diff --git "a/text_encoder/coreml_model.mlmodelc/model.mil" "b/text_encoder/coreml_model.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/text_encoder/coreml_model.mlmodelc/model.mil" @@ -0,0 +1,872 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})] +{ + func main(tensor input_ids) { + tensor var_5 = const()[name = tensor("op_5"), val = tensor(-1)]; + tensor var_6 = const()[name = tensor("op_6"), val = tensor(false)]; + tensor cast_1_dtype_0 = const()[name = tensor("cast_1_dtype_0"), val = tensor("int32")]; + tensor inputs_embeds_axis_0 = const()[name = tensor("inputs_embeds_axis_0"), val = tensor(0)]; + tensor inputs_embeds_batch_dims_0 = const()[name = tensor("inputs_embeds_batch_dims_0"), val = tensor(0)]; + tensor text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor cast_2 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor("cast_2")]; + tensor inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_2, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor("inputs_embeds_cast_fp16")]; + tensor position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75890816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75935232))), name = tensor("position_embeddings_to_fp16_palettized"), shape = tensor([1, 77, 768])]; + tensor input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16_palettized)[name = tensor("input_3_cast_fp16")]; + tensor hidden_states_1_axes_0 = const()[name = tensor("hidden_states_1_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75935424)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75937024)))]; + tensor var_15_to_fp16 = const()[name = tensor("op_15_to_fp16"), val = tensor(0x1.5p-17)]; + tensor hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("hidden_states_1_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75938624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76381056))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76381248)))]; + tensor linear_0_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_106_to_fp16 = const()[name = tensor("op_106_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_106_to_fp16)[name = tensor("tensor_5_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76382848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76825280))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76825472)))]; + tensor linear_1_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_111 = const()[name = tensor("op_111"), val = tensor([1, -1, 12, 64])]; + tensor var_112_cast_fp16 = reshape(shape = var_111, x = linear_1_cast_fp16)[name = tensor("op_112_cast_fp16")]; + tensor var_113_perm_0 = const()[name = tensor("op_113_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76827072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77269504))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77269696)))]; + tensor linear_2_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_118 = const()[name = tensor("op_118"), val = tensor([1, -1, 12, 64])]; + tensor var_119_cast_fp16 = reshape(shape = var_118, x = linear_2_cast_fp16)[name = tensor("op_119_cast_fp16")]; + tensor var_120_perm_0 = const()[name = tensor("op_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_127 = const()[name = tensor("op_127"), val = tensor([1, 77, 12, 64])]; + tensor var_128_cast_fp16 = reshape(shape = var_127, x = tensor_5_cast_fp16)[name = tensor("op_128_cast_fp16")]; + tensor var_129_perm_0 = const()[name = tensor("op_129_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_131 = const()[name = tensor("op_131"), val = tensor([12, -1, 64])]; + tensor transpose_47 = transpose(perm = var_129_perm_0, x = var_128_cast_fp16)[name = tensor("transpose_47")]; + tensor query_states_1_cast_fp16 = reshape(shape = var_131, x = transpose_47)[name = tensor("query_states_1_cast_fp16")]; + tensor var_133 = const()[name = tensor("op_133"), val = tensor([12, -1, 64])]; + tensor transpose_46 = transpose(perm = var_113_perm_0, x = var_112_cast_fp16)[name = tensor("transpose_46")]; + tensor key_states_3_cast_fp16 = reshape(shape = var_133, x = transpose_46)[name = tensor("key_states_3_cast_fp16")]; + tensor var_135 = const()[name = tensor("op_135"), val = tensor([12, -1, 64])]; + tensor transpose_45 = transpose(perm = var_120_perm_0, x = var_119_cast_fp16)[name = tensor("transpose_45")]; + tensor value_states_3_cast_fp16 = reshape(shape = var_135, x = transpose_45)[name = tensor("value_states_3_cast_fp16")]; + tensor attn_weights_1_transpose_x_1 = const()[name = tensor("attn_weights_1_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_1_transpose_y_1 = const()[name = tensor("attn_weights_1_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = query_states_1_cast_fp16, y = key_states_3_cast_fp16)[name = tensor("attn_weights_1_cast_fp16")]; + tensor var_140 = const()[name = tensor("op_140"), val = tensor([1, 12, 77, 77])]; + tensor var_141_cast_fp16 = reshape(shape = var_140, x = attn_weights_1_cast_fp16)[name = tensor("op_141_cast_fp16")]; + tensor op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77271296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77275840))), name = tensor("op_56_to_fp16_palettized"), shape = tensor([1, 1, 77, 77])]; + tensor attn_weights_3_cast_fp16 = add(x = var_141_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_3_cast_fp16")]; + tensor var_146 = const()[name = tensor("op_146"), val = tensor([12, 77, 77])]; + tensor input_5_cast_fp16 = reshape(shape = var_146, x = attn_weights_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor attn_output_1_transpose_x_0 = const()[name = tensor("attn_output_1_transpose_x_0"), val = tensor(false)]; + tensor attn_output_1_transpose_y_0 = const()[name = tensor("attn_output_1_transpose_y_0"), val = tensor(false)]; + tensor attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast_fp16, y = value_states_3_cast_fp16)[name = tensor("attn_output_1_cast_fp16")]; + tensor var_151 = const()[name = tensor("op_151"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_3_cast_fp16 = reshape(shape = var_151, x = attn_output_1_cast_fp16)[name = tensor("attn_output_3_cast_fp16")]; + tensor attn_output_5_perm_0 = const()[name = tensor("attn_output_5_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_154 = const()[name = tensor("op_154"), val = tensor([1, 77, 768])]; + tensor transpose_44 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor("transpose_44")]; + tensor input_9_cast_fp16 = reshape(shape = var_154, x = transpose_44)[name = tensor("input_9_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77276032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77718464))), name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77718656)))]; + tensor linear_3_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor input_13_axes_0 = const()[name = tensor("input_13_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77720256)))]; + tensor text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77721856)))]; + tensor input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77723456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79492992))), name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79493184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79495552))), name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor var_169_to_fp16 = const()[name = tensor("op_169_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_170_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_169_to_fp16)[name = tensor("op_170_cast_fp16")]; + tensor var_171_cast_fp16 = sigmoid(x = var_170_cast_fp16)[name = tensor("op_171_cast_fp16")]; + tensor input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_171_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79495744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81265280))), name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81265472)))]; + tensor linear_5_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor hidden_states_7_axes_0 = const()[name = tensor("hidden_states_7_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81267072)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81268672)))]; + tensor hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor("hidden_states_7_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81270272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81712704))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81712896)))]; + tensor linear_6_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_196_to_fp16)[name = tensor("tensor_11_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81714496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82156928))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82157120)))]; + tensor linear_7_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_201 = const()[name = tensor("op_201"), val = tensor([1, -1, 12, 64])]; + tensor var_202_cast_fp16 = reshape(shape = var_201, x = linear_7_cast_fp16)[name = tensor("op_202_cast_fp16")]; + tensor var_203_perm_0 = const()[name = tensor("op_203_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82158720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82601152))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82601344)))]; + tensor linear_8_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_208 = const()[name = tensor("op_208"), val = tensor([1, -1, 12, 64])]; + tensor var_209_cast_fp16 = reshape(shape = var_208, x = linear_8_cast_fp16)[name = tensor("op_209_cast_fp16")]; + tensor var_210_perm_0 = const()[name = tensor("op_210_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_217 = const()[name = tensor("op_217"), val = tensor([1, 77, 12, 64])]; + tensor var_218_cast_fp16 = reshape(shape = var_217, x = tensor_11_cast_fp16)[name = tensor("op_218_cast_fp16")]; + tensor var_219_perm_0 = const()[name = tensor("op_219_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_221 = const()[name = tensor("op_221"), val = tensor([12, -1, 64])]; + tensor transpose_43 = transpose(perm = var_219_perm_0, x = var_218_cast_fp16)[name = tensor("transpose_43")]; + tensor query_states_3_cast_fp16 = reshape(shape = var_221, x = transpose_43)[name = tensor("query_states_3_cast_fp16")]; + tensor var_223 = const()[name = tensor("op_223"), val = tensor([12, -1, 64])]; + tensor transpose_42 = transpose(perm = var_203_perm_0, x = var_202_cast_fp16)[name = tensor("transpose_42")]; + tensor key_states_7_cast_fp16 = reshape(shape = var_223, x = transpose_42)[name = tensor("key_states_7_cast_fp16")]; + tensor var_225 = const()[name = tensor("op_225"), val = tensor([12, -1, 64])]; + tensor transpose_41 = transpose(perm = var_210_perm_0, x = var_209_cast_fp16)[name = tensor("transpose_41")]; + tensor value_states_7_cast_fp16 = reshape(shape = var_225, x = transpose_41)[name = tensor("value_states_7_cast_fp16")]; + tensor attn_weights_7_transpose_x_1 = const()[name = tensor("attn_weights_7_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_7_transpose_y_1 = const()[name = tensor("attn_weights_7_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = query_states_3_cast_fp16, y = key_states_7_cast_fp16)[name = tensor("attn_weights_7_cast_fp16")]; + tensor var_230 = const()[name = tensor("op_230"), val = tensor([1, 12, 77, 77])]; + tensor var_231_cast_fp16 = reshape(shape = var_230, x = attn_weights_7_cast_fp16)[name = tensor("op_231_cast_fp16")]; + tensor attn_weights_9_cast_fp16 = add(x = var_231_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_9_cast_fp16")]; + tensor var_236 = const()[name = tensor("op_236"), val = tensor([12, 77, 77])]; + tensor input_21_cast_fp16 = reshape(shape = var_236, x = attn_weights_9_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor attn_output_7_transpose_x_0 = const()[name = tensor("attn_output_7_transpose_x_0"), val = tensor(false)]; + tensor attn_output_7_transpose_y_0 = const()[name = tensor("attn_output_7_transpose_y_0"), val = tensor(false)]; + tensor attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast_fp16, y = value_states_7_cast_fp16)[name = tensor("attn_output_7_cast_fp16")]; + tensor var_241 = const()[name = tensor("op_241"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_9_cast_fp16 = reshape(shape = var_241, x = attn_output_7_cast_fp16)[name = tensor("attn_output_9_cast_fp16")]; + tensor attn_output_11_perm_0 = const()[name = tensor("attn_output_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_244 = const()[name = tensor("op_244"), val = tensor([1, 77, 768])]; + tensor transpose_40 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor("transpose_40")]; + tensor input_25_cast_fp16 = reshape(shape = var_244, x = transpose_40)[name = tensor("input_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82602944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83045376))), name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83045568)))]; + tensor linear_9_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor input_29_axes_0 = const()[name = tensor("input_29_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83047168)))]; + tensor text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83048768)))]; + tensor input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83050368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84819904))), name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84820096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84822464))), name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor var_259_to_fp16 = const()[name = tensor("op_259_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_260_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_259_to_fp16)[name = tensor("op_260_cast_fp16")]; + tensor var_261_cast_fp16 = sigmoid(x = var_260_cast_fp16)[name = tensor("op_261_cast_fp16")]; + tensor input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_261_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84822656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86592192))), name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86592384)))]; + tensor linear_11_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor hidden_states_13_axes_0 = const()[name = tensor("hidden_states_13_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86593984)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86595584)))]; + tensor hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor("hidden_states_13_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86597184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87039616))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87039808)))]; + tensor linear_12_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_286_to_fp16)[name = tensor("tensor_17_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87041408))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87483840))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87484032)))]; + tensor linear_13_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_291 = const()[name = tensor("op_291"), val = tensor([1, -1, 12, 64])]; + tensor var_292_cast_fp16 = reshape(shape = var_291, x = linear_13_cast_fp16)[name = tensor("op_292_cast_fp16")]; + tensor var_293_perm_0 = const()[name = tensor("op_293_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87485632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87928064))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87928256)))]; + tensor linear_14_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_298 = const()[name = tensor("op_298"), val = tensor([1, -1, 12, 64])]; + tensor var_299_cast_fp16 = reshape(shape = var_298, x = linear_14_cast_fp16)[name = tensor("op_299_cast_fp16")]; + tensor var_300_perm_0 = const()[name = tensor("op_300_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_307 = const()[name = tensor("op_307"), val = tensor([1, 77, 12, 64])]; + tensor var_308_cast_fp16 = reshape(shape = var_307, x = tensor_17_cast_fp16)[name = tensor("op_308_cast_fp16")]; + tensor var_309_perm_0 = const()[name = tensor("op_309_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_311 = const()[name = tensor("op_311"), val = tensor([12, -1, 64])]; + tensor transpose_39 = transpose(perm = var_309_perm_0, x = var_308_cast_fp16)[name = tensor("transpose_39")]; + tensor query_states_5_cast_fp16 = reshape(shape = var_311, x = transpose_39)[name = tensor("query_states_5_cast_fp16")]; + tensor var_313 = const()[name = tensor("op_313"), val = tensor([12, -1, 64])]; + tensor transpose_38 = transpose(perm = var_293_perm_0, x = var_292_cast_fp16)[name = tensor("transpose_38")]; + tensor key_states_11_cast_fp16 = reshape(shape = var_313, x = transpose_38)[name = tensor("key_states_11_cast_fp16")]; + tensor var_315 = const()[name = tensor("op_315"), val = tensor([12, -1, 64])]; + tensor transpose_37 = transpose(perm = var_300_perm_0, x = var_299_cast_fp16)[name = tensor("transpose_37")]; + tensor value_states_11_cast_fp16 = reshape(shape = var_315, x = transpose_37)[name = tensor("value_states_11_cast_fp16")]; + tensor attn_weights_13_transpose_x_1 = const()[name = tensor("attn_weights_13_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_13_transpose_y_1 = const()[name = tensor("attn_weights_13_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = query_states_5_cast_fp16, y = key_states_11_cast_fp16)[name = tensor("attn_weights_13_cast_fp16")]; + tensor var_320 = const()[name = tensor("op_320"), val = tensor([1, 12, 77, 77])]; + tensor var_321_cast_fp16 = reshape(shape = var_320, x = attn_weights_13_cast_fp16)[name = tensor("op_321_cast_fp16")]; + tensor attn_weights_15_cast_fp16 = add(x = var_321_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_15_cast_fp16")]; + tensor var_326 = const()[name = tensor("op_326"), val = tensor([12, 77, 77])]; + tensor input_37_cast_fp16 = reshape(shape = var_326, x = attn_weights_15_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor attn_output_13_transpose_x_0 = const()[name = tensor("attn_output_13_transpose_x_0"), val = tensor(false)]; + tensor attn_output_13_transpose_y_0 = const()[name = tensor("attn_output_13_transpose_y_0"), val = tensor(false)]; + tensor attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast_fp16, y = value_states_11_cast_fp16)[name = tensor("attn_output_13_cast_fp16")]; + tensor var_331 = const()[name = tensor("op_331"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_15_cast_fp16 = reshape(shape = var_331, x = attn_output_13_cast_fp16)[name = tensor("attn_output_15_cast_fp16")]; + tensor attn_output_17_perm_0 = const()[name = tensor("attn_output_17_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_334 = const()[name = tensor("op_334"), val = tensor([1, 77, 768])]; + tensor transpose_36 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor("transpose_36")]; + tensor input_41_cast_fp16 = reshape(shape = var_334, x = transpose_36)[name = tensor("input_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87929856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88372288))), name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88372480)))]; + tensor linear_15_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor input_45_axes_0 = const()[name = tensor("input_45_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88374080)))]; + tensor text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88375680)))]; + tensor input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88377280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90146816))), name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90147008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90149376))), name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor var_349_to_fp16 = const()[name = tensor("op_349_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_350_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_349_to_fp16)[name = tensor("op_350_cast_fp16")]; + tensor var_351_cast_fp16 = sigmoid(x = var_350_cast_fp16)[name = tensor("op_351_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_351_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90149568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91919104))), name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91919296)))]; + tensor linear_17_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor("input_51_cast_fp16")]; + tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91920896)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91922496)))]; + tensor hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor("hidden_states_19_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91924096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92366528))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92366720)))]; + tensor linear_18_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_376_to_fp16 = const()[name = tensor("op_376_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_376_to_fp16)[name = tensor("tensor_23_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92368320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92810752))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92810944)))]; + tensor linear_19_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, -1, 12, 64])]; + tensor var_382_cast_fp16 = reshape(shape = var_381, x = linear_19_cast_fp16)[name = tensor("op_382_cast_fp16")]; + tensor var_383_perm_0 = const()[name = tensor("op_383_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92812544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93254976))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93255168)))]; + tensor linear_20_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, -1, 12, 64])]; + tensor var_389_cast_fp16 = reshape(shape = var_388, x = linear_20_cast_fp16)[name = tensor("op_389_cast_fp16")]; + tensor var_390_perm_0 = const()[name = tensor("op_390_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_397 = const()[name = tensor("op_397"), val = tensor([1, 77, 12, 64])]; + tensor var_398_cast_fp16 = reshape(shape = var_397, x = tensor_23_cast_fp16)[name = tensor("op_398_cast_fp16")]; + tensor var_399_perm_0 = const()[name = tensor("op_399_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_401 = const()[name = tensor("op_401"), val = tensor([12, -1, 64])]; + tensor transpose_35 = transpose(perm = var_399_perm_0, x = var_398_cast_fp16)[name = tensor("transpose_35")]; + tensor query_states_7_cast_fp16 = reshape(shape = var_401, x = transpose_35)[name = tensor("query_states_7_cast_fp16")]; + tensor var_403 = const()[name = tensor("op_403"), val = tensor([12, -1, 64])]; + tensor transpose_34 = transpose(perm = var_383_perm_0, x = var_382_cast_fp16)[name = tensor("transpose_34")]; + tensor key_states_15_cast_fp16 = reshape(shape = var_403, x = transpose_34)[name = tensor("key_states_15_cast_fp16")]; + tensor var_405 = const()[name = tensor("op_405"), val = tensor([12, -1, 64])]; + tensor transpose_33 = transpose(perm = var_390_perm_0, x = var_389_cast_fp16)[name = tensor("transpose_33")]; + tensor value_states_15_cast_fp16 = reshape(shape = var_405, x = transpose_33)[name = tensor("value_states_15_cast_fp16")]; + tensor attn_weights_19_transpose_x_1 = const()[name = tensor("attn_weights_19_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_19_transpose_y_1 = const()[name = tensor("attn_weights_19_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = query_states_7_cast_fp16, y = key_states_15_cast_fp16)[name = tensor("attn_weights_19_cast_fp16")]; + tensor var_410 = const()[name = tensor("op_410"), val = tensor([1, 12, 77, 77])]; + tensor var_411_cast_fp16 = reshape(shape = var_410, x = attn_weights_19_cast_fp16)[name = tensor("op_411_cast_fp16")]; + tensor attn_weights_21_cast_fp16 = add(x = var_411_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_21_cast_fp16")]; + tensor var_416 = const()[name = tensor("op_416"), val = tensor([12, 77, 77])]; + tensor input_53_cast_fp16 = reshape(shape = var_416, x = attn_weights_21_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor attn_output_19_transpose_x_0 = const()[name = tensor("attn_output_19_transpose_x_0"), val = tensor(false)]; + tensor attn_output_19_transpose_y_0 = const()[name = tensor("attn_output_19_transpose_y_0"), val = tensor(false)]; + tensor attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast_fp16, y = value_states_15_cast_fp16)[name = tensor("attn_output_19_cast_fp16")]; + tensor var_421 = const()[name = tensor("op_421"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_21_cast_fp16 = reshape(shape = var_421, x = attn_output_19_cast_fp16)[name = tensor("attn_output_21_cast_fp16")]; + tensor attn_output_23_perm_0 = const()[name = tensor("attn_output_23_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_424 = const()[name = tensor("op_424"), val = tensor([1, 77, 768])]; + tensor transpose_32 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor("transpose_32")]; + tensor input_57_cast_fp16 = reshape(shape = var_424, x = transpose_32)[name = tensor("input_57_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93256768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93699200))), name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93699392)))]; + tensor linear_21_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor input_61_axes_0 = const()[name = tensor("input_61_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93700992)))]; + tensor text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93702592)))]; + tensor input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93704192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95473728))), name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95473920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95476288))), name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor var_439_to_fp16 = const()[name = tensor("op_439_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_440_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_439_to_fp16)[name = tensor("op_440_cast_fp16")]; + tensor var_441_cast_fp16 = sigmoid(x = var_440_cast_fp16)[name = tensor("op_441_cast_fp16")]; + tensor input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_441_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95476480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97246016))), name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97246208)))]; + tensor linear_23_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97247808)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97249408)))]; + tensor hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor("hidden_states_25_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97251008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97693440))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97693632)))]; + tensor linear_24_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_466_to_fp16 = const()[name = tensor("op_466_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_466_to_fp16)[name = tensor("tensor_29_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97695232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98137664))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98137856)))]; + tensor linear_25_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_471 = const()[name = tensor("op_471"), val = tensor([1, -1, 12, 64])]; + tensor var_472_cast_fp16 = reshape(shape = var_471, x = linear_25_cast_fp16)[name = tensor("op_472_cast_fp16")]; + tensor var_473_perm_0 = const()[name = tensor("op_473_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98139456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98581888))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98582080)))]; + tensor linear_26_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_478 = const()[name = tensor("op_478"), val = tensor([1, -1, 12, 64])]; + tensor var_479_cast_fp16 = reshape(shape = var_478, x = linear_26_cast_fp16)[name = tensor("op_479_cast_fp16")]; + tensor var_480_perm_0 = const()[name = tensor("op_480_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_487 = const()[name = tensor("op_487"), val = tensor([1, 77, 12, 64])]; + tensor var_488_cast_fp16 = reshape(shape = var_487, x = tensor_29_cast_fp16)[name = tensor("op_488_cast_fp16")]; + tensor var_489_perm_0 = const()[name = tensor("op_489_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_491 = const()[name = tensor("op_491"), val = tensor([12, -1, 64])]; + tensor transpose_31 = transpose(perm = var_489_perm_0, x = var_488_cast_fp16)[name = tensor("transpose_31")]; + tensor query_states_9_cast_fp16 = reshape(shape = var_491, x = transpose_31)[name = tensor("query_states_9_cast_fp16")]; + tensor var_493 = const()[name = tensor("op_493"), val = tensor([12, -1, 64])]; + tensor transpose_30 = transpose(perm = var_473_perm_0, x = var_472_cast_fp16)[name = tensor("transpose_30")]; + tensor key_states_19_cast_fp16 = reshape(shape = var_493, x = transpose_30)[name = tensor("key_states_19_cast_fp16")]; + tensor var_495 = const()[name = tensor("op_495"), val = tensor([12, -1, 64])]; + tensor transpose_29 = transpose(perm = var_480_perm_0, x = var_479_cast_fp16)[name = tensor("transpose_29")]; + tensor value_states_19_cast_fp16 = reshape(shape = var_495, x = transpose_29)[name = tensor("value_states_19_cast_fp16")]; + tensor attn_weights_25_transpose_x_1 = const()[name = tensor("attn_weights_25_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_25_transpose_y_1 = const()[name = tensor("attn_weights_25_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = query_states_9_cast_fp16, y = key_states_19_cast_fp16)[name = tensor("attn_weights_25_cast_fp16")]; + tensor var_500 = const()[name = tensor("op_500"), val = tensor([1, 12, 77, 77])]; + tensor var_501_cast_fp16 = reshape(shape = var_500, x = attn_weights_25_cast_fp16)[name = tensor("op_501_cast_fp16")]; + tensor attn_weights_27_cast_fp16 = add(x = var_501_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_27_cast_fp16")]; + tensor var_506 = const()[name = tensor("op_506"), val = tensor([12, 77, 77])]; + tensor input_69_cast_fp16 = reshape(shape = var_506, x = attn_weights_27_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor attn_output_25_transpose_x_0 = const()[name = tensor("attn_output_25_transpose_x_0"), val = tensor(false)]; + tensor attn_output_25_transpose_y_0 = const()[name = tensor("attn_output_25_transpose_y_0"), val = tensor(false)]; + tensor attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast_fp16, y = value_states_19_cast_fp16)[name = tensor("attn_output_25_cast_fp16")]; + tensor var_511 = const()[name = tensor("op_511"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_27_cast_fp16 = reshape(shape = var_511, x = attn_output_25_cast_fp16)[name = tensor("attn_output_27_cast_fp16")]; + tensor attn_output_29_perm_0 = const()[name = tensor("attn_output_29_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_514 = const()[name = tensor("op_514"), val = tensor([1, 77, 768])]; + tensor transpose_28 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor("transpose_28")]; + tensor input_73_cast_fp16 = reshape(shape = var_514, x = transpose_28)[name = tensor("input_73_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98583680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99026112))), name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99026304)))]; + tensor linear_27_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor input_77_axes_0 = const()[name = tensor("input_77_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99027904)))]; + tensor text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99029504)))]; + tensor input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99031104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100800640))), name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100800832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100803200))), name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor var_529_to_fp16 = const()[name = tensor("op_529_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_530_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_529_to_fp16)[name = tensor("op_530_cast_fp16")]; + tensor var_531_cast_fp16 = sigmoid(x = var_530_cast_fp16)[name = tensor("op_531_cast_fp16")]; + tensor input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_531_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100803392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102572928))), name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102573120)))]; + tensor linear_29_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor("input_83_cast_fp16")]; + tensor hidden_states_31_axes_0 = const()[name = tensor("hidden_states_31_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102574720)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102576320)))]; + tensor hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor("hidden_states_31_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102577920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103020352))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103020544)))]; + tensor linear_30_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_556_to_fp16)[name = tensor("tensor_35_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103022144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103464576))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103464768)))]; + tensor linear_31_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_561 = const()[name = tensor("op_561"), val = tensor([1, -1, 12, 64])]; + tensor var_562_cast_fp16 = reshape(shape = var_561, x = linear_31_cast_fp16)[name = tensor("op_562_cast_fp16")]; + tensor var_563_perm_0 = const()[name = tensor("op_563_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103466368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103908800))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103908992)))]; + tensor linear_32_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_568 = const()[name = tensor("op_568"), val = tensor([1, -1, 12, 64])]; + tensor var_569_cast_fp16 = reshape(shape = var_568, x = linear_32_cast_fp16)[name = tensor("op_569_cast_fp16")]; + tensor var_570_perm_0 = const()[name = tensor("op_570_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_577 = const()[name = tensor("op_577"), val = tensor([1, 77, 12, 64])]; + tensor var_578_cast_fp16 = reshape(shape = var_577, x = tensor_35_cast_fp16)[name = tensor("op_578_cast_fp16")]; + tensor var_579_perm_0 = const()[name = tensor("op_579_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_581 = const()[name = tensor("op_581"), val = tensor([12, -1, 64])]; + tensor transpose_27 = transpose(perm = var_579_perm_0, x = var_578_cast_fp16)[name = tensor("transpose_27")]; + tensor query_states_11_cast_fp16 = reshape(shape = var_581, x = transpose_27)[name = tensor("query_states_11_cast_fp16")]; + tensor var_583 = const()[name = tensor("op_583"), val = tensor([12, -1, 64])]; + tensor transpose_26 = transpose(perm = var_563_perm_0, x = var_562_cast_fp16)[name = tensor("transpose_26")]; + tensor key_states_23_cast_fp16 = reshape(shape = var_583, x = transpose_26)[name = tensor("key_states_23_cast_fp16")]; + tensor var_585 = const()[name = tensor("op_585"), val = tensor([12, -1, 64])]; + tensor transpose_25 = transpose(perm = var_570_perm_0, x = var_569_cast_fp16)[name = tensor("transpose_25")]; + tensor value_states_23_cast_fp16 = reshape(shape = var_585, x = transpose_25)[name = tensor("value_states_23_cast_fp16")]; + tensor attn_weights_31_transpose_x_1 = const()[name = tensor("attn_weights_31_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_31_transpose_y_1 = const()[name = tensor("attn_weights_31_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = query_states_11_cast_fp16, y = key_states_23_cast_fp16)[name = tensor("attn_weights_31_cast_fp16")]; + tensor var_590 = const()[name = tensor("op_590"), val = tensor([1, 12, 77, 77])]; + tensor var_591_cast_fp16 = reshape(shape = var_590, x = attn_weights_31_cast_fp16)[name = tensor("op_591_cast_fp16")]; + tensor attn_weights_33_cast_fp16 = add(x = var_591_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_33_cast_fp16")]; + tensor var_596 = const()[name = tensor("op_596"), val = tensor([12, 77, 77])]; + tensor input_85_cast_fp16 = reshape(shape = var_596, x = attn_weights_33_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor attn_output_31_transpose_x_0 = const()[name = tensor("attn_output_31_transpose_x_0"), val = tensor(false)]; + tensor attn_output_31_transpose_y_0 = const()[name = tensor("attn_output_31_transpose_y_0"), val = tensor(false)]; + tensor attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast_fp16, y = value_states_23_cast_fp16)[name = tensor("attn_output_31_cast_fp16")]; + tensor var_601 = const()[name = tensor("op_601"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_33_cast_fp16 = reshape(shape = var_601, x = attn_output_31_cast_fp16)[name = tensor("attn_output_33_cast_fp16")]; + tensor attn_output_35_perm_0 = const()[name = tensor("attn_output_35_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_604 = const()[name = tensor("op_604"), val = tensor([1, 77, 768])]; + tensor transpose_24 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor("transpose_24")]; + tensor input_89_cast_fp16 = reshape(shape = var_604, x = transpose_24)[name = tensor("input_89_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103910592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104353024))), name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104353216)))]; + tensor linear_33_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor input_93_axes_0 = const()[name = tensor("input_93_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104354816)))]; + tensor text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104356416)))]; + tensor input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104358016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106127552))), name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106127744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130112))), name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor var_619_to_fp16 = const()[name = tensor("op_619_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_620_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_619_to_fp16)[name = tensor("op_620_cast_fp16")]; + tensor var_621_cast_fp16 = sigmoid(x = var_620_cast_fp16)[name = tensor("op_621_cast_fp16")]; + tensor input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_621_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107899840))), name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107900032)))]; + tensor linear_35_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor hidden_states_37_axes_0 = const()[name = tensor("hidden_states_37_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107901632)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107903232)))]; + tensor hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor("hidden_states_37_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107904832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108347264))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108347456)))]; + tensor linear_36_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_646_to_fp16)[name = tensor("tensor_41_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108349056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108791488))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108791680)))]; + tensor linear_37_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_651 = const()[name = tensor("op_651"), val = tensor([1, -1, 12, 64])]; + tensor var_652_cast_fp16 = reshape(shape = var_651, x = linear_37_cast_fp16)[name = tensor("op_652_cast_fp16")]; + tensor var_653_perm_0 = const()[name = tensor("op_653_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108793280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109235712))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109235904)))]; + tensor linear_38_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_658 = const()[name = tensor("op_658"), val = tensor([1, -1, 12, 64])]; + tensor var_659_cast_fp16 = reshape(shape = var_658, x = linear_38_cast_fp16)[name = tensor("op_659_cast_fp16")]; + tensor var_660_perm_0 = const()[name = tensor("op_660_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_667 = const()[name = tensor("op_667"), val = tensor([1, 77, 12, 64])]; + tensor var_668_cast_fp16 = reshape(shape = var_667, x = tensor_41_cast_fp16)[name = tensor("op_668_cast_fp16")]; + tensor var_669_perm_0 = const()[name = tensor("op_669_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_671 = const()[name = tensor("op_671"), val = tensor([12, -1, 64])]; + tensor transpose_23 = transpose(perm = var_669_perm_0, x = var_668_cast_fp16)[name = tensor("transpose_23")]; + tensor query_states_13_cast_fp16 = reshape(shape = var_671, x = transpose_23)[name = tensor("query_states_13_cast_fp16")]; + tensor var_673 = const()[name = tensor("op_673"), val = tensor([12, -1, 64])]; + tensor transpose_22 = transpose(perm = var_653_perm_0, x = var_652_cast_fp16)[name = tensor("transpose_22")]; + tensor key_states_27_cast_fp16 = reshape(shape = var_673, x = transpose_22)[name = tensor("key_states_27_cast_fp16")]; + tensor var_675 = const()[name = tensor("op_675"), val = tensor([12, -1, 64])]; + tensor transpose_21 = transpose(perm = var_660_perm_0, x = var_659_cast_fp16)[name = tensor("transpose_21")]; + tensor value_states_27_cast_fp16 = reshape(shape = var_675, x = transpose_21)[name = tensor("value_states_27_cast_fp16")]; + tensor attn_weights_37_transpose_x_1 = const()[name = tensor("attn_weights_37_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_37_transpose_y_1 = const()[name = tensor("attn_weights_37_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = query_states_13_cast_fp16, y = key_states_27_cast_fp16)[name = tensor("attn_weights_37_cast_fp16")]; + tensor var_680 = const()[name = tensor("op_680"), val = tensor([1, 12, 77, 77])]; + tensor var_681_cast_fp16 = reshape(shape = var_680, x = attn_weights_37_cast_fp16)[name = tensor("op_681_cast_fp16")]; + tensor attn_weights_39_cast_fp16 = add(x = var_681_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_39_cast_fp16")]; + tensor var_686 = const()[name = tensor("op_686"), val = tensor([12, 77, 77])]; + tensor input_101_cast_fp16 = reshape(shape = var_686, x = attn_weights_39_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor attn_output_37_transpose_x_0 = const()[name = tensor("attn_output_37_transpose_x_0"), val = tensor(false)]; + tensor attn_output_37_transpose_y_0 = const()[name = tensor("attn_output_37_transpose_y_0"), val = tensor(false)]; + tensor attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast_fp16, y = value_states_27_cast_fp16)[name = tensor("attn_output_37_cast_fp16")]; + tensor var_691 = const()[name = tensor("op_691"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_39_cast_fp16 = reshape(shape = var_691, x = attn_output_37_cast_fp16)[name = tensor("attn_output_39_cast_fp16")]; + tensor attn_output_41_perm_0 = const()[name = tensor("attn_output_41_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_694 = const()[name = tensor("op_694"), val = tensor([1, 77, 768])]; + tensor transpose_20 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor("transpose_20")]; + tensor input_105_cast_fp16 = reshape(shape = var_694, x = transpose_20)[name = tensor("input_105_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109237504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109679936))), name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109680128)))]; + tensor linear_39_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor input_109_axes_0 = const()[name = tensor("input_109_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109681728)))]; + tensor text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109683328)))]; + tensor input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast_fp16)[name = tensor("input_109_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109684928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111454464))), name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111454656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111457024))), name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor var_709_to_fp16 = const()[name = tensor("op_709_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_710_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_709_to_fp16)[name = tensor("op_710_cast_fp16")]; + tensor var_711_cast_fp16 = sigmoid(x = var_710_cast_fp16)[name = tensor("op_711_cast_fp16")]; + tensor input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_711_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111457216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113226752))), name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113226944)))]; + tensor linear_41_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113228544)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113230144)))]; + tensor hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor("hidden_states_43_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113231744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113674176))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113674368)))]; + tensor linear_42_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_736_to_fp16 = const()[name = tensor("op_736_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_736_to_fp16)[name = tensor("tensor_47_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(113675968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114118400))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114118592)))]; + tensor linear_43_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_741 = const()[name = tensor("op_741"), val = tensor([1, -1, 12, 64])]; + tensor var_742_cast_fp16 = reshape(shape = var_741, x = linear_43_cast_fp16)[name = tensor("op_742_cast_fp16")]; + tensor var_743_perm_0 = const()[name = tensor("op_743_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114120192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114562624))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114562816)))]; + tensor linear_44_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_748 = const()[name = tensor("op_748"), val = tensor([1, -1, 12, 64])]; + tensor var_749_cast_fp16 = reshape(shape = var_748, x = linear_44_cast_fp16)[name = tensor("op_749_cast_fp16")]; + tensor var_750_perm_0 = const()[name = tensor("op_750_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_757 = const()[name = tensor("op_757"), val = tensor([1, 77, 12, 64])]; + tensor var_758_cast_fp16 = reshape(shape = var_757, x = tensor_47_cast_fp16)[name = tensor("op_758_cast_fp16")]; + tensor var_759_perm_0 = const()[name = tensor("op_759_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_761 = const()[name = tensor("op_761"), val = tensor([12, -1, 64])]; + tensor transpose_19 = transpose(perm = var_759_perm_0, x = var_758_cast_fp16)[name = tensor("transpose_19")]; + tensor query_states_15_cast_fp16 = reshape(shape = var_761, x = transpose_19)[name = tensor("query_states_15_cast_fp16")]; + tensor var_763 = const()[name = tensor("op_763"), val = tensor([12, -1, 64])]; + tensor transpose_18 = transpose(perm = var_743_perm_0, x = var_742_cast_fp16)[name = tensor("transpose_18")]; + tensor key_states_31_cast_fp16 = reshape(shape = var_763, x = transpose_18)[name = tensor("key_states_31_cast_fp16")]; + tensor var_765 = const()[name = tensor("op_765"), val = tensor([12, -1, 64])]; + tensor transpose_17 = transpose(perm = var_750_perm_0, x = var_749_cast_fp16)[name = tensor("transpose_17")]; + tensor value_states_31_cast_fp16 = reshape(shape = var_765, x = transpose_17)[name = tensor("value_states_31_cast_fp16")]; + tensor attn_weights_43_transpose_x_1 = const()[name = tensor("attn_weights_43_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_43_transpose_y_1 = const()[name = tensor("attn_weights_43_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = query_states_15_cast_fp16, y = key_states_31_cast_fp16)[name = tensor("attn_weights_43_cast_fp16")]; + tensor var_770 = const()[name = tensor("op_770"), val = tensor([1, 12, 77, 77])]; + tensor var_771_cast_fp16 = reshape(shape = var_770, x = attn_weights_43_cast_fp16)[name = tensor("op_771_cast_fp16")]; + tensor attn_weights_45_cast_fp16 = add(x = var_771_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_45_cast_fp16")]; + tensor var_776 = const()[name = tensor("op_776"), val = tensor([12, 77, 77])]; + tensor input_117_cast_fp16 = reshape(shape = var_776, x = attn_weights_45_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor attn_output_43_transpose_x_0 = const()[name = tensor("attn_output_43_transpose_x_0"), val = tensor(false)]; + tensor attn_output_43_transpose_y_0 = const()[name = tensor("attn_output_43_transpose_y_0"), val = tensor(false)]; + tensor attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast_fp16, y = value_states_31_cast_fp16)[name = tensor("attn_output_43_cast_fp16")]; + tensor var_781 = const()[name = tensor("op_781"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_45_cast_fp16 = reshape(shape = var_781, x = attn_output_43_cast_fp16)[name = tensor("attn_output_45_cast_fp16")]; + tensor attn_output_47_perm_0 = const()[name = tensor("attn_output_47_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_784 = const()[name = tensor("op_784"), val = tensor([1, 77, 768])]; + tensor transpose_16 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor("transpose_16")]; + tensor input_121_cast_fp16 = reshape(shape = var_784, x = transpose_16)[name = tensor("input_121_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114564416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115006848))), name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115007040)))]; + tensor linear_45_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor input_125_axes_0 = const()[name = tensor("input_125_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115008640)))]; + tensor text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115010240)))]; + tensor input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115011840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116781376))), name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116781568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116783936))), name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor var_799_to_fp16 = const()[name = tensor("op_799_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_800_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_799_to_fp16)[name = tensor("op_800_cast_fp16")]; + tensor var_801_cast_fp16 = sigmoid(x = var_800_cast_fp16)[name = tensor("op_801_cast_fp16")]; + tensor input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_801_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116784128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118553664))), name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118553856)))]; + tensor linear_47_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118555456)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118557056)))]; + tensor hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor("hidden_states_49_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(118558656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119001088))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119001280)))]; + tensor linear_48_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_826_to_fp16 = const()[name = tensor("op_826_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_826_to_fp16)[name = tensor("tensor_53_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119002880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119445312))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119445504)))]; + tensor linear_49_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_831 = const()[name = tensor("op_831"), val = tensor([1, -1, 12, 64])]; + tensor var_832_cast_fp16 = reshape(shape = var_831, x = linear_49_cast_fp16)[name = tensor("op_832_cast_fp16")]; + tensor var_833_perm_0 = const()[name = tensor("op_833_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119447104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119889536))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119889728)))]; + tensor linear_50_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_838 = const()[name = tensor("op_838"), val = tensor([1, -1, 12, 64])]; + tensor var_839_cast_fp16 = reshape(shape = var_838, x = linear_50_cast_fp16)[name = tensor("op_839_cast_fp16")]; + tensor var_840_perm_0 = const()[name = tensor("op_840_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_847 = const()[name = tensor("op_847"), val = tensor([1, 77, 12, 64])]; + tensor var_848_cast_fp16 = reshape(shape = var_847, x = tensor_53_cast_fp16)[name = tensor("op_848_cast_fp16")]; + tensor var_849_perm_0 = const()[name = tensor("op_849_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_851 = const()[name = tensor("op_851"), val = tensor([12, -1, 64])]; + tensor transpose_15 = transpose(perm = var_849_perm_0, x = var_848_cast_fp16)[name = tensor("transpose_15")]; + tensor query_states_17_cast_fp16 = reshape(shape = var_851, x = transpose_15)[name = tensor("query_states_17_cast_fp16")]; + tensor var_853 = const()[name = tensor("op_853"), val = tensor([12, -1, 64])]; + tensor transpose_14 = transpose(perm = var_833_perm_0, x = var_832_cast_fp16)[name = tensor("transpose_14")]; + tensor key_states_35_cast_fp16 = reshape(shape = var_853, x = transpose_14)[name = tensor("key_states_35_cast_fp16")]; + tensor var_855 = const()[name = tensor("op_855"), val = tensor([12, -1, 64])]; + tensor transpose_13 = transpose(perm = var_840_perm_0, x = var_839_cast_fp16)[name = tensor("transpose_13")]; + tensor value_states_35_cast_fp16 = reshape(shape = var_855, x = transpose_13)[name = tensor("value_states_35_cast_fp16")]; + tensor attn_weights_49_transpose_x_1 = const()[name = tensor("attn_weights_49_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_49_transpose_y_1 = const()[name = tensor("attn_weights_49_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = query_states_17_cast_fp16, y = key_states_35_cast_fp16)[name = tensor("attn_weights_49_cast_fp16")]; + tensor var_860 = const()[name = tensor("op_860"), val = tensor([1, 12, 77, 77])]; + tensor var_861_cast_fp16 = reshape(shape = var_860, x = attn_weights_49_cast_fp16)[name = tensor("op_861_cast_fp16")]; + tensor attn_weights_51_cast_fp16 = add(x = var_861_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_51_cast_fp16")]; + tensor var_866 = const()[name = tensor("op_866"), val = tensor([12, 77, 77])]; + tensor input_133_cast_fp16 = reshape(shape = var_866, x = attn_weights_51_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor("input_135_cast_fp16")]; + tensor attn_output_49_transpose_x_0 = const()[name = tensor("attn_output_49_transpose_x_0"), val = tensor(false)]; + tensor attn_output_49_transpose_y_0 = const()[name = tensor("attn_output_49_transpose_y_0"), val = tensor(false)]; + tensor attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast_fp16, y = value_states_35_cast_fp16)[name = tensor("attn_output_49_cast_fp16")]; + tensor var_871 = const()[name = tensor("op_871"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_51_cast_fp16 = reshape(shape = var_871, x = attn_output_49_cast_fp16)[name = tensor("attn_output_51_cast_fp16")]; + tensor attn_output_53_perm_0 = const()[name = tensor("attn_output_53_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_874 = const()[name = tensor("op_874"), val = tensor([1, 77, 768])]; + tensor transpose_12 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor("transpose_12")]; + tensor input_137_cast_fp16 = reshape(shape = var_874, x = transpose_12)[name = tensor("input_137_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119891328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120333760))), name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120333952)))]; + tensor linear_51_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor input_141_axes_0 = const()[name = tensor("input_141_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120335552)))]; + tensor text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120337152)))]; + tensor input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120338752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122108288))), name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122108480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122110848))), name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor var_889_to_fp16 = const()[name = tensor("op_889_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_890_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_889_to_fp16)[name = tensor("op_890_cast_fp16")]; + tensor var_891_cast_fp16 = sigmoid(x = var_890_cast_fp16)[name = tensor("op_891_cast_fp16")]; + tensor input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_891_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(122111040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123880576))), name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123880768)))]; + tensor linear_53_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor hidden_states_55_axes_0 = const()[name = tensor("hidden_states_55_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123882368)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123883968)))]; + tensor hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor("hidden_states_55_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123885568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124328000))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124328192)))]; + tensor linear_54_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_916_to_fp16)[name = tensor("tensor_59_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124329792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124772224))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124772416)))]; + tensor linear_55_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_921 = const()[name = tensor("op_921"), val = tensor([1, -1, 12, 64])]; + tensor var_922_cast_fp16 = reshape(shape = var_921, x = linear_55_cast_fp16)[name = tensor("op_922_cast_fp16")]; + tensor var_923_perm_0 = const()[name = tensor("op_923_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124774016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125216448))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125216640)))]; + tensor linear_56_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_928 = const()[name = tensor("op_928"), val = tensor([1, -1, 12, 64])]; + tensor var_929_cast_fp16 = reshape(shape = var_928, x = linear_56_cast_fp16)[name = tensor("op_929_cast_fp16")]; + tensor var_930_perm_0 = const()[name = tensor("op_930_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_937 = const()[name = tensor("op_937"), val = tensor([1, 77, 12, 64])]; + tensor var_938_cast_fp16 = reshape(shape = var_937, x = tensor_59_cast_fp16)[name = tensor("op_938_cast_fp16")]; + tensor var_939_perm_0 = const()[name = tensor("op_939_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_941 = const()[name = tensor("op_941"), val = tensor([12, -1, 64])]; + tensor transpose_11 = transpose(perm = var_939_perm_0, x = var_938_cast_fp16)[name = tensor("transpose_11")]; + tensor query_states_19_cast_fp16 = reshape(shape = var_941, x = transpose_11)[name = tensor("query_states_19_cast_fp16")]; + tensor var_943 = const()[name = tensor("op_943"), val = tensor([12, -1, 64])]; + tensor transpose_10 = transpose(perm = var_923_perm_0, x = var_922_cast_fp16)[name = tensor("transpose_10")]; + tensor key_states_39_cast_fp16 = reshape(shape = var_943, x = transpose_10)[name = tensor("key_states_39_cast_fp16")]; + tensor var_945 = const()[name = tensor("op_945"), val = tensor([12, -1, 64])]; + tensor transpose_9 = transpose(perm = var_930_perm_0, x = var_929_cast_fp16)[name = tensor("transpose_9")]; + tensor value_states_39_cast_fp16 = reshape(shape = var_945, x = transpose_9)[name = tensor("value_states_39_cast_fp16")]; + tensor attn_weights_55_transpose_x_1 = const()[name = tensor("attn_weights_55_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_55_transpose_y_1 = const()[name = tensor("attn_weights_55_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = query_states_19_cast_fp16, y = key_states_39_cast_fp16)[name = tensor("attn_weights_55_cast_fp16")]; + tensor var_950 = const()[name = tensor("op_950"), val = tensor([1, 12, 77, 77])]; + tensor var_951_cast_fp16 = reshape(shape = var_950, x = attn_weights_55_cast_fp16)[name = tensor("op_951_cast_fp16")]; + tensor attn_weights_57_cast_fp16 = add(x = var_951_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_57_cast_fp16")]; + tensor var_956 = const()[name = tensor("op_956"), val = tensor([12, 77, 77])]; + tensor input_149_cast_fp16 = reshape(shape = var_956, x = attn_weights_57_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor attn_output_55_transpose_x_0 = const()[name = tensor("attn_output_55_transpose_x_0"), val = tensor(false)]; + tensor attn_output_55_transpose_y_0 = const()[name = tensor("attn_output_55_transpose_y_0"), val = tensor(false)]; + tensor attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast_fp16, y = value_states_39_cast_fp16)[name = tensor("attn_output_55_cast_fp16")]; + tensor var_961 = const()[name = tensor("op_961"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_57_cast_fp16 = reshape(shape = var_961, x = attn_output_55_cast_fp16)[name = tensor("attn_output_57_cast_fp16")]; + tensor attn_output_59_perm_0 = const()[name = tensor("attn_output_59_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_964 = const()[name = tensor("op_964"), val = tensor([1, 77, 768])]; + tensor transpose_8 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor("transpose_8")]; + tensor input_153_cast_fp16 = reshape(shape = var_964, x = transpose_8)[name = tensor("input_153_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125218240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125660672))), name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125660864)))]; + tensor linear_57_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor input_157_axes_0 = const()[name = tensor("input_157_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125662464)))]; + tensor text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125664064)))]; + tensor input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor("input_157_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(125665664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127435200))), name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127435392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127437760))), name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor var_979_to_fp16 = const()[name = tensor("op_979_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_980_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_979_to_fp16)[name = tensor("op_980_cast_fp16")]; + tensor var_981_cast_fp16 = sigmoid(x = var_980_cast_fp16)[name = tensor("op_981_cast_fp16")]; + tensor input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_981_cast_fp16)[name = tensor("input_161_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127437952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129207488))), name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129207680)))]; + tensor linear_59_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor hidden_states_61_axes_0 = const()[name = tensor("hidden_states_61_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129209280)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129210880)))]; + tensor hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor("hidden_states_61_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129212480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129654912))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129655104)))]; + tensor linear_60_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_1006_to_fp16)[name = tensor("tensor_65_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129656704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130099136))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130099328)))]; + tensor linear_61_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_1011 = const()[name = tensor("op_1011"), val = tensor([1, -1, 12, 64])]; + tensor var_1012_cast_fp16 = reshape(shape = var_1011, x = linear_61_cast_fp16)[name = tensor("op_1012_cast_fp16")]; + tensor var_1013_perm_0 = const()[name = tensor("op_1013_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130100928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130543360))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130543552)))]; + tensor linear_62_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_1018 = const()[name = tensor("op_1018"), val = tensor([1, -1, 12, 64])]; + tensor var_1019_cast_fp16 = reshape(shape = var_1018, x = linear_62_cast_fp16)[name = tensor("op_1019_cast_fp16")]; + tensor var_1020_perm_0 = const()[name = tensor("op_1020_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1027 = const()[name = tensor("op_1027"), val = tensor([1, 77, 12, 64])]; + tensor var_1028_cast_fp16 = reshape(shape = var_1027, x = tensor_65_cast_fp16)[name = tensor("op_1028_cast_fp16")]; + tensor var_1029_perm_0 = const()[name = tensor("op_1029_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1031 = const()[name = tensor("op_1031"), val = tensor([12, -1, 64])]; + tensor transpose_7 = transpose(perm = var_1029_perm_0, x = var_1028_cast_fp16)[name = tensor("transpose_7")]; + tensor query_states_21_cast_fp16 = reshape(shape = var_1031, x = transpose_7)[name = tensor("query_states_21_cast_fp16")]; + tensor var_1033 = const()[name = tensor("op_1033"), val = tensor([12, -1, 64])]; + tensor transpose_6 = transpose(perm = var_1013_perm_0, x = var_1012_cast_fp16)[name = tensor("transpose_6")]; + tensor key_states_43_cast_fp16 = reshape(shape = var_1033, x = transpose_6)[name = tensor("key_states_43_cast_fp16")]; + tensor var_1035 = const()[name = tensor("op_1035"), val = tensor([12, -1, 64])]; + tensor transpose_5 = transpose(perm = var_1020_perm_0, x = var_1019_cast_fp16)[name = tensor("transpose_5")]; + tensor value_states_43_cast_fp16 = reshape(shape = var_1035, x = transpose_5)[name = tensor("value_states_43_cast_fp16")]; + tensor attn_weights_61_transpose_x_1 = const()[name = tensor("attn_weights_61_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_61_transpose_y_1 = const()[name = tensor("attn_weights_61_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = query_states_21_cast_fp16, y = key_states_43_cast_fp16)[name = tensor("attn_weights_61_cast_fp16")]; + tensor var_1040 = const()[name = tensor("op_1040"), val = tensor([1, 12, 77, 77])]; + tensor var_1041_cast_fp16 = reshape(shape = var_1040, x = attn_weights_61_cast_fp16)[name = tensor("op_1041_cast_fp16")]; + tensor attn_weights_63_cast_fp16 = add(x = var_1041_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_63_cast_fp16")]; + tensor var_1046 = const()[name = tensor("op_1046"), val = tensor([12, 77, 77])]; + tensor input_165_cast_fp16 = reshape(shape = var_1046, x = attn_weights_63_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor attn_output_61_transpose_x_0 = const()[name = tensor("attn_output_61_transpose_x_0"), val = tensor(false)]; + tensor attn_output_61_transpose_y_0 = const()[name = tensor("attn_output_61_transpose_y_0"), val = tensor(false)]; + tensor attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast_fp16, y = value_states_43_cast_fp16)[name = tensor("attn_output_61_cast_fp16")]; + tensor var_1051 = const()[name = tensor("op_1051"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_63_cast_fp16 = reshape(shape = var_1051, x = attn_output_61_cast_fp16)[name = tensor("attn_output_63_cast_fp16")]; + tensor attn_output_65_perm_0 = const()[name = tensor("attn_output_65_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1054 = const()[name = tensor("op_1054"), val = tensor([1, 77, 768])]; + tensor transpose_4 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor("transpose_4")]; + tensor input_169_cast_fp16 = reshape(shape = var_1054, x = transpose_4)[name = tensor("input_169_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130545152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130987584))), name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130987776)))]; + tensor linear_63_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor input_173_axes_0 = const()[name = tensor("input_173_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130989376)))]; + tensor text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130990976)))]; + tensor input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130992576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132762112))), name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132762304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132764672))), name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1070_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1069_to_fp16)[name = tensor("op_1070_cast_fp16")]; + tensor var_1071_cast_fp16 = sigmoid(x = var_1070_cast_fp16)[name = tensor("op_1071_cast_fp16")]; + tensor input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1071_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132764864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134534400))), name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134534592)))]; + tensor linear_65_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor input_179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("input_179_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134536192)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134537792)))]; + tensor hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor("hidden_states_67_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134539392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134981824))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134982016)))]; + tensor linear_66_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1096_to_fp16 = const()[name = tensor("op_1096_to_fp16"), val = tensor(0x1p-3)]; + tensor tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1096_to_fp16)[name = tensor("tensor_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134983616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135426048))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135426240)))]; + tensor linear_67_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1101 = const()[name = tensor("op_1101"), val = tensor([1, -1, 12, 64])]; + tensor var_1102_cast_fp16 = reshape(shape = var_1101, x = linear_67_cast_fp16)[name = tensor("op_1102_cast_fp16")]; + tensor var_1103_perm_0 = const()[name = tensor("op_1103_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135427840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135870272))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135870464)))]; + tensor linear_68_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1108 = const()[name = tensor("op_1108"), val = tensor([1, -1, 12, 64])]; + tensor var_1109_cast_fp16 = reshape(shape = var_1108, x = linear_68_cast_fp16)[name = tensor("op_1109_cast_fp16")]; + tensor var_1110_perm_0 = const()[name = tensor("op_1110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1117 = const()[name = tensor("op_1117"), val = tensor([1, 77, 12, 64])]; + tensor var_1118_cast_fp16 = reshape(shape = var_1117, x = tensor_cast_fp16)[name = tensor("op_1118_cast_fp16")]; + tensor var_1119_perm_0 = const()[name = tensor("op_1119_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([12, -1, 64])]; + tensor transpose_3 = transpose(perm = var_1119_perm_0, x = var_1118_cast_fp16)[name = tensor("transpose_3")]; + tensor query_states_cast_fp16 = reshape(shape = var_1121, x = transpose_3)[name = tensor("query_states_cast_fp16")]; + tensor var_1123 = const()[name = tensor("op_1123"), val = tensor([12, -1, 64])]; + tensor transpose_2 = transpose(perm = var_1103_perm_0, x = var_1102_cast_fp16)[name = tensor("transpose_2")]; + tensor key_states_cast_fp16 = reshape(shape = var_1123, x = transpose_2)[name = tensor("key_states_cast_fp16")]; + tensor var_1125 = const()[name = tensor("op_1125"), val = tensor([12, -1, 64])]; + tensor transpose_1 = transpose(perm = var_1110_perm_0, x = var_1109_cast_fp16)[name = tensor("transpose_1")]; + tensor value_states_cast_fp16 = reshape(shape = var_1125, x = transpose_1)[name = tensor("value_states_cast_fp16")]; + tensor attn_weights_67_transpose_x_1 = const()[name = tensor("attn_weights_67_transpose_x_1"), val = tensor(false)]; + tensor attn_weights_67_transpose_y_1 = const()[name = tensor("attn_weights_67_transpose_y_1"), val = tensor(true)]; + tensor attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = query_states_cast_fp16, y = key_states_cast_fp16)[name = tensor("attn_weights_67_cast_fp16")]; + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 12, 77, 77])]; + tensor var_1131_cast_fp16 = reshape(shape = var_1130, x = attn_weights_67_cast_fp16)[name = tensor("op_1131_cast_fp16")]; + tensor attn_weights_69_cast_fp16 = add(x = var_1131_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor("attn_weights_69_cast_fp16")]; + tensor var_1136 = const()[name = tensor("op_1136"), val = tensor([12, 77, 77])]; + tensor input_181_cast_fp16 = reshape(shape = var_1136, x = attn_weights_69_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor attn_output_67_transpose_x_0 = const()[name = tensor("attn_output_67_transpose_x_0"), val = tensor(false)]; + tensor attn_output_67_transpose_y_0 = const()[name = tensor("attn_output_67_transpose_y_0"), val = tensor(false)]; + tensor attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast_fp16, y = value_states_cast_fp16)[name = tensor("attn_output_67_cast_fp16")]; + tensor var_1141 = const()[name = tensor("op_1141"), val = tensor([1, 12, 77, 64])]; + tensor attn_output_69_cast_fp16 = reshape(shape = var_1141, x = attn_output_67_cast_fp16)[name = tensor("attn_output_69_cast_fp16")]; + tensor attn_output_perm_0 = const()[name = tensor("attn_output_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 77, 768])]; + tensor transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor("transpose_0")]; + tensor input_185_cast_fp16 = reshape(shape = var_1144, x = transpose_0)[name = tensor("input_185_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135872064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136314496))), name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor([768, 768])]; + tensor text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136314688)))]; + tensor linear_69_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor("input_187_cast_fp16")]; + tensor input_189_axes_0 = const()[name = tensor("input_189_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136316288)))]; + tensor text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136317888)))]; + tensor input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136319488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138089024))), name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor([3072, 768])]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138089216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138091584))), name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized"), shape = tensor([3072])]; + tensor linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(0x1.b3cp+0)]; + tensor var_1160_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1159_to_fp16)[name = tensor("op_1160_cast_fp16")]; + tensor var_1161_cast_fp16 = sigmoid(x = var_1160_cast_fp16)[name = tensor("op_1161_cast_fp16")]; + tensor input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1161_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138091776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139861312))), name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor([768, 3072])]; + tensor text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139861504)))]; + tensor linear_71_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor input_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor("input_cast_fp16")]; + tensor last_hidden_state_axes_0 = const()[name = tensor("last_hidden_state_axes_0"), val = tensor([-1])]; + tensor text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139863104)))]; + tensor text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139864704)))]; + tensor last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor("last_hidden_state_cast_fp16")]; + tensor var_1175 = const()[name = tensor("op_1175"), val = tensor([0])]; + tensor var_1177 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_2)[name = tensor("op_1177")]; + tensor stack_0_axis_0 = const()[name = tensor("stack_0_axis_0"), val = tensor(1)]; + tensor stack_0 = stack(axis = stack_0_axis_0, values = (var_1175, var_1177))[name = tensor("stack_0")]; + tensor var_1179_transpose_batch_dims_0 = const()[name = tensor("op_1179_transpose_batch_dims_0"), val = tensor(0)]; + tensor var_1179_transpose_cast_fp16 = gather_nd(batch_dims = var_1179_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor("op_1179_transpose_cast_fp16")]; + tensor var_1179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_1179_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor hidden_embeds = cast(dtype = input_179_cast_fp16_to_fp32_dtype_0, x = input_179_cast_fp16)[name = tensor("cast_0")]; + tensor pooled_outputs = cast(dtype = var_1179_cast_fp16_to_fp32_dtype_0, x = var_1179_transpose_cast_fp16)[name = tensor("cast_1")]; + } -> (hidden_embeds, pooled_outputs); +} \ No newline at end of file