_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_7973_end_mask_0 = const()[name = tensor("op_7973_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7973_cast_fp16 = slice_by_index(begin = var_7973_begin_0, end = var_7973_end_0, end_mask = var_7973_end_mask_0, x = var_7882_cast_fp16)[name = tensor("op_7973_cast_fp16")]; tensor var_7980_begin_0 = const()[name = tensor("op_7980_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_7980_end_0 = const()[name = tensor("op_7980_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_7980_end_mask_0 = const()[name = tensor("op_7980_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7980_cast_fp16 = slice_by_index(begin = var_7980_begin_0, end = var_7980_end_0, end_mask = var_7980_end_mask_0, x = var_7882_cast_fp16)[name = tensor("op_7980_cast_fp16")]; tensor var_7987_begin_0 = const()[name = tensor("op_7987_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_7987_end_0 = const()[name = tensor("op_7987_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_7987_end_mask_0 = const()[name = tensor("op_7987_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7987_cast_fp16 = slice_by_index(begin = var_7987_begin_0, end = var_7987_end_0, end_mask = var_7987_end_mask_0, x = var_7886_cast_fp16)[name = tensor("op_7987_cast_fp16")]; tensor var_7994_begin_0 = const()[name = tensor("op_7994_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_7994_end_0 = const()[name = tensor("op_7994_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_7994_end_mask_0 = const()[name = tensor("op_7994_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_7994_cast_fp16 = slice_by_index(begin = var_7994_begin_0, end = var_7994_end_0, end_mask = var_7994_end_mask_0, x = var_7886_cast_fp16)[name = tensor("op_7994_cast_fp16")]; tensor var_8001_begin_0 = const()[name = tensor("op_8001_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8001_end_0 = const()[name = tensor("op_8001_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8001_end_mask_0 = const()[name = tensor("op_8001_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8001_cast_fp16 = slice_by_index(begin = var_8001_begin_0, end = var_8001_end_0, end_mask = var_8001_end_mask_0, x = var_7886_cast_fp16)[name = tensor("op_8001_cast_fp16")]; tensor var_8008_begin_0 = const()[name = tensor("op_8008_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8008_end_0 = const()[name = tensor("op_8008_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8008_end_mask_0 = const()[name = tensor("op_8008_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8008_cast_fp16 = slice_by_index(begin = var_8008_begin_0, end = var_8008_end_0, end_mask = var_8008_end_mask_0, x = var_7886_cast_fp16)[name = tensor("op_8008_cast_fp16")]; tensor var_8015_begin_0 = const()[name = tensor("op_8015_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8015_end_0 = const()[name = tensor("op_8015_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8015_end_mask_0 = const()[name = tensor("op_8015_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8015_cast_fp16 = slice_by_index(begin = var_8015_begin_0, end = var_8015_end_0, end_mask = var_8015_end_mask_0, x = var_7890_cast_fp16)[name = tensor("op_8015_cast_fp16")]; tensor var_8022_begin_0 = const()[name = tensor("op_8022_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8022_end_0 = const()[name = tensor("op_8022_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8022_end_mask_0 = const()[name = tensor("op_8022_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8022_cast_fp16 = slice_by_index(begin = var_8022_begin_0, end = var_8022_end_0, end_mask = var_8022_end_mask_0, x = var_7890_cast_fp16)[name = tensor("op_8022_cast_fp16")]; tensor var_8029_begin_0 = const()[name = tensor("op_8029_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8029_end_0 = const()[name = tensor("op_8029_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8029_end_mask_0 = const()[name = tensor("op_8029_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8029_cast_fp16 = slice_by_index(begin = var_8029_begin_0, end = var_8029_end_0, end_mask = var_8029_end_mask_0, x = var_7890_cast_fp16)[name = tensor("op_8029_cast_fp16")]; tensor var_8036_begin_0 = const()[name = tensor("op_8036_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8036_end_0 = const()[name = tensor("op_8036_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8036_end_mask_0 = const()[name = tensor("op_8036_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8036_cast_fp16 = slice_by_index(begin = var_8036_begin_0, end = var_8036_end_0, end_mask = var_8036_end_mask_0, x = var_7890_cast_fp16)[name = tensor("op_8036_cast_fp16")]; tensor var_8043_begin_0 = const()[name = tensor("op_8043_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8043_end_0 = const()[name = tensor("op_8043_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8043_end_mask_0 = const()[name = tensor("op_8043_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8043_cast_fp16 = slice_by_index(begin = var_8043_begin_0, end = var_8043_end_0, end_mask = var_8043_end_mask_0, x = var_7894_cast_fp16)[name = tensor("op_8043_cast_fp16")]; tensor var_8050_begin_0 = const()[name = tensor("op_8050_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8050_end_0 = const()[name = tensor("op_8050_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8050_end_mask_0 = const()[name = tensor("op_8050_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8050_cast_fp16 = slice_by_index(begin = var_8050_begin_0, end = var_8050_end_0, end_mask = var_8050_end_mask_0, x = var_7894_cast_fp16)[name = tensor("op_8050_cast_fp16")]; tensor var_8057_begin_0 = const()[name = tensor("op_8057_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8057_end_0 = const()[name = tensor("op_8057_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8057_end_mask_0 = const()[name = tensor("op_8057_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8057_cast_fp16 = slice_by_index(begin = var_8057_begin_0, end = var_8057_end_0, end_mask = var_8057_end_mask_0, x = var_7894_cast_fp16)[name = tensor("op_8057_cast_fp16")]; tensor var_8064_begin_0 = const()[name = tensor("op_8064_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8064_end_0 = const()[name = tensor("op_8064_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8064_end_mask_0 = const()[name = tensor("op_8064_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8064_cast_fp16 = slice_by_index(begin = var_8064_begin_0, end = var_8064_end_0, end_mask = var_8064_end_mask_0, x = var_7894_cast_fp16)[name = tensor("op_8064_cast_fp16")]; tensor var_8071_begin_0 = const()[name = tensor("op_8071_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8071_end_0 = const()[name = tensor("op_8071_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8071_end_mask_0 = const()[name = tensor("op_8071_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8071_cast_fp16 = slice_by_index(begin = var_8071_begin_0, end = var_8071_end_0, end_mask = var_8071_end_mask_0, x = var_7898_cast_fp16)[name = tensor("op_8071_cast_fp16")]; tensor var_8078_begin_0 = const()[name = tensor("op_8078_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8078_end_0 = const()[name = tensor("op_8078_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8078_end_mask_0 = const()[name = tensor("op_8078_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8078_cast_fp16 = slice_by_index(begin = var_8078_begin_0, end = var_8078_end_0, end_mask = var_8078_end_mask_0, x = var_7898_cast_fp16)[name = tensor("op_8078_cast_fp16")]; tensor var_8085_begin_0 = const()[name = tensor("op_8085_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8085_end_0 = const()[name = tensor("op_8085_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8085_end_mask_0 = const()[name = tensor("op_8085_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8085_cast_fp16 = slice_by_index(begin = var_8085_begin_0, end = var_8085_end_0, end_mask = var_8085_end_mask_0, x = var_7898_cast_fp16)[name = tensor("op_8085_cast_fp16")]; tensor var_8092_begin_0 = const()[name = tensor("op_8092_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8092_end_0 = const()[name = tensor("op_8092_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8092_end_mask_0 = const()[name = tensor("op_8092_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8092_cast_fp16 = slice_by_index(begin = var_8092_begin_0, end = var_8092_end_0, end_mask = var_8092_end_mask_0, x = var_7898_cast_fp16)[name = tensor("op_8092_cast_fp16")]; tensor var_8099_begin_0 = const()[name = tensor("op_8099_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8099_end_0 = const()[name = tensor("op_8099_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8099_end_mask_0 = const()[name = tensor("op_8099_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8099_cast_fp16 = slice_by_index(begin = var_8099_begin_0, end = var_8099_end_0, end_mask = var_8099_end_mask_0, x = var_7902_cast_fp16)[name = tensor("op_8099_cast_fp16")]; tensor var_8106_begin_0 = const()[name = tensor("op_8106_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8106_end_0 = const()[name = tensor("op_8106_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8106_end_mask_0 = const()[name = tensor("op_8106_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8106_cast_fp16 = slice_by_index(begin = var_8106_begin_0, end = var_8106_end_0, end_mask = var_8106_end_mask_0, x = var_7902_cast_fp16)[name = tensor("op_8106_cast_fp16")]; tensor var_8113_begin_0 = const()[name = tensor("op_8113_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8113_end_0 = const()[name = tensor("op_8113_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8113_end_mask_0 = const()[name = tensor("op_8113_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8113_cast_fp16 = slice_by_index(begin = var_8113_begin_0, end = var_8113_end_0, end_mask = var_8113_end_mask_0, x = var_7902_cast_fp16)[name = tensor("op_8113_cast_fp16")]; tensor var_8120_begin_0 = const()[name = tensor("op_8120_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8120_end_0 = const()[name = tensor("op_8120_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8120_end_mask_0 = const()[name = tensor("op_8120_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8120_cast_fp16 = slice_by_index(begin = var_8120_begin_0, end = var_8120_end_0, end_mask = var_8120_end_mask_0, x = var_7902_cast_fp16)[name = tensor("op_8120_cast_fp16")]; tensor var_8127_begin_0 = const()[name = tensor("op_8127_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8127_end_0 = const()[name = tensor("op_8127_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8127_end_mask_0 = const()[name = tensor("op_8127_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8127_cast_fp16 = slice_by_index(begin = var_8127_begin_0, end = var_8127_end_0, end_mask = var_8127_end_mask_0, x = var_7906_cast_fp16)[name = tensor("op_8127_cast_fp16")]; tensor var_8134_begin_0 = const()[name = tensor("op_8134_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8134_end_0 = const()[name = tensor("op_8134_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8134_end_mask_0 = const()[name = tensor("op_8134_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8134_cast_fp16 = slice_by_index(begin = var_8134_begin_0, end = var_8134_end_0, end_mask = var_8134_end_mask_0, x = var_7906_cast_fp16)[name = tensor("op_8134_cast_fp16")]; tensor var_8141_begin_0 = const()[name = tensor("op_8141_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8141_end_0 = const()[name = tensor("op_8141_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8141_end_mask_0 = const()[name = tensor("op_8141_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8141_cast_fp16 = slice_by_index(begin = var_8141_begin_0, end = var_8141_end_0, end_mask = var_8141_end_mask_0, x = var_7906_cast_fp16)[name = tensor("op_8141_cast_fp16")]; tensor var_8148_begin_0 = const()[name = tensor("op_8148_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8148_end_0 = const()[name = tensor("op_8148_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8148_end_mask_0 = const()[name = tensor("op_8148_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8148_cast_fp16 = slice_by_index(begin = var_8148_begin_0, end = var_8148_end_0, end_mask = var_8148_end_mask_0, x = var_7906_cast_fp16)[name = tensor("op_8148_cast_fp16")]; tensor var_8155_begin_0 = const()[name = tensor("op_8155_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8155_end_0 = const()[name = tensor("op_8155_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8155_end_mask_0 = const()[name = tensor("op_8155_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8155_cast_fp16 = slice_by_index(begin = var_8155_begin_0, end = var_8155_end_0, end_mask = var_8155_end_mask_0, x = var_7910_cast_fp16)[name = tensor("op_8155_cast_fp16")]; tensor var_8162_begin_0 = const()[name = tensor("op_8162_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8162_end_0 = const()[name = tensor("op_8162_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8162_end_mask_0 = const()[name = tensor("op_8162_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8162_cast_fp16 = slice_by_index(begin = var_8162_begin_0, end = var_8162_end_0, end_mask = var_8162_end_mask_0, x = var_7910_cast_fp16)[name = tensor("op_8162_cast_fp16")]; tensor var_8169_begin_0 = const()[name = tensor("op_8169_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8169_end_0 = const()[name = tensor("op_8169_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8169_end_mask_0 = const()[name = tensor("op_8169_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8169_cast_fp16 = slice_by_index(begin = var_8169_begin_0, end = var_8169_end_0, end_mask = var_8169_end_mask_0, x = var_7910_cast_fp16)[name = tensor("op_8169_cast_fp16")]; tensor var_8176_begin_0 = const()[name = tensor("op_8176_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8176_end_0 = const()[name = tensor("op_8176_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8176_end_mask_0 = const()[name = tensor("op_8176_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8176_cast_fp16 = slice_by_index(begin = var_8176_begin_0, end = var_8176_end_0, end_mask = var_8176_end_mask_0, x = var_7910_cast_fp16)[name = tensor("op_8176_cast_fp16")]; tensor var_8183_begin_0 = const()[name = tensor("op_8183_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8183_end_0 = const()[name = tensor("op_8183_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8183_end_mask_0 = const()[name = tensor("op_8183_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8183_cast_fp16 = slice_by_index(begin = var_8183_begin_0, end = var_8183_end_0, end_mask = var_8183_end_mask_0, x = var_7914_cast_fp16)[name = tensor("op_8183_cast_fp16")]; tensor var_8190_begin_0 = const()[name = tensor("op_8190_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8190_end_0 = const()[name = tensor("op_8190_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8190_end_mask_0 = const()[name = tensor("op_8190_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8190_cast_fp16 = slice_by_index(begin = var_8190_begin_0, end = var_8190_end_0, end_mask = var_8190_end_mask_0, x = var_7914_cast_fp16)[name = tensor("op_8190_cast_fp16")]; tensor var_8197_begin_0 = const()[name = tensor("op_8197_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8197_end_0 = const()[name = tensor("op_8197_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8197_end_mask_0 = const()[name = tensor("op_8197_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8197_cast_fp16 = slice_by_index(begin = var_8197_begin_0, end = var_8197_end_0, end_mask = var_8197_end_mask_0, x = var_7914_cast_fp16)[name = tensor("op_8197_cast_fp16")]; tensor var_8204_begin_0 = const()[name = tensor("op_8204_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8204_end_0 = const()[name = tensor("op_8204_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8204_end_mask_0 = const()[name = tensor("op_8204_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8204_cast_fp16 = slice_by_index(begin = var_8204_begin_0, end = var_8204_end_0, end_mask = var_8204_end_mask_0, x = var_7914_cast_fp16)[name = tensor("op_8204_cast_fp16")]; tensor var_8211_begin_0 = const()[name = tensor("op_8211_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8211_end_0 = const()[name = tensor("op_8211_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8211_end_mask_0 = const()[name = tensor("op_8211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8211_cast_fp16 = slice_by_index(begin = var_8211_begin_0, end = var_8211_end_0, end_mask = var_8211_end_mask_0, x = var_7918_cast_fp16)[name = tensor("op_8211_cast_fp16")]; tensor var_8218_begin_0 = const()[name = tensor("op_8218_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8218_end_0 = const()[name = tensor("op_8218_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8218_end_mask_0 = const()[name = tensor("op_8218_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8218_cast_fp16 = slice_by_index(begin = var_8218_begin_0, end = var_8218_end_0, end_mask = var_8218_end_mask_0, x = var_7918_cast_fp16)[name = tensor("op_8218_cast_fp16")]; tensor var_8225_begin_0 = const()[name = tensor("op_8225_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8225_end_0 = const()[name = tensor("op_8225_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8225_end_mask_0 = const()[name = tensor("op_8225_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8225_cast_fp16 = slice_by_index(begin = var_8225_begin_0, end = var_8225_end_0, end_mask = var_8225_end_mask_0, x = var_7918_cast_fp16)[name = tensor("op_8225_cast_fp16")]; tensor var_8232_begin_0 = const()[name = tensor("op_8232_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8232_end_0 = const()[name = tensor("op_8232_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8232_end_mask_0 = const()[name = tensor("op_8232_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8232_cast_fp16 = slice_by_index(begin = var_8232_begin_0, end = var_8232_end_0, end_mask = var_8232_end_mask_0, x = var_7918_cast_fp16)[name = tensor("op_8232_cast_fp16")]; tensor var_8239_begin_0 = const()[name = tensor("op_8239_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8239_end_0 = const()[name = tensor("op_8239_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_8239_end_mask_0 = const()[name = tensor("op_8239_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8239_cast_fp16 = slice_by_index(begin = var_8239_begin_0, end = var_8239_end_0, end_mask = var_8239_end_mask_0, x = var_7922_cast_fp16)[name = tensor("op_8239_cast_fp16")]; tensor var_8246_begin_0 = const()[name = tensor("op_8246_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_8246_end_0 = const()[name = tensor("op_8246_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_8246_end_mask_0 = const()[name = tensor("op_8246_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8246_cast_fp16 = slice_by_index(begin = var_8246_begin_0, end = var_8246_end_0, end_mask = var_8246_end_mask_0, x = var_7922_cast_fp16)[name = tensor("op_8246_cast_fp16")]; tensor var_8253_begin_0 = const()[name = tensor("op_8253_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_8253_end_0 = const()[name = tensor("op_8253_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_8253_end_mask_0 = const()[name = tensor("op_8253_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8253_cast_fp16 = slice_by_index(begin = var_8253_begin_0, end = var_8253_end_0, end_mask = var_8253_end_mask_0, x = var_7922_cast_fp16)[name = tensor("op_8253_cast_fp16")]; tensor var_8260_begin_0 = const()[name = tensor("op_8260_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_8260_end_0 = const()[name = tensor("op_8260_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8260_end_mask_0 = const()[name = tensor("op_8260_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8260_cast_fp16 = slice_by_index(begin = var_8260_begin_0, end = var_8260_end_0, end_mask = var_8260_end_mask_0, x = var_7922_cast_fp16)[name = tensor("op_8260_cast_fp16")]; tensor k_17_perm_0 = const()[name = tensor("k_17_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_8265_begin_0 = const()[name = tensor("op_8265_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8265_end_0 = const()[name = tensor("op_8265_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_8265_end_mask_0 = const()[name = tensor("op_8265_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_3 = transpose(perm = k_17_perm_0, x = key_17_cast_fp16)[name = tensor("transpose_3")]; tensor var_8265_cast_fp16 = slice_by_index(begin = var_8265_begin_0, end = var_8265_end_0, end_mask = var_8265_end_mask_0, x = transpose_3)[name = tensor("op_8265_cast_fp16")]; tensor var_8269_begin_0 = const()[name = tensor("op_8269_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_8269_end_0 = const()[name = tensor("op_8269_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_8269_end_mask_0 = const()[name = tensor("op_8269_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8269_cast_fp16 = slice_by_index(begin = var_8269_begin_0, end = var_8269_end_0, end_mask = var_8269_end_mask_0, x = transpose_3)[name = tensor("op_8269_cast_fp16")]; tensor var_8273_begin_0 = const()[name = tensor("op_8273_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_8273_end_0 = const()[name = tensor("op_8273_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_8273_end_mask_0 = const()[name = tensor("op_8273_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8273_cast_fp16 = slice_by_index(begin = var_8273_begin_0, end = var_8273_end_0, end_mask = var_8273_end_mask_0, x = transpose_3)[name = tensor("op_8273_cast_fp16")]; tensor var_8277_begin_0 = const()[name = tensor("op_8277_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_8277_end_0 = const()[name = tensor("op_8277_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_8277_end_mask_0 = const()[name = tensor("op_8277_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8277_cast_fp16 = slice_by_index(begin = var_8277_begin_0, end = var_8277_end_0, end_mask = var_8277_end_mask_0, x = transpose_3)[name = tensor("op_8277_cast_fp16")]; tensor var_8281_begin_0 = const()[name = tensor("op_8281_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_8281_end_0 = const()[name = tensor("op_8281_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_8281_end_mask_0 = const()[name = tensor("op_8281_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8281_cast_fp16 = slice_by_index(begin = var_8281_begin_0, end = var_8281_end_0, end_mask = var_8281_end_mask_0, x = transpose_3)[name = tensor("op_8281_cast_fp16")]; tensor var_8285_begin_0 = const()[name = tensor("op_8285_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_8285_end_0 = const()[name = tensor("op_8285_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_8285_end_mask_0 = const()[name = tensor("op_8285_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8285_cast_fp16 = slice_by_index(begin = var_8285_begin_0, end = var_8285_end_0, end_mask = var_8285_end_mask_0, x = transpose_3)[name = tensor("op_8285_cast_fp16")]; tensor var_8289_begin_0 = const()[name = tensor("op_8289_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_8289_end_0 = const()[name = tensor("op_8289_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_8289_end_mask_0 = const()[name = tensor("op_8289_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8289_cast_fp16 = slice_by_index(begin = var_8289_begin_0, end = var_8289_end_0, end_mask = var_8289_end_mask_0, x = transpose_3)[name = tensor("op_8289_cast_fp16")]; tensor var_8293_begin_0 = const()[name = tensor("op_8293_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_8293_end_0 = const()[name = tensor("op_8293_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_8293_end_mask_0 = const()[name = tensor("op_8293_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8293_cast_fp16 = slice_by_index(begin = var_8293_begin_0, end = var_8293_end_0, end_mask = var_8293_end_mask_0, x = transpose_3)[name = tensor("op_8293_cast_fp16")]; tensor var_8297_begin_0 = const()[name = tensor("op_8297_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_8297_end_0 = const()[name = tensor("op_8297_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_8297_end_mask_0 = const()[name = tensor("op_8297_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8297_cast_fp16 = slice_by_index(begin = var_8297_begin_0, end = var_8297_end_0, end_mask = var_8297_end_mask_0, x = transpose_3)[name = tensor("op_8297_cast_fp16")]; tensor var_8301_begin_0 = const()[name = tensor("op_8301_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_8301_end_0 = const()[name = tensor("op_8301_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_8301_end_mask_0 = const()[name = tensor("op_8301_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8301_cast_fp16 = slice_by_index(begin = var_8301_begin_0, end = var_8301_end_0, end_mask = var_8301_end_mask_0, x = transpose_3)[name = tensor("op_8301_cast_fp16")]; tensor var_8305_begin_0 = const()[name = tensor("op_8305_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_8305_end_0 = const()[name = tensor("op_8305_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_8305_end_mask_0 = const()[name = tensor("op_8305_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8305_cast_fp16 = slice_by_index(begin = var_8305_begin_0, end = var_8305_end_0, end_mask = var_8305_end_mask_0, x = transpose_3)[name = tensor("op_8305_cast_fp16")]; tensor var_8309_begin_0 = const()[name = tensor("op_8309_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_8309_end_0 = const()[name = tensor("op_8309_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_8309_end_mask_0 = const()[name = tensor("op_8309_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_8309_cast_fp16 = slice_by_index(begin = var_8309_begin_0, end = var_8309_end_0, end_mask = var_8309_end_mask_0, x = transpose_3)[name = tensor("op_8309_cast_fp16")]; tensor var_8311_begin_0 = const()[name = tensor("op_8311_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8311_end_0 = const()[name = tensor("op_8311_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8311_end_mask_0 = const()[name = tensor("op_8311_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8311_cast_fp16 = slice_by_index(begin = var_8311_begin_0, end = var_8311_end_0, end_mask = var_8311_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8311_cast_fp16")]; tensor var_8315_begin_0 = const()[name = tensor("op_8315_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8315_end_0 = const()[name = tensor("op_8315_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_8315_end_mask_0 = const()[name = tensor("op_8315_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8315_cast_fp16 = slice_by_index(begin = var_8315_begin_0, end = var_8315_end_0, end_mask = var_8315_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8315_cast_fp16")]; tensor var_8319_begin_0 = const()[name = tensor("op_8319_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8319_end_0 = const()[name = tensor("op_8319_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_8319_end_mask_0 = const()[name = tensor("op_8319_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8319_cast_fp16 = slice_by_index(begin = var_8319_begin_0, end = var_8319_end_0, end_mask = var_8319_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8319_cast_fp16")]; tensor var_8323_begin_0 = const()[name = tensor("op_8323_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8323_end_0 = const()[name = tensor("op_8323_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_8323_end_mask_0 = const()[name = tensor("op_8323_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8323_cast_fp16 = slice_by_index(begin = var_8323_begin_0, end = var_8323_end_0, end_mask = var_8323_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8323_cast_fp16")]; tensor var_8327_begin_0 = const()[name = tensor("op_8327_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8327_end_0 = const()[name = tensor("op_8327_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_8327_end_mask_0 = const()[name = tensor("op_8327_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8327_cast_fp16 = slice_by_index(begin = var_8327_begin_0, end = var_8327_end_0, end_mask = var_8327_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8327_cast_fp16")]; tensor var_8331_begin_0 = const()[name = tensor("op_8331_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8331_end_0 = const()[name = tensor("op_8331_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_8331_end_mask_0 = const()[name = tensor("op_8331_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8331_cast_fp16 = slice_by_index(begin = var_8331_begin_0, end = var_8331_end_0, end_mask = var_8331_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8331_cast_fp16")]; tensor var_8335_begin_0 = const()[name = tensor("op_8335_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8335_end_0 = const()[name = tensor("op_8335_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_8335_end_mask_0 = const()[name = tensor("op_8335_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8335_cast_fp16 = slice_by_index(begin = var_8335_begin_0, end = var_8335_end_0, end_mask = var_8335_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8335_cast_fp16")]; tensor var_8339_begin_0 = const()[name = tensor("op_8339_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8339_end_0 = const()[name = tensor("op_8339_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_8339_end_mask_0 = const()[name = tensor("op_8339_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8339_cast_fp16")]; tensor var_8343_begin_0 = const()[name = tensor("op_8343_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8343_end_0 = const()[name = tensor("op_8343_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_8343_end_mask_0 = const()[name = tensor("op_8343_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8343_cast_fp16 = slice_by_index(begin = var_8343_begin_0, end = var_8343_end_0, end_mask = var_8343_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8343_cast_fp16")]; tensor var_8347_begin_0 = const()[name = tensor("op_8347_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8347_end_0 = const()[name = tensor("op_8347_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_8347_end_mask_0 = const()[name = tensor("op_8347_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8347_cast_fp16 = slice_by_index(begin = var_8347_begin_0, end = var_8347_end_0, end_mask = var_8347_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8347_cast_fp16")]; tensor var_8351_begin_0 = const()[name = tensor("op_8351_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8351_end_0 = const()[name = tensor("op_8351_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_8351_end_mask_0 = const()[name = tensor("op_8351_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8351_cast_fp16 = slice_by_index(begin = var_8351_begin_0, end = var_8351_end_0, end_mask = var_8351_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8351_cast_fp16")]; tensor var_8355_begin_0 = const()[name = tensor("op_8355_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8355_end_0 = const()[name = tensor("op_8355_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_8355_end_mask_0 = const()[name = tensor("op_8355_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8355_cast_fp16 = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = value_17_cast_fp16)[name = tensor("op_8355_cast_fp16")]; tensor var_8359_equation_0 = const()[name = tensor("op_8359_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8359_cast_fp16 = einsum(equation = var_8359_equation_0, values = (var_8265_cast_fp16, var_7931_cast_fp16))[name = tensor("op_8359_cast_fp16")]; tensor var_8360_to_fp16 = const()[name = tensor("op_8360_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_769_cast_fp16 = mul(x = var_8359_cast_fp16, y = var_8360_to_fp16)[name = tensor("aw_chunk_769_cast_fp16")]; tensor var_8363_equation_0 = const()[name = tensor("op_8363_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8363_cast_fp16 = einsum(equation = var_8363_equation_0, values = (var_8265_cast_fp16, var_7938_cast_fp16))[name = tensor("op_8363_cast_fp16")]; tensor var_8364_to_fp16 = const()[name = tensor("op_8364_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_771_cast_fp16 = mul(x = var_8363_cast_fp16, y = var_8364_to_fp16)[name = tensor("aw_chunk_771_cast_fp16")]; tensor var_8367_equation_0 = const()[name = tensor("op_8367_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8367_cast_fp16 = einsum(equation = var_8367_equation_0, values = (var_8265_cast_fp16, var_7945_cast_fp16))[name = tensor("op_8367_cast_fp16")]; tensor var_8368_to_fp16 = const()[name = tensor("op_8368_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_773_cast_fp16 = mul(x = var_8367_cast_fp16, y = var_8368_to_fp16)[name = tensor("aw_chunk_773_cast_fp16")]; tensor var_8371_equation_0 = const()[name = tensor("op_8371_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8371_cast_fp16 = einsum(equation = var_8371_equation_0, values = (var_8265_cast_fp16, var_7952_cast_fp16))[name = tensor("op_8371_cast_fp16")]; tensor var_8372_to_fp16 = const()[name = tensor("op_8372_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_775_cast_fp16 = mul(x = var_8371_cast_fp16, y = var_8372_to_fp16)[name = tensor("aw_chunk_775_cast_fp16")]; tensor var_8375_equation_0 = const()[name = tensor("op_8375_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8375_cast_fp16 = einsum(equation = var_8375_equation_0, values = (var_8269_cast_fp16, var_7959_cast_fp16))[name = tensor("op_8375_cast_fp16")]; tensor var_8376_to_fp16 = const()[name = tensor("op_8376_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_777_cast_fp16 = mul(x = var_8375_cast_fp16, y = var_8376_to_fp16)[name = tensor("aw_chunk_777_cast_fp16")]; tensor var_8379_equation_0 = const()[name = tensor("op_8379_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8379_cast_fp16 = einsum(equation = var_8379_equation_0, values = (var_8269_cast_fp16, var_7966_cast_fp16))[name = tensor("op_8379_cast_fp16")]; tensor var_8380_to_fp16 = const()[name = tensor("op_8380_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_779_cast_fp16 = mul(x = var_8379_cast_fp16, y = var_8380_to_fp16)[name = tensor("aw_chunk_779_cast_fp16")]; tensor var_8383_equation_0 = const()[name = tensor("op_8383_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8383_cast_fp16 = einsum(equation = var_8383_equation_0, values = (var_8269_cast_fp16, var_7973_cast_fp16))[name = tensor("op_8383_cast_fp16")]; tensor var_8384_to_fp16 = const()[name = tensor("op_8384_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_781_cast_fp16 = mul(x = var_8383_cast_fp16, y = var_8384_to_fp16)[name = tensor("aw_chunk_781_cast_fp16")]; tensor var_8387_equation_0 = const()[name = tensor("op_8387_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8387_cast_fp16 = einsum(equation = var_8387_equation_0, values = (var_8269_cast_fp16, var_7980_cast_fp16))[name = tensor("op_8387_cast_fp16")]; tensor var_8388_to_fp16 = const()[name = tensor("op_8388_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_783_cast_fp16 = mul(x = var_8387_cast_fp16, y = var_8388_to_fp16)[name = tensor("aw_chunk_783_cast_fp16")]; tensor var_8391_equation_0 = const()[name = tensor("op_8391_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8391_cast_fp16 = einsum(equation = var_8391_equation_0, values = (var_8273_cast_fp16, var_7987_cast_fp16))[name = tensor("op_8391_cast_fp16")]; tensor var_8392_to_fp16 = const()[name = tensor("op_8392_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_785_cast_fp16 = mul(x = var_8391_cast_fp16, y = var_8392_to_fp16)[name = tensor("aw_chunk_785_cast_fp16")]; tensor var_8395_equation_0 = const()[name = tensor("op_8395_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8395_cast_fp16 = einsum(equation = var_8395_equation_0, values = (var_8273_cast_fp16, var_7994_cast_fp16))[name = tensor("op_8395_cast_fp16")]; tensor var_8396_to_fp16 = const()[name = tensor("op_8396_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_787_cast_fp16 = mul(x = var_8395_cast_fp16, y = var_8396_to_fp16)[name = tensor("aw_chunk_787_cast_fp16")]; tensor var_8399_equation_0 = const()[name = tensor("op_8399_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8399_cast_fp16 = einsum(equation = var_8399_equation_0, values = (var_8273_cast_fp16, var_8001_cast_fp16))[name = tensor("op_8399_cast_fp16")]; tensor var_8400_to_fp16 = const()[name = tensor("op_8400_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_789_cast_fp16 = mul(x = var_8399_cast_fp16, y = var_8400_to_fp16)[name = tensor("aw_chunk_789_cast_fp16")]; tensor var_8403_equation_0 = const()[name = tensor("op_8403_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8403_cast_fp16 = einsum(equation = var_8403_equation_0, values = (var_8273_cast_fp16, var_8008_cast_fp16))[name = tensor("op_8403_cast_fp16")]; tensor var_8404_to_fp16 = const()[name = tensor("op_8404_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_791_cast_fp16 = mul(x = var_8403_cast_fp16, y = var_8404_to_fp16)[name = tensor("aw_chunk_791_cast_fp16")]; tensor var_8407_equation_0 = const()[name = tensor("op_8407_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8407_cast_fp16 = einsum(equation = var_8407_equation_0, values = (var_8277_cast_fp16, var_8015_cast_fp16))[name = tensor("op_8407_cast_fp16")]; tensor var_8408_to_fp16 = const()[name = tensor("op_8408_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_793_cast_fp16 = mul(x = var_8407_cast_fp16, y = var_8408_to_fp16)[name = tensor("aw_chunk_793_cast_fp16")]; tensor var_8411_equation_0 = const()[name = tensor("op_8411_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8411_cast_fp16 = einsum(equation = var_8411_equation_0, values = (var_8277_cast_fp16, var_8022_cast_fp16))[name = tensor("op_8411_cast_fp16")]; tensor var_8412_to_fp16 = const()[name = tensor("op_8412_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_795_cast_fp16 = mul(x = var_8411_cast_fp16, y = var_8412_to_fp16)[name = tensor("aw_chunk_795_cast_fp16")]; tensor var_8415_equation_0 = const()[name = tensor("op_8415_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8415_cast_fp16 = einsum(equation = var_8415_equation_0, values = (var_8277_cast_fp16, var_8029_cast_fp16))[name = tensor("op_8415_cast_fp16")]; tensor var_8416_to_fp16 = const()[name = tensor("op_8416_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_797_cast_fp16 = mul(x = var_8415_cast_fp16, y = var_8416_to_fp16)[name = tensor("aw_chunk_797_cast_fp16")]; tensor var_8419_equation_0 = const()[name = tensor("op_8419_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8419_cast_fp16 = einsum(equation = var_8419_equation_0, values = (var_8277_cast_fp16, var_8036_cast_fp16))[name = tensor("op_8419_cast_fp16")]; tensor var_8420_to_fp16 = const()[name = tensor("op_8420_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_799_cast_fp16 = mul(x = var_8419_cast_fp16, y = var_8420_to_fp16)[name = tensor("aw_chunk_799_cast_fp16")]; tensor var_8423_equation_0 = const()[name = tensor("op_8423_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8423_cast_fp16 = einsum(equation = var_8423_equation_0, values = (var_8281_cast_fp16, var_8043_cast_fp16))[name = tensor("op_8423_cast_fp16")]; tensor var_8424_to_fp16 = const()[name = tensor("op_8424_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_801_cast_fp16 = mul(x = var_8423_cast_fp16, y = var_8424_to_fp16)[name = tensor("aw_chunk_801_cast_fp16")]; tensor var_8427_equation_0 = const()[name = tensor("op_8427_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8427_cast_fp16 = einsum(equation = var_8427_equation_0, values = (var_8281_cast_fp16, var_8050_cast_fp16))[name = tensor("op_8427_cast_fp16")]; tensor var_8428_to_fp16 = const()[name = tensor("op_8428_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_803_cast_fp16 = mul(x = var_8427_cast_fp16, y = var_8428_to_fp16)[name = tensor("aw_chunk_803_cast_fp16")]; tensor var_8431_equation_0 = const()[name = tensor("op_8431_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8431_cast_fp16 = einsum(equation = var_8431_equation_0, values = (var_8281_cast_fp16, var_8057_cast_fp16))[name = tensor("op_8431_cast_fp16")]; tensor var_8432_to_fp16 = const()[name = tensor("op_8432_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_805_cast_fp16 = mul(x = var_8431_cast_fp16, y = var_8432_to_fp16)[name = tensor("aw_chunk_805_cast_fp16")]; tensor var_8435_equation_0 = const()[name = tensor("op_8435_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8435_cast_fp16 = einsum(equation = var_8435_equation_0, values = (var_8281_cast_fp16, var_8064_cast_fp16))[name = tensor("op_8435_cast_fp16")]; tensor var_8436_to_fp16 = const()[name = tensor("op_8436_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_807_cast_fp16 = mul(x = var_8435_cast_fp16, y = var_8436_to_fp16)[name = tensor("aw_chunk_807_cast_fp16")]; tensor var_8439_equation_0 = const()[name = tensor("op_8439_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8439_cast_fp16 = einsum(equation = var_8439_equation_0, values = (var_8285_cast_fp16, var_8071_cast_fp16))[name = tensor("op_8439_cast_fp16")]; tensor var_8440_to_fp16 = const()[name = tensor("op_8440_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_809_cast_fp16 = mul(x = var_8439_cast_fp16, y = var_8440_to_fp16)[name = tensor("aw_chunk_809_cast_fp16")]; tensor var_8443_equation_0 = const()[name = tensor("op_8443_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8443_cast_fp16 = einsum(equation = var_8443_equation_0, values = (var_8285_cast_fp16, var_8078_cast_fp16))[name = tensor("op_8443_cast_fp16")]; tensor var_8444_to_fp16 = const()[name = tensor("op_8444_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_811_cast_fp16 = mul(x = var_8443_cast_fp16, y = var_8444_to_fp16)[name = tensor("aw_chunk_811_cast_fp16")]; tensor var_8447_equation_0 = const()[name = tensor("op_8447_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8447_cast_fp16 = einsum(equation = var_8447_equation_0, values = (var_8285_cast_fp16, var_8085_cast_fp16))[name = tensor("op_8447_cast_fp16")]; tensor var_8448_to_fp16 = const()[name = tensor("op_8448_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_813_cast_fp16 = mul(x = var_8447_cast_fp16, y = var_8448_to_fp16)[name = tensor("aw_chunk_813_cast_fp16")]; tensor var_8451_equation_0 = const()[name = tensor("op_8451_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8451_cast_fp16 = einsum(equation = var_8451_equation_0, values = (var_8285_cast_fp16, var_8092_cast_fp16))[name = tensor("op_8451_cast_fp16")]; tensor var_8452_to_fp16 = const()[name = tensor("op_8452_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_815_cast_fp16 = mul(x = var_8451_cast_fp16, y = var_8452_to_fp16)[name = tensor("aw_chunk_815_cast_fp16")]; tensor var_8455_equation_0 = const()[name = tensor("op_8455_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8455_cast_fp16 = einsum(equation = var_8455_equation_0, values = (var_8289_cast_fp16, var_8099_cast_fp16))[name = tensor("op_8455_cast_fp16")]; tensor var_8456_to_fp16 = const()[name = tensor("op_8456_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_817_cast_fp16 = mul(x = var_8455_cast_fp16, y = var_8456_to_fp16)[name = tensor("aw_chunk_817_cast_fp16")]; tensor var_8459_equation_0 = const()[name = tensor("op_8459_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8459_cast_fp16 = einsum(equation = var_8459_equation_0, values = (var_8289_cast_fp16, var_8106_cast_fp16))[name = tensor("op_8459_cast_fp16")]; tensor var_8460_to_fp16 = const()[name = tensor("op_8460_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_819_cast_fp16 = mul(x = var_8459_cast_fp16, y = var_8460_to_fp16)[name = tensor("aw_chunk_819_cast_fp16")]; tensor var_8463_equation_0 = const()[name = tensor("op_8463_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8463_cast_fp16 = einsum(equation = var_8463_equation_0, values = (var_8289_cast_fp16, var_8113_cast_fp16))[name = tensor("op_8463_cast_fp16")]; tensor var_8464_to_fp16 = const()[name = tensor("op_8464_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_821_cast_fp16 = mul(x = var_8463_cast_fp16, y = var_8464_to_fp16)[name = tensor("aw_chunk_821_cast_fp16")]; tensor var_8467_equation_0 = const()[name = tensor("op_8467_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8467_cast_fp16 = einsum(equation = var_8467_equation_0, values = (var_8289_cast_fp16, var_8120_cast_fp16))[name = tensor("op_8467_cast_fp16")]; tensor var_8468_to_fp16 = const()[name = tensor("op_8468_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_823_cast_fp16 = mul(x = var_8467_cast_fp16, y = var_8468_to_fp16)[name = tensor("aw_chunk_823_cast_fp16")]; tensor var_8471_equation_0 = const()[name = tensor("op_8471_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8471_cast_fp16 = einsum(equation = var_8471_equation_0, values = (var_8293_cast_fp16, var_8127_cast_fp16))[name = tensor("op_8471_cast_fp16")]; tensor var_8472_to_fp16 = const()[name = tensor("op_8472_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_825_cast_fp16 = mul(x = var_8471_cast_fp16, y = var_8472_to_fp16)[name = tensor("aw_chunk_825_cast_fp16")]; tensor var_8475_equation_0 = const()[name = tensor("op_8475_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8475_cast_fp16 = einsum(equation = var_8475_equation_0, values = (var_8293_cast_fp16, var_8134_cast_fp16))[name = tensor("op_8475_cast_fp16")]; tensor var_8476_to_fp16 = const()[name = tensor("op_8476_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_827_cast_fp16 = mul(x = var_8475_cast_fp16, y = var_8476_to_fp16)[name = tensor("aw_chunk_827_cast_fp16")]; tensor var_8479_equation_0 = const()[name = tensor("op_8479_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8479_cast_fp16 = einsum(equation = var_8479_equation_0, values = (var_8293_cast_fp16, var_8141_cast_fp16))[name = tensor("op_8479_cast_fp16")]; tensor var_8480_to_fp16 = const()[name = tensor("op_8480_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_829_cast_fp16 = mul(x = var_8479_cast_fp16, y = var_8480_to_fp16)[name = tensor("aw_chunk_829_cast_fp16")]; tensor var_8483_equation_0 = const()[name = tensor("op_8483_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8483_cast_fp16 = einsum(equation = var_8483_equation_0, values = (var_8293_cast_fp16, var_8148_cast_fp16))[name = tensor("op_8483_cast_fp16")]; tensor var_8484_to_fp16 = const()[name = tensor("op_8484_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_831_cast_fp16 = mul(x = var_8483_cast_fp16, y = var_8484_to_fp16)[name = tensor("aw_chunk_831_cast_fp16")]; tensor var_8487_equation_0 = const()[name = tensor("op_8487_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8487_cast_fp16 = einsum(equation = var_8487_equation_0, values = (var_8297_cast_fp16, var_8155_cast_fp16))[name = tensor("op_8487_cast_fp16")]; tensor var_8488_to_fp16 = const()[name = tensor("op_8488_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_833_cast_fp16 = mul(x = var_8487_cast_fp16, y = var_8488_to_fp16)[name = tensor("aw_chunk_833_cast_fp16")]; tensor var_8491_equation_0 = const()[name = tensor("op_8491_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8491_cast_fp16 = einsum(equation = var_8491_equation_0, values = (var_8297_cast_fp16, var_8162_cast_fp16))[name = tensor("op_8491_cast_fp16")]; tensor var_8492_to_fp16 = const()[name = tensor("op_8492_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_835_cast_fp16 = mul(x = var_8491_cast_fp16, y = var_8492_to_fp16)[name = tensor("aw_chunk_835_cast_fp16")]; tensor var_8495_equation_0 = const()[name = tensor("op_8495_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8495_cast_fp16 = einsum(equation = var_8495_equation_0, values = (var_8297_cast_fp16, var_8169_cast_fp16))[name = tensor("op_8495_cast_fp16")]; tensor var_8496_to_fp16 = const()[name = tensor("op_8496_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_837_cast_fp16 = mul(x = var_8495_cast_fp16, y = var_8496_to_fp16)[name = tensor("aw_chunk_837_cast_fp16")]; tensor var_8499_equation_0 = const()[name = tensor("op_8499_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8499_cast_fp16 = einsum(equation = var_8499_equation_0, values = (var_8297_cast_fp16, var_8176_cast_fp16))[name = tensor("op_8499_cast_fp16")]; tensor var_8500_to_fp16 = const()[name = tensor("op_8500_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_839_cast_fp16 = mul(x = var_8499_cast_fp16, y = var_8500_to_fp16)[name = tensor("aw_chunk_839_cast_fp16")]; tensor var_8503_equation_0 = const()[name = tensor("op_8503_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8503_cast_fp16 = einsum(equation = var_8503_equation_0, values = (var_8301_cast_fp16, var_8183_cast_fp16))[name = tensor("op_8503_cast_fp16")]; tensor var_8504_to_fp16 = const()[name = tensor("op_8504_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_841_cast_fp16 = mul(x = var_8503_cast_fp16, y = var_8504_to_fp16)[name = tensor("aw_chunk_841_cast_fp16")]; tensor var_8507_equation_0 = const()[name = tensor("op_8507_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8507_cast_fp16 = einsum(equation = var_8507_equation_0, values = (var_8301_cast_fp16, var_8190_cast_fp16))[name = tensor("op_8507_cast_fp16")]; tensor var_8508_to_fp16 = const()[name = tensor("op_8508_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_843_cast_fp16 = mul(x = var_8507_cast_fp16, y = var_8508_to_fp16)[name = tensor("aw_chunk_843_cast_fp16")]; tensor var_8511_equation_0 = const()[name = tensor("op_8511_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8511_cast_fp16 = einsum(equation = var_8511_equation_0, values = (var_8301_cast_fp16, var_8197_cast_fp16))[name = tensor("op_8511_cast_fp16")]; tensor var_8512_to_fp16 = const()[name = tensor("op_8512_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_845_cast_fp16 = mul(x = var_8511_cast_fp16, y = var_8512_to_fp16)[name = tensor("aw_chunk_845_cast_fp16")]; tensor var_8515_equation_0 = const()[name = tensor("op_8515_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8515_cast_fp16 = einsum(equation = var_8515_equation_0, values = (var_8301_cast_fp16, var_8204_cast_fp16))[name = tensor("op_8515_cast_fp16")]; tensor var_8516_to_fp16 = const()[name = tensor("op_8516_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_847_cast_fp16 = mul(x = var_8515_cast_fp16, y = var_8516_to_fp16)[name = tensor("aw_chunk_847_cast_fp16")]; tensor var_8519_equation_0 = const()[name = tensor("op_8519_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8519_cast_fp16 = einsum(equation = var_8519_equation_0, values = (var_8305_cast_fp16, var_8211_cast_fp16))[name = tensor("op_8519_cast_fp16")]; tensor var_8520_to_fp16 = const()[name = tensor("op_8520_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_849_cast_fp16 = mul(x = var_8519_cast_fp16, y = var_8520_to_fp16)[name = tensor("aw_chunk_849_cast_fp16")]; tensor var_8523_equation_0 = const()[name = tensor("op_8523_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8523_cast_fp16 = einsum(equation = var_8523_equation_0, values = (var_8305_cast_fp16, var_8218_cast_fp16))[name = tensor("op_8523_cast_fp16")]; tensor var_8524_to_fp16 = const()[name = tensor("op_8524_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_851_cast_fp16 = mul(x = var_8523_cast_fp16, y = var_8524_to_fp16)[name = tensor("aw_chunk_851_cast_fp16")]; tensor var_8527_equation_0 = const()[name = tensor("op_8527_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8527_cast_fp16 = einsum(equation = var_8527_equation_0, values = (var_8305_cast_fp16, var_8225_cast_fp16))[name = tensor("op_8527_cast_fp16")]; tensor var_8528_to_fp16 = const()[name = tensor("op_8528_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_853_cast_fp16 = mul(x = var_8527_cast_fp16, y = var_8528_to_fp16)[name = tensor("aw_chunk_853_cast_fp16")]; tensor var_8531_equation_0 = const()[name = tensor("op_8531_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8531_cast_fp16 = einsum(equation = var_8531_equation_0, values = (var_8305_cast_fp16, var_8232_cast_fp16))[name = tensor("op_8531_cast_fp16")]; tensor var_8532_to_fp16 = const()[name = tensor("op_8532_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_855_cast_fp16 = mul(x = var_8531_cast_fp16, y = var_8532_to_fp16)[name = tensor("aw_chunk_855_cast_fp16")]; tensor var_8535_equation_0 = const()[name = tensor("op_8535_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8535_cast_fp16 = einsum(equation = var_8535_equation_0, values = (var_8309_cast_fp16, var_8239_cast_fp16))[name = tensor("op_8535_cast_fp16")]; tensor var_8536_to_fp16 = const()[name = tensor("op_8536_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_857_cast_fp16 = mul(x = var_8535_cast_fp16, y = var_8536_to_fp16)[name = tensor("aw_chunk_857_cast_fp16")]; tensor var_8539_equation_0 = const()[name = tensor("op_8539_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8539_cast_fp16 = einsum(equation = var_8539_equation_0, values = (var_8309_cast_fp16, var_8246_cast_fp16))[name = tensor("op_8539_cast_fp16")]; tensor var_8540_to_fp16 = const()[name = tensor("op_8540_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_859_cast_fp16 = mul(x = var_8539_cast_fp16, y = var_8540_to_fp16)[name = tensor("aw_chunk_859_cast_fp16")]; tensor var_8543_equation_0 = const()[name = tensor("op_8543_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8543_cast_fp16 = einsum(equation = var_8543_equation_0, values = (var_8309_cast_fp16, var_8253_cast_fp16))[name = tensor("op_8543_cast_fp16")]; tensor var_8544_to_fp16 = const()[name = tensor("op_8544_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_861_cast_fp16 = mul(x = var_8543_cast_fp16, y = var_8544_to_fp16)[name = tensor("aw_chunk_861_cast_fp16")]; tensor var_8547_equation_0 = const()[name = tensor("op_8547_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_8547_cast_fp16 = einsum(equation = var_8547_equation_0, values = (var_8309_cast_fp16, var_8260_cast_fp16))[name = tensor("op_8547_cast_fp16")]; tensor var_8548_to_fp16 = const()[name = tensor("op_8548_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_863_cast_fp16 = mul(x = var_8547_cast_fp16, y = var_8548_to_fp16)[name = tensor("aw_chunk_863_cast_fp16")]; tensor var_8550_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_769_cast_fp16)[name = tensor("op_8550_cast_fp16")]; tensor var_8551_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_771_cast_fp16)[name = tensor("op_8551_cast_fp16")]; tensor var_8552_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_773_cast_fp16)[name = tensor("op_8552_cast_fp16")]; tensor var_8553_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_775_cast_fp16)[name = tensor("op_8553_cast_fp16")]; tensor var_8554_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_777_cast_fp16)[name = tensor("op_8554_cast_fp16")]; tensor var_8555_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_779_cast_fp16)[name = tensor("op_8555_cast_fp16")]; tensor var_8556_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_781_cast_fp16)[name = tensor("op_8556_cast_fp16")]; tensor var_8557_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_783_cast_fp16)[name = tensor("op_8557_cast_fp16")]; tensor var_8558_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_785_cast_fp16)[name = tensor("op_8558_cast_fp16")]; tensor var_8559_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_787_cast_fp16)[name = tensor("op_8559_cast_fp16")]; tensor var_8560_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_789_cast_fp16)[name = tensor("op_8560_cast_fp16")]; tensor var_8561_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_791_cast_fp16)[name = tensor("op_8561_cast_fp16")]; tensor var_8562_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_793_cast_fp16)[name = tensor("op_8562_cast_fp16")]; tensor var_8563_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_795_cast_fp16)[name = tensor("op_8563_cast_fp16")]; tensor var_8564_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_797_cast_fp16)[name = tensor("op_8564_cast_fp16")]; tensor var_8565_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_799_cast_fp16)[name = tensor("op_8565_cast_fp16")]; tensor var_8566_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_801_cast_fp16)[name = tensor("op_8566_cast_fp16")]; tensor var_8567_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_803_cast_fp16)[name = tensor("op_8567_cast_fp16")]; tensor var_8568_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_805_cast_fp16)[name = tensor("op_8568_cast_fp16")]; tensor var_8569_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_807_cast_fp16)[name = tensor("op_8569_cast_fp16")]; tensor var_8570_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_809_cast_fp16)[name = tensor("op_8570_cast_fp16")]; tensor var_8571_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_811_cast_fp16)[name = tensor("op_8571_cast_fp16")]; tensor var_8572_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_813_cast_fp16)[name = tensor("op_8572_cast_fp16")]; tensor var_8573_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_815_cast_fp16)[name = tensor("op_8573_cast_fp16")]; tensor var_8574_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_817_cast_fp16)[name = tensor("op_8574_cast_fp16")]; tensor var_8575_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_819_cast_fp16)[name = tensor("op_8575_cast_fp16")]; tensor var_8576_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_821_cast_fp16)[name = tensor("op_8576_cast_fp16")]; tensor var_8577_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_823_cast_fp16)[name = tensor("op_8577_cast_fp16")]; tensor var_8578_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_825_cast_fp16)[name = tensor("op_8578_cast_fp16")]; tensor var_8579_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_827_cast_fp16)[name = tensor("op_8579_cast_fp16")]; tensor var_8580_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_829_cast_fp16)[name = tensor("op_8580_cast_fp16")]; tensor var_8581_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_831_cast_fp16)[name = tensor("op_8581_cast_fp16")]; tensor var_8582_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_833_cast_fp16)[name = tensor("op_8582_cast_fp16")]; tensor var_8583_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_835_cast_fp16)[name = tensor("op_8583_cast_fp16")]; tensor var_8584_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_837_cast_fp16)[name = tensor("op_8584_cast_fp16")]; tensor var_8585_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_839_cast_fp16)[name = tensor("op_8585_cast_fp16")]; tensor var_8586_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_841_cast_fp16)[name = tensor("op_8586_cast_fp16")]; tensor var_8587_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_843_cast_fp16)[name = tensor("op_8587_cast_fp16")]; tensor var_8588_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_845_cast_fp16)[name = tensor("op_8588_cast_fp16")]; tensor var_8589_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_847_cast_fp16)[name = tensor("op_8589_cast_fp16")]; tensor var_8590_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_849_cast_fp16)[name = tensor("op_8590_cast_fp16")]; tensor var_8591_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_851_cast_fp16)[name = tensor("op_8591_cast_fp16")]; tensor var_8592_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_853_cast_fp16)[name = tensor("op_8592_cast_fp16")]; tensor var_8593_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_855_cast_fp16)[name = tensor("op_8593_cast_fp16")]; tensor var_8594_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_857_cast_fp16)[name = tensor("op_8594_cast_fp16")]; tensor var_8595_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_859_cast_fp16)[name = tensor("op_8595_cast_fp16")]; tensor var_8596_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_861_cast_fp16)[name = tensor("op_8596_cast_fp16")]; tensor var_8597_cast_fp16 = softmax(axis = var_7823, x = aw_chunk_863_cast_fp16)[name = tensor("op_8597_cast_fp16")]; tensor var_8599_equation_0 = const()[name = tensor("op_8599_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8599_cast_fp16 = einsum(equation = var_8599_equation_0, values = (var_8311_cast_fp16, var_8550_cast_fp16))[name = tensor("op_8599_cast_fp16")]; tensor var_8601_equation_0 = const()[name = tensor("op_8601_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8601_cast_fp16 = einsum(equation = var_8601_equation_0, values = (var_8311_cast_fp16, var_8551_cast_fp16))[name = tensor("op_8601_cast_fp16")]; tensor var_8603_equation_0 = const()[name = tensor("op_8603_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8603_cast_fp16 = einsum(equation = var_8603_equation_0, values = (var_8311_cast_fp16, var_8552_cast_fp16))[name = tensor("op_8603_cast_fp16")]; tensor var_8605_equation_0 = const()[name = tensor("op_8605_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8605_cast_fp16 = einsum(equation = var_8605_equation_0, values = (var_8311_cast_fp16, var_8553_cast_fp16))[name = tensor("op_8605_cast_fp16")]; tensor var_8607_equation_0 = const()[name = tensor("op_8607_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8607_cast_fp16 = einsum(equation = var_8607_equation_0, values = (var_8315_cast_fp16, var_8554_cast_fp16))[name = tensor("op_8607_cast_fp16")]; tensor var_8609_equation_0 = const()[name = tensor("op_8609_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8609_cast_fp16 = einsum(equation = var_8609_equation_0, values = (var_8315_cast_fp16, var_8555_cast_fp16))[name = tensor("op_8609_cast_fp16")]; tensor var_8611_equation_0 = const()[name = tensor("op_8611_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8611_cast_fp16 = einsum(equation = var_8611_equation_0, values = (var_8315_cast_fp16, var_8556_cast_fp16))[name = tensor("op_8611_cast_fp16")]; tensor var_8613_equation_0 = const()[name = tensor("op_8613_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8613_cast_fp16 = einsum(equation = var_8613_equation_0, values = (var_8315_cast_fp16, var_8557_cast_fp16))[name = tensor("op_8613_cast_fp16")]; tensor var_8615_equation_0 = const()[name = tensor("op_8615_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8615_cast_fp16 = einsum(equation = var_8615_equation_0, values = (var_8319_cast_fp16, var_8558_cast_fp16))[name = tensor("op_8615_cast_fp16")]; tensor var_8617_equation_0 = const()[name = tensor("op_8617_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8617_cast_fp16 = einsum(equation = var_8617_equation_0, values = (var_8319_cast_fp16, var_8559_cast_fp16))[name = tensor("op_8617_cast_fp16")]; tensor var_8619_equation_0 = const()[name = tensor("op_8619_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8619_cast_fp16 = einsum(equation = var_8619_equation_0, values = (var_8319_cast_fp16, var_8560_cast_fp16))[name = tensor("op_8619_cast_fp16")]; tensor var_8621_equation_0 = const()[name = tensor("op_8621_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8621_cast_fp16 = einsum(equation = var_8621_equation_0, values = (var_8319_cast_fp16, var_8561_cast_fp16))[name = tensor("op_8621_cast_fp16")]; tensor var_8623_equation_0 = const()[name = tensor("op_8623_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8623_cast_fp16 = einsum(equation = var_8623_equation_0, values = (var_8323_cast_fp16, var_8562_cast_fp16))[name = tensor("op_8623_cast_fp16")]; tensor var_8625_equation_0 = const()[name = tensor("op_8625_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8625_cast_fp16 = einsum(equation = var_8625_equation_0, values = (var_8323_cast_fp16, var_8563_cast_fp16))[name = tensor("op_8625_cast_fp16")]; tensor var_8627_equation_0 = const()[name = tensor("op_8627_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8627_cast_fp16 = einsum(equation = var_8627_equation_0, values = (var_8323_cast_fp16, var_8564_cast_fp16))[name = tensor("op_8627_cast_fp16")]; tensor var_8629_equation_0 = const()[name = tensor("op_8629_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8629_cast_fp16 = einsum(equation = var_8629_equation_0, values = (var_8323_cast_fp16, var_8565_cast_fp16))[name = tensor("op_8629_cast_fp16")]; tensor var_8631_equation_0 = const()[name = tensor("op_8631_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8631_cast_fp16 = einsum(equation = var_8631_equation_0, values = (var_8327_cast_fp16, var_8566_cast_fp16))[name = tensor("op_8631_cast_fp16")]; tensor var_8633_equation_0 = const()[name = tensor("op_8633_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8633_cast_fp16 = einsum(equation = var_8633_equation_0, values = (var_8327_cast_fp16, var_8567_cast_fp16))[name = tensor("op_8633_cast_fp16")]; tensor var_8635_equation_0 = const()[name = tensor("op_8635_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8635_cast_fp16 = einsum(equation = var_8635_equation_0, values = (var_8327_cast_fp16, var_8568_cast_fp16))[name = tensor("op_8635_cast_fp16")]; tensor var_8637_equation_0 = const()[name = tensor("op_8637_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8637_cast_fp16 = einsum(equation = var_8637_equation_0, values = (var_8327_cast_fp16, var_8569_cast_fp16))[name = tensor("op_8637_cast_fp16")]; tensor var_8639_equation_0 = const()[name = tensor("op_8639_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8639_cast_fp16 = einsum(equation = var_8639_equation_0, values = (var_8331_cast_fp16, var_8570_cast_fp16))[name = tensor("op_8639_cast_fp16")]; tensor var_8641_equation_0 = const()[name = tensor("op_8641_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8641_cast_fp16 = einsum(equation = var_8641_equation_0, values = (var_8331_cast_fp16, var_8571_cast_fp16))[name = tensor("op_8641_cast_fp16")]; tensor var_8643_equation_0 = const()[name = tensor("op_8643_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8643_cast_fp16 = einsum(equation = var_8643_equation_0, values = (var_8331_cast_fp16, var_8572_cast_fp16))[name = tensor("op_8643_cast_fp16")]; tensor var_8645_equation_0 = const()[name = tensor("op_8645_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8645_cast_fp16 = einsum(equation = var_8645_equation_0, values = (var_8331_cast_fp16, var_8573_cast_fp16))[name = tensor("op_8645_cast_fp16")]; tensor var_8647_equation_0 = const()[name = tensor("op_8647_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8647_cast_fp16 = einsum(equation = var_8647_equation_0, values = (var_8335_cast_fp16, var_8574_cast_fp16))[name = tensor("op_8647_cast_fp16")]; tensor var_8649_equation_0 = const()[name = tensor("op_8649_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8649_cast_fp16 = einsum(equation = var_8649_equation_0, values = (var_8335_cast_fp16, var_8575_cast_fp16))[name = tensor("op_8649_cast_fp16")]; tensor var_8651_equation_0 = const()[name = tensor("op_8651_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8651_cast_fp16 = einsum(equation = var_8651_equation_0, values = (var_8335_cast_fp16, var_8576_cast_fp16))[name = tensor("op_8651_cast_fp16")]; tensor var_8653_equation_0 = const()[name = tensor("op_8653_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8653_cast_fp16 = einsum(equation = var_8653_equation_0, values = (var_8335_cast_fp16, var_8577_cast_fp16))[name = tensor("op_8653_cast_fp16")]; tensor var_8655_equation_0 = const()[name = tensor("op_8655_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8655_cast_fp16 = einsum(equation = var_8655_equation_0, values = (var_8339_cast_fp16, var_8578_cast_fp16))[name = tensor("op_8655_cast_fp16")]; tensor var_8657_equation_0 = const()[name = tensor("op_8657_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8657_cast_fp16 = einsum(equation = var_8657_equation_0, values = (var_8339_cast_fp16, var_8579_cast_fp16))[name = tensor("op_8657_cast_fp16")]; tensor var_8659_equation_0 = const()[name = tensor("op_8659_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8659_cast_fp16 = einsum(equation = var_8659_equation_0, values = (var_8339_cast_fp16, var_8580_cast_fp16))[name = tensor("op_8659_cast_fp16")]; tensor var_8661_equation_0 = const()[name = tensor("op_8661_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8661_cast_fp16 = einsum(equation = var_8661_equation_0, values = (var_8339_cast_fp16, var_8581_cast_fp16))[name = tensor("op_8661_cast_fp16")]; tensor var_8663_equation_0 = const()[name = tensor("op_8663_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8663_cast_fp16 = einsum(equation = var_8663_equation_0, values = (var_8343_cast_fp16, var_8582_cast_fp16))[name = tensor("op_8663_cast_fp16")]; tensor var_8665_equation_0 = const()[name = tensor("op_8665_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8665_cast_fp16 = einsum(equation = var_8665_equation_0, values = (var_8343_cast_fp16, var_8583_cast_fp16))[name = tensor("op_8665_cast_fp16")]; tensor var_8667_equation_0 = const()[name = tensor("op_8667_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8667_cast_fp16 = einsum(equation = var_8667_equation_0, values = (var_8343_cast_fp16, var_8584_cast_fp16))[name = tensor("op_8667_cast_fp16")]; tensor var_8669_equation_0 = const()[name = tensor("op_8669_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8669_cast_fp16 = einsum(equation = var_8669_equation_0, values = (var_8343_cast_fp16, var_8585_cast_fp16))[name = tensor("op_8669_cast_fp16")]; tensor var_8671_equation_0 = const()[name = tensor("op_8671_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8671_cast_fp16 = einsum(equation = var_8671_equation_0, values = (var_8347_cast_fp16, var_8586_cast_fp16))[name = tensor("op_8671_cast_fp16")]; tensor var_8673_equation_0 = const()[name = tensor("op_8673_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8673_cast_fp16 = einsum(equation = var_8673_equation_0, values = (var_8347_cast_fp16, var_8587_cast_fp16))[name = tensor("op_8673_cast_fp16")]; tensor var_8675_equation_0 = const()[name = tensor("op_8675_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8675_cast_fp16 = einsum(equation = var_8675_equation_0, values = (var_8347_cast_fp16, var_8588_cast_fp16))[name = tensor("op_8675_cast_fp16")]; tensor var_8677_equation_0 = const()[name = tensor("op_8677_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8677_cast_fp16 = einsum(equation = var_8677_equation_0, values = (var_8347_cast_fp16, var_8589_cast_fp16))[name = tensor("op_8677_cast_fp16")]; tensor var_8679_equation_0 = const()[name = tensor("op_8679_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8679_cast_fp16 = einsum(equation = var_8679_equation_0, values = (var_8351_cast_fp16, var_8590_cast_fp16))[name = tensor("op_8679_cast_fp16")]; tensor var_8681_equation_0 = const()[name = tensor("op_8681_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8681_cast_fp16 = einsum(equation = var_8681_equation_0, values = (var_8351_cast_fp16, var_8591_cast_fp16))[name = tensor("op_8681_cast_fp16")]; tensor var_8683_equation_0 = const()[name = tensor("op_8683_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8683_cast_fp16 = einsum(equation = var_8683_equation_0, values = (var_8351_cast_fp16, var_8592_cast_fp16))[name = tensor("op_8683_cast_fp16")]; tensor var_8685_equation_0 = const()[name = tensor("op_8685_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8685_cast_fp16 = einsum(equation = var_8685_equation_0, values = (var_8351_cast_fp16, var_8593_cast_fp16))[name = tensor("op_8685_cast_fp16")]; tensor var_8687_equation_0 = const()[name = tensor("op_8687_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8687_cast_fp16 = einsum(equation = var_8687_equation_0, values = (var_8355_cast_fp16, var_8594_cast_fp16))[name = tensor("op_8687_cast_fp16")]; tensor var_8689_equation_0 = const()[name = tensor("op_8689_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8689_cast_fp16 = einsum(equation = var_8689_equation_0, values = (var_8355_cast_fp16, var_8595_cast_fp16))[name = tensor("op_8689_cast_fp16")]; tensor var_8691_equation_0 = const()[name = tensor("op_8691_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8691_cast_fp16 = einsum(equation = var_8691_equation_0, values = (var_8355_cast_fp16, var_8596_cast_fp16))[name = tensor("op_8691_cast_fp16")]; tensor var_8693_equation_0 = const()[name = tensor("op_8693_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_8693_cast_fp16 = einsum(equation = var_8693_equation_0, values = (var_8355_cast_fp16, var_8597_cast_fp16))[name = tensor("op_8693_cast_fp16")]; tensor var_8695_interleave_0 = const()[name = tensor("op_8695_interleave_0"), val = tensor(false)]; tensor var_8695_cast_fp16 = concat(axis = var_7806, interleave = var_8695_interleave_0, values = (var_8599_cast_fp16, var_8601_cast_fp16, var_8603_cast_fp16, var_8605_cast_fp16))[name = tensor("op_8695_cast_fp16")]; tensor var_8697_interleave_0 = const()[name = tensor("op_8697_interleave_0"), val = tensor(false)]; tensor var_8697_cast_fp16 = concat(axis = var_7806, interleave = var_8697_interleave_0, values = (var_8607_cast_fp16, var_8609_cast_fp16, var_8611_cast_fp16, var_8613_cast_fp16))[name = tensor("op_8697_cast_fp16")]; tensor var_8699_interleave_0 = const()[name = tensor("op_8699_interleave_0"), val = tensor(false)]; tensor var_8699_cast_fp16 = concat(axis = var_7806, interleave = var_8699_interleave_0, values = (var_8615_cast_fp16, var_8617_cast_fp16, var_8619_cast_fp16, var_8621_cast_fp16))[name = tensor("op_8699_cast_fp16")]; tensor var_8701_interleave_0 = const()[name = tensor("op_8701_interleave_0"), val = tensor(false)]; tensor var_8701_cast_fp16 = concat(axis = var_7806, interleave = var_8701_interleave_0, values = (var_8623_cast_fp16, var_8625_cast_fp16, var_8627_cast_fp16, var_8629_cast_fp16))[name = tensor("op_8701_cast_fp16")]; tensor var_8703_interleave_0 = const()[name = tensor("op_8703_interleave_0"), val = tensor(false)]; tensor var_8703_cast_fp16 = concat(axis = var_7806, interleave = var_8703_interleave_0, values = (var_8631_cast_fp16, var_8633_cast_fp16, var_8635_cast_fp16, var_8637_cast_fp16))[name = tensor("op_8703_cast_fp16")]; tensor var_8705_interleave_0 = const()[name = tensor("op_8705_interleave_0"), val = tensor(false)]; tensor var_8705_cast_fp16 = concat(axis = var_7806, interleave = var_8705_interleave_0, values = (var_8639_cast_fp16, var_8641_cast_fp16, var_8643_cast_fp16, var_8645_cast_fp16))[name = tensor("op_8705_cast_fp16")]; tensor var_8707_interleave_0 = const()[name = tensor("op_8707_interleave_0"), val = tensor(false)]; tensor var_8707_cast_fp16 = concat(axis = var_7806, interleave = var_8707_interleave_0, values = (var_8647_cast_fp16, var_8649_cast_fp16, var_8651_cast_fp16, var_8653_cast_fp16))[name = tensor("op_8707_cast_fp16")]; tensor var_8709_interleave_0 = const()[name = tensor("op_8709_interleave_0"), val = tensor(false)]; tensor var_8709_cast_fp16 = concat(axis = var_7806, interleave = var_8709_interleave_0, values = (var_8655_cast_fp16, var_8657_cast_fp16, var_8659_cast_fp16, var_8661_cast_fp16))[name = tensor("op_8709_cast_fp16")]; tensor var_8711_interleave_0 = const()[name = tensor("op_8711_interleave_0"), val = tensor(false)]; tensor var_8711_cast_fp16 = concat(axis = var_7806, interleave = var_8711_interleave_0, values = (var_8663_cast_fp16, var_8665_cast_fp16, var_8667_cast_fp16, var_8669_cast_fp16))[name = tensor("op_8711_cast_fp16")]; tensor var_8713_interleave_0 = const()[name = tensor("op_8713_interleave_0"), val = tensor(false)]; tensor var_8713_cast_fp16 = concat(axis = var_7806, interleave = var_8713_interleave_0, values = (var_8671_cast_fp16, var_8673_cast_fp16, var_8675_cast_fp16, var_8677_cast_fp16))[name = tensor("op_8713_cast_fp16")]; tensor var_8715_interleave_0 = const()[name = tensor("op_8715_interleave_0"), val = tensor(false)]; tensor var_8715_cast_fp16 = concat(axis = var_7806, interleave = var_8715_interleave_0, values = (var_8679_cast_fp16, var_8681_cast_fp16, var_8683_cast_fp16, var_8685_cast_fp16))[name = tensor("op_8715_cast_fp16")]; tensor var_8717_interleave_0 = const()[name = tensor("op_8717_interleave_0"), val = tensor(false)]; tensor var_8717_cast_fp16 = concat(axis = var_7806, interleave = var_8717_interleave_0, values = (var_8687_cast_fp16, var_8689_cast_fp16, var_8691_cast_fp16, var_8693_cast_fp16))[name = tensor("op_8717_cast_fp16")]; tensor input_65_interleave_0 = const()[name = tensor("input_65_interleave_0"), val = tensor(false)]; tensor input_65_cast_fp16 = concat(axis = var_7823, interleave = input_65_interleave_0, values = (var_8695_cast_fp16, var_8697_cast_fp16, var_8699_cast_fp16, var_8701_cast_fp16, var_8703_cast_fp16, var_8705_cast_fp16, var_8707_cast_fp16, var_8709_cast_fp16, var_8711_cast_fp16, var_8713_cast_fp16, var_8715_cast_fp16, var_8717_cast_fp16))[name = tensor("input_65_cast_fp16")]; tensor var_8722 = const()[name = tensor("op_8722"), val = tensor([1, 1])]; tensor var_8724 = const()[name = tensor("op_8724"), val = tensor([1, 1])]; tensor obj_35_pad_type_0 = const()[name = tensor("obj_35_pad_type_0"), val = tensor("custom")]; tensor obj_35_pad_0 = const()[name = tensor("obj_35_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123165120)))]; tensor layers_8_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124344832)))]; tensor obj_35_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_bias_to_fp16, dilations = var_8724, groups = var_7823, pad = obj_35_pad_0, pad_type = obj_35_pad_type_0, strides = var_8722, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_65_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_8730 = const()[name = tensor("op_8730"), val = tensor([1])]; tensor channels_mean_35_cast_fp16 = reduce_mean(axes = var_8730, keep_dims = var_7824, 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_8734 = const()[name = tensor("op_8734"), val = tensor([1])]; tensor var_8735_cast_fp16 = reduce_mean(axes = var_8734, keep_dims = var_7824, x = zero_mean_sq_35_cast_fp16)[name = tensor("op_8735_cast_fp16")]; tensor var_8736_to_fp16 = const()[name = tensor("op_8736_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8737_cast_fp16 = add(x = var_8735_cast_fp16, y = var_8736_to_fp16)[name = tensor("op_8737_cast_fp16")]; tensor denom_35_epsilon_0_to_fp16 = const()[name = tensor("denom_35_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_35_cast_fp16 = rsqrt(epsilon = denom_35_epsilon_0_to_fp16, x = var_8737_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_67_gamma_0_to_fp16 = const()[name = tensor("input_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124346432)))]; tensor input_67_beta_0_to_fp16 = const()[name = tensor("input_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124348032)))]; tensor input_67_epsilon_0_to_fp16 = const()[name = tensor("input_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_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_67_cast_fp16")]; tensor var_8748 = const()[name = tensor("op_8748"), val = tensor([1, 1])]; tensor var_8750 = const()[name = tensor("op_8750"), 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_8_fc1_weight_to_fp16 = const()[name = tensor("layers_8_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124349632)))]; tensor layers_8_fc1_bias_to_fp16 = const()[name = tensor("layers_8_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129068288)))]; tensor input_69_cast_fp16 = conv(bias = layers_8_fc1_bias_to_fp16, dilations = var_8750, groups = var_7823, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = var_8748, weight = layers_8_fc1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("input_69_cast_fp16")]; tensor input_71_mode_0 = const()[name = tensor("input_71_mode_0"), val = tensor("EXACT")]; tensor input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = tensor("input_71_cast_fp16")]; tensor var_8756 = const()[name = tensor("op_8756"), val = tensor([1, 1])]; tensor var_8758 = const()[name = tensor("op_8758"), val = tensor([1, 1])]; tensor hidden_states_21_pad_type_0 = const()[name = tensor("hidden_states_21_pad_type_0"), val = tensor("custom")]; tensor hidden_states_21_pad_0 = const()[name = tensor("hidden_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_8_fc2_weight_to_fp16 = const()[name = tensor("layers_8_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129074496)))]; tensor layers_8_fc2_bias_to_fp16 = const()[name = tensor("layers_8_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133793152)))]; tensor hidden_states_21_cast_fp16 = conv(bias = layers_8_fc2_bias_to_fp16, dilations = var_8758, groups = var_7823, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_8756, weight = layers_8_fc2_weight_to_fp16, x = input_71_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_8765 = const()[name = tensor("op_8765"), val = tensor(3)]; tensor var_8782 = const()[name = tensor("op_8782"), val = tensor(1)]; tensor var_8783 = const()[name = tensor("op_8783"), val = tensor(true)]; tensor var_8793 = const()[name = tensor("op_8793"), val = tensor([1])]; tensor channels_mean_37_cast_fp16 = reduce_mean(axes = var_8793, keep_dims = var_8783, 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_8797 = const()[name = tensor("op_8797"), val = tensor([1])]; tensor var_8798_cast_fp16 = reduce_mean(axes = var_8797, keep_dims = var_8783, x = zero_mean_sq_37_cast_fp16)[name = tensor("op_8798_cast_fp16")]; tensor var_8799_to_fp16 = const()[name = tensor("op_8799_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_8800_cast_fp16 = add(x = var_8798_cast_fp16, y = var_8799_to_fp16)[name = tensor("op_8800_cast_fp16")]; tensor denom_37_epsilon_0_to_fp16 = const()[name = tensor("denom_37_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_37_cast_fp16 = rsqrt(epsilon = denom_37_epsilon_0_to_fp16, x = var_8800_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(133794752)))]; 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(133796352)))]; 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_8815 = const()[name = tensor("op_8815"), val = tensor([1, 1])]; tensor var_8817 = const()[name = tensor("op_8817"), val = tensor([1, 1])]; tensor query_19_pad_type_0 = const()[name = tensor("query_19_pad_type_0"), val = tensor("custom")]; tensor query_19_pad_0 = const()[name = tensor("query_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133797952)))]; tensor layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134977664)))]; tensor query_19_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_bias_to_fp16, dilations = var_8817, groups = var_8782, pad = query_19_pad_0, pad_type = query_19_pad_type_0, strides = var_8815, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("query_19_cast_fp16")]; tensor var_8821 = const()[name = tensor("op_8821"), val = tensor([1, 1])]; tensor var_8823 = const()[name = tensor("op_8823"), val = tensor([1, 1])]; tensor key_19_pad_type_0 = const()[name = tensor("key_19_pad_type_0"), val = tensor("custom")]; tensor key_19_pad_0 = const()[name = tensor("key_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134979264)))]; tensor key_19_cast_fp16 = conv(dilations = var_8823, groups = var_8782, pad = key_19_pad_0, pad_type = key_19_pad_type_0, strides = var_8821, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("key_19_cast_fp16")]; tensor var_8828 = const()[name = tensor("op_8828"), val = tensor([1, 1])]; tensor var_8830 = const()[name = tensor("op_8830"), val = tensor([1, 1])]; tensor value_19_pad_type_0 = const()[name = tensor("value_19_pad_type_0"), val = tensor("custom")]; tensor value_19_pad_0 = const()[name = tensor("value_19_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136158976)))]; tensor layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137338688)))]; tensor value_19_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_bias_to_fp16, dilations = var_8830, groups = var_8782, pad = value_19_pad_0, pad_type = value_19_pad_type_0, strides = var_8828, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor("value_19_cast_fp16")]; tensor var_8837_begin_0 = const()[name = tensor("op_8837_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8837_end_0 = const()[name = tensor("op_8837_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_8837_end_mask_0 = const()[name = tensor("op_8837_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8837_cast_fp16 = slice_by_index(begin = var_8837_begin_0, end = var_8837_end_0, end_mask = var_8837_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8837_cast_fp16")]; tensor var_8841_begin_0 = const()[name = tensor("op_8841_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_8841_end_0 = const()[name = tensor("op_8841_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_8841_end_mask_0 = const()[name = tensor("op_8841_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_19_cast_fp16)[name = tensor("op_8841_cast_fp16")]; tensor var_8845_begin_0 = const()[name = tensor("op_8845_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_8845_end_0 = const()[name = tensor("op_8845_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_8845_end_mask_0 = const()[name = tensor("op_8845_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8845_cast_fp16 = slice_by_index(begin = var_8845_begin_0, end = var_8845_end_0, end_mask = var_8845_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8845_cast_fp16")]; tensor var_8849_begin_0 = const()[name = tensor("op_8849_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_8849_end_0 = const()[name = tensor("op_8849_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_8849_end_mask_0 = const()[name = tensor("op_8849_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8849_cast_fp16 = slice_by_index(begin = var_8849_begin_0, end = var_8849_end_0, end_mask = var_8849_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8849_cast_fp16")]; tensor var_8853_begin_0 = const()[name = tensor("op_8853_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_8853_end_0 = const()[name = tensor("op_8853_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_8853_end_mask_0 = const()[name = tensor("op_8853_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8853_cast_fp16 = slice_by_index(begin = var_8853_begin_0, end = var_8853_end_0, end_mask = var_8853_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8853_cast_fp16")]; tensor var_8857_begin_0 = const()[name = tensor("op_8857_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_8857_end_0 = const()[name = tensor("op_8857_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_8857_end_mask_0 = const()[name = tensor("op_8857_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8857_cast_fp16 = slice_by_index(begin = var_8857_begin_0, end = var_8857_end_0, end_mask = var_8857_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8857_cast_fp16")]; tensor var_8861_begin_0 = const()[name = tensor("op_8861_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_8861_end_0 = const()[name = tensor("op_8861_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_8861_end_mask_0 = const()[name = tensor("op_8861_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8861_cast_fp16 = slice_by_index(begin = var_8861_begin_0, end = var_8861_end_0, end_mask = var_8861_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8861_cast_fp16")]; tensor var_8865_begin_0 = const()[name = tensor("op_8865_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_8865_end_0 = const()[name = tensor("op_8865_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_8865_end_mask_0 = const()[name = tensor("op_8865_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8865_cast_fp16 = slice_by_index(begin = var_8865_begin_0, end = var_8865_end_0, end_mask = var_8865_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8865_cast_fp16")]; tensor var_8869_begin_0 = const()[name = tensor("op_8869_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_8869_end_0 = const()[name = tensor("op_8869_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_8869_end_mask_0 = const()[name = tensor("op_8869_end_mask_0"), val = tensor([true, false, true, true])]; 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 = query_19_cast_fp16)[name = tensor("op_8869_cast_fp16")]; tensor var_8873_begin_0 = const()[name = tensor("op_8873_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_8873_end_0 = const()[name = tensor("op_8873_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_8873_end_mask_0 = const()[name = tensor("op_8873_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8873_cast_fp16 = slice_by_index(begin = var_8873_begin_0, end = var_8873_end_0, end_mask = var_8873_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8873_cast_fp16")]; tensor var_8877_begin_0 = const()[name = tensor("op_8877_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_8877_end_0 = const()[name = tensor("op_8877_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_8877_end_mask_0 = const()[name = tensor("op_8877_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8877_cast_fp16 = slice_by_index(begin = var_8877_begin_0, end = var_8877_end_0, end_mask = var_8877_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8877_cast_fp16")]; tensor var_8881_begin_0 = const()[name = tensor("op_8881_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_8881_end_0 = const()[name = tensor("op_8881_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_8881_end_mask_0 = const()[name = tensor("op_8881_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_8881_cast_fp16 = slice_by_index(begin = var_8881_begin_0, end = var_8881_end_0, end_mask = var_8881_end_mask_0, x = query_19_cast_fp16)[name = tensor("op_8881_cast_fp16")]; tensor var_8890_begin_0 = const()[name = tensor("op_8890_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_8890_end_0 = const()[name = tensor("op_8890_end_0"), val = tensor([1, 64, 1, 375])]; 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_8837_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, 375])]; tensor var_8897_end_0 = const()[name = tensor("op_8897_end_0"), val = tensor([1, 64, 1, 750])]; 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_8837_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, 750])]; tensor var_8904_end_0 = const()[name = tensor("op_8904_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8837_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, 1125])]; tensor var_8911_end_0 = const()[name = tensor("op_8911_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8837_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, 0])]; tensor var_8918_end_0 = const()[name = tensor("op_8918_end_0"), val = tensor([1, 64, 1, 375])]; 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_8841_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, 375])]; tensor var_8925_end_0 = const()[name = tensor("op_8925_end_0"), val = tensor([1, 64, 1, 750])]; 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_8841_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, 750])]; tensor var_8932_end_0 = const()[name = tensor("op_8932_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8841_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, 1125])]; tensor var_8939_end_0 = const()[name = tensor("op_8939_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8841_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, 0])]; tensor var_8946_end_0 = const()[name = tensor("op_8946_end_0"), val = tensor([1, 64, 1, 375])]; 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_8845_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, 375])]; tensor var_8953_end_0 = const()[name = tensor("op_8953_end_0"), val = tensor([1, 64, 1, 750])]; 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_8845_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, 750])]; tensor var_8960_end_0 = const()[name = tensor("op_8960_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8845_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, 1125])]; tensor var_8967_end_0 = const()[name = tensor("op_8967_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8845_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, 0])]; tensor var_8974_end_0 = const()[name = tensor("op_8974_end_0"), val = tensor([1, 64, 1, 375])]; 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_8849_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, 375])]; tensor var_8981_end_0 = const()[name = tensor("op_8981_end_0"), val = tensor([1, 64, 1, 750])]; 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_8849_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, 750])]; tensor var_8988_end_0 = const()[name = tensor("op_8988_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8849_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, 1125])]; tensor var_8995_end_0 = const()[name = tensor("op_8995_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8849_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, 0])]; tensor var_9002_end_0 = const()[name = tensor("op_9002_end_0"), val = tensor([1, 64, 1, 375])]; 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_8853_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, 375])]; tensor var_9009_end_0 = const()[name = tensor("op_9009_end_0"), val = tensor([1, 64, 1, 750])]; 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_8853_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, 750])]; tensor var_9016_end_0 = const()[name = tensor("op_9016_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8853_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, 1125])]; tensor var_9023_end_0 = const()[name = tensor("op_9023_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8853_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, 0])]; tensor var_9030_end_0 = const()[name = tensor("op_9030_end_0"), val = tensor([1, 64, 1, 375])]; 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_8857_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, 375])]; tensor var_9037_end_0 = const()[name = tensor("op_9037_end_0"), val = tensor([1, 64, 1, 750])]; 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_8857_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, 750])]; tensor var_9044_end_0 = const()[name = tensor("op_9044_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8857_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, 1125])]; tensor var_9051_end_0 = const()[name = tensor("op_9051_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8857_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, 0])]; tensor var_9058_end_0 = const()[name = tensor("op_9058_end_0"), val = tensor([1, 64, 1, 375])]; 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_8861_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, 375])]; tensor var_9065_end_0 = const()[name = tensor("op_9065_end_0"), val = tensor([1, 64, 1, 750])]; 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_8861_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, 750])]; tensor var_9072_end_0 = const()[name = tensor("op_9072_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8861_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, 1125])]; tensor var_9079_end_0 = const()[name = tensor("op_9079_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8861_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, 0])]; tensor var_9086_end_0 = const()[name = tensor("op_9086_end_0"), val = tensor([1, 64, 1, 375])]; 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_8865_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, 375])]; tensor var_9093_end_0 = const()[name = tensor("op_9093_end_0"), val = tensor([1, 64, 1, 750])]; 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_8865_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, 750])]; tensor var_9100_end_0 = const()[name = tensor("op_9100_end_0"), val = tensor([1, 64, 1, 1125])]; 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_8865_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, 1125])]; tensor var_9107_end_0 = const()[name = tensor("op_9107_end_0"), val = tensor([1, 64, 1, 1500])]; 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_8865_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, 0])]; tensor var_9114_end_0 = const()[name = tensor("op_9114_end_0"), val = tensor([1, 64, 1, 375])]; 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_8869_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, 375])]; tensor var_9121_end_0 = const()[name = tensor("op_9121_end_0"), val = tensor([1, 64, 1, 750])]; 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_8869_cast_fp16)[name = tensor("op_9121_cast_fp16")]; tensor var_9128_begin_0 = const()[name = tensor("op_9128_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9128_end_0 = const()[name = tensor("op_9128_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9128_end_mask_0 = const()[name = tensor("op_9128_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9128_cast_fp16 = slice_by_index(begin = var_9128_begin_0, end = var_9128_end_0, end_mask = var_9128_end_mask_0, x = var_8869_cast_fp16)[name = tensor("op_9128_cast_fp16")]; tensor var_9135_begin_0 = const()[name = tensor("op_9135_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9135_end_0 = const()[name = tensor("op_9135_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9135_end_mask_0 = const()[name = tensor("op_9135_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9135_cast_fp16 = slice_by_index(begin = var_9135_begin_0, end = var_9135_end_0, end_mask = var_9135_end_mask_0, x = var_8869_cast_fp16)[name = tensor("op_9135_cast_fp16")]; tensor var_9142_begin_0 = const()[name = tensor("op_9142_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9142_end_0 = const()[name = tensor("op_9142_end_0"), val = tensor([1, 64, 1, 375])]; 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 = var_8873_cast_fp16)[name = tensor("op_9142_cast_fp16")]; tensor var_9149_begin_0 = const()[name = tensor("op_9149_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9149_end_0 = const()[name = tensor("op_9149_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9149_end_mask_0 = const()[name = tensor("op_9149_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9149_cast_fp16 = slice_by_index(begin = var_9149_begin_0, end = var_9149_end_0, end_mask = var_9149_end_mask_0, x = var_8873_cast_fp16)[name = tensor("op_9149_cast_fp16")]; tensor var_9156_begin_0 = const()[name = tensor("op_9156_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9156_end_0 = const()[name = tensor("op_9156_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9156_end_mask_0 = const()[name = tensor("op_9156_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9156_cast_fp16 = slice_by_index(begin = var_9156_begin_0, end = var_9156_end_0, end_mask = var_9156_end_mask_0, x = var_8873_cast_fp16)[name = tensor("op_9156_cast_fp16")]; tensor var_9163_begin_0 = const()[name = tensor("op_9163_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9163_end_0 = const()[name = tensor("op_9163_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9163_end_mask_0 = const()[name = tensor("op_9163_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9163_cast_fp16 = slice_by_index(begin = var_9163_begin_0, end = var_9163_end_0, end_mask = var_9163_end_mask_0, x = var_8873_cast_fp16)[name = tensor("op_9163_cast_fp16")]; tensor var_9170_begin_0 = const()[name = tensor("op_9170_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9170_end_0 = const()[name = tensor("op_9170_end_0"), val = tensor([1, 64, 1, 375])]; 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 = var_8877_cast_fp16)[name = tensor("op_9170_cast_fp16")]; tensor var_9177_begin_0 = const()[name = tensor("op_9177_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9177_end_0 = const()[name = tensor("op_9177_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9177_end_mask_0 = const()[name = tensor("op_9177_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9177_cast_fp16 = slice_by_index(begin = var_9177_begin_0, end = var_9177_end_0, end_mask = var_9177_end_mask_0, x = var_8877_cast_fp16)[name = tensor("op_9177_cast_fp16")]; tensor var_9184_begin_0 = const()[name = tensor("op_9184_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9184_end_0 = const()[name = tensor("op_9184_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9184_end_mask_0 = const()[name = tensor("op_9184_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9184_cast_fp16 = slice_by_index(begin = var_9184_begin_0, end = var_9184_end_0, end_mask = var_9184_end_mask_0, x = var_8877_cast_fp16)[name = tensor("op_9184_cast_fp16")]; tensor var_9191_begin_0 = const()[name = tensor("op_9191_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9191_end_0 = const()[name = tensor("op_9191_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9191_end_mask_0 = const()[name = tensor("op_9191_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9191_cast_fp16 = slice_by_index(begin = var_9191_begin_0, end = var_9191_end_0, end_mask = var_9191_end_mask_0, x = var_8877_cast_fp16)[name = tensor("op_9191_cast_fp16")]; tensor var_9198_begin_0 = const()[name = tensor("op_9198_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9198_end_0 = const()[name = tensor("op_9198_end_0"), val = tensor([1, 64, 1, 375])]; 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 = var_8881_cast_fp16)[name = tensor("op_9198_cast_fp16")]; tensor var_9205_begin_0 = const()[name = tensor("op_9205_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9205_end_0 = const()[name = tensor("op_9205_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9205_end_mask_0 = const()[name = tensor("op_9205_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9205_cast_fp16 = slice_by_index(begin = var_9205_begin_0, end = var_9205_end_0, end_mask = var_9205_end_mask_0, x = var_8881_cast_fp16)[name = tensor("op_9205_cast_fp16")]; tensor var_9212_begin_0 = const()[name = tensor("op_9212_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9212_end_0 = const()[name = tensor("op_9212_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9212_end_mask_0 = const()[name = tensor("op_9212_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_8881_cast_fp16)[name = tensor("op_9212_cast_fp16")]; tensor var_9219_begin_0 = const()[name = tensor("op_9219_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9219_end_0 = const()[name = tensor("op_9219_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9219_end_mask_0 = const()[name = tensor("op_9219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9219_cast_fp16 = slice_by_index(begin = var_9219_begin_0, end = var_9219_end_0, end_mask = var_9219_end_mask_0, x = var_8881_cast_fp16)[name = tensor("op_9219_cast_fp16")]; tensor k_19_perm_0 = const()[name = tensor("k_19_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_9224_begin_0 = const()[name = tensor("op_9224_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9224_end_0 = const()[name = tensor("op_9224_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_9224_end_mask_0 = const()[name = tensor("op_9224_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_2 = transpose(perm = k_19_perm_0, x = key_19_cast_fp16)[name = tensor("transpose_2")]; 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 = transpose_2)[name = tensor("op_9224_cast_fp16")]; tensor var_9228_begin_0 = const()[name = tensor("op_9228_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_9228_end_0 = const()[name = tensor("op_9228_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_9228_end_mask_0 = const()[name = tensor("op_9228_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9228_cast_fp16")]; tensor var_9232_begin_0 = const()[name = tensor("op_9232_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_9232_end_0 = const()[name = tensor("op_9232_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_9232_end_mask_0 = const()[name = tensor("op_9232_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9232_cast_fp16")]; tensor var_9236_begin_0 = const()[name = tensor("op_9236_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_9236_end_0 = const()[name = tensor("op_9236_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_9236_end_mask_0 = const()[name = tensor("op_9236_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9236_cast_fp16")]; tensor var_9240_begin_0 = const()[name = tensor("op_9240_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_9240_end_0 = const()[name = tensor("op_9240_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_9240_end_mask_0 = const()[name = tensor("op_9240_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9240_cast_fp16")]; tensor var_9244_begin_0 = const()[name = tensor("op_9244_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_9244_end_0 = const()[name = tensor("op_9244_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_9244_end_mask_0 = const()[name = tensor("op_9244_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9244_cast_fp16")]; tensor var_9248_begin_0 = const()[name = tensor("op_9248_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_9248_end_0 = const()[name = tensor("op_9248_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_9248_end_mask_0 = const()[name = tensor("op_9248_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9248_cast_fp16")]; tensor var_9252_begin_0 = const()[name = tensor("op_9252_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_9252_end_0 = const()[name = tensor("op_9252_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_9252_end_mask_0 = const()[name = tensor("op_9252_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9252_cast_fp16")]; tensor var_9256_begin_0 = const()[name = tensor("op_9256_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_9256_end_0 = const()[name = tensor("op_9256_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_9256_end_mask_0 = const()[name = tensor("op_9256_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9256_cast_fp16")]; tensor var_9260_begin_0 = const()[name = tensor("op_9260_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_9260_end_0 = const()[name = tensor("op_9260_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_9260_end_mask_0 = const()[name = tensor("op_9260_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9260_cast_fp16")]; tensor var_9264_begin_0 = const()[name = tensor("op_9264_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_9264_end_0 = const()[name = tensor("op_9264_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_9264_end_mask_0 = const()[name = tensor("op_9264_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9264_cast_fp16")]; tensor var_9268_begin_0 = const()[name = tensor("op_9268_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_9268_end_0 = const()[name = tensor("op_9268_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_9268_end_mask_0 = const()[name = tensor("op_9268_end_mask_0"), val = tensor([true, true, true, false])]; 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 = transpose_2)[name = tensor("op_9268_cast_fp16")]; tensor var_9270_begin_0 = const()[name = tensor("op_9270_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9270_end_0 = const()[name = tensor("op_9270_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9270_end_mask_0 = const()[name = tensor("op_9270_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9270_cast_fp16 = slice_by_index(begin = var_9270_begin_0, end = var_9270_end_0, end_mask = var_9270_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9270_cast_fp16")]; tensor var_9274_begin_0 = const()[name = tensor("op_9274_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9274_end_0 = const()[name = tensor("op_9274_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_9274_end_mask_0 = const()[name = tensor("op_9274_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9274_cast_fp16 = slice_by_index(begin = var_9274_begin_0, end = var_9274_end_0, end_mask = var_9274_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9274_cast_fp16")]; tensor var_9278_begin_0 = const()[name = tensor("op_9278_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9278_end_0 = const()[name = tensor("op_9278_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_9278_end_mask_0 = const()[name = tensor("op_9278_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9278_cast_fp16 = slice_by_index(begin = var_9278_begin_0, end = var_9278_end_0, end_mask = var_9278_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9278_cast_fp16")]; tensor var_9282_begin_0 = const()[name = tensor("op_9282_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9282_end_0 = const()[name = tensor("op_9282_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_9282_end_mask_0 = const()[name = tensor("op_9282_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9282_cast_fp16 = slice_by_index(begin = var_9282_begin_0, end = var_9282_end_0, end_mask = var_9282_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9282_cast_fp16")]; tensor var_9286_begin_0 = const()[name = tensor("op_9286_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9286_end_0 = const()[name = tensor("op_9286_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_9286_end_mask_0 = const()[name = tensor("op_9286_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9286_cast_fp16 = slice_by_index(begin = var_9286_begin_0, end = var_9286_end_0, end_mask = var_9286_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9286_cast_fp16")]; tensor var_9290_begin_0 = const()[name = tensor("op_9290_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9290_end_0 = const()[name = tensor("op_9290_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_9290_end_mask_0 = const()[name = tensor("op_9290_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9290_cast_fp16 = slice_by_index(begin = var_9290_begin_0, end = var_9290_end_0, end_mask = var_9290_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9290_cast_fp16")]; tensor var_9294_begin_0 = const()[name = tensor("op_9294_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9294_end_0 = const()[name = tensor("op_9294_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_9294_end_mask_0 = const()[name = tensor("op_9294_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9294_cast_fp16 = slice_by_index(begin = var_9294_begin_0, end = var_9294_end_0, end_mask = var_9294_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9294_cast_fp16")]; tensor var_9298_begin_0 = const()[name = tensor("op_9298_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9298_end_0 = const()[name = tensor("op_9298_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_9298_end_mask_0 = const()[name = tensor("op_9298_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9298_cast_fp16 = slice_by_index(begin = var_9298_begin_0, end = var_9298_end_0, end_mask = var_9298_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9298_cast_fp16")]; tensor var_9302_begin_0 = const()[name = tensor("op_9302_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9302_end_0 = const()[name = tensor("op_9302_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_9302_end_mask_0 = const()[name = tensor("op_9302_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9302_cast_fp16 = slice_by_index(begin = var_9302_begin_0, end = var_9302_end_0, end_mask = var_9302_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9302_cast_fp16")]; tensor var_9306_begin_0 = const()[name = tensor("op_9306_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9306_end_0 = const()[name = tensor("op_9306_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_9306_end_mask_0 = const()[name = tensor("op_9306_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9306_cast_fp16 = slice_by_index(begin = var_9306_begin_0, end = var_9306_end_0, end_mask = var_9306_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9306_cast_fp16")]; tensor var_9310_begin_0 = const()[name = tensor("op_9310_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_9310_end_0 = const()[name = tensor("op_9310_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_9310_end_mask_0 = const()[name = tensor("op_9310_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9310_cast_fp16 = slice_by_index(begin = var_9310_begin_0, end = var_9310_end_0, end_mask = var_9310_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9310_cast_fp16")]; tensor var_9314_begin_0 = const()[name = tensor("op_9314_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_9314_end_0 = const()[name = tensor("op_9314_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_9314_end_mask_0 = const()[name = tensor("op_9314_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9314_cast_fp16 = slice_by_index(begin = var_9314_begin_0, end = var_9314_end_0, end_mask = var_9314_end_mask_0, x = value_19_cast_fp16)[name = tensor("op_9314_cast_fp16")]; tensor var_9318_equation_0 = const()[name = tensor("op_9318_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9318_cast_fp16 = einsum(equation = var_9318_equation_0, values = (var_9224_cast_fp16, var_8890_cast_fp16))[name = tensor("op_9318_cast_fp16")]; tensor var_9319_to_fp16 = const()[name = tensor("op_9319_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_865_cast_fp16 = mul(x = var_9318_cast_fp16, y = var_9319_to_fp16)[name = tensor("aw_chunk_865_cast_fp16")]; tensor var_9322_equation_0 = const()[name = tensor("op_9322_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9322_cast_fp16 = einsum(equation = var_9322_equation_0, values = (var_9224_cast_fp16, var_8897_cast_fp16))[name = tensor("op_9322_cast_fp16")]; tensor var_9323_to_fp16 = const()[name = tensor("op_9323_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_867_cast_fp16 = mul(x = var_9322_cast_fp16, y = var_9323_to_fp16)[name = tensor("aw_chunk_867_cast_fp16")]; tensor var_9326_equation_0 = const()[name = tensor("op_9326_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9326_cast_fp16 = einsum(equation = var_9326_equation_0, values = (var_9224_cast_fp16, var_8904_cast_fp16))[name = tensor("op_9326_cast_fp16")]; tensor var_9327_to_fp16 = const()[name = tensor("op_9327_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_869_cast_fp16 = mul(x = var_9326_cast_fp16, y = var_9327_to_fp16)[name = tensor("aw_chunk_869_cast_fp16")]; tensor var_9330_equation_0 = const()[name = tensor("op_9330_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9330_cast_fp16 = einsum(equation = var_9330_equation_0, values = (var_9224_cast_fp16, var_8911_cast_fp16))[name = tensor("op_9330_cast_fp16")]; tensor var_9331_to_fp16 = const()[name = tensor("op_9331_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_871_cast_fp16 = mul(x = var_9330_cast_fp16, y = var_9331_to_fp16)[name = tensor("aw_chunk_871_cast_fp16")]; tensor var_9334_equation_0 = const()[name = tensor("op_9334_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9334_cast_fp16 = einsum(equation = var_9334_equation_0, values = (var_9228_cast_fp16, var_8918_cast_fp16))[name = tensor("op_9334_cast_fp16")]; tensor var_9335_to_fp16 = const()[name = tensor("op_9335_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_873_cast_fp16 = mul(x = var_9334_cast_fp16, y = var_9335_to_fp16)[name = tensor("aw_chunk_873_cast_fp16")]; tensor var_9338_equation_0 = const()[name = tensor("op_9338_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9338_cast_fp16 = einsum(equation = var_9338_equation_0, values = (var_9228_cast_fp16, var_8925_cast_fp16))[name = tensor("op_9338_cast_fp16")]; tensor var_9339_to_fp16 = const()[name = tensor("op_9339_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_875_cast_fp16 = mul(x = var_9338_cast_fp16, y = var_9339_to_fp16)[name = tensor("aw_chunk_875_cast_fp16")]; tensor var_9342_equation_0 = const()[name = tensor("op_9342_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9342_cast_fp16 = einsum(equation = var_9342_equation_0, values = (var_9228_cast_fp16, var_8932_cast_fp16))[name = tensor("op_9342_cast_fp16")]; tensor var_9343_to_fp16 = const()[name = tensor("op_9343_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_877_cast_fp16 = mul(x = var_9342_cast_fp16, y = var_9343_to_fp16)[name = tensor("aw_chunk_877_cast_fp16")]; tensor var_9346_equation_0 = const()[name = tensor("op_9346_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9346_cast_fp16 = einsum(equation = var_9346_equation_0, values = (var_9228_cast_fp16, var_8939_cast_fp16))[name = tensor("op_9346_cast_fp16")]; tensor var_9347_to_fp16 = const()[name = tensor("op_9347_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_879_cast_fp16 = mul(x = var_9346_cast_fp16, y = var_9347_to_fp16)[name = tensor("aw_chunk_879_cast_fp16")]; tensor var_9350_equation_0 = const()[name = tensor("op_9350_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9350_cast_fp16 = einsum(equation = var_9350_equation_0, values = (var_9232_cast_fp16, var_8946_cast_fp16))[name = tensor("op_9350_cast_fp16")]; tensor var_9351_to_fp16 = const()[name = tensor("op_9351_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_881_cast_fp16 = mul(x = var_9350_cast_fp16, y = var_9351_to_fp16)[name = tensor("aw_chunk_881_cast_fp16")]; tensor var_9354_equation_0 = const()[name = tensor("op_9354_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9354_cast_fp16 = einsum(equation = var_9354_equation_0, values = (var_9232_cast_fp16, var_8953_cast_fp16))[name = tensor("op_9354_cast_fp16")]; tensor var_9355_to_fp16 = const()[name = tensor("op_9355_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_883_cast_fp16 = mul(x = var_9354_cast_fp16, y = var_9355_to_fp16)[name = tensor("aw_chunk_883_cast_fp16")]; tensor var_9358_equation_0 = const()[name = tensor("op_9358_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9358_cast_fp16 = einsum(equation = var_9358_equation_0, values = (var_9232_cast_fp16, var_8960_cast_fp16))[name = tensor("op_9358_cast_fp16")]; tensor var_9359_to_fp16 = const()[name = tensor("op_9359_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_885_cast_fp16 = mul(x = var_9358_cast_fp16, y = var_9359_to_fp16)[name = tensor("aw_chunk_885_cast_fp16")]; tensor var_9362_equation_0 = const()[name = tensor("op_9362_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9362_cast_fp16 = einsum(equation = var_9362_equation_0, values = (var_9232_cast_fp16, var_8967_cast_fp16))[name = tensor("op_9362_cast_fp16")]; tensor var_9363_to_fp16 = const()[name = tensor("op_9363_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_887_cast_fp16 = mul(x = var_9362_cast_fp16, y = var_9363_to_fp16)[name = tensor("aw_chunk_887_cast_fp16")]; tensor var_9366_equation_0 = const()[name = tensor("op_9366_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9366_cast_fp16 = einsum(equation = var_9366_equation_0, values = (var_9236_cast_fp16, var_8974_cast_fp16))[name = tensor("op_9366_cast_fp16")]; tensor var_9367_to_fp16 = const()[name = tensor("op_9367_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_889_cast_fp16 = mul(x = var_9366_cast_fp16, y = var_9367_to_fp16)[name = tensor("aw_chunk_889_cast_fp16")]; tensor var_9370_equation_0 = const()[name = tensor("op_9370_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9370_cast_fp16 = einsum(equation = var_9370_equation_0, values = (var_9236_cast_fp16, var_8981_cast_fp16))[name = tensor("op_9370_cast_fp16")]; tensor var_9371_to_fp16 = const()[name = tensor("op_9371_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_891_cast_fp16 = mul(x = var_9370_cast_fp16, y = var_9371_to_fp16)[name = tensor("aw_chunk_891_cast_fp16")]; tensor var_9374_equation_0 = const()[name = tensor("op_9374_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9374_cast_fp16 = einsum(equation = var_9374_equation_0, values = (var_9236_cast_fp16, var_8988_cast_fp16))[name = tensor("op_9374_cast_fp16")]; tensor var_9375_to_fp16 = const()[name = tensor("op_9375_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_893_cast_fp16 = mul(x = var_9374_cast_fp16, y = var_9375_to_fp16)[name = tensor("aw_chunk_893_cast_fp16")]; tensor var_9378_equation_0 = const()[name = tensor("op_9378_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9378_cast_fp16 = einsum(equation = var_9378_equation_0, values = (var_9236_cast_fp16, var_8995_cast_fp16))[name = tensor("op_9378_cast_fp16")]; tensor var_9379_to_fp16 = const()[name = tensor("op_9379_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_895_cast_fp16 = mul(x = var_9378_cast_fp16, y = var_9379_to_fp16)[name = tensor("aw_chunk_895_cast_fp16")]; tensor var_9382_equation_0 = const()[name = tensor("op_9382_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9382_cast_fp16 = einsum(equation = var_9382_equation_0, values = (var_9240_cast_fp16, var_9002_cast_fp16))[name = tensor("op_9382_cast_fp16")]; tensor var_9383_to_fp16 = const()[name = tensor("op_9383_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_897_cast_fp16 = mul(x = var_9382_cast_fp16, y = var_9383_to_fp16)[name = tensor("aw_chunk_897_cast_fp16")]; tensor var_9386_equation_0 = const()[name = tensor("op_9386_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9386_cast_fp16 = einsum(equation = var_9386_equation_0, values = (var_9240_cast_fp16, var_9009_cast_fp16))[name = tensor("op_9386_cast_fp16")]; tensor var_9387_to_fp16 = const()[name = tensor("op_9387_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_899_cast_fp16 = mul(x = var_9386_cast_fp16, y = var_9387_to_fp16)[name = tensor("aw_chunk_899_cast_fp16")]; tensor var_9390_equation_0 = const()[name = tensor("op_9390_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9390_cast_fp16 = einsum(equation = var_9390_equation_0, values = (var_9240_cast_fp16, var_9016_cast_fp16))[name = tensor("op_9390_cast_fp16")]; tensor var_9391_to_fp16 = const()[name = tensor("op_9391_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_901_cast_fp16 = mul(x = var_9390_cast_fp16, y = var_9391_to_fp16)[name = tensor("aw_chunk_901_cast_fp16")]; tensor var_9394_equation_0 = const()[name = tensor("op_9394_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9394_cast_fp16 = einsum(equation = var_9394_equation_0, values = (var_9240_cast_fp16, var_9023_cast_fp16))[name = tensor("op_9394_cast_fp16")]; tensor var_9395_to_fp16 = const()[name = tensor("op_9395_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_903_cast_fp16 = mul(x = var_9394_cast_fp16, y = var_9395_to_fp16)[name = tensor("aw_chunk_903_cast_fp16")]; tensor var_9398_equation_0 = const()[name = tensor("op_9398_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9398_cast_fp16 = einsum(equation = var_9398_equation_0, values = (var_9244_cast_fp16, var_9030_cast_fp16))[name = tensor("op_9398_cast_fp16")]; tensor var_9399_to_fp16 = const()[name = tensor("op_9399_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_905_cast_fp16 = mul(x = var_9398_cast_fp16, y = var_9399_to_fp16)[name = tensor("aw_chunk_905_cast_fp16")]; tensor var_9402_equation_0 = const()[name = tensor("op_9402_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9402_cast_fp16 = einsum(equation = var_9402_equation_0, values = (var_9244_cast_fp16, var_9037_cast_fp16))[name = tensor("op_9402_cast_fp16")]; tensor var_9403_to_fp16 = const()[name = tensor("op_9403_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_907_cast_fp16 = mul(x = var_9402_cast_fp16, y = var_9403_to_fp16)[name = tensor("aw_chunk_907_cast_fp16")]; tensor var_9406_equation_0 = const()[name = tensor("op_9406_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9406_cast_fp16 = einsum(equation = var_9406_equation_0, values = (var_9244_cast_fp16, var_9044_cast_fp16))[name = tensor("op_9406_cast_fp16")]; tensor var_9407_to_fp16 = const()[name = tensor("op_9407_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_909_cast_fp16 = mul(x = var_9406_cast_fp16, y = var_9407_to_fp16)[name = tensor("aw_chunk_909_cast_fp16")]; tensor var_9410_equation_0 = const()[name = tensor("op_9410_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9410_cast_fp16 = einsum(equation = var_9410_equation_0, values = (var_9244_cast_fp16, var_9051_cast_fp16))[name = tensor("op_9410_cast_fp16")]; tensor var_9411_to_fp16 = const()[name = tensor("op_9411_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_911_cast_fp16 = mul(x = var_9410_cast_fp16, y = var_9411_to_fp16)[name = tensor("aw_chunk_911_cast_fp16")]; tensor var_9414_equation_0 = const()[name = tensor("op_9414_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9414_cast_fp16 = einsum(equation = var_9414_equation_0, values = (var_9248_cast_fp16, var_9058_cast_fp16))[name = tensor("op_9414_cast_fp16")]; tensor var_9415_to_fp16 = const()[name = tensor("op_9415_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_913_cast_fp16 = mul(x = var_9414_cast_fp16, y = var_9415_to_fp16)[name = tensor("aw_chunk_913_cast_fp16")]; tensor var_9418_equation_0 = const()[name = tensor("op_9418_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9418_cast_fp16 = einsum(equation = var_9418_equation_0, values = (var_9248_cast_fp16, var_9065_cast_fp16))[name = tensor("op_9418_cast_fp16")]; tensor var_9419_to_fp16 = const()[name = tensor("op_9419_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_915_cast_fp16 = mul(x = var_9418_cast_fp16, y = var_9419_to_fp16)[name = tensor("aw_chunk_915_cast_fp16")]; tensor var_9422_equation_0 = const()[name = tensor("op_9422_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9422_cast_fp16 = einsum(equation = var_9422_equation_0, values = (var_9248_cast_fp16, var_9072_cast_fp16))[name = tensor("op_9422_cast_fp16")]; tensor var_9423_to_fp16 = const()[name = tensor("op_9423_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_917_cast_fp16 = mul(x = var_9422_cast_fp16, y = var_9423_to_fp16)[name = tensor("aw_chunk_917_cast_fp16")]; tensor var_9426_equation_0 = const()[name = tensor("op_9426_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9426_cast_fp16 = einsum(equation = var_9426_equation_0, values = (var_9248_cast_fp16, var_9079_cast_fp16))[name = tensor("op_9426_cast_fp16")]; tensor var_9427_to_fp16 = const()[name = tensor("op_9427_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_919_cast_fp16 = mul(x = var_9426_cast_fp16, y = var_9427_to_fp16)[name = tensor("aw_chunk_919_cast_fp16")]; tensor var_9430_equation_0 = const()[name = tensor("op_9430_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9430_cast_fp16 = einsum(equation = var_9430_equation_0, values = (var_9252_cast_fp16, var_9086_cast_fp16))[name = tensor("op_9430_cast_fp16")]; tensor var_9431_to_fp16 = const()[name = tensor("op_9431_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_921_cast_fp16 = mul(x = var_9430_cast_fp16, y = var_9431_to_fp16)[name = tensor("aw_chunk_921_cast_fp16")]; tensor var_9434_equation_0 = const()[name = tensor("op_9434_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9434_cast_fp16 = einsum(equation = var_9434_equation_0, values = (var_9252_cast_fp16, var_9093_cast_fp16))[name = tensor("op_9434_cast_fp16")]; tensor var_9435_to_fp16 = const()[name = tensor("op_9435_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_923_cast_fp16 = mul(x = var_9434_cast_fp16, y = var_9435_to_fp16)[name = tensor("aw_chunk_923_cast_fp16")]; tensor var_9438_equation_0 = const()[name = tensor("op_9438_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9438_cast_fp16 = einsum(equation = var_9438_equation_0, values = (var_9252_cast_fp16, var_9100_cast_fp16))[name = tensor("op_9438_cast_fp16")]; tensor var_9439_to_fp16 = const()[name = tensor("op_9439_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_925_cast_fp16 = mul(x = var_9438_cast_fp16, y = var_9439_to_fp16)[name = tensor("aw_chunk_925_cast_fp16")]; tensor var_9442_equation_0 = const()[name = tensor("op_9442_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9442_cast_fp16 = einsum(equation = var_9442_equation_0, values = (var_9252_cast_fp16, var_9107_cast_fp16))[name = tensor("op_9442_cast_fp16")]; tensor var_9443_to_fp16 = const()[name = tensor("op_9443_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_927_cast_fp16 = mul(x = var_9442_cast_fp16, y = var_9443_to_fp16)[name = tensor("aw_chunk_927_cast_fp16")]; tensor var_9446_equation_0 = const()[name = tensor("op_9446_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9446_cast_fp16 = einsum(equation = var_9446_equation_0, values = (var_9256_cast_fp16, var_9114_cast_fp16))[name = tensor("op_9446_cast_fp16")]; tensor var_9447_to_fp16 = const()[name = tensor("op_9447_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_929_cast_fp16 = mul(x = var_9446_cast_fp16, y = var_9447_to_fp16)[name = tensor("aw_chunk_929_cast_fp16")]; tensor var_9450_equation_0 = const()[name = tensor("op_9450_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9450_cast_fp16 = einsum(equation = var_9450_equation_0, values = (var_9256_cast_fp16, var_9121_cast_fp16))[name = tensor("op_9450_cast_fp16")]; tensor var_9451_to_fp16 = const()[name = tensor("op_9451_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_931_cast_fp16 = mul(x = var_9450_cast_fp16, y = var_9451_to_fp16)[name = tensor("aw_chunk_931_cast_fp16")]; tensor var_9454_equation_0 = const()[name = tensor("op_9454_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9454_cast_fp16 = einsum(equation = var_9454_equation_0, values = (var_9256_cast_fp16, var_9128_cast_fp16))[name = tensor("op_9454_cast_fp16")]; tensor var_9455_to_fp16 = const()[name = tensor("op_9455_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_933_cast_fp16 = mul(x = var_9454_cast_fp16, y = var_9455_to_fp16)[name = tensor("aw_chunk_933_cast_fp16")]; tensor var_9458_equation_0 = const()[name = tensor("op_9458_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9458_cast_fp16 = einsum(equation = var_9458_equation_0, values = (var_9256_cast_fp16, var_9135_cast_fp16))[name = tensor("op_9458_cast_fp16")]; tensor var_9459_to_fp16 = const()[name = tensor("op_9459_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_935_cast_fp16 = mul(x = var_9458_cast_fp16, y = var_9459_to_fp16)[name = tensor("aw_chunk_935_cast_fp16")]; tensor var_9462_equation_0 = const()[name = tensor("op_9462_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9462_cast_fp16 = einsum(equation = var_9462_equation_0, values = (var_9260_cast_fp16, var_9142_cast_fp16))[name = tensor("op_9462_cast_fp16")]; tensor var_9463_to_fp16 = const()[name = tensor("op_9463_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_937_cast_fp16 = mul(x = var_9462_cast_fp16, y = var_9463_to_fp16)[name = tensor("aw_chunk_937_cast_fp16")]; tensor var_9466_equation_0 = const()[name = tensor("op_9466_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9466_cast_fp16 = einsum(equation = var_9466_equation_0, values = (var_9260_cast_fp16, var_9149_cast_fp16))[name = tensor("op_9466_cast_fp16")]; tensor var_9467_to_fp16 = const()[name = tensor("op_9467_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_939_cast_fp16 = mul(x = var_9466_cast_fp16, y = var_9467_to_fp16)[name = tensor("aw_chunk_939_cast_fp16")]; tensor var_9470_equation_0 = const()[name = tensor("op_9470_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9470_cast_fp16 = einsum(equation = var_9470_equation_0, values = (var_9260_cast_fp16, var_9156_cast_fp16))[name = tensor("op_9470_cast_fp16")]; tensor var_9471_to_fp16 = const()[name = tensor("op_9471_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_941_cast_fp16 = mul(x = var_9470_cast_fp16, y = var_9471_to_fp16)[name = tensor("aw_chunk_941_cast_fp16")]; tensor var_9474_equation_0 = const()[name = tensor("op_9474_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9474_cast_fp16 = einsum(equation = var_9474_equation_0, values = (var_9260_cast_fp16, var_9163_cast_fp16))[name = tensor("op_9474_cast_fp16")]; tensor var_9475_to_fp16 = const()[name = tensor("op_9475_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_943_cast_fp16 = mul(x = var_9474_cast_fp16, y = var_9475_to_fp16)[name = tensor("aw_chunk_943_cast_fp16")]; tensor var_9478_equation_0 = const()[name = tensor("op_9478_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9478_cast_fp16 = einsum(equation = var_9478_equation_0, values = (var_9264_cast_fp16, var_9170_cast_fp16))[name = tensor("op_9478_cast_fp16")]; tensor var_9479_to_fp16 = const()[name = tensor("op_9479_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_945_cast_fp16 = mul(x = var_9478_cast_fp16, y = var_9479_to_fp16)[name = tensor("aw_chunk_945_cast_fp16")]; tensor var_9482_equation_0 = const()[name = tensor("op_9482_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9482_cast_fp16 = einsum(equation = var_9482_equation_0, values = (var_9264_cast_fp16, var_9177_cast_fp16))[name = tensor("op_9482_cast_fp16")]; tensor var_9483_to_fp16 = const()[name = tensor("op_9483_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_947_cast_fp16 = mul(x = var_9482_cast_fp16, y = var_9483_to_fp16)[name = tensor("aw_chunk_947_cast_fp16")]; tensor var_9486_equation_0 = const()[name = tensor("op_9486_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9486_cast_fp16 = einsum(equation = var_9486_equation_0, values = (var_9264_cast_fp16, var_9184_cast_fp16))[name = tensor("op_9486_cast_fp16")]; tensor var_9487_to_fp16 = const()[name = tensor("op_9487_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_949_cast_fp16 = mul(x = var_9486_cast_fp16, y = var_9487_to_fp16)[name = tensor("aw_chunk_949_cast_fp16")]; tensor var_9490_equation_0 = const()[name = tensor("op_9490_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9490_cast_fp16 = einsum(equation = var_9490_equation_0, values = (var_9264_cast_fp16, var_9191_cast_fp16))[name = tensor("op_9490_cast_fp16")]; tensor var_9491_to_fp16 = const()[name = tensor("op_9491_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_951_cast_fp16 = mul(x = var_9490_cast_fp16, y = var_9491_to_fp16)[name = tensor("aw_chunk_951_cast_fp16")]; tensor var_9494_equation_0 = const()[name = tensor("op_9494_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9494_cast_fp16 = einsum(equation = var_9494_equation_0, values = (var_9268_cast_fp16, var_9198_cast_fp16))[name = tensor("op_9494_cast_fp16")]; tensor var_9495_to_fp16 = const()[name = tensor("op_9495_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_953_cast_fp16 = mul(x = var_9494_cast_fp16, y = var_9495_to_fp16)[name = tensor("aw_chunk_953_cast_fp16")]; tensor var_9498_equation_0 = const()[name = tensor("op_9498_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9498_cast_fp16 = einsum(equation = var_9498_equation_0, values = (var_9268_cast_fp16, var_9205_cast_fp16))[name = tensor("op_9498_cast_fp16")]; tensor var_9499_to_fp16 = const()[name = tensor("op_9499_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_955_cast_fp16 = mul(x = var_9498_cast_fp16, y = var_9499_to_fp16)[name = tensor("aw_chunk_955_cast_fp16")]; tensor var_9502_equation_0 = const()[name = tensor("op_9502_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9502_cast_fp16 = einsum(equation = var_9502_equation_0, values = (var_9268_cast_fp16, var_9212_cast_fp16))[name = tensor("op_9502_cast_fp16")]; tensor var_9503_to_fp16 = const()[name = tensor("op_9503_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_957_cast_fp16 = mul(x = var_9502_cast_fp16, y = var_9503_to_fp16)[name = tensor("aw_chunk_957_cast_fp16")]; tensor var_9506_equation_0 = const()[name = tensor("op_9506_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_9506_cast_fp16 = einsum(equation = var_9506_equation_0, values = (var_9268_cast_fp16, var_9219_cast_fp16))[name = tensor("op_9506_cast_fp16")]; tensor var_9507_to_fp16 = const()[name = tensor("op_9507_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_959_cast_fp16 = mul(x = var_9506_cast_fp16, y = var_9507_to_fp16)[name = tensor("aw_chunk_959_cast_fp16")]; tensor var_9509_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_865_cast_fp16)[name = tensor("op_9509_cast_fp16")]; tensor var_9510_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_867_cast_fp16)[name = tensor("op_9510_cast_fp16")]; tensor var_9511_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_869_cast_fp16)[name = tensor("op_9511_cast_fp16")]; tensor var_9512_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_871_cast_fp16)[name = tensor("op_9512_cast_fp16")]; tensor var_9513_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_873_cast_fp16)[name = tensor("op_9513_cast_fp16")]; tensor var_9514_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_875_cast_fp16)[name = tensor("op_9514_cast_fp16")]; tensor var_9515_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_877_cast_fp16)[name = tensor("op_9515_cast_fp16")]; tensor var_9516_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_879_cast_fp16)[name = tensor("op_9516_cast_fp16")]; tensor var_9517_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_881_cast_fp16)[name = tensor("op_9517_cast_fp16")]; tensor var_9518_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_883_cast_fp16)[name = tensor("op_9518_cast_fp16")]; tensor var_9519_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_885_cast_fp16)[name = tensor("op_9519_cast_fp16")]; tensor var_9520_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_887_cast_fp16)[name = tensor("op_9520_cast_fp16")]; tensor var_9521_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_889_cast_fp16)[name = tensor("op_9521_cast_fp16")]; tensor var_9522_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_891_cast_fp16)[name = tensor("op_9522_cast_fp16")]; tensor var_9523_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_893_cast_fp16)[name = tensor("op_9523_cast_fp16")]; tensor var_9524_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_895_cast_fp16)[name = tensor("op_9524_cast_fp16")]; tensor var_9525_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_897_cast_fp16)[name = tensor("op_9525_cast_fp16")]; tensor var_9526_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_899_cast_fp16)[name = tensor("op_9526_cast_fp16")]; tensor var_9527_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_901_cast_fp16)[name = tensor("op_9527_cast_fp16")]; tensor var_9528_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_903_cast_fp16)[name = tensor("op_9528_cast_fp16")]; tensor var_9529_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_905_cast_fp16)[name = tensor("op_9529_cast_fp16")]; tensor var_9530_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_907_cast_fp16)[name = tensor("op_9530_cast_fp16")]; tensor var_9531_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_909_cast_fp16)[name = tensor("op_9531_cast_fp16")]; tensor var_9532_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_911_cast_fp16)[name = tensor("op_9532_cast_fp16")]; tensor var_9533_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_913_cast_fp16)[name = tensor("op_9533_cast_fp16")]; tensor var_9534_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_915_cast_fp16)[name = tensor("op_9534_cast_fp16")]; tensor var_9535_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_917_cast_fp16)[name = tensor("op_9535_cast_fp16")]; tensor var_9536_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_919_cast_fp16)[name = tensor("op_9536_cast_fp16")]; tensor var_9537_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_921_cast_fp16)[name = tensor("op_9537_cast_fp16")]; tensor var_9538_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_923_cast_fp16)[name = tensor("op_9538_cast_fp16")]; tensor var_9539_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_925_cast_fp16)[name = tensor("op_9539_cast_fp16")]; tensor var_9540_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_927_cast_fp16)[name = tensor("op_9540_cast_fp16")]; tensor var_9541_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_929_cast_fp16)[name = tensor("op_9541_cast_fp16")]; tensor var_9542_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_931_cast_fp16)[name = tensor("op_9542_cast_fp16")]; tensor var_9543_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_933_cast_fp16)[name = tensor("op_9543_cast_fp16")]; tensor var_9544_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_935_cast_fp16)[name = tensor("op_9544_cast_fp16")]; tensor var_9545_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_937_cast_fp16)[name = tensor("op_9545_cast_fp16")]; tensor var_9546_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_939_cast_fp16)[name = tensor("op_9546_cast_fp16")]; tensor var_9547_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_941_cast_fp16)[name = tensor("op_9547_cast_fp16")]; tensor var_9548_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_943_cast_fp16)[name = tensor("op_9548_cast_fp16")]; tensor var_9549_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_945_cast_fp16)[name = tensor("op_9549_cast_fp16")]; tensor var_9550_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_947_cast_fp16)[name = tensor("op_9550_cast_fp16")]; tensor var_9551_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_949_cast_fp16)[name = tensor("op_9551_cast_fp16")]; tensor var_9552_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_951_cast_fp16)[name = tensor("op_9552_cast_fp16")]; tensor var_9553_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_953_cast_fp16)[name = tensor("op_9553_cast_fp16")]; tensor var_9554_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_955_cast_fp16)[name = tensor("op_9554_cast_fp16")]; tensor var_9555_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_957_cast_fp16)[name = tensor("op_9555_cast_fp16")]; tensor var_9556_cast_fp16 = softmax(axis = var_8782, x = aw_chunk_959_cast_fp16)[name = tensor("op_9556_cast_fp16")]; tensor var_9558_equation_0 = const()[name = tensor("op_9558_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9558_cast_fp16 = einsum(equation = var_9558_equation_0, values = (var_9270_cast_fp16, var_9509_cast_fp16))[name = tensor("op_9558_cast_fp16")]; tensor var_9560_equation_0 = const()[name = tensor("op_9560_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9560_cast_fp16 = einsum(equation = var_9560_equation_0, values = (var_9270_cast_fp16, var_9510_cast_fp16))[name = tensor("op_9560_cast_fp16")]; tensor var_9562_equation_0 = const()[name = tensor("op_9562_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9562_cast_fp16 = einsum(equation = var_9562_equation_0, values = (var_9270_cast_fp16, var_9511_cast_fp16))[name = tensor("op_9562_cast_fp16")]; tensor var_9564_equation_0 = const()[name = tensor("op_9564_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9564_cast_fp16 = einsum(equation = var_9564_equation_0, values = (var_9270_cast_fp16, var_9512_cast_fp16))[name = tensor("op_9564_cast_fp16")]; tensor var_9566_equation_0 = const()[name = tensor("op_9566_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9566_cast_fp16 = einsum(equation = var_9566_equation_0, values = (var_9274_cast_fp16, var_9513_cast_fp16))[name = tensor("op_9566_cast_fp16")]; tensor var_9568_equation_0 = const()[name = tensor("op_9568_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9568_cast_fp16 = einsum(equation = var_9568_equation_0, values = (var_9274_cast_fp16, var_9514_cast_fp16))[name = tensor("op_9568_cast_fp16")]; tensor var_9570_equation_0 = const()[name = tensor("op_9570_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9570_cast_fp16 = einsum(equation = var_9570_equation_0, values = (var_9274_cast_fp16, var_9515_cast_fp16))[name = tensor("op_9570_cast_fp16")]; tensor var_9572_equation_0 = const()[name = tensor("op_9572_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9572_cast_fp16 = einsum(equation = var_9572_equation_0, values = (var_9274_cast_fp16, var_9516_cast_fp16))[name = tensor("op_9572_cast_fp16")]; tensor var_9574_equation_0 = const()[name = tensor("op_9574_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9574_cast_fp16 = einsum(equation = var_9574_equation_0, values = (var_9278_cast_fp16, var_9517_cast_fp16))[name = tensor("op_9574_cast_fp16")]; tensor var_9576_equation_0 = const()[name = tensor("op_9576_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9576_cast_fp16 = einsum(equation = var_9576_equation_0, values = (var_9278_cast_fp16, var_9518_cast_fp16))[name = tensor("op_9576_cast_fp16")]; tensor var_9578_equation_0 = const()[name = tensor("op_9578_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9578_cast_fp16 = einsum(equation = var_9578_equation_0, values = (var_9278_cast_fp16, var_9519_cast_fp16))[name = tensor("op_9578_cast_fp16")]; tensor var_9580_equation_0 = const()[name = tensor("op_9580_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9580_cast_fp16 = einsum(equation = var_9580_equation_0, values = (var_9278_cast_fp16, var_9520_cast_fp16))[name = tensor("op_9580_cast_fp16")]; tensor var_9582_equation_0 = const()[name = tensor("op_9582_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9582_cast_fp16 = einsum(equation = var_9582_equation_0, values = (var_9282_cast_fp16, var_9521_cast_fp16))[name = tensor("op_9582_cast_fp16")]; tensor var_9584_equation_0 = const()[name = tensor("op_9584_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9584_cast_fp16 = einsum(equation = var_9584_equation_0, values = (var_9282_cast_fp16, var_9522_cast_fp16))[name = tensor("op_9584_cast_fp16")]; tensor var_9586_equation_0 = const()[name = tensor("op_9586_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9586_cast_fp16 = einsum(equation = var_9586_equation_0, values = (var_9282_cast_fp16, var_9523_cast_fp16))[name = tensor("op_9586_cast_fp16")]; tensor var_9588_equation_0 = const()[name = tensor("op_9588_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9588_cast_fp16 = einsum(equation = var_9588_equation_0, values = (var_9282_cast_fp16, var_9524_cast_fp16))[name = tensor("op_9588_cast_fp16")]; tensor var_9590_equation_0 = const()[name = tensor("op_9590_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9590_cast_fp16 = einsum(equation = var_9590_equation_0, values = (var_9286_cast_fp16, var_9525_cast_fp16))[name = tensor("op_9590_cast_fp16")]; tensor var_9592_equation_0 = const()[name = tensor("op_9592_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9592_cast_fp16 = einsum(equation = var_9592_equation_0, values = (var_9286_cast_fp16, var_9526_cast_fp16))[name = tensor("op_9592_cast_fp16")]; tensor var_9594_equation_0 = const()[name = tensor("op_9594_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9594_cast_fp16 = einsum(equation = var_9594_equation_0, values = (var_9286_cast_fp16, var_9527_cast_fp16))[name = tensor("op_9594_cast_fp16")]; tensor var_9596_equation_0 = const()[name = tensor("op_9596_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9596_cast_fp16 = einsum(equation = var_9596_equation_0, values = (var_9286_cast_fp16, var_9528_cast_fp16))[name = tensor("op_9596_cast_fp16")]; tensor var_9598_equation_0 = const()[name = tensor("op_9598_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9598_cast_fp16 = einsum(equation = var_9598_equation_0, values = (var_9290_cast_fp16, var_9529_cast_fp16))[name = tensor("op_9598_cast_fp16")]; tensor var_9600_equation_0 = const()[name = tensor("op_9600_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9600_cast_fp16 = einsum(equation = var_9600_equation_0, values = (var_9290_cast_fp16, var_9530_cast_fp16))[name = tensor("op_9600_cast_fp16")]; tensor var_9602_equation_0 = const()[name = tensor("op_9602_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9602_cast_fp16 = einsum(equation = var_9602_equation_0, values = (var_9290_cast_fp16, var_9531_cast_fp16))[name = tensor("op_9602_cast_fp16")]; tensor var_9604_equation_0 = const()[name = tensor("op_9604_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9604_cast_fp16 = einsum(equation = var_9604_equation_0, values = (var_9290_cast_fp16, var_9532_cast_fp16))[name = tensor("op_9604_cast_fp16")]; tensor var_9606_equation_0 = const()[name = tensor("op_9606_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9606_cast_fp16 = einsum(equation = var_9606_equation_0, values = (var_9294_cast_fp16, var_9533_cast_fp16))[name = tensor("op_9606_cast_fp16")]; tensor var_9608_equation_0 = const()[name = tensor("op_9608_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9608_cast_fp16 = einsum(equation = var_9608_equation_0, values = (var_9294_cast_fp16, var_9534_cast_fp16))[name = tensor("op_9608_cast_fp16")]; tensor var_9610_equation_0 = const()[name = tensor("op_9610_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9610_cast_fp16 = einsum(equation = var_9610_equation_0, values = (var_9294_cast_fp16, var_9535_cast_fp16))[name = tensor("op_9610_cast_fp16")]; tensor var_9612_equation_0 = const()[name = tensor("op_9612_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9612_cast_fp16 = einsum(equation = var_9612_equation_0, values = (var_9294_cast_fp16, var_9536_cast_fp16))[name = tensor("op_9612_cast_fp16")]; tensor var_9614_equation_0 = const()[name = tensor("op_9614_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9614_cast_fp16 = einsum(equation = var_9614_equation_0, values = (var_9298_cast_fp16, var_9537_cast_fp16))[name = tensor("op_9614_cast_fp16")]; tensor var_9616_equation_0 = const()[name = tensor("op_9616_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9616_cast_fp16 = einsum(equation = var_9616_equation_0, values = (var_9298_cast_fp16, var_9538_cast_fp16))[name = tensor("op_9616_cast_fp16")]; tensor var_9618_equation_0 = const()[name = tensor("op_9618_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9618_cast_fp16 = einsum(equation = var_9618_equation_0, values = (var_9298_cast_fp16, var_9539_cast_fp16))[name = tensor("op_9618_cast_fp16")]; tensor var_9620_equation_0 = const()[name = tensor("op_9620_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9620_cast_fp16 = einsum(equation = var_9620_equation_0, values = (var_9298_cast_fp16, var_9540_cast_fp16))[name = tensor("op_9620_cast_fp16")]; tensor var_9622_equation_0 = const()[name = tensor("op_9622_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9622_cast_fp16 = einsum(equation = var_9622_equation_0, values = (var_9302_cast_fp16, var_9541_cast_fp16))[name = tensor("op_9622_cast_fp16")]; tensor var_9624_equation_0 = const()[name = tensor("op_9624_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9624_cast_fp16 = einsum(equation = var_9624_equation_0, values = (var_9302_cast_fp16, var_9542_cast_fp16))[name = tensor("op_9624_cast_fp16")]; tensor var_9626_equation_0 = const()[name = tensor("op_9626_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9626_cast_fp16 = einsum(equation = var_9626_equation_0, values = (var_9302_cast_fp16, var_9543_cast_fp16))[name = tensor("op_9626_cast_fp16")]; tensor var_9628_equation_0 = const()[name = tensor("op_9628_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9628_cast_fp16 = einsum(equation = var_9628_equation_0, values = (var_9302_cast_fp16, var_9544_cast_fp16))[name = tensor("op_9628_cast_fp16")]; tensor var_9630_equation_0 = const()[name = tensor("op_9630_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9630_cast_fp16 = einsum(equation = var_9630_equation_0, values = (var_9306_cast_fp16, var_9545_cast_fp16))[name = tensor("op_9630_cast_fp16")]; tensor var_9632_equation_0 = const()[name = tensor("op_9632_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9632_cast_fp16 = einsum(equation = var_9632_equation_0, values = (var_9306_cast_fp16, var_9546_cast_fp16))[name = tensor("op_9632_cast_fp16")]; tensor var_9634_equation_0 = const()[name = tensor("op_9634_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9634_cast_fp16 = einsum(equation = var_9634_equation_0, values = (var_9306_cast_fp16, var_9547_cast_fp16))[name = tensor("op_9634_cast_fp16")]; tensor var_9636_equation_0 = const()[name = tensor("op_9636_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9636_cast_fp16 = einsum(equation = var_9636_equation_0, values = (var_9306_cast_fp16, var_9548_cast_fp16))[name = tensor("op_9636_cast_fp16")]; tensor var_9638_equation_0 = const()[name = tensor("op_9638_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9638_cast_fp16 = einsum(equation = var_9638_equation_0, values = (var_9310_cast_fp16, var_9549_cast_fp16))[name = tensor("op_9638_cast_fp16")]; tensor var_9640_equation_0 = const()[name = tensor("op_9640_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9640_cast_fp16 = einsum(equation = var_9640_equation_0, values = (var_9310_cast_fp16, var_9550_cast_fp16))[name = tensor("op_9640_cast_fp16")]; tensor var_9642_equation_0 = const()[name = tensor("op_9642_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9642_cast_fp16 = einsum(equation = var_9642_equation_0, values = (var_9310_cast_fp16, var_9551_cast_fp16))[name = tensor("op_9642_cast_fp16")]; tensor var_9644_equation_0 = const()[name = tensor("op_9644_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9644_cast_fp16 = einsum(equation = var_9644_equation_0, values = (var_9310_cast_fp16, var_9552_cast_fp16))[name = tensor("op_9644_cast_fp16")]; tensor var_9646_equation_0 = const()[name = tensor("op_9646_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9646_cast_fp16 = einsum(equation = var_9646_equation_0, values = (var_9314_cast_fp16, var_9553_cast_fp16))[name = tensor("op_9646_cast_fp16")]; tensor var_9648_equation_0 = const()[name = tensor("op_9648_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9648_cast_fp16 = einsum(equation = var_9648_equation_0, values = (var_9314_cast_fp16, var_9554_cast_fp16))[name = tensor("op_9648_cast_fp16")]; tensor var_9650_equation_0 = const()[name = tensor("op_9650_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9650_cast_fp16 = einsum(equation = var_9650_equation_0, values = (var_9314_cast_fp16, var_9555_cast_fp16))[name = tensor("op_9650_cast_fp16")]; tensor var_9652_equation_0 = const()[name = tensor("op_9652_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_9652_cast_fp16 = einsum(equation = var_9652_equation_0, values = (var_9314_cast_fp16, var_9556_cast_fp16))[name = tensor("op_9652_cast_fp16")]; tensor var_9654_interleave_0 = const()[name = tensor("op_9654_interleave_0"), val = tensor(false)]; tensor var_9654_cast_fp16 = concat(axis = var_8765, interleave = var_9654_interleave_0, values = (var_9558_cast_fp16, var_9560_cast_fp16, var_9562_cast_fp16, var_9564_cast_fp16))[name = tensor("op_9654_cast_fp16")]; tensor var_9656_interleave_0 = const()[name = tensor("op_9656_interleave_0"), val = tensor(false)]; tensor var_9656_cast_fp16 = concat(axis = var_8765, interleave = var_9656_interleave_0, values = (var_9566_cast_fp16, var_9568_cast_fp16, var_9570_cast_fp16, var_9572_cast_fp16))[name = tensor("op_9656_cast_fp16")]; tensor var_9658_interleave_0 = const()[name = tensor("op_9658_interleave_0"), val = tensor(false)]; tensor var_9658_cast_fp16 = concat(axis = var_8765, interleave = var_9658_interleave_0, values = (var_9574_cast_fp16, var_9576_cast_fp16, var_9578_cast_fp16, var_9580_cast_fp16))[name = tensor("op_9658_cast_fp16")]; tensor var_9660_interleave_0 = const()[name = tensor("op_9660_interleave_0"), val = tensor(false)]; tensor var_9660_cast_fp16 = concat(axis = var_8765, interleave = var_9660_interleave_0, values = (var_9582_cast_fp16, var_9584_cast_fp16, var_9586_cast_fp16, var_9588_cast_fp16))[name = tensor("op_9660_cast_fp16")]; tensor var_9662_interleave_0 = const()[name = tensor("op_9662_interleave_0"), val = tensor(false)]; tensor var_9662_cast_fp16 = concat(axis = var_8765, interleave = var_9662_interleave_0, values = (var_9590_cast_fp16, var_9592_cast_fp16, var_9594_cast_fp16, var_9596_cast_fp16))[name = tensor("op_9662_cast_fp16")]; tensor var_9664_interleave_0 = const()[name = tensor("op_9664_interleave_0"), val = tensor(false)]; tensor var_9664_cast_fp16 = concat(axis = var_8765, interleave = var_9664_interleave_0, values = (var_9598_cast_fp16, var_9600_cast_fp16, var_9602_cast_fp16, var_9604_cast_fp16))[name = tensor("op_9664_cast_fp16")]; tensor var_9666_interleave_0 = const()[name = tensor("op_9666_interleave_0"), val = tensor(false)]; tensor var_9666_cast_fp16 = concat(axis = var_8765, interleave = var_9666_interleave_0, values = (var_9606_cast_fp16, var_9608_cast_fp16, var_9610_cast_fp16, var_9612_cast_fp16))[name = tensor("op_9666_cast_fp16")]; tensor var_9668_interleave_0 = const()[name = tensor("op_9668_interleave_0"), val = tensor(false)]; tensor var_9668_cast_fp16 = concat(axis = var_8765, interleave = var_9668_interleave_0, values = (var_9614_cast_fp16, var_9616_cast_fp16, var_9618_cast_fp16, var_9620_cast_fp16))[name = tensor("op_9668_cast_fp16")]; tensor var_9670_interleave_0 = const()[name = tensor("op_9670_interleave_0"), val = tensor(false)]; tensor var_9670_cast_fp16 = concat(axis = var_8765, interleave = var_9670_interleave_0, values = (var_9622_cast_fp16, var_9624_cast_fp16, var_9626_cast_fp16, var_9628_cast_fp16))[name = tensor("op_9670_cast_fp16")]; tensor var_9672_interleave_0 = const()[name = tensor("op_9672_interleave_0"), val = tensor(false)]; tensor var_9672_cast_fp16 = concat(axis = var_8765, interleave = var_9672_interleave_0, values = (var_9630_cast_fp16, var_9632_cast_fp16, var_9634_cast_fp16, var_9636_cast_fp16))[name = tensor("op_9672_cast_fp16")]; tensor var_9674_interleave_0 = const()[name = tensor("op_9674_interleave_0"), val = tensor(false)]; tensor var_9674_cast_fp16 = concat(axis = var_8765, interleave = var_9674_interleave_0, values = (var_9638_cast_fp16, var_9640_cast_fp16, var_9642_cast_fp16, var_9644_cast_fp16))[name = tensor("op_9674_cast_fp16")]; tensor var_9676_interleave_0 = const()[name = tensor("op_9676_interleave_0"), val = tensor(false)]; tensor var_9676_cast_fp16 = concat(axis = var_8765, interleave = var_9676_interleave_0, values = (var_9646_cast_fp16, var_9648_cast_fp16, var_9650_cast_fp16, var_9652_cast_fp16))[name = tensor("op_9676_cast_fp16")]; tensor input_73_interleave_0 = const()[name = tensor("input_73_interleave_0"), val = tensor(false)]; tensor input_73_cast_fp16 = concat(axis = var_8782, interleave = input_73_interleave_0, values = (var_9654_cast_fp16, var_9656_cast_fp16, var_9658_cast_fp16, var_9660_cast_fp16, var_9662_cast_fp16, var_9664_cast_fp16, var_9666_cast_fp16, var_9668_cast_fp16, var_9670_cast_fp16, var_9672_cast_fp16, var_9674_cast_fp16, var_9676_cast_fp16))[name = tensor("input_73_cast_fp16")]; tensor var_9681 = const()[name = tensor("op_9681"), val = tensor([1, 1])]; tensor var_9683 = const()[name = tensor("op_9683"), val = tensor([1, 1])]; tensor obj_39_pad_type_0 = const()[name = tensor("obj_39_pad_type_0"), val = tensor("custom")]; tensor obj_39_pad_0 = const()[name = tensor("obj_39_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137340288)))]; tensor layers_9_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138520000)))]; tensor obj_39_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_bias_to_fp16, dilations = var_9683, groups = var_8782, pad = obj_39_pad_0, pad_type = obj_39_pad_type_0, strides = var_9681, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_73_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_9689 = const()[name = tensor("op_9689"), val = tensor([1])]; tensor channels_mean_39_cast_fp16 = reduce_mean(axes = var_9689, keep_dims = var_8783, 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_9693 = const()[name = tensor("op_9693"), val = tensor([1])]; tensor var_9694_cast_fp16 = reduce_mean(axes = var_9693, keep_dims = var_8783, x = zero_mean_sq_39_cast_fp16)[name = tensor("op_9694_cast_fp16")]; tensor var_9695_to_fp16 = const()[name = tensor("op_9695_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9696_cast_fp16 = add(x = var_9694_cast_fp16, y = var_9695_to_fp16)[name = tensor("op_9696_cast_fp16")]; tensor denom_39_epsilon_0_to_fp16 = const()[name = tensor("denom_39_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_39_cast_fp16 = rsqrt(epsilon = denom_39_epsilon_0_to_fp16, x = var_9696_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_75_gamma_0_to_fp16 = const()[name = tensor("input_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138521600)))]; tensor input_75_beta_0_to_fp16 = const()[name = tensor("input_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138523200)))]; tensor input_75_epsilon_0_to_fp16 = const()[name = tensor("input_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_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_75_cast_fp16")]; tensor var_9707 = const()[name = tensor("op_9707"), val = tensor([1, 1])]; tensor var_9709 = const()[name = tensor("op_9709"), val = tensor([1, 1])]; tensor input_77_pad_type_0 = const()[name = tensor("input_77_pad_type_0"), val = tensor("custom")]; tensor input_77_pad_0 = const()[name = tensor("input_77_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_fc1_weight_to_fp16 = const()[name = tensor("layers_9_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138524800)))]; tensor layers_9_fc1_bias_to_fp16 = const()[name = tensor("layers_9_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143243456)))]; tensor input_77_cast_fp16 = conv(bias = layers_9_fc1_bias_to_fp16, dilations = var_9709, groups = var_8782, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = var_9707, weight = layers_9_fc1_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; tensor input_79_mode_0 = const()[name = tensor("input_79_mode_0"), val = tensor("EXACT")]; tensor input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor("input_79_cast_fp16")]; tensor var_9715 = const()[name = tensor("op_9715"), val = tensor([1, 1])]; tensor var_9717 = const()[name = tensor("op_9717"), val = tensor([1, 1])]; tensor hidden_states_23_pad_type_0 = const()[name = tensor("hidden_states_23_pad_type_0"), val = tensor("custom")]; tensor hidden_states_23_pad_0 = const()[name = tensor("hidden_states_23_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_9_fc2_weight_to_fp16 = const()[name = tensor("layers_9_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143249664)))]; tensor layers_9_fc2_bias_to_fp16 = const()[name = tensor("layers_9_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147968320)))]; tensor hidden_states_23_cast_fp16 = conv(bias = layers_9_fc2_bias_to_fp16, dilations = var_9717, groups = var_8782, pad = hidden_states_23_pad_0, pad_type = hidden_states_23_pad_type_0, strides = var_9715, weight = layers_9_fc2_weight_to_fp16, x = input_79_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_9724 = const()[name = tensor("op_9724"), val = tensor(3)]; tensor var_9741 = const()[name = tensor("op_9741"), val = tensor(1)]; tensor var_9742 = const()[name = tensor("op_9742"), val = tensor(true)]; tensor var_9752 = const()[name = tensor("op_9752"), val = tensor([1])]; tensor channels_mean_41_cast_fp16 = reduce_mean(axes = var_9752, keep_dims = var_9742, 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_9756 = const()[name = tensor("op_9756"), val = tensor([1])]; tensor var_9757_cast_fp16 = reduce_mean(axes = var_9756, keep_dims = var_9742, x = zero_mean_sq_41_cast_fp16)[name = tensor("op_9757_cast_fp16")]; tensor var_9758_to_fp16 = const()[name = tensor("op_9758_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_9759_cast_fp16 = add(x = var_9757_cast_fp16, y = var_9758_to_fp16)[name = tensor("op_9759_cast_fp16")]; tensor denom_41_epsilon_0_to_fp16 = const()[name = tensor("denom_41_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_41_cast_fp16 = rsqrt(epsilon = denom_41_epsilon_0_to_fp16, x = var_9759_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(147969920)))]; 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(147971520)))]; 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_9774 = const()[name = tensor("op_9774"), val = tensor([1, 1])]; tensor var_9776 = const()[name = tensor("op_9776"), val = tensor([1, 1])]; tensor query_21_pad_type_0 = const()[name = tensor("query_21_pad_type_0"), val = tensor("custom")]; tensor query_21_pad_0 = const()[name = tensor("query_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147973120)))]; tensor layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149152832)))]; tensor query_21_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_bias_to_fp16, dilations = var_9776, groups = var_9741, pad = query_21_pad_0, pad_type = query_21_pad_type_0, strides = var_9774, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("query_21_cast_fp16")]; tensor var_9780 = const()[name = tensor("op_9780"), val = tensor([1, 1])]; tensor var_9782 = const()[name = tensor("op_9782"), val = tensor([1, 1])]; tensor key_21_pad_type_0 = const()[name = tensor("key_21_pad_type_0"), val = tensor("custom")]; tensor key_21_pad_0 = const()[name = tensor("key_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149154432)))]; tensor key_21_cast_fp16 = conv(dilations = var_9782, groups = var_9741, pad = key_21_pad_0, pad_type = key_21_pad_type_0, strides = var_9780, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("key_21_cast_fp16")]; tensor var_9787 = const()[name = tensor("op_9787"), val = tensor([1, 1])]; tensor var_9789 = const()[name = tensor("op_9789"), val = tensor([1, 1])]; tensor value_21_pad_type_0 = const()[name = tensor("value_21_pad_type_0"), val = tensor("custom")]; tensor value_21_pad_0 = const()[name = tensor("value_21_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150334144)))]; tensor layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151513856)))]; tensor value_21_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_bias_to_fp16, dilations = var_9789, groups = var_9741, pad = value_21_pad_0, pad_type = value_21_pad_type_0, strides = var_9787, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = obj_41_cast_fp16)[name = tensor("value_21_cast_fp16")]; tensor var_9796_begin_0 = const()[name = tensor("op_9796_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9796_end_0 = const()[name = tensor("op_9796_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9796_end_mask_0 = const()[name = tensor("op_9796_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9796_cast_fp16 = slice_by_index(begin = var_9796_begin_0, end = var_9796_end_0, end_mask = var_9796_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9796_cast_fp16")]; tensor var_9800_begin_0 = const()[name = tensor("op_9800_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_9800_end_0 = const()[name = tensor("op_9800_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_9800_end_mask_0 = const()[name = tensor("op_9800_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9800_cast_fp16 = slice_by_index(begin = var_9800_begin_0, end = var_9800_end_0, end_mask = var_9800_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9800_cast_fp16")]; tensor var_9804_begin_0 = const()[name = tensor("op_9804_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_9804_end_0 = const()[name = tensor("op_9804_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_9804_end_mask_0 = const()[name = tensor("op_9804_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9804_cast_fp16 = slice_by_index(begin = var_9804_begin_0, end = var_9804_end_0, end_mask = var_9804_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9804_cast_fp16")]; tensor var_9808_begin_0 = const()[name = tensor("op_9808_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_9808_end_0 = const()[name = tensor("op_9808_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_9808_end_mask_0 = const()[name = tensor("op_9808_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9808_cast_fp16 = slice_by_index(begin = var_9808_begin_0, end = var_9808_end_0, end_mask = var_9808_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9808_cast_fp16")]; tensor var_9812_begin_0 = const()[name = tensor("op_9812_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_9812_end_0 = const()[name = tensor("op_9812_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_9812_end_mask_0 = const()[name = tensor("op_9812_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9812_cast_fp16 = slice_by_index(begin = var_9812_begin_0, end = var_9812_end_0, end_mask = var_9812_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9812_cast_fp16")]; tensor var_9816_begin_0 = const()[name = tensor("op_9816_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_9816_end_0 = const()[name = tensor("op_9816_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_9816_end_mask_0 = const()[name = tensor("op_9816_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9816_cast_fp16 = slice_by_index(begin = var_9816_begin_0, end = var_9816_end_0, end_mask = var_9816_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9816_cast_fp16")]; tensor var_9820_begin_0 = const()[name = tensor("op_9820_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_9820_end_0 = const()[name = tensor("op_9820_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_9820_end_mask_0 = const()[name = tensor("op_9820_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9820_cast_fp16 = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9820_cast_fp16")]; tensor var_9824_begin_0 = const()[name = tensor("op_9824_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_9824_end_0 = const()[name = tensor("op_9824_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_9824_end_mask_0 = const()[name = tensor("op_9824_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9824_cast_fp16 = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9824_cast_fp16")]; tensor var_9828_begin_0 = const()[name = tensor("op_9828_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_9828_end_0 = const()[name = tensor("op_9828_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_9828_end_mask_0 = const()[name = tensor("op_9828_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9828_cast_fp16 = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9828_cast_fp16")]; tensor var_9832_begin_0 = const()[name = tensor("op_9832_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_9832_end_0 = const()[name = tensor("op_9832_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_9832_end_mask_0 = const()[name = tensor("op_9832_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9832_cast_fp16 = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9832_cast_fp16")]; tensor var_9836_begin_0 = const()[name = tensor("op_9836_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_9836_end_0 = const()[name = tensor("op_9836_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_9836_end_mask_0 = const()[name = tensor("op_9836_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9836_cast_fp16 = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9836_cast_fp16")]; tensor var_9840_begin_0 = const()[name = tensor("op_9840_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_9840_end_0 = const()[name = tensor("op_9840_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_9840_end_mask_0 = const()[name = tensor("op_9840_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_9840_cast_fp16 = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = query_21_cast_fp16)[name = tensor("op_9840_cast_fp16")]; tensor var_9849_begin_0 = const()[name = tensor("op_9849_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9849_end_0 = const()[name = tensor("op_9849_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9849_end_mask_0 = const()[name = tensor("op_9849_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9849_cast_fp16 = slice_by_index(begin = var_9849_begin_0, end = var_9849_end_0, end_mask = var_9849_end_mask_0, x = var_9796_cast_fp16)[name = tensor("op_9849_cast_fp16")]; tensor var_9856_begin_0 = const()[name = tensor("op_9856_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9856_end_0 = const()[name = tensor("op_9856_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9856_end_mask_0 = const()[name = tensor("op_9856_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9856_cast_fp16 = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = var_9796_cast_fp16)[name = tensor("op_9856_cast_fp16")]; tensor var_9863_begin_0 = const()[name = tensor("op_9863_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9863_end_0 = const()[name = tensor("op_9863_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9863_end_mask_0 = const()[name = tensor("op_9863_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9863_cast_fp16 = slice_by_index(begin = var_9863_begin_0, end = var_9863_end_0, end_mask = var_9863_end_mask_0, x = var_9796_cast_fp16)[name = tensor("op_9863_cast_fp16")]; tensor var_9870_begin_0 = const()[name = tensor("op_9870_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9870_end_0 = const()[name = tensor("op_9870_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9870_end_mask_0 = const()[name = tensor("op_9870_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9870_cast_fp16 = slice_by_index(begin = var_9870_begin_0, end = var_9870_end_0, end_mask = var_9870_end_mask_0, x = var_9796_cast_fp16)[name = tensor("op_9870_cast_fp16")]; tensor var_9877_begin_0 = const()[name = tensor("op_9877_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9877_end_0 = const()[name = tensor("op_9877_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9877_end_mask_0 = const()[name = tensor("op_9877_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9877_cast_fp16 = slice_by_index(begin = var_9877_begin_0, end = var_9877_end_0, end_mask = var_9877_end_mask_0, x = var_9800_cast_fp16)[name = tensor("op_9877_cast_fp16")]; tensor var_9884_begin_0 = const()[name = tensor("op_9884_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9884_end_0 = const()[name = tensor("op_9884_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9884_end_mask_0 = const()[name = tensor("op_9884_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9884_cast_fp16 = slice_by_index(begin = var_9884_begin_0, end = var_9884_end_0, end_mask = var_9884_end_mask_0, x = var_9800_cast_fp16)[name = tensor("op_9884_cast_fp16")]; tensor var_9891_begin_0 = const()[name = tensor("op_9891_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9891_end_0 = const()[name = tensor("op_9891_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9891_end_mask_0 = const()[name = tensor("op_9891_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9891_cast_fp16 = slice_by_index(begin = var_9891_begin_0, end = var_9891_end_0, end_mask = var_9891_end_mask_0, x = var_9800_cast_fp16)[name = tensor("op_9891_cast_fp16")]; tensor var_9898_begin_0 = const()[name = tensor("op_9898_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9898_end_0 = const()[name = tensor("op_9898_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9898_end_mask_0 = const()[name = tensor("op_9898_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9898_cast_fp16 = slice_by_index(begin = var_9898_begin_0, end = var_9898_end_0, end_mask = var_9898_end_mask_0, x = var_9800_cast_fp16)[name = tensor("op_9898_cast_fp16")]; tensor var_9905_begin_0 = const()[name = tensor("op_9905_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9905_end_0 = const()[name = tensor("op_9905_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9905_end_mask_0 = const()[name = tensor("op_9905_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9905_cast_fp16 = slice_by_index(begin = var_9905_begin_0, end = var_9905_end_0, end_mask = var_9905_end_mask_0, x = var_9804_cast_fp16)[name = tensor("op_9905_cast_fp16")]; tensor var_9912_begin_0 = const()[name = tensor("op_9912_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9912_end_0 = const()[name = tensor("op_9912_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9912_end_mask_0 = const()[name = tensor("op_9912_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9912_cast_fp16 = slice_by_index(begin = var_9912_begin_0, end = var_9912_end_0, end_mask = var_9912_end_mask_0, x = var_9804_cast_fp16)[name = tensor("op_9912_cast_fp16")]; tensor var_9919_begin_0 = const()[name = tensor("op_9919_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9919_end_0 = const()[name = tensor("op_9919_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9919_end_mask_0 = const()[name = tensor("op_9919_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9919_cast_fp16 = slice_by_index(begin = var_9919_begin_0, end = var_9919_end_0, end_mask = var_9919_end_mask_0, x = var_9804_cast_fp16)[name = tensor("op_9919_cast_fp16")]; tensor var_9926_begin_0 = const()[name = tensor("op_9926_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9926_end_0 = const()[name = tensor("op_9926_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9926_end_mask_0 = const()[name = tensor("op_9926_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9926_cast_fp16 = slice_by_index(begin = var_9926_begin_0, end = var_9926_end_0, end_mask = var_9926_end_mask_0, x = var_9804_cast_fp16)[name = tensor("op_9926_cast_fp16")]; tensor var_9933_begin_0 = const()[name = tensor("op_9933_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9933_end_0 = const()[name = tensor("op_9933_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9933_end_mask_0 = const()[name = tensor("op_9933_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9933_cast_fp16 = slice_by_index(begin = var_9933_begin_0, end = var_9933_end_0, end_mask = var_9933_end_mask_0, x = var_9808_cast_fp16)[name = tensor("op_9933_cast_fp16")]; tensor var_9940_begin_0 = const()[name = tensor("op_9940_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9940_end_0 = const()[name = tensor("op_9940_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9940_end_mask_0 = const()[name = tensor("op_9940_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9940_cast_fp16 = slice_by_index(begin = var_9940_begin_0, end = var_9940_end_0, end_mask = var_9940_end_mask_0, x = var_9808_cast_fp16)[name = tensor("op_9940_cast_fp16")]; tensor var_9947_begin_0 = const()[name = tensor("op_9947_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9947_end_0 = const()[name = tensor("op_9947_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9947_end_mask_0 = const()[name = tensor("op_9947_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9947_cast_fp16 = slice_by_index(begin = var_9947_begin_0, end = var_9947_end_0, end_mask = var_9947_end_mask_0, x = var_9808_cast_fp16)[name = tensor("op_9947_cast_fp16")]; tensor var_9954_begin_0 = const()[name = tensor("op_9954_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9954_end_0 = const()[name = tensor("op_9954_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9954_end_mask_0 = const()[name = tensor("op_9954_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9954_cast_fp16 = slice_by_index(begin = var_9954_begin_0, end = var_9954_end_0, end_mask = var_9954_end_mask_0, x = var_9808_cast_fp16)[name = tensor("op_9954_cast_fp16")]; tensor var_9961_begin_0 = const()[name = tensor("op_9961_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9961_end_0 = const()[name = tensor("op_9961_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9961_end_mask_0 = const()[name = tensor("op_9961_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9961_cast_fp16 = slice_by_index(begin = var_9961_begin_0, end = var_9961_end_0, end_mask = var_9961_end_mask_0, x = var_9812_cast_fp16)[name = tensor("op_9961_cast_fp16")]; tensor var_9968_begin_0 = const()[name = tensor("op_9968_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9968_end_0 = const()[name = tensor("op_9968_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9968_end_mask_0 = const()[name = tensor("op_9968_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9968_cast_fp16 = slice_by_index(begin = var_9968_begin_0, end = var_9968_end_0, end_mask = var_9968_end_mask_0, x = var_9812_cast_fp16)[name = tensor("op_9968_cast_fp16")]; tensor var_9975_begin_0 = const()[name = tensor("op_9975_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_9975_end_0 = const()[name = tensor("op_9975_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_9975_end_mask_0 = const()[name = tensor("op_9975_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9975_cast_fp16 = slice_by_index(begin = var_9975_begin_0, end = var_9975_end_0, end_mask = var_9975_end_mask_0, x = var_9812_cast_fp16)[name = tensor("op_9975_cast_fp16")]; tensor var_9982_begin_0 = const()[name = tensor("op_9982_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_9982_end_0 = const()[name = tensor("op_9982_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_9982_end_mask_0 = const()[name = tensor("op_9982_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9982_cast_fp16 = slice_by_index(begin = var_9982_begin_0, end = var_9982_end_0, end_mask = var_9982_end_mask_0, x = var_9812_cast_fp16)[name = tensor("op_9982_cast_fp16")]; tensor var_9989_begin_0 = const()[name = tensor("op_9989_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_9989_end_0 = const()[name = tensor("op_9989_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_9989_end_mask_0 = const()[name = tensor("op_9989_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9989_cast_fp16 = slice_by_index(begin = var_9989_begin_0, end = var_9989_end_0, end_mask = var_9989_end_mask_0, x = var_9816_cast_fp16)[name = tensor("op_9989_cast_fp16")]; tensor var_9996_begin_0 = const()[name = tensor("op_9996_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_9996_end_0 = const()[name = tensor("op_9996_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_9996_end_mask_0 = const()[name = tensor("op_9996_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_9996_cast_fp16 = slice_by_index(begin = var_9996_begin_0, end = var_9996_end_0, end_mask = var_9996_end_mask_0, x = var_9816_cast_fp16)[name = tensor("op_9996_cast_fp16")]; tensor var_10003_begin_0 = const()[name = tensor("op_10003_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10003_end_0 = const()[name = tensor("op_10003_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10003_end_mask_0 = const()[name = tensor("op_10003_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10003_cast_fp16 = slice_by_index(begin = var_10003_begin_0, end = var_10003_end_0, end_mask = var_10003_end_mask_0, x = var_9816_cast_fp16)[name = tensor("op_10003_cast_fp16")]; tensor var_10010_begin_0 = const()[name = tensor("op_10010_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10010_end_0 = const()[name = tensor("op_10010_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10010_end_mask_0 = const()[name = tensor("op_10010_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10010_cast_fp16 = slice_by_index(begin = var_10010_begin_0, end = var_10010_end_0, end_mask = var_10010_end_mask_0, x = var_9816_cast_fp16)[name = tensor("op_10010_cast_fp16")]; tensor var_10017_begin_0 = const()[name = tensor("op_10017_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10017_end_0 = const()[name = tensor("op_10017_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10017_end_mask_0 = const()[name = tensor("op_10017_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10017_cast_fp16 = slice_by_index(begin = var_10017_begin_0, end = var_10017_end_0, end_mask = var_10017_end_mask_0, x = var_9820_cast_fp16)[name = tensor("op_10017_cast_fp16")]; tensor var_10024_begin_0 = const()[name = tensor("op_10024_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10024_end_0 = const()[name = tensor("op_10024_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10024_end_mask_0 = const()[name = tensor("op_10024_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10024_cast_fp16 = slice_by_index(begin = var_10024_begin_0, end = var_10024_end_0, end_mask = var_10024_end_mask_0, x = var_9820_cast_fp16)[name = tensor("op_10024_cast_fp16")]; tensor var_10031_begin_0 = const()[name = tensor("op_10031_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10031_end_0 = const()[name = tensor("op_10031_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10031_end_mask_0 = const()[name = tensor("op_10031_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10031_cast_fp16 = slice_by_index(begin = var_10031_begin_0, end = var_10031_end_0, end_mask = var_10031_end_mask_0, x = var_9820_cast_fp16)[name = tensor("op_10031_cast_fp16")]; tensor var_10038_begin_0 = const()[name = tensor("op_10038_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10038_end_0 = const()[name = tensor("op_10038_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10038_end_mask_0 = const()[name = tensor("op_10038_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10038_cast_fp16 = slice_by_index(begin = var_10038_begin_0, end = var_10038_end_0, end_mask = var_10038_end_mask_0, x = var_9820_cast_fp16)[name = tensor("op_10038_cast_fp16")]; tensor var_10045_begin_0 = const()[name = tensor("op_10045_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10045_end_0 = const()[name = tensor("op_10045_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10045_end_mask_0 = const()[name = tensor("op_10045_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10045_cast_fp16 = slice_by_index(begin = var_10045_begin_0, end = var_10045_end_0, end_mask = var_10045_end_mask_0, x = var_9824_cast_fp16)[name = tensor("op_10045_cast_fp16")]; tensor var_10052_begin_0 = const()[name = tensor("op_10052_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10052_end_0 = const()[name = tensor("op_10052_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10052_end_mask_0 = const()[name = tensor("op_10052_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10052_cast_fp16 = slice_by_index(begin = var_10052_begin_0, end = var_10052_end_0, end_mask = var_10052_end_mask_0, x = var_9824_cast_fp16)[name = tensor("op_10052_cast_fp16")]; tensor var_10059_begin_0 = const()[name = tensor("op_10059_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10059_end_0 = const()[name = tensor("op_10059_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10059_end_mask_0 = const()[name = tensor("op_10059_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10059_cast_fp16 = slice_by_index(begin = var_10059_begin_0, end = var_10059_end_0, end_mask = var_10059_end_mask_0, x = var_9824_cast_fp16)[name = tensor("op_10059_cast_fp16")]; tensor var_10066_begin_0 = const()[name = tensor("op_10066_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10066_end_0 = const()[name = tensor("op_10066_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10066_end_mask_0 = const()[name = tensor("op_10066_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10066_cast_fp16 = slice_by_index(begin = var_10066_begin_0, end = var_10066_end_0, end_mask = var_10066_end_mask_0, x = var_9824_cast_fp16)[name = tensor("op_10066_cast_fp16")]; tensor var_10073_begin_0 = const()[name = tensor("op_10073_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10073_end_0 = const()[name = tensor("op_10073_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10073_end_mask_0 = const()[name = tensor("op_10073_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10073_cast_fp16 = slice_by_index(begin = var_10073_begin_0, end = var_10073_end_0, end_mask = var_10073_end_mask_0, x = var_9828_cast_fp16)[name = tensor("op_10073_cast_fp16")]; tensor var_10080_begin_0 = const()[name = tensor("op_10080_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10080_end_0 = const()[name = tensor("op_10080_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10080_end_mask_0 = const()[name = tensor("op_10080_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10080_cast_fp16 = slice_by_index(begin = var_10080_begin_0, end = var_10080_end_0, end_mask = var_10080_end_mask_0, x = var_9828_cast_fp16)[name = tensor("op_10080_cast_fp16")]; tensor var_10087_begin_0 = const()[name = tensor("op_10087_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10087_end_0 = const()[name = tensor("op_10087_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10087_end_mask_0 = const()[name = tensor("op_10087_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10087_cast_fp16 = slice_by_index(begin = var_10087_begin_0, end = var_10087_end_0, end_mask = var_10087_end_mask_0, x = var_9828_cast_fp16)[name = tensor("op_10087_cast_fp16")]; tensor var_10094_begin_0 = const()[name = tensor("op_10094_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10094_end_0 = const()[name = tensor("op_10094_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10094_end_mask_0 = const()[name = tensor("op_10094_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10094_cast_fp16 = slice_by_index(begin = var_10094_begin_0, end = var_10094_end_0, end_mask = var_10094_end_mask_0, x = var_9828_cast_fp16)[name = tensor("op_10094_cast_fp16")]; tensor var_10101_begin_0 = const()[name = tensor("op_10101_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10101_end_0 = const()[name = tensor("op_10101_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10101_end_mask_0 = const()[name = tensor("op_10101_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10101_cast_fp16 = slice_by_index(begin = var_10101_begin_0, end = var_10101_end_0, end_mask = var_10101_end_mask_0, x = var_9832_cast_fp16)[name = tensor("op_10101_cast_fp16")]; tensor var_10108_begin_0 = const()[name = tensor("op_10108_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10108_end_0 = const()[name = tensor("op_10108_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10108_end_mask_0 = const()[name = tensor("op_10108_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10108_cast_fp16 = slice_by_index(begin = var_10108_begin_0, end = var_10108_end_0, end_mask = var_10108_end_mask_0, x = var_9832_cast_fp16)[name = tensor("op_10108_cast_fp16")]; tensor var_10115_begin_0 = const()[name = tensor("op_10115_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10115_end_0 = const()[name = tensor("op_10115_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10115_end_mask_0 = const()[name = tensor("op_10115_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10115_cast_fp16 = slice_by_index(begin = var_10115_begin_0, end = var_10115_end_0, end_mask = var_10115_end_mask_0, x = var_9832_cast_fp16)[name = tensor("op_10115_cast_fp16")]; tensor var_10122_begin_0 = const()[name = tensor("op_10122_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10122_end_0 = const()[name = tensor("op_10122_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10122_end_mask_0 = const()[name = tensor("op_10122_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_9832_cast_fp16)[name = tensor("op_10122_cast_fp16")]; tensor var_10129_begin_0 = const()[name = tensor("op_10129_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10129_end_0 = const()[name = tensor("op_10129_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10129_end_mask_0 = const()[name = tensor("op_10129_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10129_cast_fp16 = slice_by_index(begin = var_10129_begin_0, end = var_10129_end_0, end_mask = var_10129_end_mask_0, x = var_9836_cast_fp16)[name = tensor("op_10129_cast_fp16")]; tensor var_10136_begin_0 = const()[name = tensor("op_10136_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10136_end_0 = const()[name = tensor("op_10136_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10136_end_mask_0 = const()[name = tensor("op_10136_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10136_cast_fp16 = slice_by_index(begin = var_10136_begin_0, end = var_10136_end_0, end_mask = var_10136_end_mask_0, x = var_9836_cast_fp16)[name = tensor("op_10136_cast_fp16")]; tensor var_10143_begin_0 = const()[name = tensor("op_10143_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10143_end_0 = const()[name = tensor("op_10143_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10143_end_mask_0 = const()[name = tensor("op_10143_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10143_cast_fp16 = slice_by_index(begin = var_10143_begin_0, end = var_10143_end_0, end_mask = var_10143_end_mask_0, x = var_9836_cast_fp16)[name = tensor("op_10143_cast_fp16")]; tensor var_10150_begin_0 = const()[name = tensor("op_10150_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10150_end_0 = const()[name = tensor("op_10150_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10150_end_mask_0 = const()[name = tensor("op_10150_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_9836_cast_fp16)[name = tensor("op_10150_cast_fp16")]; tensor var_10157_begin_0 = const()[name = tensor("op_10157_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10157_end_0 = const()[name = tensor("op_10157_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10157_end_mask_0 = const()[name = tensor("op_10157_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10157_cast_fp16 = slice_by_index(begin = var_10157_begin_0, end = var_10157_end_0, end_mask = var_10157_end_mask_0, x = var_9840_cast_fp16)[name = tensor("op_10157_cast_fp16")]; tensor var_10164_begin_0 = const()[name = tensor("op_10164_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10164_end_0 = const()[name = tensor("op_10164_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10164_end_mask_0 = const()[name = tensor("op_10164_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10164_cast_fp16 = slice_by_index(begin = var_10164_begin_0, end = var_10164_end_0, end_mask = var_10164_end_mask_0, x = var_9840_cast_fp16)[name = tensor("op_10164_cast_fp16")]; tensor var_10171_begin_0 = const()[name = tensor("op_10171_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10171_end_0 = const()[name = tensor("op_10171_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10171_end_mask_0 = const()[name = tensor("op_10171_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10171_cast_fp16 = slice_by_index(begin = var_10171_begin_0, end = var_10171_end_0, end_mask = var_10171_end_mask_0, x = var_9840_cast_fp16)[name = tensor("op_10171_cast_fp16")]; tensor var_10178_begin_0 = const()[name = tensor("op_10178_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10178_end_0 = const()[name = tensor("op_10178_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10178_end_mask_0 = const()[name = tensor("op_10178_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_9840_cast_fp16)[name = tensor("op_10178_cast_fp16")]; tensor k_21_perm_0 = const()[name = tensor("k_21_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_10183_begin_0 = const()[name = tensor("op_10183_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10183_end_0 = const()[name = tensor("op_10183_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_10183_end_mask_0 = const()[name = tensor("op_10183_end_mask_0"), val = tensor([true, true, true, false])]; tensor transpose_1 = transpose(perm = k_21_perm_0, x = key_21_cast_fp16)[name = tensor("transpose_1")]; tensor var_10183_cast_fp16 = slice_by_index(begin = var_10183_begin_0, end = var_10183_end_0, end_mask = var_10183_end_mask_0, x = transpose_1)[name = tensor("op_10183_cast_fp16")]; tensor var_10187_begin_0 = const()[name = tensor("op_10187_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_10187_end_0 = const()[name = tensor("op_10187_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_10187_end_mask_0 = const()[name = tensor("op_10187_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10187_cast_fp16 = slice_by_index(begin = var_10187_begin_0, end = var_10187_end_0, end_mask = var_10187_end_mask_0, x = transpose_1)[name = tensor("op_10187_cast_fp16")]; tensor var_10191_begin_0 = const()[name = tensor("op_10191_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_10191_end_0 = const()[name = tensor("op_10191_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_10191_end_mask_0 = const()[name = tensor("op_10191_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10191_cast_fp16 = slice_by_index(begin = var_10191_begin_0, end = var_10191_end_0, end_mask = var_10191_end_mask_0, x = transpose_1)[name = tensor("op_10191_cast_fp16")]; tensor var_10195_begin_0 = const()[name = tensor("op_10195_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_10195_end_0 = const()[name = tensor("op_10195_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_10195_end_mask_0 = const()[name = tensor("op_10195_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10195_cast_fp16 = slice_by_index(begin = var_10195_begin_0, end = var_10195_end_0, end_mask = var_10195_end_mask_0, x = transpose_1)[name = tensor("op_10195_cast_fp16")]; tensor var_10199_begin_0 = const()[name = tensor("op_10199_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_10199_end_0 = const()[name = tensor("op_10199_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_10199_end_mask_0 = const()[name = tensor("op_10199_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10199_cast_fp16 = slice_by_index(begin = var_10199_begin_0, end = var_10199_end_0, end_mask = var_10199_end_mask_0, x = transpose_1)[name = tensor("op_10199_cast_fp16")]; tensor var_10203_begin_0 = const()[name = tensor("op_10203_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_10203_end_0 = const()[name = tensor("op_10203_end_0"), val = tensor([1, 1500, 1, 384])]; 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 = transpose_1)[name = tensor("op_10203_cast_fp16")]; tensor var_10207_begin_0 = const()[name = tensor("op_10207_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_10207_end_0 = const()[name = tensor("op_10207_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_10207_end_mask_0 = const()[name = tensor("op_10207_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10207_cast_fp16 = slice_by_index(begin = var_10207_begin_0, end = var_10207_end_0, end_mask = var_10207_end_mask_0, x = transpose_1)[name = tensor("op_10207_cast_fp16")]; tensor var_10211_begin_0 = const()[name = tensor("op_10211_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_10211_end_0 = const()[name = tensor("op_10211_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_10211_end_mask_0 = const()[name = tensor("op_10211_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10211_cast_fp16 = slice_by_index(begin = var_10211_begin_0, end = var_10211_end_0, end_mask = var_10211_end_mask_0, x = transpose_1)[name = tensor("op_10211_cast_fp16")]; tensor var_10215_begin_0 = const()[name = tensor("op_10215_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_10215_end_0 = const()[name = tensor("op_10215_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_10215_end_mask_0 = const()[name = tensor("op_10215_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10215_cast_fp16 = slice_by_index(begin = var_10215_begin_0, end = var_10215_end_0, end_mask = var_10215_end_mask_0, x = transpose_1)[name = tensor("op_10215_cast_fp16")]; tensor var_10219_begin_0 = const()[name = tensor("op_10219_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_10219_end_0 = const()[name = tensor("op_10219_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_10219_end_mask_0 = const()[name = tensor("op_10219_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10219_cast_fp16 = slice_by_index(begin = var_10219_begin_0, end = var_10219_end_0, end_mask = var_10219_end_mask_0, x = transpose_1)[name = tensor("op_10219_cast_fp16")]; tensor var_10223_begin_0 = const()[name = tensor("op_10223_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_10223_end_0 = const()[name = tensor("op_10223_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_10223_end_mask_0 = const()[name = tensor("op_10223_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10223_cast_fp16 = slice_by_index(begin = var_10223_begin_0, end = var_10223_end_0, end_mask = var_10223_end_mask_0, x = transpose_1)[name = tensor("op_10223_cast_fp16")]; tensor var_10227_begin_0 = const()[name = tensor("op_10227_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_10227_end_0 = const()[name = tensor("op_10227_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_10227_end_mask_0 = const()[name = tensor("op_10227_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10227_cast_fp16 = slice_by_index(begin = var_10227_begin_0, end = var_10227_end_0, end_mask = var_10227_end_mask_0, x = transpose_1)[name = tensor("op_10227_cast_fp16")]; tensor var_10229_begin_0 = const()[name = tensor("op_10229_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10229_end_0 = const()[name = tensor("op_10229_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10229_end_mask_0 = const()[name = tensor("op_10229_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10229_cast_fp16 = slice_by_index(begin = var_10229_begin_0, end = var_10229_end_0, end_mask = var_10229_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10229_cast_fp16")]; tensor var_10233_begin_0 = const()[name = tensor("op_10233_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10233_end_0 = const()[name = tensor("op_10233_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_10233_end_mask_0 = const()[name = tensor("op_10233_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10233_cast_fp16 = slice_by_index(begin = var_10233_begin_0, end = var_10233_end_0, end_mask = var_10233_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10233_cast_fp16")]; tensor var_10237_begin_0 = const()[name = tensor("op_10237_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10237_end_0 = const()[name = tensor("op_10237_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_10237_end_mask_0 = const()[name = tensor("op_10237_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10237_cast_fp16 = slice_by_index(begin = var_10237_begin_0, end = var_10237_end_0, end_mask = var_10237_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10237_cast_fp16")]; tensor var_10241_begin_0 = const()[name = tensor("op_10241_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10241_end_0 = const()[name = tensor("op_10241_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_10241_end_mask_0 = const()[name = tensor("op_10241_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10241_cast_fp16 = slice_by_index(begin = var_10241_begin_0, end = var_10241_end_0, end_mask = var_10241_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10241_cast_fp16")]; tensor var_10245_begin_0 = const()[name = tensor("op_10245_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10245_end_0 = const()[name = tensor("op_10245_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_10245_end_mask_0 = const()[name = tensor("op_10245_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_21_cast_fp16)[name = tensor("op_10245_cast_fp16")]; tensor var_10249_begin_0 = const()[name = tensor("op_10249_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10249_end_0 = const()[name = tensor("op_10249_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_10249_end_mask_0 = const()[name = tensor("op_10249_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10249_cast_fp16 = slice_by_index(begin = var_10249_begin_0, end = var_10249_end_0, end_mask = var_10249_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10249_cast_fp16")]; tensor var_10253_begin_0 = const()[name = tensor("op_10253_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10253_end_0 = const()[name = tensor("op_10253_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_10253_end_mask_0 = const()[name = tensor("op_10253_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10253_cast_fp16 = slice_by_index(begin = var_10253_begin_0, end = var_10253_end_0, end_mask = var_10253_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10253_cast_fp16")]; tensor var_10257_begin_0 = const()[name = tensor("op_10257_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10257_end_0 = const()[name = tensor("op_10257_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_10257_end_mask_0 = const()[name = tensor("op_10257_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10257_cast_fp16 = slice_by_index(begin = var_10257_begin_0, end = var_10257_end_0, end_mask = var_10257_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10257_cast_fp16")]; tensor var_10261_begin_0 = const()[name = tensor("op_10261_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10261_end_0 = const()[name = tensor("op_10261_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_10261_end_mask_0 = const()[name = tensor("op_10261_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10261_cast_fp16 = slice_by_index(begin = var_10261_begin_0, end = var_10261_end_0, end_mask = var_10261_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10261_cast_fp16")]; tensor var_10265_begin_0 = const()[name = tensor("op_10265_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10265_end_0 = const()[name = tensor("op_10265_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_10265_end_mask_0 = const()[name = tensor("op_10265_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10265_cast_fp16 = slice_by_index(begin = var_10265_begin_0, end = var_10265_end_0, end_mask = var_10265_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10265_cast_fp16")]; tensor var_10269_begin_0 = const()[name = tensor("op_10269_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_10269_end_0 = const()[name = tensor("op_10269_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_10269_end_mask_0 = const()[name = tensor("op_10269_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10269_cast_fp16 = slice_by_index(begin = var_10269_begin_0, end = var_10269_end_0, end_mask = var_10269_end_mask_0, x = value_21_cast_fp16)[name = tensor("op_10269_cast_fp16")]; tensor var_10273_begin_0 = const()[name = tensor("op_10273_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_10273_end_0 = const()[name = tensor("op_10273_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_10273_end_mask_0 = const()[name = tensor("op_10273_end_mask_0"), val = tensor([true, false, true, true])]; 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 = value_21_cast_fp16)[name = tensor("op_10273_cast_fp16")]; tensor var_10277_equation_0 = const()[name = tensor("op_10277_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10277_cast_fp16 = einsum(equation = var_10277_equation_0, values = (var_10183_cast_fp16, var_9849_cast_fp16))[name = tensor("op_10277_cast_fp16")]; tensor var_10278_to_fp16 = const()[name = tensor("op_10278_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_961_cast_fp16 = mul(x = var_10277_cast_fp16, y = var_10278_to_fp16)[name = tensor("aw_chunk_961_cast_fp16")]; tensor var_10281_equation_0 = const()[name = tensor("op_10281_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10281_cast_fp16 = einsum(equation = var_10281_equation_0, values = (var_10183_cast_fp16, var_9856_cast_fp16))[name = tensor("op_10281_cast_fp16")]; tensor var_10282_to_fp16 = const()[name = tensor("op_10282_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_963_cast_fp16 = mul(x = var_10281_cast_fp16, y = var_10282_to_fp16)[name = tensor("aw_chunk_963_cast_fp16")]; tensor var_10285_equation_0 = const()[name = tensor("op_10285_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10285_cast_fp16 = einsum(equation = var_10285_equation_0, values = (var_10183_cast_fp16, var_9863_cast_fp16))[name = tensor("op_10285_cast_fp16")]; tensor var_10286_to_fp16 = const()[name = tensor("op_10286_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_965_cast_fp16 = mul(x = var_10285_cast_fp16, y = var_10286_to_fp16)[name = tensor("aw_chunk_965_cast_fp16")]; tensor var_10289_equation_0 = const()[name = tensor("op_10289_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10289_cast_fp16 = einsum(equation = var_10289_equation_0, values = (var_10183_cast_fp16, var_9870_cast_fp16))[name = tensor("op_10289_cast_fp16")]; tensor var_10290_to_fp16 = const()[name = tensor("op_10290_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_967_cast_fp16 = mul(x = var_10289_cast_fp16, y = var_10290_to_fp16)[name = tensor("aw_chunk_967_cast_fp16")]; tensor var_10293_equation_0 = const()[name = tensor("op_10293_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10293_cast_fp16 = einsum(equation = var_10293_equation_0, values = (var_10187_cast_fp16, var_9877_cast_fp16))[name = tensor("op_10293_cast_fp16")]; tensor var_10294_to_fp16 = const()[name = tensor("op_10294_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_969_cast_fp16 = mul(x = var_10293_cast_fp16, y = var_10294_to_fp16)[name = tensor("aw_chunk_969_cast_fp16")]; tensor var_10297_equation_0 = const()[name = tensor("op_10297_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10297_cast_fp16 = einsum(equation = var_10297_equation_0, values = (var_10187_cast_fp16, var_9884_cast_fp16))[name = tensor("op_10297_cast_fp16")]; tensor var_10298_to_fp16 = const()[name = tensor("op_10298_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_971_cast_fp16 = mul(x = var_10297_cast_fp16, y = var_10298_to_fp16)[name = tensor("aw_chunk_971_cast_fp16")]; tensor var_10301_equation_0 = const()[name = tensor("op_10301_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10301_cast_fp16 = einsum(equation = var_10301_equation_0, values = (var_10187_cast_fp16, var_9891_cast_fp16))[name = tensor("op_10301_cast_fp16")]; tensor var_10302_to_fp16 = const()[name = tensor("op_10302_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_973_cast_fp16 = mul(x = var_10301_cast_fp16, y = var_10302_to_fp16)[name = tensor("aw_chunk_973_cast_fp16")]; tensor var_10305_equation_0 = const()[name = tensor("op_10305_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10305_cast_fp16 = einsum(equation = var_10305_equation_0, values = (var_10187_cast_fp16, var_9898_cast_fp16))[name = tensor("op_10305_cast_fp16")]; tensor var_10306_to_fp16 = const()[name = tensor("op_10306_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_975_cast_fp16 = mul(x = var_10305_cast_fp16, y = var_10306_to_fp16)[name = tensor("aw_chunk_975_cast_fp16")]; tensor var_10309_equation_0 = const()[name = tensor("op_10309_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10309_cast_fp16 = einsum(equation = var_10309_equation_0, values = (var_10191_cast_fp16, var_9905_cast_fp16))[name = tensor("op_10309_cast_fp16")]; tensor var_10310_to_fp16 = const()[name = tensor("op_10310_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_977_cast_fp16 = mul(x = var_10309_cast_fp16, y = var_10310_to_fp16)[name = tensor("aw_chunk_977_cast_fp16")]; tensor var_10313_equation_0 = const()[name = tensor("op_10313_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10313_cast_fp16 = einsum(equation = var_10313_equation_0, values = (var_10191_cast_fp16, var_9912_cast_fp16))[name = tensor("op_10313_cast_fp16")]; tensor var_10314_to_fp16 = const()[name = tensor("op_10314_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_979_cast_fp16 = mul(x = var_10313_cast_fp16, y = var_10314_to_fp16)[name = tensor("aw_chunk_979_cast_fp16")]; tensor var_10317_equation_0 = const()[name = tensor("op_10317_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10317_cast_fp16 = einsum(equation = var_10317_equation_0, values = (var_10191_cast_fp16, var_9919_cast_fp16))[name = tensor("op_10317_cast_fp16")]; tensor var_10318_to_fp16 = const()[name = tensor("op_10318_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_981_cast_fp16 = mul(x = var_10317_cast_fp16, y = var_10318_to_fp16)[name = tensor("aw_chunk_981_cast_fp16")]; tensor var_10321_equation_0 = const()[name = tensor("op_10321_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10321_cast_fp16 = einsum(equation = var_10321_equation_0, values = (var_10191_cast_fp16, var_9926_cast_fp16))[name = tensor("op_10321_cast_fp16")]; tensor var_10322_to_fp16 = const()[name = tensor("op_10322_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_983_cast_fp16 = mul(x = var_10321_cast_fp16, y = var_10322_to_fp16)[name = tensor("aw_chunk_983_cast_fp16")]; tensor var_10325_equation_0 = const()[name = tensor("op_10325_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10325_cast_fp16 = einsum(equation = var_10325_equation_0, values = (var_10195_cast_fp16, var_9933_cast_fp16))[name = tensor("op_10325_cast_fp16")]; tensor var_10326_to_fp16 = const()[name = tensor("op_10326_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_985_cast_fp16 = mul(x = var_10325_cast_fp16, y = var_10326_to_fp16)[name = tensor("aw_chunk_985_cast_fp16")]; tensor var_10329_equation_0 = const()[name = tensor("op_10329_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10329_cast_fp16 = einsum(equation = var_10329_equation_0, values = (var_10195_cast_fp16, var_9940_cast_fp16))[name = tensor("op_10329_cast_fp16")]; tensor var_10330_to_fp16 = const()[name = tensor("op_10330_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_987_cast_fp16 = mul(x = var_10329_cast_fp16, y = var_10330_to_fp16)[name = tensor("aw_chunk_987_cast_fp16")]; tensor var_10333_equation_0 = const()[name = tensor("op_10333_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10333_cast_fp16 = einsum(equation = var_10333_equation_0, values = (var_10195_cast_fp16, var_9947_cast_fp16))[name = tensor("op_10333_cast_fp16")]; tensor var_10334_to_fp16 = const()[name = tensor("op_10334_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_989_cast_fp16 = mul(x = var_10333_cast_fp16, y = var_10334_to_fp16)[name = tensor("aw_chunk_989_cast_fp16")]; tensor var_10337_equation_0 = const()[name = tensor("op_10337_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10337_cast_fp16 = einsum(equation = var_10337_equation_0, values = (var_10195_cast_fp16, var_9954_cast_fp16))[name = tensor("op_10337_cast_fp16")]; tensor var_10338_to_fp16 = const()[name = tensor("op_10338_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_991_cast_fp16 = mul(x = var_10337_cast_fp16, y = var_10338_to_fp16)[name = tensor("aw_chunk_991_cast_fp16")]; tensor var_10341_equation_0 = const()[name = tensor("op_10341_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10341_cast_fp16 = einsum(equation = var_10341_equation_0, values = (var_10199_cast_fp16, var_9961_cast_fp16))[name = tensor("op_10341_cast_fp16")]; tensor var_10342_to_fp16 = const()[name = tensor("op_10342_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_993_cast_fp16 = mul(x = var_10341_cast_fp16, y = var_10342_to_fp16)[name = tensor("aw_chunk_993_cast_fp16")]; tensor var_10345_equation_0 = const()[name = tensor("op_10345_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10345_cast_fp16 = einsum(equation = var_10345_equation_0, values = (var_10199_cast_fp16, var_9968_cast_fp16))[name = tensor("op_10345_cast_fp16")]; tensor var_10346_to_fp16 = const()[name = tensor("op_10346_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_995_cast_fp16 = mul(x = var_10345_cast_fp16, y = var_10346_to_fp16)[name = tensor("aw_chunk_995_cast_fp16")]; tensor var_10349_equation_0 = const()[name = tensor("op_10349_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10349_cast_fp16 = einsum(equation = var_10349_equation_0, values = (var_10199_cast_fp16, var_9975_cast_fp16))[name = tensor("op_10349_cast_fp16")]; tensor var_10350_to_fp16 = const()[name = tensor("op_10350_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_997_cast_fp16 = mul(x = var_10349_cast_fp16, y = var_10350_to_fp16)[name = tensor("aw_chunk_997_cast_fp16")]; tensor var_10353_equation_0 = const()[name = tensor("op_10353_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10353_cast_fp16 = einsum(equation = var_10353_equation_0, values = (var_10199_cast_fp16, var_9982_cast_fp16))[name = tensor("op_10353_cast_fp16")]; tensor var_10354_to_fp16 = const()[name = tensor("op_10354_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_999_cast_fp16 = mul(x = var_10353_cast_fp16, y = var_10354_to_fp16)[name = tensor("aw_chunk_999_cast_fp16")]; tensor var_10357_equation_0 = const()[name = tensor("op_10357_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10357_cast_fp16 = einsum(equation = var_10357_equation_0, values = (var_10203_cast_fp16, var_9989_cast_fp16))[name = tensor("op_10357_cast_fp16")]; tensor var_10358_to_fp16 = const()[name = tensor("op_10358_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1001_cast_fp16 = mul(x = var_10357_cast_fp16, y = var_10358_to_fp16)[name = tensor("aw_chunk_1001_cast_fp16")]; tensor var_10361_equation_0 = const()[name = tensor("op_10361_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10361_cast_fp16 = einsum(equation = var_10361_equation_0, values = (var_10203_cast_fp16, var_9996_cast_fp16))[name = tensor("op_10361_cast_fp16")]; tensor var_10362_to_fp16 = const()[name = tensor("op_10362_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1003_cast_fp16 = mul(x = var_10361_cast_fp16, y = var_10362_to_fp16)[name = tensor("aw_chunk_1003_cast_fp16")]; tensor var_10365_equation_0 = const()[name = tensor("op_10365_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10365_cast_fp16 = einsum(equation = var_10365_equation_0, values = (var_10203_cast_fp16, var_10003_cast_fp16))[name = tensor("op_10365_cast_fp16")]; tensor var_10366_to_fp16 = const()[name = tensor("op_10366_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1005_cast_fp16 = mul(x = var_10365_cast_fp16, y = var_10366_to_fp16)[name = tensor("aw_chunk_1005_cast_fp16")]; tensor var_10369_equation_0 = const()[name = tensor("op_10369_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10369_cast_fp16 = einsum(equation = var_10369_equation_0, values = (var_10203_cast_fp16, var_10010_cast_fp16))[name = tensor("op_10369_cast_fp16")]; tensor var_10370_to_fp16 = const()[name = tensor("op_10370_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1007_cast_fp16 = mul(x = var_10369_cast_fp16, y = var_10370_to_fp16)[name = tensor("aw_chunk_1007_cast_fp16")]; tensor var_10373_equation_0 = const()[name = tensor("op_10373_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10373_cast_fp16 = einsum(equation = var_10373_equation_0, values = (var_10207_cast_fp16, var_10017_cast_fp16))[name = tensor("op_10373_cast_fp16")]; tensor var_10374_to_fp16 = const()[name = tensor("op_10374_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1009_cast_fp16 = mul(x = var_10373_cast_fp16, y = var_10374_to_fp16)[name = tensor("aw_chunk_1009_cast_fp16")]; tensor var_10377_equation_0 = const()[name = tensor("op_10377_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10377_cast_fp16 = einsum(equation = var_10377_equation_0, values = (var_10207_cast_fp16, var_10024_cast_fp16))[name = tensor("op_10377_cast_fp16")]; tensor var_10378_to_fp16 = const()[name = tensor("op_10378_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1011_cast_fp16 = mul(x = var_10377_cast_fp16, y = var_10378_to_fp16)[name = tensor("aw_chunk_1011_cast_fp16")]; tensor var_10381_equation_0 = const()[name = tensor("op_10381_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10381_cast_fp16 = einsum(equation = var_10381_equation_0, values = (var_10207_cast_fp16, var_10031_cast_fp16))[name = tensor("op_10381_cast_fp16")]; tensor var_10382_to_fp16 = const()[name = tensor("op_10382_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1013_cast_fp16 = mul(x = var_10381_cast_fp16, y = var_10382_to_fp16)[name = tensor("aw_chunk_1013_cast_fp16")]; tensor var_10385_equation_0 = const()[name = tensor("op_10385_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10385_cast_fp16 = einsum(equation = var_10385_equation_0, values = (var_10207_cast_fp16, var_10038_cast_fp16))[name = tensor("op_10385_cast_fp16")]; tensor var_10386_to_fp16 = const()[name = tensor("op_10386_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1015_cast_fp16 = mul(x = var_10385_cast_fp16, y = var_10386_to_fp16)[name = tensor("aw_chunk_1015_cast_fp16")]; tensor var_10389_equation_0 = const()[name = tensor("op_10389_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10389_cast_fp16 = einsum(equation = var_10389_equation_0, values = (var_10211_cast_fp16, var_10045_cast_fp16))[name = tensor("op_10389_cast_fp16")]; tensor var_10390_to_fp16 = const()[name = tensor("op_10390_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1017_cast_fp16 = mul(x = var_10389_cast_fp16, y = var_10390_to_fp16)[name = tensor("aw_chunk_1017_cast_fp16")]; tensor var_10393_equation_0 = const()[name = tensor("op_10393_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10393_cast_fp16 = einsum(equation = var_10393_equation_0, values = (var_10211_cast_fp16, var_10052_cast_fp16))[name = tensor("op_10393_cast_fp16")]; tensor var_10394_to_fp16 = const()[name = tensor("op_10394_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1019_cast_fp16 = mul(x = var_10393_cast_fp16, y = var_10394_to_fp16)[name = tensor("aw_chunk_1019_cast_fp16")]; tensor var_10397_equation_0 = const()[name = tensor("op_10397_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10397_cast_fp16 = einsum(equation = var_10397_equation_0, values = (var_10211_cast_fp16, var_10059_cast_fp16))[name = tensor("op_10397_cast_fp16")]; tensor var_10398_to_fp16 = const()[name = tensor("op_10398_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1021_cast_fp16 = mul(x = var_10397_cast_fp16, y = var_10398_to_fp16)[name = tensor("aw_chunk_1021_cast_fp16")]; tensor var_10401_equation_0 = const()[name = tensor("op_10401_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10401_cast_fp16 = einsum(equation = var_10401_equation_0, values = (var_10211_cast_fp16, var_10066_cast_fp16))[name = tensor("op_10401_cast_fp16")]; tensor var_10402_to_fp16 = const()[name = tensor("op_10402_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1023_cast_fp16 = mul(x = var_10401_cast_fp16, y = var_10402_to_fp16)[name = tensor("aw_chunk_1023_cast_fp16")]; tensor var_10405_equation_0 = const()[name = tensor("op_10405_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10405_cast_fp16 = einsum(equation = var_10405_equation_0, values = (var_10215_cast_fp16, var_10073_cast_fp16))[name = tensor("op_10405_cast_fp16")]; tensor var_10406_to_fp16 = const()[name = tensor("op_10406_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1025_cast_fp16 = mul(x = var_10405_cast_fp16, y = var_10406_to_fp16)[name = tensor("aw_chunk_1025_cast_fp16")]; tensor var_10409_equation_0 = const()[name = tensor("op_10409_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10409_cast_fp16 = einsum(equation = var_10409_equation_0, values = (var_10215_cast_fp16, var_10080_cast_fp16))[name = tensor("op_10409_cast_fp16")]; tensor var_10410_to_fp16 = const()[name = tensor("op_10410_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1027_cast_fp16 = mul(x = var_10409_cast_fp16, y = var_10410_to_fp16)[name = tensor("aw_chunk_1027_cast_fp16")]; tensor var_10413_equation_0 = const()[name = tensor("op_10413_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10413_cast_fp16 = einsum(equation = var_10413_equation_0, values = (var_10215_cast_fp16, var_10087_cast_fp16))[name = tensor("op_10413_cast_fp16")]; tensor var_10414_to_fp16 = const()[name = tensor("op_10414_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1029_cast_fp16 = mul(x = var_10413_cast_fp16, y = var_10414_to_fp16)[name = tensor("aw_chunk_1029_cast_fp16")]; tensor var_10417_equation_0 = const()[name = tensor("op_10417_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10417_cast_fp16 = einsum(equation = var_10417_equation_0, values = (var_10215_cast_fp16, var_10094_cast_fp16))[name = tensor("op_10417_cast_fp16")]; tensor var_10418_to_fp16 = const()[name = tensor("op_10418_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1031_cast_fp16 = mul(x = var_10417_cast_fp16, y = var_10418_to_fp16)[name = tensor("aw_chunk_1031_cast_fp16")]; tensor var_10421_equation_0 = const()[name = tensor("op_10421_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10421_cast_fp16 = einsum(equation = var_10421_equation_0, values = (var_10219_cast_fp16, var_10101_cast_fp16))[name = tensor("op_10421_cast_fp16")]; tensor var_10422_to_fp16 = const()[name = tensor("op_10422_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1033_cast_fp16 = mul(x = var_10421_cast_fp16, y = var_10422_to_fp16)[name = tensor("aw_chunk_1033_cast_fp16")]; tensor var_10425_equation_0 = const()[name = tensor("op_10425_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10425_cast_fp16 = einsum(equation = var_10425_equation_0, values = (var_10219_cast_fp16, var_10108_cast_fp16))[name = tensor("op_10425_cast_fp16")]; tensor var_10426_to_fp16 = const()[name = tensor("op_10426_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1035_cast_fp16 = mul(x = var_10425_cast_fp16, y = var_10426_to_fp16)[name = tensor("aw_chunk_1035_cast_fp16")]; tensor var_10429_equation_0 = const()[name = tensor("op_10429_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10429_cast_fp16 = einsum(equation = var_10429_equation_0, values = (var_10219_cast_fp16, var_10115_cast_fp16))[name = tensor("op_10429_cast_fp16")]; tensor var_10430_to_fp16 = const()[name = tensor("op_10430_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1037_cast_fp16 = mul(x = var_10429_cast_fp16, y = var_10430_to_fp16)[name = tensor("aw_chunk_1037_cast_fp16")]; tensor var_10433_equation_0 = const()[name = tensor("op_10433_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10433_cast_fp16 = einsum(equation = var_10433_equation_0, values = (var_10219_cast_fp16, var_10122_cast_fp16))[name = tensor("op_10433_cast_fp16")]; tensor var_10434_to_fp16 = const()[name = tensor("op_10434_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1039_cast_fp16 = mul(x = var_10433_cast_fp16, y = var_10434_to_fp16)[name = tensor("aw_chunk_1039_cast_fp16")]; tensor var_10437_equation_0 = const()[name = tensor("op_10437_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10437_cast_fp16 = einsum(equation = var_10437_equation_0, values = (var_10223_cast_fp16, var_10129_cast_fp16))[name = tensor("op_10437_cast_fp16")]; tensor var_10438_to_fp16 = const()[name = tensor("op_10438_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1041_cast_fp16 = mul(x = var_10437_cast_fp16, y = var_10438_to_fp16)[name = tensor("aw_chunk_1041_cast_fp16")]; tensor var_10441_equation_0 = const()[name = tensor("op_10441_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10441_cast_fp16 = einsum(equation = var_10441_equation_0, values = (var_10223_cast_fp16, var_10136_cast_fp16))[name = tensor("op_10441_cast_fp16")]; tensor var_10442_to_fp16 = const()[name = tensor("op_10442_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1043_cast_fp16 = mul(x = var_10441_cast_fp16, y = var_10442_to_fp16)[name = tensor("aw_chunk_1043_cast_fp16")]; tensor var_10445_equation_0 = const()[name = tensor("op_10445_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10445_cast_fp16 = einsum(equation = var_10445_equation_0, values = (var_10223_cast_fp16, var_10143_cast_fp16))[name = tensor("op_10445_cast_fp16")]; tensor var_10446_to_fp16 = const()[name = tensor("op_10446_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1045_cast_fp16 = mul(x = var_10445_cast_fp16, y = var_10446_to_fp16)[name = tensor("aw_chunk_1045_cast_fp16")]; tensor var_10449_equation_0 = const()[name = tensor("op_10449_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10449_cast_fp16 = einsum(equation = var_10449_equation_0, values = (var_10223_cast_fp16, var_10150_cast_fp16))[name = tensor("op_10449_cast_fp16")]; tensor var_10450_to_fp16 = const()[name = tensor("op_10450_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1047_cast_fp16 = mul(x = var_10449_cast_fp16, y = var_10450_to_fp16)[name = tensor("aw_chunk_1047_cast_fp16")]; tensor var_10453_equation_0 = const()[name = tensor("op_10453_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10453_cast_fp16 = einsum(equation = var_10453_equation_0, values = (var_10227_cast_fp16, var_10157_cast_fp16))[name = tensor("op_10453_cast_fp16")]; tensor var_10454_to_fp16 = const()[name = tensor("op_10454_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1049_cast_fp16 = mul(x = var_10453_cast_fp16, y = var_10454_to_fp16)[name = tensor("aw_chunk_1049_cast_fp16")]; tensor var_10457_equation_0 = const()[name = tensor("op_10457_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10457_cast_fp16 = einsum(equation = var_10457_equation_0, values = (var_10227_cast_fp16, var_10164_cast_fp16))[name = tensor("op_10457_cast_fp16")]; tensor var_10458_to_fp16 = const()[name = tensor("op_10458_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1051_cast_fp16 = mul(x = var_10457_cast_fp16, y = var_10458_to_fp16)[name = tensor("aw_chunk_1051_cast_fp16")]; tensor var_10461_equation_0 = const()[name = tensor("op_10461_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10461_cast_fp16 = einsum(equation = var_10461_equation_0, values = (var_10227_cast_fp16, var_10171_cast_fp16))[name = tensor("op_10461_cast_fp16")]; tensor var_10462_to_fp16 = const()[name = tensor("op_10462_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1053_cast_fp16 = mul(x = var_10461_cast_fp16, y = var_10462_to_fp16)[name = tensor("aw_chunk_1053_cast_fp16")]; tensor var_10465_equation_0 = const()[name = tensor("op_10465_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_10465_cast_fp16 = einsum(equation = var_10465_equation_0, values = (var_10227_cast_fp16, var_10178_cast_fp16))[name = tensor("op_10465_cast_fp16")]; tensor var_10466_to_fp16 = const()[name = tensor("op_10466_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1055_cast_fp16 = mul(x = var_10465_cast_fp16, y = var_10466_to_fp16)[name = tensor("aw_chunk_1055_cast_fp16")]; tensor var_10468_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_961_cast_fp16)[name = tensor("op_10468_cast_fp16")]; tensor var_10469_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_963_cast_fp16)[name = tensor("op_10469_cast_fp16")]; tensor var_10470_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_965_cast_fp16)[name = tensor("op_10470_cast_fp16")]; tensor var_10471_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_967_cast_fp16)[name = tensor("op_10471_cast_fp16")]; tensor var_10472_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_969_cast_fp16)[name = tensor("op_10472_cast_fp16")]; tensor var_10473_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_971_cast_fp16)[name = tensor("op_10473_cast_fp16")]; tensor var_10474_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_973_cast_fp16)[name = tensor("op_10474_cast_fp16")]; tensor var_10475_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_975_cast_fp16)[name = tensor("op_10475_cast_fp16")]; tensor var_10476_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_977_cast_fp16)[name = tensor("op_10476_cast_fp16")]; tensor var_10477_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_979_cast_fp16)[name = tensor("op_10477_cast_fp16")]; tensor var_10478_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_981_cast_fp16)[name = tensor("op_10478_cast_fp16")]; tensor var_10479_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_983_cast_fp16)[name = tensor("op_10479_cast_fp16")]; tensor var_10480_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_985_cast_fp16)[name = tensor("op_10480_cast_fp16")]; tensor var_10481_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_987_cast_fp16)[name = tensor("op_10481_cast_fp16")]; tensor var_10482_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_989_cast_fp16)[name = tensor("op_10482_cast_fp16")]; tensor var_10483_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_991_cast_fp16)[name = tensor("op_10483_cast_fp16")]; tensor var_10484_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_993_cast_fp16)[name = tensor("op_10484_cast_fp16")]; tensor var_10485_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_995_cast_fp16)[name = tensor("op_10485_cast_fp16")]; tensor var_10486_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_997_cast_fp16)[name = tensor("op_10486_cast_fp16")]; tensor var_10487_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_999_cast_fp16)[name = tensor("op_10487_cast_fp16")]; tensor var_10488_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1001_cast_fp16)[name = tensor("op_10488_cast_fp16")]; tensor var_10489_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1003_cast_fp16)[name = tensor("op_10489_cast_fp16")]; tensor var_10490_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1005_cast_fp16)[name = tensor("op_10490_cast_fp16")]; tensor var_10491_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1007_cast_fp16)[name = tensor("op_10491_cast_fp16")]; tensor var_10492_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1009_cast_fp16)[name = tensor("op_10492_cast_fp16")]; tensor var_10493_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1011_cast_fp16)[name = tensor("op_10493_cast_fp16")]; tensor var_10494_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1013_cast_fp16)[name = tensor("op_10494_cast_fp16")]; tensor var_10495_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1015_cast_fp16)[name = tensor("op_10495_cast_fp16")]; tensor var_10496_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1017_cast_fp16)[name = tensor("op_10496_cast_fp16")]; tensor var_10497_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1019_cast_fp16)[name = tensor("op_10497_cast_fp16")]; tensor var_10498_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1021_cast_fp16)[name = tensor("op_10498_cast_fp16")]; tensor var_10499_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1023_cast_fp16)[name = tensor("op_10499_cast_fp16")]; tensor var_10500_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1025_cast_fp16)[name = tensor("op_10500_cast_fp16")]; tensor var_10501_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1027_cast_fp16)[name = tensor("op_10501_cast_fp16")]; tensor var_10502_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1029_cast_fp16)[name = tensor("op_10502_cast_fp16")]; tensor var_10503_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1031_cast_fp16)[name = tensor("op_10503_cast_fp16")]; tensor var_10504_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1033_cast_fp16)[name = tensor("op_10504_cast_fp16")]; tensor var_10505_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1035_cast_fp16)[name = tensor("op_10505_cast_fp16")]; tensor var_10506_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1037_cast_fp16)[name = tensor("op_10506_cast_fp16")]; tensor var_10507_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1039_cast_fp16)[name = tensor("op_10507_cast_fp16")]; tensor var_10508_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1041_cast_fp16)[name = tensor("op_10508_cast_fp16")]; tensor var_10509_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1043_cast_fp16)[name = tensor("op_10509_cast_fp16")]; tensor var_10510_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1045_cast_fp16)[name = tensor("op_10510_cast_fp16")]; tensor var_10511_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1047_cast_fp16)[name = tensor("op_10511_cast_fp16")]; tensor var_10512_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1049_cast_fp16)[name = tensor("op_10512_cast_fp16")]; tensor var_10513_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1051_cast_fp16)[name = tensor("op_10513_cast_fp16")]; tensor var_10514_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1053_cast_fp16)[name = tensor("op_10514_cast_fp16")]; tensor var_10515_cast_fp16 = softmax(axis = var_9741, x = aw_chunk_1055_cast_fp16)[name = tensor("op_10515_cast_fp16")]; tensor var_10517_equation_0 = const()[name = tensor("op_10517_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10517_cast_fp16 = einsum(equation = var_10517_equation_0, values = (var_10229_cast_fp16, var_10468_cast_fp16))[name = tensor("op_10517_cast_fp16")]; tensor var_10519_equation_0 = const()[name = tensor("op_10519_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10519_cast_fp16 = einsum(equation = var_10519_equation_0, values = (var_10229_cast_fp16, var_10469_cast_fp16))[name = tensor("op_10519_cast_fp16")]; tensor var_10521_equation_0 = const()[name = tensor("op_10521_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10521_cast_fp16 = einsum(equation = var_10521_equation_0, values = (var_10229_cast_fp16, var_10470_cast_fp16))[name = tensor("op_10521_cast_fp16")]; tensor var_10523_equation_0 = const()[name = tensor("op_10523_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10523_cast_fp16 = einsum(equation = var_10523_equation_0, values = (var_10229_cast_fp16, var_10471_cast_fp16))[name = tensor("op_10523_cast_fp16")]; tensor var_10525_equation_0 = const()[name = tensor("op_10525_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10525_cast_fp16 = einsum(equation = var_10525_equation_0, values = (var_10233_cast_fp16, var_10472_cast_fp16))[name = tensor("op_10525_cast_fp16")]; tensor var_10527_equation_0 = const()[name = tensor("op_10527_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10527_cast_fp16 = einsum(equation = var_10527_equation_0, values = (var_10233_cast_fp16, var_10473_cast_fp16))[name = tensor("op_10527_cast_fp16")]; tensor var_10529_equation_0 = const()[name = tensor("op_10529_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10529_cast_fp16 = einsum(equation = var_10529_equation_0, values = (var_10233_cast_fp16, var_10474_cast_fp16))[name = tensor("op_10529_cast_fp16")]; tensor var_10531_equation_0 = const()[name = tensor("op_10531_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10531_cast_fp16 = einsum(equation = var_10531_equation_0, values = (var_10233_cast_fp16, var_10475_cast_fp16))[name = tensor("op_10531_cast_fp16")]; tensor var_10533_equation_0 = const()[name = tensor("op_10533_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10533_cast_fp16 = einsum(equation = var_10533_equation_0, values = (var_10237_cast_fp16, var_10476_cast_fp16))[name = tensor("op_10533_cast_fp16")]; tensor var_10535_equation_0 = const()[name = tensor("op_10535_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10535_cast_fp16 = einsum(equation = var_10535_equation_0, values = (var_10237_cast_fp16, var_10477_cast_fp16))[name = tensor("op_10535_cast_fp16")]; tensor var_10537_equation_0 = const()[name = tensor("op_10537_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10537_cast_fp16 = einsum(equation = var_10537_equation_0, values = (var_10237_cast_fp16, var_10478_cast_fp16))[name = tensor("op_10537_cast_fp16")]; tensor var_10539_equation_0 = const()[name = tensor("op_10539_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10539_cast_fp16 = einsum(equation = var_10539_equation_0, values = (var_10237_cast_fp16, var_10479_cast_fp16))[name = tensor("op_10539_cast_fp16")]; tensor var_10541_equation_0 = const()[name = tensor("op_10541_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10541_cast_fp16 = einsum(equation = var_10541_equation_0, values = (var_10241_cast_fp16, var_10480_cast_fp16))[name = tensor("op_10541_cast_fp16")]; tensor var_10543_equation_0 = const()[name = tensor("op_10543_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10543_cast_fp16 = einsum(equation = var_10543_equation_0, values = (var_10241_cast_fp16, var_10481_cast_fp16))[name = tensor("op_10543_cast_fp16")]; tensor var_10545_equation_0 = const()[name = tensor("op_10545_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10545_cast_fp16 = einsum(equation = var_10545_equation_0, values = (var_10241_cast_fp16, var_10482_cast_fp16))[name = tensor("op_10545_cast_fp16")]; tensor var_10547_equation_0 = const()[name = tensor("op_10547_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10547_cast_fp16 = einsum(equation = var_10547_equation_0, values = (var_10241_cast_fp16, var_10483_cast_fp16))[name = tensor("op_10547_cast_fp16")]; tensor var_10549_equation_0 = const()[name = tensor("op_10549_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10549_cast_fp16 = einsum(equation = var_10549_equation_0, values = (var_10245_cast_fp16, var_10484_cast_fp16))[name = tensor("op_10549_cast_fp16")]; tensor var_10551_equation_0 = const()[name = tensor("op_10551_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10551_cast_fp16 = einsum(equation = var_10551_equation_0, values = (var_10245_cast_fp16, var_10485_cast_fp16))[name = tensor("op_10551_cast_fp16")]; tensor var_10553_equation_0 = const()[name = tensor("op_10553_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10553_cast_fp16 = einsum(equation = var_10553_equation_0, values = (var_10245_cast_fp16, var_10486_cast_fp16))[name = tensor("op_10553_cast_fp16")]; tensor var_10555_equation_0 = const()[name = tensor("op_10555_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10555_cast_fp16 = einsum(equation = var_10555_equation_0, values = (var_10245_cast_fp16, var_10487_cast_fp16))[name = tensor("op_10555_cast_fp16")]; tensor var_10557_equation_0 = const()[name = tensor("op_10557_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10557_cast_fp16 = einsum(equation = var_10557_equation_0, values = (var_10249_cast_fp16, var_10488_cast_fp16))[name = tensor("op_10557_cast_fp16")]; tensor var_10559_equation_0 = const()[name = tensor("op_10559_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10559_cast_fp16 = einsum(equation = var_10559_equation_0, values = (var_10249_cast_fp16, var_10489_cast_fp16))[name = tensor("op_10559_cast_fp16")]; tensor var_10561_equation_0 = const()[name = tensor("op_10561_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10561_cast_fp16 = einsum(equation = var_10561_equation_0, values = (var_10249_cast_fp16, var_10490_cast_fp16))[name = tensor("op_10561_cast_fp16")]; tensor var_10563_equation_0 = const()[name = tensor("op_10563_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10563_cast_fp16 = einsum(equation = var_10563_equation_0, values = (var_10249_cast_fp16, var_10491_cast_fp16))[name = tensor("op_10563_cast_fp16")]; tensor var_10565_equation_0 = const()[name = tensor("op_10565_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10565_cast_fp16 = einsum(equation = var_10565_equation_0, values = (var_10253_cast_fp16, var_10492_cast_fp16))[name = tensor("op_10565_cast_fp16")]; tensor var_10567_equation_0 = const()[name = tensor("op_10567_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10567_cast_fp16 = einsum(equation = var_10567_equation_0, values = (var_10253_cast_fp16, var_10493_cast_fp16))[name = tensor("op_10567_cast_fp16")]; tensor var_10569_equation_0 = const()[name = tensor("op_10569_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10569_cast_fp16 = einsum(equation = var_10569_equation_0, values = (var_10253_cast_fp16, var_10494_cast_fp16))[name = tensor("op_10569_cast_fp16")]; tensor var_10571_equation_0 = const()[name = tensor("op_10571_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10571_cast_fp16 = einsum(equation = var_10571_equation_0, values = (var_10253_cast_fp16, var_10495_cast_fp16))[name = tensor("op_10571_cast_fp16")]; tensor var_10573_equation_0 = const()[name = tensor("op_10573_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10573_cast_fp16 = einsum(equation = var_10573_equation_0, values = (var_10257_cast_fp16, var_10496_cast_fp16))[name = tensor("op_10573_cast_fp16")]; tensor var_10575_equation_0 = const()[name = tensor("op_10575_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10575_cast_fp16 = einsum(equation = var_10575_equation_0, values = (var_10257_cast_fp16, var_10497_cast_fp16))[name = tensor("op_10575_cast_fp16")]; tensor var_10577_equation_0 = const()[name = tensor("op_10577_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10577_cast_fp16 = einsum(equation = var_10577_equation_0, values = (var_10257_cast_fp16, var_10498_cast_fp16))[name = tensor("op_10577_cast_fp16")]; tensor var_10579_equation_0 = const()[name = tensor("op_10579_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10579_cast_fp16 = einsum(equation = var_10579_equation_0, values = (var_10257_cast_fp16, var_10499_cast_fp16))[name = tensor("op_10579_cast_fp16")]; tensor var_10581_equation_0 = const()[name = tensor("op_10581_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10581_cast_fp16 = einsum(equation = var_10581_equation_0, values = (var_10261_cast_fp16, var_10500_cast_fp16))[name = tensor("op_10581_cast_fp16")]; tensor var_10583_equation_0 = const()[name = tensor("op_10583_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10583_cast_fp16 = einsum(equation = var_10583_equation_0, values = (var_10261_cast_fp16, var_10501_cast_fp16))[name = tensor("op_10583_cast_fp16")]; tensor var_10585_equation_0 = const()[name = tensor("op_10585_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10585_cast_fp16 = einsum(equation = var_10585_equation_0, values = (var_10261_cast_fp16, var_10502_cast_fp16))[name = tensor("op_10585_cast_fp16")]; tensor var_10587_equation_0 = const()[name = tensor("op_10587_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10587_cast_fp16 = einsum(equation = var_10587_equation_0, values = (var_10261_cast_fp16, var_10503_cast_fp16))[name = tensor("op_10587_cast_fp16")]; tensor var_10589_equation_0 = const()[name = tensor("op_10589_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10589_cast_fp16 = einsum(equation = var_10589_equation_0, values = (var_10265_cast_fp16, var_10504_cast_fp16))[name = tensor("op_10589_cast_fp16")]; tensor var_10591_equation_0 = const()[name = tensor("op_10591_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10591_cast_fp16 = einsum(equation = var_10591_equation_0, values = (var_10265_cast_fp16, var_10505_cast_fp16))[name = tensor("op_10591_cast_fp16")]; tensor var_10593_equation_0 = const()[name = tensor("op_10593_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10593_cast_fp16 = einsum(equation = var_10593_equation_0, values = (var_10265_cast_fp16, var_10506_cast_fp16))[name = tensor("op_10593_cast_fp16")]; tensor var_10595_equation_0 = const()[name = tensor("op_10595_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10595_cast_fp16 = einsum(equation = var_10595_equation_0, values = (var_10265_cast_fp16, var_10507_cast_fp16))[name = tensor("op_10595_cast_fp16")]; tensor var_10597_equation_0 = const()[name = tensor("op_10597_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10597_cast_fp16 = einsum(equation = var_10597_equation_0, values = (var_10269_cast_fp16, var_10508_cast_fp16))[name = tensor("op_10597_cast_fp16")]; tensor var_10599_equation_0 = const()[name = tensor("op_10599_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10599_cast_fp16 = einsum(equation = var_10599_equation_0, values = (var_10269_cast_fp16, var_10509_cast_fp16))[name = tensor("op_10599_cast_fp16")]; tensor var_10601_equation_0 = const()[name = tensor("op_10601_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10601_cast_fp16 = einsum(equation = var_10601_equation_0, values = (var_10269_cast_fp16, var_10510_cast_fp16))[name = tensor("op_10601_cast_fp16")]; tensor var_10603_equation_0 = const()[name = tensor("op_10603_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10603_cast_fp16 = einsum(equation = var_10603_equation_0, values = (var_10269_cast_fp16, var_10511_cast_fp16))[name = tensor("op_10603_cast_fp16")]; tensor var_10605_equation_0 = const()[name = tensor("op_10605_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10605_cast_fp16 = einsum(equation = var_10605_equation_0, values = (var_10273_cast_fp16, var_10512_cast_fp16))[name = tensor("op_10605_cast_fp16")]; tensor var_10607_equation_0 = const()[name = tensor("op_10607_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10607_cast_fp16 = einsum(equation = var_10607_equation_0, values = (var_10273_cast_fp16, var_10513_cast_fp16))[name = tensor("op_10607_cast_fp16")]; tensor var_10609_equation_0 = const()[name = tensor("op_10609_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10609_cast_fp16 = einsum(equation = var_10609_equation_0, values = (var_10273_cast_fp16, var_10514_cast_fp16))[name = tensor("op_10609_cast_fp16")]; tensor var_10611_equation_0 = const()[name = tensor("op_10611_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_10611_cast_fp16 = einsum(equation = var_10611_equation_0, values = (var_10273_cast_fp16, var_10515_cast_fp16))[name = tensor("op_10611_cast_fp16")]; tensor var_10613_interleave_0 = const()[name = tensor("op_10613_interleave_0"), val = tensor(false)]; tensor var_10613_cast_fp16 = concat(axis = var_9724, interleave = var_10613_interleave_0, values = (var_10517_cast_fp16, var_10519_cast_fp16, var_10521_cast_fp16, var_10523_cast_fp16))[name = tensor("op_10613_cast_fp16")]; tensor var_10615_interleave_0 = const()[name = tensor("op_10615_interleave_0"), val = tensor(false)]; tensor var_10615_cast_fp16 = concat(axis = var_9724, interleave = var_10615_interleave_0, values = (var_10525_cast_fp16, var_10527_cast_fp16, var_10529_cast_fp16, var_10531_cast_fp16))[name = tensor("op_10615_cast_fp16")]; tensor var_10617_interleave_0 = const()[name = tensor("op_10617_interleave_0"), val = tensor(false)]; tensor var_10617_cast_fp16 = concat(axis = var_9724, interleave = var_10617_interleave_0, values = (var_10533_cast_fp16, var_10535_cast_fp16, var_10537_cast_fp16, var_10539_cast_fp16))[name = tensor("op_10617_cast_fp16")]; tensor var_10619_interleave_0 = const()[name = tensor("op_10619_interleave_0"), val = tensor(false)]; tensor var_10619_cast_fp16 = concat(axis = var_9724, interleave = var_10619_interleave_0, values = (var_10541_cast_fp16, var_10543_cast_fp16, var_10545_cast_fp16, var_10547_cast_fp16))[name = tensor("op_10619_cast_fp16")]; tensor var_10621_interleave_0 = const()[name = tensor("op_10621_interleave_0"), val = tensor(false)]; tensor var_10621_cast_fp16 = concat(axis = var_9724, interleave = var_10621_interleave_0, values = (var_10549_cast_fp16, var_10551_cast_fp16, var_10553_cast_fp16, var_10555_cast_fp16))[name = tensor("op_10621_cast_fp16")]; tensor var_10623_interleave_0 = const()[name = tensor("op_10623_interleave_0"), val = tensor(false)]; tensor var_10623_cast_fp16 = concat(axis = var_9724, interleave = var_10623_interleave_0, values = (var_10557_cast_fp16, var_10559_cast_fp16, var_10561_cast_fp16, var_10563_cast_fp16))[name = tensor("op_10623_cast_fp16")]; tensor var_10625_interleave_0 = const()[name = tensor("op_10625_interleave_0"), val = tensor(false)]; tensor var_10625_cast_fp16 = concat(axis = var_9724, interleave = var_10625_interleave_0, values = (var_10565_cast_fp16, var_10567_cast_fp16, var_10569_cast_fp16, var_10571_cast_fp16))[name = tensor("op_10625_cast_fp16")]; tensor var_10627_interleave_0 = const()[name = tensor("op_10627_interleave_0"), val = tensor(false)]; tensor var_10627_cast_fp16 = concat(axis = var_9724, interleave = var_10627_interleave_0, values = (var_10573_cast_fp16, var_10575_cast_fp16, var_10577_cast_fp16, var_10579_cast_fp16))[name = tensor("op_10627_cast_fp16")]; tensor var_10629_interleave_0 = const()[name = tensor("op_10629_interleave_0"), val = tensor(false)]; tensor var_10629_cast_fp16 = concat(axis = var_9724, interleave = var_10629_interleave_0, values = (var_10581_cast_fp16, var_10583_cast_fp16, var_10585_cast_fp16, var_10587_cast_fp16))[name = tensor("op_10629_cast_fp16")]; tensor var_10631_interleave_0 = const()[name = tensor("op_10631_interleave_0"), val = tensor(false)]; tensor var_10631_cast_fp16 = concat(axis = var_9724, interleave = var_10631_interleave_0, values = (var_10589_cast_fp16, var_10591_cast_fp16, var_10593_cast_fp16, var_10595_cast_fp16))[name = tensor("op_10631_cast_fp16")]; tensor var_10633_interleave_0 = const()[name = tensor("op_10633_interleave_0"), val = tensor(false)]; tensor var_10633_cast_fp16 = concat(axis = var_9724, interleave = var_10633_interleave_0, values = (var_10597_cast_fp16, var_10599_cast_fp16, var_10601_cast_fp16, var_10603_cast_fp16))[name = tensor("op_10633_cast_fp16")]; tensor var_10635_interleave_0 = const()[name = tensor("op_10635_interleave_0"), val = tensor(false)]; tensor var_10635_cast_fp16 = concat(axis = var_9724, interleave = var_10635_interleave_0, values = (var_10605_cast_fp16, var_10607_cast_fp16, var_10609_cast_fp16, var_10611_cast_fp16))[name = tensor("op_10635_cast_fp16")]; tensor input_81_interleave_0 = const()[name = tensor("input_81_interleave_0"), val = tensor(false)]; tensor input_81_cast_fp16 = concat(axis = var_9741, interleave = input_81_interleave_0, values = (var_10613_cast_fp16, var_10615_cast_fp16, var_10617_cast_fp16, var_10619_cast_fp16, var_10621_cast_fp16, var_10623_cast_fp16, var_10625_cast_fp16, var_10627_cast_fp16, var_10629_cast_fp16, var_10631_cast_fp16, var_10633_cast_fp16, var_10635_cast_fp16))[name = tensor("input_81_cast_fp16")]; tensor var_10640 = const()[name = tensor("op_10640"), val = tensor([1, 1])]; tensor var_10642 = const()[name = tensor("op_10642"), val = tensor([1, 1])]; tensor obj_43_pad_type_0 = const()[name = tensor("obj_43_pad_type_0"), val = tensor("custom")]; tensor obj_43_pad_0 = const()[name = tensor("obj_43_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151515456)))]; tensor layers_10_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152695168)))]; tensor obj_43_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_bias_to_fp16, dilations = var_10642, groups = var_9741, pad = obj_43_pad_0, pad_type = obj_43_pad_type_0, strides = var_10640, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_81_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_10648 = const()[name = tensor("op_10648"), val = tensor([1])]; tensor channels_mean_43_cast_fp16 = reduce_mean(axes = var_10648, keep_dims = var_9742, 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_10652 = const()[name = tensor("op_10652"), val = tensor([1])]; tensor var_10653_cast_fp16 = reduce_mean(axes = var_10652, keep_dims = var_9742, x = zero_mean_sq_43_cast_fp16)[name = tensor("op_10653_cast_fp16")]; tensor var_10654_to_fp16 = const()[name = tensor("op_10654_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10655_cast_fp16 = add(x = var_10653_cast_fp16, y = var_10654_to_fp16)[name = tensor("op_10655_cast_fp16")]; tensor denom_43_epsilon_0_to_fp16 = const()[name = tensor("denom_43_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_43_cast_fp16 = rsqrt(epsilon = denom_43_epsilon_0_to_fp16, x = var_10655_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_83_gamma_0_to_fp16 = const()[name = tensor("input_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152696768)))]; tensor input_83_beta_0_to_fp16 = const()[name = tensor("input_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152698368)))]; tensor input_83_epsilon_0_to_fp16 = const()[name = tensor("input_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; tensor input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_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_83_cast_fp16")]; tensor var_10666 = const()[name = tensor("op_10666"), val = tensor([1, 1])]; tensor var_10668 = const()[name = tensor("op_10668"), 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_10_fc1_weight_to_fp16 = const()[name = tensor("layers_10_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152699968)))]; tensor layers_10_fc1_bias_to_fp16 = const()[name = tensor("layers_10_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157418624)))]; tensor input_85_cast_fp16 = conv(bias = layers_10_fc1_bias_to_fp16, dilations = var_10668, groups = var_9741, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = var_10666, weight = layers_10_fc1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("input_85_cast_fp16")]; tensor input_87_mode_0 = const()[name = tensor("input_87_mode_0"), val = tensor("EXACT")]; tensor input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; tensor var_10674 = const()[name = tensor("op_10674"), val = tensor([1, 1])]; tensor var_10676 = const()[name = tensor("op_10676"), val = tensor([1, 1])]; tensor hidden_states_25_pad_type_0 = const()[name = tensor("hidden_states_25_pad_type_0"), val = tensor("custom")]; tensor hidden_states_25_pad_0 = const()[name = tensor("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_10_fc2_weight_to_fp16 = const()[name = tensor("layers_10_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157424832)))]; tensor layers_10_fc2_bias_to_fp16 = const()[name = tensor("layers_10_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162143488)))]; tensor hidden_states_25_cast_fp16 = conv(bias = layers_10_fc2_bias_to_fp16, dilations = var_10676, groups = var_9741, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_10674, weight = layers_10_fc2_weight_to_fp16, x = input_87_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_10683 = const()[name = tensor("op_10683"), val = tensor(3)]; tensor var_10700 = const()[name = tensor("op_10700"), val = tensor(1)]; tensor var_10701 = const()[name = tensor("op_10701"), val = tensor(true)]; tensor var_10711 = const()[name = tensor("op_10711"), val = tensor([1])]; tensor channels_mean_45_cast_fp16 = reduce_mean(axes = var_10711, keep_dims = var_10701, 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_10715 = const()[name = tensor("op_10715"), val = tensor([1])]; tensor var_10716_cast_fp16 = reduce_mean(axes = var_10715, keep_dims = var_10701, x = zero_mean_sq_45_cast_fp16)[name = tensor("op_10716_cast_fp16")]; tensor var_10717_to_fp16 = const()[name = tensor("op_10717_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_10718_cast_fp16 = add(x = var_10716_cast_fp16, y = var_10717_to_fp16)[name = tensor("op_10718_cast_fp16")]; tensor denom_45_epsilon_0_to_fp16 = const()[name = tensor("denom_45_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_45_cast_fp16 = rsqrt(epsilon = denom_45_epsilon_0_to_fp16, x = var_10718_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(162145088)))]; 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(162146688)))]; 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_10733 = const()[name = tensor("op_10733"), val = tensor([1, 1])]; tensor var_10735 = const()[name = tensor("op_10735"), val = tensor([1, 1])]; tensor query_pad_type_0 = const()[name = tensor("query_pad_type_0"), val = tensor("custom")]; tensor query_pad_0 = const()[name = tensor("query_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162148288)))]; tensor layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163328000)))]; tensor query_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_bias_to_fp16, dilations = var_10735, groups = var_10700, pad = query_pad_0, pad_type = query_pad_type_0, strides = var_10733, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("query_cast_fp16")]; tensor var_10739 = const()[name = tensor("op_10739"), val = tensor([1, 1])]; tensor var_10741 = const()[name = tensor("op_10741"), val = tensor([1, 1])]; tensor key_pad_type_0 = const()[name = tensor("key_pad_type_0"), val = tensor("custom")]; tensor key_pad_0 = const()[name = tensor("key_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163329600)))]; tensor key_cast_fp16 = conv(dilations = var_10741, groups = var_10700, pad = key_pad_0, pad_type = key_pad_type_0, strides = var_10739, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("key_cast_fp16")]; tensor var_10746 = const()[name = tensor("op_10746"), val = tensor([1, 1])]; tensor var_10748 = const()[name = tensor("op_10748"), val = tensor([1, 1])]; tensor value_pad_type_0 = const()[name = tensor("value_pad_type_0"), val = tensor("custom")]; tensor value_pad_0 = const()[name = tensor("value_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164509312)))]; tensor layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165689024)))]; tensor value_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_bias_to_fp16, dilations = var_10748, groups = var_10700, pad = value_pad_0, pad_type = value_pad_type_0, strides = var_10746, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = obj_45_cast_fp16)[name = tensor("value_cast_fp16")]; tensor var_10755_begin_0 = const()[name = tensor("op_10755_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10755_end_0 = const()[name = tensor("op_10755_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10755_end_mask_0 = const()[name = tensor("op_10755_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10755_cast_fp16 = slice_by_index(begin = var_10755_begin_0, end = var_10755_end_0, end_mask = var_10755_end_mask_0, x = query_cast_fp16)[name = tensor("op_10755_cast_fp16")]; tensor var_10759_begin_0 = const()[name = tensor("op_10759_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_10759_end_0 = const()[name = tensor("op_10759_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_10759_end_mask_0 = const()[name = tensor("op_10759_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10759_cast_fp16 = slice_by_index(begin = var_10759_begin_0, end = var_10759_end_0, end_mask = var_10759_end_mask_0, x = query_cast_fp16)[name = tensor("op_10759_cast_fp16")]; tensor var_10763_begin_0 = const()[name = tensor("op_10763_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_10763_end_0 = const()[name = tensor("op_10763_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_10763_end_mask_0 = const()[name = tensor("op_10763_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10763_cast_fp16 = slice_by_index(begin = var_10763_begin_0, end = var_10763_end_0, end_mask = var_10763_end_mask_0, x = query_cast_fp16)[name = tensor("op_10763_cast_fp16")]; tensor var_10767_begin_0 = const()[name = tensor("op_10767_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_10767_end_0 = const()[name = tensor("op_10767_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_10767_end_mask_0 = const()[name = tensor("op_10767_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10767_cast_fp16 = slice_by_index(begin = var_10767_begin_0, end = var_10767_end_0, end_mask = var_10767_end_mask_0, x = query_cast_fp16)[name = tensor("op_10767_cast_fp16")]; tensor var_10771_begin_0 = const()[name = tensor("op_10771_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_10771_end_0 = const()[name = tensor("op_10771_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_10771_end_mask_0 = const()[name = tensor("op_10771_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10771_cast_fp16 = slice_by_index(begin = var_10771_begin_0, end = var_10771_end_0, end_mask = var_10771_end_mask_0, x = query_cast_fp16)[name = tensor("op_10771_cast_fp16")]; tensor var_10775_begin_0 = const()[name = tensor("op_10775_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_10775_end_0 = const()[name = tensor("op_10775_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_10775_end_mask_0 = const()[name = tensor("op_10775_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10775_cast_fp16 = slice_by_index(begin = var_10775_begin_0, end = var_10775_end_0, end_mask = var_10775_end_mask_0, x = query_cast_fp16)[name = tensor("op_10775_cast_fp16")]; tensor var_10779_begin_0 = const()[name = tensor("op_10779_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_10779_end_0 = const()[name = tensor("op_10779_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_10779_end_mask_0 = const()[name = tensor("op_10779_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10779_cast_fp16 = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = query_cast_fp16)[name = tensor("op_10779_cast_fp16")]; tensor var_10783_begin_0 = const()[name = tensor("op_10783_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_10783_end_0 = const()[name = tensor("op_10783_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_10783_end_mask_0 = const()[name = tensor("op_10783_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10783_cast_fp16 = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = query_cast_fp16)[name = tensor("op_10783_cast_fp16")]; tensor var_10787_begin_0 = const()[name = tensor("op_10787_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_10787_end_0 = const()[name = tensor("op_10787_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_10787_end_mask_0 = const()[name = tensor("op_10787_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10787_cast_fp16 = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = query_cast_fp16)[name = tensor("op_10787_cast_fp16")]; tensor var_10791_begin_0 = const()[name = tensor("op_10791_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_10791_end_0 = const()[name = tensor("op_10791_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_10791_end_mask_0 = const()[name = tensor("op_10791_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10791_cast_fp16 = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = query_cast_fp16)[name = tensor("op_10791_cast_fp16")]; tensor var_10795_begin_0 = const()[name = tensor("op_10795_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_10795_end_0 = const()[name = tensor("op_10795_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_10795_end_mask_0 = const()[name = tensor("op_10795_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10795_cast_fp16 = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = query_cast_fp16)[name = tensor("op_10795_cast_fp16")]; tensor var_10799_begin_0 = const()[name = tensor("op_10799_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_10799_end_0 = const()[name = tensor("op_10799_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_10799_end_mask_0 = const()[name = tensor("op_10799_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_10799_cast_fp16 = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = query_cast_fp16)[name = tensor("op_10799_cast_fp16")]; tensor var_10808_begin_0 = const()[name = tensor("op_10808_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10808_end_0 = const()[name = tensor("op_10808_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10808_end_mask_0 = const()[name = tensor("op_10808_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10808_cast_fp16 = slice_by_index(begin = var_10808_begin_0, end = var_10808_end_0, end_mask = var_10808_end_mask_0, x = var_10755_cast_fp16)[name = tensor("op_10808_cast_fp16")]; tensor var_10815_begin_0 = const()[name = tensor("op_10815_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10815_end_0 = const()[name = tensor("op_10815_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10815_end_mask_0 = const()[name = tensor("op_10815_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10815_cast_fp16 = slice_by_index(begin = var_10815_begin_0, end = var_10815_end_0, end_mask = var_10815_end_mask_0, x = var_10755_cast_fp16)[name = tensor("op_10815_cast_fp16")]; tensor var_10822_begin_0 = const()[name = tensor("op_10822_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10822_end_0 = const()[name = tensor("op_10822_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10822_end_mask_0 = const()[name = tensor("op_10822_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10822_cast_fp16 = slice_by_index(begin = var_10822_begin_0, end = var_10822_end_0, end_mask = var_10822_end_mask_0, x = var_10755_cast_fp16)[name = tensor("op_10822_cast_fp16")]; tensor var_10829_begin_0 = const()[name = tensor("op_10829_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10829_end_0 = const()[name = tensor("op_10829_end_0"), val = tensor([1, 64, 1, 1500])]; 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 = var_10755_cast_fp16)[name = tensor("op_10829_cast_fp16")]; tensor var_10836_begin_0 = const()[name = tensor("op_10836_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10836_end_0 = const()[name = tensor("op_10836_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10836_end_mask_0 = const()[name = tensor("op_10836_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10836_cast_fp16 = slice_by_index(begin = var_10836_begin_0, end = var_10836_end_0, end_mask = var_10836_end_mask_0, x = var_10759_cast_fp16)[name = tensor("op_10836_cast_fp16")]; tensor var_10843_begin_0 = const()[name = tensor("op_10843_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10843_end_0 = const()[name = tensor("op_10843_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10843_end_mask_0 = const()[name = tensor("op_10843_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_10759_cast_fp16)[name = tensor("op_10843_cast_fp16")]; tensor var_10850_begin_0 = const()[name = tensor("op_10850_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10850_end_0 = const()[name = tensor("op_10850_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10850_end_mask_0 = const()[name = tensor("op_10850_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10850_cast_fp16 = slice_by_index(begin = var_10850_begin_0, end = var_10850_end_0, end_mask = var_10850_end_mask_0, x = var_10759_cast_fp16)[name = tensor("op_10850_cast_fp16")]; tensor var_10857_begin_0 = const()[name = tensor("op_10857_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10857_end_0 = const()[name = tensor("op_10857_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10857_end_mask_0 = const()[name = tensor("op_10857_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10857_cast_fp16 = slice_by_index(begin = var_10857_begin_0, end = var_10857_end_0, end_mask = var_10857_end_mask_0, x = var_10759_cast_fp16)[name = tensor("op_10857_cast_fp16")]; tensor var_10864_begin_0 = const()[name = tensor("op_10864_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10864_end_0 = const()[name = tensor("op_10864_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10864_end_mask_0 = const()[name = tensor("op_10864_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10864_cast_fp16 = slice_by_index(begin = var_10864_begin_0, end = var_10864_end_0, end_mask = var_10864_end_mask_0, x = var_10763_cast_fp16)[name = tensor("op_10864_cast_fp16")]; tensor var_10871_begin_0 = const()[name = tensor("op_10871_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10871_end_0 = const()[name = tensor("op_10871_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10871_end_mask_0 = const()[name = tensor("op_10871_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_10763_cast_fp16)[name = tensor("op_10871_cast_fp16")]; tensor var_10878_begin_0 = const()[name = tensor("op_10878_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10878_end_0 = const()[name = tensor("op_10878_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10878_end_mask_0 = const()[name = tensor("op_10878_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10878_cast_fp16 = slice_by_index(begin = var_10878_begin_0, end = var_10878_end_0, end_mask = var_10878_end_mask_0, x = var_10763_cast_fp16)[name = tensor("op_10878_cast_fp16")]; tensor var_10885_begin_0 = const()[name = tensor("op_10885_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10885_end_0 = const()[name = tensor("op_10885_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10885_end_mask_0 = const()[name = tensor("op_10885_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10885_cast_fp16 = slice_by_index(begin = var_10885_begin_0, end = var_10885_end_0, end_mask = var_10885_end_mask_0, x = var_10763_cast_fp16)[name = tensor("op_10885_cast_fp16")]; tensor var_10892_begin_0 = const()[name = tensor("op_10892_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10892_end_0 = const()[name = tensor("op_10892_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10892_end_mask_0 = const()[name = tensor("op_10892_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10892_cast_fp16 = slice_by_index(begin = var_10892_begin_0, end = var_10892_end_0, end_mask = var_10892_end_mask_0, x = var_10767_cast_fp16)[name = tensor("op_10892_cast_fp16")]; tensor var_10899_begin_0 = const()[name = tensor("op_10899_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10899_end_0 = const()[name = tensor("op_10899_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10899_end_mask_0 = const()[name = tensor("op_10899_end_mask_0"), val = tensor([true, true, true, false])]; 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 = var_10767_cast_fp16)[name = tensor("op_10899_cast_fp16")]; tensor var_10906_begin_0 = const()[name = tensor("op_10906_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10906_end_0 = const()[name = tensor("op_10906_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10906_end_mask_0 = const()[name = tensor("op_10906_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10906_cast_fp16 = slice_by_index(begin = var_10906_begin_0, end = var_10906_end_0, end_mask = var_10906_end_mask_0, x = var_10767_cast_fp16)[name = tensor("op_10906_cast_fp16")]; tensor var_10913_begin_0 = const()[name = tensor("op_10913_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10913_end_0 = const()[name = tensor("op_10913_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10913_end_mask_0 = const()[name = tensor("op_10913_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10913_cast_fp16 = slice_by_index(begin = var_10913_begin_0, end = var_10913_end_0, end_mask = var_10913_end_mask_0, x = var_10767_cast_fp16)[name = tensor("op_10913_cast_fp16")]; tensor var_10920_begin_0 = const()[name = tensor("op_10920_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10920_end_0 = const()[name = tensor("op_10920_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10920_end_mask_0 = const()[name = tensor("op_10920_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10920_cast_fp16 = slice_by_index(begin = var_10920_begin_0, end = var_10920_end_0, end_mask = var_10920_end_mask_0, x = var_10771_cast_fp16)[name = tensor("op_10920_cast_fp16")]; tensor var_10927_begin_0 = const()[name = tensor("op_10927_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10927_end_0 = const()[name = tensor("op_10927_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10927_end_mask_0 = const()[name = tensor("op_10927_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10927_cast_fp16 = slice_by_index(begin = var_10927_begin_0, end = var_10927_end_0, end_mask = var_10927_end_mask_0, x = var_10771_cast_fp16)[name = tensor("op_10927_cast_fp16")]; tensor var_10934_begin_0 = const()[name = tensor("op_10934_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10934_end_0 = const()[name = tensor("op_10934_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10934_end_mask_0 = const()[name = tensor("op_10934_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10934_cast_fp16 = slice_by_index(begin = var_10934_begin_0, end = var_10934_end_0, end_mask = var_10934_end_mask_0, x = var_10771_cast_fp16)[name = tensor("op_10934_cast_fp16")]; tensor var_10941_begin_0 = const()[name = tensor("op_10941_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10941_end_0 = const()[name = tensor("op_10941_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10941_end_mask_0 = const()[name = tensor("op_10941_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10941_cast_fp16 = slice_by_index(begin = var_10941_begin_0, end = var_10941_end_0, end_mask = var_10941_end_mask_0, x = var_10771_cast_fp16)[name = tensor("op_10941_cast_fp16")]; tensor var_10948_begin_0 = const()[name = tensor("op_10948_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10948_end_0 = const()[name = tensor("op_10948_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10948_end_mask_0 = const()[name = tensor("op_10948_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10948_cast_fp16 = slice_by_index(begin = var_10948_begin_0, end = var_10948_end_0, end_mask = var_10948_end_mask_0, x = var_10775_cast_fp16)[name = tensor("op_10948_cast_fp16")]; tensor var_10955_begin_0 = const()[name = tensor("op_10955_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10955_end_0 = const()[name = tensor("op_10955_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10955_end_mask_0 = const()[name = tensor("op_10955_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10955_cast_fp16 = slice_by_index(begin = var_10955_begin_0, end = var_10955_end_0, end_mask = var_10955_end_mask_0, x = var_10775_cast_fp16)[name = tensor("op_10955_cast_fp16")]; tensor var_10962_begin_0 = const()[name = tensor("op_10962_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10962_end_0 = const()[name = tensor("op_10962_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10962_end_mask_0 = const()[name = tensor("op_10962_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10962_cast_fp16 = slice_by_index(begin = var_10962_begin_0, end = var_10962_end_0, end_mask = var_10962_end_mask_0, x = var_10775_cast_fp16)[name = tensor("op_10962_cast_fp16")]; tensor var_10969_begin_0 = const()[name = tensor("op_10969_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10969_end_0 = const()[name = tensor("op_10969_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10969_end_mask_0 = const()[name = tensor("op_10969_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10969_cast_fp16 = slice_by_index(begin = var_10969_begin_0, end = var_10969_end_0, end_mask = var_10969_end_mask_0, x = var_10775_cast_fp16)[name = tensor("op_10969_cast_fp16")]; tensor var_10976_begin_0 = const()[name = tensor("op_10976_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_10976_end_0 = const()[name = tensor("op_10976_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_10976_end_mask_0 = const()[name = tensor("op_10976_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10976_cast_fp16 = slice_by_index(begin = var_10976_begin_0, end = var_10976_end_0, end_mask = var_10976_end_mask_0, x = var_10779_cast_fp16)[name = tensor("op_10976_cast_fp16")]; tensor var_10983_begin_0 = const()[name = tensor("op_10983_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_10983_end_0 = const()[name = tensor("op_10983_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_10983_end_mask_0 = const()[name = tensor("op_10983_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10983_cast_fp16 = slice_by_index(begin = var_10983_begin_0, end = var_10983_end_0, end_mask = var_10983_end_mask_0, x = var_10779_cast_fp16)[name = tensor("op_10983_cast_fp16")]; tensor var_10990_begin_0 = const()[name = tensor("op_10990_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_10990_end_0 = const()[name = tensor("op_10990_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_10990_end_mask_0 = const()[name = tensor("op_10990_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10990_cast_fp16 = slice_by_index(begin = var_10990_begin_0, end = var_10990_end_0, end_mask = var_10990_end_mask_0, x = var_10779_cast_fp16)[name = tensor("op_10990_cast_fp16")]; tensor var_10997_begin_0 = const()[name = tensor("op_10997_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_10997_end_0 = const()[name = tensor("op_10997_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_10997_end_mask_0 = const()[name = tensor("op_10997_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_10997_cast_fp16 = slice_by_index(begin = var_10997_begin_0, end = var_10997_end_0, end_mask = var_10997_end_mask_0, x = var_10779_cast_fp16)[name = tensor("op_10997_cast_fp16")]; tensor var_11004_begin_0 = const()[name = tensor("op_11004_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11004_end_0 = const()[name = tensor("op_11004_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11004_end_mask_0 = const()[name = tensor("op_11004_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11004_cast_fp16 = slice_by_index(begin = var_11004_begin_0, end = var_11004_end_0, end_mask = var_11004_end_mask_0, x = var_10783_cast_fp16)[name = tensor("op_11004_cast_fp16")]; tensor var_11011_begin_0 = const()[name = tensor("op_11011_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11011_end_0 = const()[name = tensor("op_11011_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11011_end_mask_0 = const()[name = tensor("op_11011_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11011_cast_fp16 = slice_by_index(begin = var_11011_begin_0, end = var_11011_end_0, end_mask = var_11011_end_mask_0, x = var_10783_cast_fp16)[name = tensor("op_11011_cast_fp16")]; tensor var_11018_begin_0 = const()[name = tensor("op_11018_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11018_end_0 = const()[name = tensor("op_11018_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11018_end_mask_0 = const()[name = tensor("op_11018_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11018_cast_fp16 = slice_by_index(begin = var_11018_begin_0, end = var_11018_end_0, end_mask = var_11018_end_mask_0, x = var_10783_cast_fp16)[name = tensor("op_11018_cast_fp16")]; tensor var_11025_begin_0 = const()[name = tensor("op_11025_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11025_end_0 = const()[name = tensor("op_11025_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11025_end_mask_0 = const()[name = tensor("op_11025_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11025_cast_fp16 = slice_by_index(begin = var_11025_begin_0, end = var_11025_end_0, end_mask = var_11025_end_mask_0, x = var_10783_cast_fp16)[name = tensor("op_11025_cast_fp16")]; tensor var_11032_begin_0 = const()[name = tensor("op_11032_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11032_end_0 = const()[name = tensor("op_11032_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11032_end_mask_0 = const()[name = tensor("op_11032_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11032_cast_fp16 = slice_by_index(begin = var_11032_begin_0, end = var_11032_end_0, end_mask = var_11032_end_mask_0, x = var_10787_cast_fp16)[name = tensor("op_11032_cast_fp16")]; tensor var_11039_begin_0 = const()[name = tensor("op_11039_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11039_end_0 = const()[name = tensor("op_11039_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11039_end_mask_0 = const()[name = tensor("op_11039_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11039_cast_fp16 = slice_by_index(begin = var_11039_begin_0, end = var_11039_end_0, end_mask = var_11039_end_mask_0, x = var_10787_cast_fp16)[name = tensor("op_11039_cast_fp16")]; tensor var_11046_begin_0 = const()[name = tensor("op_11046_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11046_end_0 = const()[name = tensor("op_11046_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11046_end_mask_0 = const()[name = tensor("op_11046_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11046_cast_fp16 = slice_by_index(begin = var_11046_begin_0, end = var_11046_end_0, end_mask = var_11046_end_mask_0, x = var_10787_cast_fp16)[name = tensor("op_11046_cast_fp16")]; tensor var_11053_begin_0 = const()[name = tensor("op_11053_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11053_end_0 = const()[name = tensor("op_11053_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11053_end_mask_0 = const()[name = tensor("op_11053_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11053_cast_fp16 = slice_by_index(begin = var_11053_begin_0, end = var_11053_end_0, end_mask = var_11053_end_mask_0, x = var_10787_cast_fp16)[name = tensor("op_11053_cast_fp16")]; tensor var_11060_begin_0 = const()[name = tensor("op_11060_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11060_end_0 = const()[name = tensor("op_11060_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11060_end_mask_0 = const()[name = tensor("op_11060_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11060_cast_fp16 = slice_by_index(begin = var_11060_begin_0, end = var_11060_end_0, end_mask = var_11060_end_mask_0, x = var_10791_cast_fp16)[name = tensor("op_11060_cast_fp16")]; tensor var_11067_begin_0 = const()[name = tensor("op_11067_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11067_end_0 = const()[name = tensor("op_11067_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11067_end_mask_0 = const()[name = tensor("op_11067_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11067_cast_fp16 = slice_by_index(begin = var_11067_begin_0, end = var_11067_end_0, end_mask = var_11067_end_mask_0, x = var_10791_cast_fp16)[name = tensor("op_11067_cast_fp16")]; tensor var_11074_begin_0 = const()[name = tensor("op_11074_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11074_end_0 = const()[name = tensor("op_11074_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11074_end_mask_0 = const()[name = tensor("op_11074_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11074_cast_fp16 = slice_by_index(begin = var_11074_begin_0, end = var_11074_end_0, end_mask = var_11074_end_mask_0, x = var_10791_cast_fp16)[name = tensor("op_11074_cast_fp16")]; tensor var_11081_begin_0 = const()[name = tensor("op_11081_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11081_end_0 = const()[name = tensor("op_11081_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11081_end_mask_0 = const()[name = tensor("op_11081_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11081_cast_fp16 = slice_by_index(begin = var_11081_begin_0, end = var_11081_end_0, end_mask = var_11081_end_mask_0, x = var_10791_cast_fp16)[name = tensor("op_11081_cast_fp16")]; tensor var_11088_begin_0 = const()[name = tensor("op_11088_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11088_end_0 = const()[name = tensor("op_11088_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11088_end_mask_0 = const()[name = tensor("op_11088_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11088_cast_fp16 = slice_by_index(begin = var_11088_begin_0, end = var_11088_end_0, end_mask = var_11088_end_mask_0, x = var_10795_cast_fp16)[name = tensor("op_11088_cast_fp16")]; tensor var_11095_begin_0 = const()[name = tensor("op_11095_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11095_end_0 = const()[name = tensor("op_11095_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11095_end_mask_0 = const()[name = tensor("op_11095_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11095_cast_fp16 = slice_by_index(begin = var_11095_begin_0, end = var_11095_end_0, end_mask = var_11095_end_mask_0, x = var_10795_cast_fp16)[name = tensor("op_11095_cast_fp16")]; tensor var_11102_begin_0 = const()[name = tensor("op_11102_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11102_end_0 = const()[name = tensor("op_11102_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11102_end_mask_0 = const()[name = tensor("op_11102_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11102_cast_fp16 = slice_by_index(begin = var_11102_begin_0, end = var_11102_end_0, end_mask = var_11102_end_mask_0, x = var_10795_cast_fp16)[name = tensor("op_11102_cast_fp16")]; tensor var_11109_begin_0 = const()[name = tensor("op_11109_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11109_end_0 = const()[name = tensor("op_11109_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11109_end_mask_0 = const()[name = tensor("op_11109_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11109_cast_fp16 = slice_by_index(begin = var_11109_begin_0, end = var_11109_end_0, end_mask = var_11109_end_mask_0, x = var_10795_cast_fp16)[name = tensor("op_11109_cast_fp16")]; tensor var_11116_begin_0 = const()[name = tensor("op_11116_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11116_end_0 = const()[name = tensor("op_11116_end_0"), val = tensor([1, 64, 1, 375])]; tensor var_11116_end_mask_0 = const()[name = tensor("op_11116_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11116_cast_fp16 = slice_by_index(begin = var_11116_begin_0, end = var_11116_end_0, end_mask = var_11116_end_mask_0, x = var_10799_cast_fp16)[name = tensor("op_11116_cast_fp16")]; tensor var_11123_begin_0 = const()[name = tensor("op_11123_begin_0"), val = tensor([0, 0, 0, 375])]; tensor var_11123_end_0 = const()[name = tensor("op_11123_end_0"), val = tensor([1, 64, 1, 750])]; tensor var_11123_end_mask_0 = const()[name = tensor("op_11123_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11123_cast_fp16 = slice_by_index(begin = var_11123_begin_0, end = var_11123_end_0, end_mask = var_11123_end_mask_0, x = var_10799_cast_fp16)[name = tensor("op_11123_cast_fp16")]; tensor var_11130_begin_0 = const()[name = tensor("op_11130_begin_0"), val = tensor([0, 0, 0, 750])]; tensor var_11130_end_0 = const()[name = tensor("op_11130_end_0"), val = tensor([1, 64, 1, 1125])]; tensor var_11130_end_mask_0 = const()[name = tensor("op_11130_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11130_cast_fp16 = slice_by_index(begin = var_11130_begin_0, end = var_11130_end_0, end_mask = var_11130_end_mask_0, x = var_10799_cast_fp16)[name = tensor("op_11130_cast_fp16")]; tensor var_11137_begin_0 = const()[name = tensor("op_11137_begin_0"), val = tensor([0, 0, 0, 1125])]; tensor var_11137_end_0 = const()[name = tensor("op_11137_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11137_end_mask_0 = const()[name = tensor("op_11137_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11137_cast_fp16 = slice_by_index(begin = var_11137_begin_0, end = var_11137_end_0, end_mask = var_11137_end_mask_0, x = var_10799_cast_fp16)[name = tensor("op_11137_cast_fp16")]; tensor k_perm_0 = const()[name = tensor("k_perm_0"), val = tensor([0, 3, 2, 1])]; tensor var_11142_begin_0 = const()[name = tensor("op_11142_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11142_end_0 = const()[name = tensor("op_11142_end_0"), val = tensor([1, 1500, 1, 64])]; tensor var_11142_end_mask_0 = const()[name = tensor("op_11142_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_11142_cast_fp16 = slice_by_index(begin = var_11142_begin_0, end = var_11142_end_0, end_mask = var_11142_end_mask_0, x = transpose_0)[name = tensor("op_11142_cast_fp16")]; tensor var_11146_begin_0 = const()[name = tensor("op_11146_begin_0"), val = tensor([0, 0, 0, 64])]; tensor var_11146_end_0 = const()[name = tensor("op_11146_end_0"), val = tensor([1, 1500, 1, 128])]; tensor var_11146_end_mask_0 = const()[name = tensor("op_11146_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11146_cast_fp16 = slice_by_index(begin = var_11146_begin_0, end = var_11146_end_0, end_mask = var_11146_end_mask_0, x = transpose_0)[name = tensor("op_11146_cast_fp16")]; tensor var_11150_begin_0 = const()[name = tensor("op_11150_begin_0"), val = tensor([0, 0, 0, 128])]; tensor var_11150_end_0 = const()[name = tensor("op_11150_end_0"), val = tensor([1, 1500, 1, 192])]; tensor var_11150_end_mask_0 = const()[name = tensor("op_11150_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11150_cast_fp16 = slice_by_index(begin = var_11150_begin_0, end = var_11150_end_0, end_mask = var_11150_end_mask_0, x = transpose_0)[name = tensor("op_11150_cast_fp16")]; tensor var_11154_begin_0 = const()[name = tensor("op_11154_begin_0"), val = tensor([0, 0, 0, 192])]; tensor var_11154_end_0 = const()[name = tensor("op_11154_end_0"), val = tensor([1, 1500, 1, 256])]; tensor var_11154_end_mask_0 = const()[name = tensor("op_11154_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11154_cast_fp16 = slice_by_index(begin = var_11154_begin_0, end = var_11154_end_0, end_mask = var_11154_end_mask_0, x = transpose_0)[name = tensor("op_11154_cast_fp16")]; tensor var_11158_begin_0 = const()[name = tensor("op_11158_begin_0"), val = tensor([0, 0, 0, 256])]; tensor var_11158_end_0 = const()[name = tensor("op_11158_end_0"), val = tensor([1, 1500, 1, 320])]; tensor var_11158_end_mask_0 = const()[name = tensor("op_11158_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11158_cast_fp16 = slice_by_index(begin = var_11158_begin_0, end = var_11158_end_0, end_mask = var_11158_end_mask_0, x = transpose_0)[name = tensor("op_11158_cast_fp16")]; tensor var_11162_begin_0 = const()[name = tensor("op_11162_begin_0"), val = tensor([0, 0, 0, 320])]; tensor var_11162_end_0 = const()[name = tensor("op_11162_end_0"), val = tensor([1, 1500, 1, 384])]; tensor var_11162_end_mask_0 = const()[name = tensor("op_11162_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11162_cast_fp16 = slice_by_index(begin = var_11162_begin_0, end = var_11162_end_0, end_mask = var_11162_end_mask_0, x = transpose_0)[name = tensor("op_11162_cast_fp16")]; tensor var_11166_begin_0 = const()[name = tensor("op_11166_begin_0"), val = tensor([0, 0, 0, 384])]; tensor var_11166_end_0 = const()[name = tensor("op_11166_end_0"), val = tensor([1, 1500, 1, 448])]; tensor var_11166_end_mask_0 = const()[name = tensor("op_11166_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11166_cast_fp16 = slice_by_index(begin = var_11166_begin_0, end = var_11166_end_0, end_mask = var_11166_end_mask_0, x = transpose_0)[name = tensor("op_11166_cast_fp16")]; tensor var_11170_begin_0 = const()[name = tensor("op_11170_begin_0"), val = tensor([0, 0, 0, 448])]; tensor var_11170_end_0 = const()[name = tensor("op_11170_end_0"), val = tensor([1, 1500, 1, 512])]; tensor var_11170_end_mask_0 = const()[name = tensor("op_11170_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11170_cast_fp16 = slice_by_index(begin = var_11170_begin_0, end = var_11170_end_0, end_mask = var_11170_end_mask_0, x = transpose_0)[name = tensor("op_11170_cast_fp16")]; tensor var_11174_begin_0 = const()[name = tensor("op_11174_begin_0"), val = tensor([0, 0, 0, 512])]; tensor var_11174_end_0 = const()[name = tensor("op_11174_end_0"), val = tensor([1, 1500, 1, 576])]; tensor var_11174_end_mask_0 = const()[name = tensor("op_11174_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11174_cast_fp16 = slice_by_index(begin = var_11174_begin_0, end = var_11174_end_0, end_mask = var_11174_end_mask_0, x = transpose_0)[name = tensor("op_11174_cast_fp16")]; tensor var_11178_begin_0 = const()[name = tensor("op_11178_begin_0"), val = tensor([0, 0, 0, 576])]; tensor var_11178_end_0 = const()[name = tensor("op_11178_end_0"), val = tensor([1, 1500, 1, 640])]; tensor var_11178_end_mask_0 = const()[name = tensor("op_11178_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11178_cast_fp16 = slice_by_index(begin = var_11178_begin_0, end = var_11178_end_0, end_mask = var_11178_end_mask_0, x = transpose_0)[name = tensor("op_11178_cast_fp16")]; tensor var_11182_begin_0 = const()[name = tensor("op_11182_begin_0"), val = tensor([0, 0, 0, 640])]; tensor var_11182_end_0 = const()[name = tensor("op_11182_end_0"), val = tensor([1, 1500, 1, 704])]; tensor var_11182_end_mask_0 = const()[name = tensor("op_11182_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11182_cast_fp16 = slice_by_index(begin = var_11182_begin_0, end = var_11182_end_0, end_mask = var_11182_end_mask_0, x = transpose_0)[name = tensor("op_11182_cast_fp16")]; tensor var_11186_begin_0 = const()[name = tensor("op_11186_begin_0"), val = tensor([0, 0, 0, 704])]; tensor var_11186_end_0 = const()[name = tensor("op_11186_end_0"), val = tensor([1, 1500, 1, 768])]; tensor var_11186_end_mask_0 = const()[name = tensor("op_11186_end_mask_0"), val = tensor([true, true, true, false])]; tensor var_11186_cast_fp16 = slice_by_index(begin = var_11186_begin_0, end = var_11186_end_0, end_mask = var_11186_end_mask_0, x = transpose_0)[name = tensor("op_11186_cast_fp16")]; tensor var_11188_begin_0 = const()[name = tensor("op_11188_begin_0"), val = tensor([0, 0, 0, 0])]; tensor var_11188_end_0 = const()[name = tensor("op_11188_end_0"), val = tensor([1, 64, 1, 1500])]; tensor var_11188_end_mask_0 = const()[name = tensor("op_11188_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11188_cast_fp16 = slice_by_index(begin = var_11188_begin_0, end = var_11188_end_0, end_mask = var_11188_end_mask_0, x = value_cast_fp16)[name = tensor("op_11188_cast_fp16")]; tensor var_11192_begin_0 = const()[name = tensor("op_11192_begin_0"), val = tensor([0, 64, 0, 0])]; tensor var_11192_end_0 = const()[name = tensor("op_11192_end_0"), val = tensor([1, 128, 1, 1500])]; tensor var_11192_end_mask_0 = const()[name = tensor("op_11192_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11192_cast_fp16 = slice_by_index(begin = var_11192_begin_0, end = var_11192_end_0, end_mask = var_11192_end_mask_0, x = value_cast_fp16)[name = tensor("op_11192_cast_fp16")]; tensor var_11196_begin_0 = const()[name = tensor("op_11196_begin_0"), val = tensor([0, 128, 0, 0])]; tensor var_11196_end_0 = const()[name = tensor("op_11196_end_0"), val = tensor([1, 192, 1, 1500])]; tensor var_11196_end_mask_0 = const()[name = tensor("op_11196_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11196_cast_fp16 = slice_by_index(begin = var_11196_begin_0, end = var_11196_end_0, end_mask = var_11196_end_mask_0, x = value_cast_fp16)[name = tensor("op_11196_cast_fp16")]; tensor var_11200_begin_0 = const()[name = tensor("op_11200_begin_0"), val = tensor([0, 192, 0, 0])]; tensor var_11200_end_0 = const()[name = tensor("op_11200_end_0"), val = tensor([1, 256, 1, 1500])]; tensor var_11200_end_mask_0 = const()[name = tensor("op_11200_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11200_cast_fp16 = slice_by_index(begin = var_11200_begin_0, end = var_11200_end_0, end_mask = var_11200_end_mask_0, x = value_cast_fp16)[name = tensor("op_11200_cast_fp16")]; tensor var_11204_begin_0 = const()[name = tensor("op_11204_begin_0"), val = tensor([0, 256, 0, 0])]; tensor var_11204_end_0 = const()[name = tensor("op_11204_end_0"), val = tensor([1, 320, 1, 1500])]; tensor var_11204_end_mask_0 = const()[name = tensor("op_11204_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11204_cast_fp16 = slice_by_index(begin = var_11204_begin_0, end = var_11204_end_0, end_mask = var_11204_end_mask_0, x = value_cast_fp16)[name = tensor("op_11204_cast_fp16")]; tensor var_11208_begin_0 = const()[name = tensor("op_11208_begin_0"), val = tensor([0, 320, 0, 0])]; tensor var_11208_end_0 = const()[name = tensor("op_11208_end_0"), val = tensor([1, 384, 1, 1500])]; tensor var_11208_end_mask_0 = const()[name = tensor("op_11208_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11208_cast_fp16 = slice_by_index(begin = var_11208_begin_0, end = var_11208_end_0, end_mask = var_11208_end_mask_0, x = value_cast_fp16)[name = tensor("op_11208_cast_fp16")]; tensor var_11212_begin_0 = const()[name = tensor("op_11212_begin_0"), val = tensor([0, 384, 0, 0])]; tensor var_11212_end_0 = const()[name = tensor("op_11212_end_0"), val = tensor([1, 448, 1, 1500])]; tensor var_11212_end_mask_0 = const()[name = tensor("op_11212_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11212_cast_fp16 = slice_by_index(begin = var_11212_begin_0, end = var_11212_end_0, end_mask = var_11212_end_mask_0, x = value_cast_fp16)[name = tensor("op_11212_cast_fp16")]; tensor var_11216_begin_0 = const()[name = tensor("op_11216_begin_0"), val = tensor([0, 448, 0, 0])]; tensor var_11216_end_0 = const()[name = tensor("op_11216_end_0"), val = tensor([1, 512, 1, 1500])]; tensor var_11216_end_mask_0 = const()[name = tensor("op_11216_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11216_cast_fp16 = slice_by_index(begin = var_11216_begin_0, end = var_11216_end_0, end_mask = var_11216_end_mask_0, x = value_cast_fp16)[name = tensor("op_11216_cast_fp16")]; tensor var_11220_begin_0 = const()[name = tensor("op_11220_begin_0"), val = tensor([0, 512, 0, 0])]; tensor var_11220_end_0 = const()[name = tensor("op_11220_end_0"), val = tensor([1, 576, 1, 1500])]; tensor var_11220_end_mask_0 = const()[name = tensor("op_11220_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11220_cast_fp16 = slice_by_index(begin = var_11220_begin_0, end = var_11220_end_0, end_mask = var_11220_end_mask_0, x = value_cast_fp16)[name = tensor("op_11220_cast_fp16")]; tensor var_11224_begin_0 = const()[name = tensor("op_11224_begin_0"), val = tensor([0, 576, 0, 0])]; tensor var_11224_end_0 = const()[name = tensor("op_11224_end_0"), val = tensor([1, 640, 1, 1500])]; tensor var_11224_end_mask_0 = const()[name = tensor("op_11224_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11224_cast_fp16 = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = value_cast_fp16)[name = tensor("op_11224_cast_fp16")]; tensor var_11228_begin_0 = const()[name = tensor("op_11228_begin_0"), val = tensor([0, 640, 0, 0])]; tensor var_11228_end_0 = const()[name = tensor("op_11228_end_0"), val = tensor([1, 704, 1, 1500])]; tensor var_11228_end_mask_0 = const()[name = tensor("op_11228_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11228_cast_fp16 = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = value_cast_fp16)[name = tensor("op_11228_cast_fp16")]; tensor var_11232_begin_0 = const()[name = tensor("op_11232_begin_0"), val = tensor([0, 704, 0, 0])]; tensor var_11232_end_0 = const()[name = tensor("op_11232_end_0"), val = tensor([1, 768, 1, 1500])]; tensor var_11232_end_mask_0 = const()[name = tensor("op_11232_end_mask_0"), val = tensor([true, false, true, true])]; tensor var_11232_cast_fp16 = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = value_cast_fp16)[name = tensor("op_11232_cast_fp16")]; tensor var_11236_equation_0 = const()[name = tensor("op_11236_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11236_cast_fp16 = einsum(equation = var_11236_equation_0, values = (var_11142_cast_fp16, var_10808_cast_fp16))[name = tensor("op_11236_cast_fp16")]; tensor var_11237_to_fp16 = const()[name = tensor("op_11237_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1057_cast_fp16 = mul(x = var_11236_cast_fp16, y = var_11237_to_fp16)[name = tensor("aw_chunk_1057_cast_fp16")]; tensor var_11240_equation_0 = const()[name = tensor("op_11240_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11240_cast_fp16 = einsum(equation = var_11240_equation_0, values = (var_11142_cast_fp16, var_10815_cast_fp16))[name = tensor("op_11240_cast_fp16")]; tensor var_11241_to_fp16 = const()[name = tensor("op_11241_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1059_cast_fp16 = mul(x = var_11240_cast_fp16, y = var_11241_to_fp16)[name = tensor("aw_chunk_1059_cast_fp16")]; tensor var_11244_equation_0 = const()[name = tensor("op_11244_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11244_cast_fp16 = einsum(equation = var_11244_equation_0, values = (var_11142_cast_fp16, var_10822_cast_fp16))[name = tensor("op_11244_cast_fp16")]; tensor var_11245_to_fp16 = const()[name = tensor("op_11245_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1061_cast_fp16 = mul(x = var_11244_cast_fp16, y = var_11245_to_fp16)[name = tensor("aw_chunk_1061_cast_fp16")]; tensor var_11248_equation_0 = const()[name = tensor("op_11248_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11248_cast_fp16 = einsum(equation = var_11248_equation_0, values = (var_11142_cast_fp16, var_10829_cast_fp16))[name = tensor("op_11248_cast_fp16")]; tensor var_11249_to_fp16 = const()[name = tensor("op_11249_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1063_cast_fp16 = mul(x = var_11248_cast_fp16, y = var_11249_to_fp16)[name = tensor("aw_chunk_1063_cast_fp16")]; tensor var_11252_equation_0 = const()[name = tensor("op_11252_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11252_cast_fp16 = einsum(equation = var_11252_equation_0, values = (var_11146_cast_fp16, var_10836_cast_fp16))[name = tensor("op_11252_cast_fp16")]; tensor var_11253_to_fp16 = const()[name = tensor("op_11253_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1065_cast_fp16 = mul(x = var_11252_cast_fp16, y = var_11253_to_fp16)[name = tensor("aw_chunk_1065_cast_fp16")]; tensor var_11256_equation_0 = const()[name = tensor("op_11256_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11256_cast_fp16 = einsum(equation = var_11256_equation_0, values = (var_11146_cast_fp16, var_10843_cast_fp16))[name = tensor("op_11256_cast_fp16")]; tensor var_11257_to_fp16 = const()[name = tensor("op_11257_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1067_cast_fp16 = mul(x = var_11256_cast_fp16, y = var_11257_to_fp16)[name = tensor("aw_chunk_1067_cast_fp16")]; tensor var_11260_equation_0 = const()[name = tensor("op_11260_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11260_cast_fp16 = einsum(equation = var_11260_equation_0, values = (var_11146_cast_fp16, var_10850_cast_fp16))[name = tensor("op_11260_cast_fp16")]; tensor var_11261_to_fp16 = const()[name = tensor("op_11261_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1069_cast_fp16 = mul(x = var_11260_cast_fp16, y = var_11261_to_fp16)[name = tensor("aw_chunk_1069_cast_fp16")]; tensor var_11264_equation_0 = const()[name = tensor("op_11264_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11264_cast_fp16 = einsum(equation = var_11264_equation_0, values = (var_11146_cast_fp16, var_10857_cast_fp16))[name = tensor("op_11264_cast_fp16")]; tensor var_11265_to_fp16 = const()[name = tensor("op_11265_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1071_cast_fp16 = mul(x = var_11264_cast_fp16, y = var_11265_to_fp16)[name = tensor("aw_chunk_1071_cast_fp16")]; tensor var_11268_equation_0 = const()[name = tensor("op_11268_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11268_cast_fp16 = einsum(equation = var_11268_equation_0, values = (var_11150_cast_fp16, var_10864_cast_fp16))[name = tensor("op_11268_cast_fp16")]; tensor var_11269_to_fp16 = const()[name = tensor("op_11269_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1073_cast_fp16 = mul(x = var_11268_cast_fp16, y = var_11269_to_fp16)[name = tensor("aw_chunk_1073_cast_fp16")]; tensor var_11272_equation_0 = const()[name = tensor("op_11272_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11272_cast_fp16 = einsum(equation = var_11272_equation_0, values = (var_11150_cast_fp16, var_10871_cast_fp16))[name = tensor("op_11272_cast_fp16")]; tensor var_11273_to_fp16 = const()[name = tensor("op_11273_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1075_cast_fp16 = mul(x = var_11272_cast_fp16, y = var_11273_to_fp16)[name = tensor("aw_chunk_1075_cast_fp16")]; tensor var_11276_equation_0 = const()[name = tensor("op_11276_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11276_cast_fp16 = einsum(equation = var_11276_equation_0, values = (var_11150_cast_fp16, var_10878_cast_fp16))[name = tensor("op_11276_cast_fp16")]; tensor var_11277_to_fp16 = const()[name = tensor("op_11277_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1077_cast_fp16 = mul(x = var_11276_cast_fp16, y = var_11277_to_fp16)[name = tensor("aw_chunk_1077_cast_fp16")]; tensor var_11280_equation_0 = const()[name = tensor("op_11280_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11280_cast_fp16 = einsum(equation = var_11280_equation_0, values = (var_11150_cast_fp16, var_10885_cast_fp16))[name = tensor("op_11280_cast_fp16")]; tensor var_11281_to_fp16 = const()[name = tensor("op_11281_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1079_cast_fp16 = mul(x = var_11280_cast_fp16, y = var_11281_to_fp16)[name = tensor("aw_chunk_1079_cast_fp16")]; tensor var_11284_equation_0 = const()[name = tensor("op_11284_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11284_cast_fp16 = einsum(equation = var_11284_equation_0, values = (var_11154_cast_fp16, var_10892_cast_fp16))[name = tensor("op_11284_cast_fp16")]; tensor var_11285_to_fp16 = const()[name = tensor("op_11285_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1081_cast_fp16 = mul(x = var_11284_cast_fp16, y = var_11285_to_fp16)[name = tensor("aw_chunk_1081_cast_fp16")]; tensor var_11288_equation_0 = const()[name = tensor("op_11288_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11288_cast_fp16 = einsum(equation = var_11288_equation_0, values = (var_11154_cast_fp16, var_10899_cast_fp16))[name = tensor("op_11288_cast_fp16")]; tensor var_11289_to_fp16 = const()[name = tensor("op_11289_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1083_cast_fp16 = mul(x = var_11288_cast_fp16, y = var_11289_to_fp16)[name = tensor("aw_chunk_1083_cast_fp16")]; tensor var_11292_equation_0 = const()[name = tensor("op_11292_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11292_cast_fp16 = einsum(equation = var_11292_equation_0, values = (var_11154_cast_fp16, var_10906_cast_fp16))[name = tensor("op_11292_cast_fp16")]; tensor var_11293_to_fp16 = const()[name = tensor("op_11293_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1085_cast_fp16 = mul(x = var_11292_cast_fp16, y = var_11293_to_fp16)[name = tensor("aw_chunk_1085_cast_fp16")]; tensor var_11296_equation_0 = const()[name = tensor("op_11296_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11296_cast_fp16 = einsum(equation = var_11296_equation_0, values = (var_11154_cast_fp16, var_10913_cast_fp16))[name = tensor("op_11296_cast_fp16")]; tensor var_11297_to_fp16 = const()[name = tensor("op_11297_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1087_cast_fp16 = mul(x = var_11296_cast_fp16, y = var_11297_to_fp16)[name = tensor("aw_chunk_1087_cast_fp16")]; tensor var_11300_equation_0 = const()[name = tensor("op_11300_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11300_cast_fp16 = einsum(equation = var_11300_equation_0, values = (var_11158_cast_fp16, var_10920_cast_fp16))[name = tensor("op_11300_cast_fp16")]; tensor var_11301_to_fp16 = const()[name = tensor("op_11301_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1089_cast_fp16 = mul(x = var_11300_cast_fp16, y = var_11301_to_fp16)[name = tensor("aw_chunk_1089_cast_fp16")]; tensor var_11304_equation_0 = const()[name = tensor("op_11304_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11304_cast_fp16 = einsum(equation = var_11304_equation_0, values = (var_11158_cast_fp16, var_10927_cast_fp16))[name = tensor("op_11304_cast_fp16")]; tensor var_11305_to_fp16 = const()[name = tensor("op_11305_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1091_cast_fp16 = mul(x = var_11304_cast_fp16, y = var_11305_to_fp16)[name = tensor("aw_chunk_1091_cast_fp16")]; tensor var_11308_equation_0 = const()[name = tensor("op_11308_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11308_cast_fp16 = einsum(equation = var_11308_equation_0, values = (var_11158_cast_fp16, var_10934_cast_fp16))[name = tensor("op_11308_cast_fp16")]; tensor var_11309_to_fp16 = const()[name = tensor("op_11309_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1093_cast_fp16 = mul(x = var_11308_cast_fp16, y = var_11309_to_fp16)[name = tensor("aw_chunk_1093_cast_fp16")]; tensor var_11312_equation_0 = const()[name = tensor("op_11312_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11312_cast_fp16 = einsum(equation = var_11312_equation_0, values = (var_11158_cast_fp16, var_10941_cast_fp16))[name = tensor("op_11312_cast_fp16")]; tensor var_11313_to_fp16 = const()[name = tensor("op_11313_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1095_cast_fp16 = mul(x = var_11312_cast_fp16, y = var_11313_to_fp16)[name = tensor("aw_chunk_1095_cast_fp16")]; tensor var_11316_equation_0 = const()[name = tensor("op_11316_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11316_cast_fp16 = einsum(equation = var_11316_equation_0, values = (var_11162_cast_fp16, var_10948_cast_fp16))[name = tensor("op_11316_cast_fp16")]; tensor var_11317_to_fp16 = const()[name = tensor("op_11317_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1097_cast_fp16 = mul(x = var_11316_cast_fp16, y = var_11317_to_fp16)[name = tensor("aw_chunk_1097_cast_fp16")]; tensor var_11320_equation_0 = const()[name = tensor("op_11320_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11320_cast_fp16 = einsum(equation = var_11320_equation_0, values = (var_11162_cast_fp16, var_10955_cast_fp16))[name = tensor("op_11320_cast_fp16")]; tensor var_11321_to_fp16 = const()[name = tensor("op_11321_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1099_cast_fp16 = mul(x = var_11320_cast_fp16, y = var_11321_to_fp16)[name = tensor("aw_chunk_1099_cast_fp16")]; tensor var_11324_equation_0 = const()[name = tensor("op_11324_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11324_cast_fp16 = einsum(equation = var_11324_equation_0, values = (var_11162_cast_fp16, var_10962_cast_fp16))[name = tensor("op_11324_cast_fp16")]; tensor var_11325_to_fp16 = const()[name = tensor("op_11325_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1101_cast_fp16 = mul(x = var_11324_cast_fp16, y = var_11325_to_fp16)[name = tensor("aw_chunk_1101_cast_fp16")]; tensor var_11328_equation_0 = const()[name = tensor("op_11328_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11328_cast_fp16 = einsum(equation = var_11328_equation_0, values = (var_11162_cast_fp16, var_10969_cast_fp16))[name = tensor("op_11328_cast_fp16")]; tensor var_11329_to_fp16 = const()[name = tensor("op_11329_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1103_cast_fp16 = mul(x = var_11328_cast_fp16, y = var_11329_to_fp16)[name = tensor("aw_chunk_1103_cast_fp16")]; tensor var_11332_equation_0 = const()[name = tensor("op_11332_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11332_cast_fp16 = einsum(equation = var_11332_equation_0, values = (var_11166_cast_fp16, var_10976_cast_fp16))[name = tensor("op_11332_cast_fp16")]; tensor var_11333_to_fp16 = const()[name = tensor("op_11333_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1105_cast_fp16 = mul(x = var_11332_cast_fp16, y = var_11333_to_fp16)[name = tensor("aw_chunk_1105_cast_fp16")]; tensor var_11336_equation_0 = const()[name = tensor("op_11336_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11336_cast_fp16 = einsum(equation = var_11336_equation_0, values = (var_11166_cast_fp16, var_10983_cast_fp16))[name = tensor("op_11336_cast_fp16")]; tensor var_11337_to_fp16 = const()[name = tensor("op_11337_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1107_cast_fp16 = mul(x = var_11336_cast_fp16, y = var_11337_to_fp16)[name = tensor("aw_chunk_1107_cast_fp16")]; tensor var_11340_equation_0 = const()[name = tensor("op_11340_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11340_cast_fp16 = einsum(equation = var_11340_equation_0, values = (var_11166_cast_fp16, var_10990_cast_fp16))[name = tensor("op_11340_cast_fp16")]; tensor var_11341_to_fp16 = const()[name = tensor("op_11341_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1109_cast_fp16 = mul(x = var_11340_cast_fp16, y = var_11341_to_fp16)[name = tensor("aw_chunk_1109_cast_fp16")]; tensor var_11344_equation_0 = const()[name = tensor("op_11344_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11344_cast_fp16 = einsum(equation = var_11344_equation_0, values = (var_11166_cast_fp16, var_10997_cast_fp16))[name = tensor("op_11344_cast_fp16")]; tensor var_11345_to_fp16 = const()[name = tensor("op_11345_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1111_cast_fp16 = mul(x = var_11344_cast_fp16, y = var_11345_to_fp16)[name = tensor("aw_chunk_1111_cast_fp16")]; tensor var_11348_equation_0 = const()[name = tensor("op_11348_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11348_cast_fp16 = einsum(equation = var_11348_equation_0, values = (var_11170_cast_fp16, var_11004_cast_fp16))[name = tensor("op_11348_cast_fp16")]; tensor var_11349_to_fp16 = const()[name = tensor("op_11349_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1113_cast_fp16 = mul(x = var_11348_cast_fp16, y = var_11349_to_fp16)[name = tensor("aw_chunk_1113_cast_fp16")]; tensor var_11352_equation_0 = const()[name = tensor("op_11352_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11352_cast_fp16 = einsum(equation = var_11352_equation_0, values = (var_11170_cast_fp16, var_11011_cast_fp16))[name = tensor("op_11352_cast_fp16")]; tensor var_11353_to_fp16 = const()[name = tensor("op_11353_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1115_cast_fp16 = mul(x = var_11352_cast_fp16, y = var_11353_to_fp16)[name = tensor("aw_chunk_1115_cast_fp16")]; tensor var_11356_equation_0 = const()[name = tensor("op_11356_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11356_cast_fp16 = einsum(equation = var_11356_equation_0, values = (var_11170_cast_fp16, var_11018_cast_fp16))[name = tensor("op_11356_cast_fp16")]; tensor var_11357_to_fp16 = const()[name = tensor("op_11357_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1117_cast_fp16 = mul(x = var_11356_cast_fp16, y = var_11357_to_fp16)[name = tensor("aw_chunk_1117_cast_fp16")]; tensor var_11360_equation_0 = const()[name = tensor("op_11360_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11360_cast_fp16 = einsum(equation = var_11360_equation_0, values = (var_11170_cast_fp16, var_11025_cast_fp16))[name = tensor("op_11360_cast_fp16")]; tensor var_11361_to_fp16 = const()[name = tensor("op_11361_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1119_cast_fp16 = mul(x = var_11360_cast_fp16, y = var_11361_to_fp16)[name = tensor("aw_chunk_1119_cast_fp16")]; tensor var_11364_equation_0 = const()[name = tensor("op_11364_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11364_cast_fp16 = einsum(equation = var_11364_equation_0, values = (var_11174_cast_fp16, var_11032_cast_fp16))[name = tensor("op_11364_cast_fp16")]; tensor var_11365_to_fp16 = const()[name = tensor("op_11365_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1121_cast_fp16 = mul(x = var_11364_cast_fp16, y = var_11365_to_fp16)[name = tensor("aw_chunk_1121_cast_fp16")]; tensor var_11368_equation_0 = const()[name = tensor("op_11368_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11368_cast_fp16 = einsum(equation = var_11368_equation_0, values = (var_11174_cast_fp16, var_11039_cast_fp16))[name = tensor("op_11368_cast_fp16")]; tensor var_11369_to_fp16 = const()[name = tensor("op_11369_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1123_cast_fp16 = mul(x = var_11368_cast_fp16, y = var_11369_to_fp16)[name = tensor("aw_chunk_1123_cast_fp16")]; tensor var_11372_equation_0 = const()[name = tensor("op_11372_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11372_cast_fp16 = einsum(equation = var_11372_equation_0, values = (var_11174_cast_fp16, var_11046_cast_fp16))[name = tensor("op_11372_cast_fp16")]; tensor var_11373_to_fp16 = const()[name = tensor("op_11373_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1125_cast_fp16 = mul(x = var_11372_cast_fp16, y = var_11373_to_fp16)[name = tensor("aw_chunk_1125_cast_fp16")]; tensor var_11376_equation_0 = const()[name = tensor("op_11376_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11376_cast_fp16 = einsum(equation = var_11376_equation_0, values = (var_11174_cast_fp16, var_11053_cast_fp16))[name = tensor("op_11376_cast_fp16")]; tensor var_11377_to_fp16 = const()[name = tensor("op_11377_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1127_cast_fp16 = mul(x = var_11376_cast_fp16, y = var_11377_to_fp16)[name = tensor("aw_chunk_1127_cast_fp16")]; tensor var_11380_equation_0 = const()[name = tensor("op_11380_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11380_cast_fp16 = einsum(equation = var_11380_equation_0, values = (var_11178_cast_fp16, var_11060_cast_fp16))[name = tensor("op_11380_cast_fp16")]; tensor var_11381_to_fp16 = const()[name = tensor("op_11381_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1129_cast_fp16 = mul(x = var_11380_cast_fp16, y = var_11381_to_fp16)[name = tensor("aw_chunk_1129_cast_fp16")]; tensor var_11384_equation_0 = const()[name = tensor("op_11384_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11384_cast_fp16 = einsum(equation = var_11384_equation_0, values = (var_11178_cast_fp16, var_11067_cast_fp16))[name = tensor("op_11384_cast_fp16")]; tensor var_11385_to_fp16 = const()[name = tensor("op_11385_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1131_cast_fp16 = mul(x = var_11384_cast_fp16, y = var_11385_to_fp16)[name = tensor("aw_chunk_1131_cast_fp16")]; tensor var_11388_equation_0 = const()[name = tensor("op_11388_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11388_cast_fp16 = einsum(equation = var_11388_equation_0, values = (var_11178_cast_fp16, var_11074_cast_fp16))[name = tensor("op_11388_cast_fp16")]; tensor var_11389_to_fp16 = const()[name = tensor("op_11389_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1133_cast_fp16 = mul(x = var_11388_cast_fp16, y = var_11389_to_fp16)[name = tensor("aw_chunk_1133_cast_fp16")]; tensor var_11392_equation_0 = const()[name = tensor("op_11392_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11392_cast_fp16 = einsum(equation = var_11392_equation_0, values = (var_11178_cast_fp16, var_11081_cast_fp16))[name = tensor("op_11392_cast_fp16")]; tensor var_11393_to_fp16 = const()[name = tensor("op_11393_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1135_cast_fp16 = mul(x = var_11392_cast_fp16, y = var_11393_to_fp16)[name = tensor("aw_chunk_1135_cast_fp16")]; tensor var_11396_equation_0 = const()[name = tensor("op_11396_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11396_cast_fp16 = einsum(equation = var_11396_equation_0, values = (var_11182_cast_fp16, var_11088_cast_fp16))[name = tensor("op_11396_cast_fp16")]; tensor var_11397_to_fp16 = const()[name = tensor("op_11397_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1137_cast_fp16 = mul(x = var_11396_cast_fp16, y = var_11397_to_fp16)[name = tensor("aw_chunk_1137_cast_fp16")]; tensor var_11400_equation_0 = const()[name = tensor("op_11400_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11400_cast_fp16 = einsum(equation = var_11400_equation_0, values = (var_11182_cast_fp16, var_11095_cast_fp16))[name = tensor("op_11400_cast_fp16")]; tensor var_11401_to_fp16 = const()[name = tensor("op_11401_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1139_cast_fp16 = mul(x = var_11400_cast_fp16, y = var_11401_to_fp16)[name = tensor("aw_chunk_1139_cast_fp16")]; tensor var_11404_equation_0 = const()[name = tensor("op_11404_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11404_cast_fp16 = einsum(equation = var_11404_equation_0, values = (var_11182_cast_fp16, var_11102_cast_fp16))[name = tensor("op_11404_cast_fp16")]; tensor var_11405_to_fp16 = const()[name = tensor("op_11405_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1141_cast_fp16 = mul(x = var_11404_cast_fp16, y = var_11405_to_fp16)[name = tensor("aw_chunk_1141_cast_fp16")]; tensor var_11408_equation_0 = const()[name = tensor("op_11408_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11408_cast_fp16 = einsum(equation = var_11408_equation_0, values = (var_11182_cast_fp16, var_11109_cast_fp16))[name = tensor("op_11408_cast_fp16")]; tensor var_11409_to_fp16 = const()[name = tensor("op_11409_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1143_cast_fp16 = mul(x = var_11408_cast_fp16, y = var_11409_to_fp16)[name = tensor("aw_chunk_1143_cast_fp16")]; tensor var_11412_equation_0 = const()[name = tensor("op_11412_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11412_cast_fp16 = einsum(equation = var_11412_equation_0, values = (var_11186_cast_fp16, var_11116_cast_fp16))[name = tensor("op_11412_cast_fp16")]; tensor var_11413_to_fp16 = const()[name = tensor("op_11413_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1145_cast_fp16 = mul(x = var_11412_cast_fp16, y = var_11413_to_fp16)[name = tensor("aw_chunk_1145_cast_fp16")]; tensor var_11416_equation_0 = const()[name = tensor("op_11416_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11416_cast_fp16 = einsum(equation = var_11416_equation_0, values = (var_11186_cast_fp16, var_11123_cast_fp16))[name = tensor("op_11416_cast_fp16")]; tensor var_11417_to_fp16 = const()[name = tensor("op_11417_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1147_cast_fp16 = mul(x = var_11416_cast_fp16, y = var_11417_to_fp16)[name = tensor("aw_chunk_1147_cast_fp16")]; tensor var_11420_equation_0 = const()[name = tensor("op_11420_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11420_cast_fp16 = einsum(equation = var_11420_equation_0, values = (var_11186_cast_fp16, var_11130_cast_fp16))[name = tensor("op_11420_cast_fp16")]; tensor var_11421_to_fp16 = const()[name = tensor("op_11421_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_1149_cast_fp16 = mul(x = var_11420_cast_fp16, y = var_11421_to_fp16)[name = tensor("aw_chunk_1149_cast_fp16")]; tensor var_11424_equation_0 = const()[name = tensor("op_11424_equation_0"), val = tensor("bkhc,bchq->bkhq")]; tensor var_11424_cast_fp16 = einsum(equation = var_11424_equation_0, values = (var_11186_cast_fp16, var_11137_cast_fp16))[name = tensor("op_11424_cast_fp16")]; tensor var_11425_to_fp16 = const()[name = tensor("op_11425_to_fp16"), val = tensor(0x1p-3)]; tensor aw_chunk_cast_fp16 = mul(x = var_11424_cast_fp16, y = var_11425_to_fp16)[name = tensor("aw_chunk_cast_fp16")]; tensor var_11427_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1057_cast_fp16)[name = tensor("op_11427_cast_fp16")]; tensor var_11428_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1059_cast_fp16)[name = tensor("op_11428_cast_fp16")]; tensor var_11429_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1061_cast_fp16)[name = tensor("op_11429_cast_fp16")]; tensor var_11430_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1063_cast_fp16)[name = tensor("op_11430_cast_fp16")]; tensor var_11431_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1065_cast_fp16)[name = tensor("op_11431_cast_fp16")]; tensor var_11432_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1067_cast_fp16)[name = tensor("op_11432_cast_fp16")]; tensor var_11433_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1069_cast_fp16)[name = tensor("op_11433_cast_fp16")]; tensor var_11434_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1071_cast_fp16)[name = tensor("op_11434_cast_fp16")]; tensor var_11435_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1073_cast_fp16)[name = tensor("op_11435_cast_fp16")]; tensor var_11436_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1075_cast_fp16)[name = tensor("op_11436_cast_fp16")]; tensor var_11437_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1077_cast_fp16)[name = tensor("op_11437_cast_fp16")]; tensor var_11438_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1079_cast_fp16)[name = tensor("op_11438_cast_fp16")]; tensor var_11439_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1081_cast_fp16)[name = tensor("op_11439_cast_fp16")]; tensor var_11440_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1083_cast_fp16)[name = tensor("op_11440_cast_fp16")]; tensor var_11441_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1085_cast_fp16)[name = tensor("op_11441_cast_fp16")]; tensor var_11442_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1087_cast_fp16)[name = tensor("op_11442_cast_fp16")]; tensor var_11443_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1089_cast_fp16)[name = tensor("op_11443_cast_fp16")]; tensor var_11444_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1091_cast_fp16)[name = tensor("op_11444_cast_fp16")]; tensor var_11445_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1093_cast_fp16)[name = tensor("op_11445_cast_fp16")]; tensor var_11446_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1095_cast_fp16)[name = tensor("op_11446_cast_fp16")]; tensor var_11447_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1097_cast_fp16)[name = tensor("op_11447_cast_fp16")]; tensor var_11448_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1099_cast_fp16)[name = tensor("op_11448_cast_fp16")]; tensor var_11449_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1101_cast_fp16)[name = tensor("op_11449_cast_fp16")]; tensor var_11450_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1103_cast_fp16)[name = tensor("op_11450_cast_fp16")]; tensor var_11451_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1105_cast_fp16)[name = tensor("op_11451_cast_fp16")]; tensor var_11452_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1107_cast_fp16)[name = tensor("op_11452_cast_fp16")]; tensor var_11453_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1109_cast_fp16)[name = tensor("op_11453_cast_fp16")]; tensor var_11454_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1111_cast_fp16)[name = tensor("op_11454_cast_fp16")]; tensor var_11455_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1113_cast_fp16)[name = tensor("op_11455_cast_fp16")]; tensor var_11456_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1115_cast_fp16)[name = tensor("op_11456_cast_fp16")]; tensor var_11457_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1117_cast_fp16)[name = tensor("op_11457_cast_fp16")]; tensor var_11458_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1119_cast_fp16)[name = tensor("op_11458_cast_fp16")]; tensor var_11459_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1121_cast_fp16)[name = tensor("op_11459_cast_fp16")]; tensor var_11460_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1123_cast_fp16)[name = tensor("op_11460_cast_fp16")]; tensor var_11461_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1125_cast_fp16)[name = tensor("op_11461_cast_fp16")]; tensor var_11462_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1127_cast_fp16)[name = tensor("op_11462_cast_fp16")]; tensor var_11463_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1129_cast_fp16)[name = tensor("op_11463_cast_fp16")]; tensor var_11464_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1131_cast_fp16)[name = tensor("op_11464_cast_fp16")]; tensor var_11465_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1133_cast_fp16)[name = tensor("op_11465_cast_fp16")]; tensor var_11466_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1135_cast_fp16)[name = tensor("op_11466_cast_fp16")]; tensor var_11467_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1137_cast_fp16)[name = tensor("op_11467_cast_fp16")]; tensor var_11468_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1139_cast_fp16)[name = tensor("op_11468_cast_fp16")]; tensor var_11469_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1141_cast_fp16)[name = tensor("op_11469_cast_fp16")]; tensor var_11470_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1143_cast_fp16)[name = tensor("op_11470_cast_fp16")]; tensor var_11471_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1145_cast_fp16)[name = tensor("op_11471_cast_fp16")]; tensor var_11472_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1147_cast_fp16)[name = tensor("op_11472_cast_fp16")]; tensor var_11473_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_1149_cast_fp16)[name = tensor("op_11473_cast_fp16")]; tensor var_11474_cast_fp16 = softmax(axis = var_10700, x = aw_chunk_cast_fp16)[name = tensor("op_11474_cast_fp16")]; tensor var_11476_equation_0 = const()[name = tensor("op_11476_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11476_cast_fp16 = einsum(equation = var_11476_equation_0, values = (var_11188_cast_fp16, var_11427_cast_fp16))[name = tensor("op_11476_cast_fp16")]; tensor var_11478_equation_0 = const()[name = tensor("op_11478_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11478_cast_fp16 = einsum(equation = var_11478_equation_0, values = (var_11188_cast_fp16, var_11428_cast_fp16))[name = tensor("op_11478_cast_fp16")]; tensor var_11480_equation_0 = const()[name = tensor("op_11480_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11480_cast_fp16 = einsum(equation = var_11480_equation_0, values = (var_11188_cast_fp16, var_11429_cast_fp16))[name = tensor("op_11480_cast_fp16")]; tensor var_11482_equation_0 = const()[name = tensor("op_11482_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11482_cast_fp16 = einsum(equation = var_11482_equation_0, values = (var_11188_cast_fp16, var_11430_cast_fp16))[name = tensor("op_11482_cast_fp16")]; tensor var_11484_equation_0 = const()[name = tensor("op_11484_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11484_cast_fp16 = einsum(equation = var_11484_equation_0, values = (var_11192_cast_fp16, var_11431_cast_fp16))[name = tensor("op_11484_cast_fp16")]; tensor var_11486_equation_0 = const()[name = tensor("op_11486_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11486_cast_fp16 = einsum(equation = var_11486_equation_0, values = (var_11192_cast_fp16, var_11432_cast_fp16))[name = tensor("op_11486_cast_fp16")]; tensor var_11488_equation_0 = const()[name = tensor("op_11488_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11488_cast_fp16 = einsum(equation = var_11488_equation_0, values = (var_11192_cast_fp16, var_11433_cast_fp16))[name = tensor("op_11488_cast_fp16")]; tensor var_11490_equation_0 = const()[name = tensor("op_11490_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11490_cast_fp16 = einsum(equation = var_11490_equation_0, values = (var_11192_cast_fp16, var_11434_cast_fp16))[name = tensor("op_11490_cast_fp16")]; tensor var_11492_equation_0 = const()[name = tensor("op_11492_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11492_cast_fp16 = einsum(equation = var_11492_equation_0, values = (var_11196_cast_fp16, var_11435_cast_fp16))[name = tensor("op_11492_cast_fp16")]; tensor var_11494_equation_0 = const()[name = tensor("op_11494_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11494_cast_fp16 = einsum(equation = var_11494_equation_0, values = (var_11196_cast_fp16, var_11436_cast_fp16))[name = tensor("op_11494_cast_fp16")]; tensor var_11496_equation_0 = const()[name = tensor("op_11496_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11496_cast_fp16 = einsum(equation = var_11496_equation_0, values = (var_11196_cast_fp16, var_11437_cast_fp16))[name = tensor("op_11496_cast_fp16")]; tensor var_11498_equation_0 = const()[name = tensor("op_11498_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11498_cast_fp16 = einsum(equation = var_11498_equation_0, values = (var_11196_cast_fp16, var_11438_cast_fp16))[name = tensor("op_11498_cast_fp16")]; tensor var_11500_equation_0 = const()[name = tensor("op_11500_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11500_cast_fp16 = einsum(equation = var_11500_equation_0, values = (var_11200_cast_fp16, var_11439_cast_fp16))[name = tensor("op_11500_cast_fp16")]; tensor var_11502_equation_0 = const()[name = tensor("op_11502_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11502_cast_fp16 = einsum(equation = var_11502_equation_0, values = (var_11200_cast_fp16, var_11440_cast_fp16))[name = tensor("op_11502_cast_fp16")]; tensor var_11504_equation_0 = const()[name = tensor("op_11504_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11504_cast_fp16 = einsum(equation = var_11504_equation_0, values = (var_11200_cast_fp16, var_11441_cast_fp16))[name = tensor("op_11504_cast_fp16")]; tensor var_11506_equation_0 = const()[name = tensor("op_11506_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11506_cast_fp16 = einsum(equation = var_11506_equation_0, values = (var_11200_cast_fp16, var_11442_cast_fp16))[name = tensor("op_11506_cast_fp16")]; tensor var_11508_equation_0 = const()[name = tensor("op_11508_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11508_cast_fp16 = einsum(equation = var_11508_equation_0, values = (var_11204_cast_fp16, var_11443_cast_fp16))[name = tensor("op_11508_cast_fp16")]; tensor var_11510_equation_0 = const()[name = tensor("op_11510_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11510_cast_fp16 = einsum(equation = var_11510_equation_0, values = (var_11204_cast_fp16, var_11444_cast_fp16))[name = tensor("op_11510_cast_fp16")]; tensor var_11512_equation_0 = const()[name = tensor("op_11512_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11512_cast_fp16 = einsum(equation = var_11512_equation_0, values = (var_11204_cast_fp16, var_11445_cast_fp16))[name = tensor("op_11512_cast_fp16")]; tensor var_11514_equation_0 = const()[name = tensor("op_11514_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11514_cast_fp16 = einsum(equation = var_11514_equation_0, values = (var_11204_cast_fp16, var_11446_cast_fp16))[name = tensor("op_11514_cast_fp16")]; tensor var_11516_equation_0 = const()[name = tensor("op_11516_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11516_cast_fp16 = einsum(equation = var_11516_equation_0, values = (var_11208_cast_fp16, var_11447_cast_fp16))[name = tensor("op_11516_cast_fp16")]; tensor var_11518_equation_0 = const()[name = tensor("op_11518_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11518_cast_fp16 = einsum(equation = var_11518_equation_0, values = (var_11208_cast_fp16, var_11448_cast_fp16))[name = tensor("op_11518_cast_fp16")]; tensor var_11520_equation_0 = const()[name = tensor("op_11520_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11520_cast_fp16 = einsum(equation = var_11520_equation_0, values = (var_11208_cast_fp16, var_11449_cast_fp16))[name = tensor("op_11520_cast_fp16")]; tensor var_11522_equation_0 = const()[name = tensor("op_11522_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11522_cast_fp16 = einsum(equation = var_11522_equation_0, values = (var_11208_cast_fp16, var_11450_cast_fp16))[name = tensor("op_11522_cast_fp16")]; tensor var_11524_equation_0 = const()[name = tensor("op_11524_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11524_cast_fp16 = einsum(equation = var_11524_equation_0, values = (var_11212_cast_fp16, var_11451_cast_fp16))[name = tensor("op_11524_cast_fp16")]; tensor var_11526_equation_0 = const()[name = tensor("op_11526_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11526_cast_fp16 = einsum(equation = var_11526_equation_0, values = (var_11212_cast_fp16, var_11452_cast_fp16))[name = tensor("op_11526_cast_fp16")]; tensor var_11528_equation_0 = const()[name = tensor("op_11528_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11528_cast_fp16 = einsum(equation = var_11528_equation_0, values = (var_11212_cast_fp16, var_11453_cast_fp16))[name = tensor("op_11528_cast_fp16")]; tensor var_11530_equation_0 = const()[name = tensor("op_11530_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11530_cast_fp16 = einsum(equation = var_11530_equation_0, values = (var_11212_cast_fp16, var_11454_cast_fp16))[name = tensor("op_11530_cast_fp16")]; tensor var_11532_equation_0 = const()[name = tensor("op_11532_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11532_cast_fp16 = einsum(equation = var_11532_equation_0, values = (var_11216_cast_fp16, var_11455_cast_fp16))[name = tensor("op_11532_cast_fp16")]; tensor var_11534_equation_0 = const()[name = tensor("op_11534_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11534_cast_fp16 = einsum(equation = var_11534_equation_0, values = (var_11216_cast_fp16, var_11456_cast_fp16))[name = tensor("op_11534_cast_fp16")]; tensor var_11536_equation_0 = const()[name = tensor("op_11536_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11536_cast_fp16 = einsum(equation = var_11536_equation_0, values = (var_11216_cast_fp16, var_11457_cast_fp16))[name = tensor("op_11536_cast_fp16")]; tensor var_11538_equation_0 = const()[name = tensor("op_11538_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11538_cast_fp16 = einsum(equation = var_11538_equation_0, values = (var_11216_cast_fp16, var_11458_cast_fp16))[name = tensor("op_11538_cast_fp16")]; tensor var_11540_equation_0 = const()[name = tensor("op_11540_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11540_cast_fp16 = einsum(equation = var_11540_equation_0, values = (var_11220_cast_fp16, var_11459_cast_fp16))[name = tensor("op_11540_cast_fp16")]; tensor var_11542_equation_0 = const()[name = tensor("op_11542_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11542_cast_fp16 = einsum(equation = var_11542_equation_0, values = (var_11220_cast_fp16, var_11460_cast_fp16))[name = tensor("op_11542_cast_fp16")]; tensor var_11544_equation_0 = const()[name = tensor("op_11544_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11544_cast_fp16 = einsum(equation = var_11544_equation_0, values = (var_11220_cast_fp16, var_11461_cast_fp16))[name = tensor("op_11544_cast_fp16")]; tensor var_11546_equation_0 = const()[name = tensor("op_11546_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11546_cast_fp16 = einsum(equation = var_11546_equation_0, values = (var_11220_cast_fp16, var_11462_cast_fp16))[name = tensor("op_11546_cast_fp16")]; tensor var_11548_equation_0 = const()[name = tensor("op_11548_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11548_cast_fp16 = einsum(equation = var_11548_equation_0, values = (var_11224_cast_fp16, var_11463_cast_fp16))[name = tensor("op_11548_cast_fp16")]; tensor var_11550_equation_0 = const()[name = tensor("op_11550_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11550_cast_fp16 = einsum(equation = var_11550_equation_0, values = (var_11224_cast_fp16, var_11464_cast_fp16))[name = tensor("op_11550_cast_fp16")]; tensor var_11552_equation_0 = const()[name = tensor("op_11552_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11552_cast_fp16 = einsum(equation = var_11552_equation_0, values = (var_11224_cast_fp16, var_11465_cast_fp16))[name = tensor("op_11552_cast_fp16")]; tensor var_11554_equation_0 = const()[name = tensor("op_11554_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11554_cast_fp16 = einsum(equation = var_11554_equation_0, values = (var_11224_cast_fp16, var_11466_cast_fp16))[name = tensor("op_11554_cast_fp16")]; tensor var_11556_equation_0 = const()[name = tensor("op_11556_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11556_cast_fp16 = einsum(equation = var_11556_equation_0, values = (var_11228_cast_fp16, var_11467_cast_fp16))[name = tensor("op_11556_cast_fp16")]; tensor var_11558_equation_0 = const()[name = tensor("op_11558_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11558_cast_fp16 = einsum(equation = var_11558_equation_0, values = (var_11228_cast_fp16, var_11468_cast_fp16))[name = tensor("op_11558_cast_fp16")]; tensor var_11560_equation_0 = const()[name = tensor("op_11560_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11560_cast_fp16 = einsum(equation = var_11560_equation_0, values = (var_11228_cast_fp16, var_11469_cast_fp16))[name = tensor("op_11560_cast_fp16")]; tensor var_11562_equation_0 = const()[name = tensor("op_11562_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11562_cast_fp16 = einsum(equation = var_11562_equation_0, values = (var_11228_cast_fp16, var_11470_cast_fp16))[name = tensor("op_11562_cast_fp16")]; tensor var_11564_equation_0 = const()[name = tensor("op_11564_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11564_cast_fp16 = einsum(equation = var_11564_equation_0, values = (var_11232_cast_fp16, var_11471_cast_fp16))[name = tensor("op_11564_cast_fp16")]; tensor var_11566_equation_0 = const()[name = tensor("op_11566_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11566_cast_fp16 = einsum(equation = var_11566_equation_0, values = (var_11232_cast_fp16, var_11472_cast_fp16))[name = tensor("op_11566_cast_fp16")]; tensor var_11568_equation_0 = const()[name = tensor("op_11568_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11568_cast_fp16 = einsum(equation = var_11568_equation_0, values = (var_11232_cast_fp16, var_11473_cast_fp16))[name = tensor("op_11568_cast_fp16")]; tensor var_11570_equation_0 = const()[name = tensor("op_11570_equation_0"), val = tensor("bchk,bkhq->bchq")]; tensor var_11570_cast_fp16 = einsum(equation = var_11570_equation_0, values = (var_11232_cast_fp16, var_11474_cast_fp16))[name = tensor("op_11570_cast_fp16")]; tensor var_11572_interleave_0 = const()[name = tensor("op_11572_interleave_0"), val = tensor(false)]; tensor var_11572_cast_fp16 = concat(axis = var_10683, interleave = var_11572_interleave_0, values = (var_11476_cast_fp16, var_11478_cast_fp16, var_11480_cast_fp16, var_11482_cast_fp16))[name = tensor("op_11572_cast_fp16")]; tensor var_11574_interleave_0 = const()[name = tensor("op_11574_interleave_0"), val = tensor(false)]; tensor var_11574_cast_fp16 = concat(axis = var_10683, interleave = var_11574_interleave_0, values = (var_11484_cast_fp16, var_11486_cast_fp16, var_11488_cast_fp16, var_11490_cast_fp16))[name = tensor("op_11574_cast_fp16")]; tensor var_11576_interleave_0 = const()[name = tensor("op_11576_interleave_0"), val = tensor(false)]; tensor var_11576_cast_fp16 = concat(axis = var_10683, interleave = var_11576_interleave_0, values = (var_11492_cast_fp16, var_11494_cast_fp16, var_11496_cast_fp16, var_11498_cast_fp16))[name = tensor("op_11576_cast_fp16")]; tensor var_11578_interleave_0 = const()[name = tensor("op_11578_interleave_0"), val = tensor(false)]; tensor var_11578_cast_fp16 = concat(axis = var_10683, interleave = var_11578_interleave_0, values = (var_11500_cast_fp16, var_11502_cast_fp16, var_11504_cast_fp16, var_11506_cast_fp16))[name = tensor("op_11578_cast_fp16")]; tensor var_11580_interleave_0 = const()[name = tensor("op_11580_interleave_0"), val = tensor(false)]; tensor var_11580_cast_fp16 = concat(axis = var_10683, interleave = var_11580_interleave_0, values = (var_11508_cast_fp16, var_11510_cast_fp16, var_11512_cast_fp16, var_11514_cast_fp16))[name = tensor("op_11580_cast_fp16")]; tensor var_11582_interleave_0 = const()[name = tensor("op_11582_interleave_0"), val = tensor(false)]; tensor var_11582_cast_fp16 = concat(axis = var_10683, interleave = var_11582_interleave_0, values = (var_11516_cast_fp16, var_11518_cast_fp16, var_11520_cast_fp16, var_11522_cast_fp16))[name = tensor("op_11582_cast_fp16")]; tensor var_11584_interleave_0 = const()[name = tensor("op_11584_interleave_0"), val = tensor(false)]; tensor var_11584_cast_fp16 = concat(axis = var_10683, interleave = var_11584_interleave_0, values = (var_11524_cast_fp16, var_11526_cast_fp16, var_11528_cast_fp16, var_11530_cast_fp16))[name = tensor("op_11584_cast_fp16")]; tensor var_11586_interleave_0 = const()[name = tensor("op_11586_interleave_0"), val = tensor(false)]; tensor var_11586_cast_fp16 = concat(axis = var_10683, interleave = var_11586_interleave_0, values = (var_11532_cast_fp16, var_11534_cast_fp16, var_11536_cast_fp16, var_11538_cast_fp16))[name = tensor("op_11586_cast_fp16")]; tensor var_11588_interleave_0 = const()[name = tensor("op_11588_interleave_0"), val = tensor(false)]; tensor var_11588_cast_fp16 = concat(axis = var_10683, interleave = var_11588_interleave_0, values = (var_11540_cast_fp16, var_11542_cast_fp16, var_11544_cast_fp16, var_11546_cast_fp16))[name = tensor("op_11588_cast_fp16")]; tensor var_11590_interleave_0 = const()[name = tensor("op_11590_interleave_0"), val = tensor(false)]; tensor var_11590_cast_fp16 = concat(axis = var_10683, interleave = var_11590_interleave_0, values = (var_11548_cast_fp16, var_11550_cast_fp16, var_11552_cast_fp16, var_11554_cast_fp16))[name = tensor("op_11590_cast_fp16")]; tensor var_11592_interleave_0 = const()[name = tensor("op_11592_interleave_0"), val = tensor(false)]; tensor var_11592_cast_fp16 = concat(axis = var_10683, interleave = var_11592_interleave_0, values = (var_11556_cast_fp16, var_11558_cast_fp16, var_11560_cast_fp16, var_11562_cast_fp16))[name = tensor("op_11592_cast_fp16")]; tensor var_11594_interleave_0 = const()[name = tensor("op_11594_interleave_0"), val = tensor(false)]; tensor var_11594_cast_fp16 = concat(axis = var_10683, interleave = var_11594_interleave_0, values = (var_11564_cast_fp16, var_11566_cast_fp16, var_11568_cast_fp16, var_11570_cast_fp16))[name = tensor("op_11594_cast_fp16")]; tensor input_89_interleave_0 = const()[name = tensor("input_89_interleave_0"), val = tensor(false)]; tensor input_89_cast_fp16 = concat(axis = var_10700, interleave = input_89_interleave_0, values = (var_11572_cast_fp16, var_11574_cast_fp16, var_11576_cast_fp16, var_11578_cast_fp16, var_11580_cast_fp16, var_11582_cast_fp16, var_11584_cast_fp16, var_11586_cast_fp16, var_11588_cast_fp16, var_11590_cast_fp16, var_11592_cast_fp16, var_11594_cast_fp16))[name = tensor("input_89_cast_fp16")]; tensor var_11599 = const()[name = tensor("op_11599"), val = tensor([1, 1])]; tensor var_11601 = const()[name = tensor("op_11601"), val = tensor([1, 1])]; tensor obj_pad_type_0 = const()[name = tensor("obj_pad_type_0"), val = tensor("custom")]; tensor obj_pad_0 = const()[name = tensor("obj_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165690624)))]; tensor layers_11_self_attn_o_proj_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166870336)))]; tensor obj_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_bias_to_fp16, dilations = var_11601, groups = var_10700, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = var_11599, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_89_cast_fp16)[name = tensor("obj_cast_fp16")]; tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; tensor var_11607 = const()[name = tensor("op_11607"), val = tensor([1])]; tensor channels_mean_47_cast_fp16 = reduce_mean(axes = var_11607, keep_dims = var_10701, 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_11611 = const()[name = tensor("op_11611"), val = tensor([1])]; tensor var_11612_cast_fp16 = reduce_mean(axes = var_11611, keep_dims = var_10701, x = zero_mean_sq_47_cast_fp16)[name = tensor("op_11612_cast_fp16")]; tensor var_11613_to_fp16 = const()[name = tensor("op_11613_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11614_cast_fp16 = add(x = var_11612_cast_fp16, y = var_11613_to_fp16)[name = tensor("op_11614_cast_fp16")]; tensor denom_47_epsilon_0_to_fp16 = const()[name = tensor("denom_47_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_47_cast_fp16 = rsqrt(epsilon = denom_47_epsilon_0_to_fp16, x = var_11614_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_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(166871936)))]; 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(166873536)))]; 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_47_cast_fp16)[name = tensor("input_91_cast_fp16")]; tensor var_11625 = const()[name = tensor("op_11625"), val = tensor([1, 1])]; tensor var_11627 = const()[name = tensor("op_11627"), 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_11_fc1_weight_to_fp16 = const()[name = tensor("layers_11_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166875136)))]; tensor layers_11_fc1_bias_to_fp16 = const()[name = tensor("layers_11_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171593792)))]; tensor input_93_cast_fp16 = conv(bias = layers_11_fc1_bias_to_fp16, dilations = var_11627, groups = var_10700, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = var_11625, weight = layers_11_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; tensor input_mode_0 = const()[name = tensor("input_mode_0"), val = tensor("EXACT")]; tensor input_cast_fp16 = gelu(mode = input_mode_0, x = input_93_cast_fp16)[name = tensor("input_cast_fp16")]; tensor var_11633 = const()[name = tensor("op_11633"), val = tensor([1, 1])]; tensor var_11635 = const()[name = tensor("op_11635"), val = tensor([1, 1])]; tensor hidden_states_pad_type_0 = const()[name = tensor("hidden_states_pad_type_0"), val = tensor("custom")]; tensor hidden_states_pad_0 = const()[name = tensor("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; tensor layers_11_fc2_weight_to_fp16 = const()[name = tensor("layers_11_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171600000)))]; tensor layers_11_fc2_bias_to_fp16 = const()[name = tensor("layers_11_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176318656)))]; tensor hidden_states_cast_fp16 = conv(bias = layers_11_fc2_bias_to_fp16, dilations = var_11635, groups = var_10700, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_11633, weight = layers_11_fc2_weight_to_fp16, x = input_cast_fp16)[name = tensor("hidden_states_cast_fp16")]; tensor inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_cast_fp16)[name = tensor("inputs_cast_fp16")]; tensor var_11641 = const()[name = tensor("op_11641"), val = tensor(true)]; tensor var_11645 = const()[name = tensor("op_11645"), val = tensor([1])]; tensor channels_mean_cast_fp16 = reduce_mean(axes = var_11645, keep_dims = var_11641, 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_11649 = const()[name = tensor("op_11649"), val = tensor([1])]; tensor var_11650_cast_fp16 = reduce_mean(axes = var_11649, keep_dims = var_11641, x = zero_mean_sq_cast_fp16)[name = tensor("op_11650_cast_fp16")]; tensor var_11651_to_fp16 = const()[name = tensor("op_11651_to_fp16"), val = tensor(0x1.5p-17)]; tensor var_11652_cast_fp16 = add(x = var_11650_cast_fp16, y = var_11651_to_fp16)[name = tensor("op_11652_cast_fp16")]; tensor denom_epsilon_0_to_fp16 = const()[name = tensor("denom_epsilon_0_to_fp16"), val = tensor(0x1p-24)]; tensor denom_cast_fp16 = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_11652_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(176320256)))]; 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(176321856)))]; 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); }